diff options
Diffstat (limited to 'src/pluto/ipsec_doi.c')
-rw-r--r-- | src/pluto/ipsec_doi.c | 5630 |
1 files changed, 5630 insertions, 0 deletions
diff --git a/src/pluto/ipsec_doi.c b/src/pluto/ipsec_doi.c new file mode 100644 index 000000000..1c22b299b --- /dev/null +++ b/src/pluto/ipsec_doi.c @@ -0,0 +1,5630 @@ +/* IPsec DOI and Oakley resolution routines + * Copyright (C) 1997 Angelos D. Keromytis. + * Copyright (C) 1998-2002 D. Hugh Redelmeier. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License as published by the + * Free Software Foundation; either version 2 of the License, or (at your + * option) any later version. See <http://www.fsf.org/copyleft/gpl.txt>. + * + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY + * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * for more details. + * + * RCSID $Id: ipsec_doi.c,v 1.39 2006/04/22 21:59:20 as Exp $ + */ + +#include <stdio.h> +#include <string.h> +#include <stddef.h> +#include <stdlib.h> +#include <unistd.h> +#include <sys/socket.h> +#include <netinet/in.h> +#include <arpa/inet.h> +#include <resolv.h> +#include <arpa/nameser.h> /* missing from <resolv.h> on old systems */ +#include <sys/queue.h> +#include <sys/time.h> /* for gettimeofday */ + +#include <freeswan.h> +#include <ipsec_policy.h> + +#include "constants.h" +#include "defs.h" +#include "mp_defs.h" +#include "state.h" +#include "id.h" +#include "x509.h" +#include "crl.h" +#include "ca.h" +#include "certs.h" +#include "smartcard.h" +#include "connections.h" +#include "keys.h" +#include "packet.h" +#include "demux.h" /* needs packet.h */ +#include "adns.h" /* needs <resolv.h> */ +#include "dnskey.h" /* needs keys.h and adns.h */ +#include "kernel.h" +#include "log.h" +#include "cookie.h" +#include "server.h" +#include "spdb.h" +#include "timer.h" +#include "rnd.h" +#include "ipsec_doi.h" /* needs demux.h and state.h */ +#include "whack.h" +#include "fetch.h" +#include "pkcs7.h" +#include "asn1.h" + +#include "sha1.h" +#include "md5.h" +#include "crypto.h" /* requires sha1.h and md5.h */ +#include "vendor.h" +#include "alg_info.h" +#include "ike_alg.h" +#include "kernel_alg.h" +#include "nat_traversal.h" +#include "virtual.h" + +/* + * are we sending Pluto's Vendor ID? + */ +#ifdef VENDORID +#define SEND_PLUTO_VID 1 +#else /* !VENDORID */ +#define SEND_PLUTO_VID 0 +#endif /* !VENDORID */ + +/* + * are we sending a Cisco Unity VID? + */ +#ifdef CISCO_QUIRKS +#define SEND_CISCO_UNITY_VID 1 +#else /* !CISCO_QUIRKS */ +#define SEND_CISCO_UNITY_VID 0 +#endif /* !CISCO_QUIRKS */ + +/* MAGIC: perform f, a function that returns notification_t + * and return from the ENCLOSING stf_status returning function if it fails. + */ +#define RETURN_STF_FAILURE(f) \ + { int r = (f); if (r != NOTHING_WRONG) return STF_FAIL + r; } + +/* create output HDR as replica of input HDR */ +void +echo_hdr(struct msg_digest *md, bool enc, u_int8_t np) +{ + struct isakmp_hdr r_hdr = md->hdr; /* mostly same as incoming header */ + + r_hdr.isa_flags &= ~ISAKMP_FLAG_COMMIT; /* we won't ever turn on this bit */ + if (enc) + r_hdr.isa_flags |= ISAKMP_FLAG_ENCRYPTION; + /* some day, we may have to set r_hdr.isa_version */ + r_hdr.isa_np = np; + if (!out_struct(&r_hdr, &isakmp_hdr_desc, &md->reply, &md->rbody)) + impossible(); /* surely must have room and be well-formed */ +} + +/* Compute DH shared secret from our local secret and the peer's public value. + * We make the leap that the length should be that of the group + * (see quoted passage at start of ACCEPT_KE). + */ +static void +compute_dh_shared(struct state *st, const chunk_t g +, const struct oakley_group_desc *group) +{ + MP_INT mp_g, mp_shared; + struct timeval tv0, tv1; + unsigned long tv_diff; + + gettimeofday(&tv0, NULL); + passert(st->st_sec_in_use); + n_to_mpz(&mp_g, g.ptr, g.len); + mpz_init(&mp_shared); + mpz_powm(&mp_shared, &mp_g, &st->st_sec, group->modulus); + mpz_clear(&mp_g); + freeanychunk(st->st_shared); /* happens in odd error cases */ + st->st_shared = mpz_to_n(&mp_shared, group->bytes); + mpz_clear(&mp_shared); + gettimeofday(&tv1, NULL); + tv_diff=(tv1.tv_sec - tv0.tv_sec) * 1000000 + (tv1.tv_usec - tv0.tv_usec); + DBG(DBG_CRYPT, + DBG_log("compute_dh_shared(): time elapsed (%s): %ld usec" + , enum_show(&oakley_group_names, st->st_oakley.group->group) + , tv_diff); + ); + /* if took more than 200 msec ... */ + if (tv_diff > 200000) { + loglog(RC_LOG_SERIOUS, "WARNING: compute_dh_shared(): for %s took " + "%ld usec" + , enum_show(&oakley_group_names, st->st_oakley.group->group) + , tv_diff); + } + + DBG_cond_dump_chunk(DBG_CRYPT, "DH shared secret:\n", st->st_shared); +} + +/* if we haven't already done so, compute a local DH secret (st->st_sec) and + * the corresponding public value (g). This is emitted as a KE payload. + */ +static bool +build_and_ship_KE(struct state *st, chunk_t *g +, const struct oakley_group_desc *group, pb_stream *outs, u_int8_t np) +{ + if (!st->st_sec_in_use) + { + u_char tmp[LOCALSECRETSIZE]; + MP_INT mp_g; + + get_rnd_bytes(tmp, LOCALSECRETSIZE); + st->st_sec_in_use = TRUE; + n_to_mpz(&st->st_sec, tmp, LOCALSECRETSIZE); + + mpz_init(&mp_g); + mpz_powm(&mp_g, &groupgenerator, &st->st_sec, group->modulus); + freeanychunk(*g); /* happens in odd error cases */ + *g = mpz_to_n(&mp_g, group->bytes); + mpz_clear(&mp_g); + DBG(DBG_CRYPT, + DBG_dump("Local DH secret:\n", tmp, LOCALSECRETSIZE); + DBG_dump_chunk("Public DH value sent:\n", *g)); + } + return out_generic_chunk(np, &isakmp_keyex_desc, outs, *g, "keyex value"); +} + +/* accept_ke + * + * Check and accept DH public value (Gi or Gr) from peer's message. + * According to RFC2409 "The Internet key exchange (IKE)" 5: + * The Diffie-Hellman public value passed in a KE payload, in either + * a phase 1 or phase 2 exchange, MUST be the length of the negotiated + * Diffie-Hellman group enforced, if necessary, by pre-pending the + * value with zeros. + */ +static notification_t +accept_KE(chunk_t *dest, const char *val_name +, const struct oakley_group_desc *gr +, pb_stream *pbs) +{ + if (pbs_left(pbs) != gr->bytes) + { + loglog(RC_LOG_SERIOUS, "KE has %u byte DH public value; %u required" + , (unsigned) pbs_left(pbs), (unsigned) gr->bytes); + /* XXX Could send notification back */ + return INVALID_KEY_INFORMATION; + } + clonereplacechunk(*dest, pbs->cur, pbs_left(pbs), val_name); + DBG_cond_dump_chunk(DBG_CRYPT, "DH public value received:\n", *dest); + return NOTHING_WRONG; +} + +/* accept_PFS_KE + * + * Check and accept optional Quick Mode KE payload for PFS. + * Extends ACCEPT_PFS to check whether KE is allowed or required. + */ +static notification_t +accept_PFS_KE(struct msg_digest *md, chunk_t *dest +, const char *val_name, const char *msg_name) +{ + struct state *st = md->st; + struct payload_digest *const ke_pd = md->chain[ISAKMP_NEXT_KE]; + + if (ke_pd == NULL) + { + if (st->st_pfs_group != NULL) + { + loglog(RC_LOG_SERIOUS, "missing KE payload in %s message", msg_name); + return INVALID_KEY_INFORMATION; + } + } + else + { + if (st->st_pfs_group == NULL) + { + loglog(RC_LOG_SERIOUS, "%s message KE payload requires a GROUP_DESCRIPTION attribute in SA" + , msg_name); + return INVALID_KEY_INFORMATION; + } + if (ke_pd->next != NULL) + { + loglog(RC_LOG_SERIOUS, "%s message contains several KE payloads; we accept at most one", msg_name); + return INVALID_KEY_INFORMATION; /* ??? */ + } + return accept_KE(dest, val_name, st->st_pfs_group, &ke_pd->pbs); + } + return NOTHING_WRONG; +} + +static bool +build_and_ship_nonce(chunk_t *n, pb_stream *outs, u_int8_t np +, const char *name) +{ + freeanychunk(*n); + setchunk(*n, alloc_bytes(DEFAULT_NONCE_SIZE, name), DEFAULT_NONCE_SIZE); + get_rnd_bytes(n->ptr, DEFAULT_NONCE_SIZE); + return out_generic_chunk(np, &isakmp_nonce_desc, outs, *n, name); +} + +static bool +collect_rw_ca_candidates(struct msg_digest *md, generalName_t **top) +{ + struct connection *d = find_host_connection(&md->iface->addr + , pluto_port, (ip_address*)NULL, md->sender_port, LEMPTY); + + for (; d != NULL; d = d->hp_next) + { + /* must be a road warrior connection */ + if (d->kind == CK_TEMPLATE && !(d->policy & POLICY_OPPO) + && d->spd.that.ca.ptr != NULL) + { + generalName_t *gn; + bool new_entry = TRUE; + + for (gn = *top; gn != NULL; gn = gn->next) + { + if (same_dn(gn->name, d->spd.that.ca)) + { + new_entry = FALSE; + break; + } + } + if (new_entry) + { + gn = alloc_thing(generalName_t, "generalName"); + gn->kind = GN_DIRECTORY_NAME; + gn->name = d->spd.that.ca; + gn->next = *top; + *top = gn; + } + } + } + return *top != NULL; +} + +static bool +build_and_ship_CR(u_int8_t type, chunk_t ca, pb_stream *outs, u_int8_t np) +{ + pb_stream cr_pbs; + struct isakmp_cr cr_hd; + cr_hd.isacr_np = np; + cr_hd.isacr_type = type; + + /* build CR header */ + if (!out_struct(&cr_hd, &isakmp_ipsec_cert_req_desc, outs, &cr_pbs)) + return FALSE; + + if (ca.ptr != NULL) + { + /* build CR body containing the distinguished name of the CA */ + if (!out_chunk(ca, &cr_pbs, "CA")) + return FALSE; + } + close_output_pbs(&cr_pbs); + return TRUE; +} + +/* Send a notification to the peer. We could decide + * whether to send the notification, based on the type and the + * destination, if we care to. + */ +static void +send_notification(struct state *sndst, u_int16_t type, struct state *encst, + msgid_t msgid, u_char *icookie, u_char *rcookie, + u_char *spi, size_t spisize, u_char protoid) +{ + u_char buffer[1024]; + pb_stream pbs, r_hdr_pbs; + u_char *r_hashval = NULL; /* where in reply to jam hash value */ + u_char *r_hash_start = NULL; /* start of what is to be hashed */ + + passert((sndst) && (sndst->st_connection)); + + plog("sending %snotification %s to %s:%u" + , encst ? "encrypted " : "" + , enum_name(¬ification_names, type) + , ip_str(&sndst->st_connection->spd.that.host_addr) + , (unsigned)sndst->st_connection->spd.that.host_port); + + memset(buffer, 0, sizeof(buffer)); + init_pbs(&pbs, buffer, sizeof(buffer), "ISAKMP notify"); + + /* HDR* */ + { + struct isakmp_hdr hdr; + + hdr.isa_version = ISAKMP_MAJOR_VERSION << ISA_MAJ_SHIFT | ISAKMP_MINOR_VERSION; + hdr.isa_np = encst ? ISAKMP_NEXT_HASH : ISAKMP_NEXT_N; + hdr.isa_xchg = ISAKMP_XCHG_INFO; + hdr.isa_msgid = msgid; + hdr.isa_flags = encst ? ISAKMP_FLAG_ENCRYPTION : 0; + if (icookie) + memcpy(hdr.isa_icookie, icookie, COOKIE_SIZE); + if (rcookie) + memcpy(hdr.isa_rcookie, rcookie, COOKIE_SIZE); + if (!out_struct(&hdr, &isakmp_hdr_desc, &pbs, &r_hdr_pbs)) + impossible(); + } + + /* HASH -- value to be filled later */ + if (encst) + { + pb_stream hash_pbs; + if (!out_generic(ISAKMP_NEXT_N, &isakmp_hash_desc, &r_hdr_pbs, + &hash_pbs)) + impossible(); + r_hashval = hash_pbs.cur; /* remember where to plant value */ + if (!out_zero( + encst->st_oakley.hasher->hash_digest_size, &hash_pbs, "HASH")) + impossible(); + close_output_pbs(&hash_pbs); + r_hash_start = r_hdr_pbs.cur; /* hash from after HASH */ + } + + /* Notification Payload */ + { + pb_stream not_pbs; + struct isakmp_notification isan; + + isan.isan_doi = ISAKMP_DOI_IPSEC; + isan.isan_np = ISAKMP_NEXT_NONE; + isan.isan_type = type; + isan.isan_spisize = spisize; + isan.isan_protoid = protoid; + + if (!out_struct(&isan, &isakmp_notification_desc, &r_hdr_pbs, ¬_pbs) + || !out_raw(spi, spisize, ¬_pbs, "spi")) + impossible(); + close_output_pbs(¬_pbs); + } + + /* calculate hash value and patch into Hash Payload */ + if (encst) + { + struct hmac_ctx ctx; + hmac_init_chunk(&ctx, encst->st_oakley.hasher, encst->st_skeyid_a); + hmac_update(&ctx, (u_char *) &msgid, sizeof(msgid_t)); + hmac_update(&ctx, r_hash_start, r_hdr_pbs.cur-r_hash_start); + hmac_final(r_hashval, &ctx); + + DBG(DBG_CRYPT, + DBG_log("HASH computed:"); + DBG_dump("", r_hashval, ctx.hmac_digest_size); + ) + } + + /* Encrypt message (preserve st_iv and st_new_iv) */ + if (encst) + { + u_char old_iv[MAX_DIGEST_LEN]; + u_char new_iv[MAX_DIGEST_LEN]; + + u_int old_iv_len = encst->st_iv_len; + u_int new_iv_len = encst->st_new_iv_len; + + if (old_iv_len > MAX_DIGEST_LEN || new_iv_len > MAX_DIGEST_LEN) + impossible(); + + memcpy(old_iv, encst->st_iv, old_iv_len); + memcpy(new_iv, encst->st_new_iv, new_iv_len); + + if (!IS_ISAKMP_SA_ESTABLISHED(encst->st_state)) + { + memcpy(encst->st_ph1_iv, encst->st_new_iv, encst->st_new_iv_len); + encst->st_ph1_iv_len = encst->st_new_iv_len; + } + init_phase2_iv(encst, &msgid); + if (!encrypt_message(&r_hdr_pbs, encst)) + impossible(); + + /* restore preserved st_iv and st_new_iv */ + memcpy(encst->st_iv, old_iv, old_iv_len); + memcpy(encst->st_new_iv, new_iv, new_iv_len); + encst->st_iv_len = old_iv_len; + encst->st_new_iv_len = new_iv_len; + } + else + { + close_output_pbs(&r_hdr_pbs); + } + + /* Send packet (preserve st_tpacket) */ + { + chunk_t saved_tpacket = sndst->st_tpacket; + + setchunk(sndst->st_tpacket, pbs.start, pbs_offset(&pbs)); + send_packet(sndst, "ISAKMP notify"); + sndst->st_tpacket = saved_tpacket; + } +} + +void +send_notification_from_state(struct state *st, enum state_kind state, + u_int16_t type) +{ + struct state *p1st; + + passert(st); + + if (state == STATE_UNDEFINED) + state = st->st_state; + + if (IS_QUICK(state)) + { + p1st = find_phase1_state(st->st_connection, ISAKMP_SA_ESTABLISHED_STATES); + if ((p1st == NULL) || (!IS_ISAKMP_SA_ESTABLISHED(p1st->st_state))) + { + loglog(RC_LOG_SERIOUS, + "no Phase1 state for Quick mode notification"); + return; + } + send_notification(st, type, p1st, generate_msgid(p1st), + st->st_icookie, st->st_rcookie, NULL, 0, PROTO_ISAKMP); + } + else if (IS_ISAKMP_ENCRYPTED(state) && st->st_enc_key.ptr != NULL) + { + send_notification(st, type, st, generate_msgid(st), + st->st_icookie, st->st_rcookie, NULL, 0, PROTO_ISAKMP); + } + else + { + /* no ISAKMP SA established - don't encrypt notification */ + send_notification(st, type, NULL, 0, + st->st_icookie, st->st_rcookie, NULL, 0, PROTO_ISAKMP); + } +} + +void +send_notification_from_md(struct msg_digest *md, u_int16_t type) +{ + /** + * Create a dummy state to be able to use send_packet in + * send_notification + * + * we need to set: + * st_connection->that.host_addr + * st_connection->that.host_port + * st_connection->interface + */ + struct state st; + struct connection cnx; + + passert(md); + + memset(&st, 0, sizeof(st)); + memset(&cnx, 0, sizeof(cnx)); + st.st_connection = &cnx; + cnx.spd.that.host_addr = md->sender; + cnx.spd.that.host_port = md->sender_port; + cnx.interface = md->iface; + + send_notification(&st, type, NULL, 0, + md->hdr.isa_icookie, md->hdr.isa_rcookie, NULL, 0, PROTO_ISAKMP); +} + +/* Send a Delete Notification to announce deletion of ISAKMP SA or + * inbound IPSEC SAs. Does nothing if no such SAs are being deleted. + * Delete Notifications cannot announce deletion of outbound IPSEC/ISAKMP SAs. + */ +void +send_delete(struct state *st) +{ + pb_stream reply_pbs; + pb_stream r_hdr_pbs; + msgid_t msgid; + u_char buffer[8192]; + struct state *p1st; + ip_said said[EM_MAXRELSPIS]; + ip_said *ns = said; + u_char + *r_hashval, /* where in reply to jam hash value */ + *r_hash_start; /* start of what is to be hashed */ + bool isakmp_sa = FALSE; + + if (IS_IPSEC_SA_ESTABLISHED(st->st_state)) + { + p1st = find_phase1_state(st->st_connection, ISAKMP_SA_ESTABLISHED_STATES); + if (p1st == NULL) + { + DBG(DBG_CONTROL, DBG_log("no Phase 1 state for Delete")); + return; + } + + if (st->st_ah.present) + { + ns->spi = st->st_ah.our_spi; + ns->dst = st->st_connection->spd.this.host_addr; + ns->proto = PROTO_IPSEC_AH; + ns++; + } + if (st->st_esp.present) + { + ns->spi = st->st_esp.our_spi; + ns->dst = st->st_connection->spd.this.host_addr; + ns->proto = PROTO_IPSEC_ESP; + ns++; + } + + passert(ns != said); /* there must be some SAs to delete */ + } + else if (IS_ISAKMP_SA_ESTABLISHED(st->st_state)) + { + p1st = st; + isakmp_sa = TRUE; + } + else + { + return; /* nothing to do */ + } + + msgid = generate_msgid(p1st); + + zero(buffer); + init_pbs(&reply_pbs, buffer, sizeof(buffer), "delete msg"); + + /* HDR* */ + { + struct isakmp_hdr hdr; + + hdr.isa_version = ISAKMP_MAJOR_VERSION << ISA_MAJ_SHIFT | ISAKMP_MINOR_VERSION; + hdr.isa_np = ISAKMP_NEXT_HASH; + hdr.isa_xchg = ISAKMP_XCHG_INFO; + hdr.isa_msgid = msgid; + hdr.isa_flags = ISAKMP_FLAG_ENCRYPTION; + memcpy(hdr.isa_icookie, p1st->st_icookie, COOKIE_SIZE); + memcpy(hdr.isa_rcookie, p1st->st_rcookie, COOKIE_SIZE); + if (!out_struct(&hdr, &isakmp_hdr_desc, &reply_pbs, &r_hdr_pbs)) + impossible(); + } + + /* HASH -- value to be filled later */ + { + pb_stream hash_pbs; + + if (!out_generic(ISAKMP_NEXT_D, &isakmp_hash_desc, &r_hdr_pbs, &hash_pbs)) + impossible(); + r_hashval = hash_pbs.cur; /* remember where to plant value */ + if (!out_zero(p1st->st_oakley.hasher->hash_digest_size, &hash_pbs, "HASH(1)")) + impossible(); + close_output_pbs(&hash_pbs); + r_hash_start = r_hdr_pbs.cur; /* hash from after HASH(1) */ + } + + /* Delete Payloads */ + if (isakmp_sa) + { + pb_stream del_pbs; + struct isakmp_delete isad; + u_char isakmp_spi[2*COOKIE_SIZE]; + + isad.isad_doi = ISAKMP_DOI_IPSEC; + isad.isad_np = ISAKMP_NEXT_NONE; + isad.isad_spisize = (2 * COOKIE_SIZE); + isad.isad_protoid = PROTO_ISAKMP; + isad.isad_nospi = 1; + + memcpy(isakmp_spi, st->st_icookie, COOKIE_SIZE); + memcpy(isakmp_spi+COOKIE_SIZE, st->st_rcookie, COOKIE_SIZE); + + if (!out_struct(&isad, &isakmp_delete_desc, &r_hdr_pbs, &del_pbs) + || !out_raw(&isakmp_spi, (2*COOKIE_SIZE), &del_pbs, "delete payload")) + impossible(); + close_output_pbs(&del_pbs); + } + else + { + while (ns != said) + { + + pb_stream del_pbs; + struct isakmp_delete isad; + + ns--; + isad.isad_doi = ISAKMP_DOI_IPSEC; + isad.isad_np = ns == said? ISAKMP_NEXT_NONE : ISAKMP_NEXT_D; + isad.isad_spisize = sizeof(ipsec_spi_t); + isad.isad_protoid = ns->proto; + + isad.isad_nospi = 1; + if (!out_struct(&isad, &isakmp_delete_desc, &r_hdr_pbs, &del_pbs) + || !out_raw(&ns->spi, sizeof(ipsec_spi_t), &del_pbs, "delete payload")) + impossible(); + close_output_pbs(&del_pbs); + } + } + + /* calculate hash value and patch into Hash Payload */ + { + struct hmac_ctx ctx; + hmac_init_chunk(&ctx, p1st->st_oakley.hasher, p1st->st_skeyid_a); + hmac_update(&ctx, (u_char *) &msgid, sizeof(msgid_t)); + hmac_update(&ctx, r_hash_start, r_hdr_pbs.cur-r_hash_start); + hmac_final(r_hashval, &ctx); + + DBG(DBG_CRYPT, + DBG_log("HASH(1) computed:"); + DBG_dump("", r_hashval, ctx.hmac_digest_size); + ) + } + + /* Do a dance to avoid needing a new state object. + * We use the Phase 1 State. This is the one with right + * IV, for one thing. + * The tricky bits are: + * - we need to preserve (save/restore) st_iv (but not st_iv_new) + * - we need to preserve (save/restore) st_tpacket. + */ + { + u_char old_iv[MAX_DIGEST_LEN]; + chunk_t saved_tpacket = p1st->st_tpacket; + + memcpy(old_iv, p1st->st_iv, p1st->st_iv_len); + init_phase2_iv(p1st, &msgid); + + if (!encrypt_message(&r_hdr_pbs, p1st)) + impossible(); + + setchunk(p1st->st_tpacket, reply_pbs.start, pbs_offset(&reply_pbs)); + send_packet(p1st, "delete notify"); + p1st->st_tpacket = saved_tpacket; + + /* get back old IV for this state */ + memcpy(p1st->st_iv, old_iv, p1st->st_iv_len); + } +} + +void +accept_delete(struct state *st, struct msg_digest *md, struct payload_digest *p) +{ + struct isakmp_delete *d = &(p->payload.delete); + size_t sizespi; + int i; + + if (!md->encrypted) + { + loglog(RC_LOG_SERIOUS, "ignoring Delete SA payload: not encrypted"); + return; + } + + if (!IS_ISAKMP_SA_ESTABLISHED(st->st_state)) + { + /* can't happen (if msg is encrypt), but just to be sure */ + loglog(RC_LOG_SERIOUS, "ignoring Delete SA payload: " + "ISAKMP SA not established"); + return; + } + + if (d->isad_nospi == 0) + { + loglog(RC_LOG_SERIOUS, "ignoring Delete SA payload: no SPI"); + return; + } + + switch (d->isad_protoid) + { + case PROTO_ISAKMP: + sizespi = 2 * COOKIE_SIZE; + break; + case PROTO_IPSEC_AH: + case PROTO_IPSEC_ESP: + sizespi = sizeof(ipsec_spi_t); + break; + case PROTO_IPCOMP: + /* nothing interesting to delete */ + return; + default: + loglog(RC_LOG_SERIOUS + , "ignoring Delete SA payload: unknown Protocol ID (%s)" + , enum_show(&protocol_names, d->isad_protoid)); + return; + } + + if (d->isad_spisize != sizespi) + { + loglog(RC_LOG_SERIOUS + , "ignoring Delete SA payload: bad SPI size (%d) for %s" + , d->isad_spisize, enum_show(&protocol_names, d->isad_protoid)); + return; + } + + if (pbs_left(&p->pbs) != d->isad_nospi * sizespi) + { + loglog(RC_LOG_SERIOUS + , "ignoring Delete SA payload: invalid payload size"); + return; + } + + for (i = 0; i < d->isad_nospi; i++) + { + u_char *spi = p->pbs.cur + (i * sizespi); + + if (d->isad_protoid == PROTO_ISAKMP) + { + /** + * ISAKMP + */ + struct state *dst = find_state(spi /*iCookie*/ + , spi+COOKIE_SIZE /*rCookie*/ + , &st->st_connection->spd.that.host_addr + , MAINMODE_MSGID); + + if (dst == NULL) + { + loglog(RC_LOG_SERIOUS, "ignoring Delete SA payload: " + "ISAKMP SA not found (maybe expired)"); + } + else if (!same_peer_ids(st->st_connection, dst->st_connection, NULL)) + { + /* we've not authenticated the relevant identities */ + loglog(RC_LOG_SERIOUS, "ignoring Delete SA payload: " + "ISAKMP SA used to convey Delete has different IDs from ISAKMP SA it deletes"); + } + else + { + struct connection *oldc; + + oldc = cur_connection; + set_cur_connection(dst->st_connection); + + if (nat_traversal_enabled) + nat_traversal_change_port_lookup(md, dst); + + loglog(RC_LOG_SERIOUS, "received Delete SA payload: " + "deleting ISAKMP State #%lu", dst->st_serialno); + delete_state(dst); + set_cur_connection(oldc); + } + } + else + { + /** + * IPSEC (ESP/AH) + */ + bool bogus; + struct state *dst = find_phase2_state_to_delete(st + , d->isad_protoid + , *(ipsec_spi_t *)spi /* network order */ + , &bogus); + + if (dst == NULL) + { + loglog(RC_LOG_SERIOUS + , "ignoring Delete SA payload: %s SA(0x%08lx) not found (%s)" + , enum_show(&protocol_names, d->isad_protoid) + , (unsigned long)ntohl((unsigned long)*(ipsec_spi_t *)spi) + , bogus ? "our SPI - bogus implementation" : "maybe expired"); + } + else + { + struct connection *rc = dst->st_connection; + struct connection *oldc; + + oldc = cur_connection; + set_cur_connection(rc); + + if (nat_traversal_enabled) + nat_traversal_change_port_lookup(md, dst); + + if (rc->newest_ipsec_sa == dst->st_serialno + && (rc->policy & POLICY_UP)) + { + /* Last IPSec SA for a permanent connection that we + * have initiated. Replace it in a few seconds. + * + * Useful if the other peer is rebooting. + */ +#define DELETE_SA_DELAY EVENT_RETRANSMIT_DELAY_0 + if (dst->st_event != NULL + && dst->st_event->ev_type == EVENT_SA_REPLACE + && dst->st_event->ev_time <= DELETE_SA_DELAY + now()) + { + /* Patch from Angus Lees to ignore retransmited + * Delete SA. + */ + loglog(RC_LOG_SERIOUS, "received Delete SA payload: " + "already replacing IPSEC State #%lu in %d seconds" + , dst->st_serialno, (int)(dst->st_event->ev_time - now())); + } + else + { + loglog(RC_LOG_SERIOUS, "received Delete SA payload: " + "replace IPSEC State #%lu in %d seconds" + , dst->st_serialno, DELETE_SA_DELAY); + dst->st_margin = DELETE_SA_DELAY; + delete_event(dst); + event_schedule(EVENT_SA_REPLACE, DELETE_SA_DELAY, dst); + } + } + else + { + loglog(RC_LOG_SERIOUS, "received Delete SA(0x%08lx) payload: " + "deleting IPSEC State #%lu" + , (unsigned long)ntohl((unsigned long)*(ipsec_spi_t *)spi) + , dst->st_serialno); + delete_state(dst); + } + + /* reset connection */ + set_cur_connection(oldc); + } + } + } +} + +/* The whole message must be a multiple of 4 octets. + * I'm not sure where this is spelled out, but look at + * rfc2408 3.6 Transform Payload. + * Note: it talks about 4 BYTE boundaries! + */ +void +close_message(pb_stream *pbs) +{ + size_t padding = pad_up(pbs_offset(pbs), 4); + + if (padding != 0) + (void) out_zero(padding, pbs, "message padding"); + close_output_pbs(pbs); +} + +/* Initiate an Oakley Main Mode exchange. + * --> HDR;SA + * Note: this is not called from demux.c + */ +static stf_status +main_outI1(int whack_sock, struct connection *c, struct state *predecessor + , lset_t policy, unsigned long try) +{ + struct state *st = new_state(); + pb_stream reply; /* not actually a reply, but you know what I mean */ + pb_stream rbody; + + int vids_to_send = 0; + + /* set up new state */ + st->st_connection = c; + set_cur_state(st); /* we must reset before exit */ + st->st_policy = policy & ~POLICY_IPSEC_MASK; + st->st_whack_sock = whack_sock; + st->st_try = try; + st->st_state = STATE_MAIN_I1; + + /* determine how many Vendor ID payloads we will be sending */ + if (SEND_PLUTO_VID) + vids_to_send++; + if (SEND_CISCO_UNITY_VID) + vids_to_send++; + if (c->spd.this.cert.type == CERT_PGP) + vids_to_send++; + /* always send XAUTH Vendor ID */ + vids_to_send++; + /* always send DPD Vendor ID */ + vids_to_send++; + if (nat_traversal_enabled) + vids_to_send++; + + get_cookie(TRUE, st->st_icookie, COOKIE_SIZE, &c->spd.that.host_addr); + + insert_state(st); /* needs cookies, connection, and msgid (0) */ + + if (HAS_IPSEC_POLICY(policy)) + add_pending(dup_any(whack_sock), st, c, policy, 1 + , predecessor == NULL? SOS_NOBODY : predecessor->st_serialno); + + if (predecessor == NULL) + plog("initiating Main Mode"); + else + plog("initiating Main Mode to replace #%lu", predecessor->st_serialno); + + /* set up reply */ + init_pbs(&reply, reply_buffer, sizeof(reply_buffer), "reply packet"); + + /* HDR out */ + { + struct isakmp_hdr hdr; + + zero(&hdr); /* default to 0 */ + hdr.isa_version = ISAKMP_MAJOR_VERSION << ISA_MAJ_SHIFT | ISAKMP_MINOR_VERSION; + hdr.isa_np = ISAKMP_NEXT_SA; + hdr.isa_xchg = ISAKMP_XCHG_IDPROT; + memcpy(hdr.isa_icookie, st->st_icookie, COOKIE_SIZE); + /* R-cookie, flags and MessageID are left zero */ + + if (!out_struct(&hdr, &isakmp_hdr_desc, &reply, &rbody)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + } + + /* SA out */ + { + u_char *sa_start = rbody.cur; + lset_t auth_policy = policy & POLICY_ID_AUTH_MASK; + + if (!out_sa(&rbody, &oakley_sadb, st, TRUE + , vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + + /* save initiator SA for later HASH */ + passert(st->st_p1isa.ptr == NULL); /* no leak! (MUST be first time) */ + clonetochunk(st->st_p1isa, sa_start, rbody.cur - sa_start + , "sa in main_outI1"); + } + + /* if enabled send Pluto Vendor ID */ + if (SEND_PLUTO_VID) + { + if (!out_vendorid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &rbody, VID_STRONGSWAN)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + } + + /* if enabled send Cisco Unity Vendor ID */ + if (SEND_CISCO_UNITY_VID) + { + if (!out_vendorid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &rbody, VID_CISCO_UNITY)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + } + /* if we have an OpenPGP certificate we assume an + * OpenPGP peer and have to send the Vendor ID + */ + if (c->spd.this.cert.type == CERT_PGP) + { + if (!out_vendorid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &rbody, VID_OPENPGP)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + } + + /* Announce our ability to do eXtended AUTHentication to the peer */ + if (!out_vendorid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &rbody, VID_MISC_XAUTH)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + + /* Announce our ability to do Dead Peer Detection to the peer */ + { + if (!out_vendorid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &rbody, VID_MISC_DPD)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + } + + if (nat_traversal_enabled) + { + /* Add supported NAT-Traversal VID */ + if (!nat_traversal_add_vid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &rbody)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + } + + close_message(&rbody); + close_output_pbs(&reply); + + clonetochunk(st->st_tpacket, reply.start, pbs_offset(&reply) + , "reply packet for main_outI1"); + + /* Transmit */ + + send_packet(st, "main_outI1"); + + /* Set up a retransmission event, half a minute henceforth */ + delete_event(st); + event_schedule(EVENT_RETRANSMIT, EVENT_RETRANSMIT_DELAY_0, st); + + if (predecessor != NULL) + { + update_pending(predecessor, st); + whack_log(RC_NEW_STATE + STATE_MAIN_I1 + , "%s: initiate, replacing #%lu" + , enum_name(&state_names, st->st_state) + , predecessor->st_serialno); + } + else + { + whack_log(RC_NEW_STATE + STATE_MAIN_I1 + , "%s: initiate", enum_name(&state_names, st->st_state)); + } + reset_cur_state(); + return STF_OK; +} + +void +ipsecdoi_initiate(int whack_sock +, struct connection *c +, lset_t policy +, unsigned long try +, so_serial_t replacing) +{ + /* If there's already an ISAKMP SA established, use that and + * go directly to Quick Mode. We are even willing to use one + * that is still being negotiated, but only if we are the Initiator + * (thus we can be sure that the IDs are not going to change; + * other issues around intent might matter). + * Note: there is no way to initiate with a Road Warrior. + */ + struct state *st = find_phase1_state(c + , ISAKMP_SA_ESTABLISHED_STATES | PHASE1_INITIATOR_STATES); + + if (st == NULL) + { + (void) main_outI1(whack_sock, c, NULL, policy, try); + } + else if (HAS_IPSEC_POLICY(policy)) + { + if (!IS_ISAKMP_SA_ESTABLISHED(st->st_state)) + { + /* leave our Phase 2 negotiation pending */ + add_pending(whack_sock, st, c, policy, try, replacing); + } + else + { + /* ??? we assume that peer_nexthop_sin isn't important: + * we already have it from when we negotiated the ISAKMP SA! + * It isn't clear what to do with the error return. + */ + (void) quick_outI1(whack_sock, st, c, policy, try, replacing); + } + } + else + { + close_any(whack_sock); + } +} + +/* Replace SA with a fresh one that is similar + * + * Shares some logic with ipsecdoi_initiate, but not the same! + * - we must not reuse the ISAKMP SA if we are trying to replace it! + * - if trying to replace IPSEC SA, use ipsecdoi_initiate to build + * ISAKMP SA if needed. + * - duplicate whack fd, if live. + * Does not delete the old state -- someone else will do that. + */ +void +ipsecdoi_replace(struct state *st, unsigned long try) +{ + int whack_sock = dup_any(st->st_whack_sock); + lset_t policy = st->st_policy; + + if (IS_PHASE1(st->st_state)) + { + passert(!HAS_IPSEC_POLICY(policy)); + (void) main_outI1(whack_sock, st->st_connection, st, policy, try); + } + else + { + /* Add features of actual old state to policy. This ensures + * that rekeying doesn't downgrade security. I admit that + * this doesn't capture everything. + */ + if (st->st_pfs_group != NULL) + policy |= POLICY_PFS; + if (st->st_ah.present) + { + policy |= POLICY_AUTHENTICATE; + if (st->st_ah.attrs.encapsulation == ENCAPSULATION_MODE_TUNNEL) + policy |= POLICY_TUNNEL; + } + if (st->st_esp.present && st->st_esp.attrs.transid != ESP_NULL) + { + policy |= POLICY_ENCRYPT; + if (st->st_esp.attrs.encapsulation == ENCAPSULATION_MODE_TUNNEL) + policy |= POLICY_TUNNEL; + } + if (st->st_ipcomp.present) + { + policy |= POLICY_COMPRESS; + if (st->st_ipcomp.attrs.encapsulation == ENCAPSULATION_MODE_TUNNEL) + policy |= POLICY_TUNNEL; + } + passert(HAS_IPSEC_POLICY(policy)); + ipsecdoi_initiate(whack_sock, st->st_connection, policy, try + , st->st_serialno); + } +} + +/* SKEYID for preshared keys. + * See draft-ietf-ipsec-ike-01.txt 4.1 + */ +static bool +skeyid_preshared(struct state *st) +{ + const chunk_t *pss = get_preshared_secret(st->st_connection); + + if (pss == NULL) + { + loglog(RC_LOG_SERIOUS, "preshared secret disappeared!"); + return FALSE; + } + else + { + struct hmac_ctx ctx; + + hmac_init_chunk(&ctx, st->st_oakley.hasher, *pss); + hmac_update_chunk(&ctx, st->st_ni); + hmac_update_chunk(&ctx, st->st_nr); + hmac_final_chunk(st->st_skeyid, "st_skeyid in skeyid_preshared()", &ctx); + return TRUE; + } +} + +static bool +skeyid_digisig(struct state *st) +{ + struct hmac_ctx ctx; + chunk_t nir; + + /* We need to hmac_init with the concatenation of Ni_b and Nr_b, + * so we have to build a temporary concatentation. + */ + nir.len = st->st_ni.len + st->st_nr.len; + nir.ptr = alloc_bytes(nir.len, "Ni + Nr in skeyid_digisig"); + memcpy(nir.ptr, st->st_ni.ptr, st->st_ni.len); + memcpy(nir.ptr+st->st_ni.len, st->st_nr.ptr, st->st_nr.len); + hmac_init_chunk(&ctx, st->st_oakley.hasher, nir); + pfree(nir.ptr); + + hmac_update_chunk(&ctx, st->st_shared); + hmac_final_chunk(st->st_skeyid, "st_skeyid in skeyid_digisig()", &ctx); + return TRUE; +} + +/* Generate the SKEYID_* and new IV + * See draft-ietf-ipsec-ike-01.txt 4.1 + */ +static bool +generate_skeyids_iv(struct state *st) +{ + /* Generate the SKEYID */ + switch (st->st_oakley.auth) + { + case OAKLEY_PRESHARED_KEY: + case XAUTHInitPreShared: + case XAUTHRespPreShared: + if (!skeyid_preshared(st)) + return FALSE; + break; + + case OAKLEY_RSA_SIG: + case XAUTHInitRSA: + case XAUTHRespRSA: + if (!skeyid_digisig(st)) + return FALSE; + break; + + case OAKLEY_DSS_SIG: + /* XXX */ + + case OAKLEY_RSA_ENC: + case OAKLEY_RSA_ENC_REV: + case OAKLEY_ELGAMAL_ENC: + case OAKLEY_ELGAMAL_ENC_REV: + /* XXX */ + + default: + bad_case(st->st_oakley.auth); + } + + /* generate SKEYID_* from SKEYID */ + { + struct hmac_ctx ctx; + + hmac_init_chunk(&ctx, st->st_oakley.hasher, st->st_skeyid); + + /* SKEYID_D */ + hmac_update_chunk(&ctx, st->st_shared); + hmac_update(&ctx, st->st_icookie, COOKIE_SIZE); + hmac_update(&ctx, st->st_rcookie, COOKIE_SIZE); + hmac_update(&ctx, "\0", 1); + hmac_final_chunk(st->st_skeyid_d, "st_skeyid_d in generate_skeyids_iv()", &ctx); + + /* SKEYID_A */ + hmac_reinit(&ctx); + hmac_update_chunk(&ctx, st->st_skeyid_d); + hmac_update_chunk(&ctx, st->st_shared); + hmac_update(&ctx, st->st_icookie, COOKIE_SIZE); + hmac_update(&ctx, st->st_rcookie, COOKIE_SIZE); + hmac_update(&ctx, "\1", 1); + hmac_final_chunk(st->st_skeyid_a, "st_skeyid_a in generate_skeyids_iv()", &ctx); + + /* SKEYID_E */ + hmac_reinit(&ctx); + hmac_update_chunk(&ctx, st->st_skeyid_a); + hmac_update_chunk(&ctx, st->st_shared); + hmac_update(&ctx, st->st_icookie, COOKIE_SIZE); + hmac_update(&ctx, st->st_rcookie, COOKIE_SIZE); + hmac_update(&ctx, "\2", 1); + hmac_final_chunk(st->st_skeyid_e, "st_skeyid_e in generate_skeyids_iv()", &ctx); + } + + /* generate IV */ + { + union hash_ctx hash_ctx; + const struct hash_desc *h = st->st_oakley.hasher; + + st->st_new_iv_len = h->hash_digest_size; + passert(st->st_new_iv_len <= sizeof(st->st_new_iv)); + + DBG(DBG_CRYPT, + DBG_dump_chunk("DH_i:", st->st_gi); + DBG_dump_chunk("DH_r:", st->st_gr); + ); + h->hash_init(&hash_ctx); + h->hash_update(&hash_ctx, st->st_gi.ptr, st->st_gi.len); + h->hash_update(&hash_ctx, st->st_gr.ptr, st->st_gr.len); + h->hash_final(st->st_new_iv, &hash_ctx); + } + + /* Oakley Keying Material + * Derived from Skeyid_e: if it is not big enough, generate more + * using the PRF. + * See RFC 2409 "IKE" Appendix B + */ + { + /* const size_t keysize = st->st_oakley.encrypter->keydeflen/BITS_PER_BYTE; */ + const size_t keysize = st->st_oakley.enckeylen/BITS_PER_BYTE; + u_char keytemp[MAX_OAKLEY_KEY_LEN + MAX_DIGEST_LEN]; + u_char *k = st->st_skeyid_e.ptr; + + if (keysize > st->st_skeyid_e.len) + { + struct hmac_ctx ctx; + size_t i = 0; + + hmac_init_chunk(&ctx, st->st_oakley.hasher, st->st_skeyid_e); + hmac_update(&ctx, "\0", 1); + for (;;) + { + hmac_final(&keytemp[i], &ctx); + i += ctx.hmac_digest_size; + if (i >= keysize) + break; + hmac_reinit(&ctx); + hmac_update(&ctx, &keytemp[i - ctx.hmac_digest_size], ctx.hmac_digest_size); + } + k = keytemp; + } + clonereplacechunk(st->st_enc_key, k, keysize, "st_enc_key"); + } + + DBG(DBG_CRYPT, + DBG_dump_chunk("Skeyid: ", st->st_skeyid); + DBG_dump_chunk("Skeyid_d:", st->st_skeyid_d); + DBG_dump_chunk("Skeyid_a:", st->st_skeyid_a); + DBG_dump_chunk("Skeyid_e:", st->st_skeyid_e); + DBG_dump_chunk("enc key:", st->st_enc_key); + DBG_dump("IV:", st->st_new_iv, st->st_new_iv_len)); + return TRUE; +} + +/* Generate HASH_I or HASH_R for ISAKMP Phase I. + * This will *not* generate other hash payloads (eg. Phase II or Quick Mode, + * New Group Mode, or ISAKMP Informational Exchanges). + * If the hashi argument is TRUE, generate HASH_I; if FALSE generate HASH_R. + * If hashus argument is TRUE, we're generating a hash for our end. + * See RFC2409 IKE 5. + * + * Generating the SIG_I and SIG_R for DSS is an odd perversion of this: + * Most of the logic is the same, but SHA-1 is used in place of HMAC-whatever. + * The extensive common logic is embodied in main_mode_hash_body(). + * See draft-ietf-ipsec-ike-01.txt 4.1 and 6.1.1.2 + */ + +typedef void (*hash_update_t)(union hash_ctx *, const u_char *, size_t) ; +static void +main_mode_hash_body(struct state *st +, bool hashi /* Initiator? */ +, const pb_stream *idpl /* ID payload, as PBS */ +, union hash_ctx *ctx +, void (*hash_update_void)(void *, const u_char *input, size_t)) +{ +#define HASH_UPDATE_T (union hash_ctx *, const u_char *input, unsigned int len) + hash_update_t hash_update=(hash_update_t) hash_update_void; +#if 0 /* if desperate to debug hashing */ +# define hash_update(ctx, input, len) { \ + DBG_dump("hash input", input, len); \ + (hash_update)(ctx, input, len); \ + } +#endif + +# define hash_update_chunk(ctx, ch) hash_update((ctx), (ch).ptr, (ch).len) + + if (hashi) + { + hash_update_chunk(ctx, st->st_gi); + hash_update_chunk(ctx, st->st_gr); + hash_update(ctx, st->st_icookie, COOKIE_SIZE); + hash_update(ctx, st->st_rcookie, COOKIE_SIZE); + } + else + { + hash_update_chunk(ctx, st->st_gr); + hash_update_chunk(ctx, st->st_gi); + hash_update(ctx, st->st_rcookie, COOKIE_SIZE); + hash_update(ctx, st->st_icookie, COOKIE_SIZE); + } + + DBG(DBG_CRYPT, DBG_log("hashing %lu bytes of SA" + , (unsigned long) (st->st_p1isa.len - sizeof(struct isakmp_generic)))); + + /* SA_b */ + hash_update(ctx, st->st_p1isa.ptr + sizeof(struct isakmp_generic) + , st->st_p1isa.len - sizeof(struct isakmp_generic)); + + /* Hash identification payload, without generic payload header. + * We used to reconstruct ID Payload for this purpose, but now + * we use the bytes as they appear on the wire to avoid + * "spelling problems". + */ + hash_update(ctx + , idpl->start + sizeof(struct isakmp_generic) + , pbs_offset(idpl) - sizeof(struct isakmp_generic)); + +# undef hash_update_chunk +# undef hash_update +} + +static size_t /* length of hash */ +main_mode_hash(struct state *st +, u_char *hash_val /* resulting bytes */ +, bool hashi /* Initiator? */ +, const pb_stream *idpl) /* ID payload, as PBS; cur must be at end */ +{ + struct hmac_ctx ctx; + + hmac_init_chunk(&ctx, st->st_oakley.hasher, st->st_skeyid); + main_mode_hash_body(st, hashi, idpl, &ctx.hash_ctx, ctx.h->hash_update); + hmac_final(hash_val, &ctx); + return ctx.hmac_digest_size; +} + +#if 0 /* only needed for DSS */ +static void +main_mode_sha1(struct state *st +, u_char *hash_val /* resulting bytes */ +, size_t *hash_len /* length of hash */ +, bool hashi /* Initiator? */ +, const pb_stream *idpl) /* ID payload, as PBS */ +{ + union hash_ctx ctx; + + SHA1Init(&ctx.ctx_sha1); + SHA1Update(&ctx.ctx_sha1, st->st_skeyid.ptr, st->st_skeyid.len); + *hash_len = SHA1_DIGEST_SIZE; + main_mode_hash_body(st, hashi, idpl, &ctx + , (void (*)(union hash_ctx *, const u_char *, unsigned int))&SHA1Update); + SHA1Final(hash_val, &ctx.ctx_sha1); +} +#endif + +/* Create an RSA signature of a hash. + * Poorly specified in draft-ietf-ipsec-ike-01.txt 6.1.1.2. + * Use PKCS#1 version 1.5 encryption of hash (called + * RSAES-PKCS1-V1_5) in PKCS#2. + */ +static size_t +RSA_sign_hash(struct connection *c +, u_char sig_val[RSA_MAX_OCTETS] +, const u_char *hash_val, size_t hash_len) +{ + size_t sz = 0; + smartcard_t *sc = c->spd.this.sc; + + if (sc == NULL) /* no smartcard */ + { + const struct RSA_private_key *k = get_RSA_private_key(c); + + if (k == NULL) + return 0; /* failure: no key to use */ + + sz = k->pub.k; + passert(RSA_MIN_OCTETS <= sz && 4 + hash_len < sz && sz <= RSA_MAX_OCTETS); + sign_hash(k, hash_val, hash_len, sig_val, sz); + } + else if (sc->valid) /* if valid pin then sign hash on the smartcard */ + { + lock_certs_and_keys("RSA_sign_hash"); + if (!scx_establish_context(sc) || !scx_login(sc)) + { + scx_release_context(sc); + unlock_certs_and_keys("RSA_sign_hash"); + return 0; + } + + sz = scx_get_keylength(sc); + if (sz == 0) + { + plog("failed to get keylength from smartcard"); + scx_release_context(sc); + unlock_certs_and_keys("RSA_sign_hash"); + return 0; + } + + DBG(DBG_CONTROL | DBG_CRYPT, + DBG_log("signing hash with RSA key from smartcard (slot: %d, id: %s)" + , (int)sc->slot, sc->id) + ) + sz = scx_sign_hash(sc, hash_val, hash_len, sig_val, sz) ? sz : 0; + if (!pkcs11_keep_state) + scx_release_context(sc); + unlock_certs_and_keys("RSA_sign_hash"); + } + return sz; +} + +/* Check a Main Mode RSA Signature against computed hash using RSA public key k. + * + * As a side effect, on success, the public key is copied into the + * state object to record the authenticator. + * + * Can fail because wrong public key is used or because hash disagrees. + * We distinguish because diagnostics should also. + * + * The result is NULL if the Signature checked out. + * Otherwise, the first character of the result indicates + * how far along failure occurred. A greater character signifies + * greater progress. + * + * Classes: + * 0 reserved for caller + * 1 SIG length doesn't match key length -- wrong key + * 2-8 malformed ECB after decryption -- probably wrong key + * 9 decrypted hash != computed hash -- probably correct key + * + * Although the math should be the same for generating and checking signatures, + * it is not: the knowledge of the private key allows more efficient (i.e. + * different) computation for encryption. + */ +static err_t +try_RSA_signature(const u_char hash_val[MAX_DIGEST_LEN], size_t hash_len +, const pb_stream *sig_pbs, pubkey_t *kr +, struct state *st) +{ + const u_char *sig_val = sig_pbs->cur; + size_t sig_len = pbs_left(sig_pbs); + u_char s[RSA_MAX_OCTETS]; /* for decrypted sig_val */ + u_char *hash_in_s = &s[sig_len - hash_len]; + const struct RSA_public_key *k = &kr->u.rsa; + + /* decrypt the signature -- reversing RSA_sign_hash */ + if (sig_len != k->k) + { + /* XXX notification: INVALID_KEY_INFORMATION */ + return "1" "SIG length does not match public key length"; + } + + /* actual exponentiation; see PKCS#1 v2.0 5.1 */ + { + chunk_t temp_s; + mpz_t c; + + n_to_mpz(c, sig_val, sig_len); + mpz_powm(c, c, &k->e, &k->n); + + temp_s = mpz_to_n(c, sig_len); /* back to octets */ + memcpy(s, temp_s.ptr, sig_len); + pfree(temp_s.ptr); + mpz_clear(c); + } + + /* sanity check on signature: see if it matches + * PKCS#1 v1.5 8.1 encryption-block formatting + */ + { + err_t ugh = NULL; + + if (s[0] != 0x00) + ugh = "2" "no leading 00"; + else if (hash_in_s[-1] != 0x00) + ugh = "3" "00 separator not present"; + else if (s[1] == 0x01) + { + const u_char *p; + + for (p = &s[2]; p != hash_in_s - 1; p++) + { + if (*p != 0xFF) + { + ugh = "4" "invalid Padding String"; + break; + } + } + } + else if (s[1] == 0x02) + { + const u_char *p; + + for (p = &s[2]; p != hash_in_s - 1; p++) + { + if (*p == 0x00) + { + ugh = "5" "invalid Padding String"; + break; + } + } + } + else + ugh = "6" "Block Type not 01 or 02"; + + if (ugh != NULL) + { + /* note: it might be a good idea to make sure that + * an observer cannot tell what kind of failure happened. + * I don't know what this means in practice. + */ + /* We probably selected the wrong public key for peer: + * SIG Payload decrypted into malformed ECB + */ + /* XXX notification: INVALID_KEY_INFORMATION */ + return ugh; + } + } + + /* We have the decoded hash: see if it matches. */ + if (memcmp(hash_val, hash_in_s, hash_len) != 0) + { + /* good: header, hash, signature, and other payloads well-formed + * good: we could find an RSA Sig key for the peer. + * bad: hash doesn't match + * Guess: sides disagree about key to be used. + */ + DBG_cond_dump(DBG_CRYPT, "decrypted SIG", s, sig_len); + DBG_cond_dump(DBG_CRYPT, "computed HASH", hash_val, hash_len); + /* XXX notification: INVALID_HASH_INFORMATION */ + return "9" "authentication failure: received SIG does not match computed HASH, but message is well-formed"; + } + + /* Success: copy successful key into state. + * There might be an old one if we previously aborted this + * state transition. + */ + unreference_key(&st->st_peer_pubkey); + st->st_peer_pubkey = reference_key(kr); + + return NULL; /* happy happy */ +} + +/* Check signature against all RSA public keys we can find. + * If we need keys from DNS KEY records, and they haven't been fetched, + * return STF_SUSPEND to ask for asynch DNS lookup. + * + * Note: parameter keys_from_dns contains results of DNS lookup for key + * or is NULL indicating lookup not yet tried. + * + * take_a_crack is a helper function. Mostly forensic. + * If only we had coroutines. + */ +struct tac_state { + /* RSA_check_signature's args that take_a_crack needs */ + struct state *st; + const u_char *hash_val; + size_t hash_len; + const pb_stream *sig_pbs; + + /* state carried between calls */ + err_t best_ugh; /* most successful failure */ + int tried_cnt; /* number of keys tried */ + char tried[50]; /* keyids of tried public keys */ + char *tn; /* roof of tried[] */ +}; + +static bool +take_a_crack(struct tac_state *s +, pubkey_t *kr +, const char *story USED_BY_DEBUG) +{ + err_t ugh = try_RSA_signature(s->hash_val, s->hash_len, s->sig_pbs + , kr, s->st); + const struct RSA_public_key *k = &kr->u.rsa; + + s->tried_cnt++; + if (ugh == NULL) + { + DBG(DBG_CRYPT | DBG_CONTROL + , DBG_log("an RSA Sig check passed with *%s [%s]" + , k->keyid, story)); + return TRUE; + } + else + { + DBG(DBG_CRYPT + , DBG_log("an RSA Sig check failure %s with *%s [%s]" + , ugh + 1, k->keyid, story)); + if (s->best_ugh == NULL || s->best_ugh[0] < ugh[0]) + s->best_ugh = ugh; + if (ugh[0] > '0' + && s->tn - s->tried + KEYID_BUF + 2 < (ptrdiff_t)sizeof(s->tried)) + { + strcpy(s->tn, " *"); + strcpy(s->tn + 2, k->keyid); + s->tn += strlen(s->tn); + } + return FALSE; + } +} + +static stf_status +RSA_check_signature(const struct id* peer +, struct state *st +, const u_char hash_val[MAX_DIGEST_LEN] +, size_t hash_len +, const pb_stream *sig_pbs +#ifdef USE_KEYRR +, const pubkey_list_t *keys_from_dns +#endif /* USE_KEYRR */ +, const struct gw_info *gateways_from_dns +) +{ + const struct connection *c = st->st_connection; + struct tac_state s; + err_t dns_ugh = NULL; + + s.st = st; + s.hash_val = hash_val; + s.hash_len = hash_len; + s.sig_pbs = sig_pbs; + + s.best_ugh = NULL; + s.tried_cnt = 0; + s.tn = s.tried; + + /* try all gateway records hung off c */ + if (c->policy & POLICY_OPPO) + { + struct gw_info *gw; + + for (gw = c->gw_info; gw != NULL; gw = gw->next) + { + /* only consider entries that have a key and are for our peer */ + if (gw->gw_key_present + && same_id(&gw->gw_id, &c->spd.that.id) + && take_a_crack(&s, gw->key, "key saved from DNS TXT")) + return STF_OK; + } + } + + /* try all appropriate Public keys */ + { + pubkey_list_t *p, **pp; + + pp = &pubkeys; + + for (p = pubkeys; p != NULL; p = *pp) + { + pubkey_t *key = p->key; + + if (key->alg == PUBKEY_ALG_RSA && same_id(peer, &key->id)) + { + time_t now = time(NULL); + + /* check if found public key has expired */ + if (key->until_time != UNDEFINED_TIME && key->until_time < now) + { + loglog(RC_LOG_SERIOUS, + "cached RSA public key has expired and has been deleted"); + *pp = free_public_keyentry(p); + continue; /* continue with next public key */ + } + + if (take_a_crack(&s, key, "preloaded key")) + return STF_OK; + } + pp = &p->next; + } + } + + /* if no key was found (evidenced by best_ugh == NULL) + * and that side of connection is key_from_DNS_on_demand + * then go search DNS for keys for peer. + */ + if (s.best_ugh == NULL && c->spd.that.key_from_DNS_on_demand) + { + if (gateways_from_dns != NULL) + { + /* TXT keys */ + const struct gw_info *gwp; + + for (gwp = gateways_from_dns; gwp != NULL; gwp = gwp->next) + if (gwp->gw_key_present + && take_a_crack(&s, gwp->key, "key from DNS TXT")) + return STF_OK; + } +#ifdef USE_KEYRR + else if (keys_from_dns != NULL) + { + /* KEY keys */ + const pubkey_list_t *kr; + + for (kr = keys_from_dns; kr != NULL; kr = kr->next) + if (kr->key->alg == PUBKEY_ALG_RSA + && take_a_crack(&s, kr->key, "key from DNS KEY")) + return STF_OK; + } +#endif /* USE_KEYRR */ + else + { + /* nothing yet: ask for asynch DNS lookup */ + return STF_SUSPEND; + } + } + + /* no acceptable key was found: diagnose */ + { + char id_buf[BUF_LEN]; /* arbitrary limit on length of ID reported */ + + (void) idtoa(peer, id_buf, sizeof(id_buf)); + + if (s.best_ugh == NULL) + { + if (dns_ugh == NULL) + loglog(RC_LOG_SERIOUS, "no RSA public key known for '%s'" + , id_buf); + else + loglog(RC_LOG_SERIOUS, "no RSA public key known for '%s'" + "; DNS search for KEY failed (%s)" + , id_buf, dns_ugh); + + /* ??? is this the best code there is? */ + return STF_FAIL + INVALID_KEY_INFORMATION; + } + + if (s.best_ugh[0] == '9') + { + loglog(RC_LOG_SERIOUS, "%s", s.best_ugh + 1); + /* XXX Could send notification back */ + return STF_FAIL + INVALID_HASH_INFORMATION; + } + else + { + if (s.tried_cnt == 1) + { + loglog(RC_LOG_SERIOUS + , "Signature check (on %s) failed (wrong key?); tried%s" + , id_buf, s.tried); + DBG(DBG_CONTROL, + DBG_log("public key for %s failed:" + " decrypted SIG payload into a malformed ECB (%s)" + , id_buf, s.best_ugh + 1)); + } + else + { + loglog(RC_LOG_SERIOUS + , "Signature check (on %s) failed:" + " tried%s keys but none worked." + , id_buf, s.tried); + DBG(DBG_CONTROL, + DBG_log("all %d public keys for %s failed:" + " best decrypted SIG payload into a malformed ECB (%s)" + , s.tried_cnt, id_buf, s.best_ugh + 1)); + } + return STF_FAIL + INVALID_KEY_INFORMATION; + } + } +} + +static notification_t +accept_nonce(struct msg_digest *md, chunk_t *dest, const char *name) +{ + pb_stream *nonce_pbs = &md->chain[ISAKMP_NEXT_NONCE]->pbs; + size_t len = pbs_left(nonce_pbs); + + if (len < MINIMUM_NONCE_SIZE || MAXIMUM_NONCE_SIZE < len) + { + loglog(RC_LOG_SERIOUS, "%s length not between %d and %d" + , name , MINIMUM_NONCE_SIZE, MAXIMUM_NONCE_SIZE); + return PAYLOAD_MALFORMED; /* ??? */ + } + clonereplacechunk(*dest, nonce_pbs->cur, len, "nonce"); + return NOTHING_WRONG; +} + +/* encrypt message, sans fixed part of header + * IV is fetched from st->st_new_iv and stored into st->st_iv. + * The theory is that there will be no "backing out", so we commit to IV. + * We also close the pbs. + */ +bool +encrypt_message(pb_stream *pbs, struct state *st) +{ + const struct encrypt_desc *e = st->st_oakley.encrypter; + u_int8_t *enc_start = pbs->start + sizeof(struct isakmp_hdr); + size_t enc_len = pbs_offset(pbs) - sizeof(struct isakmp_hdr); + + DBG_cond_dump(DBG_CRYPT | DBG_RAW, "encrypting:\n", enc_start, enc_len); + + /* Pad up to multiple of encryption blocksize. + * See the description associated with the definition of + * struct isakmp_hdr in packet.h. + */ + { + size_t padding = pad_up(enc_len, e->enc_blocksize); + + if (padding != 0) + { + if (!out_zero(padding, pbs, "encryption padding")) + return FALSE; + enc_len += padding; + } + } + + DBG(DBG_CRYPT, DBG_log("encrypting using %s", enum_show(&oakley_enc_names, st->st_oakley.encrypt))); + + /* e->crypt(TRUE, enc_start, enc_len, st); */ + crypto_cbc_encrypt(e, TRUE, enc_start, enc_len, st); + + update_iv(st); + DBG_cond_dump(DBG_CRYPT, "next IV:", st->st_iv, st->st_iv_len); + close_message(pbs); + return TRUE; +} + +/* Compute HASH(1), HASH(2) of Quick Mode. + * HASH(1) is part of Quick I1 message. + * HASH(2) is part of Quick R1 message. + * Used by: quick_outI1, quick_inI1_outR1 (twice), quick_inR1_outI2 + * (see RFC 2409 "IKE" 5.5, pg. 18 or draft-ietf-ipsec-ike-01.txt 6.2 pg 25) + */ +static size_t +quick_mode_hash12(u_char *dest, const u_char *start, const u_char *roof +, const struct state *st, const msgid_t *msgid, bool hash2) +{ + struct hmac_ctx ctx; + +#if 0 /* if desperate to debug hashing */ +# define hmac_update(ctx, ptr, len) { \ + DBG_dump("hash input", (ptr), (len)); \ + (hmac_update)((ctx), (ptr), (len)); \ + } + DBG_dump("hash key", st->st_skeyid_a.ptr, st->st_skeyid_a.len); +#endif + hmac_init_chunk(&ctx, st->st_oakley.hasher, st->st_skeyid_a); + hmac_update(&ctx, (const void *) msgid, sizeof(msgid_t)); + if (hash2) + hmac_update_chunk(&ctx, st->st_ni); /* include Ni_b in the hash */ + hmac_update(&ctx, start, roof-start); + hmac_final(dest, &ctx); + + DBG(DBG_CRYPT, + DBG_log("HASH(%d) computed:", hash2 + 1); + DBG_dump("", dest, ctx.hmac_digest_size)); + return ctx.hmac_digest_size; +# undef hmac_update +} + +/* Compute HASH(3) in Quick Mode (part of Quick I2 message). + * Used by: quick_inR1_outI2, quick_inI2 + * See RFC2409 "The Internet Key Exchange (IKE)" 5.5. + * NOTE: this hash (unlike HASH(1) and HASH(2)) ONLY covers the + * Message ID and Nonces. This is a mistake. + */ +static size_t +quick_mode_hash3(u_char *dest, struct state *st) +{ + struct hmac_ctx ctx; + + hmac_init_chunk(&ctx, st->st_oakley.hasher, st->st_skeyid_a); + hmac_update(&ctx, "\0", 1); + hmac_update(&ctx, (u_char *) &st->st_msgid, sizeof(st->st_msgid)); + hmac_update_chunk(&ctx, st->st_ni); + hmac_update_chunk(&ctx, st->st_nr); + hmac_final(dest, &ctx); + DBG_cond_dump(DBG_CRYPT, "HASH(3) computed:", dest, ctx.hmac_digest_size); + return ctx.hmac_digest_size; +} + +/* Compute Phase 2 IV. + * Uses Phase 1 IV from st_iv; puts result in st_new_iv. + */ +void +init_phase2_iv(struct state *st, const msgid_t *msgid) +{ + const struct hash_desc *h = st->st_oakley.hasher; + union hash_ctx ctx; + + DBG_cond_dump(DBG_CRYPT, "last Phase 1 IV:" + , st->st_ph1_iv, st->st_ph1_iv_len); + + st->st_new_iv_len = h->hash_digest_size; + passert(st->st_new_iv_len <= sizeof(st->st_new_iv)); + + h->hash_init(&ctx); + h->hash_update(&ctx, st->st_ph1_iv, st->st_ph1_iv_len); + passert(*msgid != 0); + h->hash_update(&ctx, (const u_char *)msgid, sizeof(*msgid)); + h->hash_final(st->st_new_iv, &ctx); + + DBG_cond_dump(DBG_CRYPT, "computed Phase 2 IV:" + , st->st_new_iv, st->st_new_iv_len); +} + +/* Initiate quick mode. + * --> HDR*, HASH(1), SA, Nr [, KE ] [, IDci, IDcr ] + * (see RFC 2409 "IKE" 5.5) + * Note: this is not called from demux.c + */ + +static bool +emit_subnet_id(ip_subnet *net +, u_int8_t np, u_int8_t protoid, u_int16_t port, pb_stream *outs) +{ + struct isakmp_ipsec_id id; + pb_stream id_pbs; + ip_address ta; + const unsigned char *tbp; + size_t tal; + + id.isaiid_np = np; + id.isaiid_idtype = subnetishost(net) + ? aftoinfo(subnettypeof(net))->id_addr + : aftoinfo(subnettypeof(net))->id_subnet; + id.isaiid_protoid = protoid; + id.isaiid_port = port; + + if (!out_struct(&id, &isakmp_ipsec_identification_desc, outs, &id_pbs)) + return FALSE; + + networkof(net, &ta); + tal = addrbytesptr(&ta, &tbp); + if (!out_raw(tbp, tal, &id_pbs, "client network")) + return FALSE; + + if (!subnetishost(net)) + { + maskof(net, &ta); + tal = addrbytesptr(&ta, &tbp); + if (!out_raw(tbp, tal, &id_pbs, "client mask")) + return FALSE; + } + + close_output_pbs(&id_pbs); + return TRUE; +} + +stf_status +quick_outI1(int whack_sock +, struct state *isakmp_sa +, struct connection *c +, lset_t policy +, unsigned long try +, so_serial_t replacing) +{ + struct state *st = duplicate_state(isakmp_sa); + pb_stream reply; /* not really a reply */ + pb_stream rbody; + u_char /* set by START_HASH_PAYLOAD: */ + *r_hashval, /* where in reply to jam hash value */ + *r_hash_start; /* start of what is to be hashed */ + bool has_client = c->spd.this.has_client || c->spd.that.has_client || + c->spd.this.protocol || c->spd.that.protocol || + c->spd.this.port || c->spd.that.port; + + bool send_natoa = FALSE; + u_int8_t np = ISAKMP_NEXT_NONE; + + st->st_whack_sock = whack_sock; + st->st_connection = c; + set_cur_state(st); /* we must reset before exit */ + st->st_policy = policy; + st->st_try = try; + + st->st_myuserprotoid = c->spd.this.protocol; + st->st_peeruserprotoid = c->spd.that.protocol; + st->st_myuserport = c->spd.this.port; + st->st_peeruserport = c->spd.that.port; + + st->st_msgid = generate_msgid(isakmp_sa); + st->st_state = STATE_QUICK_I1; + + insert_state(st); /* needs cookies, connection, and msgid */ + + if (replacing == SOS_NOBODY) + plog("initiating Quick Mode %s {using isakmp#%lu}" + , prettypolicy(policy) + , isakmp_sa->st_serialno); + else + plog("initiating Quick Mode %s to replace #%lu {using isakmp#%lu}" + , prettypolicy(policy) + , replacing + , isakmp_sa->st_serialno); + + if (isakmp_sa->nat_traversal & NAT_T_DETECTED) + { + /* Duplicate nat_traversal status in new state */ + st->nat_traversal = isakmp_sa->nat_traversal; + + if (isakmp_sa->nat_traversal & LELEM(NAT_TRAVERSAL_NAT_BHND_ME)) + has_client = TRUE; + + nat_traversal_change_port_lookup(NULL, st); + } + else + st->nat_traversal = 0; + + /* are we going to send a NAT-OA payload? */ + if ((st->nat_traversal & NAT_T_WITH_NATOA) + && !(st->st_policy & POLICY_TUNNEL) + && (st->nat_traversal & LELEM(NAT_TRAVERSAL_NAT_BHND_ME))) + { + send_natoa = TRUE; + np = (st->nat_traversal & NAT_T_WITH_RFC_VALUES) ? + ISAKMP_NEXT_NATOA_RFC : ISAKMP_NEXT_NATOA_DRAFTS; + } + + /* set up reply */ + init_pbs(&reply, reply_buffer, sizeof(reply_buffer), "reply packet"); + + /* HDR* out */ + { + struct isakmp_hdr hdr; + + hdr.isa_version = ISAKMP_MAJOR_VERSION << ISA_MAJ_SHIFT | ISAKMP_MINOR_VERSION; + hdr.isa_np = ISAKMP_NEXT_HASH; + hdr.isa_xchg = ISAKMP_XCHG_QUICK; + hdr.isa_msgid = st->st_msgid; + hdr.isa_flags = ISAKMP_FLAG_ENCRYPTION; + memcpy(hdr.isa_icookie, st->st_icookie, COOKIE_SIZE); + memcpy(hdr.isa_rcookie, st->st_rcookie, COOKIE_SIZE); + if (!out_struct(&hdr, &isakmp_hdr_desc, &reply, &rbody)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + } + + /* HASH(1) -- create and note space to be filled later */ + START_HASH_PAYLOAD(rbody, ISAKMP_NEXT_SA); + + /* SA out */ + + /* + * See if pfs_group has been specified for this conn, + * if not, fallback to old use-same-as-P1 behaviour + */ +#ifndef NO_IKE_ALG + if (st->st_connection) + st->st_pfs_group = ike_alg_pfsgroup(st->st_connection, policy); + if (!st->st_pfs_group) +#endif + /* If PFS specified, use the same group as during Phase 1: + * since no negotiation is possible, we pick one that is + * very likely supported. + */ + st->st_pfs_group = policy & POLICY_PFS? isakmp_sa->st_oakley.group : NULL; + + /* Emit SA payload based on a subset of the policy bits. + * POLICY_COMPRESS is considered iff we can do IPcomp. + */ + { + lset_t pm = POLICY_ENCRYPT | POLICY_AUTHENTICATE; + + if (can_do_IPcomp) + pm |= POLICY_COMPRESS; + + if (!out_sa(&rbody + , &ipsec_sadb[(st->st_policy & pm) >> POLICY_IPSEC_SHIFT] + , st, FALSE, ISAKMP_NEXT_NONCE)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + } + + /* Ni out */ + if (!build_and_ship_nonce(&st->st_ni, &rbody + , policy & POLICY_PFS? ISAKMP_NEXT_KE : has_client? ISAKMP_NEXT_ID : np + , "Ni")) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + + /* [ KE ] out (for PFS) */ + + if (st->st_pfs_group != NULL) + { + if (!build_and_ship_KE(st, &st->st_gi, st->st_pfs_group + , &rbody, has_client? ISAKMP_NEXT_ID : np)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + } + + /* [ IDci, IDcr ] out */ + if (has_client) + { + /* IDci (we are initiator), then IDcr (peer is responder) */ + if (!emit_subnet_id(&c->spd.this.client + , ISAKMP_NEXT_ID, st->st_myuserprotoid, st->st_myuserport, &rbody) + || !emit_subnet_id(&c->spd.that.client + , np, st->st_peeruserprotoid, st->st_peeruserport, &rbody)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + } + + /* Send NAT-OA if our address is NATed */ + if (send_natoa) + { + if (!nat_traversal_add_natoa(ISAKMP_NEXT_NONE, &rbody, st)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + } + + /* finish computing HASH(1), inserting it in output */ + (void) quick_mode_hash12(r_hashval, r_hash_start, rbody.cur + , st, &st->st_msgid, FALSE); + + /* encrypt message, except for fixed part of header */ + + init_phase2_iv(isakmp_sa, &st->st_msgid); + st->st_new_iv_len = isakmp_sa->st_new_iv_len; + memcpy(st->st_new_iv, isakmp_sa->st_new_iv, st->st_new_iv_len); + + if (!encrypt_message(&rbody, st)) + { + reset_cur_state(); + return STF_INTERNAL_ERROR; + } + + /* save packet, now that we know its size */ + clonetochunk(st->st_tpacket, reply.start, pbs_offset(&reply) + , "reply packet from quick_outI1"); + + /* send the packet */ + + send_packet(st, "quick_outI1"); + + delete_event(st); + event_schedule(EVENT_RETRANSMIT, EVENT_RETRANSMIT_DELAY_0, st); + + if (replacing == SOS_NOBODY) + whack_log(RC_NEW_STATE + STATE_QUICK_I1 + , "%s: initiate" + , enum_name(&state_names, st->st_state)); + else + whack_log(RC_NEW_STATE + STATE_QUICK_I1 + , "%s: initiate to replace #%lu" + , enum_name(&state_names, st->st_state) + , replacing); + reset_cur_state(); + return STF_OK; +} + + +/* + * Decode the CERT payload of Phase 1. + */ +static void +decode_cert(struct msg_digest *md) +{ + struct payload_digest *p; + + for (p = md->chain[ISAKMP_NEXT_CERT]; p != NULL; p = p->next) + { + struct isakmp_cert *const cert = &p->payload.cert; + chunk_t blob; + time_t valid_until; + blob.ptr = p->pbs.cur; + blob.len = pbs_left(&p->pbs); + if (cert->isacert_type == CERT_X509_SIGNATURE) + { + x509cert_t cert = empty_x509cert; + if (parse_x509cert(blob, 0, &cert)) + { + if (verify_x509cert(&cert, strict_crl_policy, &valid_until)) + { + DBG(DBG_PARSING, + DBG_log("Public key validated") + ) + add_x509_public_key(&cert, valid_until, DAL_SIGNED); + } + else + { + plog("X.509 certificate rejected"); + } + free_generalNames(cert.subjectAltName, FALSE); + free_generalNames(cert.crlDistributionPoints, FALSE); + } + else + plog("Syntax error in X.509 certificate"); + } + else if (cert->isacert_type == CERT_PKCS7_WRAPPED_X509) + { + x509cert_t *cert = NULL; + + if (pkcs7_parse_signedData(blob, NULL, &cert, NULL, NULL)) + store_x509certs(&cert, strict_crl_policy); + else + plog("Syntax error in PKCS#7 wrapped X.509 certificates"); + } + else + { + loglog(RC_LOG_SERIOUS, "ignoring %s certificate payload", + enum_show(&cert_type_names, cert->isacert_type)); + DBG_cond_dump_chunk(DBG_PARSING, "CERT:\n", blob); + } + } +} + +/* + * Decode the CR payload of Phase 1. + */ +static void +decode_cr(struct msg_digest *md, struct connection *c) +{ + struct payload_digest *p; + + for (p = md->chain[ISAKMP_NEXT_CR]; p != NULL; p = p->next) + { + struct isakmp_cr *const cr = &p->payload.cr; + chunk_t ca_name; + + ca_name.len = pbs_left(&p->pbs); + ca_name.ptr = (ca_name.len > 0)? p->pbs.cur : NULL; + + DBG_cond_dump_chunk(DBG_PARSING, "CR", ca_name); + + if (cr->isacr_type == CERT_X509_SIGNATURE) + { + char buf[BUF_LEN]; + + if (ca_name.len > 0) + { + generalName_t *gn; + + if (!is_asn1(ca_name)) + continue; + + gn = alloc_thing(generalName_t, "generalName"); + clonetochunk(ca_name, ca_name.ptr,ca_name.len, "ca name"); + gn->kind = GN_DIRECTORY_NAME; + gn->name = ca_name; + gn->next = c->requested_ca; + c->requested_ca = gn; + } + c->got_certrequest = TRUE; + + DBG(DBG_PARSING | DBG_CONTROL, + dntoa_or_null(buf, BUF_LEN, ca_name, "%any"); + DBG_log("requested CA: '%s'", buf); + ) + } + else + loglog(RC_LOG_SERIOUS, "ignoring %s certificate request payload", + enum_show(&cert_type_names, cr->isacr_type)); + } +} + +/* Decode the ID payload of Phase 1 (main_inI3_outR3 and main_inR3) + * Note: we may change connections as a result. + * We must be called before SIG or HASH are decoded since we + * may change the peer's RSA key or ID. + */ +static bool +decode_peer_id(struct msg_digest *md, struct id *peer) +{ + struct state *const st = md->st; + struct payload_digest *const id_pld = md->chain[ISAKMP_NEXT_ID]; + const pb_stream *const id_pbs = &id_pld->pbs; + struct isakmp_id *const id = &id_pld->payload.id; + + /* I think that RFC2407 (IPSEC DOI) 4.6.2 is confused. + * It talks about the protocol ID and Port fields of the ID + * Payload, but they don't exist as such in Phase 1. + * We use more appropriate names. + * isaid_doi_specific_a is in place of Protocol ID. + * isaid_doi_specific_b is in place of Port. + * Besides, there is no good reason for allowing these to be + * other than 0 in Phase 1. + */ + if ((st->nat_traversal & NAT_T_WITH_PORT_FLOATING) + && id->isaid_doi_specific_a == IPPROTO_UDP + && (id->isaid_doi_specific_b == 0 || id->isaid_doi_specific_b == NAT_T_IKE_FLOAT_PORT)) + { + DBG_log("protocol/port in Phase 1 ID Payload is %d/%d. " + "accepted with port_floating NAT-T", + id->isaid_doi_specific_a, id->isaid_doi_specific_b); + } + else if (!(id->isaid_doi_specific_a == 0 && id->isaid_doi_specific_b == 0) + && !(id->isaid_doi_specific_a == IPPROTO_UDP && id->isaid_doi_specific_b == IKE_UDP_PORT)) + { + loglog(RC_LOG_SERIOUS, "protocol/port in Phase 1 ID Payload must be 0/0 or %d/%d" + " but are %d/%d" + , IPPROTO_UDP, IKE_UDP_PORT + , id->isaid_doi_specific_a, id->isaid_doi_specific_b); + return FALSE; + } + + peer->kind = id->isaid_idtype; + + switch (peer->kind) + { + case ID_IPV4_ADDR: + case ID_IPV6_ADDR: + /* failure mode for initaddr is probably inappropriate address length */ + { + err_t ugh = initaddr(id_pbs->cur, pbs_left(id_pbs) + , peer->kind == ID_IPV4_ADDR? AF_INET : AF_INET6 + , &peer->ip_addr); + + if (ugh != NULL) + { + loglog(RC_LOG_SERIOUS, "improper %s identification payload: %s" + , enum_show(&ident_names, peer->kind), ugh); + /* XXX Could send notification back */ + return FALSE; + } + } + break; + + case ID_USER_FQDN: + if (memchr(id_pbs->cur, '@', pbs_left(id_pbs)) == NULL) + { + loglog(RC_LOG_SERIOUS, "peer's ID_USER_FQDN contains no @"); + return FALSE; + } + /* FALLTHROUGH */ + case ID_FQDN: + if (memchr(id_pbs->cur, '\0', pbs_left(id_pbs)) != NULL) + { + loglog(RC_LOG_SERIOUS, "Phase 1 ID Payload of type %s contains a NUL" + , enum_show(&ident_names, peer->kind)); + return FALSE; + } + + /* ??? ought to do some more sanity check, but what? */ + + setchunk(peer->name, id_pbs->cur, pbs_left(id_pbs)); + break; + + case ID_KEY_ID: + setchunk(peer->name, id_pbs->cur, pbs_left(id_pbs)); + DBG(DBG_PARSING, + DBG_dump_chunk("KEY ID:", peer->name)); + break; + + case ID_DER_ASN1_DN: + setchunk(peer->name, id_pbs->cur, pbs_left(id_pbs)); + DBG(DBG_PARSING, + DBG_dump_chunk("DER ASN1 DN:", peer->name)); + break; + + default: + /* XXX Could send notification back */ + loglog(RC_LOG_SERIOUS, "Unacceptable identity type (%s) in Phase 1 ID Payload" + , enum_show(&ident_names, peer->kind)); + return FALSE; + } + + { + char buf[BUF_LEN]; + + idtoa(peer, buf, sizeof(buf)); + plog("Peer ID is %s: '%s'", + enum_show(&ident_names, id->isaid_idtype), buf); + } + + /* check for certificates */ + decode_cert(md); + return TRUE; +} + +/* Now that we've decoded the ID payload, let's see if we + * need to switch connections. + * We must not switch horses if we initiated: + * - if the initiation was explicit, we'd be ignoring user's intent + * - if opportunistic, we'll lose our HOLD info + */ +static bool +switch_connection(struct msg_digest *md, struct id *peer, bool initiator) +{ + struct state *const st = md->st; + struct connection *c = st->st_connection; + + chunk_t peer_ca = (st->st_peer_pubkey != NULL) + ? st->st_peer_pubkey->issuer : empty_chunk; + + DBG(DBG_CONTROL, + char buf[BUF_LEN]; + + dntoa_or_null(buf, BUF_LEN, peer_ca, "%none"); + DBG_log("peer CA: '%s'", buf); + ) + + if (initiator) + { + int pathlen; + + if (!same_id(&c->spd.that.id, peer)) + { + char expect[BUF_LEN] + , found[BUF_LEN]; + + idtoa(&c->spd.that.id, expect, sizeof(expect)); + idtoa(peer, found, sizeof(found)); + loglog(RC_LOG_SERIOUS + , "we require peer to have ID '%s', but peer declares '%s'" + , expect, found); + return FALSE; + } + + DBG(DBG_CONTROL, + char buf[BUF_LEN]; + + dntoa_or_null(buf, BUF_LEN, c->spd.this.ca, "%none"); + DBG_log("required CA: '%s'", buf); + ) + + if (!trusted_ca(peer_ca, c->spd.that.ca, &pathlen)) + { + loglog(RC_LOG_SERIOUS + , "we don't accept the peer's CA"); + return FALSE; + } + } + else + { + struct connection *r; + + /* check for certificate requests */ + decode_cr(md, c); + + r = refine_host_connection(st, peer, peer_ca); + + /* delete the collected certificate requests */ + free_generalNames(c->requested_ca, TRUE); + c->requested_ca = NULL; + + if (r == NULL) + { + char buf[BUF_LEN]; + + idtoa(peer, buf, sizeof(buf)); + loglog(RC_LOG_SERIOUS, "no suitable connection for peer '%s'", buf); + return FALSE; + } + + DBG(DBG_CONTROL, + char buf[BUF_LEN]; + + dntoa_or_null(buf, BUF_LEN, r->spd.this.ca, "%none"); + DBG_log("offered CA: '%s'", buf); + ) + + if (r != c) + { + /* apparently, r is an improvement on c -- replace */ + + DBG(DBG_CONTROL + , DBG_log("switched from \"%s\" to \"%s\"", c->name, r->name)); + if (r->kind == CK_TEMPLATE) + { + /* instantiate it, filling in peer's ID */ + r = rw_instantiate(r, &c->spd.that.host_addr + , c->spd.that.host_port, NULL, peer); + } + + /* copy certificate request info */ + r->got_certrequest = c->got_certrequest; + + st->st_connection = r; /* kill reference to c */ + set_cur_connection(r); + connection_discard(c); + } + else if (c->spd.that.has_id_wildcards) + { + free_id_content(&c->spd.that.id); + c->spd.that.id = *peer; + c->spd.that.has_id_wildcards = FALSE; + unshare_id_content(&c->spd.that.id); + } + } + return TRUE; +} + +/* Decode the variable part of an ID packet (during Quick Mode). + * This is designed for packets that identify clients, not peers. + * Rejects 0.0.0.0/32 or IPv6 equivalent because + * (1) it is wrong and (2) we use this value for inband signalling. + */ +static bool +decode_net_id(struct isakmp_ipsec_id *id +, pb_stream *id_pbs +, ip_subnet *net +, const char *which) +{ + const struct af_info *afi = NULL; + + /* Note: the following may be a pointer into static memory + * that may be recycled, but only if the type is not known. + * That case is disposed of very early -- in the first switch. + */ + const char *idtypename = enum_show(&ident_names, id->isaiid_idtype); + + switch (id->isaiid_idtype) + { + case ID_IPV4_ADDR: + case ID_IPV4_ADDR_SUBNET: + case ID_IPV4_ADDR_RANGE: + afi = &af_inet4_info; + break; + case ID_IPV6_ADDR: + case ID_IPV6_ADDR_SUBNET: + case ID_IPV6_ADDR_RANGE: + afi = &af_inet6_info; + break; + case ID_FQDN: + return TRUE; + default: + /* XXX support more */ + loglog(RC_LOG_SERIOUS, "unsupported ID type %s" + , idtypename); + /* XXX Could send notification back */ + return FALSE; + } + + switch (id->isaiid_idtype) + { + case ID_IPV4_ADDR: + case ID_IPV6_ADDR: + { + ip_address temp_address; + err_t ugh; + + ugh = initaddr(id_pbs->cur, pbs_left(id_pbs), afi->af, &temp_address); + + if (ugh != NULL) + { + loglog(RC_LOG_SERIOUS, "%s ID payload %s has wrong length in Quick I1 (%s)" + , which, idtypename, ugh); + /* XXX Could send notification back */ + return FALSE; + } + if (isanyaddr(&temp_address)) + { + loglog(RC_LOG_SERIOUS, "%s ID payload %s is invalid (%s) in Quick I1" + , which, idtypename, ip_str(&temp_address)); + /* XXX Could send notification back */ + return FALSE; + } + happy(addrtosubnet(&temp_address, net)); + DBG(DBG_PARSING | DBG_CONTROL + , DBG_log("%s is %s", which, ip_str(&temp_address))); + break; + } + + case ID_IPV4_ADDR_SUBNET: + case ID_IPV6_ADDR_SUBNET: + { + ip_address temp_address, temp_mask; + err_t ugh; + + if (pbs_left(id_pbs) != 2 * afi->ia_sz) + { + loglog(RC_LOG_SERIOUS, "%s ID payload %s wrong length in Quick I1" + , which, idtypename); + /* XXX Could send notification back */ + return FALSE; + } + ugh = initaddr(id_pbs->cur + , afi->ia_sz, afi->af, &temp_address); + if (ugh == NULL) + ugh = initaddr(id_pbs->cur + afi->ia_sz + , afi->ia_sz, afi->af, &temp_mask); + if (ugh == NULL) + ugh = initsubnet(&temp_address, masktocount(&temp_mask) + , '0', net); + if (ugh == NULL && subnetisnone(net)) + ugh = "contains only anyaddr"; + if (ugh != NULL) + { + loglog(RC_LOG_SERIOUS, "%s ID payload %s bad subnet in Quick I1 (%s)" + , which, idtypename, ugh); + /* XXX Could send notification back */ + return FALSE; + } + DBG(DBG_PARSING | DBG_CONTROL, + { + char temp_buff[SUBNETTOT_BUF]; + + subnettot(net, 0, temp_buff, sizeof(temp_buff)); + DBG_log("%s is subnet %s", which, temp_buff); + }); + break; + } + + case ID_IPV4_ADDR_RANGE: + case ID_IPV6_ADDR_RANGE: + { + ip_address temp_address_from, temp_address_to; + err_t ugh; + + if (pbs_left(id_pbs) != 2 * afi->ia_sz) + { + loglog(RC_LOG_SERIOUS, "%s ID payload %s wrong length in Quick I1" + , which, idtypename); + /* XXX Could send notification back */ + return FALSE; + } + ugh = initaddr(id_pbs->cur, afi->ia_sz, afi->af, &temp_address_from); + if (ugh == NULL) + ugh = initaddr(id_pbs->cur + afi->ia_sz + , afi->ia_sz, afi->af, &temp_address_to); + if (ugh != NULL) + { + loglog(RC_LOG_SERIOUS, "%s ID payload %s malformed (%s) in Quick I1" + , which, idtypename, ugh); + /* XXX Could send notification back */ + return FALSE; + } + + ugh = rangetosubnet(&temp_address_from, &temp_address_to, net); + if (ugh == NULL && subnetisnone(net)) + ugh = "contains only anyaddr"; + if (ugh != NULL) + { + char temp_buff1[ADDRTOT_BUF], temp_buff2[ADDRTOT_BUF]; + + addrtot(&temp_address_from, 0, temp_buff1, sizeof(temp_buff1)); + addrtot(&temp_address_to, 0, temp_buff2, sizeof(temp_buff2)); + loglog(RC_LOG_SERIOUS, "%s ID payload in Quick I1, %s" + " %s - %s unacceptable: %s" + , which, idtypename, temp_buff1, temp_buff2, ugh); + return FALSE; + } + DBG(DBG_PARSING | DBG_CONTROL, + { + char temp_buff[SUBNETTOT_BUF]; + + subnettot(net, 0, temp_buff, sizeof(temp_buff)); + DBG_log("%s is subnet %s (received as range)" + , which, temp_buff); + }); + break; + } + } + + /* set the port selector */ + setportof(htons(id->isaiid_port), &net->addr); + + DBG(DBG_PARSING | DBG_CONTROL, + DBG_log("%s protocol/port is %d/%d", which, id->isaiid_protoid, id->isaiid_port) + ) + + return TRUE; +} + +/* like decode, but checks that what is received matches what was sent */ +static bool + +check_net_id(struct isakmp_ipsec_id *id +, pb_stream *id_pbs +, u_int8_t *protoid +, u_int16_t *port +, ip_subnet *net +, const char *which) +{ + ip_subnet net_temp; + + if (!decode_net_id(id, id_pbs, &net_temp, which)) + return FALSE; + + if (!samesubnet(net, &net_temp) + || *protoid != id->isaiid_protoid || *port != id->isaiid_port) + { + loglog(RC_LOG_SERIOUS, "%s ID returned doesn't match my proposal", which); + return FALSE; + } + return TRUE; +} + +/* + * look for the existence of a non-expiring preloaded public key + */ +static bool +has_preloaded_public_key(struct state *st) +{ + struct connection *c = st->st_connection; + + /* do not consider rw connections since + * the peer's identity must be known + */ + if (c->kind == CK_PERMANENT) + { + pubkey_list_t *p; + + /* look for a matching RSA public key */ + for (p = pubkeys; p != NULL; p = p->next) + { + pubkey_t *key = p->key; + + if (key->alg == PUBKEY_ALG_RSA && + same_id(&c->spd.that.id, &key->id) && + key->until_time == UNDEFINED_TIME) + { + /* found a preloaded public key */ + return TRUE; + } + } + } + return FALSE; +} + +/* + * Produce the new key material of Quick Mode. + * RFC 2409 "IKE" section 5.5 + * specifies how this is to be done. + */ +static void +compute_proto_keymat(struct state *st +, u_int8_t protoid +, struct ipsec_proto_info *pi) +{ + size_t needed_len; /* bytes of keying material needed */ + + /* Add up the requirements for keying material + * (It probably doesn't matter if we produce too much!) + */ + switch (protoid) + { + case PROTO_IPSEC_ESP: + switch (pi->attrs.transid) + { + case ESP_NULL: + needed_len = 0; + break; + case ESP_DES: + needed_len = DES_CBC_BLOCK_SIZE; + break; + case ESP_3DES: + needed_len = DES_CBC_BLOCK_SIZE * 3; + break; + default: +#ifndef NO_KERNEL_ALG + if((needed_len=kernel_alg_esp_enc_keylen(pi->attrs.transid))>0) { + /* XXX: check key_len "coupling with kernel.c's */ + if (pi->attrs.key_len) { + needed_len=pi->attrs.key_len/8; + DBG(DBG_PARSING, DBG_log("compute_proto_keymat:" + "key_len=%d from peer", + (int)needed_len)); + } + break; + } +#endif + bad_case(pi->attrs.transid); + } + +#ifndef NO_KERNEL_ALG + DBG(DBG_PARSING, DBG_log("compute_proto_keymat:" + "needed_len (after ESP enc)=%d", + (int)needed_len)); + if (kernel_alg_esp_auth_ok(pi->attrs.auth, NULL)) { + needed_len += kernel_alg_esp_auth_keylen(pi->attrs.auth); + } else +#endif + switch (pi->attrs.auth) + { + case AUTH_ALGORITHM_NONE: + break; + case AUTH_ALGORITHM_HMAC_MD5: + needed_len += HMAC_MD5_KEY_LEN; + break; + case AUTH_ALGORITHM_HMAC_SHA1: + needed_len += HMAC_SHA1_KEY_LEN; + break; + case AUTH_ALGORITHM_DES_MAC: + default: + bad_case(pi->attrs.auth); + } + DBG(DBG_PARSING, DBG_log("compute_proto_keymat:" + "needed_len (after ESP auth)=%d", + (int)needed_len)); + break; + + case PROTO_IPSEC_AH: + switch (pi->attrs.transid) + { + case AH_MD5: + needed_len = HMAC_MD5_KEY_LEN; + break; + case AH_SHA: + needed_len = HMAC_SHA1_KEY_LEN; + break; + default: + bad_case(pi->attrs.transid); + } + break; + + default: + bad_case(protoid); + } + + pi->keymat_len = needed_len; + + /* Allocate space for the keying material. + * Although only needed_len bytes are desired, we + * must round up to a multiple of ctx.hmac_digest_size + * so that our buffer isn't overrun. + */ + { + struct hmac_ctx ctx_me, ctx_peer; + size_t needed_space; /* space needed for keying material (rounded up) */ + size_t i; + + hmac_init_chunk(&ctx_me, st->st_oakley.hasher, st->st_skeyid_d); + ctx_peer = ctx_me; /* duplicate initial conditions */ + + needed_space = needed_len + pad_up(needed_len, ctx_me.hmac_digest_size); + replace(pi->our_keymat, alloc_bytes(needed_space, "keymat in compute_keymat()")); + replace(pi->peer_keymat, alloc_bytes(needed_space, "peer_keymat in quick_inI1_outR1()")); + + for (i = 0;; ) + { + if (st->st_shared.ptr != NULL) + { + /* PFS: include the g^xy */ + hmac_update_chunk(&ctx_me, st->st_shared); + hmac_update_chunk(&ctx_peer, st->st_shared); + } + hmac_update(&ctx_me, &protoid, sizeof(protoid)); + hmac_update(&ctx_peer, &protoid, sizeof(protoid)); + + hmac_update(&ctx_me, (u_char *)&pi->our_spi, sizeof(pi->our_spi)); + hmac_update(&ctx_peer, (u_char *)&pi->attrs.spi, sizeof(pi->attrs.spi)); + + hmac_update_chunk(&ctx_me, st->st_ni); + hmac_update_chunk(&ctx_peer, st->st_ni); + + hmac_update_chunk(&ctx_me, st->st_nr); + hmac_update_chunk(&ctx_peer, st->st_nr); + + hmac_final(pi->our_keymat + i, &ctx_me); + hmac_final(pi->peer_keymat + i, &ctx_peer); + + i += ctx_me.hmac_digest_size; + if (i >= needed_space) + break; + + /* more keying material needed: prepare to go around again */ + + hmac_reinit(&ctx_me); + hmac_reinit(&ctx_peer); + + hmac_update(&ctx_me, pi->our_keymat + i - ctx_me.hmac_digest_size + , ctx_me.hmac_digest_size); + hmac_update(&ctx_peer, pi->peer_keymat + i - ctx_peer.hmac_digest_size + , ctx_peer.hmac_digest_size); + } + } + + DBG(DBG_CRYPT, + DBG_dump("KEYMAT computed:\n", pi->our_keymat, pi->keymat_len); + DBG_dump("Peer KEYMAT computed:\n", pi->peer_keymat, pi->keymat_len)); +} + +static void +compute_keymats(struct state *st) +{ + if (st->st_ah.present) + compute_proto_keymat(st, PROTO_IPSEC_AH, &st->st_ah); + if (st->st_esp.present) + compute_proto_keymat(st, PROTO_IPSEC_ESP, &st->st_esp); +} + +/* State Transition Functions. + * + * The definition of state_microcode_table in demux.c is a good + * overview of these routines. + * + * - Called from process_packet; result handled by complete_state_transition + * - struct state_microcode member "processor" points to these + * - these routine definitionss are in state order + * - these routines must be restartable from any point of error return: + * beware of memory allocated before any error. + * - output HDR is usually emitted by process_packet (if state_microcode + * member first_out_payload isn't ISAKMP_NEXT_NONE). + * + * The transition functions' functions include: + * - process and judge payloads + * - update st_iv (result of decryption is in st_new_iv) + * - build reply packet + */ + +/* Handle a Main Mode Oakley first packet (responder side). + * HDR;SA --> HDR;SA + */ +stf_status +main_inI1_outR1(struct msg_digest *md) +{ + struct payload_digest *const sa_pd = md->chain[ISAKMP_NEXT_SA]; + struct state *st; + struct connection *c; + struct isakmp_proposal proposal; + pb_stream proposal_pbs; + pb_stream r_sa_pbs; + u_int32_t ipsecdoisit; + lset_t policy = LEMPTY; + int vids_to_send = 0; + + /* We preparse the peer's proposal in order to determine + * the requested authentication policy (RSA or PSK) + */ + RETURN_STF_FAILURE(preparse_isakmp_sa_body(&sa_pd->payload.sa + , &sa_pd->pbs, &ipsecdoisit, &proposal_pbs, &proposal)); + + backup_pbs(&proposal_pbs); + RETURN_STF_FAILURE(parse_isakmp_policy(&proposal_pbs + , proposal.isap_notrans, &policy)); + restore_pbs(&proposal_pbs); + + /* We are only considering candidate connections that match + * the requested authentication policy (RSA or PSK) + */ + c = find_host_connection(&md->iface->addr, pluto_port + , &md->sender, md->sender_port, policy); + + if (c == NULL && md->iface->ike_float) + { + c = find_host_connection(&md->iface->addr, NAT_T_IKE_FLOAT_PORT + , &md->sender, md->sender_port, policy); + } + + if (c == NULL) + { + /* See if a wildcarded connection can be found. + * We cannot pick the right connection, so we're making a guess. + * All Road Warrior connections are fair game: + * we pick the first we come across (if any). + * If we don't find any, we pick the first opportunistic + * with the smallest subnet that includes the peer. + * There is, of course, no necessary relationship between + * an Initiator's address and that of its client, + * but Food Groups kind of assumes one. + */ + { + struct connection *d; + + d = find_host_connection(&md->iface->addr + , pluto_port, (ip_address*)NULL, md->sender_port, policy); + + for (; d != NULL; d = d->hp_next) + { + if (d->kind == CK_GROUP) + { + /* ignore */ + } + else + { + if (d->kind == CK_TEMPLATE && !(d->policy & POLICY_OPPO)) + { + /* must be Road Warrior: we have a winner */ + c = d; + break; + } + + /* Opportunistic or Shunt: pick tightest match */ + if (addrinsubnet(&md->sender, &d->spd.that.client) + && (c == NULL || !subnetinsubnet(&c->spd.that.client, &d->spd.that.client))) + c = d; + } + } + } + + if (c == NULL) + { + loglog(RC_LOG_SERIOUS, "initial Main Mode message received on %s:%u" + " but no connection has been authorized%s%s" + , ip_str(&md->iface->addr), ntohs(portof(&md->iface->addr)) + , (policy != LEMPTY) ? " with policy=" : "" + , (policy != LEMPTY) ? bitnamesof(sa_policy_bit_names, policy) : ""); + /* XXX notification is in order! */ + return STF_IGNORE; + } + else if (c->kind != CK_TEMPLATE) + { + loglog(RC_LOG_SERIOUS, "initial Main Mode message received on %s:%u" + " but \"%s\" forbids connection" + , ip_str(&md->iface->addr), pluto_port, c->name); + /* XXX notification is in order! */ + return STF_IGNORE; + } + else + { + /* Create a temporary connection that is a copy of this one. + * His ID isn't declared yet. + */ + c = rw_instantiate(c, &md->sender, md->sender_port, NULL, NULL); + } + } + else if (c->kind == CK_TEMPLATE) + { + /* Create an instance + * This is a rare case: wildcard peer ID but static peer IP address + */ + c = rw_instantiate(c, &md->sender, md->sender_port, NULL, &c->spd.that.id); + } + + /* Set up state */ + md->st = st = new_state(); + st->st_connection = c; + set_cur_state(st); /* (caller will reset cur_state) */ + st->st_try = 0; /* not our job to try again from start */ + st->st_policy = c->policy & ~POLICY_IPSEC_MASK; /* only as accurate as connection */ + + memcpy(st->st_icookie, md->hdr.isa_icookie, COOKIE_SIZE); + get_cookie(FALSE, st->st_rcookie, COOKIE_SIZE, &md->sender); + + insert_state(st); /* needs cookies, connection, and msgid (0) */ + + st->st_doi = ISAKMP_DOI_IPSEC; + st->st_situation = SIT_IDENTITY_ONLY; /* We only support this */ + + if ((c->kind == CK_INSTANCE) && (c->spd.that.host_port != pluto_port)) + { + plog("responding to Main Mode from unknown peer %s:%u" + , ip_str(&c->spd.that.host_addr), c->spd.that.host_port); + } + else if (c->kind == CK_INSTANCE) + { + plog("responding to Main Mode from unknown peer %s" + , ip_str(&c->spd.that.host_addr)); + } + else + { + plog("responding to Main Mode"); + } + + /* parse_isakmp_sa also spits out a winning SA into our reply, + * so we have to build our md->reply and emit HDR before calling it. + */ + + /* determine how many Vendor ID payloads we will be sending */ + if (SEND_PLUTO_VID) + vids_to_send++; + if (SEND_CISCO_UNITY_VID) + vids_to_send++; + if (md->openpgp) + vids_to_send++; + /* always send XAUTH Vendor ID */ + vids_to_send++; + /* always send DPD Vendor ID */ + vids_to_send++; + if (md->nat_traversal_vid && nat_traversal_enabled) + vids_to_send++; + + /* HDR out. + * We can't leave this to comm_handle() because we must + * fill in the cookie. + */ + { + struct isakmp_hdr r_hdr = md->hdr; + + r_hdr.isa_flags &= ~ISAKMP_FLAG_COMMIT; /* we won't ever turn on this bit */ + memcpy(r_hdr.isa_rcookie, st->st_rcookie, COOKIE_SIZE); + r_hdr.isa_np = ISAKMP_NEXT_SA; + if (!out_struct(&r_hdr, &isakmp_hdr_desc, &md->reply, &md->rbody)) + return STF_INTERNAL_ERROR; + } + + /* start of SA out */ + { + struct isakmp_sa r_sa = sa_pd->payload.sa; + + r_sa.isasa_np = vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE; + + if (!out_struct(&r_sa, &isakmp_sa_desc, &md->rbody, &r_sa_pbs)) + return STF_INTERNAL_ERROR; + } + + /* SA body in and out */ + RETURN_STF_FAILURE(parse_isakmp_sa_body(ipsecdoisit, &proposal_pbs + ,&proposal, &r_sa_pbs, st, FALSE)); + + /* if enabled send Pluto Vendor ID */ + if (SEND_PLUTO_VID) + { + if (!out_vendorid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &md->rbody, VID_STRONGSWAN)) + { + return STF_INTERNAL_ERROR; + } + } + + /* if enabled send Cisco Unity Vendor ID */ + if (SEND_CISCO_UNITY_VID) + { + if (!out_vendorid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &md->rbody, VID_CISCO_UNITY)) + { + return STF_INTERNAL_ERROR; + } + } + + /* + * if the peer sent an OpenPGP Vendor ID we offer the same capability + */ + if (md->openpgp) + { + if (!out_vendorid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &md->rbody, VID_OPENPGP)) + { + return STF_INTERNAL_ERROR; + } + } + + /* Announce our ability to do eXtended AUTHentication to the peer */ + if (!out_vendorid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &md->rbody, VID_MISC_XAUTH)) + { + return STF_INTERNAL_ERROR; + } + + /* Announce our ability to do Dead Peer Detection to the peer */ + if (!out_vendorid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &md->rbody, VID_MISC_DPD)) + { + return STF_INTERNAL_ERROR; + } + + if (md->nat_traversal_vid && nat_traversal_enabled) + { + /* reply if NAT-Traversal draft is supported */ + st->nat_traversal = nat_traversal_vid_to_method(md->nat_traversal_vid); + + if (st->nat_traversal + && !out_vendorid(vids_to_send-- ? ISAKMP_NEXT_VID : ISAKMP_NEXT_NONE + , &md->rbody, md->nat_traversal_vid)) + { + return STF_INTERNAL_ERROR; + } + } + + close_message(&md->rbody); + + /* save initiator SA for HASH */ + clonereplacechunk(st->st_p1isa, sa_pd->pbs.start, pbs_room(&sa_pd->pbs), "sa in main_inI1_outR1()"); + + return STF_OK; +} + +/* STATE_MAIN_I1: HDR, SA --> auth dependent + * PSK_AUTH, DS_AUTH: --> HDR, KE, Ni + * + * The following are not yet implemented: + * PKE_AUTH: --> HDR, KE, [ HASH(1), ] <IDi1_b>PubKey_r, <Ni_b>PubKey_r + * RPKE_AUTH: --> HDR, [ HASH(1), ] <Ni_b>Pubkey_r, <KE_b>Ke_i, + * <IDi1_b>Ke_i [,<<Cert-I_b>Ke_i] + * + * We must verify that the proposal received matches one we sent. + */ +stf_status +main_inR1_outI2(struct msg_digest *md) +{ + struct state *const st = md->st; + + u_int8_t np = ISAKMP_NEXT_NONE; + + /* verify echoed SA */ + { + u_int32_t ipsecdoisit; + pb_stream proposal_pbs; + struct isakmp_proposal proposal; + struct payload_digest *const sapd = md->chain[ISAKMP_NEXT_SA]; + + RETURN_STF_FAILURE(preparse_isakmp_sa_body(&sapd->payload.sa + ,&sapd->pbs, &ipsecdoisit, &proposal_pbs, &proposal)); + if (proposal.isap_notrans != 1) + { + loglog(RC_LOG_SERIOUS, "a single Transform is required in a selecting Oakley Proposal; found %u" + , (unsigned)proposal.isap_notrans); + RETURN_STF_FAILURE(BAD_PROPOSAL_SYNTAX); + } + RETURN_STF_FAILURE(parse_isakmp_sa_body(ipsecdoisit + , &proposal_pbs, &proposal, NULL, st, TRUE)); + } + + if (nat_traversal_enabled && md->nat_traversal_vid) + { + st->nat_traversal = nat_traversal_vid_to_method(md->nat_traversal_vid); + plog("enabling possible NAT-traversal with method %s" + , bitnamesof(natt_type_bitnames, st->nat_traversal)); + } + if (st->nat_traversal & NAT_T_WITH_NATD) + { + np = (st->nat_traversal & NAT_T_WITH_RFC_VALUES) ? + ISAKMP_NEXT_NATD_RFC : ISAKMP_NEXT_NATD_DRAFTS; + } + + /**************** build output packet HDR;KE;Ni ****************/ + + /* HDR out. + * We can't leave this to comm_handle() because the isa_np + * depends on the type of Auth (eventually). + */ + echo_hdr(md, FALSE, ISAKMP_NEXT_KE); + + /* KE out */ + if (!build_and_ship_KE(st, &st->st_gi, st->st_oakley.group + , &md->rbody, ISAKMP_NEXT_NONCE)) + return STF_INTERNAL_ERROR; + +#ifdef DEBUG + /* Ni out */ + if (!build_and_ship_nonce(&st->st_ni, &md->rbody + , (cur_debugging & IMPAIR_BUST_MI2)? ISAKMP_NEXT_VID : np, "Ni")) + return STF_INTERNAL_ERROR; + + if (cur_debugging & IMPAIR_BUST_MI2) + { + /* generate a pointless large VID payload to push message over MTU */ + pb_stream vid_pbs; + + if (!out_generic(np, &isakmp_vendor_id_desc, &md->rbody, &vid_pbs)) + return STF_INTERNAL_ERROR; + if (!out_zero(1500 /*MTU?*/, &vid_pbs, "Filler VID")) + return STF_INTERNAL_ERROR; + close_output_pbs(&vid_pbs); + } +#else + /* Ni out */ + if (!build_and_ship_nonce(&st->st_ni, &md->rbody, np, "Ni")) + return STF_INTERNAL_ERROR; +#endif + + if (st->nat_traversal & NAT_T_WITH_NATD) + { + if (!nat_traversal_add_natd(ISAKMP_NEXT_NONE, &md->rbody, md)) + return STF_INTERNAL_ERROR; + } + + /* finish message */ + close_message(&md->rbody); + + /* Reinsert the state, using the responder cookie we just received */ + unhash_state(st); + memcpy(st->st_rcookie, md->hdr.isa_rcookie, COOKIE_SIZE); + insert_state(st); /* needs cookies, connection, and msgid (0) */ + + return STF_OK; +} + +/* STATE_MAIN_R1: + * PSK_AUTH, DS_AUTH: HDR, KE, Ni --> HDR, KE, Nr + * + * The following are not yet implemented: + * PKE_AUTH: HDR, KE, [ HASH(1), ] <IDi1_b>PubKey_r, <Ni_b>PubKey_r + * --> HDR, KE, <IDr1_b>PubKey_i, <Nr_b>PubKey_i + * RPKE_AUTH: + * HDR, [ HASH(1), ] <Ni_b>Pubkey_r, <KE_b>Ke_i, <IDi1_b>Ke_i [,<<Cert-I_b>Ke_i] + * --> HDR, <Nr_b>PubKey_i, <KE_b>Ke_r, <IDr1_b>Ke_r + */ +stf_status +main_inI2_outR2(struct msg_digest *md) +{ + struct state *const st = md->st; + pb_stream *keyex_pbs = &md->chain[ISAKMP_NEXT_KE]->pbs; + + /* send CR if auth is RSA and no preloaded RSA public key exists*/ + bool RSA_auth = st->st_oakley.auth == OAKLEY_RSA_SIG + || st->st_oakley.auth == XAUTHInitRSA + || st->st_oakley.auth == XAUTHRespRSA; + bool send_cr = !no_cr_send && RSA_auth && !has_preloaded_public_key(st); + + u_int8_t np = ISAKMP_NEXT_NONE; + + /* KE in */ + RETURN_STF_FAILURE(accept_KE(&st->st_gi, "Gi", st->st_oakley.group, keyex_pbs)); + + /* Ni in */ + RETURN_STF_FAILURE(accept_nonce(md, &st->st_ni, "Ni")); + + if (st->nat_traversal & NAT_T_WITH_NATD) + { + nat_traversal_natd_lookup(md); + + np = (st->nat_traversal & NAT_T_WITH_RFC_VALUES) ? + ISAKMP_NEXT_NATD_RFC : ISAKMP_NEXT_NATD_DRAFTS; + } + if (st->nat_traversal) + { + nat_traversal_show_result(st->nat_traversal, md->sender_port); + } + if (st->nat_traversal & NAT_T_WITH_KA) + { + nat_traversal_new_ka_event(); + } + + /* decode certificate requests */ + st->st_connection->got_certrequest = FALSE; + decode_cr(md, st->st_connection); + + /**************** build output packet HDR;KE;Nr ****************/ + + /* HDR out done */ + + /* KE out */ + if (!build_and_ship_KE(st, &st->st_gr, st->st_oakley.group + , &md->rbody, ISAKMP_NEXT_NONCE)) + return STF_INTERNAL_ERROR; + +#ifdef DEBUG + /* Nr out */ + if (!build_and_ship_nonce(&st->st_nr, &md->rbody + , (cur_debugging & IMPAIR_BUST_MR2)? ISAKMP_NEXT_VID + : (send_cr? ISAKMP_NEXT_CR : np), "Nr")) + return STF_INTERNAL_ERROR; + + if (cur_debugging & IMPAIR_BUST_MR2) + { + /* generate a pointless large VID payload to push message over MTU */ + pb_stream vid_pbs; + + if (!out_generic((send_cr)? ISAKMP_NEXT_CR : np, + &isakmp_vendor_id_desc, &md->rbody, &vid_pbs)) + return STF_INTERNAL_ERROR; + if (!out_zero(1500 /*MTU?*/, &vid_pbs, "Filler VID")) + return STF_INTERNAL_ERROR; + close_output_pbs(&vid_pbs); + } +#else + /* Nr out */ + if (!build_and_ship_nonce(&st->st_nr, &md->rbody, + (send_cr)? ISAKMP_NEXT_CR : np, "Nr")) + return STF_INTERNAL_ERROR; +#endif + + /* CR out */ + if (send_cr) + { + if (st->st_connection->kind == CK_PERMANENT) + { + if (!build_and_ship_CR(CERT_X509_SIGNATURE + , st->st_connection->spd.that.ca + , &md->rbody, np)) + return STF_INTERNAL_ERROR; + } + else + { + generalName_t *ca = NULL; + + if (collect_rw_ca_candidates(md, &ca)) + { + generalName_t *gn; + + for (gn = ca; gn != NULL; gn = gn->next) + { + if (!build_and_ship_CR(CERT_X509_SIGNATURE, gn->name + , &md->rbody + , gn->next == NULL ? np : ISAKMP_NEXT_CR)) + return STF_INTERNAL_ERROR; + } + free_generalNames(ca, FALSE); + } + else + { + if (!build_and_ship_CR(CERT_X509_SIGNATURE, empty_chunk + , &md->rbody, np)) + return STF_INTERNAL_ERROR; + } + } + } + + if (st->nat_traversal & NAT_T_WITH_NATD) + { + if (!nat_traversal_add_natd(ISAKMP_NEXT_NONE, &md->rbody, md)) + return STF_INTERNAL_ERROR; + } + + /* finish message */ + close_message(&md->rbody); + + /* next message will be encrypted, but not this one. + * We could defer this calculation. + */ + compute_dh_shared(st, st->st_gi, st->st_oakley.group); + if (!generate_skeyids_iv(st)) + return STF_FAIL + AUTHENTICATION_FAILED; + update_iv(st); + + return STF_OK; +} + +/* STATE_MAIN_I2: + * SMF_PSK_AUTH: HDR, KE, Nr --> HDR*, IDi1, HASH_I + * SMF_DS_AUTH: HDR, KE, Nr --> HDR*, IDi1, [ CERT, ] SIG_I + * + * The following are not yet implemented. + * SMF_PKE_AUTH: HDR, KE, <IDr1_b>PubKey_i, <Nr_b>PubKey_i + * --> HDR*, HASH_I + * SMF_RPKE_AUTH: HDR, <Nr_b>PubKey_i, <KE_b>Ke_r, <IDr1_b>Ke_r + * --> HDR*, HASH_I + */ +stf_status +main_inR2_outI3(struct msg_digest *md) +{ + struct state *const st = md->st; + pb_stream *const keyex_pbs = &md->chain[ISAKMP_NEXT_KE]->pbs; + pb_stream id_pbs; /* ID Payload; also used for hash calculation */ + + certpolicy_t cert_policy = st->st_connection->spd.this.sendcert; + cert_t mycert = st->st_connection->spd.this.cert; + bool requested, send_cert, send_cr; + + bool RSA_auth = st->st_oakley.auth == OAKLEY_RSA_SIG + || st->st_oakley.auth == XAUTHInitRSA + || st->st_oakley.auth == XAUTHRespRSA; + + int auth_payload = RSA_auth ? ISAKMP_NEXT_SIG : ISAKMP_NEXT_HASH; + + /* KE in */ + RETURN_STF_FAILURE(accept_KE(&st->st_gr, "Gr", st->st_oakley.group, keyex_pbs)); + + /* Nr in */ + RETURN_STF_FAILURE(accept_nonce(md, &st->st_nr, "Nr")); + + /* decode certificate requests */ + st->st_connection->got_certrequest = FALSE; + decode_cr(md, st->st_connection); + + /* free collected certificate requests since as initiator + * we don't heed them anyway + */ + free_generalNames(st->st_connection->requested_ca, TRUE); + st->st_connection->requested_ca = NULL; + + /* send certificate if auth is RSA, we have one and we want + * or are requested to send it + */ + requested = cert_policy == CERT_SEND_IF_ASKED + && st->st_connection->got_certrequest; + send_cert = RSA_auth && mycert.type != CERT_NONE + && (cert_policy == CERT_ALWAYS_SEND || requested); + + /* send certificate request if we don't have a preloaded RSA public key */ + send_cr = !no_cr_send && send_cert && !has_preloaded_public_key(st); + + /* done parsing; initialize crypto */ + compute_dh_shared(st, st->st_gr, st->st_oakley.group); + if (!generate_skeyids_iv(st)) + return STF_FAIL + AUTHENTICATION_FAILED; + + if (st->nat_traversal & NAT_T_WITH_NATD) + { + nat_traversal_natd_lookup(md); + } + if (st->nat_traversal) + { + nat_traversal_show_result(st->nat_traversal, md->sender_port); + } + if (st->nat_traversal & NAT_T_WITH_KA) + { + nat_traversal_new_ka_event(); + } + + /*************** build output packet HDR*;IDii;HASH/SIG_I ***************/ + /* ??? NOTE: this is almost the same as main_inI3_outR3's code */ + + /* HDR* out done */ + + /* IDii out */ + { + struct isakmp_ipsec_id id_hd; + chunk_t id_b; + + build_id_payload(&id_hd, &id_b, &st->st_connection->spd.this); + id_hd.isaiid_np = (send_cert)? ISAKMP_NEXT_CERT : auth_payload; + if (!out_struct(&id_hd, &isakmp_ipsec_identification_desc, &md->rbody, &id_pbs) + || !out_chunk(id_b, &id_pbs, "my identity")) + return STF_INTERNAL_ERROR; + close_output_pbs(&id_pbs); + } + + /* CERT out */ + if (RSA_auth) + { + DBG(DBG_CONTROL, + DBG_log("our certificate policy is %s" + , enum_name(&cert_policy_names, cert_policy)) + ) + if (mycert.type != CERT_NONE) + { + const char *request_text = ""; + + if (cert_policy == CERT_SEND_IF_ASKED) + request_text = (send_cert)? "upon request":"without request"; + plog("we have a cert %s sending it %s" + , send_cert? "and are":"but are not", request_text); + } + else + { + plog("we don't have a cert"); + } + } + if (send_cert) + { + pb_stream cert_pbs; + + struct isakmp_cert cert_hd; + cert_hd.isacert_np = (send_cr)? ISAKMP_NEXT_CR : ISAKMP_NEXT_SIG; + cert_hd.isacert_type = mycert.type; + + if (!out_struct(&cert_hd, &isakmp_ipsec_certificate_desc, &md->rbody, &cert_pbs)) + return STF_INTERNAL_ERROR; + if (!out_chunk(get_mycert(mycert), &cert_pbs, "CERT")) + return STF_INTERNAL_ERROR; + close_output_pbs(&cert_pbs); + } + + /* CR out */ + if (send_cr) + { + if (!build_and_ship_CR(mycert.type, st->st_connection->spd.that.ca + , &md->rbody, ISAKMP_NEXT_SIG)) + return STF_INTERNAL_ERROR; + } + + /* HASH_I or SIG_I out */ + { + u_char hash_val[MAX_DIGEST_LEN]; + size_t hash_len = main_mode_hash(st, hash_val, TRUE, &id_pbs); + + if (auth_payload == ISAKMP_NEXT_HASH) + { + /* HASH_I out */ + if (!out_generic_raw(ISAKMP_NEXT_NONE, &isakmp_hash_desc, &md->rbody + , hash_val, hash_len, "HASH_I")) + return STF_INTERNAL_ERROR; + } + else + { + /* SIG_I out */ + u_char sig_val[RSA_MAX_OCTETS]; + size_t sig_len = RSA_sign_hash(st->st_connection + , sig_val, hash_val, hash_len); + + if (sig_len == 0) + { + loglog(RC_LOG_SERIOUS, "unable to locate my private key for RSA Signature"); + return STF_FAIL + AUTHENTICATION_FAILED; + } + + if (!out_generic_raw(ISAKMP_NEXT_NONE, &isakmp_signature_desc + , &md->rbody, sig_val, sig_len, "SIG_I")) + return STF_INTERNAL_ERROR; + } + } + + /* encrypt message, except for fixed part of header */ + + /* st_new_iv was computed by generate_skeyids_iv */ + if (!encrypt_message(&md->rbody, st)) + return STF_INTERNAL_ERROR; /* ??? we may be partly committed */ + + return STF_OK; +} + +/* Shared logic for asynchronous lookup of DNS KEY records. + * Used for STATE_MAIN_R2 and STATE_MAIN_I3. + */ + +enum key_oppo_step { + kos_null, + kos_his_txt +#ifdef USE_KEYRR + , kos_his_key +#endif +}; + +struct key_continuation { + struct adns_continuation ac; /* common prefix */ + struct msg_digest *md; + enum key_oppo_step step; + bool failure_ok; + err_t last_ugh; +}; + +typedef stf_status (key_tail_fn)(struct msg_digest *md + , struct key_continuation *kc); +static void +report_key_dns_failure(struct id *id, err_t ugh) +{ + char id_buf[BUF_LEN]; /* arbitrary limit on length of ID reported */ + + (void) idtoa(id, id_buf, sizeof(id_buf)); + loglog(RC_LOG_SERIOUS, "no RSA public key known for '%s'" + "; DNS search for KEY failed (%s)", id_buf, ugh); +} + + +/* Processs the Main Mode ID Payload and the Authenticator + * (Hash or Signature Payload). + * If a DNS query is still needed to get the other host's public key, + * the query is initiated and STF_SUSPEND is returned. + * Note: parameter kc is a continuation containing the results from + * the previous DNS query, or NULL indicating no query has been issued. + */ +static stf_status +main_id_and_auth(struct msg_digest *md + , bool initiator /* are we the Initiator? */ + , cont_fn_t cont_fn /* continuation function */ + , const struct key_continuation *kc /* current state, can be NULL */ +) +{ + struct state *st = md->st; + u_char hash_val[MAX_DIGEST_LEN]; + size_t hash_len; + struct id peer; + stf_status r = STF_OK; + + /* ID Payload in */ + if (!decode_peer_id(md, &peer)) + return STF_FAIL + INVALID_ID_INFORMATION; + + /* Hash the ID Payload. + * main_mode_hash requires idpl->cur to be at end of payload + * so we temporarily set if so. + */ + { + pb_stream *idpl = &md->chain[ISAKMP_NEXT_ID]->pbs; + u_int8_t *old_cur = idpl->cur; + + idpl->cur = idpl->roof; + hash_len = main_mode_hash(st, hash_val, !initiator, idpl); + idpl->cur = old_cur; + } + + switch (st->st_oakley.auth) + { + case OAKLEY_PRESHARED_KEY: + case XAUTHInitPreShared: + case XAUTHRespPreShared: + { + pb_stream *const hash_pbs = &md->chain[ISAKMP_NEXT_HASH]->pbs; + + if (pbs_left(hash_pbs) != hash_len + || memcmp(hash_pbs->cur, hash_val, hash_len) != 0) + { + DBG_cond_dump(DBG_CRYPT, "received HASH:" + , hash_pbs->cur, pbs_left(hash_pbs)); + loglog(RC_LOG_SERIOUS, "received Hash Payload does not match computed value"); + /* XXX Could send notification back */ + r = STF_FAIL + INVALID_HASH_INFORMATION; + } + } + break; + + case OAKLEY_RSA_SIG: + case XAUTHInitRSA: + case XAUTHRespRSA: + r = RSA_check_signature(&peer, st, hash_val, hash_len + , &md->chain[ISAKMP_NEXT_SIG]->pbs +#ifdef USE_KEYRR + , kc == NULL? NULL : kc->ac.keys_from_dns +#endif /* USE_KEYRR */ + , kc == NULL? NULL : kc->ac.gateways_from_dns + ); + + if (r == STF_SUSPEND) + { + /* initiate/resume asynchronous DNS lookup for key */ + struct key_continuation *nkc + = alloc_thing(struct key_continuation, "key continuation"); + enum key_oppo_step step_done = kc == NULL? kos_null : kc->step; + err_t ugh; + + /* Record that state is used by a suspended md */ + passert(st->st_suspended_md == NULL); + st->st_suspended_md = md; + + nkc->failure_ok = FALSE; + nkc->md = md; + + switch (step_done) + { + case kos_null: + /* first try: look for the TXT records */ + nkc->step = kos_his_txt; +#ifdef USE_KEYRR + nkc->failure_ok = TRUE; +#endif + ugh = start_adns_query(&peer + , &peer /* SG itself */ + , T_TXT + , cont_fn + , &nkc->ac); + break; + +#ifdef USE_KEYRR + case kos_his_txt: + /* second try: look for the KEY records */ + nkc->step = kos_his_key; + ugh = start_adns_query(&peer + , NULL /* no sgw for KEY */ + , T_KEY + , cont_fn + , &nkc->ac); + break; +#endif /* USE_KEYRR */ + + default: + bad_case(step_done); + } + + if (ugh != NULL) + { + report_key_dns_failure(&peer, ugh); + st->st_suspended_md = NULL; + r = STF_FAIL + INVALID_KEY_INFORMATION; + } + } + break; + + default: + bad_case(st->st_oakley.auth); + } + if (r != STF_OK) + return r; + + DBG(DBG_CRYPT, DBG_log("authentication succeeded")); + + /* + * With the peer ID known, let's see if we need to switch connections. + */ + if (!switch_connection(md, &peer, initiator)) + return STF_FAIL + INVALID_ID_INFORMATION; + + return r; +} + +/* This continuation is called as part of either + * the main_inI3_outR3 state or main_inR3 state. + * + * The "tail" function is the corresponding tail + * function main_inI3_outR3_tail | main_inR3_tail, + * either directly when the state is started, or via + * adns continuation. + * + * Basically, we go around in a circle: + * main_in?3* -> key_continue + * ^ \ + * / V + * adns main_in?3*_tail + * ^ | + * \ V + * main_id_and_auth + * + * until such time as main_id_and_auth is able + * to find authentication, or we run out of things + * to try. + */ +static void +key_continue(struct adns_continuation *cr +, err_t ugh +, key_tail_fn *tail) +{ + struct key_continuation *kc = (void *)cr; + struct state *st = kc->md->st; + + passert(cur_state == NULL); + + /* if st == NULL, our state has been deleted -- just clean up */ + if (st != NULL) + { + stf_status r; + + passert(st->st_suspended_md == kc->md); + st->st_suspended_md = NULL; /* no longer connected or suspended */ + cur_state = st; + + if (!kc->failure_ok && ugh != NULL) + { + report_key_dns_failure(&st->st_connection->spd.that.id, ugh); + r = STF_FAIL + INVALID_KEY_INFORMATION; + } + else + { + +#ifdef USE_KEYRR + passert(kc->step == kos_his_txt || kc->step == kos_his_key); +#else + passert(kc->step == kos_his_txt); +#endif + kc->last_ugh = ugh; /* record previous error in case we need it */ + r = (*tail)(kc->md, kc); + } + complete_state_transition(&kc->md, r); + } + if (kc->md != NULL) + release_md(kc->md); + cur_state = NULL; +} + +/* STATE_MAIN_R2: + * PSK_AUTH: HDR*, IDi1, HASH_I --> HDR*, IDr1, HASH_R + * DS_AUTH: HDR*, IDi1, [ CERT, ] SIG_I --> HDR*, IDr1, [ CERT, ] SIG_R + * PKE_AUTH, RPKE_AUTH: HDR*, HASH_I --> HDR*, HASH_R + * + * Broken into parts to allow asynchronous DNS lookup. + * + * - main_inI3_outR3 to start + * - main_inI3_outR3_tail to finish or suspend for DNS lookup + * - main_inI3_outR3_continue to start main_inI3_outR3_tail again + */ +static key_tail_fn main_inI3_outR3_tail; /* forward */ + +stf_status +main_inI3_outR3(struct msg_digest *md) +{ + return main_inI3_outR3_tail(md, NULL); +} + +static void +main_inI3_outR3_continue(struct adns_continuation *cr, err_t ugh) +{ + key_continue(cr, ugh, main_inI3_outR3_tail); +} + +static stf_status +main_inI3_outR3_tail(struct msg_digest *md +, struct key_continuation *kc) +{ + struct state *const st = md->st; + u_int8_t auth_payload; + pb_stream r_id_pbs; /* ID Payload; also used for hash calculation */ + certpolicy_t cert_policy; + cert_t mycert; + bool RSA_auth; + bool send_cert; + bool requested; + + /* ID and HASH_I or SIG_I in + * Note: this may switch the connection being used! + */ + { + stf_status r = main_id_and_auth(md, FALSE + , main_inI3_outR3_continue + , kc); + + if (r != STF_OK) + return r; + } + + /* send certificate if auth is RSA, we have one and we want + * or are requested to send it + */ + cert_policy = st->st_connection->spd.this.sendcert; + mycert = st->st_connection->spd.this.cert; + requested = cert_policy == CERT_SEND_IF_ASKED + && st->st_connection->got_certrequest; + RSA_auth = st->st_oakley.auth == OAKLEY_RSA_SIG + || st->st_oakley.auth == XAUTHInitRSA + || st->st_oakley.auth == XAUTHRespRSA; + send_cert = RSA_auth + && mycert.type != CERT_NONE + && (cert_policy == CERT_ALWAYS_SEND || requested); + + /*************** build output packet HDR*;IDir;HASH/SIG_R ***************/ + /* proccess_packet() would automatically generate the HDR* + * payload if smc->first_out_payload is not ISAKMP_NEXT_NONE. + * We don't do this because we wish there to be no partially + * built output packet if we need to suspend for asynch DNS. + */ + /* ??? NOTE: this is almost the same as main_inR2_outI3's code */ + + /* HDR* out + * If auth were PKE_AUTH or RPKE_AUTH, ISAKMP_NEXT_HASH would + * be first payload. + */ + echo_hdr(md, TRUE, ISAKMP_NEXT_ID); + + auth_payload = RSA_auth ? ISAKMP_NEXT_SIG : ISAKMP_NEXT_HASH; + + /* IDir out */ + { + /* id_hd should be struct isakmp_id, but struct isakmp_ipsec_id + * allows build_id_payload() to work for both phases. + */ + struct isakmp_ipsec_id id_hd; + chunk_t id_b; + + build_id_payload(&id_hd, &id_b, &st->st_connection->spd.this); + id_hd.isaiid_np = (send_cert)? ISAKMP_NEXT_CERT : auth_payload; + if (!out_struct(&id_hd, &isakmp_ipsec_identification_desc, &md->rbody, &r_id_pbs) + || !out_chunk(id_b, &r_id_pbs, "my identity")) + return STF_INTERNAL_ERROR; + close_output_pbs(&r_id_pbs); + } + + /* CERT out */ + if (RSA_auth) + { + DBG(DBG_CONTROL, + DBG_log("our certificate policy is %s" + , enum_name(&cert_policy_names, cert_policy)) + ) + if (mycert.type != CERT_NONE) + { + const char *request_text = ""; + + if (cert_policy == CERT_SEND_IF_ASKED) + request_text = (send_cert)? "upon request":"without request"; + plog("we have a cert %s sending it %s" + , send_cert? "and are":"but are not", request_text); + } + else + { + plog("we don't have a cert"); + } + } + if (send_cert) + { + pb_stream cert_pbs; + + struct isakmp_cert cert_hd; + cert_hd.isacert_np = ISAKMP_NEXT_SIG; + cert_hd.isacert_type = mycert.type; + + if (!out_struct(&cert_hd, &isakmp_ipsec_certificate_desc, &md->rbody, &cert_pbs)) + return STF_INTERNAL_ERROR; + if (!out_chunk(get_mycert(mycert), &cert_pbs, "CERT")) + return STF_INTERNAL_ERROR; + close_output_pbs(&cert_pbs); + } + + /* HASH_R or SIG_R out */ + { + u_char hash_val[MAX_DIGEST_LEN]; + size_t hash_len = main_mode_hash(st, hash_val, FALSE, &r_id_pbs); + + if (auth_payload == ISAKMP_NEXT_HASH) + { + /* HASH_R out */ + if (!out_generic_raw(ISAKMP_NEXT_NONE, &isakmp_hash_desc, &md->rbody + , hash_val, hash_len, "HASH_R")) + return STF_INTERNAL_ERROR; + } + else + { + /* SIG_R out */ + u_char sig_val[RSA_MAX_OCTETS]; + size_t sig_len = RSA_sign_hash(st->st_connection + , sig_val, hash_val, hash_len); + + if (sig_len == 0) + { + loglog(RC_LOG_SERIOUS, "unable to locate my private key for RSA Signature"); + return STF_FAIL + AUTHENTICATION_FAILED; + } + + if (!out_generic_raw(ISAKMP_NEXT_NONE, &isakmp_signature_desc + , &md->rbody, sig_val, sig_len, "SIG_R")) + return STF_INTERNAL_ERROR; + } + } + + /* encrypt message, sans fixed part of header */ + + if (!encrypt_message(&md->rbody, st)) + return STF_INTERNAL_ERROR; /* ??? we may be partly committed */ + + /* Last block of Phase 1 (R3), kept for Phase 2 IV generation */ + DBG_cond_dump(DBG_CRYPT, "last encrypted block of Phase 1:" + , st->st_new_iv, st->st_new_iv_len); + + ISAKMP_SA_established(st->st_connection, st->st_serialno); + + /* Save Phase 1 IV */ + st->st_ph1_iv_len = st->st_new_iv_len; + set_ph1_iv(st, st->st_new_iv); + + return STF_OK; +} + +/* STATE_MAIN_I3: + * Handle HDR*;IDir;HASH/SIG_R from responder. + * + * Broken into parts to allow asynchronous DNS for KEY records. + * + * - main_inR3 to start + * - main_inR3_tail to finish or suspend for DNS lookup + * - main_inR3_continue to start main_inR3_tail again + */ + +static key_tail_fn main_inR3_tail; /* forward */ + +stf_status +main_inR3(struct msg_digest *md) +{ + return main_inR3_tail(md, NULL); +} + +static void +main_inR3_continue(struct adns_continuation *cr, err_t ugh) +{ + key_continue(cr, ugh, main_inR3_tail); +} + +static stf_status +main_inR3_tail(struct msg_digest *md +, struct key_continuation *kc) +{ + struct state *const st = md->st; + + /* ID and HASH_R or SIG_R in + * Note: this may switch the connection being used! + */ + { + stf_status r = main_id_and_auth(md, TRUE, main_inR3_continue, kc); + + if (r != STF_OK) + return r; + } + + /**************** done input ****************/ + + ISAKMP_SA_established(st->st_connection, st->st_serialno); + + /* Save Phase 1 IV */ + st->st_ph1_iv_len = st->st_new_iv_len; + set_ph1_iv(st, st->st_new_iv); + + + update_iv(st); /* finalize our Phase 1 IV */ + + return STF_OK; +} + +/* Handle first message of Phase 2 -- Quick Mode. + * HDR*, HASH(1), SA, Ni [, KE ] [, IDci, IDcr ] --> + * HDR*, HASH(2), SA, Nr [, KE ] [, IDci, IDcr ] + * (see RFC 2409 "IKE" 5.5) + * Installs inbound IPsec SAs. + * Although this seems early, we know enough to do so, and + * this way we know that it is soon enough to catch all + * packets that other side could send using this IPsec SA. + * + * Broken into parts to allow asynchronous DNS for TXT records: + * + * - quick_inI1_outR1 starts the ball rolling. + * It checks and parses enough to learn the Phase 2 IDs + * + * - quick_inI1_outR1_tail does the rest of the job + * unless DNS must be consulted. In that case, + * it starts a DNS query, salts away what is needed + * to continue, and suspends. Calls + * + quick_inI1_outR1_start_query + * + quick_inI1_outR1_process_answer + * + * - quick_inI1_outR1_continue will restart quick_inI1_outR1_tail + * when DNS comes back with an answer. + * + * A big chunk of quick_inI1_outR1_tail is executed twice. + * This is necessary because the set of connections + * might change while we are awaiting DNS. + * When first called, gateways_from_dns == NULL. If DNS is + * consulted asynchronously, gateways_from_dns != NULL the second time. + * Remember that our state object might disappear too! + * + * + * If the connection is opportunistic, we must verify delegation. + * + * 1. Check that we are authorized to be SG for + * our client. We look for the TXT record that + * delegates us. We also check that the public + * key (if present) matches the private key we used. + * Eventually, we should probably require DNSsec + * authentication for our side. + * + * 2. If our client TXT record did not include a + * public key, check the KEY record indicated + * by the identity in the TXT record. + * + * 3. If the peer's client is the peer itself, we + * consider it authenticated. Otherwise, we check + * the TXT record for the client to see that + * the identity of the SG matches the peer and + * that some public key (if present in the TXT) + * matches. We need not check the public key if + * it isn't in the TXT record. + * + * Since p isn't yet instantiated, we need to look + * in c for description of peer. + * + * We cannot afford to block waiting for a DNS query. + * The code here is structured as two halves: + * - process the result of just completed + * DNS query (if any) + * - if another query is needed, initiate the next + * DNS query and suspend + */ + +enum verify_oppo_step { + vos_fail, + vos_start, + vos_our_client, + vos_our_txt, +#ifdef USE_KEYRR + vos_our_key, +#endif /* USE_KEYRR */ + vos_his_client, + vos_done +}; + +static const char *const verify_step_name[] = { + "vos_fail", + "vos_start", + "vos_our_client", + "vos_our_txt", +#ifdef USE_KEYRR + "vos_our_key", +#endif /* USE_KEYRR */ + "vos_his_client", + "vos_done" +}; + +/* hold anything we can handle of a Phase 2 ID */ +struct p2id { + ip_subnet net; + u_int8_t proto; + u_int16_t port; +}; + +struct verify_oppo_bundle { + enum verify_oppo_step step; + bool failure_ok; /* if true, quick_inI1_outR1_continue will try + * other things on DNS failure */ + struct msg_digest *md; + struct p2id my, his; + unsigned int new_iv_len; /* p1st's might change */ + u_char new_iv[MAX_DIGEST_LEN]; + /* int whackfd; */ /* not needed because we are Responder */ +}; + +struct verify_oppo_continuation { + struct adns_continuation ac; /* common prefix */ + struct verify_oppo_bundle b; +}; + +static stf_status quick_inI1_outR1_tail(struct verify_oppo_bundle *b + , struct adns_continuation *ac); + +stf_status +quick_inI1_outR1(struct msg_digest *md) +{ + const struct state *const p1st = md->st; + struct connection *c = p1st->st_connection; + struct payload_digest *const id_pd = md->chain[ISAKMP_NEXT_ID]; + struct verify_oppo_bundle b; + + /* HASH(1) in */ + CHECK_QUICK_HASH(md + , quick_mode_hash12(hash_val, hash_pbs->roof, md->message_pbs.roof + , p1st, &md->hdr.isa_msgid, FALSE) + , "HASH(1)", "Quick I1"); + + /* [ IDci, IDcr ] in + * We do this now (probably out of physical order) because + * we wish to select the correct connection before we consult + * it for policy. + */ + + if (id_pd != NULL) + { + /* ??? we are assuming IPSEC_DOI */ + + /* IDci (initiator is peer) */ + + if (!decode_net_id(&id_pd->payload.ipsec_id, &id_pd->pbs + , &b.his.net, "peer client")) + return STF_FAIL + INVALID_ID_INFORMATION; + + /* Hack for MS 818043 NAT-T Update */ + + if (id_pd->payload.ipsec_id.isaiid_idtype == ID_FQDN) + happy(addrtosubnet(&c->spd.that.host_addr, &b.his.net)); + + /* End Hack for MS 818043 NAT-T Update */ + + b.his.proto = id_pd->payload.ipsec_id.isaiid_protoid; + b.his.port = id_pd->payload.ipsec_id.isaiid_port; + b.his.net.addr.u.v4.sin_port = htons(b.his.port); + + /* IDcr (we are responder) */ + + if (!decode_net_id(&id_pd->next->payload.ipsec_id, &id_pd->next->pbs + , &b.my.net, "our client")) + return STF_FAIL + INVALID_ID_INFORMATION; + + b.my.proto = id_pd->next->payload.ipsec_id.isaiid_protoid; + b.my.port = id_pd->next->payload.ipsec_id.isaiid_port; + b.my.net.addr.u.v4.sin_port = htons(b.my.port); + } + else + { + /* implicit IDci and IDcr: peer and self */ + if (!sameaddrtype(&c->spd.this.host_addr, &c->spd.that.host_addr)) + return STF_FAIL; + + happy(addrtosubnet(&c->spd.this.host_addr, &b.my.net)); + happy(addrtosubnet(&c->spd.that.host_addr, &b.his.net)); + b.his.proto = b.my.proto = 0; + b.his.port = b.my.port = 0; + } + b.step = vos_start; + b.md = md; + b.new_iv_len = p1st->st_new_iv_len; + memcpy(b.new_iv, p1st->st_new_iv, p1st->st_new_iv_len); + return quick_inI1_outR1_tail(&b, NULL); +} + +static void +report_verify_failure(struct verify_oppo_bundle *b, err_t ugh) +{ + struct state *st = b->md->st; + char fgwb[ADDRTOT_BUF] + , cb[ADDRTOT_BUF]; + ip_address client; + err_t which; + + switch (b->step) + { + case vos_our_client: + case vos_our_txt: +#ifdef USE_KEYRR + case vos_our_key: +#endif /* USE_KEYRR */ + which = "our"; + networkof(&b->my.net, &client); + break; + + case vos_his_client: + which = "his"; + networkof(&b->his.net, &client); + break; + + case vos_start: + case vos_done: + case vos_fail: + default: + bad_case(b->step); + } + + addrtot(&st->st_connection->spd.that.host_addr, 0, fgwb, sizeof(fgwb)); + addrtot(&client, 0, cb, sizeof(cb)); + loglog(RC_OPPOFAILURE + , "gateway %s wants connection with %s as %s client, but DNS fails to confirm delegation: %s" + , fgwb, cb, which, ugh); +} + +static void +quick_inI1_outR1_continue(struct adns_continuation *cr, err_t ugh) +{ + stf_status r; + struct verify_oppo_continuation *vc = (void *)cr; + struct verify_oppo_bundle *b = &vc->b; + struct state *st = b->md->st; + + passert(cur_state == NULL); + /* if st == NULL, our state has been deleted -- just clean up */ + if (st != NULL) + { + passert(st->st_suspended_md == b->md); + st->st_suspended_md = NULL; /* no longer connected or suspended */ + cur_state = st; + if (!b->failure_ok && ugh != NULL) + { + report_verify_failure(b, ugh); + r = STF_FAIL + INVALID_ID_INFORMATION; + } + else + { + r = quick_inI1_outR1_tail(b, cr); + } + complete_state_transition(&b->md, r); + } + if (b->md != NULL) + release_md(b->md); + cur_state = NULL; +} + +static stf_status +quick_inI1_outR1_start_query(struct verify_oppo_bundle *b +, enum verify_oppo_step next_step) +{ + struct msg_digest *md = b->md; + struct state *p1st = md->st; + struct connection *c = p1st->st_connection; + struct verify_oppo_continuation *vc + = alloc_thing(struct verify_oppo_continuation, "verify continuation"); + struct id id /* subject of query */ + , *our_id /* needed for myid playing */ + , our_id_space; /* ephemeral: no need for unshare_id_content */ + ip_address client; + err_t ugh; + + /* Record that state is used by a suspended md */ + b->step = next_step; /* not just vc->b.step */ + vc->b = *b; + passert(p1st->st_suspended_md == NULL); + p1st->st_suspended_md = b->md; + + DBG(DBG_CONTROL, + { + char ours[SUBNETTOT_BUF]; + char his[SUBNETTOT_BUF]; + + subnettot(&c->spd.this.client, 0, ours, sizeof(ours)); + subnettot(&c->spd.that.client, 0, his, sizeof(his)); + + DBG_log("responding with DNS query - from %s to %s new state: %s" + , ours, his, verify_step_name[b->step]); + }); + + /* Resolve %myid in a cheesy way. + * We have to do the resolution because start_adns_query + * et al have insufficient information to do so. + * If %myid is already known, we'll use that value + * (XXX this may be a mistake: it could be stale). + * If %myid is unknown, we should check to see if + * there are credentials for the IP address or the FQDN. + * Instead, we'll just assume the IP address since we are + * acting as the responder and only the IP address would + * have gotten it to us. + * We don't even try to do this for the other side: + * %myid makes no sense for the other side (but it is syntactically + * legal). + */ + our_id = resolve_myid(&c->spd.this.id); + if (our_id->kind == ID_NONE) + { + iptoid(&c->spd.this.host_addr, &our_id_space); + our_id = &our_id_space; + } + + switch (next_step) + { + case vos_our_client: + networkof(&b->my.net, &client); + iptoid(&client, &id); + vc->b.failure_ok = b->failure_ok = FALSE; + ugh = start_adns_query(&id + , our_id + , T_TXT + , quick_inI1_outR1_continue + , &vc->ac); + break; + + case vos_our_txt: + vc->b.failure_ok = b->failure_ok = TRUE; + ugh = start_adns_query(our_id + , our_id /* self as SG */ + , T_TXT + , quick_inI1_outR1_continue + , &vc->ac); + break; + +#ifdef USE_KEYRR + case vos_our_key: + vc->b.failure_ok = b->failure_ok = FALSE; + ugh = start_adns_query(our_id + , NULL + , T_KEY + , quick_inI1_outR1_continue + , &vc->ac); + break; +#endif + + case vos_his_client: + networkof(&b->his.net, &client); + iptoid(&client, &id); + vc->b.failure_ok = b->failure_ok = FALSE; + ugh = start_adns_query(&id + , &c->spd.that.id + , T_TXT + , quick_inI1_outR1_continue + , &vc->ac); + break; + + default: + bad_case(next_step); + } + + if (ugh != NULL) + { + /* note: we'd like to use vc->b but vc has been freed + * so we have to use b. This is why we plunked next_state + * into b, not just vc->b. + */ + report_verify_failure(b, ugh); + p1st->st_suspended_md = NULL; + return STF_FAIL + INVALID_ID_INFORMATION; + } + else + { + return STF_SUSPEND; + } +} + +static enum verify_oppo_step +quick_inI1_outR1_process_answer(struct verify_oppo_bundle *b +, struct adns_continuation *ac +, struct state *p1st) +{ + struct connection *c = p1st->st_connection; + enum verify_oppo_step next_step; + err_t ugh = NULL; + + DBG(DBG_CONTROL, + { + char ours[SUBNETTOT_BUF]; + char his[SUBNETTOT_BUF]; + + subnettot(&c->spd.this.client, 0, ours, sizeof(ours)); + subnettot(&c->spd.that.client, 0, his, sizeof(his)); + DBG_log("responding on demand from %s to %s state: %s" + , ours, his, verify_step_name[b->step]); + }); + + /* process just completed DNS query (if any) */ + switch (b->step) + { + case vos_start: + /* no query to digest */ + next_step = vos_our_client; + break; + + case vos_our_client: + next_step = vos_his_client; + { + const struct RSA_private_key *pri = get_RSA_private_key(c); + struct gw_info *gwp; + + if (pri == NULL) + { + ugh = "we don't know our own key"; + break; + } + ugh = "our client does not delegate us as its Security Gateway"; + for (gwp = ac->gateways_from_dns; gwp != NULL; gwp = gwp->next) + { + ugh = "our client delegates us as its Security Gateway but with the wrong public key"; + /* If there is no key in the TXT record, + * we count it as a win, but we will have + * to separately fetch and check the KEY record. + * If there is a key from the TXT record, + * we count it as a win if we match the key. + */ + if (!gwp->gw_key_present) + { + next_step = vos_our_txt; + ugh = NULL; /* good! */ + break; + } + else if (same_RSA_public_key(&pri->pub, &gwp->key->u.rsa)) + { + ugh = NULL; /* good! */ + break; + } + } + } + break; + + case vos_our_txt: + next_step = vos_his_client; + { + const struct RSA_private_key *pri = get_RSA_private_key(c); + + if (pri == NULL) + { + ugh = "we don't know our own key"; + break; + } + { + struct gw_info *gwp; + + for (gwp = ac->gateways_from_dns; gwp != NULL; gwp = gwp->next) + { +#ifdef USE_KEYRR + /* not an error yet, because we have to check KEY RR as well */ + ugh = NULL; +#else + ugh = "our client delegation depends on our " RRNAME " record, but it has the wrong public key"; +#endif + if (gwp->gw_key_present + && same_RSA_public_key(&pri->pub, &gwp->key->u.rsa)) + { + ugh = NULL; /* good! */ + break; + } +#ifdef USE_KEYRR + next_step = vos_our_key; +#endif + } + } + } + break; + +#ifdef USE_KEYRR + case vos_our_key: + next_step = vos_his_client; + { + const struct RSA_private_key *pri = get_RSA_private_key(c); + + if (pri == NULL) + { + ugh = "we don't know our own key"; + break; + } + { + pubkey_list_t *kp; + + ugh = "our client delegation depends on our missing " RRNAME " record"; + for (kp = ac->keys_from_dns; kp != NULL; kp = kp->next) + { + ugh = "our client delegation depends on our " RRNAME " record, but it has the wrong public key"; + if (same_RSA_public_key(&pri->pub, &kp->key->u.rsa)) + { + /* do this only once a day */ + if (!logged_txt_warning) + { + loglog(RC_LOG_SERIOUS, "found KEY RR but not TXT RR. See http://www.freeswan.org/err/txt-change.html."); + logged_txt_warning = TRUE; + } + ugh = NULL; /* good! */ + break; + } + } + } + } + break; +#endif /* USE_KEYRR */ + + case vos_his_client: + next_step = vos_done; + { + struct gw_info *gwp; + + /* check that the public key that authenticated + * the ISAKMP SA (p1st) will do for this gateway. + */ + + ugh = "peer's client does not delegate to peer"; + for (gwp = ac->gateways_from_dns; gwp != NULL; gwp = gwp->next) + { + ugh = "peer and its client disagree about public key"; + /* If there is a key from the TXT record, + * we count it as a win if we match the key. + * If there was no key, we claim a match since + * it implies fetching a KEY from the same + * place we must have gotten it. + */ + if (!gwp->gw_key_present + || same_RSA_public_key(&p1st->st_peer_pubkey->u.rsa + , &gwp->key->u.rsa)) + { + ugh = NULL; /* good! */ + break; + } + } + } + break; + + default: + bad_case(b->step); + } + + if (ugh != NULL) + { + report_verify_failure(b, ugh); + next_step = vos_fail; + } + return next_step; +} + +static stf_status +quick_inI1_outR1_tail(struct verify_oppo_bundle *b +, struct adns_continuation *ac) +{ + struct msg_digest *md = b->md; + struct state *const p1st = md->st; + struct connection *c = p1st->st_connection; + struct payload_digest *const id_pd = md->chain[ISAKMP_NEXT_ID]; + ip_subnet *our_net = &b->my.net + , *his_net = &b->his.net; + + u_char /* set by START_HASH_PAYLOAD: */ + *r_hashval, /* where in reply to jam hash value */ + *r_hash_start; /* from where to start hashing */ + + /* Now that we have identities of client subnets, we must look for + * a suitable connection (our current one only matches for hosts). + */ + { + struct connection *p = find_client_connection(c + , our_net, his_net, b->my.proto, b->my.port, b->his.proto, b->his.port); + + if (p == NULL) + { + /* This message occurs in very puzzling circumstances + * so we must add as much information and beauty as we can. + */ + struct end + me = c->spd.this, + he = c->spd.that; + char buf[2*SUBNETTOT_BUF + 2*ADDRTOT_BUF + 2*BUF_LEN + 2*ADDRTOT_BUF + 12]; /* + 12 for separating */ + size_t l; + + me.client = *our_net; + me.has_client = !subnetisaddr(our_net, &me.host_addr); + me.protocol = b->my.proto; + me.port = b->my.port; + + he.client = *his_net; + he.has_client = !subnetisaddr(his_net, &he.host_addr); + he.protocol = b->his.proto; + he.port = b->his.port; + + l = format_end(buf, sizeof(buf), &me, NULL, TRUE, LEMPTY); + l += snprintf(buf + l, sizeof(buf) - l, "..."); + (void)format_end(buf + l, sizeof(buf) - l, &he, NULL, FALSE, LEMPTY); + plog("cannot respond to IPsec SA request" + " because no connection is known for %s" + , buf); + return STF_FAIL + INVALID_ID_INFORMATION; + } + else if (p != c) + { + /* We've got a better connection: it can support the + * specified clients. But it may need instantiation. + */ + if (p->kind == CK_TEMPLATE) + { + /* Yup, it needs instantiation. How much? + * Is it a Road Warrior connection (simple) + * or is it an Opportunistic connection (needing gw validation)? + */ + if (p->policy & POLICY_OPPO) + { + /* Opportunistic case: delegation must be verified. + * Here be dragons. + */ + enum verify_oppo_step next_step; + ip_address our_client, his_client; + + passert(subnetishost(our_net) && subnetishost(his_net)); + networkof(our_net, &our_client); + networkof(his_net, &his_client); + + next_step = quick_inI1_outR1_process_answer(b, ac, p1st); + if (next_step == vos_fail) + return STF_FAIL + INVALID_ID_INFORMATION; + + /* short circuit: if peer's client is self, + * accept that we've verified delegation in Phase 1 + */ + if (next_step == vos_his_client + && sameaddr(&c->spd.that.host_addr, &his_client)) + next_step = vos_done; + + /* the second chunk: initiate the next DNS query (if any) */ + DBG(DBG_CONTROL, + { + char ours[SUBNETTOT_BUF]; + char his[SUBNETTOT_BUF]; + + subnettot(&c->spd.this.client, 0, ours, sizeof(ours)); + subnettot(&c->spd.that.client, 0, his, sizeof(his)); + + DBG_log("responding on demand from %s to %s new state: %s" + , ours, his, verify_step_name[next_step]); + }); + + /* start next DNS query and suspend (if necessary) */ + if (next_step != vos_done) + return quick_inI1_outR1_start_query(b, next_step); + + /* Instantiate inbound Opportunistic connection, + * carrying over authenticated peer ID + * and filling in a few more details. + * We used to include gateways_from_dns, but that + * seems pointless at this stage of negotiation. + * We should record DNS sec use, if any -- belongs in + * state during perhaps. + */ + p = oppo_instantiate(p, &c->spd.that.host_addr, &c->spd.that.id + , NULL, &our_client, &his_client); + } + else + { + /* Plain Road Warrior: + * instantiate, carrying over authenticated peer ID + */ + p = rw_instantiate(p, &c->spd.that.host_addr, md->sender_port + , his_net, &c->spd.that.id); + } + } +#ifdef DEBUG + /* temporarily bump up cur_debugging to get "using..." message + * printed if we'd want it with new connection. + */ + { + lset_t old_cur_debugging = cur_debugging; + + cur_debugging |= p->extra_debugging; + DBG(DBG_CONTROL, DBG_log("using connection \"%s\"", p->name)); + cur_debugging = old_cur_debugging; + } +#endif + c = p; + } + /* fill in the client's true ip address/subnet */ + if (p->spd.that.has_client_wildcard) + { + p->spd.that.client = *his_net; + p->spd.that.has_client_wildcard = FALSE; + } + else if (is_virtual_connection(c)) + { + c->spd.that.client = *his_net; + c->spd.that.virt = NULL; + if (subnetishost(his_net) && addrinsubnet(&c->spd.that.host_addr, his_net)) + c->spd.that.has_client = FALSE; + } + + /* fill in the client's true port */ + if (p->spd.that.has_port_wildcard) + { + int port = htons(b->his.port); + + setportof(port, &p->spd.that.host_addr); + setportof(port, &p->spd.that.client.addr); + + p->spd.that.port = b->his.port; + p->spd.that.has_port_wildcard = FALSE; + } + } + + /* now that we are sure of our connection, create our new state */ + { + struct state *const st = duplicate_state(p1st); + + /* first: fill in missing bits of our new state object + * note: we don't copy over st_peer_pubkey, the public key + * that authenticated the ISAKMP SA. We only need it in this + * routine, so we can "reach back" to p1st to get it. + */ + + if (st->st_connection != c) + { + struct connection *t = st->st_connection; + + st->st_connection = c; + set_cur_connection(c); + connection_discard(t); + } + + st->st_try = 0; /* not our job to try again from start */ + + st->st_msgid = md->hdr.isa_msgid; + + st->st_new_iv_len = b->new_iv_len; + memcpy(st->st_new_iv, b->new_iv, b->new_iv_len); + + set_cur_state(st); /* (caller will reset) */ + md->st = st; /* feed back new state */ + + st->st_peeruserprotoid = b->his.proto; + st->st_peeruserport = b->his.port; + st->st_myuserprotoid = b->my.proto; + st->st_myuserport = b->my.port; + + insert_state(st); /* needs cookies, connection, and msgid */ + + /* copy the connection's + * IPSEC policy into our state. The ISAKMP policy is water under + * the bridge, I think. It will reflect the ISAKMP SA that we + * are using. + */ + st->st_policy = (p1st->st_policy & POLICY_ISAKMP_MASK) + | (c->policy & ~POLICY_ISAKMP_MASK); + + if (p1st->nat_traversal & NAT_T_DETECTED) + { + st->nat_traversal = p1st->nat_traversal; + nat_traversal_change_port_lookup(md, md->st); + } + else + { + st->nat_traversal = 0; + } + if ((st->nat_traversal & NAT_T_DETECTED) + && (st->nat_traversal & NAT_T_WITH_NATOA)) + { + nat_traversal_natoa_lookup(md); + } + + /* Start the output packet. + * + * proccess_packet() would automatically generate the HDR* + * payload if smc->first_out_payload is not ISAKMP_NEXT_NONE. + * We don't do this because we wish there to be no partially + * built output packet if we need to suspend for asynch DNS. + * + * We build the reply packet as we parse the message since + * the parse_ipsec_sa_body emits the reply SA + */ + + /* HDR* out */ + echo_hdr(md, TRUE, ISAKMP_NEXT_HASH); + + /* HASH(2) out -- first pass */ + START_HASH_PAYLOAD(md->rbody, ISAKMP_NEXT_SA); + + /* process SA (in and out) */ + { + struct payload_digest *const sapd = md->chain[ISAKMP_NEXT_SA]; + pb_stream r_sa_pbs; + struct isakmp_sa sa = sapd->payload.sa; + + /* sa header is unchanged -- except for np */ + sa.isasa_np = ISAKMP_NEXT_NONCE; + if (!out_struct(&sa, &isakmp_sa_desc, &md->rbody, &r_sa_pbs)) + return STF_INTERNAL_ERROR; + + /* parse and accept body */ + st->st_pfs_group = &unset_group; + RETURN_STF_FAILURE(parse_ipsec_sa_body(&sapd->pbs + , &sapd->payload.sa, &r_sa_pbs, FALSE, st)); + } + + passert(st->st_pfs_group != &unset_group); + + if ((st->st_policy & POLICY_PFS) && st->st_pfs_group == NULL) + { + loglog(RC_LOG_SERIOUS, "we require PFS but Quick I1 SA specifies no GROUP_DESCRIPTION"); + return STF_FAIL + NO_PROPOSAL_CHOSEN; /* ??? */ + } + + /* Ni in */ + RETURN_STF_FAILURE(accept_nonce(md, &st->st_ni, "Ni")); + + /* [ KE ] in (for PFS) */ + RETURN_STF_FAILURE(accept_PFS_KE(md, &st->st_gi, "Gi", "Quick Mode I1")); + + plog("responding to Quick Mode"); + + /**** finish reply packet: Nr [, KE ] [, IDci, IDcr ] ****/ + + /* Nr out */ + if (!build_and_ship_nonce(&st->st_nr, &md->rbody + , st->st_pfs_group != NULL? ISAKMP_NEXT_KE : id_pd != NULL? ISAKMP_NEXT_ID : ISAKMP_NEXT_NONE + , "Nr")) + return STF_INTERNAL_ERROR; + + /* [ KE ] out (for PFS) */ + + if (st->st_pfs_group != NULL) + { + if (!build_and_ship_KE(st, &st->st_gr, st->st_pfs_group + , &md->rbody, id_pd != NULL? ISAKMP_NEXT_ID : ISAKMP_NEXT_NONE)) + return STF_INTERNAL_ERROR; + + /* MPZ-Operations might be done after sending the packet... */ + compute_dh_shared(st, st->st_gi, st->st_pfs_group); + } + + /* [ IDci, IDcr ] out */ + if (id_pd != NULL) + { + struct isakmp_ipsec_id *p = (void *)md->rbody.cur; /* UGH! */ + + if (!out_raw(id_pd->pbs.start, pbs_room(&id_pd->pbs), &md->rbody, "IDci")) + return STF_INTERNAL_ERROR; + p->isaiid_np = ISAKMP_NEXT_ID; + + p = (void *)md->rbody.cur; /* UGH! */ + + if (!out_raw(id_pd->next->pbs.start, pbs_room(&id_pd->next->pbs), &md->rbody, "IDcr")) + return STF_INTERNAL_ERROR; + p->isaiid_np = ISAKMP_NEXT_NONE; + } + + if ((st->nat_traversal & NAT_T_WITH_NATOA) + && (st->nat_traversal & LELEM(NAT_TRAVERSAL_NAT_BHND_ME)) + && (st->st_esp.attrs.encapsulation == ENCAPSULATION_MODE_TRANSPORT)) + { + /** Send NAT-OA if our address is NATed and if we use Transport Mode */ + if (!nat_traversal_add_natoa(ISAKMP_NEXT_NONE, &md->rbody, md->st)) + { + return STF_INTERNAL_ERROR; + } + } + if ((st->nat_traversal & NAT_T_DETECTED) + && (st->st_esp.attrs.encapsulation == ENCAPSULATION_MODE_TRANSPORT) + && (c->spd.that.has_client)) + { + /** Remove client **/ + addrtosubnet(&c->spd.that.host_addr, &c->spd.that.client); + c->spd.that.has_client = FALSE; + } + + /* Compute reply HASH(2) and insert in output */ + (void)quick_mode_hash12(r_hashval, r_hash_start, md->rbody.cur + , st, &st->st_msgid, TRUE); + + /* Derive new keying material */ + compute_keymats(st); + + /* Tell the kernel to establish the new inbound SA + * (unless the commit bit is set -- which we don't support). + * We do this before any state updating so that + * failure won't look like success. + */ + if (!install_inbound_ipsec_sa(st)) + return STF_INTERNAL_ERROR; /* ??? we may be partly committed */ + + /* encrypt message, except for fixed part of header */ + + if (!encrypt_message(&md->rbody, st)) + return STF_INTERNAL_ERROR; /* ??? we may be partly committed */ + + return STF_OK; + } +} + +/* + * Initialize RFC 3706 Dead Peer Detection + */ +static void +dpd_init(struct state *st) +{ + struct state *p1st = find_state(st->st_icookie, st->st_rcookie + , &st->st_connection->spd.that.host_addr, 0); + + if (p1st == NULL) + loglog(RC_LOG_SERIOUS, "could not find phase 1 state for DPD"); + else if (p1st->st_dpd) + { + plog("Dead Peer Detection (RFC 3706) enabled"); + /* randomize the first DPD event */ + + event_schedule(EVENT_DPD + , (0.5 + rand()/(RAND_MAX + 1.E0)) * st->st_connection->dpd_delay + , st); + } +} + +/* Handle (the single) message from Responder in Quick Mode. + * HDR*, HASH(2), SA, Nr [, KE ] [, IDci, IDcr ] --> + * HDR*, HASH(3) + * (see RFC 2409 "IKE" 5.5) + * Installs inbound and outbound IPsec SAs, routing, etc. + */ +stf_status +quick_inR1_outI2(struct msg_digest *md) +{ + struct state *const st = md->st; + const struct connection *c = st->st_connection; + + /* HASH(2) in */ + CHECK_QUICK_HASH(md + , quick_mode_hash12(hash_val, hash_pbs->roof, md->message_pbs.roof + , st, &st->st_msgid, TRUE) + , "HASH(2)", "Quick R1"); + + /* SA in */ + { + struct payload_digest *const sa_pd = md->chain[ISAKMP_NEXT_SA]; + + RETURN_STF_FAILURE(parse_ipsec_sa_body(&sa_pd->pbs + , &sa_pd->payload.sa, NULL, TRUE, st)); + } + + /* Nr in */ + RETURN_STF_FAILURE(accept_nonce(md, &st->st_nr, "Nr")); + + /* [ KE ] in (for PFS) */ + RETURN_STF_FAILURE(accept_PFS_KE(md, &st->st_gr, "Gr", "Quick Mode R1")); + + if (st->st_pfs_group != NULL) + compute_dh_shared(st, st->st_gr, st->st_pfs_group); + + /* [ IDci, IDcr ] in; these must match what we sent */ + + { + struct payload_digest *const id_pd = md->chain[ISAKMP_NEXT_ID]; + + if (id_pd != NULL) + { + /* ??? we are assuming IPSEC_DOI */ + + /* IDci (we are initiator) */ + + if (!check_net_id(&id_pd->payload.ipsec_id, &id_pd->pbs + , &st->st_myuserprotoid, &st->st_myuserport + , &st->st_connection->spd.this.client + , "our client")) + return STF_FAIL + INVALID_ID_INFORMATION; + + /* IDcr (responder is peer) */ + + if (!check_net_id(&id_pd->next->payload.ipsec_id, &id_pd->next->pbs + , &st->st_peeruserprotoid, &st->st_peeruserport + , &st->st_connection->spd.that.client + , "peer client")) + return STF_FAIL + INVALID_ID_INFORMATION; + } + else + { + /* no IDci, IDcr: we must check that the defaults match our proposal */ + if (!subnetisaddr(&c->spd.this.client, &c->spd.this.host_addr) + || !subnetisaddr(&c->spd.that.client, &c->spd.that.host_addr)) + { + loglog(RC_LOG_SERIOUS, "IDci, IDcr payloads missing in message" + " but default does not match proposal"); + return STF_FAIL + INVALID_ID_INFORMATION; + } + } + } + + /* check the peer's group attributes */ + + { + const ietfAttrList_t *peer_list = NULL; + + get_peer_ca_and_groups(st->st_connection, &peer_list); + + if (!group_membership(peer_list, st->st_connection->name + , st->st_connection->spd.that.groups)) + { + char buf[BUF_LEN]; + + format_groups(st->st_connection->spd.that.groups, buf, BUF_LEN); + loglog(RC_LOG_SERIOUS, "peer is not member of one of the groups: %s" + , buf); + return STF_FAIL + INVALID_ID_INFORMATION; + } + } + + if ((st->nat_traversal & NAT_T_DETECTED) + && (st->nat_traversal & NAT_T_WITH_NATOA)) + { + nat_traversal_natoa_lookup(md); + } + + /* ??? We used to copy the accepted proposal into the state, but it was + * never used. From sa_pd->pbs.start, length pbs_room(&sa_pd->pbs). + */ + + /**************** build reply packet HDR*, HASH(3) ****************/ + + /* HDR* out done */ + + /* HASH(3) out -- since this is the only content, no passes needed */ + { + u_char /* set by START_HASH_PAYLOAD: */ + *r_hashval, /* where in reply to jam hash value */ + *r_hash_start; /* start of what is to be hashed */ + + START_HASH_PAYLOAD(md->rbody, ISAKMP_NEXT_NONE); + (void)quick_mode_hash3(r_hashval, st); + } + + /* Derive new keying material */ + compute_keymats(st); + + /* Tell the kernel to establish the inbound, outbound, and routing part + * of the new SA (unless the commit bit is set -- which we don't support). + * We do this before any state updating so that + * failure won't look like success. + */ + if (!install_ipsec_sa(st, TRUE)) + return STF_INTERNAL_ERROR; + + /* encrypt message, except for fixed part of header */ + + if (!encrypt_message(&md->rbody, st)) + return STF_INTERNAL_ERROR; /* ??? we may be partly committed */ + + { + DBG(DBG_CONTROLMORE, DBG_log("inR1_outI2: instance %s[%ld], setting newest_ipsec_sa to #%ld (was #%ld) (spd.eroute=#%ld)" + , st->st_connection->name + , st->st_connection->instance_serial + , st->st_serialno + , st->st_connection->newest_ipsec_sa + , st->st_connection->spd.eroute_owner)); + } + + st->st_connection->newest_ipsec_sa = st->st_serialno; + + /* note (presumed) success */ + if (c->gw_info != NULL) + c->gw_info->key->last_worked_time = now(); + + /* If we want DPD on this connection then initialize it */ + if (st->st_connection->dpd_action != DPD_ACTION_NONE) + dpd_init(st); + + return STF_OK; +} + +/* Handle last message of Quick Mode. + * HDR*, HASH(3) -> done + * (see RFC 2409 "IKE" 5.5) + * Installs outbound IPsec SAs, routing, etc. + */ +stf_status +quick_inI2(struct msg_digest *md) +{ + struct state *const st = md->st; + + /* HASH(3) in */ + CHECK_QUICK_HASH(md, quick_mode_hash3(hash_val, st) + , "HASH(3)", "Quick I2"); + + /* Tell the kernel to establish the outbound and routing part of the new SA + * (the previous state established inbound) + * (unless the commit bit is set -- which we don't support). + * We do this before any state updating so that + * failure won't look like success. + */ + if (!install_ipsec_sa(st, FALSE)) + return STF_INTERNAL_ERROR; + + { + DBG(DBG_CONTROLMORE, DBG_log("inI2: instance %s[%ld], setting newest_ipsec_sa to #%ld (was #%ld) (spd.eroute=#%ld)" + , st->st_connection->name + , st->st_connection->instance_serial + , st->st_serialno + , st->st_connection->newest_ipsec_sa + , st->st_connection->spd.eroute_owner)); + } + + st->st_connection->newest_ipsec_sa = st->st_serialno; + + update_iv(st); /* not actually used, but tidy */ + + /* note (presumed) success */ + { + struct gw_info *gw = st->st_connection->gw_info; + + if (gw != NULL) + gw->key->last_worked_time = now(); + } + + /* If we want DPD on this connection then initialize it */ + if (st->st_connection->dpd_action != DPD_ACTION_NONE) + dpd_init(st); + + return STF_OK; +} + +static stf_status +send_isakmp_notification(struct state *st, u_int16_t type + , const void *data, size_t len) +{ + msgid_t msgid; + pb_stream reply; + pb_stream rbody; + u_char + *r_hashval, /* where in reply to jam hash value */ + *r_hash_start; /* start of what is to be hashed */ + + msgid = generate_msgid(st); + + init_pbs(&reply, reply_buffer, sizeof(reply_buffer), "ISAKMP notify"); + + /* HDR* */ + { + struct isakmp_hdr hdr; + + hdr.isa_version = ISAKMP_MAJOR_VERSION << ISA_MAJ_SHIFT | ISAKMP_MINOR_VERSION; + hdr.isa_np = ISAKMP_NEXT_HASH; + hdr.isa_xchg = ISAKMP_XCHG_INFO; + hdr.isa_msgid = msgid; + hdr.isa_flags = ISAKMP_FLAG_ENCRYPTION; + memcpy(hdr.isa_icookie, st->st_icookie, COOKIE_SIZE); + memcpy(hdr.isa_rcookie, st->st_rcookie, COOKIE_SIZE); + if (!out_struct(&hdr, &isakmp_hdr_desc, &reply, &rbody)) + impossible(); + } + /* HASH -- create and note space to be filled later */ + START_HASH_PAYLOAD(rbody, ISAKMP_NEXT_N); + + /* NOTIFY */ + { + pb_stream notify_pbs; + struct isakmp_notification isan; + + isan.isan_np = ISAKMP_NEXT_NONE; + isan.isan_doi = ISAKMP_DOI_IPSEC; + isan.isan_protoid = PROTO_ISAKMP; + isan.isan_spisize = COOKIE_SIZE * 2; + isan.isan_type = type; + if (!out_struct(&isan, &isakmp_notification_desc, &rbody, ¬ify_pbs)) + return STF_INTERNAL_ERROR; + if (!out_raw(st->st_icookie, COOKIE_SIZE, ¬ify_pbs, "notify icookie")) + return STF_INTERNAL_ERROR; + if (!out_raw(st->st_rcookie, COOKIE_SIZE, ¬ify_pbs, "notify rcookie")) + return STF_INTERNAL_ERROR; + if (data != NULL && len > 0) + if (!out_raw(data, len, ¬ify_pbs, "notify data")) + return STF_INTERNAL_ERROR; + close_output_pbs(¬ify_pbs); + } + + { + /* finish computing HASH */ + struct hmac_ctx ctx; + hmac_init_chunk(&ctx, st->st_oakley.hasher, st->st_skeyid_a); + hmac_update(&ctx, (const u_char *) &msgid, sizeof(msgid_t)); + hmac_update(&ctx, r_hash_start, rbody.cur-r_hash_start); + hmac_final(r_hashval, &ctx); + + DBG(DBG_CRYPT, + DBG_log("HASH computed:"); + DBG_dump("", r_hashval, ctx.hmac_digest_size)); + } + + /* Encrypt message (preserve st_iv and st_new_iv) */ + { + u_char old_iv[MAX_DIGEST_LEN]; + u_char new_iv[MAX_DIGEST_LEN]; + + u_int old_iv_len = st->st_iv_len; + u_int new_iv_len = st->st_new_iv_len; + + if (old_iv_len > MAX_DIGEST_LEN || new_iv_len > MAX_DIGEST_LEN) + return STF_INTERNAL_ERROR; + + memcpy(old_iv, st->st_iv, old_iv_len); + memcpy(new_iv, st->st_new_iv, new_iv_len); + + init_phase2_iv(st, &msgid); + if (!encrypt_message(&rbody, st)) + return STF_INTERNAL_ERROR; + + /* restore preserved st_iv and st_new_iv */ + memcpy(st->st_iv, old_iv, old_iv_len); + memcpy(st->st_new_iv, new_iv, new_iv_len); + st->st_iv_len = old_iv_len; + st->st_new_iv_len = new_iv_len; + } + + /* Send packet (preserve st_tpacket) */ + { + chunk_t saved_tpacket = st->st_tpacket; + + setchunk(st->st_tpacket, reply.start, pbs_offset(&reply)); + send_packet(st, "ISAKMP notify"); + st->st_tpacket = saved_tpacket; + } + + return STF_IGNORE; +} + +/* + * DPD Out Initiator + */ +void +dpd_outI(struct state *p2st) +{ + struct state *st; + u_int32_t seqno; + time_t tm; + time_t idle_time; + time_t delay = p2st->st_connection->dpd_delay; + time_t timeout = p2st->st_connection->dpd_timeout; + + /* find the newest related Phase 1 state */ + st = find_phase1_state(p2st->st_connection, ISAKMP_SA_ESTABLISHED_STATES); + + if (st == NULL) + { + loglog(RC_LOG_SERIOUS, "DPD: Could not find newest phase 1 state"); + return; + } + + /* If no DPD, then get out of here */ + if (!st->st_dpd) + return; + + /* schedule the next periodic DPD event */ + event_schedule(EVENT_DPD, delay, p2st); + + /* Current time */ + tm = now(); + + /* Make sure we really need to invoke DPD */ + if (!was_eroute_idle(p2st, delay, &idle_time)) + { + DBG(DBG_CONTROL, + DBG_log("recent eroute activity %u seconds ago, " + "no need to send DPD notification" + , (int)idle_time) + ) + st->st_last_dpd = tm; + delete_dpd_event(st); + return; + } + + /* If an R_U_THERE has been sent or received recently, or if a + * companion Phase 2 SA has shown eroute activity, + * then we don't need to invoke DPD. + */ + if (tm < st->st_last_dpd + delay) + { + DBG(DBG_CONTROL, + DBG_log("recent DPD activity %u seconds ago, " + "no need to send DPD notification" + , (int)(tm - st->st_last_dpd)) + ) + return; + } + + if (!IS_ISAKMP_SA_ESTABLISHED(st->st_state)) + return; + + if (!st->st_dpd_seqno) + { + /* Get a non-zero random value that has room to grow */ + get_rnd_bytes((u_char *)&st->st_dpd_seqno, sizeof(st->st_dpd_seqno)); + st->st_dpd_seqno &= 0x7fff; + st->st_dpd_seqno++; + } + seqno = htonl(st->st_dpd_seqno); + + if (send_isakmp_notification(st, R_U_THERE, &seqno, sizeof(seqno)) != STF_IGNORE) + { + loglog(RC_LOG_SERIOUS, "DPD: Could not send R_U_THERE"); + return; + } + DBG(DBG_CONTROL, + DBG_log("sent DPD notification R_U_THERE with seqno = %u", st->st_dpd_seqno) + ) + st->st_dpd_expectseqno = st->st_dpd_seqno++; + st->st_last_dpd = tm; + /* Only schedule a new timeout if there isn't one currently, + * or if it would be sooner than the current timeout. */ + if (st->st_dpd_event == NULL + || st->st_dpd_event->ev_time > tm + timeout) + { + delete_dpd_event(st); + event_schedule(EVENT_DPD_TIMEOUT, timeout, st); + } +} + +/* + * DPD in Initiator, out Responder + */ +stf_status +dpd_inI_outR(struct state *st, struct isakmp_notification *const n, pb_stream *pbs) +{ + time_t tm = now(); + u_int32_t seqno; + + if (!IS_ISAKMP_SA_ESTABLISHED(st->st_state)) + { + loglog(RC_LOG_SERIOUS, "DPD: Received R_U_THERE for unestablished ISKAMP SA"); + return STF_IGNORE; + } + if (n->isan_spisize != COOKIE_SIZE * 2 || pbs_left(pbs) < COOKIE_SIZE * 2) + { + loglog(RC_LOG_SERIOUS, "DPD: R_U_THERE has invalid SPI length (%d)", n->isan_spisize); + return STF_FAIL + PAYLOAD_MALFORMED; + } + + if (memcmp(pbs->cur, st->st_icookie, COOKIE_SIZE) != 0) + { +#ifdef APPLY_CRISCO + /* Ignore it, cisco sends odd icookies */ +#else + loglog(RC_LOG_SERIOUS, "DPD: R_U_THERE has invalid icookie (broken Cisco?)"); + return STF_FAIL + INVALID_COOKIE; +#endif + } + pbs->cur += COOKIE_SIZE; + + if (memcmp(pbs->cur, st->st_rcookie, COOKIE_SIZE) != 0) + { + loglog(RC_LOG_SERIOUS, "DPD: R_U_THERE has invalid rcookie (broken Cisco?)"); + return STF_FAIL + INVALID_COOKIE; + } + pbs->cur += COOKIE_SIZE; + + if (pbs_left(pbs) != sizeof(seqno)) + { + loglog(RC_LOG_SERIOUS, "DPD: R_U_THERE has invalid data length (%d)" + , (int) pbs_left(pbs)); + return STF_FAIL + PAYLOAD_MALFORMED; + } + + seqno = ntohl(*(u_int32_t *)pbs->cur); + DBG(DBG_CONTROL, + DBG_log("received DPD notification R_U_THERE with seqno = %u", seqno) + ) + + if (st->st_dpd_peerseqno && seqno <= st->st_dpd_peerseqno) { + loglog(RC_LOG_SERIOUS, "DPD: Received old or duplicate R_U_THERE"); + return STF_IGNORE; + } + + st->st_dpd_peerseqno = seqno; + delete_dpd_event(st); + + if (send_isakmp_notification(st, R_U_THERE_ACK, pbs->cur, pbs_left(pbs)) != STF_IGNORE) + { + loglog(RC_LOG_SERIOUS, "DPD Info: could not send R_U_THERE_ACK"); + return STF_IGNORE; + } + DBG(DBG_CONTROL, + DBG_log("sent DPD notification R_U_THERE_ACK with seqno = %u", seqno) + ) + + st->st_last_dpd = tm; + return STF_IGNORE; +} + +/* + * DPD out Responder + */ +stf_status +dpd_inR(struct state *st, struct isakmp_notification *const n, pb_stream *pbs) +{ + u_int32_t seqno; + + if (!IS_ISAKMP_SA_ESTABLISHED(st->st_state)) + { + loglog(RC_LOG_SERIOUS + , "DPD: Received R_U_THERE_ACK for unestablished ISKAMP SA"); + return STF_FAIL; + } + + if (n->isan_spisize != COOKIE_SIZE * 2 || pbs_left(pbs) < COOKIE_SIZE * 2) + { + loglog(RC_LOG_SERIOUS + , "DPD: R_U_THERE_ACK has invalid SPI length (%d)" + , n->isan_spisize); + return STF_FAIL + PAYLOAD_MALFORMED; + } + + if (memcmp(pbs->cur, st->st_icookie, COOKIE_SIZE) != 0) + { +#ifdef APPLY_CRISCO + /* Ignore it, cisco sends odd icookies */ +#else + loglog(RC_LOG_SERIOUS, "DPD: R_U_THERE_ACK has invalid icookie"); + return STF_FAIL + INVALID_COOKIE; +#endif + } + pbs->cur += COOKIE_SIZE; + + if (memcmp(pbs->cur, st->st_rcookie, COOKIE_SIZE) != 0) + { +#ifdef APPLY_CRISCO + /* Ignore it, cisco sends odd icookies */ +#else + loglog(RC_LOG_SERIOUS, "DPD: R_U_THERE_ACK has invalid rcookie"); + return STF_FAIL + INVALID_COOKIE; +#endif + } + pbs->cur += COOKIE_SIZE; + + if (pbs_left(pbs) != sizeof(seqno)) + { + loglog(RC_LOG_SERIOUS + , " DPD: R_U_THERE_ACK has invalid data length (%d)" + , (int) pbs_left(pbs)); + return STF_FAIL + PAYLOAD_MALFORMED; + } + + seqno = ntohl(*(u_int32_t *)pbs->cur); + DBG(DBG_CONTROL, + DBG_log("received DPD notification R_U_THERE_ACK with seqno = %u" + , seqno) + ) + + if (!st->st_dpd_expectseqno && seqno != st->st_dpd_expectseqno) + { + loglog(RC_LOG_SERIOUS + , "DPD: R_U_THERE_ACK has unexpected sequence number"); + return STF_FAIL + PAYLOAD_MALFORMED; + } + + st->st_dpd_expectseqno = 0; + delete_dpd_event(st); + return STF_IGNORE; +} + +/* + * DPD Timeout Function + * + * This function is called when a timeout DPD_EVENT occurs. We set clear/trap + * both the SA and the eroutes, depending on what the connection definition + * tells us (either 'hold' or 'clear') + */ +void +dpd_timeout(struct state *st) +{ + struct state *newest_phase1_st; + struct connection *c = st->st_connection; + int action = st->st_connection->dpd_action; + + passert(action == DPD_ACTION_HOLD + || action == DPD_ACTION_CLEAR + || DPD_ACTION_RESTART); + + /* is there a newer phase1_state? */ + newest_phase1_st = find_phase1_state(c, ISAKMP_SA_ESTABLISHED_STATES); + if (newest_phase1_st != NULL && newest_phase1_st != st) + { + plog("DPD: Phase1 state #%ld has been superseded by #%ld" + " - timeout ignored" + , st->st_serialno, newest_phase1_st->st_serialno); + return; + } + + loglog(RC_LOG_SERIOUS, "DPD: No response from peer - declaring peer dead"); + + /* delete the state, which is probably in phase 2 */ + set_cur_connection(c); + plog("DPD: Terminating all SAs using this connection"); + delete_states_by_connection(c, TRUE); + reset_cur_connection(); + + switch (action) + { + case DPD_ACTION_HOLD: + /* dpdaction=hold - Wipe the SA's but %trap the eroute so we don't + * leak traffic. Also, being in %trap means new packets will + * force an initiation of the conn again. + */ + loglog(RC_LOG_SERIOUS, "DPD: Putting connection into %%trap"); + break; + case DPD_ACTION_CLEAR: + /* dpdaction=clear - Wipe the SA & eroute - everything */ + loglog(RC_LOG_SERIOUS, "DPD: Clearing connection"); + unroute_connection(c); + break; + case DPD_ACTION_RESTART: + /* dpdaction=restart - Restart connection, + * except if roadwarrior connection + */ + loglog(RC_LOG_SERIOUS, "DPD: Restarting connection"); + unroute_connection(c); + initiate_connection(c->name, NULL_FD); + break; + default: + loglog(RC_LOG_SERIOUS, "DPD: unknown action"); + } +} + |