1 8485 Peter /* 2 8485 Peter * CDDL HEADER START 3 8485 Peter * 4 8485 Peter * The contents of this file are subject to the terms of the 5 8485 Peter * Common Development and Distribution License (the "License"). 6 8485 Peter * You may not use this file except in compliance with the License. 7 8485 Peter * 8 8485 Peter * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 8485 Peter * or http://www.opensolaris.org/os/licensing. 10 8485 Peter * See the License for the specific language governing permissions 11 8485 Peter * and limitations under the License. 12 8485 Peter * 13 8485 Peter * When distributing Covered Code, include this CDDL HEADER in each 14 8485 Peter * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 8485 Peter * If applicable, add the following below this CDDL HEADER, with the 16 8485 Peter * fields enclosed by brackets "[]" replaced with your own identifying 17 8485 Peter * information: Portions Copyright [yyyy] [name of copyright owner] 18 8485 Peter * 19 8485 Peter * CDDL HEADER END 20 8485 Peter * 21 8485 Peter * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 22 8485 Peter * Use is subject to license terms. 23 8485 Peter */ 24 8485 Peter 25 8485 Peter #include <inet/ip.h> 26 8485 Peter #include <inet/ip6.h> 27 8485 Peter #include <inet/ip_if.h> 28 8485 Peter #include <inet/ip_ire.h> 29 8485 Peter #include <inet/ip_multi.h> 30 11042 Erik #include <inet/ip_ndp.h> 31 8485 Peter #include <inet/ip_rts.h> 32 8485 Peter #include <inet/mi.h> 33 8485 Peter #include <net/if_types.h> 34 8485 Peter #include <sys/dlpi.h> 35 8485 Peter #include <sys/kmem.h> 36 8485 Peter #include <sys/modhash.h> 37 8485 Peter #include <sys/sdt.h> 38 8485 Peter #include <sys/strsun.h> 39 8485 Peter #include <sys/sunddi.h> 40 8485 Peter #include <sys/types.h> 41 8485 Peter 42 8485 Peter /* 43 8485 Peter * Convenience macros for getting the ip_stack_t associated with an 44 8485 Peter * ipmp_illgrp_t or ipmp_grp_t. 45 8485 Peter */ 46 8485 Peter #define IPMP_GRP_TO_IPST(grp) PHYINT_TO_IPST((grp)->gr_phyint) 47 8485 Peter #define IPMP_ILLGRP_TO_IPST(illg) ((illg)->ig_ipmp_ill->ill_ipst) 48 8485 Peter 49 8485 Peter /* 50 8485 Peter * Assorted constants that aren't important enough to be tunable. 51 8485 Peter */ 52 8485 Peter #define IPMP_GRP_HASH_SIZE 64 53 8485 Peter #define IPMP_ILL_REFRESH_TIMEOUT 120 /* seconds */ 54 8485 Peter 55 8485 Peter 56 8485 Peter /* 57 8485 Peter * IPMP meta-interface kstats (based on those in PSARC/1997/198). 58 8485 Peter */ 59 8485 Peter static const kstat_named_t ipmp_kstats[IPMP_KSTAT_MAX] = { 60 8485 Peter { "obytes", KSTAT_DATA_UINT32 }, 61 8485 Peter { "obytes64", KSTAT_DATA_UINT64 }, 62 8485 Peter { "rbytes", KSTAT_DATA_UINT32 }, 63 8485 Peter { "rbytes64", KSTAT_DATA_UINT64 }, 64 8485 Peter { "opackets", KSTAT_DATA_UINT32 }, 65 8485 Peter { "opackets64", KSTAT_DATA_UINT64 }, 66 8485 Peter { "oerrors", KSTAT_DATA_UINT32 }, 67 8485 Peter { "ipackets", KSTAT_DATA_UINT32 }, 68 8485 Peter { "ipackets64", KSTAT_DATA_UINT64 }, 69 8485 Peter { "ierrors", KSTAT_DATA_UINT32 }, 70 8485 Peter { "multircv", KSTAT_DATA_UINT32 }, 71 8485 Peter { "multixmt", KSTAT_DATA_UINT32 }, 72 8485 Peter { "brdcstrcv", KSTAT_DATA_UINT32 }, 73 8485 Peter { "brdcstxmt", KSTAT_DATA_UINT32 }, 74 8485 Peter { "link_up", KSTAT_DATA_UINT32 } 75 8485 Peter }; 76 8485 Peter 77 8485 Peter static void ipmp_grp_insert(ipmp_grp_t *, mod_hash_hndl_t); 78 8485 Peter static int ipmp_grp_create_kstats(ipmp_grp_t *); 79 8485 Peter static int ipmp_grp_update_kstats(kstat_t *, int); 80 8485 Peter static void ipmp_grp_destroy_kstats(ipmp_grp_t *); 81 8485 Peter static ill_t *ipmp_illgrp_min_ill(ipmp_illgrp_t *); 82 8485 Peter static ill_t *ipmp_illgrp_max_ill(ipmp_illgrp_t *); 83 8485 Peter static void ipmp_illgrp_set_cast(ipmp_illgrp_t *, ill_t *); 84 8485 Peter static void ipmp_illgrp_set_mtu(ipmp_illgrp_t *, uint_t); 85 8485 Peter static boolean_t ipmp_ill_activate(ill_t *); 86 8485 Peter static void ipmp_ill_deactivate(ill_t *); 87 8485 Peter static void ipmp_ill_ire_mark_testhidden(ire_t *, char *); 88 8485 Peter static void ipmp_ill_ire_clear_testhidden(ire_t *, char *); 89 8485 Peter static void ipmp_ill_refresh_active_timer_start(ill_t *); 90 8485 Peter static void ipmp_ill_rtsaddrmsg(ill_t *, int); 91 8485 Peter static void ipmp_ill_bind_ipif(ill_t *, ipif_t *, enum ip_resolver_action); 92 8485 Peter static ipif_t *ipmp_ill_unbind_ipif(ill_t *, ipif_t *, boolean_t); 93 8485 Peter static void ipmp_phyint_get_kstats(phyint_t *, uint64_t *); 94 8485 Peter static boolean_t ipmp_ipif_is_up_dataaddr(const ipif_t *); 95 8485 Peter 96 8485 Peter /* 97 8485 Peter * Initialize IPMP state for IP stack `ipst'; called from ip_stack_init(). 98 8485 Peter */ 99 8485 Peter void 100 8485 Peter ipmp_init(ip_stack_t *ipst) 101 8485 Peter { 102 8485 Peter ipst->ips_ipmp_grp_hash = mod_hash_create_extended("ipmp_grp_hash", 103 8485 Peter IPMP_GRP_HASH_SIZE, mod_hash_null_keydtor, mod_hash_null_valdtor, 104 8485 Peter mod_hash_bystr, NULL, mod_hash_strkey_cmp, KM_SLEEP); 105 8485 Peter rw_init(&ipst->ips_ipmp_lock, NULL, RW_DEFAULT, 0); 106 8485 Peter } 107 8485 Peter 108 8485 Peter /* 109 8485 Peter * Destroy IPMP state for IP stack `ipst'; called from ip_stack_fini(). 110 8485 Peter */ 111 8485 Peter void 112 8485 Peter ipmp_destroy(ip_stack_t *ipst) 113 8485 Peter { 114 8485 Peter mod_hash_destroy_hash(ipst->ips_ipmp_grp_hash); 115 8485 Peter rw_destroy(&ipst->ips_ipmp_lock); 116 8485 Peter } 117 8485 Peter 118 8485 Peter /* 119 8485 Peter * Create an IPMP group named `grname', associate it with IPMP phyint `phyi', 120 8485 Peter * and add it to the hash. On success, return a pointer to the created group. 121 8485 Peter * Caller must ensure `grname' is not yet in the hash. Assumes that the IPMP 122 8485 Peter * meta-interface associated with the group also has the same name (but they 123 8485 Peter * may differ later via ipmp_grp_rename()). 124 8485 Peter */ 125 8485 Peter ipmp_grp_t * 126 8485 Peter ipmp_grp_create(const char *grname, phyint_t *phyi) 127 8485 Peter { 128 8485 Peter ipmp_grp_t *grp; 129 8485 Peter ip_stack_t *ipst = PHYINT_TO_IPST(phyi); 130 8485 Peter mod_hash_hndl_t mh; 131 8485 Peter 132 8485 Peter ASSERT(RW_WRITE_HELD(&ipst->ips_ipmp_lock)); 133 8485 Peter 134 8485 Peter if ((grp = kmem_zalloc(sizeof (ipmp_grp_t), KM_NOSLEEP)) == NULL) 135 8485 Peter return (NULL); 136 8485 Peter 137 8485 Peter (void) strlcpy(grp->gr_name, grname, sizeof (grp->gr_name)); 138 8485 Peter (void) strlcpy(grp->gr_ifname, grname, sizeof (grp->gr_ifname)); 139 8485 Peter 140 8485 Peter /* 141 8485 Peter * Cache the group's phyint. This is safe since a phyint_t will 142 8485 Peter * outlive its ipmp_grp_t. 143 8485 Peter */ 144 8485 Peter grp->gr_phyint = phyi; 145 8485 Peter 146 8485 Peter /* 147 8485 Peter * Create IPMP group kstats. 148 8485 Peter */ 149 8485 Peter if (ipmp_grp_create_kstats(grp) != 0) { 150 8485 Peter kmem_free(grp, sizeof (ipmp_grp_t)); 151 8485 Peter return (NULL); 152 8485 Peter } 153 8485 Peter 154 8485 Peter /* 155 8485 Peter * Insert the group into the hash. 156 8485 Peter */ 157 8485 Peter if (mod_hash_reserve_nosleep(ipst->ips_ipmp_grp_hash, &mh) != 0) { 158 8485 Peter ipmp_grp_destroy_kstats(grp); 159 8485 Peter kmem_free(grp, sizeof (ipmp_grp_t)); 160 8485 Peter return (NULL); 161 8485 Peter } 162 8485 Peter ipmp_grp_insert(grp, mh); 163 8485 Peter 164 8485 Peter return (grp); 165 8485 Peter } 166 8485 Peter 167 8485 Peter /* 168 8485 Peter * Create IPMP kstat structures for `grp'. Return an errno upon failure. 169 8485 Peter */ 170 8485 Peter static int 171 8485 Peter ipmp_grp_create_kstats(ipmp_grp_t *grp) 172 8485 Peter { 173 8485 Peter kstat_t *ksp; 174 8485 Peter netstackid_t id = IPMP_GRP_TO_IPST(grp)->ips_netstack->netstack_stackid; 175 8485 Peter 176 8485 Peter ksp = kstat_create_netstack("ipmp", 0, grp->gr_ifname, "net", 177 8485 Peter KSTAT_TYPE_NAMED, IPMP_KSTAT_MAX, 0, id); 178 8485 Peter if (ksp == NULL) 179 8485 Peter return (ENOMEM); 180 8485 Peter 181 8485 Peter ksp->ks_update = ipmp_grp_update_kstats; 182 8485 Peter ksp->ks_private = grp; 183 8485 Peter bcopy(ipmp_kstats, ksp->ks_data, sizeof (ipmp_kstats)); 184 8485 Peter 185 8485 Peter kstat_install(ksp); 186 8485 Peter grp->gr_ksp = ksp; 187 8485 Peter return (0); 188 8485 Peter } 189 8485 Peter 190 8485 Peter /* 191 8485 Peter * Update the IPMP kstats tracked by `ksp'; called by the kstats framework. 192 8485 Peter */ 193 8485 Peter static int 194 8485 Peter ipmp_grp_update_kstats(kstat_t *ksp, int rw) 195 8485 Peter { 196 8485 Peter uint_t i; 197 8485 Peter kstat_named_t *kn = KSTAT_NAMED_PTR(ksp); 198 8485 Peter ipmp_grp_t *grp = ksp->ks_private; 199 8485 Peter ip_stack_t *ipst = IPMP_GRP_TO_IPST(grp); 200 8485 Peter ipsq_t *ipsq, *grp_ipsq = grp->gr_phyint->phyint_ipsq; 201 8485 Peter phyint_t *phyi; 202 8485 Peter uint64_t phyi_kstats[IPMP_KSTAT_MAX]; 203 8485 Peter 204 8485 Peter if (rw == KSTAT_WRITE) 205 8485 Peter return (EACCES); 206 8485 Peter 207 8485 Peter /* 208 8485 Peter * Start with the group's baseline values. 209 8485 Peter */ 210 8485 Peter for (i = 0; i < IPMP_KSTAT_MAX; i++) { 211 8485 Peter if (kn[i].data_type == KSTAT_DATA_UINT32) { 212 8485 Peter kn[i].value.ui32 = grp->gr_kstats0[i]; 213 8485 Peter } else { 214 8485 Peter ASSERT(kn[i].data_type == KSTAT_DATA_UINT64); 215 8485 Peter kn[i].value.ui64 = grp->gr_kstats0[i]; 216 8485 Peter } 217 8485 Peter } 218 8485 Peter 219 8485 Peter /* 220 8485 Peter * Add in the stats of each phyint currently in the group. Since we 221 8485 Peter * don't directly track the phyints in a group, we cheat by walking 222 8485 Peter * the IPSQ set under ill_g_lock. (The IPSQ list cannot change while 223 8485 Peter * ill_g_lock is held.) 224 8485 Peter */ 225 8485 Peter rw_enter(&ipst->ips_ill_g_lock, RW_READER); 226 8485 Peter ipsq = grp_ipsq->ipsq_next; 227 8485 Peter for (; ipsq != grp_ipsq; ipsq = ipsq->ipsq_next) { 228 8485 Peter phyi = ipsq->ipsq_phyint; 229 8485 Peter 230 8485 Peter /* 231 8485 Peter * If a phyint in a group is being unplumbed, it's possible 232 8485 Peter * that ill_glist_delete() -> phyint_free() already freed the 233 8485 Peter * phyint (and set ipsq_phyint to NULL), but the unplumb 234 8485 Peter * operation has yet to complete (and thus ipsq_dq() has yet 235 8485 Peter * to remove the phyint's IPSQ from the group IPSQ's phyint 236 8485 Peter * list). We skip those phyints here (note that their kstats 237 8485 Peter * have already been added to gr_kstats0[]). 238 8485 Peter */ 239 8485 Peter if (phyi == NULL) 240 8485 Peter continue; 241 8485 Peter 242 8485 Peter ipmp_phyint_get_kstats(phyi, phyi_kstats); 243 8485 Peter 244 8485 Peter for (i = 0; i < IPMP_KSTAT_MAX; i++) { 245 8485 Peter phyi_kstats[i] -= phyi->phyint_kstats0[i]; 246 8485 Peter if (kn[i].data_type == KSTAT_DATA_UINT32) 247 8485 Peter kn[i].value.ui32 += phyi_kstats[i]; 248 8485 Peter else 249 8485 Peter kn[i].value.ui64 += phyi_kstats[i]; 250 8485 Peter } 251 8485 Peter } 252 8485 Peter 253 8485 Peter kn[IPMP_KSTAT_LINK_UP].value.ui32 = 254 8485 Peter (grp->gr_phyint->phyint_flags & PHYI_RUNNING) != 0; 255 8485 Peter 256 8485 Peter rw_exit(&ipst->ips_ill_g_lock); 257 8485 Peter return (0); 258 8485 Peter } 259 8485 Peter 260 8485 Peter /* 261 8485 Peter * Destroy IPMP kstat structures for `grp'. 262 8485 Peter */ 263 8485 Peter static void 264 8485 Peter ipmp_grp_destroy_kstats(ipmp_grp_t *grp) 265 8485 Peter { 266 8485 Peter netstackid_t id = IPMP_GRP_TO_IPST(grp)->ips_netstack->netstack_stackid; 267 8485 Peter 268 8485 Peter kstat_delete_netstack(grp->gr_ksp, id); 269 8485 Peter bzero(grp->gr_kstats0, sizeof (grp->gr_kstats0)); 270 8485 Peter grp->gr_ksp = NULL; 271 8485 Peter } 272 8485 Peter 273 8485 Peter /* 274 8485 Peter * Look up an IPMP group named `grname' on IP stack `ipst'. Return NULL if it 275 8485 Peter * does not exist. 276 8485 Peter */ 277 8485 Peter ipmp_grp_t * 278 8485 Peter ipmp_grp_lookup(const char *grname, ip_stack_t *ipst) 279 8485 Peter { 280 8485 Peter ipmp_grp_t *grp; 281 8485 Peter 282 8485 Peter ASSERT(RW_LOCK_HELD(&ipst->ips_ipmp_lock)); 283 8485 Peter 284 8485 Peter if (mod_hash_find(ipst->ips_ipmp_grp_hash, (mod_hash_key_t)grname, 285 8485 Peter (mod_hash_val_t *)&grp) == 0) 286 8485 Peter return (grp); 287 8485 Peter 288 8485 Peter return (NULL); 289 8485 Peter } 290 8485 Peter 291 8485 Peter /* 292 8485 Peter * Place information about group `grp' into `lifgr'. 293 8485 Peter */ 294 8485 Peter void 295 8485 Peter ipmp_grp_info(const ipmp_grp_t *grp, lifgroupinfo_t *lifgr) 296 8485 Peter { 297 8485 Peter ill_t *ill; 298 8485 Peter ip_stack_t *ipst = IPMP_GRP_TO_IPST(grp); 299 8485 Peter 300 8485 Peter ASSERT(RW_LOCK_HELD(&ipst->ips_ipmp_lock)); 301 8485 Peter 302 8485 Peter lifgr->gi_v4 = (grp->gr_v4 != NULL); 303 8485 Peter lifgr->gi_v6 = (grp->gr_v6 != NULL); 304 8485 Peter lifgr->gi_nv4 = grp->gr_nv4 + grp->gr_pendv4; 305 8485 Peter lifgr->gi_nv6 = grp->gr_nv6 + grp->gr_pendv6; 306 8485 Peter lifgr->gi_mactype = grp->gr_nif > 0 ? grp->gr_mactype : SUNW_DL_IPMP; 307 8485 Peter (void) strlcpy(lifgr->gi_grifname, grp->gr_ifname, LIFNAMSIZ); 308 8485 Peter lifgr->gi_m4ifname[0] = '\0'; 309 8485 Peter lifgr->gi_m6ifname[0] = '\0'; 310 8485 Peter lifgr->gi_bcifname[0] = '\0'; 311 8485 Peter 312 8485 Peter if (grp->gr_v4 != NULL && (ill = grp->gr_v4->ig_cast_ill) != NULL) { 313 8485 Peter (void) strlcpy(lifgr->gi_m4ifname, ill->ill_name, LIFNAMSIZ); 314 8485 Peter (void) strlcpy(lifgr->gi_bcifname, ill->ill_name, LIFNAMSIZ); 315 8485 Peter } 316 8485 Peter 317 8485 Peter if (grp->gr_v6 != NULL && (ill = grp->gr_v6->ig_cast_ill) != NULL) 318 8485 Peter (void) strlcpy(lifgr->gi_m6ifname, ill->ill_name, LIFNAMSIZ); 319 8485 Peter } 320 8485 Peter 321 8485 Peter /* 322 8485 Peter * Insert `grp' into the hash using the reserved hash entry `mh'. 323 8485 Peter * Caller must ensure `grp' is not yet in the hash. 324 8485 Peter */ 325 8485 Peter static void 326 8485 Peter ipmp_grp_insert(ipmp_grp_t *grp, mod_hash_hndl_t mh) 327 8485 Peter { 328 8485 Peter int err; 329 8485 Peter ip_stack_t *ipst = IPMP_GRP_TO_IPST(grp); 330 8485 Peter 331 8485 Peter ASSERT(RW_WRITE_HELD(&ipst->ips_ipmp_lock)); 332 8485 Peter 333 8485 Peter /* 334 8485 Peter * Since grp->gr_name will exist at least as long as `grp' is in the 335 8485 Peter * hash, we use it directly as the key. 336 8485 Peter */ 337 8485 Peter err = mod_hash_insert_reserve(ipst->ips_ipmp_grp_hash, 338 8485 Peter (mod_hash_key_t)grp->gr_name, (mod_hash_val_t)grp, mh); 339 8485 Peter if (err != 0) { 340 8485 Peter /* 341 8485 Peter * This should never happen since `mh' was preallocated. 342 8485 Peter */ 343 8485 Peter panic("cannot insert IPMP group \"%s\" (err %d)", 344 8485 Peter grp->gr_name, err); 345 8485 Peter } 346 8485 Peter } 347 8485 Peter 348 8485 Peter /* 349 8485 Peter * Remove `grp' from the hash. Caller must ensure `grp' is in it. 350 8485 Peter */ 351 8485 Peter static void 352 8485 Peter ipmp_grp_remove(ipmp_grp_t *grp) 353 8485 Peter { 354 8485 Peter int err; 355 8485 Peter mod_hash_val_t val; 356 8485 Peter mod_hash_key_t key = (mod_hash_key_t)grp->gr_name; 357 8485 Peter ip_stack_t *ipst = IPMP_GRP_TO_IPST(grp); 358 8485 Peter 359 8485 Peter ASSERT(RW_WRITE_HELD(&ipst->ips_ipmp_lock)); 360 8485 Peter 361 8485 Peter err = mod_hash_remove(ipst->ips_ipmp_grp_hash, key, &val); 362 8485 Peter if (err != 0 || val != grp) { 363 8485 Peter panic("cannot remove IPMP group \"%s\" (err %d)", 364 8485 Peter grp->gr_name, err); 365 8485 Peter } 366 8485 Peter } 367 8485 Peter 368 8485 Peter /* 369 8485 Peter * Attempt to rename `grp' to new name `grname'. Return an errno if the new 370 8485 Peter * group name already exists or is invalid, or if there isn't enough memory. 371 8485 Peter */ 372 8485 Peter int 373 8485 Peter ipmp_grp_rename(ipmp_grp_t *grp, const char *grname) 374 8485 Peter { 375 8485 Peter mod_hash_hndl_t mh; 376 8485 Peter ip_stack_t *ipst = IPMP_GRP_TO_IPST(grp); 377 8485 Peter 378 8485 Peter ASSERT(RW_WRITE_HELD(&ipst->ips_ipmp_lock)); 379 8485 Peter 380 8485 Peter if (grname[0] == '\0') 381 8485 Peter return (EINVAL); 382 8485 Peter 383 8485 Peter if (mod_hash_find(ipst->ips_ipmp_grp_hash, (mod_hash_key_t)grname, 384 8485 Peter (mod_hash_val_t *)&grp) != MH_ERR_NOTFOUND) 385 8485 Peter return (EEXIST); 386 8485 Peter 387 8485 Peter /* 388 8485 Peter * Before we remove the group from the hash, ensure we'll be able to 389 8485 Peter * re-insert it by reserving space. 390 8485 Peter */ 391 8485 Peter if (mod_hash_reserve_nosleep(ipst->ips_ipmp_grp_hash, &mh) != 0) 392 8485 Peter return (ENOMEM); 393 8485 Peter 394 8485 Peter ipmp_grp_remove(grp); 395 8485 Peter (void) strlcpy(grp->gr_name, grname, sizeof (grp->gr_name)); 396 8485 Peter ipmp_grp_insert(grp, mh); 397 8485 Peter 398 8485 Peter return (0); 399 8485 Peter } 400 8485 Peter 401 8485 Peter /* 402 8485 Peter * Destroy `grp' and remove it from the hash. Caller must ensure `grp' is in 403 8485 Peter * the hash, and that there are no interfaces on it. 404 8485 Peter */ 405 8485 Peter void 406 8485 Peter ipmp_grp_destroy(ipmp_grp_t *grp) 407 8485 Peter { 408 8485 Peter ip_stack_t *ipst = IPMP_GRP_TO_IPST(grp); 409 8485 Peter 410 8485 Peter ASSERT(RW_WRITE_HELD(&ipst->ips_ipmp_lock)); 411 8485 Peter 412 8485 Peter /* 413 8485 Peter * If there are still interfaces using this group, panic before things 414 8485 Peter * go really off the rails. 415 8485 Peter */ 416 8485 Peter if (grp->gr_nif != 0) 417 8485 Peter panic("cannot destroy IPMP group \"%s\": in use", grp->gr_name); 418 8485 Peter 419 8485 Peter ipmp_grp_remove(grp); 420 8485 Peter ipmp_grp_destroy_kstats(grp); 421 8485 Peter 422 8485 Peter ASSERT(grp->gr_v4 == NULL); 423 8485 Peter ASSERT(grp->gr_v6 == NULL); 424 8485 Peter ASSERT(grp->gr_nv4 == 0); 425 8485 Peter ASSERT(grp->gr_nv6 == 0); 426 8485 Peter ASSERT(grp->gr_nactif == 0); 427 8485 Peter ASSERT(grp->gr_linkdownmp == NULL); 428 8485 Peter grp->gr_phyint = NULL; 429 8485 Peter 430 8485 Peter kmem_free(grp, sizeof (ipmp_grp_t)); 431 8485 Peter } 432 8485 Peter 433 8485 Peter /* 434 8485 Peter * Check whether `ill' is suitable for inclusion into `grp', and return an 435 8485 Peter * errno describing the problem (if any). NOTE: many of these errno values 436 8485 Peter * are interpreted by ifconfig, which will take corrective action and retry 437 8485 Peter * the SIOCSLIFGROUPNAME, so please exercise care when changing them. 438 8485 Peter */ 439 8485 Peter static int 440 8485 Peter ipmp_grp_vet_ill(ipmp_grp_t *grp, ill_t *ill) 441 8485 Peter { 442 8485 Peter ip_stack_t *ipst = IPMP_GRP_TO_IPST(grp); 443 8485 Peter 444 8485 Peter ASSERT(IAM_WRITER_ILL(ill)); 445 8485 Peter ASSERT(RW_LOCK_HELD(&ipst->ips_ipmp_lock)); 446 8485 Peter 447 8485 Peter /* 448 8485 Peter * To sidestep complicated address migration logic in the kernel and 449 8485 Peter * to force the kernel's all-hosts multicast memberships to be blown 450 8485 Peter * away, all addresses that had been brought up must be brought back 451 8485 Peter * down prior to adding an interface to a group. (This includes 452 8485 Peter * addresses currently down due to DAD.) Once the interface has been 453 8485 Peter * added to the group, its addresses can then be brought back up, at 454 8485 Peter * which point they will be moved to the IPMP meta-interface. 455 8485 Peter * NOTE: we do this before ill_appaddr_cnt() since bringing down the 456 8485 Peter * link-local causes in.ndpd to remove its ADDRCONF'd addresses. 457 8485 Peter */ 458 8485 Peter if (ill->ill_ipif_up_count + ill->ill_ipif_dup_count > 0) 459 8485 Peter return (EADDRINUSE); 460 8485 Peter 461 8485 Peter /* 462 8485 Peter * To avoid confusing applications by changing addresses that are 463 8485 Peter * under their control, all such control must be removed prior to 464 8485 Peter * adding an interface into a group. 465 8485 Peter */ 466 8485 Peter if (ill_appaddr_cnt(ill) != 0) 467 8485 Peter return (EADDRNOTAVAIL); 468 8485 Peter 469 8485 Peter /* 470 8485 Peter * Since PTP addresses do not share the same broadcast domain, they 471 8485 Peter * are not allowed to be in an IPMP group. 472 8485 Peter */ 473 8485 Peter if (ill_ptpaddr_cnt(ill) != 0) 474 8485 Peter return (EINVAL); 475 8485 Peter 476 8485 Peter /* 477 8485 Peter * An ill must support multicast to be allowed into a group. 478 8485 Peter */ 479 8485 Peter if (!(ill->ill_flags & ILLF_MULTICAST)) 480 8485 Peter return (ENOTSUP); 481 8485 Peter 482 8485 Peter /* 483 8485 Peter * An ill must strictly be using ARP and/or ND for address 484 8485 Peter * resolution for it to be allowed into a group. 485 8485 Peter */ 486 11042 Erik if (ill->ill_flags & (ILLF_NONUD | ILLF_NOARP)) 487 8485 Peter return (ENOTSUP); 488 8485 Peter 489 8485 Peter /* 490 8485 Peter * An ill cannot also be using usesrc groups. (Although usesrc uses 491 8485 Peter * ill_g_usesrc_lock, we don't need to grab it since usesrc also does 492 8485 Peter * all its modifications as writer.) 493 8485 Peter */ 494 8485 Peter if (IS_USESRC_ILL(ill) || IS_USESRC_CLI_ILL(ill)) 495 8485 Peter return (ENOTSUP); 496 8485 Peter 497 8485 Peter /* 498 8485 Peter * All ills in a group must be the same mactype. 499 8485 Peter */ 500 8485 Peter if (grp->gr_nif > 0 && grp->gr_mactype != ill->ill_mactype) 501 8485 Peter return (EINVAL); 502 8485 Peter 503 8485 Peter return (0); 504 8485 Peter } 505 8485 Peter 506 8485 Peter /* 507 8485 Peter * Check whether `phyi' is suitable for inclusion into `grp', and return an 508 8485 Peter * errno describing the problem (if any). See comment above ipmp_grp_vet_ill() 509 8485 Peter * regarding errno values. 510 8485 Peter */ 511 8485 Peter int 512 8485 Peter ipmp_grp_vet_phyint(ipmp_grp_t *grp, phyint_t *phyi) 513 8485 Peter { 514 8485 Peter int err = 0; 515 8485 Peter ip_stack_t *ipst = IPMP_GRP_TO_IPST(grp); 516 8485 Peter 517 8485 Peter ASSERT(IAM_WRITER_IPSQ(phyi->phyint_ipsq)); 518 8485 Peter ASSERT(RW_LOCK_HELD(&ipst->ips_ipmp_lock)); 519 8485 Peter 520 8485 Peter /* 521 8485 Peter * An interface cannot have address families plumbed that are not 522 8485 Peter * configured in the group. 523 8485 Peter */ 524 8485 Peter if (phyi->phyint_illv4 != NULL && grp->gr_v4 == NULL || 525 8485 Peter phyi->phyint_illv6 != NULL && grp->gr_v6 == NULL) 526 8485 Peter return (EAFNOSUPPORT); 527 8485 Peter 528 8485 Peter if (phyi->phyint_illv4 != NULL) 529 8485 Peter err = ipmp_grp_vet_ill(grp, phyi->phyint_illv4); 530 8485 Peter if (err == 0 && phyi->phyint_illv6 != NULL) 531 8485 Peter err = ipmp_grp_vet_ill(grp, phyi->phyint_illv6); 532 8485 Peter 533 8485 Peter return (err); 534 8485 Peter } 535 8485 Peter 536 8485 Peter /* 537 8485 Peter * Create a new illgrp on IPMP meta-interface `ill'. 538 8485 Peter */ 539 8485 Peter ipmp_illgrp_t * 540 8485 Peter ipmp_illgrp_create(ill_t *ill) 541 8485 Peter { 542 8485 Peter uint_t mtu = ill->ill_isv6 ? IPV6_MIN_MTU : IP_MIN_MTU; 543 8485 Peter ipmp_illgrp_t *illg; 544 8485 Peter 545 8485 Peter ASSERT(IAM_WRITER_ILL(ill)); 546 8485 Peter ASSERT(IS_IPMP(ill)); 547 8485 Peter ASSERT(ill->ill_grp == NULL); 548 8485 Peter 549 8485 Peter if ((illg = kmem_zalloc(sizeof (ipmp_illgrp_t), KM_NOSLEEP)) == NULL) 550 8485 Peter return (NULL); 551 8485 Peter 552 8485 Peter list_create(&illg->ig_if, sizeof (ill_t), offsetof(ill_t, ill_grpnode)); 553 8485 Peter list_create(&illg->ig_actif, sizeof (ill_t), 554 8485 Peter offsetof(ill_t, ill_actnode)); 555 8485 Peter list_create(&illg->ig_arpent, sizeof (ipmp_arpent_t), 556 8485 Peter offsetof(ipmp_arpent_t, ia_node)); 557 8485 Peter 558 8485 Peter illg->ig_ipmp_ill = ill; 559 8485 Peter ill->ill_grp = illg; 560 8485 Peter ipmp_illgrp_set_mtu(illg, mtu); 561 8485 Peter 562 8485 Peter return (illg); 563 8485 Peter } 564 8485 Peter 565 8485 Peter /* 566 8485 Peter * Destroy illgrp `illg', and disconnect it from its IPMP meta-interface. 567 8485 Peter */ 568 8485 Peter void 569 8485 Peter ipmp_illgrp_destroy(ipmp_illgrp_t *illg) 570 8485 Peter { 571 8485 Peter ASSERT(IAM_WRITER_ILL(illg->ig_ipmp_ill)); 572 8485 Peter ASSERT(IS_IPMP(illg->ig_ipmp_ill)); 573 8485 Peter 574 8485 Peter /* 575 8485 Peter * Verify `illg' is empty. 576 8485 Peter */ 577 8485 Peter ASSERT(illg->ig_next_ill == NULL); 578 8485 Peter ASSERT(illg->ig_cast_ill == NULL); 579 8485 Peter ASSERT(list_is_empty(&illg->ig_arpent)); 580 8485 Peter ASSERT(list_is_empty(&illg->ig_if)); 581 8485 Peter ASSERT(list_is_empty(&illg->ig_actif)); 582 8485 Peter ASSERT(illg->ig_nactif == 0); 583 8485 Peter 584 8485 Peter /* 585 8485 Peter * Destroy `illg'. 586 8485 Peter */ 587 8485 Peter illg->ig_ipmp_ill->ill_grp = NULL; 588 8485 Peter illg->ig_ipmp_ill = NULL; 589 8485 Peter list_destroy(&illg->ig_if); 590 8485 Peter list_destroy(&illg->ig_actif); 591 8485 Peter list_destroy(&illg->ig_arpent); 592 8485 Peter kmem_free(illg, sizeof (ipmp_illgrp_t)); 593 8485 Peter } 594 8485 Peter 595 8485 Peter /* 596 8485 Peter * Add `ipif' to the pool of usable data addresses on `illg' and attempt to 597 8485 Peter * bind it to an underlying ill, while keeping an even address distribution. 598 8485 Peter * If the bind is successful, return a pointer to the bound ill. 599 8485 Peter */ 600 8485 Peter ill_t * 601 8485 Peter ipmp_illgrp_add_ipif(ipmp_illgrp_t *illg, ipif_t *ipif) 602 8485 Peter { 603 8485 Peter ill_t *minill; 604 8485 Peter ipmp_arpent_t *entp; 605 8485 Peter 606 8485 Peter ASSERT(IAM_WRITER_IPIF(ipif)); 607 8485 Peter ASSERT(ipmp_ipif_is_dataaddr(ipif)); 608 8485 Peter 609 8485 Peter /* 610 8485 Peter * IPMP data address mappings are internally managed by IP itself, so 611 8485 Peter * delete any existing ARP entries associated with the address. 612 8485 Peter */ 613 8485 Peter if (!ipif->ipif_isv6) { 614 8485 Peter entp = ipmp_illgrp_lookup_arpent(illg, &ipif->ipif_lcl_addr); 615 8485 Peter if (entp != NULL) 616 8485 Peter ipmp_illgrp_destroy_arpent(illg, entp); 617 8485 Peter } 618 8485 Peter 619 8485 Peter if ((minill = ipmp_illgrp_min_ill(illg)) != NULL) 620 8485 Peter ipmp_ill_bind_ipif(minill, ipif, Res_act_none); 621 8485 Peter 622 8485 Peter return (ipif->ipif_bound ? ipif->ipif_bound_ill : NULL); 623 8485 Peter } 624 8485 Peter 625 8485 Peter /* 626 8485 Peter * Delete `ipif' from the pool of usable data addresses on `illg'. If it's 627 8485 Peter * bound, unbind it from the underlying ill while keeping an even address 628 8485 Peter * distribution. 629 8485 Peter */ 630 8485 Peter void 631 8485 Peter ipmp_illgrp_del_ipif(ipmp_illgrp_t *illg, ipif_t *ipif) 632 8485 Peter { 633 8485 Peter ill_t *maxill, *boundill = ipif->ipif_bound_ill; 634 8485 Peter 635 8485 Peter ASSERT(IAM_WRITER_IPIF(ipif)); 636 8485 Peter 637 8485 Peter if (boundill != NULL) { 638 8485 Peter (void) ipmp_ill_unbind_ipif(boundill, ipif, B_FALSE); 639 8485 Peter 640 8485 Peter maxill = ipmp_illgrp_max_ill(illg); 641 8485 Peter if (maxill->ill_bound_cnt > boundill->ill_bound_cnt + 1) { 642 8485 Peter ipif = ipmp_ill_unbind_ipif(maxill, NULL, B_TRUE); 643 8485 Peter ipmp_ill_bind_ipif(boundill, ipif, Res_act_rebind); 644 8485 Peter } 645 8485 Peter } 646 8485 Peter } 647 8485 Peter 648 8485 Peter /* 649 8485 Peter * Return the active ill with the greatest number of data addresses in `illg'. 650 8485 Peter */ 651 8485 Peter static ill_t * 652 8485 Peter ipmp_illgrp_max_ill(ipmp_illgrp_t *illg) 653 8485 Peter { 654 8485 Peter ill_t *ill, *bestill = NULL; 655 8485 Peter 656 8485 Peter ASSERT(IAM_WRITER_ILL(illg->ig_ipmp_ill)); 657 8485 Peter 658 8485 Peter ill = list_head(&illg->ig_actif); 659 8485 Peter for (; ill != NULL; ill = list_next(&illg->ig_actif, ill)) { 660 8485 Peter if (bestill == NULL || 661 8485 Peter ill->ill_bound_cnt > bestill->ill_bound_cnt) { 662 8485 Peter bestill = ill; 663 8485 Peter } 664 8485 Peter } 665 8485 Peter return (bestill); 666 8485 Peter } 667 8485 Peter 668 8485 Peter /* 669 8485 Peter * Return the active ill with the fewest number of data addresses in `illg'. 670 8485 Peter */ 671 8485 Peter static ill_t * 672 8485 Peter ipmp_illgrp_min_ill(ipmp_illgrp_t *illg) 673 8485 Peter { 674 8485 Peter ill_t *ill, *bestill = NULL; 675 8485 Peter 676 8485 Peter ASSERT(IAM_WRITER_ILL(illg->ig_ipmp_ill)); 677 8485 Peter 678 8485 Peter ill = list_head(&illg->ig_actif); 679 8485 Peter for (; ill != NULL; ill = list_next(&illg->ig_actif, ill)) { 680 8485 Peter if (bestill == NULL || 681 8485 Peter ill->ill_bound_cnt < bestill->ill_bound_cnt) { 682 8485 Peter if (ill->ill_bound_cnt == 0) 683 8485 Peter return (ill); /* can't get better */ 684 8485 Peter bestill = ill; 685 8485 Peter } 686 8485 Peter } 687 8485 Peter return (bestill); 688 8485 Peter } 689 8485 Peter 690 8485 Peter /* 691 8485 Peter * Return a pointer to IPMP meta-interface for `illg' (which must exist). 692 8485 Peter * Since ig_ipmp_ill never changes for a given illg, no locks are needed. 693 8485 Peter */ 694 8485 Peter ill_t * 695 8485 Peter ipmp_illgrp_ipmp_ill(ipmp_illgrp_t *illg) 696 8485 Peter { 697 8485 Peter return (illg->ig_ipmp_ill); 698 8485 Peter } 699 8485 Peter 700 8485 Peter /* 701 8485 Peter * Return a pointer to the next available underlying ill in `illg', or NULL if 702 8485 Peter * one doesn't exist. Caller must be inside the IPSQ. 703 8485 Peter */ 704 8485 Peter ill_t * 705 8485 Peter ipmp_illgrp_next_ill(ipmp_illgrp_t *illg) 706 8485 Peter { 707 8485 Peter ill_t *ill; 708 8485 Peter ip_stack_t *ipst = IPMP_ILLGRP_TO_IPST(illg); 709 8485 Peter 710 8485 Peter ASSERT(IAM_WRITER_ILL(illg->ig_ipmp_ill)); 711 8485 Peter 712 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_WRITER); 713 8485 Peter if ((ill = illg->ig_next_ill) != NULL) { 714 8485 Peter illg->ig_next_ill = list_next(&illg->ig_actif, ill); 715 8485 Peter if (illg->ig_next_ill == NULL) 716 8485 Peter illg->ig_next_ill = list_head(&illg->ig_actif); 717 8485 Peter } 718 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 719 8485 Peter 720 8485 Peter return (ill); 721 8485 Peter } 722 8485 Peter 723 8485 Peter /* 724 8485 Peter * Return a held pointer to the next available underlying ill in `illg', or 725 8485 Peter * NULL if one doesn't exist. Caller need not be inside the IPSQ. 726 8485 Peter */ 727 8485 Peter ill_t * 728 8485 Peter ipmp_illgrp_hold_next_ill(ipmp_illgrp_t *illg) 729 8485 Peter { 730 8485 Peter ill_t *ill; 731 8485 Peter uint_t i; 732 8485 Peter ip_stack_t *ipst = IPMP_ILLGRP_TO_IPST(illg); 733 8485 Peter 734 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_WRITER); 735 8485 Peter for (i = 0; i < illg->ig_nactif; i++) { 736 8485 Peter ill = illg->ig_next_ill; 737 8485 Peter illg->ig_next_ill = list_next(&illg->ig_actif, ill); 738 8485 Peter if (illg->ig_next_ill == NULL) 739 8485 Peter illg->ig_next_ill = list_head(&illg->ig_actif); 740 8485 Peter 741 11042 Erik if (ill_check_and_refhold(ill)) { 742 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 743 8485 Peter return (ill); 744 8485 Peter } 745 8485 Peter } 746 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 747 8485 Peter 748 8485 Peter return (NULL); 749 8485 Peter } 750 8485 Peter 751 8485 Peter /* 752 8485 Peter * Return a held pointer to the nominated multicast ill in `illg', or NULL if 753 8485 Peter * one doesn't exist. Caller need not be inside the IPSQ. 754 8485 Peter */ 755 8485 Peter ill_t * 756 8485 Peter ipmp_illgrp_hold_cast_ill(ipmp_illgrp_t *illg) 757 8485 Peter { 758 8485 Peter ill_t *castill; 759 8485 Peter ip_stack_t *ipst = IPMP_ILLGRP_TO_IPST(illg); 760 8485 Peter 761 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_READER); 762 8485 Peter castill = illg->ig_cast_ill; 763 11042 Erik if (castill != NULL && ill_check_and_refhold(castill)) { 764 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 765 8485 Peter return (castill); 766 8485 Peter } 767 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 768 8485 Peter return (NULL); 769 11042 Erik } 770 11042 Erik 771 11042 Erik /* 772 11042 Erik * Callback routine for ncec_walk() that deletes `nce' if it is associated with 773 11042 Erik * the `(ill_t *)arg' and it is not one of the local addresses. Caller must be 774 11042 Erik * inside the IPSQ. 775 11042 Erik */ 776 11042 Erik static void 777 11042 Erik ipmp_ncec_delete_nonlocal(ncec_t *ncec, uchar_t *arg) 778 11042 Erik { 779 11042 Erik if ((ncec != NULL) && !NCE_MYADDR(ncec) && 780 11042 Erik ncec->ncec_ill == (ill_t *)arg) { 781 11042 Erik ncec_delete(ncec); 782 11042 Erik } 783 8485 Peter } 784 8485 Peter 785 8485 Peter /* 786 8485 Peter * Set the nominated cast ill on `illg' to `castill'. If `castill' is NULL, 787 8485 Peter * any existing nomination is removed. Caller must be inside the IPSQ. 788 8485 Peter */ 789 8485 Peter static void 790 8485 Peter ipmp_illgrp_set_cast(ipmp_illgrp_t *illg, ill_t *castill) 791 8485 Peter { 792 8485 Peter ill_t *ocastill = illg->ig_cast_ill; 793 8485 Peter ill_t *ipmp_ill = illg->ig_ipmp_ill; 794 8485 Peter ip_stack_t *ipst = IPMP_ILLGRP_TO_IPST(illg); 795 8485 Peter 796 8485 Peter ASSERT(IAM_WRITER_ILL(ipmp_ill)); 797 8485 Peter 798 8485 Peter /* 799 8485 Peter * Disable old nominated ill (if any). 800 8485 Peter */ 801 8485 Peter if (ocastill != NULL) { 802 8485 Peter DTRACE_PROBE2(ipmp__illgrp__cast__disable, ipmp_illgrp_t *, 803 8485 Peter illg, ill_t *, ocastill); 804 8485 Peter ASSERT(ocastill->ill_nom_cast); 805 8485 Peter ocastill->ill_nom_cast = B_FALSE; 806 8485 Peter /* 807 8485 Peter * If the IPMP meta-interface is down, we never did the join, 808 8485 Peter * so we must not try to leave. 809 8485 Peter */ 810 8485 Peter if (ipmp_ill->ill_dl_up) 811 8485 Peter ill_leave_multicast(ipmp_ill); 812 11042 Erik 813 11042 Erik /* 814 11042 Erik * Delete any NCEs tied to the old nomination. We must do this 815 11042 Erik * last since ill_leave_multicast() may trigger IREs to be 816 11042 Erik * built using ig_cast_ill. 817 11042 Erik */ 818 11042 Erik ncec_walk(ocastill, (pfi_t)ipmp_ncec_delete_nonlocal, ocastill, 819 11042 Erik ocastill->ill_ipst); 820 8485 Peter } 821 8485 Peter 822 8485 Peter /* 823 8485 Peter * Set new nomination. 824 8485 Peter */ 825 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_WRITER); 826 8485 Peter illg->ig_cast_ill = castill; 827 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 828 8485 Peter 829 8485 Peter /* 830 8485 Peter * Enable new nominated ill (if any). 831 8485 Peter */ 832 8485 Peter if (castill != NULL) { 833 8485 Peter DTRACE_PROBE2(ipmp__illgrp__cast__enable, ipmp_illgrp_t *, 834 8485 Peter illg, ill_t *, castill); 835 8485 Peter ASSERT(!castill->ill_nom_cast); 836 8485 Peter castill->ill_nom_cast = B_TRUE; 837 8485 Peter /* 838 8485 Peter * If the IPMP meta-interface is down, the attempt to recover 839 8485 Peter * will silently fail but ill_need_recover_multicast will be 840 8485 Peter * erroneously cleared -- so check first. 841 8485 Peter */ 842 8485 Peter if (ipmp_ill->ill_dl_up) 843 8485 Peter ill_recover_multicast(ipmp_ill); 844 8485 Peter } 845 8485 Peter } 846 8485 Peter 847 8485 Peter /* 848 8485 Peter * Create an IPMP ARP entry and add it to the set tracked on `illg'. If an 849 8485 Peter * entry for the same IP address already exists, destroy it first. Return the 850 8485 Peter * created IPMP ARP entry, or NULL on failure. 851 8485 Peter */ 852 8485 Peter ipmp_arpent_t * 853 11042 Erik ipmp_illgrp_create_arpent(ipmp_illgrp_t *illg, boolean_t proxyarp, 854 11042 Erik ipaddr_t ipaddr, uchar_t *lladdr, size_t lladdr_len, uint16_t flags) 855 8485 Peter { 856 8485 Peter ipmp_arpent_t *entp, *oentp; 857 8485 Peter 858 8485 Peter ASSERT(IAM_WRITER_ILL(illg->ig_ipmp_ill)); 859 8485 Peter 860 11042 Erik if ((entp = kmem_alloc(sizeof (ipmp_arpent_t) + lladdr_len, 861 11042 Erik KM_NOSLEEP)) == NULL) 862 8485 Peter return (NULL); 863 8485 Peter 864 11042 Erik /* 865 11042 Erik * Delete any existing ARP entry for this address. 866 11042 Erik */ 867 8485 Peter if ((oentp = ipmp_illgrp_lookup_arpent(illg, &entp->ia_ipaddr)) != NULL) 868 8485 Peter ipmp_illgrp_destroy_arpent(illg, oentp); 869 8485 Peter 870 11042 Erik /* 871 11042 Erik * Prepend the new entry. 872 11042 Erik */ 873 11042 Erik entp->ia_ipaddr = ipaddr; 874 11042 Erik entp->ia_flags = flags; 875 11042 Erik entp->ia_lladdr_len = lladdr_len; 876 11042 Erik entp->ia_lladdr = (uchar_t *)&entp[1]; 877 11042 Erik bcopy(lladdr, entp->ia_lladdr, lladdr_len); 878 11042 Erik entp->ia_proxyarp = proxyarp; 879 11042 Erik entp->ia_notified = B_TRUE; 880 8485 Peter list_insert_head(&illg->ig_arpent, entp); 881 8485 Peter return (entp); 882 8485 Peter } 883 8485 Peter 884 8485 Peter /* 885 8485 Peter * Remove IPMP ARP entry `entp' from the set tracked on `illg' and destroy it. 886 8485 Peter */ 887 8485 Peter void 888 8485 Peter ipmp_illgrp_destroy_arpent(ipmp_illgrp_t *illg, ipmp_arpent_t *entp) 889 8485 Peter { 890 8485 Peter ASSERT(IAM_WRITER_ILL(illg->ig_ipmp_ill)); 891 8485 Peter 892 8485 Peter list_remove(&illg->ig_arpent, entp); 893 11042 Erik kmem_free(entp, sizeof (ipmp_arpent_t) + entp->ia_lladdr_len); 894 8485 Peter } 895 8485 Peter 896 8485 Peter /* 897 8485 Peter * Mark that ARP has been notified about the IP address on `entp'; `illg' is 898 8485 Peter * taken as a debugging aid for DTrace FBT probes. 899 8485 Peter */ 900 8485 Peter /* ARGSUSED */ 901 8485 Peter void 902 8485 Peter ipmp_illgrp_mark_arpent(ipmp_illgrp_t *illg, ipmp_arpent_t *entp) 903 8485 Peter { 904 8485 Peter entp->ia_notified = B_TRUE; 905 8485 Peter } 906 8485 Peter 907 8485 Peter /* 908 8485 Peter * Look up the IPMP ARP entry for IP address `addrp' on `illg'; if `addrp' is 909 8485 Peter * NULL, any IPMP ARP entry is requested. Return NULL if it does not exist. 910 8485 Peter */ 911 8485 Peter ipmp_arpent_t * 912 8485 Peter ipmp_illgrp_lookup_arpent(ipmp_illgrp_t *illg, ipaddr_t *addrp) 913 8485 Peter { 914 8485 Peter ipmp_arpent_t *entp = list_head(&illg->ig_arpent); 915 8485 Peter 916 8485 Peter ASSERT(IAM_WRITER_ILL(illg->ig_ipmp_ill)); 917 8485 Peter 918 8485 Peter if (addrp == NULL) 919 8485 Peter return (entp); 920 8485 Peter 921 8485 Peter for (; entp != NULL; entp = list_next(&illg->ig_arpent, entp)) 922 8485 Peter if (entp->ia_ipaddr == *addrp) 923 8485 Peter break; 924 8485 Peter return (entp); 925 8485 Peter } 926 8485 Peter 927 8485 Peter /* 928 8485 Peter * Refresh ARP entries on `illg' to be distributed across its active 929 8485 Peter * interfaces. Entries that cannot be refreshed (e.g., because there are no 930 8485 Peter * active interfaces) are marked so that subsequent calls can try again. 931 8485 Peter */ 932 8485 Peter void 933 8485 Peter ipmp_illgrp_refresh_arpent(ipmp_illgrp_t *illg) 934 8485 Peter { 935 8485 Peter ill_t *ill, *ipmp_ill = illg->ig_ipmp_ill; 936 8485 Peter uint_t paddrlen = ipmp_ill->ill_phys_addr_length; 937 8485 Peter ipmp_arpent_t *entp; 938 11042 Erik ncec_t *ncec; 939 11042 Erik nce_t *nce; 940 8485 Peter 941 8485 Peter ASSERT(IAM_WRITER_ILL(ipmp_ill)); 942 8485 Peter ASSERT(!ipmp_ill->ill_isv6); 943 8485 Peter 944 8485 Peter ill = list_head(&illg->ig_actif); 945 8485 Peter entp = list_head(&illg->ig_arpent); 946 8485 Peter for (; entp != NULL; entp = list_next(&illg->ig_arpent, entp)) { 947 8485 Peter if (ill == NULL || ipmp_ill->ill_ipif_up_count == 0) { 948 8485 Peter entp->ia_notified = B_FALSE; 949 8485 Peter continue; 950 8485 Peter } 951 8485 Peter 952 8485 Peter ASSERT(paddrlen == ill->ill_phys_addr_length); 953 8485 Peter 954 8485 Peter /* 955 8485 Peter * If this is a proxy ARP entry, we can skip notifying ARP if 956 8485 Peter * the entry is already up-to-date. If it has changed, we 957 8485 Peter * update the entry's hardware address before notifying ARP. 958 8485 Peter */ 959 8485 Peter if (entp->ia_proxyarp) { 960 11042 Erik if (bcmp(ill->ill_phys_addr, entp->ia_lladdr, 961 11042 Erik paddrlen) == 0 && entp->ia_notified) 962 8485 Peter continue; 963 11042 Erik bcopy(ill->ill_phys_addr, entp->ia_lladdr, paddrlen); 964 8485 Peter } 965 8485 Peter 966 11042 Erik (void) nce_lookup_then_add_v4(ipmp_ill, entp->ia_lladdr, 967 11042 Erik paddrlen, &entp->ia_ipaddr, entp->ia_flags, ND_UNCHANGED, 968 11042 Erik &nce); 969 11042 Erik if (nce == NULL || !entp->ia_proxyarp) { 970 11042 Erik if (nce != NULL) 971 11042 Erik nce_refrele(nce); 972 8485 Peter continue; 973 8485 Peter } 974 11042 Erik ncec = nce->nce_common; 975 11042 Erik mutex_enter(&ncec->ncec_lock); 976 11042 Erik nce_update(ncec, ND_UNCHANGED, ill->ill_phys_addr); 977 11042 Erik mutex_exit(&ncec->ncec_lock); 978 11042 Erik nce_refrele(nce); 979 8485 Peter ipmp_illgrp_mark_arpent(illg, entp); 980 8485 Peter 981 8485 Peter if ((ill = list_next(&illg->ig_actif, ill)) == NULL) 982 8485 Peter ill = list_head(&illg->ig_actif); 983 8485 Peter } 984 8485 Peter } 985 8485 Peter 986 8485 Peter /* 987 8485 Peter * Return an interface in `illg' with the specified `physaddr', or NULL if one 988 8485 Peter * doesn't exist. Caller must hold ill_g_lock if it's not inside the IPSQ. 989 8485 Peter */ 990 8485 Peter ill_t * 991 8485 Peter ipmp_illgrp_find_ill(ipmp_illgrp_t *illg, uchar_t *physaddr, uint_t paddrlen) 992 8485 Peter { 993 8485 Peter ill_t *ill; 994 8485 Peter ill_t *ipmp_ill = illg->ig_ipmp_ill; 995 8485 Peter ip_stack_t *ipst = IPMP_ILLGRP_TO_IPST(illg); 996 8485 Peter 997 8485 Peter ASSERT(IAM_WRITER_ILL(ipmp_ill) || RW_LOCK_HELD(&ipst->ips_ill_g_lock)); 998 8485 Peter 999 8485 Peter ill = list_head(&illg->ig_if); 1000 8485 Peter for (; ill != NULL; ill = list_next(&illg->ig_if, ill)) { 1001 8485 Peter if (ill->ill_phys_addr_length == paddrlen && 1002 8485 Peter bcmp(ill->ill_phys_addr, physaddr, paddrlen) == 0) 1003 8485 Peter return (ill); 1004 8485 Peter } 1005 8485 Peter return (NULL); 1006 8485 Peter } 1007 8485 Peter 1008 8485 Peter /* 1009 8485 Peter * Asynchronously update the MTU for an IPMP ill by injecting a DL_NOTIFY_IND. 1010 8485 Peter * Caller must be inside the IPSQ unless this is initialization. 1011 8485 Peter */ 1012 8485 Peter static void 1013 8485 Peter ipmp_illgrp_set_mtu(ipmp_illgrp_t *illg, uint_t mtu) 1014 8485 Peter { 1015 8485 Peter ill_t *ill = illg->ig_ipmp_ill; 1016 8485 Peter mblk_t *mp; 1017 8485 Peter 1018 8485 Peter ASSERT(illg->ig_mtu == 0 || IAM_WRITER_ILL(ill)); 1019 8485 Peter 1020 8485 Peter /* 1021 8485 Peter * If allocation fails, we have bigger problems than MTU. 1022 8485 Peter */ 1023 8485 Peter if ((mp = ip_dlnotify_alloc(DL_NOTE_SDU_SIZE, mtu)) != NULL) { 1024 8485 Peter illg->ig_mtu = mtu; 1025 8485 Peter put(ill->ill_rq, mp); 1026 8485 Peter } 1027 8485 Peter } 1028 8485 Peter 1029 8485 Peter /* 1030 8485 Peter * Recalculate the IPMP group MTU for `illg', and update its associated IPMP 1031 8485 Peter * ill MTU if necessary. 1032 8485 Peter */ 1033 8485 Peter void 1034 8485 Peter ipmp_illgrp_refresh_mtu(ipmp_illgrp_t *illg) 1035 8485 Peter { 1036 8485 Peter ill_t *ill; 1037 8485 Peter ill_t *ipmp_ill = illg->ig_ipmp_ill; 1038 8485 Peter uint_t mtu = 0; 1039 8485 Peter 1040 8485 Peter ASSERT(IAM_WRITER_ILL(ipmp_ill)); 1041 8485 Peter 1042 8485 Peter /* 1043 11042 Erik * Since ill_mtu can only change under ill_lock, we hold ill_lock 1044 8485 Peter * for each ill as we iterate through the list. Any changes to the 1045 11042 Erik * ill_mtu will also trigger an update, so even if we missed it 1046 8485 Peter * this time around, the update will catch it. 1047 8485 Peter */ 1048 8485 Peter ill = list_head(&illg->ig_if); 1049 8485 Peter for (; ill != NULL; ill = list_next(&illg->ig_if, ill)) { 1050 8485 Peter mutex_enter(&ill->ill_lock); 1051 11042 Erik if (mtu == 0 || ill->ill_mtu < mtu) 1052 11042 Erik mtu = ill->ill_mtu; 1053 8485 Peter mutex_exit(&ill->ill_lock); 1054 8485 Peter } 1055 8485 Peter 1056 8485 Peter /* 1057 8485 Peter * MTU must be at least the minimum MTU. 1058 8485 Peter */ 1059 8485 Peter mtu = MAX(mtu, ipmp_ill->ill_isv6 ? IPV6_MIN_MTU : IP_MIN_MTU); 1060 8485 Peter 1061 8485 Peter if (illg->ig_mtu != mtu) 1062 8485 Peter ipmp_illgrp_set_mtu(illg, mtu); 1063 8485 Peter } 1064 8485 Peter 1065 8485 Peter /* 1066 8485 Peter * Link illgrp `illg' to IPMP group `grp'. To simplify the caller, silently 1067 8485 Peter * allow the same link to be established more than once. 1068 8485 Peter */ 1069 8485 Peter void 1070 8485 Peter ipmp_illgrp_link_grp(ipmp_illgrp_t *illg, ipmp_grp_t *grp) 1071 8485 Peter { 1072 8485 Peter ip_stack_t *ipst = IPMP_ILLGRP_TO_IPST(illg); 1073 8485 Peter 1074 8485 Peter ASSERT(RW_WRITE_HELD(&ipst->ips_ipmp_lock)); 1075 8485 Peter 1076 8485 Peter if (illg->ig_ipmp_ill->ill_isv6) { 1077 8485 Peter ASSERT(grp->gr_v6 == NULL || grp->gr_v6 == illg); 1078 8485 Peter grp->gr_v6 = illg; 1079 8485 Peter } else { 1080 8485 Peter ASSERT(grp->gr_v4 == NULL || grp->gr_v4 == illg); 1081 8485 Peter grp->gr_v4 = illg; 1082 8485 Peter } 1083 8485 Peter } 1084 8485 Peter 1085 8485 Peter /* 1086 8485 Peter * Unlink illgrp `illg' from its IPMP group. Return an errno if the illgrp 1087 8485 Peter * cannot be unlinked (e.g., because there are still interfaces using it). 1088 8485 Peter */ 1089 8485 Peter int 1090 8485 Peter ipmp_illgrp_unlink_grp(ipmp_illgrp_t *illg) 1091 8485 Peter { 1092 8485 Peter ipmp_grp_t *grp = illg->ig_ipmp_ill->ill_phyint->phyint_grp; 1093 8485 Peter ip_stack_t *ipst = IPMP_ILLGRP_TO_IPST(illg); 1094 8485 Peter 1095 8485 Peter ASSERT(RW_WRITE_HELD(&ipst->ips_ipmp_lock)); 1096 8485 Peter 1097 8485 Peter if (illg->ig_ipmp_ill->ill_isv6) { 1098 8485 Peter if (grp->gr_nv6 + grp->gr_pendv6 != 0) 1099 8485 Peter return (EBUSY); 1100 8485 Peter grp->gr_v6 = NULL; 1101 8485 Peter } else { 1102 8485 Peter if (grp->gr_nv4 + grp->gr_pendv4 != 0) 1103 8485 Peter return (EBUSY); 1104 8485 Peter grp->gr_v4 = NULL; 1105 8485 Peter } 1106 8485 Peter return (0); 1107 8485 Peter } 1108 8485 Peter 1109 8485 Peter /* 1110 8485 Peter * Place `ill' into `illg', and rebalance the data addresses on `illg' 1111 8485 Peter * to be spread evenly across the ills now in it. Also, adjust the IPMP 1112 8485 Peter * ill as necessary to account for `ill' (e.g., MTU). 1113 8485 Peter */ 1114 8485 Peter void 1115 8485 Peter ipmp_ill_join_illgrp(ill_t *ill, ipmp_illgrp_t *illg) 1116 8485 Peter { 1117 8485 Peter ill_t *ipmp_ill; 1118 8485 Peter ipif_t *ipif; 1119 8485 Peter ip_stack_t *ipst = ill->ill_ipst; 1120 8485 Peter 1121 8485 Peter /* IS_UNDER_IPMP() requires ill_grp to be non-NULL */ 1122 8485 Peter ASSERT(!IS_IPMP(ill) && ill->ill_phyint->phyint_grp != NULL); 1123 8485 Peter ASSERT(IAM_WRITER_ILL(ill)); 1124 8485 Peter ASSERT(ill->ill_grp == NULL); 1125 8485 Peter 1126 8485 Peter ipmp_ill = illg->ig_ipmp_ill; 1127 8485 Peter 1128 8485 Peter /* 1129 8485 Peter * Account for `ill' joining the illgrp. 1130 8485 Peter */ 1131 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_WRITER); 1132 8485 Peter if (ill->ill_isv6) 1133 8485 Peter ill->ill_phyint->phyint_grp->gr_nv6++; 1134 8485 Peter else 1135 8485 Peter ill->ill_phyint->phyint_grp->gr_nv4++; 1136 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 1137 8485 Peter 1138 8485 Peter /* 1139 8485 Peter * Ensure the ILLF_ROUTER flag remains consistent across the group. 1140 8485 Peter */ 1141 8485 Peter mutex_enter(&ill->ill_lock); 1142 8485 Peter if (ipmp_ill->ill_flags & ILLF_ROUTER) 1143 8485 Peter ill->ill_flags |= ILLF_ROUTER; 1144 8485 Peter else 1145 8485 Peter ill->ill_flags &= ~ILLF_ROUTER; 1146 8485 Peter mutex_exit(&ill->ill_lock); 1147 8485 Peter 1148 8485 Peter /* 1149 8485 Peter * Blow away all multicast memberships that currently exist on `ill'. 1150 8485 Peter * This may seem odd, but it's consistent with the application view 1151 8485 Peter * that `ill' no longer exists (e.g., due to ipmp_ill_rtsaddrmsg()). 1152 8485 Peter */ 1153 11042 Erik update_conn_ill(ill, ill->ill_ipst); 1154 8485 Peter if (ill->ill_isv6) { 1155 8485 Peter reset_mrt_ill(ill); 1156 8485 Peter } else { 1157 8485 Peter ipif = ill->ill_ipif; 1158 8485 Peter for (; ipif != NULL; ipif = ipif->ipif_next) { 1159 8485 Peter reset_mrt_vif_ipif(ipif); 1160 8485 Peter } 1161 8485 Peter } 1162 8485 Peter ip_purge_allmulti(ill); 1163 8485 Peter 1164 8485 Peter /* 1165 8485 Peter * Borrow the first ill's ill_phys_addr_length value for the illgrp's 1166 8485 Peter * physical address length. All other ills must have the same value, 1167 8485 Peter * since they are required to all be the same mactype. Also update 1168 8485 Peter * the IPMP ill's MTU and CoS marking, if necessary. 1169 8485 Peter */ 1170 8485 Peter if (list_is_empty(&illg->ig_if)) { 1171 8485 Peter ASSERT(ipmp_ill->ill_phys_addr_length == 0); 1172 8485 Peter /* 1173 8485 Peter * NOTE: we leave ill_phys_addr NULL since the IPMP group 1174 8485 Peter * doesn't have a physical address. This means that code must 1175 8485 Peter * not assume that ill_phys_addr is non-NULL just because 1176 8485 Peter * ill_phys_addr_length is non-zero. Likewise for ill_nd_lla. 1177 8485 Peter */ 1178 8485 Peter ipmp_ill->ill_phys_addr_length = ill->ill_phys_addr_length; 1179 8485 Peter ipmp_ill->ill_nd_lla_len = ill->ill_phys_addr_length; 1180 8485 Peter ipmp_ill->ill_type = ill->ill_type; 1181 8485 Peter 1182 8485 Peter if (ill->ill_flags & ILLF_COS_ENABLED) { 1183 8485 Peter mutex_enter(&ipmp_ill->ill_lock); 1184 8485 Peter ipmp_ill->ill_flags |= ILLF_COS_ENABLED; 1185 8485 Peter mutex_exit(&ipmp_ill->ill_lock); 1186 8485 Peter } 1187 11042 Erik ipmp_illgrp_set_mtu(illg, ill->ill_mtu); 1188 8485 Peter } else { 1189 8485 Peter ASSERT(ipmp_ill->ill_phys_addr_length == 1190 8485 Peter ill->ill_phys_addr_length); 1191 8485 Peter ASSERT(ipmp_ill->ill_type == ill->ill_type); 1192 8485 Peter 1193 8485 Peter if (!(ill->ill_flags & ILLF_COS_ENABLED)) { 1194 8485 Peter mutex_enter(&ipmp_ill->ill_lock); 1195 8485 Peter ipmp_ill->ill_flags &= ~ILLF_COS_ENABLED; 1196 8485 Peter mutex_exit(&ipmp_ill->ill_lock); 1197 8485 Peter } 1198 11042 Erik if (illg->ig_mtu > ill->ill_mtu) 1199 11042 Erik ipmp_illgrp_set_mtu(illg, ill->ill_mtu); 1200 8485 Peter } 1201 8485 Peter 1202 8485 Peter rw_enter(&ipst->ips_ill_g_lock, RW_WRITER); 1203 8485 Peter list_insert_tail(&illg->ig_if, ill); 1204 8485 Peter ill->ill_grp = illg; 1205 8485 Peter rw_exit(&ipst->ips_ill_g_lock); 1206 8485 Peter 1207 8485 Peter /* 1208 8485 Peter * Hide the IREs on `ill' so that we don't accidentally find them when 1209 8485 Peter * sending data traffic. 1210 8485 Peter */ 1211 8485 Peter ire_walk_ill(MATCH_IRE_ILL, 0, ipmp_ill_ire_mark_testhidden, ill, ill); 1212 8485 Peter 1213 8485 Peter ipmp_ill_refresh_active(ill); 1214 8485 Peter } 1215 8485 Peter 1216 8485 Peter /* 1217 8485 Peter * Remove `ill' from its illgrp, and rebalance the data addresses in that 1218 8485 Peter * illgrp to be spread evenly across the remaining ills. Also, adjust the 1219 8485 Peter * IPMP ill as necessary now that `ill' is removed (e.g., MTU). 1220 8485 Peter */ 1221 8485 Peter void 1222 8485 Peter ipmp_ill_leave_illgrp(ill_t *ill) 1223 8485 Peter { 1224 8485 Peter ill_t *ipmp_ill; 1225 8485 Peter ipif_t *ipif; 1226 8485 Peter ipmp_arpent_t *entp; 1227 8485 Peter ipmp_illgrp_t *illg = ill->ill_grp; 1228 8485 Peter ip_stack_t *ipst = IPMP_ILLGRP_TO_IPST(illg); 1229 8485 Peter 1230 8485 Peter ASSERT(IS_UNDER_IPMP(ill)); 1231 8485 Peter ASSERT(IAM_WRITER_ILL(ill)); 1232 8485 Peter ASSERT(illg != NULL); 1233 8485 Peter 1234 8485 Peter ipmp_ill = illg->ig_ipmp_ill; 1235 8485 Peter 1236 8485 Peter /* 1237 8485 Peter * Cancel IPMP-specific ill timeouts. 1238 8485 Peter */ 1239 8485 Peter (void) untimeout(ill->ill_refresh_tid); 1240 8485 Peter 1241 8485 Peter /* 1242 8485 Peter * Expose any previously-hidden IREs on `ill'. 1243 8485 Peter */ 1244 8485 Peter ire_walk_ill(MATCH_IRE_ILL, 0, ipmp_ill_ire_clear_testhidden, ill, ill); 1245 8485 Peter 1246 8485 Peter /* 1247 8485 Peter * Ensure the multicast state for each ipif on `ill' is down so that 1248 8485 Peter * our ipif_multicast_up() (once `ill' leaves the group) will rejoin 1249 8485 Peter * all eligible groups. 1250 8485 Peter */ 1251 8485 Peter for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) 1252 8485 Peter if (ipif->ipif_flags & IPIF_UP) 1253 8485 Peter ipif_multicast_down(ipif); 1254 8485 Peter 1255 8485 Peter /* 1256 8485 Peter * Account for `ill' leaving the illgrp. 1257 8485 Peter */ 1258 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_WRITER); 1259 8485 Peter if (ill->ill_isv6) 1260 8485 Peter ill->ill_phyint->phyint_grp->gr_nv6--; 1261 8485 Peter else 1262 8485 Peter ill->ill_phyint->phyint_grp->gr_nv4--; 1263 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 1264 8485 Peter 1265 8485 Peter /* 1266 8485 Peter * Pull `ill' out of the interface lists. 1267 8485 Peter */ 1268 8485 Peter if (list_link_active(&ill->ill_actnode)) 1269 8485 Peter ipmp_ill_deactivate(ill); 1270 8485 Peter rw_enter(&ipst->ips_ill_g_lock, RW_WRITER); 1271 8485 Peter list_remove(&illg->ig_if, ill); 1272 8485 Peter ill->ill_grp = NULL; 1273 8485 Peter rw_exit(&ipst->ips_ill_g_lock); 1274 8485 Peter 1275 8485 Peter /* 1276 8485 Peter * Re-establish multicast memberships that were previously being 1277 8485 Peter * handled by the IPMP meta-interface. 1278 8485 Peter */ 1279 8485 Peter for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) 1280 8485 Peter if (ipif->ipif_flags & IPIF_UP) 1281 8485 Peter ipif_multicast_up(ipif); 1282 8485 Peter 1283 8485 Peter /* 1284 8485 Peter * Refresh the group MTU based on the new interface list. 1285 8485 Peter */ 1286 8485 Peter ipmp_illgrp_refresh_mtu(illg); 1287 8485 Peter 1288 8485 Peter if (list_is_empty(&illg->ig_if)) { 1289 8485 Peter /* 1290 8485 Peter * No ills left in the illgrp; we no longer have a physical 1291 8485 Peter * address length, nor can we support ARP, CoS, or anything 1292 8485 Peter * else that depends on knowing the link layer type. 1293 8485 Peter */ 1294 8485 Peter while ((entp = ipmp_illgrp_lookup_arpent(illg, NULL)) != NULL) 1295 8485 Peter ipmp_illgrp_destroy_arpent(illg, entp); 1296 8485 Peter 1297 8485 Peter ipmp_ill->ill_phys_addr_length = 0; 1298 8485 Peter ipmp_ill->ill_nd_lla_len = 0; 1299 8485 Peter ipmp_ill->ill_type = IFT_OTHER; 1300 8485 Peter mutex_enter(&ipmp_ill->ill_lock); 1301 8485 Peter ipmp_ill->ill_flags &= ~ILLF_COS_ENABLED; 1302 8485 Peter mutex_exit(&ipmp_ill->ill_lock); 1303 8485 Peter } else { 1304 8485 Peter /* 1305 8485 Peter * If `ill' didn't support CoS, see if it can now be enabled. 1306 8485 Peter */ 1307 8485 Peter if (!(ill->ill_flags & ILLF_COS_ENABLED)) { 1308 8485 Peter ASSERT(!(ipmp_ill->ill_flags & ILLF_COS_ENABLED)); 1309 8485 Peter 1310 8485 Peter ill = list_head(&illg->ig_if); 1311 8485 Peter do { 1312 8485 Peter if (!(ill->ill_flags & ILLF_COS_ENABLED)) 1313 8485 Peter break; 1314 8485 Peter } while ((ill = list_next(&illg->ig_if, ill)) != NULL); 1315 8485 Peter 1316 8485 Peter if (ill == NULL) { 1317 8485 Peter mutex_enter(&ipmp_ill->ill_lock); 1318 8485 Peter ipmp_ill->ill_flags |= ILLF_COS_ENABLED; 1319 8485 Peter mutex_exit(&ipmp_ill->ill_lock); 1320 8485 Peter } 1321 8485 Peter } 1322 8485 Peter } 1323 8485 Peter } 1324 8485 Peter 1325 8485 Peter /* 1326 8485 Peter * Check if `ill' should be active, and activate or deactivate if need be. 1327 8485 Peter * Return B_FALSE if a refresh was necessary but could not be performed. 1328 8485 Peter */ 1329 8485 Peter static boolean_t 1330 8485 Peter ipmp_ill_try_refresh_active(ill_t *ill) 1331 8485 Peter { 1332 8485 Peter boolean_t refreshed = B_TRUE; 1333 8485 Peter 1334 8485 Peter ASSERT(IAM_WRITER_ILL(ill)); 1335 8485 Peter ASSERT(IS_UNDER_IPMP(ill)); 1336 8485 Peter 1337 8485 Peter if (ipmp_ill_is_active(ill)) { 1338 8485 Peter if (!list_link_active(&ill->ill_actnode)) 1339 8485 Peter refreshed = ipmp_ill_activate(ill); 1340 8485 Peter } else { 1341 8485 Peter if (list_link_active(&ill->ill_actnode)) 1342 8485 Peter ipmp_ill_deactivate(ill); 1343 8485 Peter } 1344 8485 Peter 1345 8485 Peter return (refreshed); 1346 8485 Peter } 1347 8485 Peter 1348 8485 Peter /* 1349 8485 Peter * Check if `ill' should be active, and activate or deactivate if need be. 1350 8485 Peter * If the refresh fails, schedule a timer to try again later. 1351 8485 Peter */ 1352 8485 Peter void 1353 8485 Peter ipmp_ill_refresh_active(ill_t *ill) 1354 8485 Peter { 1355 8485 Peter if (!ipmp_ill_try_refresh_active(ill)) 1356 8485 Peter ipmp_ill_refresh_active_timer_start(ill); 1357 8485 Peter } 1358 8485 Peter 1359 8485 Peter /* 1360 8485 Peter * Retry ipmp_ill_try_refresh_active() on the ill named by `ill_arg'. 1361 8485 Peter */ 1362 8485 Peter static void 1363 8485 Peter ipmp_ill_refresh_active_timer(void *ill_arg) 1364 8485 Peter { 1365 8485 Peter ill_t *ill = ill_arg; 1366 8485 Peter boolean_t refreshed = B_FALSE; 1367 8485 Peter 1368 8485 Peter /* 1369 8485 Peter * Clear ill_refresh_tid to indicate that no timeout is pending 1370 8485 Peter * (another thread could schedule a new timeout while we're still 1371 8485 Peter * running, but that's harmless). If the ill is going away, bail. 1372 8485 Peter */ 1373 8485 Peter mutex_enter(&ill->ill_lock); 1374 8485 Peter ill->ill_refresh_tid = 0; 1375 8485 Peter if (ill->ill_state_flags & ILL_CONDEMNED) { 1376 8485 Peter mutex_exit(&ill->ill_lock); 1377 8485 Peter return; 1378 8485 Peter } 1379 8485 Peter mutex_exit(&ill->ill_lock); 1380 8485 Peter 1381 8485 Peter if (ipsq_try_enter(NULL, ill, NULL, NULL, NULL, NEW_OP, B_FALSE)) { 1382 8485 Peter refreshed = ipmp_ill_try_refresh_active(ill); 1383 8485 Peter ipsq_exit(ill->ill_phyint->phyint_ipsq); 1384 8485 Peter } 1385 8485 Peter 1386 8485 Peter /* 1387 8485 Peter * If the refresh failed, schedule another attempt. 1388 8485 Peter */ 1389 8485 Peter if (!refreshed) 1390 8485 Peter ipmp_ill_refresh_active_timer_start(ill); 1391 8485 Peter } 1392 8485 Peter 1393 8485 Peter /* 1394 8485 Peter * Retry an ipmp_ill_try_refresh_active() on the ill named by `arg'. 1395 8485 Peter */ 1396 8485 Peter static void 1397 8485 Peter ipmp_ill_refresh_active_timer_start(ill_t *ill) 1398 8485 Peter { 1399 8485 Peter mutex_enter(&ill->ill_lock); 1400 8485 Peter 1401 8485 Peter /* 1402 8485 Peter * If the ill is going away or a refresh is already scheduled, bail. 1403 8485 Peter */ 1404 8485 Peter if (ill->ill_refresh_tid != 0 || 1405 8485 Peter (ill->ill_state_flags & ILL_CONDEMNED)) { 1406 8485 Peter mutex_exit(&ill->ill_lock); 1407 8485 Peter return; 1408 8485 Peter } 1409 8485 Peter 1410 8485 Peter ill->ill_refresh_tid = timeout(ipmp_ill_refresh_active_timer, ill, 1411 8485 Peter SEC_TO_TICK(IPMP_ILL_REFRESH_TIMEOUT)); 1412 8485 Peter 1413 8485 Peter mutex_exit(&ill->ill_lock); 1414 8485 Peter } 1415 8485 Peter 1416 8485 Peter /* 1417 8485 Peter * Activate `ill' so it will be used to send and receive data traffic. Return 1418 8485 Peter * B_FALSE if `ill' cannot be activated. Note that we allocate any messages 1419 8485 Peter * needed to deactivate `ill' here as well so that deactivation cannot fail. 1420 8485 Peter */ 1421 8485 Peter static boolean_t 1422 8485 Peter ipmp_ill_activate(ill_t *ill) 1423 8485 Peter { 1424 8485 Peter ipif_t *ipif; 1425 8485 Peter mblk_t *linkupmp = NULL, *linkdownmp = NULL; 1426 8485 Peter ipmp_grp_t *grp = ill->ill_phyint->phyint_grp; 1427 8485 Peter ipmp_illgrp_t *illg = ill->ill_grp; 1428 8485 Peter ill_t *maxill; 1429 8485 Peter ip_stack_t *ipst = IPMP_ILLGRP_TO_IPST(illg); 1430 8485 Peter 1431 8485 Peter ASSERT(IAM_WRITER_ILL(ill)); 1432 8485 Peter ASSERT(IS_UNDER_IPMP(ill)); 1433 8485 Peter 1434 8485 Peter /* 1435 8485 Peter * If this will be the first active interface in the group, allocate 1436 8485 Peter * the link-up and link-down messages. 1437 8485 Peter */ 1438 8485 Peter if (grp->gr_nactif == 0) { 1439 8485 Peter linkupmp = ip_dlnotify_alloc(DL_NOTE_LINK_UP, 0); 1440 8485 Peter linkdownmp = ip_dlnotify_alloc(DL_NOTE_LINK_DOWN, 0); 1441 8485 Peter if (linkupmp == NULL || linkdownmp == NULL) 1442 8485 Peter goto fail; 1443 8485 Peter } 1444 8485 Peter 1445 8485 Peter if (list_is_empty(&illg->ig_actif)) { 1446 8485 Peter /* 1447 8485 Peter * Now that we have an active ill, nominate it for multicast 1448 8485 Peter * and broadcast duties. Do this before ipmp_ill_bind_ipif() 1449 8485 Peter * since that may need to send multicast packets (e.g., IPv6 1450 8485 Peter * neighbor discovery probes). 1451 8485 Peter */ 1452 8485 Peter ipmp_illgrp_set_cast(illg, ill); 1453 8485 Peter 1454 8485 Peter /* 1455 8485 Peter * This is the first active ill in the illgrp -- add 'em all. 1456 8485 Peter * We can access/walk ig_ipmp_ill's ipif list since we're 1457 8485 Peter * writer on its IPSQ as well. 1458 8485 Peter */ 1459 8485 Peter ipif = illg->ig_ipmp_ill->ill_ipif; 1460 8485 Peter for (; ipif != NULL; ipif = ipif->ipif_next) 1461 8485 Peter if (ipmp_ipif_is_up_dataaddr(ipif)) 1462 8485 Peter ipmp_ill_bind_ipif(ill, ipif, Res_act_initial); 1463 8485 Peter } else { 1464 8485 Peter /* 1465 8485 Peter * Redistribute the addresses by moving them from the ill with 1466 8485 Peter * the most addresses until the ill being activated is at the 1467 8485 Peter * same level as the rest of the ills. 1468 8485 Peter */ 1469 8485 Peter for (;;) { 1470 8485 Peter maxill = ipmp_illgrp_max_ill(illg); 1471 8485 Peter ASSERT(maxill != NULL); 1472 8485 Peter if (ill->ill_bound_cnt + 1 >= maxill->ill_bound_cnt) 1473 8485 Peter break; 1474 8485 Peter ipif = ipmp_ill_unbind_ipif(maxill, NULL, B_TRUE); 1475 8485 Peter ipmp_ill_bind_ipif(ill, ipif, Res_act_rebind); 1476 8485 Peter } 1477 8485 Peter } 1478 8485 Peter 1479 8485 Peter /* 1480 8485 Peter * Put the interface in the active list. 1481 8485 Peter */ 1482 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_WRITER); 1483 8485 Peter list_insert_tail(&illg->ig_actif, ill); 1484 8485 Peter illg->ig_nactif++; 1485 8485 Peter illg->ig_next_ill = ill; 1486 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 1487 8485 Peter 1488 8485 Peter /* 1489 11042 Erik * Refresh static/proxy ARP entries to use `ill', if need be. 1490 8485 Peter */ 1491 8485 Peter if (!ill->ill_isv6) 1492 8485 Peter ipmp_illgrp_refresh_arpent(illg); 1493 8485 Peter 1494 8485 Peter /* 1495 8485 Peter * Finally, mark the group link up, if necessary. 1496 8485 Peter */ 1497 8485 Peter if (grp->gr_nactif++ == 0) { 1498 8485 Peter ASSERT(grp->gr_linkdownmp == NULL); 1499 8485 Peter grp->gr_linkdownmp = linkdownmp; 1500 8485 Peter put(illg->ig_ipmp_ill->ill_rq, linkupmp); 1501 8485 Peter } 1502 8485 Peter return (B_TRUE); 1503 8485 Peter fail: 1504 8485 Peter freemsg(linkupmp); 1505 8485 Peter freemsg(linkdownmp); 1506 8485 Peter return (B_FALSE); 1507 8485 Peter } 1508 8485 Peter 1509 8485 Peter /* 1510 8485 Peter * Deactivate `ill' so it will not be used to send or receive data traffic. 1511 8485 Peter */ 1512 8485 Peter static void 1513 8485 Peter ipmp_ill_deactivate(ill_t *ill) 1514 8485 Peter { 1515 8485 Peter ill_t *minill; 1516 8485 Peter ipif_t *ipif, *ubnextipif, *ubheadipif = NULL; 1517 8485 Peter mblk_t *mp; 1518 8485 Peter ipmp_grp_t *grp = ill->ill_phyint->phyint_grp; 1519 8485 Peter ipmp_illgrp_t *illg = ill->ill_grp; 1520 8485 Peter ip_stack_t *ipst = IPMP_ILLGRP_TO_IPST(illg); 1521 8485 Peter 1522 8485 Peter ASSERT(IAM_WRITER_ILL(ill)); 1523 8485 Peter ASSERT(IS_UNDER_IPMP(ill)); 1524 8485 Peter 1525 8485 Peter /* 1526 8485 Peter * Pull the interface out of the active list. 1527 8485 Peter */ 1528 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_WRITER); 1529 8485 Peter list_remove(&illg->ig_actif, ill); 1530 8485 Peter illg->ig_nactif--; 1531 8485 Peter illg->ig_next_ill = list_head(&illg->ig_actif); 1532 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 1533 8485 Peter 1534 8485 Peter /* 1535 8485 Peter * If the ill that's being deactivated had been nominated for 1536 8485 Peter * multicast/broadcast, nominate a new one. 1537 8485 Peter */ 1538 8485 Peter if (ill == illg->ig_cast_ill) 1539 8485 Peter ipmp_illgrp_set_cast(illg, list_head(&illg->ig_actif)); 1540 8485 Peter 1541 8485 Peter /* 1542 11042 Erik * Delete all nce_t entries using this ill, so that the next attempt 1543 11042 Erik * to send data traffic will revalidate cached nce's. 1544 11042 Erik */ 1545 11042 Erik nce_flush(ill, B_TRUE); 1546 11042 Erik 1547 11042 Erik /* 1548 8485 Peter * Unbind all of the ipifs bound to this ill, and save 'em in a list; 1549 8485 Peter * we'll rebind them after we tell the resolver the ill is no longer 1550 8485 Peter * active. We must do things in this order or the resolver could 1551 8485 Peter * accidentally rebind to the ill we're trying to remove if multiple 1552 8485 Peter * ills in the group have the same hardware address (which is 1553 8485 Peter * unsupported, but shouldn't lead to a wedged machine). 1554 8485 Peter */ 1555 8485 Peter while ((ipif = ipmp_ill_unbind_ipif(ill, NULL, B_TRUE)) != NULL) { 1556 8485 Peter ipif->ipif_bound_next = ubheadipif; 1557 8485 Peter ubheadipif = ipif; 1558 8485 Peter } 1559 8485 Peter if (!ill->ill_isv6) { 1560 8485 Peter 1561 8485 Peter /* 1562 11042 Erik * Refresh static/proxy ARP entries that had been using `ill'. 1563 8485 Peter */ 1564 8485 Peter ipmp_illgrp_refresh_arpent(illg); 1565 8485 Peter } 1566 8485 Peter 1567 8485 Peter /* 1568 8485 Peter * Rebind each ipif from the deactivated ill to the active ill with 1569 8485 Peter * the fewest ipifs. If there are no active ills, the ipifs will 1570 8485 Peter * remain unbound. 1571 8485 Peter */ 1572 8485 Peter for (ipif = ubheadipif; ipif != NULL; ipif = ubnextipif) { 1573 8485 Peter ubnextipif = ipif->ipif_bound_next; 1574 8485 Peter ipif->ipif_bound_next = NULL; 1575 8485 Peter 1576 8485 Peter if ((minill = ipmp_illgrp_min_ill(illg)) != NULL) 1577 8485 Peter ipmp_ill_bind_ipif(minill, ipif, Res_act_rebind); 1578 8485 Peter } 1579 11042 Erik 1580 11042 Erik if (list_is_empty(&illg->ig_actif)) { 1581 11042 Erik ill_t *ipmp_ill = illg->ig_ipmp_ill; 1582 11042 Erik 1583 11042 Erik ncec_walk(ipmp_ill, (pfi_t)ncec_delete_per_ill, 1584 11042 Erik (uchar_t *)ipmp_ill, ipmp_ill->ill_ipst); 1585 11042 Erik } 1586 11042 Erik 1587 11042 Erik /* 1588 11042 Erik * Remove any IRE_IF_CLONE for this ill since they might have 1589 11042 Erik * an ire_nce_cache/nce_common which refers to another ill in the group. 1590 11042 Erik */ 1591 11042 Erik ire_walk_ill(MATCH_IRE_TYPE, IRE_IF_CLONE, ill_downi_if_clone, 1592 11042 Erik ill, ill); 1593 8485 Peter 1594 8485 Peter /* 1595 8485 Peter * Finally, mark the group link down, if necessary. 1596 8485 Peter */ 1597 8485 Peter if (--grp->gr_nactif == 0) { 1598 8485 Peter mp = grp->gr_linkdownmp; 1599 8485 Peter grp->gr_linkdownmp = NULL; 1600 8485 Peter ASSERT(mp != NULL); 1601 8485 Peter put(illg->ig_ipmp_ill->ill_rq, mp); 1602 8485 Peter } 1603 8485 Peter } 1604 8485 Peter 1605 8485 Peter /* 1606 8485 Peter * Send the routing socket messages needed to make `ill' "appear" (RTM_ADD) 1607 8485 Peter * or "disappear" (RTM_DELETE) to non-IPMP-aware routing socket listeners. 1608 8485 Peter */ 1609 8485 Peter static void 1610 8485 Peter ipmp_ill_rtsaddrmsg(ill_t *ill, int cmd) 1611 8485 Peter { 1612 8485 Peter ipif_t *ipif; 1613 8485 Peter 1614 8485 Peter ASSERT(IAM_WRITER_ILL(ill)); 1615 8485 Peter ASSERT(cmd == RTM_ADD || cmd == RTM_DELETE); 1616 8485 Peter 1617 8485 Peter /* 1618 8485 Peter * If `ill' is truly down, there are no messages to generate since: 1619 8485 Peter * 1620 8485 Peter * 1. If cmd == RTM_DELETE, then we're supposed to hide the interface 1621 8485 Peter * and its addresses by bringing them down. But that's already 1622 8485 Peter * true, so there's nothing to hide. 1623 8485 Peter * 1624 8485 Peter * 2. If cmd == RTM_ADD, then we're supposed to generate messages 1625 8485 Peter * indicating that any previously-hidden up addresses are again 1626 8485 Peter * back up (along with the interface). But they aren't, so 1627 8485 Peter * there's nothing to expose. 1628 8485 Peter */ 1629 8485 Peter if (ill->ill_ipif_up_count == 0) 1630 8485 Peter return; 1631 8485 Peter 1632 8485 Peter if (cmd == RTM_ADD) 1633 8485 Peter ip_rts_xifmsg(ill->ill_ipif, IPIF_UP, 0, RTSQ_NORMAL); 1634 8485 Peter 1635 8485 Peter for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) 1636 8485 Peter if (ipif->ipif_flags & IPIF_UP) 1637 8485 Peter ip_rts_newaddrmsg(cmd, 0, ipif, RTSQ_NORMAL); 1638 8485 Peter 1639 8485 Peter if (cmd == RTM_DELETE) 1640 8485 Peter ip_rts_xifmsg(ill->ill_ipif, 0, IPIF_UP, RTSQ_NORMAL); 1641 8485 Peter } 1642 8485 Peter 1643 8485 Peter /* 1644 8485 Peter * Bind the address named by `ipif' to the underlying ill named by `ill'. 1645 8485 Peter * If `act' is Res_act_none, don't notify the resolver. Otherwise, `act' 1646 8485 Peter * will indicate to the resolver whether this is an initial bringup of 1647 8485 Peter * `ipif', or just a rebind to another ill. 1648 8485 Peter */ 1649 8485 Peter static void 1650 8485 Peter ipmp_ill_bind_ipif(ill_t *ill, ipif_t *ipif, enum ip_resolver_action act) 1651 8485 Peter { 1652 8485 Peter int err = 0; 1653 8485 Peter ip_stack_t *ipst = ill->ill_ipst; 1654 8485 Peter 1655 8485 Peter ASSERT(IAM_WRITER_ILL(ill) && IAM_WRITER_IPIF(ipif)); 1656 8485 Peter ASSERT(IS_UNDER_IPMP(ill) && IS_IPMP(ipif->ipif_ill)); 1657 8485 Peter ASSERT(act == Res_act_none || ipmp_ipif_is_up_dataaddr(ipif)); 1658 8485 Peter ASSERT(ipif->ipif_bound_ill == NULL); 1659 8485 Peter ASSERT(ipif->ipif_bound_next == NULL); 1660 8485 Peter 1661 8485 Peter ipif->ipif_bound_next = ill->ill_bound_ipif; 1662 8485 Peter ill->ill_bound_ipif = ipif; 1663 8485 Peter ill->ill_bound_cnt++; 1664 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_WRITER); 1665 8485 Peter ipif->ipif_bound_ill = ill; 1666 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 1667 8485 Peter 1668 8485 Peter /* 1669 8485 Peter * If necessary, tell ARP/NDP about the new mapping. Note that 1670 11042 Erik * ipif_resolver_up() cannot fail for IPv6 ills. 1671 8485 Peter */ 1672 8485 Peter if (act != Res_act_none) { 1673 8485 Peter if (ill->ill_isv6) { 1674 8485 Peter VERIFY(ipif_resolver_up(ipif, act) == 0); 1675 8485 Peter err = ipif_ndp_up(ipif, act == Res_act_initial); 1676 8485 Peter } else { 1677 8485 Peter err = ipif_resolver_up(ipif, act); 1678 8485 Peter } 1679 8485 Peter 1680 8485 Peter /* 1681 8485 Peter * Since ipif_ndp_up() never returns EINPROGRESS and 1682 8485 Peter * ipif_resolver_up() only returns EINPROGRESS when the 1683 8485 Peter * associated ill is not up, we should never be here with 1684 8485 Peter * EINPROGRESS. We rely on this to simplify the design. 1685 8485 Peter */ 1686 8485 Peter ASSERT(err != EINPROGRESS); 1687 8485 Peter } 1688 8485 Peter /* TODO: retry binding on failure? when? */ 1689 8485 Peter ipif->ipif_bound = (err == 0); 1690 8485 Peter } 1691 8485 Peter 1692 8485 Peter /* 1693 8485 Peter * Unbind the address named by `ipif' from the underlying ill named by `ill'. 1694 8485 Peter * If `ipif' is NULL, then an arbitrary ipif on `ill' is unbound and returned. 1695 8485 Peter * If no ipifs are bound to `ill', NULL is returned. If `notifyres' is 1696 8485 Peter * B_TRUE, notify the resolver about the change. 1697 8485 Peter */ 1698 8485 Peter static ipif_t * 1699 8485 Peter ipmp_ill_unbind_ipif(ill_t *ill, ipif_t *ipif, boolean_t notifyres) 1700 8485 Peter { 1701 8485 Peter ipif_t *previpif; 1702 8485 Peter ip_stack_t *ipst = ill->ill_ipst; 1703 8485 Peter 1704 8485 Peter ASSERT(IAM_WRITER_ILL(ill)); 1705 8485 Peter ASSERT(IS_UNDER_IPMP(ill)); 1706 8485 Peter 1707 8485 Peter /* 1708 8485 Peter * If necessary, find an ipif to unbind. 1709 8485 Peter */ 1710 8485 Peter if (ipif == NULL) { 1711 8485 Peter if ((ipif = ill->ill_bound_ipif) == NULL) { 1712 8485 Peter ASSERT(ill->ill_bound_cnt == 0); 1713 8485 Peter return (NULL); 1714 8485 Peter } 1715 8485 Peter } 1716 8485 Peter 1717 8485 Peter ASSERT(IAM_WRITER_IPIF(ipif)); 1718 8485 Peter ASSERT(IS_IPMP(ipif->ipif_ill)); 1719 8485 Peter ASSERT(ipif->ipif_bound_ill == ill); 1720 8485 Peter ASSERT(ill->ill_bound_cnt > 0); 1721 8485 Peter 1722 8485 Peter /* 1723 8485 Peter * Unbind it. 1724 8485 Peter */ 1725 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_WRITER); 1726 8485 Peter ipif->ipif_bound_ill = NULL; 1727 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 1728 8485 Peter ill->ill_bound_cnt--; 1729 8485 Peter 1730 8485 Peter if (ill->ill_bound_ipif == ipif) { 1731 8485 Peter ill->ill_bound_ipif = ipif->ipif_bound_next; 1732 8485 Peter } else { 1733 8485 Peter previpif = ill->ill_bound_ipif; 1734 8485 Peter while (previpif->ipif_bound_next != ipif) 1735 8485 Peter previpif = previpif->ipif_bound_next; 1736 8485 Peter 1737 8485 Peter previpif->ipif_bound_next = ipif->ipif_bound_next; 1738 8485 Peter } 1739 8485 Peter ipif->ipif_bound_next = NULL; 1740 8485 Peter 1741 8485 Peter /* 1742 8485 Peter * If requested, notify the resolvers (provided we're bound). 1743 8485 Peter */ 1744 8485 Peter if (notifyres && ipif->ipif_bound) { 1745 11042 Erik if (ill->ill_isv6) 1746 8485 Peter ipif_ndp_down(ipif); 1747 11042 Erik else 1748 11042 Erik (void) ipif_arp_down(ipif); 1749 8485 Peter } 1750 8485 Peter ipif->ipif_bound = B_FALSE; 1751 8485 Peter 1752 8485 Peter return (ipif); 1753 8485 Peter } 1754 8485 Peter 1755 8485 Peter /* 1756 8485 Peter * Check if `ill' is active. Caller must hold ill_lock and phyint_lock if 1757 8485 Peter * it's not inside the IPSQ. Since ipmp_ill_try_refresh_active() calls this 1758 8485 Peter * to determine whether an ill should be considered active, other consumers 1759 8485 Peter * may race and learn about an ill that should be deactivated/activated before 1760 8485 Peter * IPMP has performed the activation/deactivation. This should be safe though 1761 8485 Peter * since at worst e.g. ire_atomic_start() will prematurely delete an IRE that 1762 8485 Peter * would've been cleaned up by ipmp_ill_deactivate(). 1763 8485 Peter */ 1764 8485 Peter boolean_t 1765 8485 Peter ipmp_ill_is_active(ill_t *ill) 1766 8485 Peter { 1767 8485 Peter phyint_t *phyi = ill->ill_phyint; 1768 8485 Peter 1769 8485 Peter ASSERT(IS_UNDER_IPMP(ill)); 1770 8485 Peter ASSERT(IAM_WRITER_ILL(ill) || 1771 8485 Peter (MUTEX_HELD(&ill->ill_lock) && MUTEX_HELD(&phyi->phyint_lock))); 1772 8485 Peter 1773 8485 Peter /* 1774 8485 Peter * Note that PHYI_RUNNING isn't checked since we rely on in.mpathd to 1775 8485 Peter * set PHYI_FAILED whenever PHYI_RUNNING is cleared. This allows the 1776 8485 Peter * link flapping logic to be just in in.mpathd and allows us to ignore 1777 8485 Peter * changes to PHYI_RUNNING. 1778 8485 Peter */ 1779 8485 Peter return (!(ill->ill_ipif_up_count == 0 || 1780 8485 Peter (phyi->phyint_flags & (PHYI_OFFLINE|PHYI_INACTIVE|PHYI_FAILED)))); 1781 8485 Peter } 1782 8485 Peter 1783 8485 Peter /* 1784 11042 Erik * IRE walker callback: set ire_testhidden on IRE_HIDDEN_TYPE IREs associated 1785 11042 Erik * with `ill_arg'. 1786 8485 Peter */ 1787 8485 Peter static void 1788 8485 Peter ipmp_ill_ire_mark_testhidden(ire_t *ire, char *ill_arg) 1789 8485 Peter { 1790 8485 Peter ill_t *ill = (ill_t *)ill_arg; 1791 8485 Peter 1792 8485 Peter ASSERT(IAM_WRITER_ILL(ill)); 1793 8485 Peter ASSERT(!IS_IPMP(ill)); 1794 8485 Peter 1795 11042 Erik if (ire->ire_ill != ill) 1796 8485 Peter return; 1797 8485 Peter 1798 11042 Erik if (IRE_HIDDEN_TYPE(ire->ire_type)) { 1799 8485 Peter DTRACE_PROBE1(ipmp__mark__testhidden, ire_t *, ire); 1800 11042 Erik ire->ire_testhidden = B_TRUE; 1801 8485 Peter } 1802 8485 Peter } 1803 8485 Peter 1804 8485 Peter /* 1805 11042 Erik * IRE walker callback: clear ire_testhidden if the IRE has a source address 1806 11042 Erik * on `ill_arg'. 1807 8485 Peter */ 1808 8485 Peter static void 1809 8485 Peter ipmp_ill_ire_clear_testhidden(ire_t *ire, char *ill_arg) 1810 8485 Peter { 1811 8485 Peter ill_t *ill = (ill_t *)ill_arg; 1812 8485 Peter 1813 8485 Peter ASSERT(IAM_WRITER_ILL(ill)); 1814 8485 Peter ASSERT(!IS_IPMP(ill)); 1815 8485 Peter 1816 11042 Erik if (ire->ire_ill == ill) { 1817 8485 Peter DTRACE_PROBE1(ipmp__clear__testhidden, ire_t *, ire); 1818 11042 Erik ire->ire_testhidden = B_FALSE; 1819 8485 Peter } 1820 8485 Peter } 1821 8485 Peter 1822 8485 Peter /* 1823 8485 Peter * Return a held pointer to the IPMP ill for underlying interface `ill', or 1824 8485 Peter * NULL if one doesn't exist. (Unfortunately, this function needs to take an 1825 8485 Peter * underlying ill rather than an ipmp_illgrp_t because an underlying ill's 1826 8564 Peter * ill_grp pointer may become stale when not inside an IPSQ and not holding 1827 8485 Peter * ipmp_lock.) Caller need not be inside the IPSQ. 1828 8485 Peter */ 1829 8485 Peter ill_t * 1830 8485 Peter ipmp_ill_hold_ipmp_ill(ill_t *ill) 1831 8485 Peter { 1832 8485 Peter ip_stack_t *ipst = ill->ill_ipst; 1833 8485 Peter ipmp_illgrp_t *illg; 1834 8485 Peter 1835 8485 Peter ASSERT(!IS_IPMP(ill)); 1836 8485 Peter 1837 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_READER); 1838 8485 Peter illg = ill->ill_grp; 1839 11042 Erik if (illg != NULL && ill_check_and_refhold(illg->ig_ipmp_ill)) { 1840 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 1841 8485 Peter return (illg->ig_ipmp_ill); 1842 8485 Peter } 1843 8485 Peter /* 1844 8485 Peter * Assume `ill' was removed from the illgrp in the meantime. 1845 8485 Peter */ 1846 8485 Peter rw_exit(&ill->ill_ipst->ips_ipmp_lock); 1847 8485 Peter return (NULL); 1848 8485 Peter } 1849 8485 Peter 1850 8485 Peter /* 1851 8485 Peter * Return the interface index for the IPMP ill tied to underlying interface 1852 8485 Peter * `ill', or zero if one doesn't exist. Caller need not be inside the IPSQ. 1853 8485 Peter */ 1854 8485 Peter uint_t 1855 8485 Peter ipmp_ill_get_ipmp_ifindex(const ill_t *ill) 1856 8485 Peter { 1857 8485 Peter uint_t ifindex = 0; 1858 8485 Peter ip_stack_t *ipst = ill->ill_ipst; 1859 8485 Peter ipmp_grp_t *grp; 1860 8485 Peter 1861 8485 Peter ASSERT(!IS_IPMP(ill)); 1862 8485 Peter 1863 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_READER); 1864 8485 Peter if ((grp = ill->ill_phyint->phyint_grp) != NULL) 1865 8485 Peter ifindex = grp->gr_phyint->phyint_ifindex; 1866 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 1867 8485 Peter return (ifindex); 1868 8485 Peter } 1869 8485 Peter 1870 8485 Peter /* 1871 8485 Peter * Place phyint `phyi' into IPMP group `grp'. 1872 8485 Peter */ 1873 8485 Peter void 1874 8485 Peter ipmp_phyint_join_grp(phyint_t *phyi, ipmp_grp_t *grp) 1875 8485 Peter { 1876 8485 Peter ill_t *ill; 1877 8485 Peter ipsq_t *ipsq = phyi->phyint_ipsq; 1878 8485 Peter ipsq_t *grp_ipsq = grp->gr_phyint->phyint_ipsq; 1879 8485 Peter ip_stack_t *ipst = PHYINT_TO_IPST(phyi); 1880 8485 Peter 1881 8485 Peter ASSERT(IAM_WRITER_IPSQ(ipsq)); 1882 8485 Peter ASSERT(phyi->phyint_illv4 != NULL || phyi->phyint_illv6 != NULL); 1883 8485 Peter 1884 8485 Peter /* 1885 8485 Peter * Send routing socket messages indicating that the phyint's ills 1886 8485 Peter * and ipifs vanished. 1887 8485 Peter */ 1888 8485 Peter if (phyi->phyint_illv4 != NULL) { 1889 8485 Peter ill = phyi->phyint_illv4; 1890 8485 Peter ipmp_ill_rtsaddrmsg(ill, RTM_DELETE); 1891 8485 Peter } 1892 8485 Peter 1893 8485 Peter if (phyi->phyint_illv6 != NULL) { 1894 8485 Peter ill = phyi->phyint_illv6; 1895 8485 Peter ipmp_ill_rtsaddrmsg(ill, RTM_DELETE); 1896 8485 Peter } 1897 8485 Peter 1898 8485 Peter /* 1899 8485 Peter * Snapshot the phyint's initial kstats as a baseline. 1900 8485 Peter */ 1901 8485 Peter ipmp_phyint_get_kstats(phyi, phyi->phyint_kstats0); 1902 8485 Peter 1903 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_WRITER); 1904 8485 Peter 1905 8485 Peter phyi->phyint_grp = grp; 1906 8485 Peter if (++grp->gr_nif == 1) 1907 8485 Peter grp->gr_mactype = ill->ill_mactype; 1908 8485 Peter else 1909 8485 Peter ASSERT(grp->gr_mactype == ill->ill_mactype); 1910 8485 Peter 1911 8485 Peter /* 1912 8485 Peter * Now that we're in the group, request a switch to the group's xop 1913 8485 Peter * when we ipsq_exit(). All future operations will be exclusive on 1914 8485 Peter * the group xop until ipmp_phyint_leave_grp() is called. 1915 8485 Peter */ 1916 8485 Peter ASSERT(ipsq->ipsq_swxop == NULL); 1917 8485 Peter ASSERT(grp_ipsq->ipsq_xop == &grp_ipsq->ipsq_ownxop); 1918 8485 Peter ipsq->ipsq_swxop = &grp_ipsq->ipsq_ownxop; 1919 8485 Peter 1920 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 1921 8485 Peter } 1922 8485 Peter 1923 8485 Peter /* 1924 8485 Peter * Remove phyint `phyi' from its current IPMP group. 1925 8485 Peter */ 1926 8485 Peter void 1927 8485 Peter ipmp_phyint_leave_grp(phyint_t *phyi) 1928 8485 Peter { 1929 8485 Peter uint_t i; 1930 8485 Peter ipsq_t *ipsq = phyi->phyint_ipsq; 1931 8485 Peter ip_stack_t *ipst = PHYINT_TO_IPST(phyi); 1932 8485 Peter uint64_t phyi_kstats[IPMP_KSTAT_MAX]; 1933 8485 Peter 1934 8485 Peter ASSERT(IAM_WRITER_IPSQ(ipsq)); 1935 8485 Peter 1936 8485 Peter /* 1937 8485 Peter * If any of the phyint's ills are still in an illgrp, kick 'em out. 1938 8485 Peter */ 1939 8485 Peter if (phyi->phyint_illv4 != NULL && IS_UNDER_IPMP(phyi->phyint_illv4)) 1940 8485 Peter ipmp_ill_leave_illgrp(phyi->phyint_illv4); 1941 8485 Peter if (phyi->phyint_illv6 != NULL && IS_UNDER_IPMP(phyi->phyint_illv6)) 1942 8485 Peter ipmp_ill_leave_illgrp(phyi->phyint_illv6); 1943 8485 Peter 1944 8485 Peter /* 1945 8485 Peter * Send routing socket messages indicating that the phyint's ills 1946 8485 Peter * and ipifs have reappeared. 1947 8485 Peter */ 1948 8485 Peter if (phyi->phyint_illv4 != NULL) 1949 8485 Peter ipmp_ill_rtsaddrmsg(phyi->phyint_illv4, RTM_ADD); 1950 8485 Peter if (phyi->phyint_illv6 != NULL) 1951 8485 Peter ipmp_ill_rtsaddrmsg(phyi->phyint_illv6, RTM_ADD); 1952 8485 Peter 1953 8485 Peter /* 1954 8485 Peter * Calculate the phyint's cumulative kstats while it was in the group, 1955 8485 Peter * and add that to the group's baseline. 1956 8485 Peter */ 1957 8485 Peter ipmp_phyint_get_kstats(phyi, phyi_kstats); 1958 8485 Peter for (i = 0; i < IPMP_KSTAT_MAX; i++) { 1959 8485 Peter phyi_kstats[i] -= phyi->phyint_kstats0[i]; 1960 8485 Peter atomic_add_64(&phyi->phyint_grp->gr_kstats0[i], phyi_kstats[i]); 1961 8485 Peter } 1962 8485 Peter 1963 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_WRITER); 1964 8485 Peter 1965 8485 Peter phyi->phyint_grp->gr_nif--; 1966 8485 Peter phyi->phyint_grp = NULL; 1967 8485 Peter 1968 8485 Peter /* 1969 8485 Peter * As our final act in leaving the group, request a switch back to our 1970 8485 Peter * IPSQ's own xop when we ipsq_exit(). 1971 8485 Peter */ 1972 8485 Peter ASSERT(ipsq->ipsq_swxop == NULL); 1973 8485 Peter ipsq->ipsq_swxop = &ipsq->ipsq_ownxop; 1974 8485 Peter 1975 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 1976 8485 Peter } 1977 8485 Peter 1978 8485 Peter /* 1979 8485 Peter * Store the IPMP-related kstats for `phyi' into the array named by `kstats'. 1980 8485 Peter * Assumes that `kstats' has at least IPMP_KSTAT_MAX elements. 1981 8485 Peter */ 1982 8485 Peter static void 1983 8485 Peter ipmp_phyint_get_kstats(phyint_t *phyi, uint64_t kstats[]) 1984 8485 Peter { 1985 8485 Peter uint_t i, j; 1986 8485 Peter const char *name; 1987 8485 Peter kstat_t *ksp; 1988 8485 Peter kstat_named_t *kn; 1989 10616 Sebastien ip_stack_t *ipst = PHYINT_TO_IPST(phyi); 1990 10616 Sebastien zoneid_t zoneid; 1991 8485 Peter 1992 8485 Peter bzero(kstats, sizeof (kstats[0]) * IPMP_KSTAT_MAX); 1993 10616 Sebastien zoneid = netstackid_to_zoneid(ipst->ips_netstack->netstack_stackid); 1994 10616 Sebastien ksp = kstat_hold_byname("link", 0, phyi->phyint_name, zoneid); 1995 8485 Peter if (ksp == NULL) 1996 8485 Peter return; 1997 8485 Peter 1998 8485 Peter KSTAT_ENTER(ksp); 1999 8485 Peter 2000 8485 Peter if (ksp->ks_data != NULL && ksp->ks_type == KSTAT_TYPE_NAMED) { 2001 8485 Peter /* 2002 8485 Peter * Bring kstats up-to-date before recording. 2003 8485 Peter */ 2004 8485 Peter (void) KSTAT_UPDATE(ksp, KSTAT_READ); 2005 8485 Peter 2006 8485 Peter kn = KSTAT_NAMED_PTR(ksp); 2007 8485 Peter for (i = 0; i < IPMP_KSTAT_MAX; i++) { 2008 8485 Peter name = ipmp_kstats[i].name; 2009 8485 Peter kstats[i] = 0; 2010 8485 Peter for (j = 0; j < ksp->ks_ndata; j++) { 2011 8485 Peter if (strcmp(kn[j].name, name) != 0) 2012 8485 Peter continue; 2013 8485 Peter 2014 8485 Peter switch (kn[j].data_type) { 2015 8485 Peter case KSTAT_DATA_INT32: 2016 8485 Peter case KSTAT_DATA_UINT32: 2017 8485 Peter kstats[i] = kn[j].value.ui32; 2018 8485 Peter break; 2019 8485 Peter #ifdef _LP64 2020 8485 Peter case KSTAT_DATA_LONG: 2021 8485 Peter case KSTAT_DATA_ULONG: 2022 8485 Peter kstats[i] = kn[j].value.ul; 2023 8485 Peter break; 2024 8485 Peter #endif 2025 8485 Peter case KSTAT_DATA_INT64: 2026 8485 Peter case KSTAT_DATA_UINT64: 2027 8485 Peter kstats[i] = kn[j].value.ui64; 2028 8485 Peter break; 2029 8485 Peter } 2030 8485 Peter break; 2031 8485 Peter } 2032 8485 Peter } 2033 8485 Peter } 2034 8485 Peter 2035 8485 Peter KSTAT_EXIT(ksp); 2036 8485 Peter kstat_rele(ksp); 2037 8485 Peter } 2038 8485 Peter 2039 8485 Peter /* 2040 8485 Peter * Refresh the active state of all ills on `phyi'. 2041 8485 Peter */ 2042 8485 Peter void 2043 8485 Peter ipmp_phyint_refresh_active(phyint_t *phyi) 2044 8485 Peter { 2045 8485 Peter if (phyi->phyint_illv4 != NULL) 2046 8485 Peter ipmp_ill_refresh_active(phyi->phyint_illv4); 2047 8485 Peter if (phyi->phyint_illv6 != NULL) 2048 8485 Peter ipmp_ill_refresh_active(phyi->phyint_illv6); 2049 8485 Peter } 2050 8485 Peter 2051 8485 Peter /* 2052 8485 Peter * Return a held pointer to the underlying ill bound to `ipif', or NULL if one 2053 8485 Peter * doesn't exist. Caller need not be inside the IPSQ. 2054 8485 Peter */ 2055 8485 Peter ill_t * 2056 8485 Peter ipmp_ipif_hold_bound_ill(const ipif_t *ipif) 2057 8485 Peter { 2058 8485 Peter ill_t *boundill; 2059 8485 Peter ip_stack_t *ipst = ipif->ipif_ill->ill_ipst; 2060 8485 Peter 2061 8485 Peter ASSERT(IS_IPMP(ipif->ipif_ill)); 2062 8485 Peter 2063 8485 Peter rw_enter(&ipst->ips_ipmp_lock, RW_READER); 2064 8485 Peter boundill = ipif->ipif_bound_ill; 2065 11042 Erik if (boundill != NULL && ill_check_and_refhold(boundill)) { 2066 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 2067 8485 Peter return (boundill); 2068 8485 Peter } 2069 8485 Peter rw_exit(&ipst->ips_ipmp_lock); 2070 8485 Peter return (NULL); 2071 8485 Peter } 2072 8485 Peter 2073 8485 Peter /* 2074 8485 Peter * Return a pointer to the underlying ill bound to `ipif', or NULL if one 2075 8485 Peter * doesn't exist. Caller must be inside the IPSQ. 2076 8485 Peter */ 2077 8485 Peter ill_t * 2078 8485 Peter ipmp_ipif_bound_ill(const ipif_t *ipif) 2079 8485 Peter { 2080 8485 Peter ASSERT(IAM_WRITER_ILL(ipif->ipif_ill)); 2081 8485 Peter ASSERT(IS_IPMP(ipif->ipif_ill)); 2082 8485 Peter 2083 8485 Peter return (ipif->ipif_bound_ill); 2084 8485 Peter } 2085 8485 Peter 2086 8485 Peter /* 2087 8485 Peter * Check if `ipif' is a "stub" (placeholder address not being used). 2088 8485 Peter */ 2089 8485 Peter boolean_t 2090 8485 Peter ipmp_ipif_is_stubaddr(const ipif_t *ipif) 2091 8485 Peter { 2092 8485 Peter if (ipif->ipif_flags & IPIF_UP) 2093 8485 Peter return (B_FALSE); 2094 8485 Peter if (ipif->ipif_ill->ill_isv6) 2095 8485 Peter return (IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6lcl_addr)); 2096 8485 Peter else 2097 8485 Peter return (ipif->ipif_lcl_addr == INADDR_ANY); 2098 8485 Peter } 2099 8485 Peter 2100 8485 Peter /* 2101 8485 Peter * Check if `ipif' is an IPMP data address. 2102 8485 Peter */ 2103 8485 Peter boolean_t 2104 8485 Peter ipmp_ipif_is_dataaddr(const ipif_t *ipif) 2105 8485 Peter { 2106 8485 Peter if (ipif->ipif_flags & IPIF_NOFAILOVER) 2107 8485 Peter return (B_FALSE); 2108 8485 Peter if (ipif->ipif_ill->ill_isv6) 2109 8485 Peter return (!IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6lcl_addr)); 2110 8485 Peter else 2111 8485 Peter return (ipif->ipif_lcl_addr != INADDR_ANY); 2112 8485 Peter } 2113 8485 Peter 2114 8485 Peter /* 2115 8485 Peter * Check if `ipif' is an IPIF_UP IPMP data address. 2116 8485 Peter */ 2117 8485 Peter static boolean_t 2118 8485 Peter ipmp_ipif_is_up_dataaddr(const ipif_t *ipif) 2119 8485 Peter { 2120 8485 Peter return (ipmp_ipif_is_dataaddr(ipif) && (ipif->ipif_flags & IPIF_UP)); 2121 8485 Peter } 2122 11042 Erik 2123 11042 Erik /* 2124 11042 Erik * Check if `mp' contains a probe packet by verifying if the IP source address 2125 11042 Erik * is a test address on an underlying interface `ill'. Caller need not be inside 2126 11042 Erik * the IPSQ. 2127 11042 Erik */ 2128 11042 Erik boolean_t 2129 11042 Erik ipmp_packet_is_probe(mblk_t *mp, ill_t *ill) 2130 11042 Erik { 2131 11042 Erik ip6_t *ip6h = (ip6_t *)mp->b_rptr; 2132 11042 Erik ipha_t *ipha = (ipha_t *)mp->b_rptr; 2133 11042 Erik 2134 11042 Erik ASSERT(DB_TYPE(mp) != M_CTL); 2135 11042 Erik 2136 11042 Erik if (!IS_UNDER_IPMP(ill)) 2137 11042 Erik return (B_FALSE); 2138 11042 Erik 2139 11042 Erik if (ill->ill_isv6) { 2140 11042 Erik if (!IN6_IS_ADDR_UNSPECIFIED(&ip6h->ip6_src) && 2141 11042 Erik ipif_lookup_testaddr_v6(ill, &ip6h->ip6_src, NULL)) 2142 11042 Erik return (B_TRUE); 2143 11042 Erik } else { 2144 11042 Erik if ((ipha->ipha_src != INADDR_ANY) && 2145 11042 Erik ipif_lookup_testaddr_v4(ill, &ipha->ipha_src, NULL)) 2146 11042 Erik return (B_TRUE); 2147 11042 Erik } 2148 11042 Erik return (B_FALSE); 2149 11042 Erik } 2150 11042 Erik 2151 11042 Erik /* 2152 11042 Erik * Pick out an appropriate underlying interface for packet transmit. This 2153 11042 Erik * function may be called from the data path, so we need to verify that the 2154 11042 Erik * IPMP group associated with `ill' is non-null after holding the ill_g_lock. 2155 11042 Erik * Caller need not be inside the IPSQ. 2156 11042 Erik */ 2157 11042 Erik ill_t * 2158 11042 Erik ipmp_ill_get_xmit_ill(ill_t *ill, boolean_t is_unicast) 2159 11042 Erik { 2160 11042 Erik ill_t *xmit_ill; 2161 11042 Erik ip_stack_t *ipst = ill->ill_ipst; 2162 11042 Erik 2163 11042 Erik rw_enter(&ipst->ips_ill_g_lock, RW_READER); 2164 11042 Erik if (ill->ill_grp == NULL) { 2165 11042 Erik /* 2166 11042 Erik * The interface was taken out of the group. Return ill itself, 2167 11042 Erik * but take a ref so that callers will always be able to do 2168 11042 Erik * ill_refrele(ill); 2169 11042 Erik */ 2170 11042 Erik rw_exit(&ipst->ips_ill_g_lock); 2171 11042 Erik ill_refhold(ill); 2172 11042 Erik return (ill); 2173 11042 Erik } 2174 11042 Erik if (!is_unicast) 2175 11042 Erik xmit_ill = ipmp_illgrp_hold_cast_ill(ill->ill_grp); 2176 11042 Erik else 2177 11042 Erik xmit_ill = ipmp_illgrp_hold_next_ill(ill->ill_grp); 2178 11042 Erik rw_exit(&ipst->ips_ill_g_lock); 2179 11042 Erik return (xmit_ill); 2180 11042 Erik } 2181 11042 Erik 2182 11042 Erik /* 2183 11042 Erik * Flush out any nce that points at `ncec' from an underlying interface 2184 11042 Erik */ 2185 11042 Erik void 2186 11042 Erik ipmp_ncec_flush_nce(ncec_t *ncec) 2187 11042 Erik { 2188 11042 Erik ill_t *ncec_ill = ncec->ncec_ill; 2189 11042 Erik ill_t *ill; 2190 11042 Erik ipmp_illgrp_t *illg; 2191 11042 Erik ip_stack_t *ipst = ncec_ill->ill_ipst; 2192 11042 Erik list_t dead; 2193 11042 Erik nce_t *nce; 2194 11042 Erik 2195 11042 Erik if (!IS_IPMP(ncec_ill)) 2196 11042 Erik return; 2197 11042 Erik 2198 11042 Erik illg = ncec_ill->ill_grp; 2199 11042 Erik list_create(&dead, sizeof (nce_t), offsetof(nce_t, nce_node)); 2200 11042 Erik 2201 11042 Erik rw_enter(&ipst->ips_ill_g_lock, RW_READER); 2202 11042 Erik ill = list_head(&illg->ig_if); 2203 11042 Erik for (; ill != NULL; ill = list_next(&illg->ig_if, ill)) { 2204 11042 Erik nce_fastpath_list_delete(ill, ncec, &dead); 2205 11042 Erik } 2206 11042 Erik rw_exit(&ipst->ips_ill_g_lock); 2207 11042 Erik 2208 11042 Erik /* 2209 11042 Erik * we may now nce_refrele() all dead entries since all locks have been 2210 11042 Erik * dropped. 2211 11042 Erik */ 2212 11042 Erik while ((nce = list_head(&dead)) != NULL) { 2213 11042 Erik list_remove(&dead, nce); 2214 11042 Erik nce_refrele(nce); 2215 11042 Erik } 2216 11042 Erik ASSERT(list_is_empty(&dead)); 2217 11042 Erik list_destroy(&dead); 2218 11042 Erik } 2219 11042 Erik 2220 11042 Erik /* 2221 11042 Erik * For each interface in the IPMP group, if there are nce_t entries for the IP 2222 11042 Erik * address corresponding to `ncec', then their dl_unitdata_req_t and fastpath 2223 11042 Erik * information must be updated to match the link-layer address information in 2224 11042 Erik * `ncec'. 2225 11042 Erik */ 2226 11042 Erik void 2227 11042 Erik ipmp_ncec_fastpath(ncec_t *ncec, ill_t *ipmp_ill) 2228 11042 Erik { 2229 11042 Erik ill_t *ill; 2230 11042 Erik ipmp_illgrp_t *illg = ipmp_ill->ill_grp; 2231 11042 Erik ip_stack_t *ipst = ipmp_ill->ill_ipst; 2232 11042 Erik nce_t *nce, *nce_next; 2233 11042 Erik list_t replace; 2234 11042 Erik 2235 11042 Erik ASSERT(IS_IPMP(ipmp_ill)); 2236 11042 Erik 2237 11042 Erik /* 2238 11042 Erik * if ncec itself is not reachable, there is no use in creating nce_t 2239 11042 Erik * entries on the underlying interfaces in the group. 2240 11042 Erik */ 2241 11042 Erik if (!NCE_ISREACHABLE(ncec)) 2242 11042 Erik return; 2243 11042 Erik 2244 11042 Erik list_create(&replace, sizeof (nce_t), offsetof(nce_t, nce_node)); 2245 11042 Erik rw_enter(&ipst->ips_ipmp_lock, RW_READER); 2246 11042 Erik ill = list_head(&illg->ig_actif); 2247 11042 Erik for (; ill != NULL; ill = list_next(&illg->ig_actif, ill)) { 2248 11042 Erik /* 2249 11042 Erik * For each underlying interface, we first check if there is an 2250 11042 Erik * nce_t for the address in ncec->ncec_addr. If one exists, 2251 11042 Erik * we should trigger nce_fastpath for that nce_t. However, the 2252 11042 Erik * catch is that we are holding the ips_ipmp_lock to prevent 2253 11042 Erik * changes to the IPMP group membership, so that we cannot 2254 11042 Erik * putnext() to the driver. So we nce_delete the 2255 11042 Erik * list nce_t entries that need to be updated into the 2256 11042 Erik * `replace' list, and then process the `replace' list 2257 11042 Erik * after dropping the ips_ipmp_lock. 2258 11042 Erik */ 2259 11042 Erik mutex_enter(&ill->ill_lock); 2260 11042 Erik for (nce = list_head(&ill->ill_nce); nce != NULL; ) { 2261 11042 Erik nce_next = list_next(&ill->ill_nce, nce); 2262 11042 Erik if (!IN6_ARE_ADDR_EQUAL(&nce->nce_addr, 2263 11042 Erik &ncec->ncec_addr)) { 2264 11042 Erik nce = nce_next; 2265 11042 Erik continue; 2266 11042 Erik } 2267 11042 Erik nce_refhold(nce); 2268 11042 Erik nce_delete(nce); 2269 11042 Erik list_insert_tail(&replace, nce); 2270 11042 Erik nce = nce_next; 2271 11042 Erik } 2272 11042 Erik mutex_exit(&ill->ill_lock); 2273 11042 Erik } 2274 11042 Erik rw_exit(&ipst->ips_ipmp_lock); 2275 11042 Erik /* 2276 11042 Erik * `replace' now has the list of nce's on which we should be triggering 2277 11042 Erik * nce_fastpath(). We now retrigger fastpath by setting up the nce 2278 11042 Erik * again. The code in nce_lookup_then_add_v* ensures that nce->nce_ill 2279 11042 Erik * is still in the group for ncec->ncec_ill 2280 11042 Erik */ 2281 11042 Erik while ((nce = list_head(&replace)) != NULL) { 2282 11042 Erik list_remove(&replace, nce); 2283 11042 Erik if (ncec->ncec_ill->ill_isv6) { 2284 11042 Erik (void) nce_lookup_then_add_v6(nce->nce_ill, 2285 11042 Erik ncec->ncec_lladdr, ncec->ncec_lladdr_length, 2286 11042 Erik &nce->nce_addr, ncec->ncec_flags, ND_UNCHANGED, 2287 11042 Erik NULL); 2288 11042 Erik } else { 2289 11042 Erik ipaddr_t ipaddr; 2290 11042 Erik 2291 11042 Erik IN6_V4MAPPED_TO_IPADDR(&ncec->ncec_addr, ipaddr); 2292 11042 Erik (void) nce_lookup_then_add_v4(nce->nce_ill, 2293 11042 Erik ncec->ncec_lladdr, ncec->ncec_lladdr_length, 2294 11042 Erik &ipaddr, ncec->ncec_flags, ND_UNCHANGED, NULL); 2295 11042 Erik } 2296 11042 Erik nce_refrele(nce); 2297 11042 Erik } 2298 11042 Erik ASSERT(list_is_empty(&replace)); 2299 11042 Erik list_destroy(&replace); 2300 11042 Erik } 2301
Indexes created Tue Nov 24 05:22:20 UTC 2009
Terms of Use |
Privacy |
Trademarks |
Copyright Policy |
Site Guidelines |
Help
Your use of this web site or any of its content or software indicates your agreement to be bound by these Terms of Use.
Copyright © 1995-2008 Sun Microsystems, Inc.