static bool
npf_log(npf_cache_t *npc, nbuf_t *nbuf, void *meta, int *decision)
{
struct mbuf *m = nbuf_head_mbuf(nbuf);
const npf_ext_log_t *log = meta;
ifnet_t *ifp;
int family;
/* Set the address family. */
if (npf_iscached(npc, NPC_IP4)) {
family = AF_INET;
} else if (npf_iscached(npc, NPC_IP6)) {
family = AF_INET6;
} else {
family = AF_UNSPEC;
}
KERNEL_LOCK(1, NULL);
/* Find a pseudo-interface to log. */
ifp = if_byindex(log->if_idx);
if (ifp == NULL) {
/* No interface. */
KERNEL_UNLOCK_ONE(NULL);
return true;
}
/* Pass through BPF. */
ifp->if_opackets++;
ifp->if_obytes += m->m_pkthdr.len;
bpf_mtap_af(ifp, family, m);
KERNEL_UNLOCK_ONE(NULL);
return true;
}
static void
mbuf_consistency_check(nbuf_t *nbuf)
{
struct mbuf *m = nbuf_head_mbuf(nbuf);
while (m) {
assert(m->m_type != MT_FREE);
m = m->m_next;
}
}
static int
npf_reassembly(npf_t *npf, npf_cache_t *npc, struct mbuf **mp)
{
nbuf_t *nbuf = npc->npc_nbuf;
int error = EINVAL;
/* Reset the mbuf as it may have changed. */
*mp = nbuf_head_mbuf(nbuf);
nbuf_reset(nbuf);
if (npf_iscached(npc, NPC_IP4)) {
struct ip *ip = nbuf_dataptr(nbuf);
error = ip_reass_packet(mp, ip);
} else if (npf_iscached(npc, NPC_IP6)) {
/*
* Note: ip6_reass_packet() offset is the start of
* the fragment header.
*/
error = ip6_reass_packet(mp, npc->npc_hlen);
if (error && *mp == NULL) {
memset(nbuf, 0, sizeof(nbuf_t));
}
}
if (error) {
npf_stats_inc(npf, NPF_STAT_REASSFAIL);
return error;
}
if (*mp == NULL) {
/* More fragments should come. */
npf_stats_inc(npf, NPF_STAT_FRAGMENTS);
return 0;
}
/*
* Reassembly is complete, we have the final packet.
* Cache again, since layer 4 data is accessible now.
*/
nbuf_init(npf, nbuf, *mp, nbuf->nb_ifp);
npc->npc_info = 0;
if (npf_cache_all(npc) & NPC_IPFRAG) {
return EINVAL;
}
npf_stats_inc(npf, NPF_STAT_REASSEMBLY);
return 0;
}
/*
* npf_packet_handler: main packet handling routine for layer 3.
*
* Note: packet flow and inspection logic is in strict order.
*/
int
npf_packet_handler(void *arg, struct mbuf **mp, ifnet_t *ifp, int di)
{
nbuf_t nbuf;
npf_cache_t npc;
npf_session_t *se;
npf_rule_t *rl;
npf_rproc_t *rp;
int error, retfl;
int decision;
/*
* Initialise packet information cache.
* Note: it is enough to clear the info bits.
*/
KASSERT(ifp != NULL);
nbuf_init(&nbuf, *mp, ifp);
npc.npc_info = 0;
decision = NPF_DECISION_BLOCK;
error = 0;
retfl = 0;
rp = NULL;
/* Cache everything. Determine whether it is an IP fragment. */
if (npf_cache_all(&npc, &nbuf) & NPC_IPFRAG) {
/*
* Pass to IPv4 or IPv6 reassembly mechanism.
*/
error = npf_reassembly(&npc, &nbuf, mp);
if (error) {
se = NULL;
goto out;
}
if (*mp == NULL) {
/* More fragments should come; return. */
return 0;
}
}
/* Inspect the list of sessions (if found, acquires a reference). */
se = npf_session_inspect(&npc, &nbuf, di, &error);
/* If "passing" session found - skip the ruleset inspection. */
if (se && npf_session_pass(se, &rp)) {
npf_stats_inc(NPF_STAT_PASS_SESSION);
KASSERT(error == 0);
goto pass;
}
if (error) {
if (error == ENETUNREACH)
goto block;
goto out;
}
/* Acquire the lock, inspect the ruleset using this packet. */
int slock = npf_config_read_enter();
npf_ruleset_t *rlset = npf_config_ruleset();
rl = npf_ruleset_inspect(&npc, &nbuf, rlset, di, NPF_LAYER_3);
if (rl == NULL) {
const bool pass = npf_default_pass();
npf_config_read_exit(slock);
if (pass) {
npf_stats_inc(NPF_STAT_PASS_DEFAULT);
goto pass;
}
npf_stats_inc(NPF_STAT_BLOCK_DEFAULT);
goto block;
}
/*
* Get the rule procedure (acquires a reference) for association
* with a session (if any) and execution.
*/
KASSERT(rp == NULL);
rp = npf_rule_getrproc(rl);
/* Conclude with the rule and release the lock. */
error = npf_rule_conclude(rl, &retfl);
npf_config_read_exit(slock);
if (error) {
npf_stats_inc(NPF_STAT_BLOCK_RULESET);
goto block;
}
npf_stats_inc(NPF_STAT_PASS_RULESET);
/*
* Establish a "pass" session, if required. Just proceed,
* if session creation fails (e.g. due to unsupported protocol).
*/
if ((retfl & NPF_RULE_STATEFUL) != 0 && !se) {
se = npf_session_establish(&npc, &nbuf, di,
(retfl & NPF_RULE_MULTIENDS) == 0);
if (se) {
/*
* Note: the reference on the rule procedure is
* transfered to the session. It will be released
* on session destruction.
*/
npf_session_setpass(se, rp);
}
}
pass:
decision = NPF_DECISION_PASS;
KASSERT(error == 0);
/*
* Perform NAT.
*/
error = npf_do_nat(&npc, se, &nbuf, di);
block:
/*
* Execute the rule procedure, if any is associated.
* It may reverse the decision from pass to block.
*/
if (rp && !npf_rproc_run(&npc, &nbuf, rp, &decision)) {
if (se) {
npf_session_release(se);
}
npf_rproc_release(rp);
*mp = NULL;
return 0;
}
out:
/*
* Release the reference on a session. Release the reference on a
* rule procedure only if there was no association.
*/
if (se) {
npf_session_release(se);
} else if (rp) {
npf_rproc_release(rp);
}
/* Reset mbuf pointer before returning to the caller. */
if ((*mp = nbuf_head_mbuf(&nbuf)) == NULL) {
return error ? error : ENOMEM;
}
/* Pass the packet if decided and there is no error. */
if (decision == NPF_DECISION_PASS && !error) {
/*
* XXX: Disable for now, it will be set accordingly later,
* for optimisations (to reduce inspection).
*/
(*mp)->m_flags &= ~M_CANFASTFWD;
return 0;
}
/*
* Block the packet. ENETUNREACH is used to indicate blocking.
* Depending on the flags and protocol, return TCP reset (RST) or
* ICMP destination unreachable.
*/
if (retfl && npf_return_block(&npc, &nbuf, retfl)) {
*mp = NULL;
}
if (!error) {
error = ENETUNREACH;
}
if (*mp) {
m_freem(*mp);
*mp = NULL;
}
return error;
}
