Print this page
Bayard's initial drop, needs finishing, or at least testing.
        
@@ -19,10 +19,11 @@
  * CDDL HEADER END
  */
 /*
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright (c) 2012 Nexenta Systems, Inc. All rights reserved.
  */
 
 #include <sys/types.h>
 #include <sys/stream.h>
 #include <sys/stropts.h>
@@ -93,35 +94,10 @@
         { 0,    0xffffffffU,    0,      "ipsecesp_default_hard_usetime"},
         { 0,    1,              0,      "ipsecesp_log_unknown_spi"},
         { 0,    2,              1,      "ipsecesp_padding_check"},
         { 0,    600,            20,     "ipsecesp_nat_keepalive_interval"},
 };
-#define ipsecesp_debug  ipsecesp_params[0].ipsecesp_param_value
-#define ipsecesp_age_interval ipsecesp_params[1].ipsecesp_param_value
-#define ipsecesp_age_int_max    ipsecesp_params[1].ipsecesp_param_max
-#define ipsecesp_reap_delay     ipsecesp_params[2].ipsecesp_param_value
-#define ipsecesp_replay_size    ipsecesp_params[3].ipsecesp_param_value
-#define ipsecesp_acquire_timeout        \
-        ipsecesp_params[4].ipsecesp_param_value
-#define ipsecesp_larval_timeout \
-        ipsecesp_params[5].ipsecesp_param_value
-#define ipsecesp_default_soft_bytes     \
-        ipsecesp_params[6].ipsecesp_param_value
-#define ipsecesp_default_hard_bytes     \
-        ipsecesp_params[7].ipsecesp_param_value
-#define ipsecesp_default_soft_addtime   \
-        ipsecesp_params[8].ipsecesp_param_value
-#define ipsecesp_default_hard_addtime   \
-        ipsecesp_params[9].ipsecesp_param_value
-#define ipsecesp_default_soft_usetime   \
-        ipsecesp_params[10].ipsecesp_param_value
-#define ipsecesp_default_hard_usetime   \
-        ipsecesp_params[11].ipsecesp_param_value
-#define ipsecesp_log_unknown_spi        \
-        ipsecesp_params[12].ipsecesp_param_value
-#define ipsecesp_padding_check  \
-        ipsecesp_params[13].ipsecesp_param_value
 /* For ipsecesp_nat_keepalive_interval, see ipsecesp.h. */
 
 #define esp0dbg(a)      printf a
 /* NOTE:  != 0 instead of > 0 so lint doesn't complain. */
 #define esp1dbg(espstack, a)    if (espstack->ipsecesp_debug != 0) printf a
@@ -131,11 +107,10 @@
 static int ipsecesp_open(queue_t *, dev_t *, int, int, cred_t *);
 static int ipsecesp_close(queue_t *);
 static void ipsecesp_wput(queue_t *, mblk_t *);
 static void     *ipsecesp_stack_init(netstackid_t stackid, netstack_t *ns);
 static void     ipsecesp_stack_fini(netstackid_t stackid, void *arg);
-static void esp_send_acquire(ipsacq_t *, mblk_t *, netstack_t *);
 
 static void esp_prepare_udp(netstack_t *, mblk_t *, ipha_t *);
 static void esp_outbound_finish(mblk_t *, ip_xmit_attr_t *);
 static void esp_inbound_restart(mblk_t *, ip_recv_attr_t *);
 
@@ -179,56 +154,10 @@
  *
  * Answer:      Yes, because I need to know which queue is BOUND to
  *              IPPROTO_ESP
  */
 
-/*
- * Stats.  This may eventually become a full-blown SNMP MIB once that spec
- * stabilizes.
- */
-
-typedef struct esp_kstats_s {
-        kstat_named_t esp_stat_num_aalgs;
-        kstat_named_t esp_stat_good_auth;
-        kstat_named_t esp_stat_bad_auth;
-        kstat_named_t esp_stat_bad_padding;
-        kstat_named_t esp_stat_replay_failures;
-        kstat_named_t esp_stat_replay_early_failures;
-        kstat_named_t esp_stat_keysock_in;
-        kstat_named_t esp_stat_out_requests;
-        kstat_named_t esp_stat_acquire_requests;
-        kstat_named_t esp_stat_bytes_expired;
-        kstat_named_t esp_stat_out_discards;
-        kstat_named_t esp_stat_crypto_sync;
-        kstat_named_t esp_stat_crypto_async;
-        kstat_named_t esp_stat_crypto_failures;
-        kstat_named_t esp_stat_num_ealgs;
-        kstat_named_t esp_stat_bad_decrypt;
-        kstat_named_t esp_stat_sa_port_renumbers;
-} esp_kstats_t;
-
-/*
- * espstack->esp_kstats is equal to espstack->esp_ksp->ks_data if
- * kstat_create_netstack for espstack->esp_ksp succeeds, but when it
- * fails, it will be NULL. Note this is done for all stack instances,
- * so it *could* fail. hence a non-NULL checking is done for
- * ESP_BUMP_STAT and ESP_DEBUMP_STAT
- */
-#define ESP_BUMP_STAT(espstack, x)                                      \
-do {                                                                    \
-        if (espstack->esp_kstats != NULL)                               \
-                (espstack->esp_kstats->esp_stat_ ## x).value.ui64++;    \
-_NOTE(CONSTCOND)                                                        \
-} while (0)
-
-#define ESP_DEBUMP_STAT(espstack, x)                                    \
-do {                                                                    \
-        if (espstack->esp_kstats != NULL)                               \
-                (espstack->esp_kstats->esp_stat_ ## x).value.ui64--;    \
-_NOTE(CONSTCOND)                                                        \
-} while (0)
-
 static int      esp_kstat_update(kstat_t *, int);
 
 static boolean_t
 esp_kstat_init(ipsecesp_stack_t *espstack, netstackid_t stackid)
 {
@@ -296,16 +225,16 @@
                 netstack_rele(ns);
                 return (-1);
         }
         ekp = (esp_kstats_t *)kp->ks_data;
 
-        mutex_enter(&ipss->ipsec_alg_lock);
+        rw_enter(&ipss->ipsec_alg_lock, RW_READER);
         ekp->esp_stat_num_aalgs.value.ui64 =
             ipss->ipsec_nalgs[IPSEC_ALG_AUTH];
         ekp->esp_stat_num_ealgs.value.ui64 =
             ipss->ipsec_nalgs[IPSEC_ALG_ENCR];
-        mutex_exit(&ipss->ipsec_alg_lock);
+        rw_exit(&ipss->ipsec_alg_lock);
 
         netstack_rele(ns);
         return (0);
 }
 
@@ -501,10 +430,11 @@
                         }
                 }
         }
         return (B_TRUE);
 }
+
 /*
  * Initialize things for ESP for each stack instance
  */
 static void *
 ipsecesp_stack_init(netstackid_t stackid, netstack_t *ns)
@@ -525,11 +455,10 @@
 
         (void) esp_kstat_init(espstack, stackid);
 
         espstack->esp_sadb.s_acquire_timeout =
             &espstack->ipsecesp_acquire_timeout;
-        espstack->esp_sadb.s_acqfn = esp_send_acquire;
         sadbp_init("ESP", &espstack->esp_sadb, SADB_SATYPE_ESP, esp_hash_size,
             espstack->ipsecesp_netstack);
 
         mutex_init(&espstack->ipsecesp_param_lock, NULL, MUTEX_DEFAULT, 0);
 
@@ -556,11 +485,10 @@
         ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)arg;
 
         if (espstack->esp_pfkey_q != NULL) {
                 (void) quntimeout(espstack->esp_pfkey_q, espstack->esp_event);
         }
-        espstack->esp_sadb.s_acqfn = NULL;
         espstack->esp_sadb.s_acquire_timeout = NULL;
         sadbp_destroy(&espstack->esp_sadb, espstack->ipsecesp_netstack);
         ip_drop_unregister(&espstack->esp_dropper);
         mutex_destroy(&espstack->ipsecesp_param_lock);
         nd_free(&espstack->ipsecesp_g_nd);
@@ -1179,183 +1107,10 @@
         /* submit the request to the crypto framework */
         return (esp_submit_req_inbound(data_mp, ira, ipsa,
             (uint8_t *)esph - data_mp->b_rptr));
 }
 
-/*
- * Perform the really difficult work of inserting the proposed situation.
- * Called while holding the algorithm lock.
- */
-static void
-esp_insert_prop(sadb_prop_t *prop, ipsacq_t *acqrec, uint_t combs,
-    netstack_t *ns)
-{
-        sadb_comb_t *comb = (sadb_comb_t *)(prop + 1);
-        ipsec_action_t *ap;
-        ipsec_prot_t *prot;
-        ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
-        ipsec_stack_t   *ipss = ns->netstack_ipsec;
-
-        ASSERT(MUTEX_HELD(&ipss->ipsec_alg_lock));
-
-        prop->sadb_prop_exttype = SADB_EXT_PROPOSAL;
-        prop->sadb_prop_len = SADB_8TO64(sizeof (sadb_prop_t));
-        *(uint32_t *)(&prop->sadb_prop_replay) = 0;     /* Quick zero-out! */
-
-        prop->sadb_prop_replay = espstack->ipsecesp_replay_size;
-
-        /*
-         * Based upon algorithm properties, and what-not, prioritize a
-         * proposal, based on the ordering of the ESP algorithms in the
-         * alternatives in the policy rule or socket that was placed
-         * in the acquire record.
-         *
-         * For each action in policy list
-         *   Add combination.  If I've hit limit, return.
-         */
-
-        for (ap = acqrec->ipsacq_act; ap != NULL;
-            ap = ap->ipa_next) {
-                ipsec_alginfo_t *ealg = NULL;
-                ipsec_alginfo_t *aalg = NULL;
-
-                if (ap->ipa_act.ipa_type != IPSEC_POLICY_APPLY)
-                        continue;
-
-                prot = &ap->ipa_act.ipa_apply;
-
-                if (!(prot->ipp_use_esp))
-                        continue;
-
-                if (prot->ipp_esp_auth_alg != 0) {
-                        aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
-                            [prot->ipp_esp_auth_alg];
-                        if (aalg == NULL || !ALG_VALID(aalg))
-                                continue;
-                }
-
-                ASSERT(prot->ipp_encr_alg > 0);
-                ealg = ipss->ipsec_alglists[IPSEC_ALG_ENCR]
-                    [prot->ipp_encr_alg];
-                if (ealg == NULL || !ALG_VALID(ealg))
-                        continue;
-
-                comb->sadb_comb_flags = 0;
-                comb->sadb_comb_reserved = 0;
-                comb->sadb_comb_encrypt = ealg->alg_id;
-                comb->sadb_comb_encrypt_minbits =
-                    MAX(prot->ipp_espe_minbits, ealg->alg_ef_minbits);
-                comb->sadb_comb_encrypt_maxbits =
-                    MIN(prot->ipp_espe_maxbits, ealg->alg_ef_maxbits);
-
-                if (aalg == NULL) {
-                        comb->sadb_comb_auth = 0;
-                        comb->sadb_comb_auth_minbits = 0;
-                        comb->sadb_comb_auth_maxbits = 0;
-                } else {
-                        comb->sadb_comb_auth = aalg->alg_id;
-                        comb->sadb_comb_auth_minbits =
-                            MAX(prot->ipp_espa_minbits, aalg->alg_ef_minbits);
-                        comb->sadb_comb_auth_maxbits =
-                            MIN(prot->ipp_espa_maxbits, aalg->alg_ef_maxbits);
-                }
-
-                /*
-                 * The following may be based on algorithm
-                 * properties, but in the meantime, we just pick
-                 * some good, sensible numbers.  Key mgmt. can
-                 * (and perhaps should) be the place to finalize
-                 * such decisions.
-                 */
-
-                /*
-                 * No limits on allocations, since we really don't
-                 * support that concept currently.
-                 */
-                comb->sadb_comb_soft_allocations = 0;
-                comb->sadb_comb_hard_allocations = 0;
-
-                /*
-                 * These may want to come from policy rule..
-                 */
-                comb->sadb_comb_soft_bytes =
-                    espstack->ipsecesp_default_soft_bytes;
-                comb->sadb_comb_hard_bytes =
-                    espstack->ipsecesp_default_hard_bytes;
-                comb->sadb_comb_soft_addtime =
-                    espstack->ipsecesp_default_soft_addtime;
-                comb->sadb_comb_hard_addtime =
-                    espstack->ipsecesp_default_hard_addtime;
-                comb->sadb_comb_soft_usetime =
-                    espstack->ipsecesp_default_soft_usetime;
-                comb->sadb_comb_hard_usetime =
-                    espstack->ipsecesp_default_hard_usetime;
-
-                prop->sadb_prop_len += SADB_8TO64(sizeof (*comb));
-                if (--combs == 0)
-                        break;  /* out of space.. */
-                comb++;
-        }
-}
-
-/*
- * Prepare and actually send the SADB_ACQUIRE message to PF_KEY.
- */
-static void
-esp_send_acquire(ipsacq_t *acqrec, mblk_t *extended, netstack_t *ns)
-{
-        uint_t combs;
-        sadb_msg_t *samsg;
-        sadb_prop_t *prop;
-        mblk_t *pfkeymp, *msgmp;
-        ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
-        ipsec_stack_t   *ipss = ns->netstack_ipsec;
-
-        ESP_BUMP_STAT(espstack, acquire_requests);
-
-        if (espstack->esp_pfkey_q == NULL) {
-                mutex_exit(&acqrec->ipsacq_lock);
-                return;
-        }
-
-        /* Set up ACQUIRE. */
-        pfkeymp = sadb_setup_acquire(acqrec, SADB_SATYPE_ESP,
-            ns->netstack_ipsec);
-        if (pfkeymp == NULL) {
-                esp0dbg(("sadb_setup_acquire failed.\n"));
-                mutex_exit(&acqrec->ipsacq_lock);
-                return;
-        }
-        ASSERT(MUTEX_HELD(&ipss->ipsec_alg_lock));
-        combs = ipss->ipsec_nalgs[IPSEC_ALG_AUTH] *
-            ipss->ipsec_nalgs[IPSEC_ALG_ENCR];
-        msgmp = pfkeymp->b_cont;
-        samsg = (sadb_msg_t *)(msgmp->b_rptr);
-
-        /* Insert proposal here. */
-
-        prop = (sadb_prop_t *)(((uint64_t *)samsg) + samsg->sadb_msg_len);
-        esp_insert_prop(prop, acqrec, combs, ns);
-        samsg->sadb_msg_len += prop->sadb_prop_len;
-        msgmp->b_wptr += SADB_64TO8(samsg->sadb_msg_len);
-
-        mutex_exit(&ipss->ipsec_alg_lock);
-
-        /*
-         * Must mutex_exit() before sending PF_KEY message up, in
-         * order to avoid recursive mutex_enter() if there are no registered
-         * listeners.
-         *
-         * Once I've sent the message, I'm cool anyway.
-         */
-        mutex_exit(&acqrec->ipsacq_lock);
-        if (extended != NULL) {
-                putnext(espstack->esp_pfkey_q, extended);
-        }
-        putnext(espstack->esp_pfkey_q, pfkeymp);
-}
-
 /* XXX refactor me */
 /*
  * Handle the SADB_GETSPI message.  Create a larval SA.
  */
 static void
@@ -3041,11 +2796,11 @@
 
         /*
          * Allocate the PF_KEY message that follows KEYSOCK_OUT.
          */
 
-        mutex_enter(&ipss->ipsec_alg_lock);
+        rw_enter(&ipss->ipsec_alg_lock, RW_READER);
         /*
          * Fill SADB_REGISTER message's algorithm descriptors.  Hold
          * down the lock while filling it.
          *
          * Return only valid algorithms, so the number of algorithms
@@ -3070,11 +2825,11 @@
                 allocsize += (num_ealgs * sizeof (*saalg));
                 allocsize += sizeof (*sasupp_encr);
         }
         keysock_out_mp->b_cont = allocb(allocsize, BPRI_HI);
         if (keysock_out_mp->b_cont == NULL) {
-                mutex_exit(&ipss->ipsec_alg_lock);
+                rw_exit(&ipss->ipsec_alg_lock);
                 freemsg(keysock_out_mp);
                 return (B_FALSE);
         }
         pfkey_msg_mp = keysock_out_mp->b_cont;
         pfkey_msg_mp->b_wptr += allocsize;
@@ -3164,11 +2919,11 @@
         }
 
         current_aalgs = num_aalgs;
         current_ealgs = num_ealgs;
 
-        mutex_exit(&ipss->ipsec_alg_lock);
+        rw_exit(&ipss->ipsec_alg_lock);
 
         if (sens_tsl != NULL) {
                 sens = (sadb_sens_t *)nextext;
                 sadb_sens_from_label(sens, SADB_EXT_SENSITIVITY,
                     sens_tsl, sens_len);
@@ -3689,11 +3444,11 @@
          * XXX Policy :  I'm not checking identities at this time,
          * but if I did, I'd do them here, before I sent
          * the weak key check up to the algorithm.
          */
 
-        mutex_enter(&ipss->ipsec_alg_lock);
+        rw_enter(&ipss->ipsec_alg_lock, RW_READER);
 
         /*
          * First locate the authentication algorithm.
          */
 #ifdef IPSEC_LATENCY_TEST
@@ -3704,11 +3459,11 @@
                 ipsec_alginfo_t *aalg;
 
                 aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
                     [assoc->sadb_sa_auth];
                 if (aalg == NULL || !ALG_VALID(aalg)) {
-                        mutex_exit(&ipss->ipsec_alg_lock);
+                        rw_exit(&ipss->ipsec_alg_lock);
                         esp1dbg(espstack, ("Couldn't find auth alg #%d.\n",
                             assoc->sadb_sa_auth));
                         *diagnostic = SADB_X_DIAGNOSTIC_BAD_AALG;
                         return (EINVAL);
                 }
@@ -3719,20 +3474,20 @@
                  * this auth_alg is not defined with ALG_FLAG_VALID. If this
                  * ever changes, the same check for SADB_AALG_NONE and
                  * a auth_key != NULL should be made here ( see below).
                  */
                 if (!ipsec_valid_key_size(akey->sadb_key_bits, aalg)) {
-                        mutex_exit(&ipss->ipsec_alg_lock);
+                        rw_exit(&ipss->ipsec_alg_lock);
                         *diagnostic = SADB_X_DIAGNOSTIC_BAD_AKEYBITS;
                         return (EINVAL);
                 }
                 ASSERT(aalg->alg_mech_type != CRYPTO_MECHANISM_INVALID);
 
                 /* check key and fix parity if needed */
                 if (ipsec_check_key(aalg->alg_mech_type, akey, B_TRUE,
                     diagnostic) != 0) {
-                        mutex_exit(&ipss->ipsec_alg_lock);
+                        rw_exit(&ipss->ipsec_alg_lock);
                         return (EINVAL);
                 }
         }
 
         /*
@@ -3743,11 +3498,11 @@
                 ipsec_alginfo_t *ealg;
 
                 ealg = ipss->ipsec_alglists[IPSEC_ALG_ENCR]
                     [assoc->sadb_sa_encrypt];
                 if (ealg == NULL || !ALG_VALID(ealg)) {
-                        mutex_exit(&ipss->ipsec_alg_lock);
+                        rw_exit(&ipss->ipsec_alg_lock);
                         esp1dbg(espstack, ("Couldn't find encr alg #%d.\n",
                             assoc->sadb_sa_encrypt));
                         *diagnostic = SADB_X_DIAGNOSTIC_BAD_EALG;
                         return (EINVAL);
                 }
@@ -3764,24 +3519,24 @@
                 keybits = ekey->sadb_key_bits;
                 keybits -= ekey->sadb_key_reserved;
                 keybits -= SADB_8TO1(ealg->alg_saltlen);
                 if ((assoc->sadb_sa_encrypt == SADB_EALG_NULL) ||
                     (!ipsec_valid_key_size(keybits, ealg))) {
-                        mutex_exit(&ipss->ipsec_alg_lock);
+                        rw_exit(&ipss->ipsec_alg_lock);
                         *diagnostic = SADB_X_DIAGNOSTIC_BAD_EKEYBITS;
                         return (EINVAL);
                 }
                 ASSERT(ealg->alg_mech_type != CRYPTO_MECHANISM_INVALID);
 
                 /* check key */
                 if (ipsec_check_key(ealg->alg_mech_type, ekey, B_FALSE,
                     diagnostic) != 0) {
-                        mutex_exit(&ipss->ipsec_alg_lock);
+                        rw_exit(&ipss->ipsec_alg_lock);
                         return (EINVAL);
                 }
         }
-        mutex_exit(&ipss->ipsec_alg_lock);
+        rw_exit(&ipss->ipsec_alg_lock);
 
         return (esp_add_sa_finish(mp, (sadb_msg_t *)mp->b_cont->b_rptr, ksi,
             diagnostic, espstack));
 }