include/inet/ip_stack.h

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  24  * Use is subject to license terms.
  25  */
  26
  27 #ifndef _INET_IP_STACK_H
  28 #define _INET_IP_STACK_H
  29
  30 #ifdef  __cplusplus
  31 extern "C" {
  32 #endif
  33
  34 #include <sys/netstack.h>
  35 #include <netinet/igmp_var.h>
  36 #include <sys/modhash.h>
  37
  38 #ifdef _KERNEL
  39 #include <sys/list.h>
  40
  41
  42 /*
  43  * IP statistics.
  44  */
  45 #define IP_STAT(ipst, x)        ((ipst)->ips_ip_statistics.x.value.ui64++)
  46 #define IP_STAT_UPDATE(ipst, x, n) \
  47                 ((ipst)->ips_ip_statistics.x.value.ui64 += (n))
  48
  49 typedef struct ip_stat {
  50         kstat_named_t   ip_udp_fannorm;
  51         kstat_named_t   ip_udp_fanmb;
  52         kstat_named_t   ip_recv_pullup;
  53         kstat_named_t   ip_db_ref;
  54         kstat_named_t   ip_notaligned;
  55         kstat_named_t   ip_multimblk;
  56         kstat_named_t   ip_opt;
  57         kstat_named_t   ipsec_proto_ahesp;
  58         kstat_named_t   ip_conn_flputbq;
  59         kstat_named_t   ip_conn_walk_drain;
  60         kstat_named_t   ip_out_sw_cksum;
  61         kstat_named_t   ip_out_sw_cksum_bytes;
  62         kstat_named_t   ip_in_sw_cksum;
  63         kstat_named_t   ip_ire_reclaim_calls;
  64         kstat_named_t   ip_ire_reclaim_deleted;
  65         kstat_named_t   ip_nce_reclaim_calls;
  66         kstat_named_t   ip_nce_reclaim_deleted;
  67         kstat_named_t   ip_dce_reclaim_calls;
  68         kstat_named_t   ip_dce_reclaim_deleted;
  69         kstat_named_t   ip_tcp_in_full_hw_cksum_err;
  70         kstat_named_t   ip_tcp_in_part_hw_cksum_err;
  71         kstat_named_t   ip_tcp_in_sw_cksum_err;
  72         kstat_named_t   ip_udp_in_full_hw_cksum_err;
  73         kstat_named_t   ip_udp_in_part_hw_cksum_err;
  74         kstat_named_t   ip_udp_in_sw_cksum_err;
  75         kstat_named_t   conn_in_recvdstaddr;
  76         kstat_named_t   conn_in_recvopts;
  77         kstat_named_t   conn_in_recvif;
  78         kstat_named_t   conn_in_recvslla;
  79         kstat_named_t   conn_in_recvucred;
  80         kstat_named_t   conn_in_recvttl;
  81         kstat_named_t   conn_in_recvhopopts;
  82         kstat_named_t   conn_in_recvhoplimit;
  83         kstat_named_t   conn_in_recvdstopts;
  84         kstat_named_t   conn_in_recvrthdrdstopts;
  85         kstat_named_t   conn_in_recvrthdr;
  86         kstat_named_t   conn_in_recvpktinfo;
  87         kstat_named_t   conn_in_recvtclass;
  88         kstat_named_t   conn_in_timestamp;
  89 } ip_stat_t;
  90
  91
  92 /*
  93  * IP6 statistics.
  94  */
  95 #define IP6_STAT(ipst, x)       ((ipst)->ips_ip6_statistics.x.value.ui64++)
  96 #define IP6_STAT_UPDATE(ipst, x, n)     \
  97         ((ipst)->ips_ip6_statistics.x.value.ui64 += (n))
  98
  99 typedef struct ip6_stat {
 100         kstat_named_t   ip6_udp_fannorm;
 101         kstat_named_t   ip6_udp_fanmb;
 102         kstat_named_t   ip6_recv_pullup;
 103         kstat_named_t   ip6_db_ref;
 104         kstat_named_t   ip6_notaligned;
 105         kstat_named_t   ip6_multimblk;
 106         kstat_named_t   ipsec_proto_ahesp;
 107         kstat_named_t   ip6_out_sw_cksum;
 108         kstat_named_t   ip6_out_sw_cksum_bytes;
 109         kstat_named_t   ip6_in_sw_cksum;
 110         kstat_named_t   ip6_tcp_in_full_hw_cksum_err;
 111         kstat_named_t   ip6_tcp_in_part_hw_cksum_err;
 112         kstat_named_t   ip6_tcp_in_sw_cksum_err;
 113         kstat_named_t   ip6_udp_in_full_hw_cksum_err;
 114         kstat_named_t   ip6_udp_in_part_hw_cksum_err;
 115         kstat_named_t   ip6_udp_in_sw_cksum_err;
 116 } ip6_stat_t;
 117
 118 typedef struct ire_stats {
 119         uint64_t ire_stats_alloced;     /* # of ires alloced */
 120         uint64_t ire_stats_freed;       /* # of ires freed */
 121         uint64_t ire_stats_inserted;    /* # of ires inserted in the bucket */
 122         uint64_t ire_stats_deleted;     /* # of ires deleted from the bucket */
 123 } ire_stats_t;
 124
 125 #define TX_FANOUT_SIZE  128
 126 #define IDLHASHINDEX(X) \
 127         ((((uintptr_t)(X) >> 2) + ((uintptr_t)(X) >> 9)) & (TX_FANOUT_SIZE - 1))
 128
 129 /* Data structure to represent addresses */
 130 typedef struct srcid_map {
 131         struct srcid_map        *sm_next;
 132         in6_addr_t              sm_addr;        /* Local address */
 133         uint_t                  sm_srcid;       /* source id */
 134         uint_t                  sm_refcnt;      /* > 1 ipif with same addr? */
 135         zoneid_t                sm_zoneid;      /* zone id */
 136 } srcid_map_t;
 137
 138 /*
 139  * IP stack instances
 140  */
 141 struct ip_stack {
 142         netstack_t      *ips_netstack;  /* Common netstack */
 143
 144         uint_t                  ips_src_generation;     /* Both IPv4 and IPv6 */
 145
 146         struct mod_prop_info_s  *ips_propinfo_tbl;      /* ip tunables table */
 147
 148         mib2_ipIfStatsEntry_t   ips_ip_mib;     /* SNMP fixed size info */
 149         mib2_icmp_t     ips_icmp_mib;
 150         /*
 151          * IPv6 mibs when the interface (ill) is not known.
 152          * When the ill is known the per-interface mib in the ill is used.
 153          */
 154         mib2_ipIfStatsEntry_t   ips_ip6_mib;
 155         mib2_ipv6IfIcmpEntry_t  ips_icmp6_mib;
 156
 157         struct igmpstat         ips_igmpstat;
 158
 159         kstat_t         *ips_ip_mibkp;  /* kstat exporting ip_mib data */
 160         kstat_t         *ips_icmp_mibkp; /* kstat exporting icmp_mib data */
 161         kstat_t         *ips_ip_kstat;
 162         ip_stat_t       ips_ip_statistics;
 163         kstat_t         *ips_ip6_kstat;
 164         ip6_stat_t      ips_ip6_statistics;
 165
 166 /* ip.c */
 167         kmutex_t        ips_igmp_timer_lock;
 168         kmutex_t        ips_mld_timer_lock;
 169         kmutex_t        ips_ip_mi_lock;
 170         kmutex_t        ips_ip_addr_avail_lock;
 171         krwlock_t       ips_ill_g_lock;
 172
 173         krwlock_t       ips_ill_g_usesrc_lock;
 174
 175         /* Taskq dispatcher for capability operations */
 176         kmutex_t        ips_capab_taskq_lock;
 177         kcondvar_t      ips_capab_taskq_cv;
 178         mblk_t          *ips_capab_taskq_head;
 179         mblk_t          *ips_capab_taskq_tail;
 180         kthread_t       *ips_capab_taskq_thread;
 181         boolean_t       ips_capab_taskq_quit;
 182
 183 /* ipclassifier.c - keep in ip_stack_t */
 184         /* ipclassifier hash tables */
 185         struct connf_s  *ips_rts_clients;
 186         struct connf_s  *ips_ipcl_conn_fanout;
 187         struct connf_s  *ips_ipcl_bind_fanout;
 188         struct connf_s  *ips_ipcl_proto_fanout_v4;
 189         struct connf_s  *ips_ipcl_proto_fanout_v6;
 190         struct connf_s  *ips_ipcl_udp_fanout;
 191         struct connf_s  *ips_ipcl_raw_fanout;           /* RAW SCTP sockets */
 192         struct connf_s  *ips_ipcl_iptun_fanout;
 193         uint_t          ips_ipcl_conn_fanout_size;
 194         uint_t          ips_ipcl_bind_fanout_size;
 195         uint_t          ips_ipcl_udp_fanout_size;
 196         uint_t          ips_ipcl_raw_fanout_size;
 197         uint_t          ips_ipcl_iptun_fanout_size;
 198         struct connf_s  *ips_ipcl_globalhash_fanout;
 199         int             ips_conn_g_index;
 200
 201 /* ip.c */
 202         /* Following protected by igmp_timer_lock */
 203         int             ips_igmp_time_to_next;  /* Time since last timeout */
 204         int             ips_igmp_timer_scheduled_last;
 205         int             ips_igmp_deferred_next;
 206         timeout_id_t    ips_igmp_timeout_id;
 207         boolean_t       ips_igmp_timer_setter_active;
 208         boolean_t       ips_igmp_timer_quiesce;
 209
 210         /* Following protected by mld_timer_lock */
 211         int             ips_mld_time_to_next;   /* Time since last timeout */
 212         int             ips_mld_timer_scheduled_last;
 213         int             ips_mld_deferred_next;
 214         timeout_id_t    ips_mld_timeout_id;
 215         boolean_t       ips_mld_timer_setter_active;
 216         boolean_t       ips_mld_timer_quiesce;
 217
 218         /* Protected by igmp_slowtimeout_lock */
 219         timeout_id_t    ips_igmp_slowtimeout_id;
 220         kmutex_t        ips_igmp_slowtimeout_lock;
 221         boolean_t       ips_igmp_slowtimeout_quiesce;
 222
 223         /* Protected by mld_slowtimeout_lock */
 224         timeout_id_t    ips_mld_slowtimeout_id;
 225         kmutex_t        ips_mld_slowtimeout_lock;
 226         boolean_t       ips_mld_slowtimeout_quiesce;
 227
 228         /* IPv4 forwarding table */
 229         struct radix_node_head *ips_ip_ftable;
 230
 231 #define IPV6_ABITS              128
 232 #define IP6_MASK_TABLE_SIZE     (IPV6_ABITS + 1)        /* 129 ptrs */
 233         struct irb      *ips_ip_forwarding_table_v6[IP6_MASK_TABLE_SIZE];
 234
 235         /*
 236          * ire_ft_init_lock is used while initializing ip_forwarding_table
 237          * dynamically in ire_add.
 238          */
 239         kmutex_t        ips_ire_ft_init_lock;
 240
 241         /*
 242          * This is the IPv6 counterpart of RADIX_NODE_HEAD_LOCK. It is used
 243          * to prevent adds and deletes while we are doing a ftable_lookup
 244          * and extracting the ire_generation.
 245          */
 246         krwlock_t       ips_ip6_ire_head_lock;
 247
 248         uint32_t        ips_ip6_ftable_hash_size;
 249
 250         ire_stats_t     ips_ire_stats_v4;       /* IPv4 ire statistics */
 251         ire_stats_t     ips_ire_stats_v6;       /* IPv6 ire statistics */
 252
 253         /* Count how many condemned objects for kmem_cache callbacks */
 254         uint32_t        ips_num_ire_condemned;
 255         uint32_t        ips_num_nce_condemned;
 256         uint32_t        ips_num_dce_condemned;
 257
 258         struct ire_s    *ips_ire_reject_v4;     /* For unreachable dests */
 259         struct ire_s    *ips_ire_reject_v6;     /* For unreachable dests */
 260         struct ire_s    *ips_ire_blackhole_v4;  /* For temporary failures */
 261         struct ire_s    *ips_ire_blackhole_v6;  /* For temporary failures */
 262
 263         /* ips_ire_dep_lock protects ire_dep_* relationship between IREs */
 264         krwlock_t       ips_ire_dep_lock;
 265
 266         /* Destination Cache Entries */
 267         struct dce_s    *ips_dce_default;
 268         uint_t          ips_dce_hashsize;
 269         struct dcb_s    *ips_dce_hash_v4;
 270         struct dcb_s    *ips_dce_hash_v6;
 271         uint_t          ips_dce_reclaim_needed;
 272
 273         /* pending binds */
 274         mblk_t          *ips_ip6_asp_pending_ops;
 275         mblk_t          *ips_ip6_asp_pending_ops_tail;
 276
 277         /* Synchronize updates with table usage */
 278         mblk_t          *ips_ip6_asp_pending_update; /* pending table updates */
 279
 280         boolean_t       ips_ip6_asp_uip;        /* table update in progress */
 281         kmutex_t        ips_ip6_asp_lock;       /* protect all the above */
 282         uint32_t        ips_ip6_asp_refcnt;     /* outstanding references */
 283
 284         struct ip6_asp  *ips_ip6_asp_table;
 285         /* The number of policy entries in the table */
 286         uint_t          ips_ip6_asp_table_count;
 287
 288         struct conn_s   *ips_ip_g_mrouter;
 289
 290         /* Time since last icmp_pkt_err */
 291         clock_t         ips_icmp_pkt_err_last;
 292         /* Number of packets sent in burst */
 293         uint_t          ips_icmp_pkt_err_sent;
 294
 295         /* Protected by ip_mi_lock */
 296         void            *ips_ip_g_head; /* IP Instance Data List Head */
 297         void            *ips_arp_g_head; /* ARP Instance Data List Head */
 298
 299         struct ipsq_s   *ips_ipsq_g_head;
 300         uint_t          ips_ill_index;  /* Used to assign interface indicies */
 301         /* When set search for unused index */
 302         boolean_t       ips_ill_index_wrap;
 303
 304         uint_t          ips_loopback_packets;
 305
 306         /* NDP/NCE structures for IPv4 and IPv6 */
 307         struct ndp_g_s  *ips_ndp4;
 308         struct ndp_g_s  *ips_ndp6;
 309
 310         /* ip_mroute stuff */
 311         kmutex_t        ips_ip_g_mrouter_mutex;
 312
 313         struct mrtstat  *ips_mrtstat;   /* Stats for netstat */
 314         int             ips_saved_ip_forwarding;
 315
 316         /* numvifs is only a hint about the max interface being used. */
 317         ushort_t        ips_numvifs;
 318         kmutex_t        ips_numvifs_mutex;
 319
 320         struct vif      *ips_vifs;
 321         struct mfcb     *ips_mfcs;      /* kernel routing table */
 322         struct tbf      *ips_tbfs;
 323         /*
 324          * One-back cache used to locate a tunnel's vif,
 325          * given a datagram's src ip address.
 326          */
 327         ipaddr_t        ips_last_encap_src;
 328         struct vif      *ips_last_encap_vif;
 329         kmutex_t        ips_last_encap_lock;    /* Protects the above */
 330
 331         /*
 332          * reg_vif_num is protected by numvifs_mutex
 333          */
 334         /* Whether or not special PIM assert processing is enabled. */
 335         ushort_t        ips_reg_vif_num;        /* Index to Register vif */
 336         int             ips_pim_assert;
 337
 338         union ill_g_head_u *ips_ill_g_heads;   /* ILL List Head */
 339
 340         kstat_t         *ips_loopback_ksp;
 341
 342         /* Array of conn drain lists */
 343         struct idl_tx_list_s    *ips_idl_tx_list;
 344         uint_t          ips_conn_drain_list_cnt; /* Count of conn_drain_list */
 345
 346         /*
 347          * ID used to assign next free one.
 348          * Increases by one. Once it wraps we search for an unused ID.
 349          */
 350         uint_t          ips_ip_src_id;
 351         boolean_t       ips_srcid_wrapped;
 352
 353         struct srcid_map *ips_srcid_head;
 354         krwlock_t       ips_srcid_lock;
 355
 356         uint64_t        ips_ipif_g_seqid;       /* Used only for sctp_addr.c */
 357         union phyint_list_u *ips_phyint_g_list; /* start of phyint list */
 358
 359 /* ip_netinfo.c */
 360         hook_family_t   ips_ipv4root;
 361         hook_family_t   ips_ipv6root;
 362         hook_family_t   ips_arproot;
 363
 364         net_handle_t            ips_ipv4_net_data;
 365         net_handle_t            ips_ipv6_net_data;
 366         net_handle_t            ips_arp_net_data;
 367
 368         /*
 369          * Hooks for firewalling
 370          */
 371         hook_event_t            ips_ip4_physical_in_event;
 372         hook_event_t            ips_ip4_physical_out_event;
 373         hook_event_t            ips_ip4_forwarding_event;
 374         hook_event_t            ips_ip4_loopback_in_event;
 375         hook_event_t            ips_ip4_loopback_out_event;
 376
 377         hook_event_t            ips_ip6_physical_in_event;
 378         hook_event_t            ips_ip6_physical_out_event;
 379         hook_event_t            ips_ip6_forwarding_event;
 380         hook_event_t            ips_ip6_loopback_in_event;
 381         hook_event_t            ips_ip6_loopback_out_event;
 382
 383         hook_event_t            ips_arp_physical_in_event;
 384         hook_event_t            ips_arp_physical_out_event;
 385         hook_event_t            ips_arp_nic_events;
 386
 387         hook_event_token_t      ips_ipv4firewall_physical_in;
 388         hook_event_token_t      ips_ipv4firewall_physical_out;
 389         hook_event_token_t      ips_ipv4firewall_forwarding;
 390         hook_event_token_t      ips_ipv4firewall_loopback_in;
 391         hook_event_token_t      ips_ipv4firewall_loopback_out;
 392
 393         hook_event_token_t      ips_ipv6firewall_physical_in;
 394         hook_event_token_t      ips_ipv6firewall_physical_out;
 395         hook_event_token_t      ips_ipv6firewall_forwarding;
 396         hook_event_token_t      ips_ipv6firewall_loopback_in;
 397         hook_event_token_t      ips_ipv6firewall_loopback_out;
 398
 399         hook_event_t            ips_ip4_nic_events;
 400         hook_event_t            ips_ip6_nic_events;
 401         hook_event_token_t      ips_ipv4nicevents;
 402         hook_event_token_t      ips_ipv6nicevents;
 403
 404         hook_event_token_t      ips_arp_physical_in;
 405         hook_event_token_t      ips_arp_physical_out;
 406         hook_event_token_t      ips_arpnicevents;
 407
 408         net_handle_t            ips_ip4_observe_pr;
 409         net_handle_t            ips_ip6_observe_pr;
 410         hook_event_t            ips_ip4_observe;
 411         hook_event_t            ips_ip6_observe;
 412         hook_event_token_t      ips_ipv4observing;
 413         hook_event_token_t      ips_ipv6observing;
 414
 415         struct __ldi_ident      *ips_ldi_ident;
 416
 417 /* ipmp.c */
 418         krwlock_t               ips_ipmp_lock;
 419         mod_hash_t              *ips_ipmp_grp_hash;
 420
 421 };
 422 typedef struct ip_stack ip_stack_t;
 423
 424 /* Finding an ip_stack_t */
 425 #define CONNQ_TO_IPST(_q)       (Q_TO_CONN(_q)->conn_netstack->netstack_ip)
 426 #define ILLQ_TO_IPST(_q)        (((ill_t *)(_q)->q_ptr)->ill_ipst)
 427 #define PHYINT_TO_IPST(phyi)    ((phyi)->phyint_ipsq->ipsq_ipst)
 428
 429 #else /* _KERNEL */
 430 typedef int ip_stack_t;
 431 #endif /* _KERNEL */
 432
 433 #ifdef  __cplusplus
 434 }
 435 #endif
 436
 437 #endif  /* _INET_IP_STACK_H */