Print this page
3914 ill_frag_hash_tbl not allocated for loopback interfaces
Reviewed by: Sebastien Roy <sebastien.roy@delphix.com>

@@ -20,10 +20,13 @@
  */
 /*
  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 1990 Mentat Inc.
  */
+/*
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ */
 
 /*
  * This file contains the interface control functions for IP.
  */
 

@@ -220,10 +223,12 @@
 static  void    ipif_trace_cleanup(const ipif_t *);
 #endif
 
 static  void    ill_dlpi_clear_deferred(ill_t *ill);
 
+static  void    phyint_flags_init(phyint_t *, t_uscalar_t);
+
 /*
  * if we go over the memory footprint limit more than once in this msec
  * interval, we'll start pruning aggressively.
  */
 int ip_min_frag_prune_time = 0;

@@ -278,11 +283,10 @@
         { DL_OTHER, IFT_OTHER, ETHERTYPE_IP, ETHERTYPE_IPV6,
             ip_ether_v4_mapping, ip_ether_v6_mapping, ip_nodef_v6intfid,
             ip_nodef_v6intfid }
 };
 
-static ill_t    ill_null;               /* Empty ILL for init. */
 char    ipif_loopback_name[] = "lo0";
 
 /* These are used by all IP network modules. */
 sin6_t  sin6_null;      /* Zero address for quick clears */
 sin_t   sin_null;       /* Zero address for quick clears */

@@ -3327,54 +3331,46 @@
         }
         return (B_TRUE);
 }
 
 /*
- * ill_init is called by ip_open when a device control stream is opened.
- * It does a few initializations, and shoots a DL_INFO_REQ message down
- * to the driver.  The response is later picked up in ip_rput_dlpi and
- * used to set up default mechanisms for talking to the driver.  (Always
- * called as writer.)
- *
- * If this function returns error, ip_open will call ip_close which in
- * turn will call ill_delete to clean up any memory allocated here that
- * is not yet freed.
+ * Here we perform initialisation of the ill_t common to both regular
+ * interface ILLs and the special loopback ILL created by ill_lookup_on_name.
  */
-int
-ill_init(queue_t *q, ill_t *ill)
+static int
+ill_init_common(ill_t *ill, queue_t *q, boolean_t isv6, boolean_t is_loopback,
+    boolean_t ipsq_enter)
 {
         int     count;
-        dl_info_req_t   *dlir;
-        mblk_t  *info_mp;
         uchar_t *frag_ptr;
 
-        /*
-         * The ill is initialized to zero by mi_alloc*(). In addition
-         * some fields already contain valid values, initialized in
-         * ip_open(), before we reach here.
-         */
         mutex_init(&ill->ill_lock, NULL, MUTEX_DEFAULT, 0);
         mutex_init(&ill->ill_saved_ire_lock, NULL, MUTEX_DEFAULT, NULL);
         ill->ill_saved_ire_cnt = 0;
 
+        if (is_loopback) {
+                ill->ill_max_frag = isv6 ? ip_loopback_mtu_v6plus :
+                    ip_loopback_mtuplus;
+                /*
+                 * No resolver here.
+                 */
+                ill->ill_net_type = IRE_LOOPBACK;
+        } else {
         ill->ill_rq = q;
         ill->ill_wq = WR(q);
+                ill->ill_ppa = UINT_MAX;
+        }
 
-        info_mp = allocb(MAX(sizeof (dl_info_req_t), sizeof (dl_info_ack_t)),
-            BPRI_HI);
-        if (info_mp == NULL)
-                return (ENOMEM);
+        ill->ill_isv6 = isv6;
 
         /*
          * Allocate sufficient space to contain our fragment hash table and
          * the device name.
          */
         frag_ptr = (uchar_t *)mi_zalloc(ILL_FRAG_HASH_TBL_SIZE + 2 * LIFNAMSIZ);
-        if (frag_ptr == NULL) {
-                freemsg(info_mp);
+        if (frag_ptr == NULL)
                 return (ENOMEM);
-        }
         ill->ill_frag_ptr = frag_ptr;
         ill->ill_frag_free_num_pkts = 0;
         ill->ill_last_frag_clean_time = 0;
         ill->ill_frag_hash_tbl = (ipfb_t *)frag_ptr;
         ill->ill_name = (char *)(frag_ptr + ILL_FRAG_HASH_TBL_SIZE);

@@ -3383,39 +3379,34 @@
                     NULL, MUTEX_DEFAULT, NULL);
         }
 
         ill->ill_phyint = (phyint_t *)mi_zalloc(sizeof (phyint_t));
         if (ill->ill_phyint == NULL) {
-                freemsg(info_mp);
                 mi_free(frag_ptr);
                 return (ENOMEM);
         }
 
         mutex_init(&ill->ill_phyint->phyint_lock, NULL, MUTEX_DEFAULT, 0);
-        /*
-         * For now pretend this is a v4 ill. We need to set phyint_ill*
-         * at this point because of the following reason. If we can't
-         * enter the ipsq at some point and cv_wait, the writer that
-         * wakes us up tries to locate us using the list of all phyints
-         * in an ipsq and the ills from the phyint thru the phyint_ill*.
-         * If we don't set it now, we risk a missed wakeup.
-         */
+        if (isv6) {
+                ill->ill_phyint->phyint_illv6 = ill;
+        } else {
         ill->ill_phyint->phyint_illv4 = ill;
-        ill->ill_ppa = UINT_MAX;
+        }
+        if (is_loopback) {
+                phyint_flags_init(ill->ill_phyint, DL_LOOP);
+        }
+
         list_create(&ill->ill_nce, sizeof (nce_t), offsetof(nce_t, nce_node));
 
         ill_set_inputfn(ill);
 
-        if (!ipsq_init(ill, B_TRUE)) {
-                freemsg(info_mp);
+        if (!ipsq_init(ill, ipsq_enter)) {
                 mi_free(frag_ptr);
                 mi_free(ill->ill_phyint);
                 return (ENOMEM);
         }
 
-        ill->ill_state_flags |= ILL_LL_SUBNET_PENDING;
-
         /* Frag queue limit stuff */
         ill->ill_frag_count = 0;
         ill->ill_ipf_gen = 0;
 
         rw_init(&ill->ill_mcast_lock, NULL, RW_DEFAULT, NULL);

@@ -3436,10 +3427,57 @@
         ill->ill_reachable_time = ND_REACHABLE_TIME;
         ill->ill_xmit_count = ND_MAX_MULTICAST_SOLICIT;
         ill->ill_max_buf = ND_MAX_Q;
         ill->ill_refcnt = 0;
 
+        return (0);
+}
+
+/*
+ * ill_init is called by ip_open when a device control stream is opened.
+ * It does a few initializations, and shoots a DL_INFO_REQ message down
+ * to the driver.  The response is later picked up in ip_rput_dlpi and
+ * used to set up default mechanisms for talking to the driver.  (Always
+ * called as writer.)
+ *
+ * If this function returns error, ip_open will call ip_close which in
+ * turn will call ill_delete to clean up any memory allocated here that
+ * is not yet freed.
+ *
+ * Note: ill_ipst and ill_zoneid must be set before calling ill_init.
+ */
+int
+ill_init(queue_t *q, ill_t *ill)
+{
+        int ret;
+        dl_info_req_t   *dlir;
+        mblk_t  *info_mp;
+
+        info_mp = allocb(MAX(sizeof (dl_info_req_t), sizeof (dl_info_ack_t)),
+            BPRI_HI);
+        if (info_mp == NULL)
+                return (ENOMEM);
+
+        /*
+         * The ill is initialized to zero by mi_alloc*(). In addition
+         * some fields already contain valid values, initialized in
+         * ip_open(), before we reach here.
+         *
+         * For now pretend this is a v4 ill. We need to set phyint_ill*
+         * at this point because of the following reason. If we can't
+         * enter the ipsq at some point and cv_wait, the writer that
+         * wakes us up tries to locate us using the list of all phyints
+         * in an ipsq and the ills from the phyint thru the phyint_ill*.
+         * If we don't set it now, we risk a missed wakeup.
+         */
+        if ((ret = ill_init_common(ill, q, B_FALSE, B_FALSE, B_TRUE)) != 0) {
+                freemsg(info_mp);
+                return (ret);
+        }
+
+        ill->ill_state_flags |= ILL_LL_SUBNET_PENDING;
+
         /* Send down the Info Request to the driver. */
         info_mp->b_datap->db_type = M_PCPROTO;
         dlir = (dl_info_req_t *)info_mp->b_rptr;
         info_mp->b_wptr = (uchar_t *)&dlir[1];
         dlir->dl_primitive = DL_INFO_REQ;

@@ -3683,40 +3721,25 @@
         ill = (ill_t *)(mi_alloc(sizeof (ill_t) +
             sizeof (ipif_loopback_name), BPRI_MED));
         if (ill == NULL)
                 goto done;
 
-        *ill = ill_null;
-        mutex_init(&ill->ill_lock, NULL, MUTEX_DEFAULT, NULL);
+        bzero(ill, sizeof (*ill));
         ill->ill_ipst = ipst;
-        list_create(&ill->ill_nce, sizeof (nce_t), offsetof(nce_t, nce_node));
         netstack_hold(ipst->ips_netstack);
         /*
          * For exclusive stacks we set the zoneid to zero
          * to make IP operate as if in the global zone.
          */
         ill->ill_zoneid = GLOBAL_ZONEID;
 
-        ill->ill_phyint = (phyint_t *)mi_zalloc(sizeof (phyint_t));
-        if (ill->ill_phyint == NULL)
+        if (ill_init_common(ill, NULL, isv6, B_TRUE, B_FALSE) != 0)
                 goto done;
 
-        if (isv6)
-                ill->ill_phyint->phyint_illv6 = ill;
-        else
-                ill->ill_phyint->phyint_illv4 = ill;
-        mutex_init(&ill->ill_phyint->phyint_lock, NULL, MUTEX_DEFAULT, 0);
-        phyint_flags_init(ill->ill_phyint, DL_LOOP);
-
-        if (isv6) {
-                ill->ill_isv6 = B_TRUE;
-                ill->ill_max_frag = ip_loopback_mtu_v6plus;
-        } else {
-                ill->ill_max_frag = ip_loopback_mtuplus;
-        }
         if (!ill_allocate_mibs(ill))
                 goto done;
+
         ill->ill_current_frag = ill->ill_max_frag;
         ill->ill_mtu = ill->ill_max_frag;       /* Initial value */
         ill->ill_mc_mtu = ill->ill_mtu;
         /*
          * ipif_loopback_name can't be pointed at directly because its used

@@ -3728,25 +3751,10 @@
         (void) strcpy(ill->ill_name, ipif_loopback_name);
         ill->ill_name_length = sizeof (ipif_loopback_name);
         /* Set ill_dlpi_pending for ipsq_current_finish() to work properly */
         ill->ill_dlpi_pending = DL_PRIM_INVAL;
 
-        rw_init(&ill->ill_mcast_lock, NULL, RW_DEFAULT, NULL);
-        mutex_init(&ill->ill_mcast_serializer, NULL, MUTEX_DEFAULT, NULL);
-        ill->ill_global_timer = INFINITY;
-        ill->ill_mcast_v1_time = ill->ill_mcast_v2_time = 0;
-        ill->ill_mcast_v1_tset = ill->ill_mcast_v2_tset = 0;
-        ill->ill_mcast_rv = MCAST_DEF_ROBUSTNESS;
-        ill->ill_mcast_qi = MCAST_DEF_QUERY_INTERVAL;
-
-        /* No resolver here. */
-        ill->ill_net_type = IRE_LOOPBACK;
-
-        /* Initialize the ipsq */
-        if (!ipsq_init(ill, B_FALSE))
-                goto done;
-
         ipif = ipif_allocate(ill, 0L, IRE_LOOPBACK, B_TRUE, B_TRUE, NULL);
         if (ipif == NULL)
                 goto done;
 
         ill->ill_flags = ILLF_MULTICAST;

@@ -3771,21 +3779,14 @@
 
         /*
          * Chain us in at the end of the ill list. hold the ill
          * before we make it globally visible. 1 for the lookup.
          */
-        ill->ill_refcnt = 0;
         ill_refhold(ill);
 
-        ill->ill_frag_count = 0;
-        ill->ill_frag_free_num_pkts = 0;
-        ill->ill_last_frag_clean_time = 0;
-
         ipsq = ill->ill_phyint->phyint_ipsq;
 
-        ill_set_inputfn(ill);
-
         if (ill_glist_insert(ill, "lo", isv6) != 0)
                 cmn_err(CE_PANIC, "cannot insert loopback interface");
 
         /* Let SCTP know so that it can add this to its list */
         sctp_update_ill(ill, SCTP_ILL_INSERT);