1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright 2019 Joyent, Inc. 25 * Copyright 2017 RackTop Systems. 26 * Copyright 2022 OmniOS Community Edition (OmniOSce) Association. 27 * Copyright 2025 Oxide Computer Company 28 */ 29 30 /* 31 * - General Introduction: 32 * 33 * This file contains the implementation of the MAC client kernel 34 * API and related code. The MAC client API allows a kernel module 35 * to gain access to a MAC instance (physical NIC, link aggregation, etc). 36 * It allows a MAC client to associate itself with a MAC address, 37 * VLANs, callback functions for data traffic and for promiscuous mode. 38 * The MAC client API is also used to specify the properties associated 39 * with a MAC client, such as bandwidth limits, priority, CPUS, etc. 40 * These properties are further used to determine the hardware resources 41 * to allocate to the various MAC clients. 42 * 43 * - Primary MAC clients: 44 * 45 * The MAC client API refers to "primary MAC clients". A primary MAC 46 * client is a client which "owns" the primary MAC address of 47 * the underlying MAC instance. The primary MAC address is called out 48 * since it is associated with specific semantics: the primary MAC 49 * address is the MAC address which is assigned to the IP interface 50 * when it is plumbed, and the primary MAC address is assigned 51 * to VLAN data-links. The primary address of a MAC instance can 52 * also change dynamically from under the MAC client, for example 53 * as a result of a change of state of a link aggregation. In that 54 * case the MAC layer automatically updates all data-structures which 55 * refer to the current value of the primary MAC address. Typical 56 * primary MAC clients are dls, aggr, and xnb. A typical non-primary 57 * MAC client is the vnic driver. 58 * 59 * - Virtual Switching: 60 * 61 * The MAC layer implements a virtual switch between the MAC clients 62 * (primary and non-primary) defined on top of the same underlying 63 * NIC (physical, link aggregation, etc). The virtual switch is 64 * VLAN-aware, i.e. it allows multiple MAC clients to be member 65 * of one or more VLANs, and the virtual switch will distribute 66 * multicast tagged packets only to the member of the corresponding 67 * VLANs. 68 * 69 * - Upper vs Lower MAC: 70 * 71 * Creating a VNIC on top of a MAC instance effectively causes 72 * two MAC instances to be layered on top of each other, one for 73 * the VNIC(s), one for the underlying MAC instance (physical NIC, 74 * link aggregation, etc). In the code below we refer to the 75 * underlying NIC as the "lower MAC", and we refer to VNICs as 76 * the "upper MAC". 77 * 78 * - Pass-through for VNICs: 79 * 80 * When VNICs are created on top of an underlying MAC, this causes 81 * a layering of two MAC instances. Since the lower MAC already 82 * does the switching and demultiplexing to its MAC clients, the 83 * upper MAC would simply have to pass packets to the layer below 84 * or above it, which would introduce overhead. In order to avoid 85 * this overhead, the MAC layer implements a pass-through mechanism 86 * for VNICs. When a VNIC opens the lower MAC instance, it saves 87 * the MAC client handle it optains from the MAC layer. When a MAC 88 * client opens a VNIC (upper MAC), the MAC layer detects that 89 * the MAC being opened is a VNIC, and gets the MAC client handle 90 * that the VNIC driver obtained from the lower MAC. This exchange 91 * is done through a private capability between the MAC layer 92 * and the VNIC driver. The upper MAC then returns that handle 93 * directly to its MAC client. Any operation done by the upper 94 * MAC client is now done on the lower MAC client handle, which 95 * allows the VNIC driver to be completely bypassed for the 96 * performance sensitive data-path. 97 * 98 * - Secondary MACs for VNICs: 99 * 100 * VNICs support multiple upper mac clients to enable support for 101 * multiple MAC addresses on the VNIC. When the VNIC is created the 102 * initial mac client is the primary upper mac. Any additional mac 103 * clients are secondary macs. These are kept in sync with the primary 104 * (for things such as the rx function and resource control settings) 105 * using the same private capability interface between the MAC layer 106 * and the VNIC layer. 107 * 108 */ 109 110 #include <sys/types.h> 111 #include <sys/conf.h> 112 #include <sys/id_space.h> 113 #include <sys/esunddi.h> 114 #include <sys/stat.h> 115 #include <sys/mkdev.h> 116 #include <sys/stream.h> 117 #include <sys/strsun.h> 118 #include <sys/strsubr.h> 119 #include <sys/pattr.h> 120 #include <sys/dlpi.h> 121 #include <sys/modhash.h> 122 #include <sys/mac_impl.h> 123 #include <sys/mac_client_impl.h> 124 #include <sys/mac_soft_ring.h> 125 #include <sys/mac_stat.h> 126 #include <sys/dls.h> 127 #include <sys/dld.h> 128 #include <sys/modctl.h> 129 #include <sys/fs/dv_node.h> 130 #include <sys/thread.h> 131 #include <sys/proc.h> 132 #include <sys/callb.h> 133 #include <sys/cpuvar.h> 134 #include <sys/atomic.h> 135 #include <sys/sdt.h> 136 #include <sys/mac_flow.h> 137 #include <sys/ddi_intr_impl.h> 138 #include <sys/disp.h> 139 #include <sys/sdt.h> 140 #include <sys/vnic.h> 141 #include <sys/vnic_impl.h> 142 #include <sys/vlan.h> 143 #include <inet/ip.h> 144 #include <inet/ip6.h> 145 #include <sys/exacct.h> 146 #include <sys/exacct_impl.h> 147 #include <inet/nd.h> 148 #include <sys/ethernet.h> 149 150 kmem_cache_t *mac_client_impl_cache; 151 kmem_cache_t *mac_promisc_impl_cache; 152 153 static boolean_t mac_client_single_rcvr(mac_client_impl_t *); 154 static flow_entry_t *mac_client_swap_mciflent(mac_client_impl_t *); 155 static flow_entry_t *mac_client_get_flow(mac_client_impl_t *, 156 mac_unicast_impl_t *); 157 static void mac_client_remove_flow_from_list(mac_client_impl_t *, 158 flow_entry_t *); 159 static void mac_client_add_to_flow_list(mac_client_impl_t *, flow_entry_t *); 160 static void mac_rename_flow_names(mac_client_impl_t *, const char *); 161 static void mac_virtual_link_update(mac_impl_t *); 162 static int mac_client_datapath_setup(mac_client_impl_t *, uint16_t, 163 uint8_t *, mac_resource_props_t *, boolean_t, mac_unicast_impl_t *); 164 static void mac_client_datapath_teardown(mac_client_handle_t, 165 mac_unicast_impl_t *, flow_entry_t *); 166 static int mac_resource_ctl_set(mac_client_handle_t, mac_resource_props_t *); 167 168 /* ARGSUSED */ 169 static int 170 i_mac_client_impl_ctor(void *buf, void *arg, int kmflag) 171 { 172 int i; 173 mac_client_impl_t *mcip = buf; 174 175 bzero(buf, MAC_CLIENT_IMPL_SIZE); 176 mutex_init(&mcip->mci_tx_cb_lock, NULL, MUTEX_DRIVER, NULL); 177 mcip->mci_tx_notify_cb_info.mcbi_lockp = &mcip->mci_tx_cb_lock; 178 179 ASSERT(mac_tx_percpu_cnt >= 0); 180 for (i = 0; i <= mac_tx_percpu_cnt; i++) { 181 mutex_init(&mcip->mci_tx_pcpu[i].pcpu_tx_lock, NULL, 182 MUTEX_DRIVER, NULL); 183 } 184 cv_init(&mcip->mci_tx_cv, NULL, CV_DRIVER, NULL); 185 186 return (0); 187 } 188 189 /* ARGSUSED */ 190 static void 191 i_mac_client_impl_dtor(void *buf, void *arg) 192 { 193 int i; 194 mac_client_impl_t *mcip = buf; 195 196 ASSERT(mcip->mci_promisc_list == NULL); 197 ASSERT(mcip->mci_unicast_list == NULL); 198 ASSERT(mcip->mci_state_flags == 0); 199 ASSERT(mcip->mci_tx_flag == 0); 200 201 mutex_destroy(&mcip->mci_tx_cb_lock); 202 203 ASSERT(mac_tx_percpu_cnt >= 0); 204 for (i = 0; i <= mac_tx_percpu_cnt; i++) { 205 ASSERT(mcip->mci_tx_pcpu[i].pcpu_tx_refcnt == 0); 206 mutex_destroy(&mcip->mci_tx_pcpu[i].pcpu_tx_lock); 207 } 208 cv_destroy(&mcip->mci_tx_cv); 209 } 210 211 /* ARGSUSED */ 212 static int 213 i_mac_promisc_impl_ctor(void *buf, void *arg, int kmflag) 214 { 215 mac_promisc_impl_t *mpip = buf; 216 217 bzero(buf, sizeof (mac_promisc_impl_t)); 218 mpip->mpi_mci_link.mcb_objp = buf; 219 mpip->mpi_mci_link.mcb_objsize = sizeof (mac_promisc_impl_t); 220 mpip->mpi_mi_link.mcb_objp = buf; 221 mpip->mpi_mi_link.mcb_objsize = sizeof (mac_promisc_impl_t); 222 return (0); 223 } 224 225 /* ARGSUSED */ 226 static void 227 i_mac_promisc_impl_dtor(void *buf, void *arg) 228 { 229 mac_promisc_impl_t *mpip = buf; 230 231 ASSERT(mpip->mpi_mci_link.mcb_objp != NULL); 232 ASSERT(mpip->mpi_mci_link.mcb_objsize == sizeof (mac_promisc_impl_t)); 233 ASSERT(mpip->mpi_mi_link.mcb_objp == mpip->mpi_mci_link.mcb_objp); 234 ASSERT(mpip->mpi_mi_link.mcb_objsize == sizeof (mac_promisc_impl_t)); 235 236 mpip->mpi_mci_link.mcb_objp = NULL; 237 mpip->mpi_mci_link.mcb_objsize = 0; 238 mpip->mpi_mi_link.mcb_objp = NULL; 239 mpip->mpi_mi_link.mcb_objsize = 0; 240 241 ASSERT(mpip->mpi_mci_link.mcb_flags == 0); 242 mpip->mpi_mci_link.mcb_objsize = 0; 243 } 244 245 void 246 mac_client_init(void) 247 { 248 ASSERT(mac_tx_percpu_cnt >= 0); 249 250 mac_client_impl_cache = kmem_cache_create("mac_client_impl_cache", 251 MAC_CLIENT_IMPL_SIZE, 0, i_mac_client_impl_ctor, 252 i_mac_client_impl_dtor, NULL, NULL, NULL, 0); 253 ASSERT(mac_client_impl_cache != NULL); 254 255 mac_promisc_impl_cache = kmem_cache_create("mac_promisc_impl_cache", 256 sizeof (mac_promisc_impl_t), 0, i_mac_promisc_impl_ctor, 257 i_mac_promisc_impl_dtor, NULL, NULL, NULL, 0); 258 ASSERT(mac_promisc_impl_cache != NULL); 259 } 260 261 void 262 mac_client_fini(void) 263 { 264 kmem_cache_destroy(mac_client_impl_cache); 265 kmem_cache_destroy(mac_promisc_impl_cache); 266 } 267 268 /* 269 * Return the lower MAC client handle from the VNIC driver for the 270 * specified VNIC MAC instance. 271 */ 272 mac_client_impl_t * 273 mac_vnic_lower(mac_impl_t *mip) 274 { 275 mac_capab_vnic_t cap; 276 mac_client_impl_t *mcip; 277 278 VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap)); 279 mcip = cap.mcv_mac_client_handle(cap.mcv_arg); 280 281 return (mcip); 282 } 283 284 /* 285 * Update the secondary macs 286 */ 287 void 288 mac_vnic_secondary_update(mac_impl_t *mip) 289 { 290 mac_capab_vnic_t cap; 291 292 VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap)); 293 cap.mcv_mac_secondary_update(cap.mcv_arg); 294 } 295 296 /* 297 * Return the MAC client handle of the primary MAC client for the 298 * specified MAC instance, or NULL otherwise. 299 */ 300 mac_client_impl_t * 301 mac_primary_client_handle(mac_impl_t *mip) 302 { 303 mac_client_impl_t *mcip; 304 305 if (mip->mi_state_flags & MIS_IS_VNIC) 306 return (mac_vnic_lower(mip)); 307 308 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 309 310 for (mcip = mip->mi_clients_list; mcip != NULL; 311 mcip = mcip->mci_client_next) { 312 if (MCIP_DATAPATH_SETUP(mcip) && mac_is_primary_client(mcip)) 313 return (mcip); 314 } 315 return (NULL); 316 } 317 318 /* 319 * Open a MAC specified by its MAC name. 320 */ 321 int 322 mac_open(const char *macname, mac_handle_t *mhp) 323 { 324 mac_impl_t *mip; 325 int err; 326 327 /* 328 * Look up its entry in the global hash table. 329 */ 330 if ((err = mac_hold(macname, &mip)) != 0) 331 return (err); 332 333 /* 334 * Hold the dip associated to the MAC to prevent it from being 335 * detached. For a softmac, its underlying dip is held by the 336 * mi_open() callback. 337 * 338 * This is done to be more tolerant with some defective drivers, 339 * which incorrectly handle mac_unregister() failure in their 340 * xxx_detach() routine. For example, some drivers ignore the 341 * failure of mac_unregister() and free all resources that 342 * that are needed for data transmition. 343 */ 344 e_ddi_hold_devi(mip->mi_dip); 345 346 if (!(mip->mi_callbacks->mc_callbacks & MC_OPEN)) { 347 *mhp = (mac_handle_t)mip; 348 return (0); 349 } 350 351 /* 352 * The mac perimeter is used in both mac_open and mac_close by the 353 * framework to single thread the MC_OPEN/MC_CLOSE of drivers. 354 */ 355 i_mac_perim_enter(mip); 356 mip->mi_oref++; 357 if (mip->mi_oref != 1 || ((err = mip->mi_open(mip->mi_driver)) == 0)) { 358 *mhp = (mac_handle_t)mip; 359 i_mac_perim_exit(mip); 360 return (0); 361 } 362 mip->mi_oref--; 363 ddi_release_devi(mip->mi_dip); 364 mac_rele(mip); 365 i_mac_perim_exit(mip); 366 return (err); 367 } 368 369 /* 370 * Open a MAC specified by its linkid. 371 */ 372 int 373 mac_open_by_linkid(datalink_id_t linkid, mac_handle_t *mhp) 374 { 375 dls_dl_handle_t dlh; 376 int err; 377 378 if ((err = dls_devnet_hold_tmp(linkid, &dlh)) != 0) 379 return (err); 380 381 dls_devnet_prop_task_wait(dlh); 382 383 err = mac_open(dls_devnet_mac(dlh), mhp); 384 385 dls_devnet_rele_tmp(dlh); 386 return (err); 387 } 388 389 /* 390 * Open a MAC specified by its link name. 391 */ 392 int 393 mac_open_by_linkname(const char *link, mac_handle_t *mhp) 394 { 395 datalink_id_t linkid; 396 int err; 397 398 if ((err = dls_mgmt_get_linkid(link, &linkid)) != 0) 399 return (err); 400 return (mac_open_by_linkid(linkid, mhp)); 401 } 402 403 /* 404 * Close the specified MAC. 405 */ 406 void 407 mac_close(mac_handle_t mh) 408 { 409 mac_impl_t *mip = (mac_impl_t *)mh; 410 411 i_mac_perim_enter(mip); 412 /* 413 * The mac perimeter is used in both mac_open and mac_close by the 414 * framework to single thread the MC_OPEN/MC_CLOSE of drivers. 415 */ 416 if (mip->mi_callbacks->mc_callbacks & MC_OPEN) { 417 ASSERT(mip->mi_oref != 0); 418 if (--mip->mi_oref == 0) { 419 if ((mip->mi_callbacks->mc_callbacks & MC_CLOSE)) 420 mip->mi_close(mip->mi_driver); 421 } 422 } 423 i_mac_perim_exit(mip); 424 ddi_release_devi(mip->mi_dip); 425 mac_rele(mip); 426 } 427 428 /* 429 * Misc utility functions to retrieve various information about a MAC 430 * instance or a MAC client. 431 */ 432 433 const mac_info_t * 434 mac_info(mac_handle_t mh) 435 { 436 return (&((mac_impl_t *)mh)->mi_info); 437 } 438 439 dev_info_t * 440 mac_devinfo_get(mac_handle_t mh) 441 { 442 return (((mac_impl_t *)mh)->mi_dip); 443 } 444 445 void * 446 mac_driver(mac_handle_t mh) 447 { 448 return (((mac_impl_t *)mh)->mi_driver); 449 } 450 451 const char * 452 mac_name(mac_handle_t mh) 453 { 454 return (((mac_impl_t *)mh)->mi_name); 455 } 456 457 int 458 mac_type(mac_handle_t mh) 459 { 460 return (((mac_impl_t *)mh)->mi_type->mt_type); 461 } 462 463 int 464 mac_nativetype(mac_handle_t mh) 465 { 466 return (((mac_impl_t *)mh)->mi_type->mt_nativetype); 467 } 468 469 char * 470 mac_client_name(mac_client_handle_t mch) 471 { 472 return (((mac_client_impl_t *)mch)->mci_name); 473 } 474 475 minor_t 476 mac_minor(mac_handle_t mh) 477 { 478 return (((mac_impl_t *)mh)->mi_minor); 479 } 480 481 /* 482 * Return the VID associated with a MAC client. This function should 483 * be called for clients which are associated with only one VID. 484 */ 485 uint16_t 486 mac_client_vid(mac_client_handle_t mch) 487 { 488 uint16_t vid = VLAN_ID_NONE; 489 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 490 flow_desc_t flow_desc; 491 492 if (mcip->mci_nflents == 0) 493 return (vid); 494 495 ASSERT(MCIP_DATAPATH_SETUP(mcip) && mac_client_single_rcvr(mcip)); 496 497 mac_flow_get_desc(mcip->mci_flent, &flow_desc); 498 if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0) 499 vid = flow_desc.fd_vid; 500 501 return (vid); 502 } 503 504 /* 505 * Return whether the specified MAC client corresponds to a VLAN VNIC. 506 */ 507 boolean_t 508 mac_client_is_vlan_vnic(mac_client_handle_t mch) 509 { 510 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 511 512 return (((mcip->mci_state_flags & MCIS_IS_VNIC) != 0) && 513 ((mcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) != 0)); 514 } 515 516 /* 517 * Return the link speed associated with the specified MAC client. 518 * 519 * The link speed of a MAC client is equal to the smallest value of 520 * 1) the current link speed of the underlying NIC, or 521 * 2) the bandwidth limit set for the MAC client. 522 * 523 * Note that the bandwidth limit can be higher than the speed 524 * of the underlying NIC. This is allowed to avoid spurious 525 * administration action failures or artifically lowering the 526 * bandwidth limit of a link that may have temporarily lowered 527 * its link speed due to hardware problem or administrator action. 528 */ 529 static uint64_t 530 mac_client_ifspeed(mac_client_impl_t *mcip) 531 { 532 mac_impl_t *mip = mcip->mci_mip; 533 uint64_t nic_speed; 534 535 nic_speed = mac_stat_get((mac_handle_t)mip, MAC_STAT_IFSPEED); 536 537 if (nic_speed == 0) { 538 return (0); 539 } else { 540 uint64_t policy_limit = (uint64_t)-1; 541 542 if (MCIP_RESOURCE_PROPS_MASK(mcip) & MRP_MAXBW) 543 policy_limit = MCIP_RESOURCE_PROPS_MAXBW(mcip); 544 545 return (MIN(policy_limit, nic_speed)); 546 } 547 } 548 549 /* 550 * Return the link state of the specified client. If here are more 551 * than one clients of the underying mac_impl_t, the link state 552 * will always be UP regardless of the link state of the underlying 553 * mac_impl_t. This is needed to allow the MAC clients to continue 554 * to communicate with each other even when the physical link of 555 * their mac_impl_t is down. 556 */ 557 static uint64_t 558 mac_client_link_state(mac_client_impl_t *mcip) 559 { 560 mac_impl_t *mip = mcip->mci_mip; 561 uint16_t vid; 562 mac_client_impl_t *mci_list; 563 mac_unicast_impl_t *mui_list, *oth_mui_list; 564 565 /* 566 * Returns LINK_STATE_UP if there are other MAC clients defined on 567 * mac_impl_t which share same VLAN ID as that of mcip. Note that 568 * if 'mcip' has more than one VID's then we match ANY one of the 569 * VID's with other MAC client's VID's and return LINK_STATE_UP. 570 */ 571 rw_enter(&mcip->mci_rw_lock, RW_READER); 572 for (mui_list = mcip->mci_unicast_list; mui_list != NULL; 573 mui_list = mui_list->mui_next) { 574 vid = mui_list->mui_vid; 575 for (mci_list = mip->mi_clients_list; mci_list != NULL; 576 mci_list = mci_list->mci_client_next) { 577 if (mci_list == mcip) 578 continue; 579 for (oth_mui_list = mci_list->mci_unicast_list; 580 oth_mui_list != NULL; oth_mui_list = oth_mui_list-> 581 mui_next) { 582 if (vid == oth_mui_list->mui_vid) { 583 rw_exit(&mcip->mci_rw_lock); 584 return (LINK_STATE_UP); 585 } 586 } 587 } 588 } 589 rw_exit(&mcip->mci_rw_lock); 590 591 return (mac_stat_get((mac_handle_t)mip, MAC_STAT_LINK_STATE)); 592 } 593 594 /* 595 * These statistics are consumed by dladm show-link -s <vnic>, 596 * dladm show-vnic -s and netstat. With the introduction of dlstat, 597 * dladm show-link -s and dladm show-vnic -s witll be EOL'ed while 598 * netstat will consume from kstats introduced for dlstat. This code 599 * will be removed at that time. 600 */ 601 602 /* 603 * Return the statistics of a MAC client. These statistics are different 604 * then the statistics of the underlying MAC which are returned by 605 * mac_stat_get(). 606 * 607 * Note that for things based on the tx and rx stats, mac will end up clobbering 608 * those stats when the underlying set of rings in the srs changes. As such, we 609 * need to source not only the current set, but also the historical set when 610 * returning to the client, lest our counters appear to go backwards. 611 */ 612 uint64_t 613 mac_client_stat_get(mac_client_handle_t mch, uint_t stat) 614 { 615 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 616 mac_impl_t *mip = mcip->mci_mip; 617 flow_entry_t *flent = mcip->mci_flent; 618 mac_soft_ring_set_t *mac_srs; 619 mac_rx_stats_t *mac_rx_stat, *old_rx_stat; 620 mac_tx_stats_t *mac_tx_stat, *old_tx_stat; 621 int i; 622 uint64_t val = 0; 623 624 mac_srs = (mac_soft_ring_set_t *)(flent->fe_tx_srs); 625 mac_tx_stat = &mac_srs->srs_tx.st_stat; 626 old_rx_stat = &mcip->mci_misc_stat.mms_defunctrxlanestats; 627 old_tx_stat = &mcip->mci_misc_stat.mms_defuncttxlanestats; 628 629 switch (stat) { 630 case MAC_STAT_LINK_STATE: 631 val = mac_client_link_state(mcip); 632 break; 633 case MAC_STAT_LINK_UP: 634 val = (mac_client_link_state(mcip) == LINK_STATE_UP); 635 break; 636 case MAC_STAT_PROMISC: 637 val = mac_stat_get((mac_handle_t)mip, MAC_STAT_PROMISC); 638 break; 639 case MAC_STAT_LOWLINK_STATE: 640 val = mac_stat_get((mac_handle_t)mip, MAC_STAT_LOWLINK_STATE); 641 break; 642 case MAC_STAT_IFSPEED: 643 val = mac_client_ifspeed(mcip); 644 break; 645 case MAC_STAT_MULTIRCV: 646 val = mcip->mci_misc_stat.mms_multircv; 647 break; 648 case MAC_STAT_BRDCSTRCV: 649 val = mcip->mci_misc_stat.mms_brdcstrcv; 650 break; 651 case MAC_STAT_MULTIXMT: 652 val = mcip->mci_misc_stat.mms_multixmt; 653 break; 654 case MAC_STAT_BRDCSTXMT: 655 val = mcip->mci_misc_stat.mms_brdcstxmt; 656 break; 657 case MAC_STAT_OBYTES: 658 val = mac_tx_stat->mts_obytes; 659 val += old_tx_stat->mts_obytes; 660 break; 661 case MAC_STAT_OPACKETS: 662 val = mac_tx_stat->mts_opackets; 663 val += old_tx_stat->mts_opackets; 664 break; 665 case MAC_STAT_OERRORS: 666 val = mac_tx_stat->mts_oerrors; 667 val += old_tx_stat->mts_oerrors; 668 break; 669 case MAC_STAT_IPACKETS: 670 for (i = 0; i < flent->fe_rx_srs_cnt; i++) { 671 mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i]; 672 mac_rx_stat = &mac_srs->srs_rx.sr_stat; 673 val += mac_rx_stat->mrs_intrcnt + 674 mac_rx_stat->mrs_pollcnt + mac_rx_stat->mrs_lclcnt; 675 } 676 val += old_rx_stat->mrs_intrcnt + old_rx_stat->mrs_pollcnt + 677 old_rx_stat->mrs_lclcnt; 678 break; 679 case MAC_STAT_RBYTES: 680 for (i = 0; i < flent->fe_rx_srs_cnt; i++) { 681 mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i]; 682 mac_rx_stat = &mac_srs->srs_rx.sr_stat; 683 val += mac_rx_stat->mrs_intrbytes + 684 mac_rx_stat->mrs_pollbytes + 685 mac_rx_stat->mrs_lclbytes; 686 } 687 val += old_rx_stat->mrs_intrbytes + old_rx_stat->mrs_pollbytes + 688 old_rx_stat->mrs_lclbytes; 689 break; 690 case MAC_STAT_IERRORS: 691 for (i = 0; i < flent->fe_rx_srs_cnt; i++) { 692 mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i]; 693 mac_rx_stat = &mac_srs->srs_rx.sr_stat; 694 val += mac_rx_stat->mrs_ierrors; 695 } 696 val += old_rx_stat->mrs_ierrors; 697 break; 698 default: 699 val = mac_driver_stat_default(mip, stat); 700 break; 701 } 702 703 return (val); 704 } 705 706 /* 707 * Return the statistics of the specified MAC instance. 708 */ 709 uint64_t 710 mac_stat_get(mac_handle_t mh, uint_t stat) 711 { 712 mac_impl_t *mip = (mac_impl_t *)mh; 713 uint64_t val; 714 int ret; 715 716 /* 717 * The range of stat determines where it is maintained. Stat 718 * values from 0 up to (but not including) MAC_STAT_MIN are 719 * mainteined by the mac module itself. Everything else is 720 * maintained by the driver. 721 * 722 * If the mac_impl_t being queried corresponds to a VNIC, 723 * the stats need to be queried from the lower MAC client 724 * corresponding to the VNIC. (The mac_link_update() 725 * invoked by the driver to the lower MAC causes the *lower 726 * MAC* to update its mi_linkstate, and send a notification 727 * to its MAC clients. Due to the VNIC passthrough, 728 * these notifications are sent to the upper MAC clients 729 * of the VNIC directly, and the upper mac_impl_t of the VNIC 730 * does not have a valid mi_linkstate. 731 */ 732 if (stat < MAC_STAT_MIN && !(mip->mi_state_flags & MIS_IS_VNIC)) { 733 /* these stats are maintained by the mac module itself */ 734 switch (stat) { 735 case MAC_STAT_LINK_STATE: 736 return (mip->mi_linkstate); 737 case MAC_STAT_LINK_UP: 738 return (mip->mi_linkstate == LINK_STATE_UP); 739 case MAC_STAT_PROMISC: 740 return (mip->mi_devpromisc != 0); 741 case MAC_STAT_LOWLINK_STATE: 742 return (mip->mi_lowlinkstate); 743 default: 744 ASSERT(B_FALSE); 745 } 746 } 747 748 /* 749 * Call the driver to get the given statistic. 750 */ 751 ret = mip->mi_getstat(mip->mi_driver, stat, &val); 752 if (ret != 0) { 753 /* 754 * The driver doesn't support this statistic. Get the 755 * statistic's default value. 756 */ 757 val = mac_driver_stat_default(mip, stat); 758 } 759 return (val); 760 } 761 762 /* 763 * Query hardware rx ring corresponding to the pseudo ring. 764 */ 765 uint64_t 766 mac_pseudo_rx_ring_stat_get(mac_ring_handle_t handle, uint_t stat) 767 { 768 return (mac_rx_ring_stat_get(handle, stat)); 769 } 770 771 /* 772 * Query hardware tx ring corresponding to the pseudo ring. 773 */ 774 uint64_t 775 mac_pseudo_tx_ring_stat_get(mac_ring_handle_t handle, uint_t stat) 776 { 777 return (mac_tx_ring_stat_get(handle, stat)); 778 } 779 780 /* 781 * Utility function which returns the VID associated with a flow entry. 782 */ 783 uint16_t 784 i_mac_flow_vid(flow_entry_t *flent) 785 { 786 flow_desc_t flow_desc; 787 788 mac_flow_get_desc(flent, &flow_desc); 789 790 if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0) 791 return (flow_desc.fd_vid); 792 return (VLAN_ID_NONE); 793 } 794 795 /* 796 * Verify the validity of the specified unicast MAC address. Returns B_TRUE 797 * if the address is valid, B_FALSE otherwise (multicast address, or incorrect 798 * length. 799 */ 800 boolean_t 801 mac_unicst_verify(mac_handle_t mh, const uint8_t *addr, uint_t len) 802 { 803 mac_impl_t *mip = (mac_impl_t *)mh; 804 805 /* 806 * Verify the address. No lock is needed since mi_type and plugin 807 * details don't change after mac_register(). 808 */ 809 if ((len != mip->mi_type->mt_addr_length) || 810 (mip->mi_type->mt_ops.mtops_unicst_verify(addr, 811 mip->mi_pdata)) != 0) { 812 return (B_FALSE); 813 } else { 814 return (B_TRUE); 815 } 816 } 817 818 void 819 mac_sdu_get(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu) 820 { 821 mac_impl_t *mip = (mac_impl_t *)mh; 822 823 if (min_sdu != NULL) 824 *min_sdu = mip->mi_sdu_min; 825 if (max_sdu != NULL) 826 *max_sdu = mip->mi_sdu_max; 827 } 828 829 void 830 mac_sdu_get2(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu, 831 uint_t *multicast_sdu) 832 { 833 mac_impl_t *mip = (mac_impl_t *)mh; 834 835 if (min_sdu != NULL) 836 *min_sdu = mip->mi_sdu_min; 837 if (max_sdu != NULL) 838 *max_sdu = mip->mi_sdu_max; 839 if (multicast_sdu != NULL) 840 *multicast_sdu = mip->mi_sdu_multicast; 841 } 842 843 /* 844 * Update the MAC unicast address of the specified client's flows. Currently 845 * only one unicast MAC unicast address is allowed per client. 846 */ 847 static void 848 mac_unicast_update_client_flow(mac_client_impl_t *mcip) 849 { 850 mac_impl_t *mip = mcip->mci_mip; 851 flow_entry_t *flent = mcip->mci_flent; 852 mac_address_t *map = mcip->mci_unicast; 853 flow_desc_t flow_desc; 854 855 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 856 ASSERT(flent != NULL); 857 858 mac_flow_get_desc(flent, &flow_desc); 859 ASSERT(flow_desc.fd_mask & FLOW_LINK_DST); 860 861 bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len); 862 mac_flow_set_desc(flent, &flow_desc); 863 864 /* 865 * The v6 local and SLAAC addrs (used by mac protection) need to be 866 * regenerated because our mac address has changed. 867 */ 868 mac_protect_update_mac_token(mcip); 869 870 /* 871 * When there are multiple VLANs sharing the same MAC address, 872 * each gets its own MAC client, except when running on sun4v 873 * vsw. In that case the mci_flent_list is used to place 874 * multiple VLAN flows on one MAC client. If we ever get rid 875 * of vsw then this code can go, but until then we need to 876 * update all flow entries. 877 */ 878 for (flent = mcip->mci_flent_list; flent != NULL; 879 flent = flent->fe_client_next) { 880 mac_flow_get_desc(flent, &flow_desc); 881 if (!(flent->fe_type & FLOW_PRIMARY_MAC || 882 flent->fe_type & FLOW_VNIC_MAC)) 883 continue; 884 885 bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len); 886 mac_flow_set_desc(flent, &flow_desc); 887 } 888 } 889 890 /* 891 * Update all clients that share the same unicast address. 892 */ 893 void 894 mac_unicast_update_clients(mac_impl_t *mip, mac_address_t *map) 895 { 896 mac_client_impl_t *mcip; 897 898 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 899 900 /* 901 * Find all clients that share the same unicast MAC address and update 902 * them appropriately. 903 */ 904 for (mcip = mip->mi_clients_list; mcip != NULL; 905 mcip = mcip->mci_client_next) { 906 /* 907 * Ignore clients that don't share this MAC address. 908 */ 909 if (map != mcip->mci_unicast) 910 continue; 911 912 /* 913 * Update those clients with same old unicast MAC address. 914 */ 915 mac_unicast_update_client_flow(mcip); 916 } 917 } 918 919 /* 920 * Update the unicast MAC address of the specified VNIC MAC client. 921 * 922 * Check whether the operation is valid. Any of following cases should fail: 923 * 924 * 1. It's a VLAN type of VNIC. 925 * 2. The new value is current "primary" MAC address. 926 * 3. The current MAC address is shared with other clients. 927 * 4. The new MAC address has been used. This case will be valid when 928 * client migration is fully supported. 929 */ 930 int 931 mac_vnic_unicast_set(mac_client_handle_t mch, const uint8_t *addr) 932 { 933 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 934 mac_impl_t *mip = mcip->mci_mip; 935 mac_address_t *map = mcip->mci_unicast; 936 int err; 937 938 ASSERT(!(mip->mi_state_flags & MIS_IS_VNIC)); 939 ASSERT(mcip->mci_state_flags & MCIS_IS_VNIC); 940 ASSERT(mcip->mci_flags != MAC_CLIENT_FLAGS_PRIMARY); 941 942 i_mac_perim_enter(mip); 943 944 /* 945 * If this is a VLAN type of VNIC, it's using "primary" MAC address 946 * of the underlying interface. Must fail here. Refer to case 1 above. 947 */ 948 if (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) == 0) { 949 i_mac_perim_exit(mip); 950 return (ENOTSUP); 951 } 952 953 /* 954 * If the new address is the "primary" one, must fail. Refer to 955 * case 2 above. 956 */ 957 if (bcmp(addr, mip->mi_addr, map->ma_len) == 0) { 958 i_mac_perim_exit(mip); 959 return (EACCES); 960 } 961 962 /* 963 * If the address is shared by multiple clients, must fail. Refer 964 * to case 3 above. 965 */ 966 if (mac_check_macaddr_shared(map)) { 967 i_mac_perim_exit(mip); 968 return (EBUSY); 969 } 970 971 /* 972 * If the new address has been used, must fail for now. Refer to 973 * case 4 above. 974 */ 975 if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) { 976 i_mac_perim_exit(mip); 977 return (ENOTSUP); 978 } 979 980 /* 981 * Update the MAC address. 982 */ 983 err = mac_update_macaddr(map, (uint8_t *)addr); 984 985 if (err != 0) { 986 i_mac_perim_exit(mip); 987 return (err); 988 } 989 990 /* 991 * Update all flows of this MAC client. 992 */ 993 mac_unicast_update_client_flow(mcip); 994 995 i_mac_perim_exit(mip); 996 return (0); 997 } 998 999 /* 1000 * Program the new primary unicast address of the specified MAC. 1001 * 1002 * Function mac_update_macaddr() takes care different types of underlying 1003 * MAC. If the underlying MAC is VNIC, the VNIC driver must have registerd 1004 * mi_unicst() entry point, that indirectly calls mac_vnic_unicast_set() 1005 * which will take care of updating the MAC address of the corresponding 1006 * MAC client. 1007 * 1008 * This is the only interface that allow the client to update the "primary" 1009 * MAC address of the underlying MAC. The new value must have not been 1010 * used by other clients. 1011 */ 1012 int 1013 mac_unicast_primary_set(mac_handle_t mh, const uint8_t *addr) 1014 { 1015 mac_impl_t *mip = (mac_impl_t *)mh; 1016 mac_address_t *map; 1017 int err; 1018 1019 /* verify the address validity */ 1020 if (!mac_unicst_verify(mh, addr, mip->mi_type->mt_addr_length)) 1021 return (EINVAL); 1022 1023 i_mac_perim_enter(mip); 1024 1025 /* 1026 * If the new value is the same as the current primary address value, 1027 * there's nothing to do. 1028 */ 1029 if (bcmp(addr, mip->mi_addr, mip->mi_type->mt_addr_length) == 0) { 1030 i_mac_perim_exit(mip); 1031 return (0); 1032 } 1033 1034 if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) { 1035 i_mac_perim_exit(mip); 1036 return (EBUSY); 1037 } 1038 1039 map = mac_find_macaddr(mip, mip->mi_addr); 1040 ASSERT(map != NULL); 1041 1042 /* 1043 * Update the MAC address. 1044 */ 1045 if (mip->mi_state_flags & MIS_IS_AGGR) { 1046 mac_capab_aggr_t aggr_cap; 1047 1048 /* 1049 * If the MAC is an aggregation, other than the unicast 1050 * addresses programming, aggr must be informed about this 1051 * primary unicst address change to change its MAC address 1052 * policy to be user-specified. 1053 */ 1054 ASSERT(map->ma_type == MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED); 1055 VERIFY(i_mac_capab_get(mh, MAC_CAPAB_AGGR, &aggr_cap)); 1056 err = aggr_cap.mca_unicst(mip->mi_driver, addr); 1057 if (err == 0) 1058 bcopy(addr, map->ma_addr, map->ma_len); 1059 } else { 1060 err = mac_update_macaddr(map, (uint8_t *)addr); 1061 } 1062 1063 if (err != 0) { 1064 i_mac_perim_exit(mip); 1065 return (err); 1066 } 1067 1068 mac_unicast_update_clients(mip, map); 1069 1070 /* 1071 * Save the new primary MAC address in mac_impl_t. 1072 */ 1073 bcopy(addr, mip->mi_addr, mip->mi_type->mt_addr_length); 1074 1075 i_mac_perim_exit(mip); 1076 1077 if (err == 0) 1078 i_mac_notify(mip, MAC_NOTE_UNICST); 1079 1080 return (err); 1081 } 1082 1083 /* 1084 * Return the current primary MAC address of the specified MAC. 1085 */ 1086 void 1087 mac_unicast_primary_get(mac_handle_t mh, uint8_t *addr) 1088 { 1089 mac_impl_t *mip = (mac_impl_t *)mh; 1090 1091 rw_enter(&mip->mi_rw_lock, RW_READER); 1092 bcopy(mip->mi_addr, addr, mip->mi_type->mt_addr_length); 1093 rw_exit(&mip->mi_rw_lock); 1094 } 1095 1096 /* 1097 * Return the secondary MAC address for the specified handle 1098 */ 1099 void 1100 mac_unicast_secondary_get(mac_client_handle_t mh, uint8_t *addr) 1101 { 1102 mac_client_impl_t *mcip = (mac_client_impl_t *)mh; 1103 1104 ASSERT(mcip->mci_unicast != NULL); 1105 bcopy(mcip->mci_unicast->ma_addr, addr, mcip->mci_unicast->ma_len); 1106 } 1107 1108 /* 1109 * Return information about the use of the primary MAC address of the 1110 * specified MAC instance: 1111 * 1112 * - if client_name is non-NULL, it must point to a string of at 1113 * least MAXNAMELEN bytes, and will be set to the name of the MAC 1114 * client which uses the primary MAC address. 1115 * 1116 * - if in_use is non-NULL, used to return whether the primary MAC 1117 * address is currently in use. 1118 */ 1119 void 1120 mac_unicast_primary_info(mac_handle_t mh, char *client_name, boolean_t *in_use) 1121 { 1122 mac_impl_t *mip = (mac_impl_t *)mh; 1123 mac_client_impl_t *cur_client; 1124 1125 if (in_use != NULL) 1126 *in_use = B_FALSE; 1127 if (client_name != NULL) 1128 bzero(client_name, MAXNAMELEN); 1129 1130 /* 1131 * The mi_rw_lock is used to protect threads that don't hold the 1132 * mac perimeter to get a consistent view of the mi_clients_list. 1133 * Threads that modify the list must hold both the mac perimeter and 1134 * mi_rw_lock(RW_WRITER) 1135 */ 1136 rw_enter(&mip->mi_rw_lock, RW_READER); 1137 for (cur_client = mip->mi_clients_list; cur_client != NULL; 1138 cur_client = cur_client->mci_client_next) { 1139 if (mac_is_primary_client(cur_client) || 1140 (mip->mi_state_flags & MIS_IS_VNIC)) { 1141 rw_exit(&mip->mi_rw_lock); 1142 if (in_use != NULL) 1143 *in_use = B_TRUE; 1144 if (client_name != NULL) { 1145 bcopy(cur_client->mci_name, client_name, 1146 MAXNAMELEN); 1147 } 1148 return; 1149 } 1150 } 1151 rw_exit(&mip->mi_rw_lock); 1152 } 1153 1154 /* 1155 * Return the current destination MAC address of the specified MAC. 1156 */ 1157 boolean_t 1158 mac_dst_get(mac_handle_t mh, uint8_t *addr) 1159 { 1160 mac_impl_t *mip = (mac_impl_t *)mh; 1161 1162 rw_enter(&mip->mi_rw_lock, RW_READER); 1163 if (mip->mi_dstaddr_set) 1164 bcopy(mip->mi_dstaddr, addr, mip->mi_type->mt_addr_length); 1165 rw_exit(&mip->mi_rw_lock); 1166 return (mip->mi_dstaddr_set); 1167 } 1168 1169 /* 1170 * Add the specified MAC client to the list of clients which opened 1171 * the specified MAC. 1172 */ 1173 static void 1174 mac_client_add(mac_client_impl_t *mcip) 1175 { 1176 mac_impl_t *mip = mcip->mci_mip; 1177 1178 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 1179 1180 /* add VNIC to the front of the list */ 1181 rw_enter(&mip->mi_rw_lock, RW_WRITER); 1182 mcip->mci_client_next = mip->mi_clients_list; 1183 mip->mi_clients_list = mcip; 1184 mip->mi_nclients++; 1185 rw_exit(&mip->mi_rw_lock); 1186 } 1187 1188 /* 1189 * Remove the specified MAC client from the list of clients which opened 1190 * the specified MAC. 1191 */ 1192 static void 1193 mac_client_remove(mac_client_impl_t *mcip) 1194 { 1195 mac_impl_t *mip = mcip->mci_mip; 1196 mac_client_impl_t **prev, *cclient; 1197 1198 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 1199 1200 rw_enter(&mip->mi_rw_lock, RW_WRITER); 1201 prev = &mip->mi_clients_list; 1202 cclient = *prev; 1203 while (cclient != NULL && cclient != mcip) { 1204 prev = &cclient->mci_client_next; 1205 cclient = *prev; 1206 } 1207 ASSERT(cclient != NULL); 1208 *prev = cclient->mci_client_next; 1209 mip->mi_nclients--; 1210 rw_exit(&mip->mi_rw_lock); 1211 } 1212 1213 static mac_unicast_impl_t * 1214 mac_client_find_vid(mac_client_impl_t *mcip, uint16_t vid) 1215 { 1216 mac_unicast_impl_t *muip = mcip->mci_unicast_list; 1217 1218 while ((muip != NULL) && (muip->mui_vid != vid)) 1219 muip = muip->mui_next; 1220 1221 return (muip); 1222 } 1223 1224 /* 1225 * Return whether the specified (MAC address, VID) tuple is already used by 1226 * one of the MAC clients associated with the specified MAC. 1227 */ 1228 static boolean_t 1229 mac_addr_in_use(mac_impl_t *mip, uint8_t *mac_addr, uint16_t vid) 1230 { 1231 mac_client_impl_t *client; 1232 mac_address_t *map; 1233 1234 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 1235 1236 for (client = mip->mi_clients_list; client != NULL; 1237 client = client->mci_client_next) { 1238 1239 /* 1240 * Ignore clients that don't have unicast address. 1241 */ 1242 if (client->mci_unicast_list == NULL) 1243 continue; 1244 1245 map = client->mci_unicast; 1246 1247 if ((bcmp(mac_addr, map->ma_addr, map->ma_len) == 0) && 1248 (mac_client_find_vid(client, vid) != NULL)) { 1249 return (B_TRUE); 1250 } 1251 } 1252 1253 return (B_FALSE); 1254 } 1255 1256 /* 1257 * Generate a random MAC address. The MAC address prefix is 1258 * stored in the array pointed to by mac_addr, and its length, in bytes, 1259 * is specified by prefix_len. The least significant bits 1260 * after prefix_len bytes are generated, and stored after the prefix 1261 * in the mac_addr array. 1262 */ 1263 int 1264 mac_addr_random(mac_client_handle_t mch, uint_t prefix_len, 1265 uint8_t *mac_addr, mac_diag_t *diag) 1266 { 1267 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 1268 mac_impl_t *mip = mcip->mci_mip; 1269 size_t addr_len = mip->mi_type->mt_addr_length; 1270 1271 if (prefix_len >= addr_len) { 1272 *diag = MAC_DIAG_MACPREFIXLEN_INVALID; 1273 return (EINVAL); 1274 } 1275 1276 /* check the prefix value */ 1277 if (prefix_len > 0) { 1278 bzero(mac_addr + prefix_len, addr_len - prefix_len); 1279 if (!mac_unicst_verify((mac_handle_t)mip, mac_addr, 1280 addr_len)) { 1281 *diag = MAC_DIAG_MACPREFIX_INVALID; 1282 return (EINVAL); 1283 } 1284 } 1285 1286 /* generate the MAC address */ 1287 if (prefix_len < addr_len) { 1288 (void) random_get_pseudo_bytes(mac_addr + 1289 prefix_len, addr_len - prefix_len); 1290 } 1291 1292 *diag = MAC_DIAG_NONE; 1293 return (0); 1294 } 1295 1296 /* 1297 * Set the priority range for this MAC client. This will be used to 1298 * determine the absolute priority for the threads created for this 1299 * MAC client using the specified "low", "medium" and "high" level. 1300 * This will also be used for any subflows on this MAC client. 1301 */ 1302 #define MAC_CLIENT_SET_PRIORITY_RANGE(mcip, pri) { \ 1303 (mcip)->mci_min_pri = FLOW_MIN_PRIORITY(MINCLSYSPRI, \ 1304 MAXCLSYSPRI, (pri)); \ 1305 (mcip)->mci_max_pri = FLOW_MAX_PRIORITY(MINCLSYSPRI, \ 1306 MAXCLSYSPRI, (mcip)->mci_min_pri); \ 1307 } 1308 1309 /* 1310 * MAC client open entry point. Return a new MAC client handle. Each 1311 * MAC client is associated with a name, specified through the 'name' 1312 * argument. 1313 */ 1314 int 1315 mac_client_open(mac_handle_t mh, mac_client_handle_t *mchp, char *name, 1316 uint16_t flags) 1317 { 1318 mac_impl_t *mip = (mac_impl_t *)mh; 1319 mac_client_impl_t *mcip; 1320 int err = 0; 1321 boolean_t share_desired; 1322 flow_entry_t *flent = NULL; 1323 1324 share_desired = (flags & MAC_OPEN_FLAGS_SHARES_DESIRED) != 0; 1325 *mchp = NULL; 1326 1327 i_mac_perim_enter(mip); 1328 1329 if (mip->mi_state_flags & MIS_IS_VNIC) { 1330 /* 1331 * The underlying MAC is a VNIC. Return the MAC client 1332 * handle of the lower MAC which was obtained by 1333 * the VNIC driver when it did its mac_client_open(). 1334 */ 1335 1336 mcip = mac_vnic_lower(mip); 1337 1338 /* 1339 * Note that multiple mac clients share the same mcip in 1340 * this case. 1341 */ 1342 if (flags & MAC_OPEN_FLAGS_EXCLUSIVE) 1343 mcip->mci_state_flags |= MCIS_EXCLUSIVE; 1344 1345 if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY) 1346 mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY; 1347 1348 mip->mi_clients_list = mcip; 1349 i_mac_perim_exit(mip); 1350 *mchp = (mac_client_handle_t)mcip; 1351 1352 DTRACE_PROBE2(mac__client__open__nonallocated, mac_impl_t *, 1353 mcip->mci_mip, mac_client_impl_t *, mcip); 1354 1355 return (err); 1356 } 1357 1358 mcip = kmem_cache_alloc(mac_client_impl_cache, KM_SLEEP); 1359 1360 mcip->mci_mip = mip; 1361 mcip->mci_upper_mip = NULL; 1362 mcip->mci_rx_fn = mac_rx_def; 1363 mcip->mci_rx_arg = NULL; 1364 mcip->mci_rx_p_fn = NULL; 1365 mcip->mci_rx_p_arg = NULL; 1366 mcip->mci_p_unicast_list = NULL; 1367 mcip->mci_direct_rx.mdrx_v4 = NULL; 1368 mcip->mci_direct_rx.mdrx_v6 = NULL; 1369 mcip->mci_direct_rx.mdrx_arg_v4 = NULL; 1370 mcip->mci_direct_rx.mdrx_arg_v6 = NULL; 1371 mcip->mci_vidcache = MCIP_VIDCACHE_INVALID; 1372 1373 mcip->mci_unicast_list = NULL; 1374 1375 if ((flags & MAC_OPEN_FLAGS_IS_VNIC) != 0) 1376 mcip->mci_state_flags |= MCIS_IS_VNIC; 1377 1378 if ((flags & MAC_OPEN_FLAGS_EXCLUSIVE) != 0) 1379 mcip->mci_state_flags |= MCIS_EXCLUSIVE; 1380 1381 if ((flags & MAC_OPEN_FLAGS_IS_AGGR_PORT) != 0) 1382 mcip->mci_state_flags |= MCIS_IS_AGGR_PORT; 1383 1384 if (mip->mi_state_flags & MIS_IS_AGGR) 1385 mcip->mci_state_flags |= MCIS_IS_AGGR_CLIENT; 1386 1387 if ((flags & MAC_OPEN_FLAGS_USE_DATALINK_NAME) != 0) { 1388 datalink_id_t linkid; 1389 1390 ASSERT(name == NULL); 1391 if ((err = dls_devnet_macname2linkid(mip->mi_name, 1392 &linkid)) != 0) { 1393 goto done; 1394 } 1395 if ((err = dls_mgmt_get_linkinfo(linkid, mcip->mci_name, NULL, 1396 NULL, NULL)) != 0) { 1397 /* 1398 * Use mac name if dlmgmtd is not available. 1399 */ 1400 if (err == EBADF) { 1401 (void) strlcpy(mcip->mci_name, mip->mi_name, 1402 sizeof (mcip->mci_name)); 1403 err = 0; 1404 } else { 1405 goto done; 1406 } 1407 } 1408 mcip->mci_state_flags |= MCIS_USE_DATALINK_NAME; 1409 } else { 1410 ASSERT(name != NULL); 1411 if (strlen(name) > MAXNAMELEN) { 1412 err = EINVAL; 1413 goto done; 1414 } 1415 (void) strlcpy(mcip->mci_name, name, sizeof (mcip->mci_name)); 1416 } 1417 1418 if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY) 1419 mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY; 1420 1421 if (flags & MAC_OPEN_FLAGS_NO_UNICAST_ADDR) 1422 mcip->mci_state_flags |= MCIS_NO_UNICAST_ADDR; 1423 1424 mac_protect_init(mcip); 1425 1426 /* the subflow table will be created dynamically */ 1427 mcip->mci_subflow_tab = NULL; 1428 1429 mcip->mci_misc_stat.mms_multircv = 0; 1430 mcip->mci_misc_stat.mms_brdcstrcv = 0; 1431 mcip->mci_misc_stat.mms_multixmt = 0; 1432 mcip->mci_misc_stat.mms_brdcstxmt = 0; 1433 1434 /* Create an initial flow */ 1435 1436 err = mac_flow_create(NULL, NULL, mcip->mci_name, NULL, 1437 mcip->mci_state_flags & MCIS_IS_VNIC ? FLOW_VNIC_MAC : 1438 FLOW_PRIMARY_MAC, &flent); 1439 if (err != 0) 1440 goto done; 1441 mcip->mci_flent = flent; 1442 FLOW_MARK(flent, FE_MC_NO_DATAPATH); 1443 flent->fe_mcip = mcip; 1444 1445 /* 1446 * Place initial creation reference on the flow. This reference 1447 * is released in the corresponding delete action viz. 1448 * mac_unicast_remove after waiting for all transient refs to 1449 * to go away. The wait happens in mac_flow_wait. 1450 */ 1451 FLOW_REFHOLD(flent); 1452 1453 /* 1454 * Do this ahead of the mac_bcast_add() below so that the mi_nclients 1455 * will have the right value for mac_rx_srs_setup(). 1456 */ 1457 mac_client_add(mcip); 1458 1459 mcip->mci_share = 0; 1460 if (share_desired) 1461 i_mac_share_alloc(mcip); 1462 1463 /* 1464 * We will do mimimal datapath setup to allow a MAC client to 1465 * transmit or receive non-unicast packets without waiting 1466 * for mac_unicast_add. 1467 */ 1468 if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) { 1469 if ((err = mac_client_datapath_setup(mcip, VLAN_ID_NONE, 1470 NULL, NULL, B_TRUE, NULL)) != 0) { 1471 goto done; 1472 } 1473 } 1474 1475 DTRACE_PROBE2(mac__client__open__allocated, mac_impl_t *, 1476 mcip->mci_mip, mac_client_impl_t *, mcip); 1477 1478 *mchp = (mac_client_handle_t)mcip; 1479 i_mac_perim_exit(mip); 1480 return (0); 1481 1482 done: 1483 i_mac_perim_exit(mip); 1484 mcip->mci_state_flags = 0; 1485 mcip->mci_tx_flag = 0; 1486 kmem_cache_free(mac_client_impl_cache, mcip); 1487 return (err); 1488 } 1489 1490 /* 1491 * Close the specified MAC client handle. 1492 */ 1493 void 1494 mac_client_close(mac_client_handle_t mch, uint16_t flags) 1495 { 1496 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 1497 mac_impl_t *mip = mcip->mci_mip; 1498 flow_entry_t *flent; 1499 1500 i_mac_perim_enter(mip); 1501 1502 if (flags & MAC_CLOSE_FLAGS_EXCLUSIVE) 1503 mcip->mci_state_flags &= ~MCIS_EXCLUSIVE; 1504 1505 if ((mcip->mci_state_flags & MCIS_IS_VNIC) && 1506 !(flags & MAC_CLOSE_FLAGS_IS_VNIC)) { 1507 /* 1508 * This is an upper VNIC client initiated operation. 1509 * The lower MAC client will be closed by the VNIC driver 1510 * when the VNIC is deleted. 1511 */ 1512 1513 i_mac_perim_exit(mip); 1514 return; 1515 } 1516 1517 /* If we have only setup up minimal datapth setup, tear it down */ 1518 if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) { 1519 mac_client_datapath_teardown((mac_client_handle_t)mcip, NULL, 1520 mcip->mci_flent); 1521 mcip->mci_state_flags &= ~MCIS_NO_UNICAST_ADDR; 1522 } 1523 1524 /* 1525 * Remove the flent associated with the MAC client 1526 */ 1527 flent = mcip->mci_flent; 1528 mcip->mci_flent = NULL; 1529 FLOW_FINAL_REFRELE(flent); 1530 1531 /* 1532 * MAC clients must remove the unicast addresses and promisc callbacks 1533 * they added before issuing a mac_client_close(). 1534 */ 1535 ASSERT(mcip->mci_unicast_list == NULL); 1536 ASSERT(mcip->mci_promisc_list == NULL); 1537 ASSERT(mcip->mci_tx_notify_cb_list == NULL); 1538 1539 i_mac_share_free(mcip); 1540 mac_protect_fini(mcip); 1541 mac_client_remove(mcip); 1542 1543 i_mac_perim_exit(mip); 1544 mcip->mci_subflow_tab = NULL; 1545 mcip->mci_state_flags = 0; 1546 mcip->mci_tx_flag = 0; 1547 kmem_cache_free(mac_client_impl_cache, mch); 1548 } 1549 1550 /* 1551 * Set the Rx bypass receive callback and return B_TRUE. Return 1552 * B_FALSE if it's not possible to enable bypass. 1553 */ 1554 boolean_t 1555 mac_rx_bypass_set(mac_client_handle_t mch, mac_direct_rx_t rx_fn, void *arg1, 1556 boolean_t v6) 1557 { 1558 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 1559 mac_impl_t *mip = mcip->mci_mip; 1560 1561 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 1562 1563 /* 1564 * If the client has more than one VLAN then process packets 1565 * through DLS. This should happen only when sun4v vsw is on 1566 * the scene. 1567 */ 1568 if (mcip->mci_nvids > 1) 1569 return (B_FALSE); 1570 1571 /* 1572 * These are not accessed directly in the data path, and hence 1573 * don't need any protection 1574 */ 1575 if (v6) { 1576 mcip->mci_direct_rx.mdrx_v6 = rx_fn; 1577 mcip->mci_direct_rx.mdrx_arg_v6 = arg1; 1578 } else { 1579 mcip->mci_direct_rx.mdrx_v4 = rx_fn; 1580 mcip->mci_direct_rx.mdrx_arg_v4 = arg1; 1581 } 1582 return (B_TRUE); 1583 } 1584 1585 /* 1586 * Enable/Disable rx bypass. By default, bypass is assumed to be enabled. 1587 */ 1588 void 1589 mac_rx_bypass_enable(mac_client_handle_t mch) 1590 { 1591 ((mac_client_impl_t *)mch)->mci_state_flags &= ~MCIS_RX_BYPASS_DISABLE; 1592 } 1593 1594 void 1595 mac_rx_bypass_disable(mac_client_handle_t mch) 1596 { 1597 ((mac_client_impl_t *)mch)->mci_state_flags |= MCIS_RX_BYPASS_DISABLE; 1598 } 1599 1600 /* 1601 * Set the receive callback for the specified MAC client. There can be 1602 * at most one such callback per MAC client. 1603 */ 1604 void 1605 mac_rx_set(mac_client_handle_t mch, mac_rx_t rx_fn, void *arg) 1606 { 1607 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 1608 mac_impl_t *mip = mcip->mci_mip; 1609 mac_impl_t *umip = mcip->mci_upper_mip; 1610 1611 /* 1612 * Instead of adding an extra set of locks and refcnts in 1613 * the datapath at the mac client boundary, we temporarily quiesce 1614 * the SRS and related entities. We then change the receive function 1615 * without interference from any receive data thread and then reenable 1616 * the data flow subsequently. 1617 */ 1618 i_mac_perim_enter(mip); 1619 mac_rx_client_quiesce(mch); 1620 1621 mcip->mci_rx_fn = rx_fn; 1622 mcip->mci_rx_arg = arg; 1623 mac_rx_client_restart(mch); 1624 1625 /* 1626 * If we're changing the Rx function on the primary MAC of a VNIC, 1627 * make sure any secondary addresses on the VNIC are updated as well. 1628 */ 1629 if (umip != NULL) { 1630 ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0); 1631 mac_vnic_secondary_update(umip); 1632 } 1633 1634 i_mac_perim_exit(mip); 1635 } 1636 1637 /* 1638 * Reset the receive callback for the specified MAC client. 1639 */ 1640 void 1641 mac_rx_clear(mac_client_handle_t mch) 1642 { 1643 mac_rx_set(mch, mac_rx_def, NULL); 1644 } 1645 1646 void 1647 mac_rx_barrier(mac_client_handle_t mch) 1648 { 1649 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 1650 mac_impl_t *mip = mcip->mci_mip; 1651 1652 i_mac_perim_enter(mip); 1653 1654 /* If a RX callback is set, quiesce and restart that datapath */ 1655 if (mcip->mci_rx_fn != mac_rx_def) { 1656 mac_rx_client_quiesce(mch); 1657 mac_rx_client_restart(mch); 1658 } 1659 1660 /* If any promisc callbacks are registered, perform a barrier there */ 1661 if (mcip->mci_promisc_list != NULL || mip->mi_promisc_list != NULL) { 1662 mac_cb_info_t *mcbi = &mip->mi_promisc_cb_info; 1663 1664 mutex_enter(mcbi->mcbi_lockp); 1665 mac_callback_barrier(mcbi); 1666 mutex_exit(mcbi->mcbi_lockp); 1667 } 1668 1669 i_mac_perim_exit(mip); 1670 } 1671 1672 void 1673 mac_secondary_dup(mac_client_handle_t smch, mac_client_handle_t dmch) 1674 { 1675 mac_client_impl_t *smcip = (mac_client_impl_t *)smch; 1676 mac_client_impl_t *dmcip = (mac_client_impl_t *)dmch; 1677 flow_entry_t *flent = dmcip->mci_flent; 1678 1679 /* This should only be called to setup secondary macs */ 1680 ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0); 1681 1682 mac_rx_set(dmch, smcip->mci_rx_fn, smcip->mci_rx_arg); 1683 dmcip->mci_promisc_list = smcip->mci_promisc_list; 1684 1685 /* 1686 * Duplicate the primary mac resources to the secondary. 1687 * Since we already validated the resource controls when setting 1688 * them on the primary, we can ignore errors here. 1689 */ 1690 (void) mac_resource_ctl_set(dmch, MCIP_RESOURCE_PROPS(smcip)); 1691 } 1692 1693 /* 1694 * Called when removing a secondary MAC. Currently only clears the promisc_list 1695 * since we share the primary mac's promisc_list. 1696 */ 1697 void 1698 mac_secondary_cleanup(mac_client_handle_t mch) 1699 { 1700 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 1701 flow_entry_t *flent = mcip->mci_flent; 1702 1703 /* This should only be called for secondary macs */ 1704 ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0); 1705 mcip->mci_promisc_list = NULL; 1706 } 1707 1708 /* 1709 * Walk the MAC client subflow table and updates their priority values. 1710 */ 1711 static int 1712 mac_update_subflow_priority_cb(flow_entry_t *flent, void *arg) 1713 { 1714 mac_flow_update_priority(arg, flent); 1715 return (0); 1716 } 1717 1718 void 1719 mac_update_subflow_priority(mac_client_impl_t *mcip) 1720 { 1721 (void) mac_flow_walk(mcip->mci_subflow_tab, 1722 mac_update_subflow_priority_cb, mcip); 1723 } 1724 1725 /* 1726 * Modify the TX or RX ring properties. We could either just move around 1727 * rings, i.e add/remove rings given to a client. Or this might cause the 1728 * client to move from hardware based to software or the other way around. 1729 * If we want to reset this property, then we clear the mask, additionally 1730 * if the client was given a non-default group we remove all rings except 1731 * for 1 and give it back to the default group. 1732 */ 1733 int 1734 mac_client_set_rings_prop(mac_client_impl_t *mcip, mac_resource_props_t *mrp, 1735 mac_resource_props_t *tmrp) 1736 { 1737 mac_impl_t *mip = mcip->mci_mip; 1738 flow_entry_t *flent = mcip->mci_flent; 1739 uint8_t *mac_addr; 1740 int err = 0; 1741 mac_group_t *defgrp; 1742 mac_group_t *group; 1743 mac_group_t *ngrp; 1744 mac_resource_props_t *cmrp = MCIP_RESOURCE_PROPS(mcip); 1745 uint_t ringcnt; 1746 boolean_t unspec; 1747 1748 if (mcip->mci_share != 0) 1749 return (EINVAL); 1750 1751 if (mrp->mrp_mask & MRP_RX_RINGS) { 1752 unspec = mrp->mrp_mask & MRP_RXRINGS_UNSPEC; 1753 group = flent->fe_rx_ring_group; 1754 defgrp = MAC_DEFAULT_RX_GROUP(mip); 1755 mac_addr = flent->fe_flow_desc.fd_dst_mac; 1756 1757 /* 1758 * No resulting change. If we are resetting on a client on 1759 * which there was no rx rings property. For dynamic group 1760 * if we are setting the same number of rings already set. 1761 * For static group if we are requesting a group again. 1762 */ 1763 if (mrp->mrp_mask & MRP_RINGS_RESET) { 1764 if (!(tmrp->mrp_mask & MRP_RX_RINGS)) 1765 return (0); 1766 } else { 1767 if (unspec) { 1768 if (tmrp->mrp_mask & MRP_RXRINGS_UNSPEC) 1769 return (0); 1770 } else if (mip->mi_rx_group_type == 1771 MAC_GROUP_TYPE_DYNAMIC) { 1772 if ((tmrp->mrp_mask & MRP_RX_RINGS) && 1773 !(tmrp->mrp_mask & MRP_RXRINGS_UNSPEC) && 1774 mrp->mrp_nrxrings == tmrp->mrp_nrxrings) { 1775 return (0); 1776 } 1777 } 1778 } 1779 /* Resetting the prop */ 1780 if (mrp->mrp_mask & MRP_RINGS_RESET) { 1781 /* 1782 * We will just keep one ring and give others back if 1783 * we are not the primary. For the primary we give 1784 * all the rings in the default group except the 1785 * default ring. If it is a static group, then 1786 * we don't do anything, but clear the MRP_RX_RINGS 1787 * flag. 1788 */ 1789 if (group != defgrp) { 1790 if (mip->mi_rx_group_type == 1791 MAC_GROUP_TYPE_DYNAMIC) { 1792 /* 1793 * This group has reserved rings 1794 * that need to be released now, 1795 * so does the group. 1796 */ 1797 MAC_RX_RING_RELEASED(mip, 1798 group->mrg_cur_count); 1799 MAC_RX_GRP_RELEASED(mip); 1800 if ((flent->fe_type & 1801 FLOW_PRIMARY_MAC) != 0) { 1802 if (mip->mi_nactiveclients == 1803 1) { 1804 (void) 1805 mac_rx_switch_group( 1806 mcip, group, 1807 defgrp); 1808 return (0); 1809 } else { 1810 cmrp->mrp_nrxrings = 1811 group-> 1812 mrg_cur_count + 1813 defgrp-> 1814 mrg_cur_count - 1; 1815 } 1816 } else { 1817 cmrp->mrp_nrxrings = 1; 1818 } 1819 (void) mac_group_ring_modify(mcip, 1820 group, defgrp); 1821 } else { 1822 /* 1823 * If this is a static group, we 1824 * need to release the group. The 1825 * client will remain in the same 1826 * group till some other client 1827 * needs this group. 1828 */ 1829 MAC_RX_GRP_RELEASED(mip); 1830 } 1831 /* Let check if we can give this an excl group */ 1832 } else if (group == defgrp) { 1833 /* 1834 * If multiple clients share an 1835 * address then they must stay on the 1836 * default group. 1837 */ 1838 if (mac_check_macaddr_shared(mcip->mci_unicast)) 1839 return (0); 1840 1841 ngrp = mac_reserve_rx_group(mcip, mac_addr, 1842 B_TRUE); 1843 /* Couldn't give it a group, that's fine */ 1844 if (ngrp == NULL) 1845 return (0); 1846 /* Switch to H/W */ 1847 if (mac_rx_switch_group(mcip, defgrp, ngrp) != 1848 0) { 1849 mac_stop_group(ngrp); 1850 return (0); 1851 } 1852 } 1853 /* 1854 * If the client is in the default group, we will 1855 * just clear the MRP_RX_RINGS and leave it as 1856 * it rather than look for an exclusive group 1857 * for it. 1858 */ 1859 return (0); 1860 } 1861 1862 if (group == defgrp && ((mrp->mrp_nrxrings > 0) || unspec)) { 1863 /* 1864 * We are requesting Rx rings. Try to reserve 1865 * a non-default group. 1866 * 1867 * If multiple clients share an address then 1868 * they must stay on the default group. 1869 */ 1870 if (mac_check_macaddr_shared(mcip->mci_unicast)) 1871 return (EINVAL); 1872 1873 ngrp = mac_reserve_rx_group(mcip, mac_addr, B_TRUE); 1874 if (ngrp == NULL) 1875 return (ENOSPC); 1876 1877 /* Switch to H/W */ 1878 if (mac_rx_switch_group(mcip, defgrp, ngrp) != 0) { 1879 mac_release_rx_group(mcip, ngrp); 1880 return (ENOSPC); 1881 } 1882 MAC_RX_GRP_RESERVED(mip); 1883 if (mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC) 1884 MAC_RX_RING_RESERVED(mip, ngrp->mrg_cur_count); 1885 } else if (group != defgrp && !unspec && 1886 mrp->mrp_nrxrings == 0) { 1887 /* Switch to S/W */ 1888 ringcnt = group->mrg_cur_count; 1889 if (mac_rx_switch_group(mcip, group, defgrp) != 0) 1890 return (ENOSPC); 1891 if (tmrp->mrp_mask & MRP_RX_RINGS) { 1892 MAC_RX_GRP_RELEASED(mip); 1893 if (mip->mi_rx_group_type == 1894 MAC_GROUP_TYPE_DYNAMIC) { 1895 MAC_RX_RING_RELEASED(mip, ringcnt); 1896 } 1897 } 1898 } else if (group != defgrp && mip->mi_rx_group_type == 1899 MAC_GROUP_TYPE_DYNAMIC) { 1900 ringcnt = group->mrg_cur_count; 1901 err = mac_group_ring_modify(mcip, group, defgrp); 1902 if (err != 0) 1903 return (err); 1904 /* 1905 * Update the accounting. If this group 1906 * already had explicitly reserved rings, 1907 * we need to update the rings based on 1908 * the new ring count. If this group 1909 * had not explicitly reserved rings, 1910 * then we just reserve the rings asked for 1911 * and reserve the group. 1912 */ 1913 if (tmrp->mrp_mask & MRP_RX_RINGS) { 1914 if (ringcnt > group->mrg_cur_count) { 1915 MAC_RX_RING_RELEASED(mip, 1916 ringcnt - group->mrg_cur_count); 1917 } else { 1918 MAC_RX_RING_RESERVED(mip, 1919 group->mrg_cur_count - ringcnt); 1920 } 1921 } else { 1922 MAC_RX_RING_RESERVED(mip, group->mrg_cur_count); 1923 MAC_RX_GRP_RESERVED(mip); 1924 } 1925 } 1926 } 1927 if (mrp->mrp_mask & MRP_TX_RINGS) { 1928 unspec = mrp->mrp_mask & MRP_TXRINGS_UNSPEC; 1929 group = flent->fe_tx_ring_group; 1930 defgrp = MAC_DEFAULT_TX_GROUP(mip); 1931 1932 /* 1933 * For static groups we only allow rings=0 or resetting the 1934 * rings property. 1935 */ 1936 if (mrp->mrp_ntxrings > 0 && 1937 mip->mi_tx_group_type != MAC_GROUP_TYPE_DYNAMIC) { 1938 return (ENOTSUP); 1939 } 1940 if (mrp->mrp_mask & MRP_RINGS_RESET) { 1941 if (!(tmrp->mrp_mask & MRP_TX_RINGS)) 1942 return (0); 1943 } else { 1944 if (unspec) { 1945 if (tmrp->mrp_mask & MRP_TXRINGS_UNSPEC) 1946 return (0); 1947 } else if (mip->mi_tx_group_type == 1948 MAC_GROUP_TYPE_DYNAMIC) { 1949 if ((tmrp->mrp_mask & MRP_TX_RINGS) && 1950 !(tmrp->mrp_mask & MRP_TXRINGS_UNSPEC) && 1951 mrp->mrp_ntxrings == tmrp->mrp_ntxrings) { 1952 return (0); 1953 } 1954 } 1955 } 1956 /* Resetting the prop */ 1957 if (mrp->mrp_mask & MRP_RINGS_RESET) { 1958 if (group != defgrp) { 1959 if (mip->mi_tx_group_type == 1960 MAC_GROUP_TYPE_DYNAMIC) { 1961 ringcnt = group->mrg_cur_count; 1962 if ((flent->fe_type & 1963 FLOW_PRIMARY_MAC) != 0) { 1964 mac_tx_client_quiesce( 1965 (mac_client_handle_t) 1966 mcip); 1967 mac_tx_switch_group(mcip, 1968 group, defgrp); 1969 mac_tx_client_restart( 1970 (mac_client_handle_t) 1971 mcip); 1972 MAC_TX_GRP_RELEASED(mip); 1973 MAC_TX_RING_RELEASED(mip, 1974 ringcnt); 1975 return (0); 1976 } 1977 cmrp->mrp_ntxrings = 1; 1978 (void) mac_group_ring_modify(mcip, 1979 group, defgrp); 1980 /* 1981 * This group has reserved rings 1982 * that need to be released now. 1983 */ 1984 MAC_TX_RING_RELEASED(mip, ringcnt); 1985 } 1986 /* 1987 * If this is a static group, we 1988 * need to release the group. The 1989 * client will remain in the same 1990 * group till some other client 1991 * needs this group. 1992 */ 1993 MAC_TX_GRP_RELEASED(mip); 1994 } else if (group == defgrp && 1995 (flent->fe_type & FLOW_PRIMARY_MAC) == 0) { 1996 ngrp = mac_reserve_tx_group(mcip, B_TRUE); 1997 if (ngrp == NULL) 1998 return (0); 1999 mac_tx_client_quiesce( 2000 (mac_client_handle_t)mcip); 2001 mac_tx_switch_group(mcip, defgrp, ngrp); 2002 mac_tx_client_restart( 2003 (mac_client_handle_t)mcip); 2004 } 2005 /* 2006 * If the client is in the default group, we will 2007 * just clear the MRP_TX_RINGS and leave it as 2008 * it rather than look for an exclusive group 2009 * for it. 2010 */ 2011 return (0); 2012 } 2013 2014 /* Switch to H/W */ 2015 if (group == defgrp && ((mrp->mrp_ntxrings > 0) || unspec)) { 2016 ngrp = mac_reserve_tx_group(mcip, B_TRUE); 2017 if (ngrp == NULL) 2018 return (ENOSPC); 2019 mac_tx_client_quiesce((mac_client_handle_t)mcip); 2020 mac_tx_switch_group(mcip, defgrp, ngrp); 2021 mac_tx_client_restart((mac_client_handle_t)mcip); 2022 MAC_TX_GRP_RESERVED(mip); 2023 if (mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC) 2024 MAC_TX_RING_RESERVED(mip, ngrp->mrg_cur_count); 2025 /* Switch to S/W */ 2026 } else if (group != defgrp && !unspec && 2027 mrp->mrp_ntxrings == 0) { 2028 /* Switch to S/W */ 2029 ringcnt = group->mrg_cur_count; 2030 mac_tx_client_quiesce((mac_client_handle_t)mcip); 2031 mac_tx_switch_group(mcip, group, defgrp); 2032 mac_tx_client_restart((mac_client_handle_t)mcip); 2033 if (tmrp->mrp_mask & MRP_TX_RINGS) { 2034 MAC_TX_GRP_RELEASED(mip); 2035 if (mip->mi_tx_group_type == 2036 MAC_GROUP_TYPE_DYNAMIC) { 2037 MAC_TX_RING_RELEASED(mip, ringcnt); 2038 } 2039 } 2040 } else if (group != defgrp && mip->mi_tx_group_type == 2041 MAC_GROUP_TYPE_DYNAMIC) { 2042 ringcnt = group->mrg_cur_count; 2043 err = mac_group_ring_modify(mcip, group, defgrp); 2044 if (err != 0) 2045 return (err); 2046 /* 2047 * Update the accounting. If this group 2048 * already had explicitly reserved rings, 2049 * we need to update the rings based on 2050 * the new ring count. If this group 2051 * had not explicitly reserved rings, 2052 * then we just reserve the rings asked for 2053 * and reserve the group. 2054 */ 2055 if (tmrp->mrp_mask & MRP_TX_RINGS) { 2056 if (ringcnt > group->mrg_cur_count) { 2057 MAC_TX_RING_RELEASED(mip, 2058 ringcnt - group->mrg_cur_count); 2059 } else { 2060 MAC_TX_RING_RESERVED(mip, 2061 group->mrg_cur_count - ringcnt); 2062 } 2063 } else { 2064 MAC_TX_RING_RESERVED(mip, group->mrg_cur_count); 2065 MAC_TX_GRP_RESERVED(mip); 2066 } 2067 } 2068 } 2069 return (0); 2070 } 2071 2072 /* 2073 * When the MAC client is being brought up (i.e. we do a unicast_add) we need 2074 * to initialize the cpu and resource control structure in the 2075 * mac_client_impl_t from the mac_impl_t (i.e if there are any cached 2076 * properties before the flow entry for the unicast address was created). 2077 */ 2078 static int 2079 mac_resource_ctl_set(mac_client_handle_t mch, mac_resource_props_t *mrp) 2080 { 2081 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 2082 mac_impl_t *mip = (mac_impl_t *)mcip->mci_mip; 2083 mac_impl_t *umip = mcip->mci_upper_mip; 2084 int err = 0; 2085 flow_entry_t *flent = mcip->mci_flent; 2086 mac_resource_props_t *omrp, *nmrp = MCIP_RESOURCE_PROPS(mcip); 2087 2088 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 2089 2090 err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ? 2091 mcip->mci_upper_mip : mip, mrp); 2092 if (err != 0) 2093 return (err); 2094 2095 /* 2096 * Copy over the existing properties since mac_update_resources 2097 * will modify the client's mrp. Currently, the saved property 2098 * is used to determine the difference between existing and 2099 * modified rings property. 2100 */ 2101 omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP); 2102 bcopy(nmrp, omrp, sizeof (*omrp)); 2103 mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE); 2104 if (MCIP_DATAPATH_SETUP(mcip)) { 2105 /* 2106 * We support rings only for primary client when there are 2107 * multiple clients sharing the same MAC address (e.g. VLAN). 2108 */ 2109 if (mrp->mrp_mask & MRP_RX_RINGS || 2110 mrp->mrp_mask & MRP_TX_RINGS) { 2111 2112 if ((err = mac_client_set_rings_prop(mcip, mrp, 2113 omrp)) != 0) { 2114 if (omrp->mrp_mask & MRP_RX_RINGS) { 2115 nmrp->mrp_mask |= MRP_RX_RINGS; 2116 nmrp->mrp_nrxrings = omrp->mrp_nrxrings; 2117 } else { 2118 nmrp->mrp_mask &= ~MRP_RX_RINGS; 2119 nmrp->mrp_nrxrings = 0; 2120 } 2121 if (omrp->mrp_mask & MRP_TX_RINGS) { 2122 nmrp->mrp_mask |= MRP_TX_RINGS; 2123 nmrp->mrp_ntxrings = omrp->mrp_ntxrings; 2124 } else { 2125 nmrp->mrp_mask &= ~MRP_TX_RINGS; 2126 nmrp->mrp_ntxrings = 0; 2127 } 2128 if (omrp->mrp_mask & MRP_RXRINGS_UNSPEC) 2129 omrp->mrp_mask |= MRP_RXRINGS_UNSPEC; 2130 else 2131 omrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC; 2132 2133 if (omrp->mrp_mask & MRP_TXRINGS_UNSPEC) 2134 omrp->mrp_mask |= MRP_TXRINGS_UNSPEC; 2135 else 2136 omrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC; 2137 kmem_free(omrp, sizeof (*omrp)); 2138 return (err); 2139 } 2140 2141 /* 2142 * If we modified the rings property of the primary 2143 * we need to update the property fields of its 2144 * VLANs as they inherit the primary's properites. 2145 */ 2146 if (mac_is_primary_client(mcip)) { 2147 mac_set_prim_vlan_rings(mip, 2148 MCIP_RESOURCE_PROPS(mcip)); 2149 } 2150 } 2151 /* 2152 * We have to set this prior to calling mac_flow_modify. 2153 */ 2154 if (mrp->mrp_mask & MRP_PRIORITY) { 2155 if (mrp->mrp_priority == MPL_RESET) { 2156 MAC_CLIENT_SET_PRIORITY_RANGE(mcip, 2157 MPL_LINK_DEFAULT); 2158 } else { 2159 MAC_CLIENT_SET_PRIORITY_RANGE(mcip, 2160 mrp->mrp_priority); 2161 } 2162 } 2163 2164 mac_flow_modify(mip->mi_flow_tab, flent, mrp); 2165 if (mrp->mrp_mask & MRP_PRIORITY) 2166 mac_update_subflow_priority(mcip); 2167 2168 /* Apply these resource settings to any secondary macs */ 2169 if (umip != NULL) { 2170 ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0); 2171 mac_vnic_secondary_update(umip); 2172 } 2173 } 2174 kmem_free(omrp, sizeof (*omrp)); 2175 return (0); 2176 } 2177 2178 static int 2179 mac_unicast_flow_create(mac_client_impl_t *mcip, uint8_t *mac_addr, 2180 uint16_t vid, boolean_t is_primary, boolean_t first_flow, 2181 flow_entry_t **flent, mac_resource_props_t *mrp) 2182 { 2183 mac_impl_t *mip = (mac_impl_t *)mcip->mci_mip; 2184 flow_desc_t flow_desc; 2185 char flowname[MAXFLOWNAMELEN]; 2186 int err; 2187 uint_t flent_flags; 2188 2189 /* 2190 * First unicast address being added, create a new flow 2191 * for that MAC client. 2192 */ 2193 bzero(&flow_desc, sizeof (flow_desc)); 2194 2195 ASSERT(mac_addr != NULL || 2196 (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR)); 2197 if (mac_addr != NULL) { 2198 flow_desc.fd_mac_len = mip->mi_type->mt_addr_length; 2199 bcopy(mac_addr, flow_desc.fd_dst_mac, flow_desc.fd_mac_len); 2200 } 2201 flow_desc.fd_mask = FLOW_LINK_DST; 2202 if (vid != 0) { 2203 flow_desc.fd_vid = vid; 2204 flow_desc.fd_mask |= FLOW_LINK_VID; 2205 } 2206 2207 /* 2208 * XXX-nicolas. For now I'm keeping the FLOW_PRIMARY_MAC 2209 * and FLOW_VNIC. Even though they're a hack inherited 2210 * from the SRS code, we'll keep them for now. They're currently 2211 * consumed by mac_datapath_setup() to create the SRS. 2212 * That code should be eventually moved out of 2213 * mac_datapath_setup() and moved to a mac_srs_create() 2214 * function of some sort to keep things clean. 2215 * 2216 * Also, there's no reason why the SRS for the primary MAC 2217 * client should be different than any other MAC client. Until 2218 * this is cleaned-up, we support only one MAC unicast address 2219 * per client. 2220 * 2221 * We set FLOW_PRIMARY_MAC for the primary MAC address, 2222 * FLOW_VNIC for everything else. 2223 */ 2224 if (is_primary) 2225 flent_flags = FLOW_PRIMARY_MAC; 2226 else 2227 flent_flags = FLOW_VNIC_MAC; 2228 2229 /* 2230 * For the first flow we use the MAC client's name - mci_name, for 2231 * subsequent ones we just create a name with the VID. This is 2232 * so that we can add these flows to the same flow table. This is 2233 * fine as the flow name (except for the one with the MAC client's 2234 * name) is not visible. When the first flow is removed, we just replace 2235 * its fdesc with another from the list, so we will still retain the 2236 * flent with the MAC client's flow name. 2237 */ 2238 if (first_flow) { 2239 bcopy(mcip->mci_name, flowname, MAXFLOWNAMELEN); 2240 } else { 2241 (void) sprintf(flowname, "%s%u", mcip->mci_name, vid); 2242 flent_flags = FLOW_NO_STATS; 2243 } 2244 2245 if ((err = mac_flow_create(&flow_desc, mrp, flowname, NULL, 2246 flent_flags, flent)) != 0) 2247 return (err); 2248 2249 mac_misc_stat_create(*flent); 2250 FLOW_MARK(*flent, FE_INCIPIENT); 2251 (*flent)->fe_mcip = mcip; 2252 2253 /* 2254 * Place initial creation reference on the flow. This reference 2255 * is released in the corresponding delete action viz. 2256 * mac_unicast_remove after waiting for all transient refs to 2257 * to go away. The wait happens in mac_flow_wait. 2258 * We have already held the reference in mac_client_open(). 2259 */ 2260 if (!first_flow) 2261 FLOW_REFHOLD(*flent); 2262 return (0); 2263 } 2264 2265 /* Refresh the multicast grouping for this VID. */ 2266 int 2267 mac_client_update_mcast(void *arg, boolean_t add, const uint8_t *addrp) 2268 { 2269 flow_entry_t *flent = arg; 2270 mac_client_impl_t *mcip = flent->fe_mcip; 2271 uint16_t vid; 2272 flow_desc_t flow_desc; 2273 2274 mac_flow_get_desc(flent, &flow_desc); 2275 vid = (flow_desc.fd_mask & FLOW_LINK_VID) != 0 ? 2276 flow_desc.fd_vid : VLAN_ID_NONE; 2277 2278 /* 2279 * We don't call mac_multicast_add()/mac_multicast_remove() as 2280 * we want to add/remove for this specific vid. 2281 */ 2282 if (add) { 2283 return (mac_bcast_add(mcip, addrp, vid, 2284 MAC_ADDRTYPE_MULTICAST)); 2285 } else { 2286 mac_bcast_delete(mcip, addrp, vid); 2287 return (0); 2288 } 2289 } 2290 2291 static void 2292 mac_update_single_active_client(mac_impl_t *mip) 2293 { 2294 mac_client_impl_t *client = NULL; 2295 2296 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 2297 2298 rw_enter(&mip->mi_rw_lock, RW_WRITER); 2299 if (mip->mi_nactiveclients == 1) { 2300 /* 2301 * Find the one active MAC client from the list of MAC 2302 * clients. The active MAC client has at least one 2303 * unicast address. 2304 */ 2305 for (client = mip->mi_clients_list; client != NULL; 2306 client = client->mci_client_next) { 2307 if (client->mci_unicast_list != NULL) 2308 break; 2309 } 2310 ASSERT(client != NULL); 2311 } 2312 2313 /* 2314 * mi_single_active_client is protected by the MAC impl's read/writer 2315 * lock, which allows mac_rx() to check the value of that pointer 2316 * as a reader. 2317 */ 2318 mip->mi_single_active_client = client; 2319 rw_exit(&mip->mi_rw_lock); 2320 } 2321 2322 /* 2323 * Set up the data path. Called from i_mac_unicast_add after having 2324 * done all the validations including making sure this is an active 2325 * client (i.e that is ready to process packets.) 2326 */ 2327 static int 2328 mac_client_datapath_setup(mac_client_impl_t *mcip, uint16_t vid, 2329 uint8_t *mac_addr, mac_resource_props_t *mrp, boolean_t isprimary, 2330 mac_unicast_impl_t *muip) 2331 { 2332 mac_impl_t *mip = mcip->mci_mip; 2333 boolean_t mac_started = B_FALSE; 2334 boolean_t bcast_added = B_FALSE; 2335 boolean_t nactiveclients_added = B_FALSE; 2336 flow_entry_t *flent; 2337 int err = 0; 2338 boolean_t no_unicast; 2339 2340 no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR; 2341 2342 if ((err = mac_start((mac_handle_t)mip)) != 0) 2343 goto bail; 2344 2345 mac_started = B_TRUE; 2346 2347 /* add the MAC client to the broadcast address group by default */ 2348 if (mip->mi_type->mt_brdcst_addr != NULL) { 2349 err = mac_bcast_add(mcip, mip->mi_type->mt_brdcst_addr, vid, 2350 MAC_ADDRTYPE_BROADCAST); 2351 if (err != 0) 2352 goto bail; 2353 bcast_added = B_TRUE; 2354 } 2355 2356 /* 2357 * If this is the first unicast address addition for this 2358 * client, reuse the pre-allocated larval flow entry associated with 2359 * the MAC client. 2360 */ 2361 flent = (mcip->mci_nflents == 0) ? mcip->mci_flent : NULL; 2362 2363 /* We are configuring the unicast flow now */ 2364 if (!MCIP_DATAPATH_SETUP(mcip)) { 2365 2366 if (mrp != NULL) { 2367 MAC_CLIENT_SET_PRIORITY_RANGE(mcip, 2368 (mrp->mrp_mask & MRP_PRIORITY) ? mrp->mrp_priority : 2369 MPL_LINK_DEFAULT); 2370 } 2371 if ((err = mac_unicast_flow_create(mcip, mac_addr, vid, 2372 isprimary, B_TRUE, &flent, mrp)) != 0) 2373 goto bail; 2374 2375 mip->mi_nactiveclients++; 2376 nactiveclients_added = B_TRUE; 2377 2378 /* 2379 * This will allocate the RX ring group if possible for the 2380 * flow and program the software classifier as needed. 2381 */ 2382 if ((err = mac_datapath_setup(mcip, flent, SRST_LINK)) != 0) 2383 goto bail; 2384 2385 if (no_unicast) 2386 goto done_setup; 2387 /* 2388 * The unicast MAC address must have been added successfully. 2389 */ 2390 ASSERT(mcip->mci_unicast != NULL); 2391 2392 /* 2393 * Push down the sub-flows that were defined on this link 2394 * hitherto. The flows are added to the active flow table 2395 * and SRS, softrings etc. are created as needed. 2396 */ 2397 mac_link_init_flows((mac_client_handle_t)mcip); 2398 } else { 2399 mac_address_t *map = mcip->mci_unicast; 2400 2401 ASSERT(!no_unicast); 2402 /* 2403 * A unicast flow already exists for that MAC client 2404 * so this flow must be the same MAC address but with 2405 * a different VID. It has been checked by 2406 * mac_addr_in_use(). 2407 * 2408 * We will use the SRS etc. from the initial 2409 * mci_flent. We don't need to create a kstat for 2410 * this, as except for the fdesc, everything will be 2411 * used from the first flent. 2412 * 2413 * The only time we should see multiple flents on the 2414 * same MAC client is on the sun4v vsw. If we removed 2415 * that code we should be able to remove the entire 2416 * notion of multiple flents on a MAC client (this 2417 * doesn't affect sub/user flows because they have 2418 * their own list unrelated to mci_flent_list). 2419 */ 2420 if (bcmp(mac_addr, map->ma_addr, map->ma_len) != 0) { 2421 err = EINVAL; 2422 goto bail; 2423 } 2424 2425 if ((err = mac_unicast_flow_create(mcip, mac_addr, vid, 2426 isprimary, B_FALSE, &flent, NULL)) != 0) { 2427 goto bail; 2428 } 2429 if ((err = mac_flow_add(mip->mi_flow_tab, flent)) != 0) { 2430 FLOW_FINAL_REFRELE(flent); 2431 goto bail; 2432 } 2433 2434 /* update the multicast group for this vid */ 2435 mac_client_bcast_refresh(mcip, mac_client_update_mcast, 2436 (void *)flent, B_TRUE); 2437 2438 } 2439 2440 /* populate the shared MAC address */ 2441 muip->mui_map = mcip->mci_unicast; 2442 2443 rw_enter(&mcip->mci_rw_lock, RW_WRITER); 2444 muip->mui_next = mcip->mci_unicast_list; 2445 mcip->mci_unicast_list = muip; 2446 rw_exit(&mcip->mci_rw_lock); 2447 2448 done_setup: 2449 /* 2450 * First add the flent to the flow list of this mcip. Then set 2451 * the mip's mi_single_active_client if needed. The Rx path assumes 2452 * that mip->mi_single_active_client will always have an associated 2453 * flent. 2454 */ 2455 mac_client_add_to_flow_list(mcip, flent); 2456 if (nactiveclients_added) 2457 mac_update_single_active_client(mip); 2458 /* 2459 * Trigger a renegotiation of the capabilities when the number of 2460 * active clients changes from 1 to 2, since some of the capabilities 2461 * might have to be disabled. Also send a MAC_NOTE_LINK notification 2462 * to all the MAC clients whenever physical link is DOWN. 2463 */ 2464 if (mip->mi_nactiveclients == 2) { 2465 mac_capab_update((mac_handle_t)mip); 2466 mac_virtual_link_update(mip); 2467 } 2468 /* 2469 * Now that the setup is complete, clear the INCIPIENT flag. 2470 * The flag was set to avoid incoming packets seeing inconsistent 2471 * structures while the setup was in progress. Clear the mci_tx_flag 2472 * by calling mac_tx_client_block. It is possible that 2473 * mac_unicast_remove was called prior to this mac_unicast_add which 2474 * could have set the MCI_TX_QUIESCE flag. 2475 */ 2476 if (flent->fe_rx_ring_group != NULL) 2477 mac_rx_group_unmark(flent->fe_rx_ring_group, MR_INCIPIENT); 2478 FLOW_UNMARK(flent, FE_INCIPIENT); 2479 2480 /* 2481 * If this is an aggr port client, don't enable the flow's 2482 * datapath at this stage. Otherwise, bcast traffic could 2483 * arrive while the aggr port is in the process of 2484 * initializing. Instead, the flow's datapath is started later 2485 * when mac_client_set_flow_cb() is called. 2486 */ 2487 if ((mcip->mci_state_flags & MCIS_IS_AGGR_PORT) == 0) 2488 FLOW_UNMARK(flent, FE_MC_NO_DATAPATH); 2489 2490 mac_tx_client_unblock(mcip); 2491 return (0); 2492 bail: 2493 if (bcast_added) 2494 mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr, vid); 2495 2496 if (nactiveclients_added) 2497 mip->mi_nactiveclients--; 2498 2499 if (mac_started) 2500 mac_stop((mac_handle_t)mip); 2501 2502 return (err); 2503 } 2504 2505 /* 2506 * Return the passive primary MAC client, if present. The passive client is 2507 * a stand-by client that has the same unicast address as another that is 2508 * currenly active. Once the active client goes away, the passive client 2509 * becomes active. 2510 */ 2511 static mac_client_impl_t * 2512 mac_get_passive_primary_client(mac_impl_t *mip) 2513 { 2514 mac_client_impl_t *mcip; 2515 2516 for (mcip = mip->mi_clients_list; mcip != NULL; 2517 mcip = mcip->mci_client_next) { 2518 if (mac_is_primary_client(mcip) && 2519 (mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) { 2520 return (mcip); 2521 } 2522 } 2523 return (NULL); 2524 } 2525 2526 /* 2527 * Add a new unicast address to the MAC client. 2528 * 2529 * The MAC address can be specified either by value, or the MAC client 2530 * can specify that it wants to use the primary MAC address of the 2531 * underlying MAC. See the introductory comments at the beginning 2532 * of this file for more more information on primary MAC addresses. 2533 * 2534 * Note also the tuple (MAC address, VID) must be unique 2535 * for the MAC clients defined on top of the same underlying MAC 2536 * instance, unless the MAC_UNICAST_NODUPCHECK is specified. 2537 * 2538 * In no case can a client use the PVID for the MAC, if the MAC has one set. 2539 */ 2540 int 2541 i_mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags, 2542 mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag) 2543 { 2544 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 2545 mac_impl_t *mip = mcip->mci_mip; 2546 int err; 2547 uint_t mac_len = mip->mi_type->mt_addr_length; 2548 boolean_t check_dups = !(flags & MAC_UNICAST_NODUPCHECK); 2549 boolean_t fastpath_disabled = B_FALSE; 2550 boolean_t is_primary = (flags & MAC_UNICAST_PRIMARY); 2551 boolean_t is_unicast_hw = (flags & MAC_UNICAST_HW); 2552 mac_resource_props_t *mrp; 2553 boolean_t passive_client = B_FALSE; 2554 mac_unicast_impl_t *muip; 2555 boolean_t is_vnic_primary = 2556 (flags & MAC_UNICAST_VNIC_PRIMARY); 2557 2558 /* 2559 * When the VID is non-zero the underlying MAC cannot be a 2560 * VNIC. I.e., dladm create-vlan cannot take a VNIC as 2561 * argument, only the primary MAC client. 2562 */ 2563 ASSERT(!((mip->mi_state_flags & MIS_IS_VNIC) && (vid != VLAN_ID_NONE))); 2564 2565 *diag = MAC_DIAG_NONE; 2566 2567 /* 2568 * Can't unicast add if the client asked only for minimal datapath 2569 * setup. 2570 */ 2571 if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) 2572 return (ENOTSUP); 2573 2574 /* 2575 * Check for an attempted use of the current Port VLAN ID, if enabled. 2576 * No client may use it. 2577 */ 2578 if (mip->mi_pvid != VLAN_ID_NONE && vid == mip->mi_pvid) 2579 return (EBUSY); 2580 2581 /* 2582 * Check whether it's the primary client and flag it. 2583 */ 2584 if (!(mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary && 2585 vid == VLAN_ID_NONE) 2586 mcip->mci_flags |= MAC_CLIENT_FLAGS_PRIMARY; 2587 2588 /* 2589 * is_vnic_primary is true when we come here as a VLAN VNIC 2590 * which uses the primary MAC client's address but with a non-zero 2591 * VID. In this case the MAC address is not specified by an upper 2592 * MAC client. 2593 */ 2594 if ((mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary && 2595 !is_vnic_primary) { 2596 /* 2597 * The address is being set by the upper MAC client 2598 * of a VNIC. The MAC address was already set by the 2599 * VNIC driver during VNIC creation. 2600 * 2601 * Note: a VNIC has only one MAC address. We return 2602 * the MAC unicast address handle of the lower MAC client 2603 * corresponding to the VNIC. We allocate a new entry 2604 * which is flagged appropriately, so that mac_unicast_remove() 2605 * doesn't attempt to free the original entry that 2606 * was allocated by the VNIC driver. 2607 */ 2608 ASSERT(mcip->mci_unicast != NULL); 2609 2610 /* Check for VLAN flags, if present */ 2611 if ((flags & MAC_UNICAST_TAG_DISABLE) != 0) 2612 mcip->mci_state_flags |= MCIS_TAG_DISABLE; 2613 2614 if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0) 2615 mcip->mci_state_flags |= MCIS_STRIP_DISABLE; 2616 2617 if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0) 2618 mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK; 2619 2620 /* 2621 * Ensure that the primary unicast address of the VNIC 2622 * is added only once unless we have the 2623 * MAC_CLIENT_FLAGS_MULTI_PRIMARY set (and this is not 2624 * a passive MAC client). 2625 */ 2626 if ((mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) != 0) { 2627 if ((mcip->mci_flags & 2628 MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 || 2629 (mcip->mci_flags & 2630 MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) { 2631 return (EBUSY); 2632 } 2633 mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY; 2634 passive_client = B_TRUE; 2635 } 2636 2637 mcip->mci_flags |= MAC_CLIENT_FLAGS_VNIC_PRIMARY; 2638 2639 /* 2640 * Create a handle for vid 0. 2641 */ 2642 ASSERT(vid == VLAN_ID_NONE); 2643 muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP); 2644 muip->mui_vid = vid; 2645 *mah = (mac_unicast_handle_t)muip; 2646 /* 2647 * This will be used by the caller to defer setting the 2648 * rx functions. 2649 */ 2650 if (passive_client) 2651 return (EAGAIN); 2652 return (0); 2653 } 2654 2655 /* primary MAC clients cannot be opened on top of anchor VNICs */ 2656 if ((is_vnic_primary || is_primary) && 2657 i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_ANCHOR_VNIC, NULL)) { 2658 return (ENXIO); 2659 } 2660 2661 /* 2662 * If this is a VNIC/VLAN, disable softmac fast-path. This is 2663 * only relevant to legacy devices which use softmac to 2664 * interface with GLDv3. 2665 */ 2666 if (mcip->mci_state_flags & MCIS_IS_VNIC) { 2667 err = mac_fastpath_disable((mac_handle_t)mip); 2668 if (err != 0) 2669 return (err); 2670 fastpath_disabled = B_TRUE; 2671 } 2672 2673 /* 2674 * Return EBUSY if: 2675 * - there is an exclusively active mac client exists. 2676 * - this is an exclusive active mac client but 2677 * a. there is already active mac clients exist, or 2678 * b. fastpath streams are already plumbed on this legacy device 2679 * - the mac creator has disallowed active mac clients. 2680 */ 2681 if (mip->mi_state_flags & (MIS_EXCLUSIVE|MIS_NO_ACTIVE)) { 2682 if (fastpath_disabled) 2683 mac_fastpath_enable((mac_handle_t)mip); 2684 return (EBUSY); 2685 } 2686 2687 if (mcip->mci_state_flags & MCIS_EXCLUSIVE) { 2688 ASSERT(!fastpath_disabled); 2689 if (mip->mi_nactiveclients != 0) 2690 return (EBUSY); 2691 2692 if ((mip->mi_state_flags & MIS_LEGACY) && 2693 !(mip->mi_capab_legacy.ml_active_set(mip->mi_driver))) { 2694 return (EBUSY); 2695 } 2696 mip->mi_state_flags |= MIS_EXCLUSIVE; 2697 } 2698 2699 mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP); 2700 if (is_primary && !(mcip->mci_state_flags & (MCIS_IS_VNIC | 2701 MCIS_IS_AGGR_PORT))) { 2702 /* 2703 * Apply the property cached in the mac_impl_t to the primary 2704 * mac client. If the mac client is a VNIC or an aggregation 2705 * port, its property should be set in the mcip when the 2706 * VNIC/aggr was created. 2707 */ 2708 mac_get_resources((mac_handle_t)mip, mrp); 2709 (void) mac_client_set_resources(mch, mrp); 2710 } else if (mcip->mci_state_flags & MCIS_IS_VNIC) { 2711 /* 2712 * This is a VLAN client sharing the address of the 2713 * primary MAC client; i.e., one created via dladm 2714 * create-vlan. We don't support specifying ring 2715 * properties for this type of client as it inherits 2716 * these from the primary MAC client. 2717 */ 2718 if (is_vnic_primary) { 2719 mac_resource_props_t *vmrp; 2720 2721 vmrp = MCIP_RESOURCE_PROPS(mcip); 2722 if (vmrp->mrp_mask & MRP_RX_RINGS || 2723 vmrp->mrp_mask & MRP_TX_RINGS) { 2724 if (fastpath_disabled) 2725 mac_fastpath_enable((mac_handle_t)mip); 2726 kmem_free(mrp, sizeof (*mrp)); 2727 return (ENOTSUP); 2728 } 2729 /* 2730 * Additionally we also need to inherit any 2731 * rings property from the MAC. 2732 */ 2733 mac_get_resources((mac_handle_t)mip, mrp); 2734 if (mrp->mrp_mask & MRP_RX_RINGS) { 2735 vmrp->mrp_mask |= MRP_RX_RINGS; 2736 vmrp->mrp_nrxrings = mrp->mrp_nrxrings; 2737 } 2738 if (mrp->mrp_mask & MRP_TX_RINGS) { 2739 vmrp->mrp_mask |= MRP_TX_RINGS; 2740 vmrp->mrp_ntxrings = mrp->mrp_ntxrings; 2741 } 2742 } 2743 bcopy(MCIP_RESOURCE_PROPS(mcip), mrp, sizeof (*mrp)); 2744 } 2745 2746 muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP); 2747 muip->mui_vid = vid; 2748 2749 if (is_primary || is_vnic_primary) { 2750 mac_addr = mip->mi_addr; 2751 } else { 2752 2753 /* 2754 * Verify the validity of the specified MAC addresses value. 2755 */ 2756 if (!mac_unicst_verify((mac_handle_t)mip, mac_addr, mac_len)) { 2757 *diag = MAC_DIAG_MACADDR_INVALID; 2758 err = EINVAL; 2759 goto bail_out; 2760 } 2761 2762 /* 2763 * Make sure that the specified MAC address is different 2764 * than the unicast MAC address of the underlying NIC. 2765 */ 2766 if (check_dups && bcmp(mip->mi_addr, mac_addr, mac_len) == 0) { 2767 *diag = MAC_DIAG_MACADDR_NIC; 2768 err = EINVAL; 2769 goto bail_out; 2770 } 2771 } 2772 2773 /* 2774 * Set the flags here so that if this is a passive client, we 2775 * can return and set it when we call mac_client_datapath_setup 2776 * when this becomes the active client. If we defer to using these 2777 * flags to mac_client_datapath_setup, then for a passive client, 2778 * we'd have to store the flags somewhere (probably fe_flags) 2779 * and then use it. 2780 */ 2781 if (!MCIP_DATAPATH_SETUP(mcip)) { 2782 if (is_unicast_hw) { 2783 /* 2784 * The client requires a hardware MAC address slot 2785 * for that unicast address. Since we support only 2786 * one unicast MAC address per client, flag the 2787 * MAC client itself. 2788 */ 2789 mcip->mci_state_flags |= MCIS_UNICAST_HW; 2790 } 2791 2792 /* Check for VLAN flags, if present */ 2793 if ((flags & MAC_UNICAST_TAG_DISABLE) != 0) 2794 mcip->mci_state_flags |= MCIS_TAG_DISABLE; 2795 2796 if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0) 2797 mcip->mci_state_flags |= MCIS_STRIP_DISABLE; 2798 2799 if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0) 2800 mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK; 2801 } else { 2802 /* 2803 * Assert that the specified flags are consistent with the 2804 * flags specified by previous calls to mac_unicast_add(). 2805 */ 2806 ASSERT(((flags & MAC_UNICAST_TAG_DISABLE) != 0 && 2807 (mcip->mci_state_flags & MCIS_TAG_DISABLE) != 0) || 2808 ((flags & MAC_UNICAST_TAG_DISABLE) == 0 && 2809 (mcip->mci_state_flags & MCIS_TAG_DISABLE) == 0)); 2810 2811 ASSERT(((flags & MAC_UNICAST_STRIP_DISABLE) != 0 && 2812 (mcip->mci_state_flags & MCIS_STRIP_DISABLE) != 0) || 2813 ((flags & MAC_UNICAST_STRIP_DISABLE) == 0 && 2814 (mcip->mci_state_flags & MCIS_STRIP_DISABLE) == 0)); 2815 2816 ASSERT(((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0 && 2817 (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) != 0) || 2818 ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) == 0 && 2819 (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) == 0)); 2820 2821 /* 2822 * Make sure the client is consistent about its requests 2823 * for MAC addresses. I.e. all requests from the clients 2824 * must have the MAC_UNICAST_HW flag set or clear. 2825 */ 2826 if (((mcip->mci_state_flags & MCIS_UNICAST_HW) != 0 && 2827 !is_unicast_hw) || 2828 ((mcip->mci_state_flags & MCIS_UNICAST_HW) == 0 && 2829 is_unicast_hw)) { 2830 err = EINVAL; 2831 goto bail_out; 2832 } 2833 } 2834 /* 2835 * Make sure the MAC address is not already used by 2836 * another MAC client defined on top of the same 2837 * underlying NIC. Unless we have MAC_CLIENT_FLAGS_MULTI_PRIMARY 2838 * set when we allow a passive client to be present which will 2839 * be activated when the currently active client goes away - this 2840 * works only with primary addresses. 2841 */ 2842 if ((check_dups || is_primary || is_vnic_primary) && 2843 mac_addr_in_use(mip, mac_addr, vid)) { 2844 /* 2845 * Must have set the multiple primary address flag when 2846 * we did a mac_client_open AND this should be a primary 2847 * MAC client AND there should not already be a passive 2848 * primary. If all is true then we let this succeed 2849 * even if the address is a dup. 2850 */ 2851 if ((mcip->mci_flags & MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 || 2852 (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) == 0 || 2853 mac_get_passive_primary_client(mip) != NULL) { 2854 *diag = MAC_DIAG_MACADDR_INUSE; 2855 err = EEXIST; 2856 goto bail_out; 2857 } 2858 ASSERT((mcip->mci_flags & 2859 MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) == 0); 2860 mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY; 2861 kmem_free(mrp, sizeof (*mrp)); 2862 2863 /* 2864 * Stash the unicast address handle, we will use it when 2865 * we set up the passive client. 2866 */ 2867 mcip->mci_p_unicast_list = muip; 2868 *mah = (mac_unicast_handle_t)muip; 2869 return (0); 2870 } 2871 2872 err = mac_client_datapath_setup(mcip, vid, mac_addr, mrp, 2873 is_primary || is_vnic_primary, muip); 2874 if (err != 0) 2875 goto bail_out; 2876 2877 kmem_free(mrp, sizeof (*mrp)); 2878 *mah = (mac_unicast_handle_t)muip; 2879 return (0); 2880 2881 bail_out: 2882 if (fastpath_disabled) 2883 mac_fastpath_enable((mac_handle_t)mip); 2884 if (mcip->mci_state_flags & MCIS_EXCLUSIVE) { 2885 mip->mi_state_flags &= ~MIS_EXCLUSIVE; 2886 if (mip->mi_state_flags & MIS_LEGACY) { 2887 mip->mi_capab_legacy.ml_active_clear( 2888 mip->mi_driver); 2889 } 2890 } 2891 kmem_free(mrp, sizeof (*mrp)); 2892 kmem_free(muip, sizeof (mac_unicast_impl_t)); 2893 return (err); 2894 } 2895 2896 /* 2897 * Wrapper function to mac_unicast_add when we want to have the same mac 2898 * client open for two instances, one that is currently active and another 2899 * that will become active when the current one is removed. In this case 2900 * mac_unicast_add will return EGAIN and we will save the rx function and 2901 * arg which will be used when we activate the passive client in 2902 * mac_unicast_remove. 2903 */ 2904 int 2905 mac_unicast_add_set_rx(mac_client_handle_t mch, uint8_t *mac_addr, 2906 uint16_t flags, mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag, 2907 mac_rx_t rx_fn, void *arg) 2908 { 2909 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 2910 uint_t err; 2911 2912 err = mac_unicast_add(mch, mac_addr, flags, mah, vid, diag); 2913 if (err != 0 && err != EAGAIN) 2914 return (err); 2915 if (err == EAGAIN) { 2916 if (rx_fn != NULL) { 2917 mcip->mci_rx_p_fn = rx_fn; 2918 mcip->mci_rx_p_arg = arg; 2919 } 2920 return (0); 2921 } 2922 if (rx_fn != NULL) 2923 mac_rx_set(mch, rx_fn, arg); 2924 return (err); 2925 } 2926 2927 int 2928 mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags, 2929 mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag) 2930 { 2931 mac_impl_t *mip = ((mac_client_impl_t *)mch)->mci_mip; 2932 uint_t err; 2933 2934 i_mac_perim_enter(mip); 2935 err = i_mac_unicast_add(mch, mac_addr, flags, mah, vid, diag); 2936 i_mac_perim_exit(mip); 2937 2938 return (err); 2939 } 2940 2941 static void 2942 mac_client_datapath_teardown(mac_client_handle_t mch, mac_unicast_impl_t *muip, 2943 flow_entry_t *flent) 2944 { 2945 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 2946 mac_impl_t *mip = mcip->mci_mip; 2947 boolean_t no_unicast; 2948 2949 /* 2950 * If we have not added a unicast address for this MAC client, just 2951 * teardown the datapath. 2952 */ 2953 no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR; 2954 2955 if (!no_unicast) { 2956 /* 2957 * We would have initialized subflows etc. only if we brought 2958 * up the primary client and set the unicast unicast address 2959 * etc. Deactivate the flows. The flow entry will be removed 2960 * from the active flow tables, and the associated SRS, 2961 * softrings etc will be deleted. But the flow entry itself 2962 * won't be destroyed, instead it will continue to be archived 2963 * off the the global flow hash list, for a possible future 2964 * activation when say IP is plumbed again. 2965 */ 2966 mac_link_release_flows(mch); 2967 } 2968 mip->mi_nactiveclients--; 2969 mac_update_single_active_client(mip); 2970 2971 /* Tear down the data path */ 2972 mac_datapath_teardown(mcip, mcip->mci_flent, SRST_LINK); 2973 2974 /* 2975 * Prevent any future access to the flow entry through the mci_flent 2976 * pointer by setting the mci_flent to NULL. Access to mci_flent in 2977 * mac_bcast_send is also under mi_rw_lock. 2978 */ 2979 rw_enter(&mip->mi_rw_lock, RW_WRITER); 2980 flent = mcip->mci_flent; 2981 mac_client_remove_flow_from_list(mcip, flent); 2982 2983 if (mcip->mci_state_flags & MCIS_DESC_LOGGED) 2984 mcip->mci_state_flags &= ~MCIS_DESC_LOGGED; 2985 2986 /* 2987 * This is the last unicast address being removed and there shouldn't 2988 * be any outbound data threads at this point coming down from mac 2989 * clients. We have waited for the data threads to finish before 2990 * starting dld_str_detach. Non-data threads must access TX SRS 2991 * under mi_rw_lock. 2992 */ 2993 rw_exit(&mip->mi_rw_lock); 2994 2995 /* 2996 * Don't use FLOW_MARK with FE_MC_NO_DATAPATH, as the flow might 2997 * contain other flags, such as FE_CONDEMNED, which we need to 2998 * cleared. We don't call mac_flow_cleanup() for this unicast 2999 * flow as we have a already cleaned up SRSs etc. (via the teadown 3000 * path). We just clear the stats and reset the initial callback 3001 * function, the rest will be set when we call mac_flow_create, 3002 * if at all. 3003 */ 3004 mutex_enter(&flent->fe_lock); 3005 ASSERT(flent->fe_refcnt == 1 && flent->fe_mbg == NULL && 3006 flent->fe_tx_srs == NULL && flent->fe_rx_srs_cnt == 0); 3007 flent->fe_flags = FE_MC_NO_DATAPATH; 3008 flow_stat_destroy(flent); 3009 mac_misc_stat_delete(flent); 3010 3011 /* Initialize the receiver function to a safe routine */ 3012 flent->fe_cb_fn = (flow_fn_t)mac_rx_def; 3013 flent->fe_cb_arg1 = NULL; 3014 flent->fe_cb_arg2 = NULL; 3015 3016 flent->fe_index = -1; 3017 mutex_exit(&flent->fe_lock); 3018 3019 if (mip->mi_type->mt_brdcst_addr != NULL) { 3020 ASSERT(muip != NULL || no_unicast); 3021 mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr, 3022 muip != NULL ? muip->mui_vid : VLAN_ID_NONE); 3023 } 3024 3025 if (mip->mi_nactiveclients == 1) { 3026 mac_capab_update((mac_handle_t)mip); 3027 mac_virtual_link_update(mip); 3028 } 3029 3030 if (mcip->mci_state_flags & MCIS_EXCLUSIVE) { 3031 mip->mi_state_flags &= ~MIS_EXCLUSIVE; 3032 3033 if (mip->mi_state_flags & MIS_LEGACY) 3034 mip->mi_capab_legacy.ml_active_clear(mip->mi_driver); 3035 } 3036 3037 mcip->mci_state_flags &= ~MCIS_UNICAST_HW; 3038 3039 if (mcip->mci_state_flags & MCIS_TAG_DISABLE) 3040 mcip->mci_state_flags &= ~MCIS_TAG_DISABLE; 3041 3042 if (mcip->mci_state_flags & MCIS_STRIP_DISABLE) 3043 mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE; 3044 3045 if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) 3046 mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK; 3047 3048 if (muip != NULL) 3049 kmem_free(muip, sizeof (mac_unicast_impl_t)); 3050 mac_protect_cancel_timer(mcip); 3051 mac_protect_flush_dynamic(mcip); 3052 3053 bzero(&mcip->mci_misc_stat, sizeof (mcip->mci_misc_stat)); 3054 /* 3055 * Disable fastpath if this is a VNIC or a VLAN. 3056 */ 3057 if (mcip->mci_state_flags & MCIS_IS_VNIC) 3058 mac_fastpath_enable((mac_handle_t)mip); 3059 mac_stop((mac_handle_t)mip); 3060 } 3061 3062 /* 3063 * Remove a MAC address which was previously added by mac_unicast_add(). 3064 */ 3065 int 3066 mac_unicast_remove(mac_client_handle_t mch, mac_unicast_handle_t mah) 3067 { 3068 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 3069 mac_unicast_impl_t *muip = (mac_unicast_impl_t *)mah; 3070 mac_unicast_impl_t *pre; 3071 mac_impl_t *mip = mcip->mci_mip; 3072 flow_entry_t *flent; 3073 uint16_t mui_vid; 3074 3075 i_mac_perim_enter(mip); 3076 if (mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) { 3077 /* 3078 * Call made by the upper MAC client of a VNIC. 3079 * There's nothing much to do, the unicast address will 3080 * be removed by the VNIC driver when the VNIC is deleted, 3081 * but let's ensure that all our transmit is done before 3082 * the client does a mac_client_stop lest it trigger an 3083 * assert in the driver. 3084 */ 3085 ASSERT(muip->mui_vid == VLAN_ID_NONE); 3086 3087 mac_tx_client_flush(mcip); 3088 3089 if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) { 3090 mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY; 3091 if (mcip->mci_rx_p_fn != NULL) { 3092 mac_rx_set(mch, mcip->mci_rx_p_fn, 3093 mcip->mci_rx_p_arg); 3094 mcip->mci_rx_p_fn = NULL; 3095 mcip->mci_rx_p_arg = NULL; 3096 } 3097 kmem_free(muip, sizeof (mac_unicast_impl_t)); 3098 i_mac_perim_exit(mip); 3099 return (0); 3100 } 3101 mcip->mci_flags &= ~MAC_CLIENT_FLAGS_VNIC_PRIMARY; 3102 3103 if (mcip->mci_state_flags & MCIS_TAG_DISABLE) 3104 mcip->mci_state_flags &= ~MCIS_TAG_DISABLE; 3105 3106 if (mcip->mci_state_flags & MCIS_STRIP_DISABLE) 3107 mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE; 3108 3109 if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) 3110 mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK; 3111 3112 kmem_free(muip, sizeof (mac_unicast_impl_t)); 3113 i_mac_perim_exit(mip); 3114 return (0); 3115 } 3116 3117 ASSERT(muip != NULL); 3118 3119 /* 3120 * We are removing a passive client, we haven't setup the datapath 3121 * for this yet, so nothing much to do. 3122 */ 3123 if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) { 3124 3125 ASSERT((mcip->mci_flent->fe_flags & FE_MC_NO_DATAPATH) != 0); 3126 ASSERT(mcip->mci_p_unicast_list == muip); 3127 3128 mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY; 3129 3130 mcip->mci_p_unicast_list = NULL; 3131 mcip->mci_rx_p_fn = NULL; 3132 mcip->mci_rx_p_arg = NULL; 3133 3134 mcip->mci_state_flags &= ~MCIS_UNICAST_HW; 3135 3136 if (mcip->mci_state_flags & MCIS_TAG_DISABLE) 3137 mcip->mci_state_flags &= ~MCIS_TAG_DISABLE; 3138 3139 if (mcip->mci_state_flags & MCIS_STRIP_DISABLE) 3140 mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE; 3141 3142 if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) 3143 mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK; 3144 3145 kmem_free(muip, sizeof (mac_unicast_impl_t)); 3146 i_mac_perim_exit(mip); 3147 return (0); 3148 } 3149 3150 /* 3151 * Remove the VID from the list of client's VIDs. 3152 */ 3153 pre = mcip->mci_unicast_list; 3154 if (muip == pre) { 3155 mcip->mci_unicast_list = muip->mui_next; 3156 } else { 3157 while ((pre->mui_next != NULL) && (pre->mui_next != muip)) 3158 pre = pre->mui_next; 3159 ASSERT(pre->mui_next == muip); 3160 rw_enter(&mcip->mci_rw_lock, RW_WRITER); 3161 pre->mui_next = muip->mui_next; 3162 rw_exit(&mcip->mci_rw_lock); 3163 } 3164 3165 if (!mac_client_single_rcvr(mcip)) { 3166 /* 3167 * This MAC client is shared by more than one unicast 3168 * addresses, so we will just remove the flent 3169 * corresponding to the address being removed. We don't invoke 3170 * mac_rx_classify_flow_rem() since the additional flow is 3171 * not associated with its own separate set of SRS and rings, 3172 * and these constructs are still needed for the remaining 3173 * flows. 3174 */ 3175 flent = mac_client_get_flow(mcip, muip); 3176 VERIFY3P(flent, !=, NULL); 3177 3178 /* 3179 * The first one is disappearing, need to make sure 3180 * we replace it with another from the list of 3181 * shared clients. 3182 */ 3183 if (flent == mcip->mci_flent) 3184 flent = mac_client_swap_mciflent(mcip); 3185 mac_client_remove_flow_from_list(mcip, flent); 3186 mac_flow_remove(mip->mi_flow_tab, flent, B_FALSE); 3187 mac_flow_wait(flent, FLOW_DRIVER_UPCALL); 3188 3189 /* 3190 * The multicast groups that were added by the client so 3191 * far must be removed from the brodcast domain corresponding 3192 * to the VID being removed. 3193 */ 3194 mac_client_bcast_refresh(mcip, mac_client_update_mcast, 3195 (void *)flent, B_FALSE); 3196 3197 if (mip->mi_type->mt_brdcst_addr != NULL) { 3198 mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr, 3199 muip->mui_vid); 3200 } 3201 3202 FLOW_FINAL_REFRELE(flent); 3203 ASSERT(!(mcip->mci_state_flags & MCIS_EXCLUSIVE)); 3204 3205 /* 3206 * Enable fastpath if this is a VNIC or a VLAN. 3207 */ 3208 if (mcip->mci_state_flags & MCIS_IS_VNIC) 3209 mac_fastpath_enable((mac_handle_t)mip); 3210 mac_stop((mac_handle_t)mip); 3211 i_mac_perim_exit(mip); 3212 return (0); 3213 } 3214 3215 mui_vid = muip->mui_vid; 3216 mac_client_datapath_teardown(mch, muip, flent); 3217 3218 if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) && 3219 mui_vid == VLAN_ID_NONE) { 3220 mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PRIMARY; 3221 } else { 3222 i_mac_perim_exit(mip); 3223 return (0); 3224 } 3225 3226 /* 3227 * If we are removing the primary, check if we have a passive primary 3228 * client that we need to activate now. 3229 */ 3230 mcip = mac_get_passive_primary_client(mip); 3231 if (mcip != NULL) { 3232 mac_resource_props_t *mrp; 3233 mac_unicast_impl_t *muip; 3234 3235 mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY; 3236 mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP); 3237 3238 /* 3239 * Apply the property cached in the mac_impl_t to the 3240 * primary mac client. 3241 */ 3242 mac_get_resources((mac_handle_t)mip, mrp); 3243 (void) mac_client_set_resources(mch, mrp); 3244 ASSERT(mcip->mci_p_unicast_list != NULL); 3245 muip = mcip->mci_p_unicast_list; 3246 mcip->mci_p_unicast_list = NULL; 3247 if (mac_client_datapath_setup(mcip, VLAN_ID_NONE, 3248 mip->mi_addr, mrp, B_TRUE, muip) == 0) { 3249 if (mcip->mci_rx_p_fn != NULL) { 3250 mac_rx_set(mch, mcip->mci_rx_p_fn, 3251 mcip->mci_rx_p_arg); 3252 mcip->mci_rx_p_fn = NULL; 3253 mcip->mci_rx_p_arg = NULL; 3254 } 3255 } else { 3256 kmem_free(muip, sizeof (mac_unicast_impl_t)); 3257 } 3258 kmem_free(mrp, sizeof (*mrp)); 3259 } 3260 i_mac_perim_exit(mip); 3261 return (0); 3262 } 3263 3264 /* 3265 * Multicast add function invoked by MAC clients. 3266 */ 3267 int 3268 mac_multicast_add(mac_client_handle_t mch, const uint8_t *addr) 3269 { 3270 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 3271 mac_impl_t *mip = mcip->mci_mip; 3272 flow_entry_t *flent = mcip->mci_flent_list; 3273 flow_entry_t *prev_fe = NULL; 3274 uint16_t vid; 3275 int err = 0; 3276 3277 /* Verify the address is a valid multicast address */ 3278 if ((err = mip->mi_type->mt_ops.mtops_multicst_verify(addr, 3279 mip->mi_pdata)) != 0) 3280 return (err); 3281 3282 i_mac_perim_enter(mip); 3283 while (flent != NULL) { 3284 vid = i_mac_flow_vid(flent); 3285 3286 err = mac_bcast_add((mac_client_impl_t *)mch, addr, vid, 3287 MAC_ADDRTYPE_MULTICAST); 3288 if (err != 0) 3289 break; 3290 prev_fe = flent; 3291 flent = flent->fe_client_next; 3292 } 3293 3294 /* 3295 * If we failed adding, then undo all, rather than partial 3296 * success. 3297 */ 3298 if (flent != NULL && prev_fe != NULL) { 3299 flent = mcip->mci_flent_list; 3300 while (flent != prev_fe->fe_client_next) { 3301 vid = i_mac_flow_vid(flent); 3302 mac_bcast_delete((mac_client_impl_t *)mch, addr, vid); 3303 flent = flent->fe_client_next; 3304 } 3305 } 3306 i_mac_perim_exit(mip); 3307 return (err); 3308 } 3309 3310 /* 3311 * Multicast delete function invoked by MAC clients. 3312 */ 3313 void 3314 mac_multicast_remove(mac_client_handle_t mch, const uint8_t *addr) 3315 { 3316 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 3317 mac_impl_t *mip = mcip->mci_mip; 3318 flow_entry_t *flent; 3319 uint16_t vid; 3320 3321 i_mac_perim_enter(mip); 3322 for (flent = mcip->mci_flent_list; flent != NULL; 3323 flent = flent->fe_client_next) { 3324 vid = i_mac_flow_vid(flent); 3325 mac_bcast_delete((mac_client_impl_t *)mch, addr, vid); 3326 } 3327 i_mac_perim_exit(mip); 3328 } 3329 3330 /* 3331 * When a MAC client desires to capture packets on an interface, 3332 * it registers a promiscuous call back with mac_promisc_add(). 3333 * There are three types of promiscuous callbacks: 3334 * 3335 * * MAC_CLIENT_PROMISC_ALL 3336 * Captures all packets sent and received by the MAC client, 3337 * the physical interface, as well as all other MAC clients 3338 * defined on top of the same MAC. 3339 * 3340 * * MAC_CLIENT_PROMISC_FILTERED 3341 * Captures all packets sent and received by the MAC client, 3342 * plus all multicast traffic sent and received by the phyisical 3343 * interface and the other MAC clients. 3344 * 3345 * * MAC_CLIENT_PROMISC_MULTI 3346 * Captures all broadcast and multicast packets sent and 3347 * received by the MAC clients as well as the physical interface. 3348 * 3349 * In all cases, the underlying MAC is put in promiscuous mode. 3350 */ 3351 int 3352 mac_promisc_add(mac_client_handle_t mch, mac_client_promisc_type_t type, 3353 mac_rx_t fn, void *arg, mac_promisc_handle_t *mphp, uint16_t flags) 3354 { 3355 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 3356 mac_impl_t *mip = mcip->mci_mip; 3357 mac_promisc_impl_t *mpip; 3358 mac_cb_info_t *mcbi; 3359 int rc; 3360 3361 i_mac_perim_enter(mip); 3362 3363 if ((rc = mac_start((mac_handle_t)mip)) != 0) { 3364 i_mac_perim_exit(mip); 3365 return (rc); 3366 } 3367 3368 if ((mcip->mci_state_flags & MCIS_IS_VNIC) && 3369 type == MAC_CLIENT_PROMISC_ALL && 3370 (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED)) { 3371 /* 3372 * The function is being invoked by the upper MAC client 3373 * of a VNIC. The VNIC should only see the traffic 3374 * it is entitled to. 3375 */ 3376 type = MAC_CLIENT_PROMISC_FILTERED; 3377 } 3378 3379 3380 /* 3381 * Turn on promiscuous mode for the underlying NIC. 3382 * This is needed even for filtered callbacks which 3383 * expect to receive all multicast traffic on the wire. 3384 * 3385 * Physical promiscuous mode should not be turned on if 3386 * MAC_PROMISC_FLAGS_NO_PHYS is set. 3387 */ 3388 if ((flags & MAC_PROMISC_FLAGS_NO_PHYS) == 0) { 3389 if ((rc = i_mac_promisc_set(mip, B_TRUE)) != 0) { 3390 mac_stop((mac_handle_t)mip); 3391 i_mac_perim_exit(mip); 3392 return (rc); 3393 } 3394 } 3395 3396 mpip = kmem_cache_alloc(mac_promisc_impl_cache, KM_SLEEP); 3397 3398 mpip->mpi_type = type; 3399 mpip->mpi_fn = fn; 3400 mpip->mpi_arg = arg; 3401 mpip->mpi_mcip = mcip; 3402 mpip->mpi_no_tx_loop = ((flags & MAC_PROMISC_FLAGS_NO_TX_LOOP) != 0); 3403 mpip->mpi_no_phys = ((flags & MAC_PROMISC_FLAGS_NO_PHYS) != 0); 3404 mpip->mpi_strip_vlan_tag = 3405 ((flags & MAC_PROMISC_FLAGS_VLAN_TAG_STRIP) != 0); 3406 mpip->mpi_no_copy = ((flags & MAC_PROMISC_FLAGS_NO_COPY) != 0); 3407 3408 mcbi = &mip->mi_promisc_cb_info; 3409 mutex_enter(mcbi->mcbi_lockp); 3410 3411 mac_callback_add(&mip->mi_promisc_cb_info, &mcip->mci_promisc_list, 3412 &mpip->mpi_mci_link); 3413 mac_callback_add(&mip->mi_promisc_cb_info, &mip->mi_promisc_list, 3414 &mpip->mpi_mi_link); 3415 3416 mutex_exit(mcbi->mcbi_lockp); 3417 3418 *mphp = (mac_promisc_handle_t)mpip; 3419 3420 if (mcip->mci_state_flags & MCIS_IS_VNIC) { 3421 mac_impl_t *umip = mcip->mci_upper_mip; 3422 3423 ASSERT(umip != NULL); 3424 mac_vnic_secondary_update(umip); 3425 } 3426 3427 i_mac_perim_exit(mip); 3428 3429 return (0); 3430 } 3431 3432 /* 3433 * Remove a multicast address previously aded through mac_promisc_add(). 3434 */ 3435 void 3436 mac_promisc_remove(mac_promisc_handle_t mph) 3437 { 3438 mac_promisc_impl_t *mpip = (mac_promisc_impl_t *)mph; 3439 mac_client_impl_t *mcip = mpip->mpi_mcip; 3440 mac_impl_t *mip = mcip->mci_mip; 3441 mac_cb_info_t *mcbi; 3442 int rv; 3443 3444 i_mac_perim_enter(mip); 3445 3446 /* 3447 * Even if the device can't be reset into normal mode, we still 3448 * need to clear the client promisc callbacks. The client may want 3449 * to close the mac end point and we can't have stale callbacks. 3450 */ 3451 if (!(mpip->mpi_no_phys)) { 3452 if ((rv = i_mac_promisc_set(mip, B_FALSE)) != 0) { 3453 cmn_err(CE_WARN, "%s: failed to switch OFF promiscuous" 3454 " mode because of error 0x%x", mip->mi_name, rv); 3455 } 3456 } 3457 mcbi = &mip->mi_promisc_cb_info; 3458 mutex_enter(mcbi->mcbi_lockp); 3459 if (mac_callback_remove(mcbi, &mip->mi_promisc_list, 3460 &mpip->mpi_mi_link)) { 3461 VERIFY(mac_callback_remove(&mip->mi_promisc_cb_info, 3462 &mcip->mci_promisc_list, &mpip->mpi_mci_link)); 3463 kmem_cache_free(mac_promisc_impl_cache, mpip); 3464 } else { 3465 mac_callback_remove_wait(&mip->mi_promisc_cb_info); 3466 } 3467 3468 if (mcip->mci_state_flags & MCIS_IS_VNIC) { 3469 mac_impl_t *umip = mcip->mci_upper_mip; 3470 3471 ASSERT(umip != NULL); 3472 mac_vnic_secondary_update(umip); 3473 } 3474 3475 mutex_exit(mcbi->mcbi_lockp); 3476 mac_stop((mac_handle_t)mip); 3477 3478 i_mac_perim_exit(mip); 3479 } 3480 3481 /* 3482 * Reference count the number of active Tx threads. MCI_TX_QUIESCE indicates 3483 * that a control operation wants to quiesce the Tx data flow in which case 3484 * we return an error. Holding any of the per cpu locks ensures that the 3485 * mci_tx_flag won't change. 3486 * 3487 * 'CPU' must be accessed just once and used to compute the index into the 3488 * percpu array, and that index must be used for the entire duration of the 3489 * packet send operation. Note that the thread may be preempted and run on 3490 * another cpu any time and so we can't use 'CPU' more than once for the 3491 * operation. 3492 */ 3493 #define MAC_TX_TRY_HOLD(mcip, mytx, error) \ 3494 { \ 3495 (error) = 0; \ 3496 (mytx) = &(mcip)->mci_tx_pcpu[CPU->cpu_seqid & mac_tx_percpu_cnt]; \ 3497 mutex_enter(&(mytx)->pcpu_tx_lock); \ 3498 if (!((mcip)->mci_tx_flag & MCI_TX_QUIESCE)) { \ 3499 (mytx)->pcpu_tx_refcnt++; \ 3500 } else { \ 3501 (error) = -1; \ 3502 } \ 3503 mutex_exit(&(mytx)->pcpu_tx_lock); \ 3504 } 3505 3506 /* 3507 * Release the reference. If needed, signal any control operation waiting 3508 * for Tx quiescence. The wait and signal are always done using the 3509 * mci_tx_pcpu[0]'s lock 3510 */ 3511 #define MAC_TX_RELE(mcip, mytx) { \ 3512 mutex_enter(&(mytx)->pcpu_tx_lock); \ 3513 if (--(mytx)->pcpu_tx_refcnt == 0 && \ 3514 (mcip)->mci_tx_flag & MCI_TX_QUIESCE) { \ 3515 mutex_exit(&(mytx)->pcpu_tx_lock); \ 3516 mutex_enter(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock); \ 3517 cv_signal(&(mcip)->mci_tx_cv); \ 3518 mutex_exit(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock); \ 3519 } else { \ 3520 mutex_exit(&(mytx)->pcpu_tx_lock); \ 3521 } \ 3522 } 3523 3524 /* 3525 * Send function invoked by MAC clients. 3526 */ 3527 mac_tx_cookie_t 3528 mac_tx(mac_client_handle_t mch, mblk_t *mp_chain, uintptr_t hint, 3529 uint16_t flag, mblk_t **ret_mp) 3530 { 3531 mac_tx_cookie_t cookie = 0; 3532 int error; 3533 mac_tx_percpu_t *mytx; 3534 mac_soft_ring_set_t *srs; 3535 flow_entry_t *flent; 3536 boolean_t is_subflow = B_FALSE; 3537 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 3538 mac_impl_t *mip = mcip->mci_mip; 3539 mac_srs_tx_t *srs_tx; 3540 3541 /* 3542 * Check whether the active Tx threads count is bumped already. 3543 */ 3544 if (!(flag & MAC_TX_NO_HOLD)) { 3545 MAC_TX_TRY_HOLD(mcip, mytx, error); 3546 if (error != 0) { 3547 freemsgchain(mp_chain); 3548 return (0); 3549 } 3550 } 3551 3552 /* 3553 * If mac protection is enabled, only the permissible packets will be 3554 * returned by mac_protect_check(). 3555 */ 3556 if ((mcip->mci_flent-> 3557 fe_resource_props.mrp_mask & MRP_PROTECT) != 0 && 3558 (mp_chain = mac_protect_check(mch, mp_chain)) == NULL) 3559 goto done; 3560 3561 if (mcip->mci_subflow_tab != NULL && 3562 mcip->mci_subflow_tab->ft_flow_count > 0 && 3563 mac_flow_lookup(mcip->mci_subflow_tab, mp_chain, 3564 FLOW_OUTBOUND, &flent) == 0) { 3565 /* 3566 * The main assumption here is that if in the event 3567 * we get a chain, all the packets will be classified 3568 * to the same Flow/SRS. If this changes for any 3569 * reason, the following logic should change as well. 3570 * I suppose the fanout_hint also assumes this . 3571 */ 3572 ASSERT(flent != NULL); 3573 is_subflow = B_TRUE; 3574 } else { 3575 flent = mcip->mci_flent; 3576 } 3577 3578 srs = flent->fe_tx_srs; 3579 /* 3580 * This is to avoid panics with PF_PACKET that can call mac_tx() 3581 * against an interface that is not capable of sending. A rewrite 3582 * of the mac datapath is required to remove this limitation. 3583 */ 3584 if (srs == NULL) { 3585 freemsgchain(mp_chain); 3586 goto done; 3587 } 3588 3589 srs_tx = &srs->srs_tx; 3590 if (srs_tx->st_mode == SRS_TX_DEFAULT && 3591 (srs->srs_state & SRS_ENQUEUED) == 0 && 3592 mip->mi_nactiveclients == 1 && 3593 mp_chain->b_next == NULL && 3594 (DB_CKSUMFLAGS(mp_chain) & HW_LSO) == 0) { 3595 uint64_t obytes; 3596 3597 /* 3598 * Since dls always opens the underlying MAC, nclients equals 3599 * to 1 means that the only active client is dls itself acting 3600 * as a primary client of the MAC instance. Since dls will not 3601 * send tagged packets in that case, and dls is trusted to send 3602 * packets for its allowed VLAN(s), the VLAN tag insertion and 3603 * check is required only if nclients is greater than 1. 3604 */ 3605 if (mip->mi_nclients > 1) { 3606 if (MAC_VID_CHECK_NEEDED(mcip)) { 3607 int err = 0; 3608 3609 MAC_VID_CHECK(mcip, mp_chain, err); 3610 if (err != 0) { 3611 freemsg(mp_chain); 3612 mcip->mci_misc_stat.mms_txerrors++; 3613 goto done; 3614 } 3615 } 3616 if (MAC_TAG_NEEDED(mcip)) { 3617 mp_chain = mac_add_vlan_tag(mp_chain, 0, 3618 mac_client_vid(mch)); 3619 if (mp_chain == NULL) { 3620 mcip->mci_misc_stat.mms_txerrors++; 3621 goto done; 3622 } 3623 } 3624 } 3625 3626 obytes = (mp_chain->b_cont == NULL ? MBLKL(mp_chain) : 3627 msgdsize(mp_chain)); 3628 3629 mp_chain = mac_provider_tx(mip, srs_tx->st_arg2, mp_chain, 3630 mcip); 3631 3632 if (mp_chain == NULL) { 3633 cookie = 0; 3634 SRS_TX_STAT_UPDATE(srs, opackets, 1); 3635 SRS_TX_STAT_UPDATE(srs, obytes, obytes); 3636 } else { 3637 mutex_enter(&srs->srs_lock); 3638 cookie = mac_tx_srs_no_desc(srs, mp_chain, 3639 flag, ret_mp); 3640 mutex_exit(&srs->srs_lock); 3641 } 3642 } else { 3643 mblk_t *mp = mp_chain; 3644 mblk_t *new_head = NULL; 3645 mblk_t *new_tail = NULL; 3646 3647 /* 3648 * There are occasions where the packets arriving here 3649 * may request hardware offloads that are not 3650 * available from the underlying MAC provider. This 3651 * currently only happens when a packet is sent across 3652 * the MAC-loopback path of one MAC and then forwarded 3653 * (via IP) to another MAC that lacks one or more of 3654 * the hardware offloads provided by the first one. 3655 * However, in the future, we may choose to pretend 3656 * all MAC providers support all offloads, performing 3657 * emulation on Tx as needed. 3658 * 3659 * We iterate each mblk in-turn, emulating hardware 3660 * offloads as required. From this process, we create 3661 * a new chain. The new chain may be the same as the 3662 * original chain (no hardware emulation needed), a 3663 * collection of new mblks (hardware emulation 3664 * needed), or a mix. At this point, the chain is safe 3665 * for consumption by the underlying MAC provider and 3666 * is passed down to the SRS. 3667 */ 3668 while (mp != NULL) { 3669 mblk_t *next = mp->b_next; 3670 mblk_t *tail = NULL; 3671 const uint16_t needed = 3672 (DB_CKSUMFLAGS(mp) ^ mip->mi_tx_cksum_flags) & 3673 DB_CKSUMFLAGS(mp); 3674 3675 mp->b_next = NULL; 3676 3677 if ((needed & (HCK_TX_FLAGS | HW_LSO_FLAGS)) != 0) { 3678 mac_emul_t emul = 0; 3679 3680 if (needed & HCK_IPV4_HDRCKSUM) 3681 emul |= MAC_IPCKSUM_EMUL; 3682 if (needed & (HCK_PARTIALCKSUM | HCK_FULLCKSUM)) 3683 emul |= MAC_HWCKSUM_EMUL; 3684 if (needed & HW_LSO) 3685 emul = MAC_LSO_EMUL; 3686 3687 mac_hw_emul(&mp, &tail, NULL, emul); 3688 3689 if (mp == NULL) { 3690 mp = next; 3691 continue; 3692 } 3693 } 3694 3695 if (new_head == NULL) { 3696 new_head = mp; 3697 } else { 3698 new_tail->b_next = mp; 3699 } 3700 3701 new_tail = (tail == NULL) ? mp : tail; 3702 mp = next; 3703 } 3704 3705 if (new_head == NULL) { 3706 cookie = 0; 3707 goto done; 3708 } 3709 3710 cookie = srs_tx->st_func(srs, new_head, hint, flag, ret_mp); 3711 } 3712 3713 done: 3714 if (is_subflow) 3715 FLOW_REFRELE(flent); 3716 3717 if (!(flag & MAC_TX_NO_HOLD)) 3718 MAC_TX_RELE(mcip, mytx); 3719 3720 return (cookie); 3721 } 3722 3723 /* 3724 * mac_tx_is_blocked 3725 * 3726 * Given a cookie, it returns if the ring identified by the cookie is 3727 * flow-controlled or not. If NULL is passed in place of a cookie, 3728 * then it finds out if any of the underlying rings belonging to the 3729 * SRS is flow controlled or not and returns that status. 3730 */ 3731 /* ARGSUSED */ 3732 boolean_t 3733 mac_tx_is_flow_blocked(mac_client_handle_t mch, mac_tx_cookie_t cookie) 3734 { 3735 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 3736 mac_soft_ring_set_t *mac_srs; 3737 mac_soft_ring_t *sringp; 3738 boolean_t blocked = B_FALSE; 3739 mac_tx_percpu_t *mytx; 3740 int err; 3741 int i; 3742 3743 /* 3744 * Bump the reference count so that mac_srs won't be deleted. 3745 * If the client is currently quiesced and we failed to bump 3746 * the reference, return B_TRUE so that flow control stays 3747 * as enabled. 3748 * 3749 * Flow control will then be disabled once the client is no 3750 * longer quiesced. 3751 */ 3752 MAC_TX_TRY_HOLD(mcip, mytx, err); 3753 if (err != 0) 3754 return (B_TRUE); 3755 3756 if ((mac_srs = MCIP_TX_SRS(mcip)) == NULL) { 3757 MAC_TX_RELE(mcip, mytx); 3758 return (B_FALSE); 3759 } 3760 3761 mutex_enter(&mac_srs->srs_lock); 3762 /* 3763 * Only in the case of TX_FANOUT and TX_AGGR, the underlying 3764 * softring (s_ring_state) will have the HIWAT set. This is 3765 * the multiple Tx ring flow control case. For all other 3766 * case, SRS (srs_state) will store the condition. 3767 */ 3768 if (mac_srs->srs_tx.st_mode == SRS_TX_FANOUT || 3769 mac_srs->srs_tx.st_mode == SRS_TX_AGGR) { 3770 if (cookie != 0) { 3771 sringp = (mac_soft_ring_t *)cookie; 3772 mutex_enter(&sringp->s_ring_lock); 3773 if (sringp->s_ring_state & S_RING_TX_HIWAT) 3774 blocked = B_TRUE; 3775 mutex_exit(&sringp->s_ring_lock); 3776 } else { 3777 for (i = 0; i < mac_srs->srs_tx_ring_count; i++) { 3778 sringp = mac_srs->srs_tx_soft_rings[i]; 3779 mutex_enter(&sringp->s_ring_lock); 3780 if (sringp->s_ring_state & S_RING_TX_HIWAT) { 3781 blocked = B_TRUE; 3782 mutex_exit(&sringp->s_ring_lock); 3783 break; 3784 } 3785 mutex_exit(&sringp->s_ring_lock); 3786 } 3787 } 3788 } else { 3789 blocked = (mac_srs->srs_state & SRS_TX_HIWAT); 3790 } 3791 mutex_exit(&mac_srs->srs_lock); 3792 MAC_TX_RELE(mcip, mytx); 3793 return (blocked); 3794 } 3795 3796 /* 3797 * Check if the MAC client is the primary MAC client. 3798 */ 3799 boolean_t 3800 mac_is_primary_client(mac_client_impl_t *mcip) 3801 { 3802 return (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY); 3803 } 3804 3805 void 3806 mac_ioctl(mac_handle_t mh, queue_t *wq, mblk_t *bp) 3807 { 3808 mac_impl_t *mip = (mac_impl_t *)mh; 3809 int cmd = ((struct iocblk *)bp->b_rptr)->ioc_cmd; 3810 3811 if ((cmd == ND_GET && (mip->mi_callbacks->mc_callbacks & MC_GETPROP)) || 3812 (cmd == ND_SET && (mip->mi_callbacks->mc_callbacks & MC_SETPROP))) { 3813 /* 3814 * If ndd props were registered, call them. 3815 * Note that ndd ioctls are Obsolete 3816 */ 3817 mac_ndd_ioctl(mip, wq, bp); 3818 return; 3819 } 3820 3821 /* 3822 * Call the driver to handle the ioctl. The driver may not support 3823 * any ioctls, in which case we reply with a NAK on its behalf. 3824 */ 3825 if (mip->mi_callbacks->mc_callbacks & MC_IOCTL) 3826 mip->mi_ioctl(mip->mi_driver, wq, bp); 3827 else 3828 miocnak(wq, bp, 0, EINVAL); 3829 } 3830 3831 /* 3832 * Return the link state of the specified MAC instance. 3833 */ 3834 link_state_t 3835 mac_link_get(mac_handle_t mh) 3836 { 3837 return (((mac_impl_t *)mh)->mi_linkstate); 3838 } 3839 3840 /* 3841 * Add a mac client specified notification callback. Please see the comments 3842 * above mac_callback_add() for general information about mac callback 3843 * addition/deletion in the presence of mac callback list walkers 3844 */ 3845 mac_notify_handle_t 3846 mac_notify_add(mac_handle_t mh, mac_notify_t notify_fn, void *arg) 3847 { 3848 mac_impl_t *mip = (mac_impl_t *)mh; 3849 mac_notify_cb_t *mncb; 3850 mac_cb_info_t *mcbi; 3851 3852 /* 3853 * Allocate a notify callback structure, fill in the details and 3854 * use the mac callback list manipulation functions to chain into 3855 * the list of callbacks. 3856 */ 3857 mncb = kmem_zalloc(sizeof (mac_notify_cb_t), KM_SLEEP); 3858 mncb->mncb_fn = notify_fn; 3859 mncb->mncb_arg = arg; 3860 mncb->mncb_mip = mip; 3861 mncb->mncb_link.mcb_objp = mncb; 3862 mncb->mncb_link.mcb_objsize = sizeof (mac_notify_cb_t); 3863 mncb->mncb_link.mcb_flags = MCB_NOTIFY_CB_T; 3864 3865 mcbi = &mip->mi_notify_cb_info; 3866 3867 i_mac_perim_enter(mip); 3868 mutex_enter(mcbi->mcbi_lockp); 3869 3870 mac_callback_add(&mip->mi_notify_cb_info, &mip->mi_notify_cb_list, 3871 &mncb->mncb_link); 3872 3873 mutex_exit(mcbi->mcbi_lockp); 3874 i_mac_perim_exit(mip); 3875 return ((mac_notify_handle_t)mncb); 3876 } 3877 3878 void 3879 mac_notify_remove_wait(mac_handle_t mh) 3880 { 3881 mac_impl_t *mip = (mac_impl_t *)mh; 3882 mac_cb_info_t *mcbi = &mip->mi_notify_cb_info; 3883 3884 mutex_enter(mcbi->mcbi_lockp); 3885 mac_callback_remove_wait(&mip->mi_notify_cb_info); 3886 mutex_exit(mcbi->mcbi_lockp); 3887 } 3888 3889 /* 3890 * Remove a mac client specified notification callback 3891 */ 3892 int 3893 mac_notify_remove(mac_notify_handle_t mnh, boolean_t wait) 3894 { 3895 mac_notify_cb_t *mncb = (mac_notify_cb_t *)mnh; 3896 mac_impl_t *mip = mncb->mncb_mip; 3897 mac_cb_info_t *mcbi; 3898 int err = 0; 3899 3900 mcbi = &mip->mi_notify_cb_info; 3901 3902 i_mac_perim_enter(mip); 3903 mutex_enter(mcbi->mcbi_lockp); 3904 3905 ASSERT(mncb->mncb_link.mcb_objp == mncb); 3906 /* 3907 * If there aren't any list walkers, the remove would succeed 3908 * inline, else we wait for the deferred remove to complete 3909 */ 3910 if (mac_callback_remove(&mip->mi_notify_cb_info, 3911 &mip->mi_notify_cb_list, &mncb->mncb_link)) { 3912 kmem_free(mncb, sizeof (mac_notify_cb_t)); 3913 } else { 3914 err = EBUSY; 3915 } 3916 3917 mutex_exit(mcbi->mcbi_lockp); 3918 i_mac_perim_exit(mip); 3919 3920 /* 3921 * If we failed to remove the notification callback and "wait" is set 3922 * to be B_TRUE, wait for the callback to finish after we exit the 3923 * mac perimeter. 3924 */ 3925 if (err != 0 && wait) { 3926 mac_notify_remove_wait((mac_handle_t)mip); 3927 return (0); 3928 } 3929 3930 return (err); 3931 } 3932 3933 /* 3934 * Associate resource management callbacks with the specified MAC 3935 * clients. 3936 */ 3937 3938 void 3939 mac_resource_set_common(mac_client_handle_t mch, mac_resource_add_t add, 3940 mac_resource_remove_t remove, mac_resource_quiesce_t quiesce, 3941 mac_resource_restart_t restart, mac_resource_bind_t bind, 3942 void *arg) 3943 { 3944 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 3945 3946 mcip->mci_resource_add = add; 3947 mcip->mci_resource_remove = remove; 3948 mcip->mci_resource_quiesce = quiesce; 3949 mcip->mci_resource_restart = restart; 3950 mcip->mci_resource_bind = bind; 3951 mcip->mci_resource_arg = arg; 3952 } 3953 3954 void 3955 mac_resource_set(mac_client_handle_t mch, mac_resource_add_t add, void *arg) 3956 { 3957 /* update the 'resource_add' callback */ 3958 mac_resource_set_common(mch, add, NULL, NULL, NULL, NULL, arg); 3959 } 3960 3961 /* 3962 * Sets up the client resources and enable the polling interface over all the 3963 * SRS's and the soft rings of the client 3964 */ 3965 void 3966 mac_client_poll_enable(mac_client_handle_t mch) 3967 { 3968 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 3969 mac_soft_ring_set_t *mac_srs; 3970 flow_entry_t *flent; 3971 int i; 3972 3973 flent = mcip->mci_flent; 3974 ASSERT(flent != NULL); 3975 3976 mcip->mci_state_flags |= MCIS_CLIENT_POLL_CAPABLE; 3977 for (i = 0; i < flent->fe_rx_srs_cnt; i++) { 3978 mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i]; 3979 ASSERT(mac_srs->srs_mcip == mcip); 3980 mac_srs_client_poll_enable(mcip, mac_srs); 3981 } 3982 } 3983 3984 /* 3985 * Tears down the client resources and disable the polling interface over all 3986 * the SRS's and the soft rings of the client 3987 */ 3988 void 3989 mac_client_poll_disable(mac_client_handle_t mch) 3990 { 3991 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 3992 mac_soft_ring_set_t *mac_srs; 3993 flow_entry_t *flent; 3994 int i; 3995 3996 flent = mcip->mci_flent; 3997 ASSERT(flent != NULL); 3998 3999 mcip->mci_state_flags &= ~MCIS_CLIENT_POLL_CAPABLE; 4000 for (i = 0; i < flent->fe_rx_srs_cnt; i++) { 4001 mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i]; 4002 ASSERT(mac_srs->srs_mcip == mcip); 4003 mac_srs_client_poll_disable(mcip, mac_srs); 4004 } 4005 } 4006 4007 /* 4008 * Associate the CPUs specified by the given property with a MAC client. 4009 */ 4010 int 4011 mac_cpu_set(mac_client_handle_t mch, mac_resource_props_t *mrp) 4012 { 4013 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 4014 mac_impl_t *mip = mcip->mci_mip; 4015 int err = 0; 4016 4017 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 4018 4019 if ((err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ? 4020 mcip->mci_upper_mip : mip, mrp)) != 0) { 4021 return (err); 4022 } 4023 if (MCIP_DATAPATH_SETUP(mcip)) 4024 mac_flow_modify(mip->mi_flow_tab, mcip->mci_flent, mrp); 4025 4026 mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE); 4027 return (0); 4028 } 4029 4030 /* 4031 * Apply the specified properties to the specified MAC client. 4032 */ 4033 int 4034 mac_client_set_resources(mac_client_handle_t mch, mac_resource_props_t *mrp) 4035 { 4036 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 4037 mac_impl_t *mip = mcip->mci_mip; 4038 int err = 0; 4039 4040 i_mac_perim_enter(mip); 4041 4042 if ((mrp->mrp_mask & MRP_MAXBW) || (mrp->mrp_mask & MRP_PRIORITY)) { 4043 err = mac_resource_ctl_set(mch, mrp); 4044 if (err != 0) 4045 goto done; 4046 } 4047 4048 if (mrp->mrp_mask & (MRP_CPUS|MRP_POOL)) { 4049 err = mac_cpu_set(mch, mrp); 4050 if (err != 0) 4051 goto done; 4052 } 4053 4054 if (mrp->mrp_mask & MRP_PROTECT) { 4055 err = mac_protect_set(mch, mrp); 4056 if (err != 0) 4057 goto done; 4058 } 4059 4060 if ((mrp->mrp_mask & MRP_RX_RINGS) || (mrp->mrp_mask & MRP_TX_RINGS)) 4061 err = mac_resource_ctl_set(mch, mrp); 4062 4063 done: 4064 i_mac_perim_exit(mip); 4065 return (err); 4066 } 4067 4068 /* 4069 * Return the properties currently associated with the specified MAC client. 4070 */ 4071 void 4072 mac_client_get_resources(mac_client_handle_t mch, mac_resource_props_t *mrp) 4073 { 4074 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 4075 mac_resource_props_t *mcip_mrp = MCIP_RESOURCE_PROPS(mcip); 4076 4077 bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t)); 4078 } 4079 4080 /* 4081 * Return the effective properties currently associated with the specified 4082 * MAC client. 4083 */ 4084 void 4085 mac_client_get_effective_resources(mac_client_handle_t mch, 4086 mac_resource_props_t *mrp) 4087 { 4088 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 4089 mac_resource_props_t *mcip_mrp = MCIP_EFFECTIVE_PROPS(mcip); 4090 4091 bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t)); 4092 } 4093 4094 /* 4095 * Pass a copy of the specified packet to the promiscuous callbacks 4096 * of the specified MAC. 4097 * 4098 * If sender is NULL, the function is being invoked for a packet chain 4099 * received from the wire. If sender is non-NULL, it points to 4100 * the MAC client from which the packet is being sent. 4101 * 4102 * The packets are distributed to the promiscuous callbacks as follows: 4103 * 4104 * - all packets are sent to the MAC_CLIENT_PROMISC_ALL callbacks 4105 * - all broadcast and multicast packets are sent to the 4106 * MAC_CLIENT_PROMISC_FILTER and MAC_CLIENT_PROMISC_MULTI. 4107 * 4108 * The unicast packets of MAC_CLIENT_PROMISC_FILTER callbacks are dispatched 4109 * after classification by mac_rx_deliver(). 4110 */ 4111 static void 4112 mac_promisc_dispatch_one(mac_promisc_impl_t *mpip, mblk_t *mp, 4113 boolean_t loopback, boolean_t local) 4114 { 4115 mblk_t *mp_next; 4116 4117 if (!mpip->mpi_no_copy || mpip->mpi_strip_vlan_tag) { 4118 mblk_t *mp_copy; 4119 4120 mp_copy = copymsg(mp); 4121 if (mp_copy == NULL) 4122 return; 4123 4124 if (mpip->mpi_strip_vlan_tag) { 4125 mp_copy = mac_strip_vlan_tag_chain(mp_copy); 4126 if (mp_copy == NULL) 4127 return; 4128 } 4129 4130 /* 4131 * There is code upstack that can't deal with message 4132 * chains. 4133 */ 4134 for (mblk_t *tmp = mp_copy; tmp != NULL; tmp = mp_next) { 4135 mp_next = tmp->b_next; 4136 tmp->b_next = NULL; 4137 mpip->mpi_fn(mpip->mpi_arg, NULL, tmp, loopback); 4138 } 4139 4140 return; 4141 } 4142 4143 mp_next = mp->b_next; 4144 mp->b_next = NULL; 4145 mpip->mpi_fn(mpip->mpi_arg, NULL, mp, loopback); 4146 mp->b_next = mp_next; 4147 } 4148 4149 /* 4150 * Return the VID of a packet. Zero if the packet is not tagged. 4151 */ 4152 static uint16_t 4153 mac_ether_vid(mblk_t *mp) 4154 { 4155 struct ether_header *eth = (struct ether_header *)mp->b_rptr; 4156 4157 if (ntohs(eth->ether_type) == ETHERTYPE_VLAN) { 4158 struct ether_vlan_header *t_evhp = 4159 (struct ether_vlan_header *)mp->b_rptr; 4160 return (VLAN_ID(ntohs(t_evhp->ether_tci))); 4161 } 4162 4163 return (0); 4164 } 4165 4166 /* 4167 * Return whether the specified packet contains a multicast or broadcast 4168 * destination MAC address. 4169 */ 4170 static boolean_t 4171 mac_is_mcast(mac_impl_t *mip, mblk_t *mp) 4172 { 4173 mac_header_info_t hdr_info; 4174 4175 if (mac_header_info((mac_handle_t)mip, mp, &hdr_info) != 0) 4176 return (B_FALSE); 4177 return ((hdr_info.mhi_dsttype == MAC_ADDRTYPE_BROADCAST) || 4178 (hdr_info.mhi_dsttype == MAC_ADDRTYPE_MULTICAST)); 4179 } 4180 4181 /* 4182 * Send a copy of an mblk chain to the MAC clients of the specified MAC. 4183 * "sender" points to the sender MAC client for outbound packets, and 4184 * is set to NULL for inbound packets. 4185 */ 4186 void 4187 mac_promisc_dispatch(mac_impl_t *mip, mblk_t *mp_chain, 4188 mac_client_impl_t *sender, boolean_t local) 4189 { 4190 mac_promisc_impl_t *mpip; 4191 mac_cb_t *mcb; 4192 mblk_t *mp; 4193 boolean_t is_mcast, is_sender; 4194 4195 MAC_PROMISC_WALKER_INC(mip); 4196 for (mp = mp_chain; mp != NULL; mp = mp->b_next) { 4197 is_mcast = mac_is_mcast(mip, mp); 4198 /* send packet to interested callbacks */ 4199 for (mcb = mip->mi_promisc_list; mcb != NULL; 4200 mcb = mcb->mcb_nextp) { 4201 mpip = (mac_promisc_impl_t *)mcb->mcb_objp; 4202 is_sender = (mpip->mpi_mcip == sender); 4203 4204 if (is_sender && mpip->mpi_no_tx_loop) 4205 /* 4206 * The sender doesn't want to receive 4207 * copies of the packets it sends. 4208 */ 4209 continue; 4210 4211 /* this client doesn't need any packets (bridge) */ 4212 if (mpip->mpi_fn == NULL) 4213 continue; 4214 4215 /* 4216 * For an ethernet MAC, don't displatch a multicast 4217 * packet to a non-PROMISC_ALL callbacks unless the VID 4218 * of the packet matches the VID of the client. 4219 */ 4220 if (is_mcast && 4221 mpip->mpi_type != MAC_CLIENT_PROMISC_ALL && 4222 !mac_client_check_flow_vid(mpip->mpi_mcip, 4223 mac_ether_vid(mp))) 4224 continue; 4225 4226 if (is_sender || 4227 mpip->mpi_type == MAC_CLIENT_PROMISC_ALL || 4228 is_mcast) { 4229 mac_promisc_dispatch_one(mpip, mp, is_sender, 4230 local); 4231 } 4232 } 4233 } 4234 MAC_PROMISC_WALKER_DCR(mip); 4235 } 4236 4237 void 4238 mac_promisc_client_dispatch(mac_client_impl_t *mcip, mblk_t *mp_chain) 4239 { 4240 mac_impl_t *mip = mcip->mci_mip; 4241 mac_promisc_impl_t *mpip; 4242 boolean_t is_mcast; 4243 mblk_t *mp; 4244 mac_cb_t *mcb; 4245 4246 /* 4247 * The unicast packets for the MAC client still 4248 * need to be delivered to the MAC_CLIENT_PROMISC_FILTERED 4249 * promiscuous callbacks. The broadcast and multicast 4250 * packets were delivered from mac_rx(). 4251 */ 4252 MAC_PROMISC_WALKER_INC(mip); 4253 for (mp = mp_chain; mp != NULL; mp = mp->b_next) { 4254 is_mcast = mac_is_mcast(mip, mp); 4255 for (mcb = mcip->mci_promisc_list; mcb != NULL; 4256 mcb = mcb->mcb_nextp) { 4257 mpip = (mac_promisc_impl_t *)mcb->mcb_objp; 4258 if (mpip->mpi_type == MAC_CLIENT_PROMISC_FILTERED && 4259 !is_mcast) { 4260 mac_promisc_dispatch_one(mpip, mp, B_FALSE, 4261 B_FALSE); 4262 } 4263 } 4264 } 4265 MAC_PROMISC_WALKER_DCR(mip); 4266 } 4267 4268 /* 4269 * Return the margin value currently assigned to the specified MAC instance. 4270 */ 4271 void 4272 mac_margin_get(mac_handle_t mh, uint32_t *marginp) 4273 { 4274 mac_impl_t *mip = (mac_impl_t *)mh; 4275 4276 rw_enter(&(mip->mi_rw_lock), RW_READER); 4277 *marginp = mip->mi_margin; 4278 rw_exit(&(mip->mi_rw_lock)); 4279 } 4280 4281 /* 4282 * mac_info_get() is used for retrieving the mac_info when a DL_INFO_REQ is 4283 * issued before a DL_ATTACH_REQ. we walk the i_mac_impl_hash table and find 4284 * the first mac_impl_t with a matching driver name; then we copy its mac_info_t 4285 * to the caller. we do all this with i_mac_impl_lock held so the mac_impl_t 4286 * cannot disappear while we are accessing it. 4287 */ 4288 typedef struct i_mac_info_state_s { 4289 const char *mi_name; 4290 mac_info_t *mi_infop; 4291 } i_mac_info_state_t; 4292 4293 /*ARGSUSED*/ 4294 static uint_t 4295 i_mac_info_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg) 4296 { 4297 i_mac_info_state_t *statep = arg; 4298 mac_impl_t *mip = (mac_impl_t *)val; 4299 4300 if (mip->mi_state_flags & MIS_DISABLED) 4301 return (MH_WALK_CONTINUE); 4302 4303 if (strcmp(statep->mi_name, 4304 ddi_driver_name(mip->mi_dip)) != 0) 4305 return (MH_WALK_CONTINUE); 4306 4307 statep->mi_infop = &mip->mi_info; 4308 return (MH_WALK_TERMINATE); 4309 } 4310 4311 boolean_t 4312 mac_info_get(const char *name, mac_info_t *minfop) 4313 { 4314 i_mac_info_state_t state; 4315 4316 rw_enter(&i_mac_impl_lock, RW_READER); 4317 state.mi_name = name; 4318 state.mi_infop = NULL; 4319 mod_hash_walk(i_mac_impl_hash, i_mac_info_walker, &state); 4320 if (state.mi_infop == NULL) { 4321 rw_exit(&i_mac_impl_lock); 4322 return (B_FALSE); 4323 } 4324 *minfop = *state.mi_infop; 4325 rw_exit(&i_mac_impl_lock); 4326 return (B_TRUE); 4327 } 4328 4329 /* 4330 * To get the capabilities that MAC layer cares about, such as rings, factory 4331 * mac address, vnic or not, it should directly invoke this function. If the 4332 * link is part of a bridge, then the only "capability" it has is the inability 4333 * to do zero copy. 4334 */ 4335 boolean_t 4336 i_mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data) 4337 { 4338 mac_impl_t *mip = (mac_impl_t *)mh; 4339 4340 if (mip->mi_bridge_link != NULL) { 4341 return (cap == MAC_CAPAB_NO_ZCOPY); 4342 } else if (mip->mi_callbacks->mc_callbacks & MC_GETCAPAB) { 4343 return (mip->mi_getcapab(mip->mi_driver, cap, cap_data)); 4344 } else { 4345 return (B_FALSE); 4346 } 4347 } 4348 4349 /* 4350 * Capability query function. If number of active mac clients is greater than 4351 * 1, only limited capabilities can be advertised to the caller no matter the 4352 * driver has certain capability or not. Else, we query the driver to get the 4353 * capability. 4354 */ 4355 boolean_t 4356 mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data) 4357 { 4358 mac_impl_t *mip = (mac_impl_t *)mh; 4359 4360 /* 4361 * Some capabilities are restricted when there are more than one active 4362 * clients on the MAC resource. The ones noted below are safe, 4363 * independent of that count. 4364 */ 4365 if (mip->mi_nactiveclients > 1) { 4366 switch (cap) { 4367 case MAC_CAPAB_NO_ZCOPY: 4368 return (B_TRUE); 4369 case MAC_CAPAB_LEGACY: 4370 case MAC_CAPAB_HCKSUM: 4371 case MAC_CAPAB_LSO: 4372 case MAC_CAPAB_NO_NATIVEVLAN: 4373 break; 4374 default: 4375 return (B_FALSE); 4376 } 4377 } 4378 4379 /* else get capab from driver */ 4380 return (i_mac_capab_get(mh, cap, cap_data)); 4381 } 4382 4383 boolean_t 4384 mac_sap_verify(mac_handle_t mh, uint32_t sap, uint32_t *bind_sap) 4385 { 4386 mac_impl_t *mip = (mac_impl_t *)mh; 4387 4388 return (mip->mi_type->mt_ops.mtops_sap_verify(sap, bind_sap, 4389 mip->mi_pdata)); 4390 } 4391 4392 mblk_t * 4393 mac_header(mac_handle_t mh, const uint8_t *daddr, uint32_t sap, mblk_t *payload, 4394 size_t extra_len) 4395 { 4396 mac_impl_t *mip = (mac_impl_t *)mh; 4397 const uint8_t *hdr_daddr; 4398 4399 /* 4400 * If the MAC is point-to-point with a fixed destination address, then 4401 * we must always use that destination in the MAC header. 4402 */ 4403 hdr_daddr = (mip->mi_dstaddr_set ? mip->mi_dstaddr : daddr); 4404 return (mip->mi_type->mt_ops.mtops_header(mip->mi_addr, hdr_daddr, sap, 4405 mip->mi_pdata, payload, extra_len)); 4406 } 4407 4408 int 4409 mac_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip) 4410 { 4411 mac_impl_t *mip = (mac_impl_t *)mh; 4412 4413 return (mip->mi_type->mt_ops.mtops_header_info(mp, mip->mi_pdata, 4414 mhip)); 4415 } 4416 4417 int 4418 mac_vlan_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip) 4419 { 4420 mac_impl_t *mip = (mac_impl_t *)mh; 4421 boolean_t is_ethernet = (mip->mi_info.mi_media == DL_ETHER); 4422 int err = 0; 4423 4424 /* 4425 * Packets should always be at least 16 bit aligned. 4426 */ 4427 ASSERT(IS_P2ALIGNED(mp->b_rptr, sizeof (uint16_t))); 4428 4429 if ((err = mac_header_info(mh, mp, mhip)) != 0) 4430 return (err); 4431 4432 /* 4433 * If this is a VLAN-tagged Ethernet packet, then the SAP in the 4434 * mac_header_info_t as returned by mac_header_info() is 4435 * ETHERTYPE_VLAN. We need to grab the ethertype from the VLAN header. 4436 */ 4437 if (is_ethernet && (mhip->mhi_bindsap == ETHERTYPE_VLAN)) { 4438 struct ether_vlan_header *evhp; 4439 uint16_t sap; 4440 mblk_t *tmp = NULL; 4441 size_t size; 4442 4443 size = sizeof (struct ether_vlan_header); 4444 if (MBLKL(mp) < size) { 4445 /* 4446 * Pullup the message in order to get the MAC header 4447 * infomation. Note that this is a read-only function, 4448 * we keep the input packet intact. 4449 */ 4450 if ((tmp = msgpullup(mp, size)) == NULL) 4451 return (EINVAL); 4452 4453 mp = tmp; 4454 } 4455 evhp = (struct ether_vlan_header *)mp->b_rptr; 4456 sap = ntohs(evhp->ether_type); 4457 (void) mac_sap_verify(mh, sap, &mhip->mhi_bindsap); 4458 mhip->mhi_hdrsize = sizeof (struct ether_vlan_header); 4459 mhip->mhi_tci = ntohs(evhp->ether_tci); 4460 mhip->mhi_istagged = B_TRUE; 4461 freemsg(tmp); 4462 4463 if (VLAN_CFI(mhip->mhi_tci) != ETHER_CFI) 4464 return (EINVAL); 4465 } else { 4466 mhip->mhi_istagged = B_FALSE; 4467 mhip->mhi_tci = 0; 4468 } 4469 4470 return (0); 4471 } 4472 4473 mblk_t * 4474 mac_header_cook(mac_handle_t mh, mblk_t *mp) 4475 { 4476 mac_impl_t *mip = (mac_impl_t *)mh; 4477 4478 if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_COOK) { 4479 if (DB_REF(mp) > 1) { 4480 mblk_t *newmp = copymsg(mp); 4481 if (newmp == NULL) 4482 return (NULL); 4483 freemsg(mp); 4484 mp = newmp; 4485 } 4486 return (mip->mi_type->mt_ops.mtops_header_cook(mp, 4487 mip->mi_pdata)); 4488 } 4489 return (mp); 4490 } 4491 4492 mblk_t * 4493 mac_header_uncook(mac_handle_t mh, mblk_t *mp) 4494 { 4495 mac_impl_t *mip = (mac_impl_t *)mh; 4496 4497 if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_UNCOOK) { 4498 if (DB_REF(mp) > 1) { 4499 mblk_t *newmp = copymsg(mp); 4500 if (newmp == NULL) 4501 return (NULL); 4502 freemsg(mp); 4503 mp = newmp; 4504 } 4505 return (mip->mi_type->mt_ops.mtops_header_uncook(mp, 4506 mip->mi_pdata)); 4507 } 4508 return (mp); 4509 } 4510 4511 uint_t 4512 mac_addr_len(mac_handle_t mh) 4513 { 4514 mac_impl_t *mip = (mac_impl_t *)mh; 4515 4516 return (mip->mi_type->mt_addr_length); 4517 } 4518 4519 /* True if a MAC is a VNIC */ 4520 boolean_t 4521 mac_is_vnic(mac_handle_t mh) 4522 { 4523 return ((((mac_impl_t *)mh)->mi_state_flags & MIS_IS_VNIC) != 0); 4524 } 4525 4526 boolean_t 4527 mac_is_overlay(mac_handle_t mh) 4528 { 4529 return ((((mac_impl_t *)mh)->mi_state_flags & MIS_IS_OVERLAY) != 0); 4530 } 4531 4532 mac_handle_t 4533 mac_get_lower_mac_handle(mac_handle_t mh) 4534 { 4535 mac_impl_t *mip = (mac_impl_t *)mh; 4536 4537 ASSERT(mac_is_vnic(mh)); 4538 return (((vnic_t *)mip->mi_driver)->vn_lower_mh); 4539 } 4540 4541 boolean_t 4542 mac_is_vnic_primary(mac_handle_t mh) 4543 { 4544 mac_impl_t *mip = (mac_impl_t *)mh; 4545 4546 ASSERT(mac_is_vnic(mh)); 4547 return (((vnic_t *)mip->mi_driver)->vn_addr_type == 4548 VNIC_MAC_ADDR_TYPE_PRIMARY); 4549 } 4550 4551 void 4552 mac_update_resources(mac_resource_props_t *nmrp, mac_resource_props_t *cmrp, 4553 boolean_t is_user_flow) 4554 { 4555 if (nmrp != NULL && cmrp != NULL) { 4556 if (nmrp->mrp_mask & MRP_PRIORITY) { 4557 if (nmrp->mrp_priority == MPL_RESET) { 4558 cmrp->mrp_mask &= ~MRP_PRIORITY; 4559 if (is_user_flow) { 4560 cmrp->mrp_priority = 4561 MPL_SUBFLOW_DEFAULT; 4562 } else { 4563 cmrp->mrp_priority = MPL_LINK_DEFAULT; 4564 } 4565 } else { 4566 cmrp->mrp_mask |= MRP_PRIORITY; 4567 cmrp->mrp_priority = nmrp->mrp_priority; 4568 } 4569 } 4570 if (nmrp->mrp_mask & MRP_MAXBW) { 4571 if (nmrp->mrp_maxbw == MRP_MAXBW_RESETVAL) { 4572 cmrp->mrp_mask &= ~MRP_MAXBW; 4573 cmrp->mrp_maxbw = 0; 4574 } else { 4575 cmrp->mrp_mask |= MRP_MAXBW; 4576 cmrp->mrp_maxbw = nmrp->mrp_maxbw; 4577 } 4578 } 4579 if (nmrp->mrp_mask & MRP_CPUS) 4580 MAC_COPY_CPUS(nmrp, cmrp); 4581 4582 if (nmrp->mrp_mask & MRP_POOL) { 4583 if (strlen(nmrp->mrp_pool) == 0) { 4584 cmrp->mrp_mask &= ~MRP_POOL; 4585 bzero(cmrp->mrp_pool, sizeof (cmrp->mrp_pool)); 4586 } else { 4587 cmrp->mrp_mask |= MRP_POOL; 4588 (void) strncpy(cmrp->mrp_pool, nmrp->mrp_pool, 4589 sizeof (cmrp->mrp_pool)); 4590 } 4591 4592 } 4593 4594 if (nmrp->mrp_mask & MRP_PROTECT) 4595 mac_protect_update(nmrp, cmrp); 4596 4597 /* 4598 * Update the rings specified. 4599 */ 4600 if (nmrp->mrp_mask & MRP_RX_RINGS) { 4601 if (nmrp->mrp_mask & MRP_RINGS_RESET) { 4602 cmrp->mrp_mask &= ~MRP_RX_RINGS; 4603 if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC) 4604 cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC; 4605 cmrp->mrp_nrxrings = 0; 4606 } else { 4607 cmrp->mrp_mask |= MRP_RX_RINGS; 4608 cmrp->mrp_nrxrings = nmrp->mrp_nrxrings; 4609 } 4610 } 4611 if (nmrp->mrp_mask & MRP_TX_RINGS) { 4612 if (nmrp->mrp_mask & MRP_RINGS_RESET) { 4613 cmrp->mrp_mask &= ~MRP_TX_RINGS; 4614 if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC) 4615 cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC; 4616 cmrp->mrp_ntxrings = 0; 4617 } else { 4618 cmrp->mrp_mask |= MRP_TX_RINGS; 4619 cmrp->mrp_ntxrings = nmrp->mrp_ntxrings; 4620 } 4621 } 4622 if (nmrp->mrp_mask & MRP_RXRINGS_UNSPEC) 4623 cmrp->mrp_mask |= MRP_RXRINGS_UNSPEC; 4624 else if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC) 4625 cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC; 4626 4627 if (nmrp->mrp_mask & MRP_TXRINGS_UNSPEC) 4628 cmrp->mrp_mask |= MRP_TXRINGS_UNSPEC; 4629 else if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC) 4630 cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC; 4631 } 4632 } 4633 4634 /* 4635 * i_mac_set_resources: 4636 * 4637 * This routine associates properties with the primary MAC client of 4638 * the specified MAC instance. 4639 * - Cache the properties in mac_impl_t 4640 * - Apply the properties to the primary MAC client if exists 4641 */ 4642 int 4643 i_mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp) 4644 { 4645 mac_impl_t *mip = (mac_impl_t *)mh; 4646 mac_client_impl_t *mcip; 4647 int err = 0; 4648 uint32_t resmask, newresmask; 4649 mac_resource_props_t *tmrp, *umrp; 4650 4651 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 4652 4653 err = mac_validate_props(mip, mrp); 4654 if (err != 0) 4655 return (err); 4656 4657 umrp = kmem_zalloc(sizeof (*umrp), KM_SLEEP); 4658 bcopy(&mip->mi_resource_props, umrp, sizeof (*umrp)); 4659 resmask = umrp->mrp_mask; 4660 mac_update_resources(mrp, umrp, B_FALSE); 4661 newresmask = umrp->mrp_mask; 4662 4663 if (resmask == 0 && newresmask != 0) { 4664 /* 4665 * Bandwidth, priority, cpu or pool link properties configured, 4666 * must disable fastpath. 4667 */ 4668 if ((err = mac_fastpath_disable((mac_handle_t)mip)) != 0) { 4669 kmem_free(umrp, sizeof (*umrp)); 4670 return (err); 4671 } 4672 } 4673 4674 /* 4675 * Since bind_cpu may be modified by mac_client_set_resources() 4676 * we use a copy of bind_cpu and finally cache bind_cpu in mip. 4677 * This allows us to cache only user edits in mip. 4678 */ 4679 tmrp = kmem_zalloc(sizeof (*tmrp), KM_SLEEP); 4680 bcopy(mrp, tmrp, sizeof (*tmrp)); 4681 mcip = mac_primary_client_handle(mip); 4682 if (mcip != NULL && (mcip->mci_state_flags & MCIS_IS_AGGR_PORT) == 0) { 4683 err = mac_client_set_resources((mac_client_handle_t)mcip, tmrp); 4684 } else if ((mrp->mrp_mask & MRP_RX_RINGS || 4685 mrp->mrp_mask & MRP_TX_RINGS)) { 4686 mac_client_impl_t *vmcip; 4687 4688 /* 4689 * If the primary is not up, we need to check if there 4690 * are any VLANs on this primary. If there are then 4691 * we need to set this property on the VLANs since 4692 * VLANs follow the primary they are based on. Just 4693 * look for the first VLAN and change its properties, 4694 * all the other VLANs should be in the same group. 4695 */ 4696 for (vmcip = mip->mi_clients_list; vmcip != NULL; 4697 vmcip = vmcip->mci_client_next) { 4698 if ((vmcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) && 4699 mac_client_vid((mac_client_handle_t)vmcip) != 4700 VLAN_ID_NONE) { 4701 break; 4702 } 4703 } 4704 if (vmcip != NULL) { 4705 mac_resource_props_t *omrp; 4706 mac_resource_props_t *vmrp; 4707 4708 omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP); 4709 bcopy(MCIP_RESOURCE_PROPS(vmcip), omrp, sizeof (*omrp)); 4710 /* 4711 * We dont' call mac_update_resources since we 4712 * want to take only the ring properties and 4713 * not all the properties that may have changed. 4714 */ 4715 vmrp = MCIP_RESOURCE_PROPS(vmcip); 4716 if (mrp->mrp_mask & MRP_RX_RINGS) { 4717 if (mrp->mrp_mask & MRP_RINGS_RESET) { 4718 vmrp->mrp_mask &= ~MRP_RX_RINGS; 4719 if (vmrp->mrp_mask & 4720 MRP_RXRINGS_UNSPEC) { 4721 vmrp->mrp_mask &= 4722 ~MRP_RXRINGS_UNSPEC; 4723 } 4724 vmrp->mrp_nrxrings = 0; 4725 } else { 4726 vmrp->mrp_mask |= MRP_RX_RINGS; 4727 vmrp->mrp_nrxrings = mrp->mrp_nrxrings; 4728 } 4729 } 4730 if (mrp->mrp_mask & MRP_TX_RINGS) { 4731 if (mrp->mrp_mask & MRP_RINGS_RESET) { 4732 vmrp->mrp_mask &= ~MRP_TX_RINGS; 4733 if (vmrp->mrp_mask & 4734 MRP_TXRINGS_UNSPEC) { 4735 vmrp->mrp_mask &= 4736 ~MRP_TXRINGS_UNSPEC; 4737 } 4738 vmrp->mrp_ntxrings = 0; 4739 } else { 4740 vmrp->mrp_mask |= MRP_TX_RINGS; 4741 vmrp->mrp_ntxrings = mrp->mrp_ntxrings; 4742 } 4743 } 4744 if (mrp->mrp_mask & MRP_RXRINGS_UNSPEC) 4745 vmrp->mrp_mask |= MRP_RXRINGS_UNSPEC; 4746 4747 if (mrp->mrp_mask & MRP_TXRINGS_UNSPEC) 4748 vmrp->mrp_mask |= MRP_TXRINGS_UNSPEC; 4749 4750 if ((err = mac_client_set_rings_prop(vmcip, mrp, 4751 omrp)) != 0) { 4752 bcopy(omrp, MCIP_RESOURCE_PROPS(vmcip), 4753 sizeof (*omrp)); 4754 } else { 4755 mac_set_prim_vlan_rings(mip, vmrp); 4756 } 4757 kmem_free(omrp, sizeof (*omrp)); 4758 } 4759 } 4760 4761 /* Only update the values if mac_client_set_resources succeeded */ 4762 if (err == 0) { 4763 bcopy(umrp, &mip->mi_resource_props, sizeof (*umrp)); 4764 /* 4765 * If bandwidth, priority or cpu link properties cleared, 4766 * renable fastpath. 4767 */ 4768 if (resmask != 0 && newresmask == 0) 4769 mac_fastpath_enable((mac_handle_t)mip); 4770 } else if (resmask == 0 && newresmask != 0) { 4771 mac_fastpath_enable((mac_handle_t)mip); 4772 } 4773 kmem_free(tmrp, sizeof (*tmrp)); 4774 kmem_free(umrp, sizeof (*umrp)); 4775 return (err); 4776 } 4777 4778 int 4779 mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp) 4780 { 4781 int err; 4782 4783 i_mac_perim_enter((mac_impl_t *)mh); 4784 err = i_mac_set_resources(mh, mrp); 4785 i_mac_perim_exit((mac_impl_t *)mh); 4786 return (err); 4787 } 4788 4789 /* 4790 * Get the properties cached for the specified MAC instance. 4791 */ 4792 void 4793 mac_get_resources(mac_handle_t mh, mac_resource_props_t *mrp) 4794 { 4795 mac_impl_t *mip = (mac_impl_t *)mh; 4796 mac_client_impl_t *mcip; 4797 4798 mcip = mac_primary_client_handle(mip); 4799 if (mcip != NULL) { 4800 mac_client_get_resources((mac_client_handle_t)mcip, mrp); 4801 return; 4802 } 4803 bcopy(&mip->mi_resource_props, mrp, sizeof (mac_resource_props_t)); 4804 } 4805 4806 /* 4807 * Get the effective properties from the primary client of the 4808 * specified MAC instance. 4809 */ 4810 void 4811 mac_get_effective_resources(mac_handle_t mh, mac_resource_props_t *mrp) 4812 { 4813 mac_impl_t *mip = (mac_impl_t *)mh; 4814 mac_client_impl_t *mcip; 4815 4816 mcip = mac_primary_client_handle(mip); 4817 if (mcip != NULL) { 4818 mac_client_get_effective_resources((mac_client_handle_t)mcip, 4819 mrp); 4820 return; 4821 } 4822 bzero(mrp, sizeof (mac_resource_props_t)); 4823 } 4824 4825 int 4826 mac_set_pvid(mac_handle_t mh, uint16_t pvid) 4827 { 4828 mac_impl_t *mip = (mac_impl_t *)mh; 4829 mac_client_impl_t *mcip; 4830 mac_unicast_impl_t *muip; 4831 4832 i_mac_perim_enter(mip); 4833 if (pvid != 0) { 4834 for (mcip = mip->mi_clients_list; mcip != NULL; 4835 mcip = mcip->mci_client_next) { 4836 for (muip = mcip->mci_unicast_list; muip != NULL; 4837 muip = muip->mui_next) { 4838 if (muip->mui_vid == pvid) { 4839 i_mac_perim_exit(mip); 4840 return (EBUSY); 4841 } 4842 } 4843 } 4844 } 4845 mip->mi_pvid = pvid; 4846 i_mac_perim_exit(mip); 4847 return (0); 4848 } 4849 4850 uint16_t 4851 mac_get_pvid(mac_handle_t mh) 4852 { 4853 mac_impl_t *mip = (mac_impl_t *)mh; 4854 4855 return (mip->mi_pvid); 4856 } 4857 4858 uint32_t 4859 mac_get_llimit(mac_handle_t mh) 4860 { 4861 mac_impl_t *mip = (mac_impl_t *)mh; 4862 4863 return (mip->mi_llimit); 4864 } 4865 4866 uint32_t 4867 mac_get_ldecay(mac_handle_t mh) 4868 { 4869 mac_impl_t *mip = (mac_impl_t *)mh; 4870 4871 return (mip->mi_ldecay); 4872 } 4873 4874 /* 4875 * Rename a mac client, its flow, and the kstat. 4876 */ 4877 int 4878 mac_rename_primary(mac_handle_t mh, const char *new_name) 4879 { 4880 mac_impl_t *mip = (mac_impl_t *)mh; 4881 mac_client_impl_t *cur_clnt = NULL; 4882 flow_entry_t *fep; 4883 4884 i_mac_perim_enter(mip); 4885 4886 /* 4887 * VNICs: we need to change the sys flow name and 4888 * the associated flow kstat. 4889 */ 4890 if (mip->mi_state_flags & MIS_IS_VNIC) { 4891 mac_client_impl_t *mcip = mac_vnic_lower(mip); 4892 ASSERT(new_name != NULL); 4893 mac_rename_flow_names(mcip, new_name); 4894 mac_stat_rename(mcip); 4895 goto done; 4896 } 4897 /* 4898 * This mac may itself be an aggr link, or it may have some client 4899 * which is an aggr port. For both cases, we need to change the 4900 * aggr port's mac client name, its flow name and the associated flow 4901 * kstat. 4902 */ 4903 if (mip->mi_state_flags & MIS_IS_AGGR) { 4904 mac_capab_aggr_t aggr_cap; 4905 mac_rename_fn_t rename_fn; 4906 boolean_t ret; 4907 4908 ASSERT(new_name != NULL); 4909 ret = i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_AGGR, 4910 (void *)(&aggr_cap)); 4911 ASSERT(ret == B_TRUE); 4912 rename_fn = aggr_cap.mca_rename_fn; 4913 rename_fn(new_name, mip->mi_driver); 4914 /* 4915 * The aggr's client name and kstat flow name will be 4916 * updated below, i.e. via mac_rename_flow_names. 4917 */ 4918 } 4919 4920 for (cur_clnt = mip->mi_clients_list; cur_clnt != NULL; 4921 cur_clnt = cur_clnt->mci_client_next) { 4922 if (cur_clnt->mci_state_flags & MCIS_IS_AGGR_PORT) { 4923 if (new_name != NULL) { 4924 char *str_st = cur_clnt->mci_name; 4925 char *str_del = strchr(str_st, '-'); 4926 4927 ASSERT(str_del != NULL); 4928 bzero(str_del + 1, MAXNAMELEN - 4929 (str_del - str_st + 1)); 4930 bcopy(new_name, str_del + 1, 4931 strlen(new_name)); 4932 } 4933 fep = cur_clnt->mci_flent; 4934 mac_rename_flow(fep, cur_clnt->mci_name); 4935 break; 4936 } else if (new_name != NULL && 4937 cur_clnt->mci_state_flags & MCIS_USE_DATALINK_NAME) { 4938 mac_rename_flow_names(cur_clnt, new_name); 4939 break; 4940 } 4941 } 4942 4943 /* Recreate kstats associated with aggr pseudo rings */ 4944 if (mip->mi_state_flags & MIS_IS_AGGR) 4945 mac_pseudo_ring_stat_rename(mip); 4946 4947 done: 4948 i_mac_perim_exit(mip); 4949 return (0); 4950 } 4951 4952 /* 4953 * Rename the MAC client's flow names 4954 */ 4955 static void 4956 mac_rename_flow_names(mac_client_impl_t *mcip, const char *new_name) 4957 { 4958 flow_entry_t *flent; 4959 uint16_t vid; 4960 char flowname[MAXFLOWNAMELEN]; 4961 mac_impl_t *mip = mcip->mci_mip; 4962 4963 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip)); 4964 4965 /* 4966 * Use mi_rw_lock to ensure that threads not in the mac perimeter 4967 * see a self-consistent value for mci_name 4968 */ 4969 rw_enter(&mip->mi_rw_lock, RW_WRITER); 4970 (void) strlcpy(mcip->mci_name, new_name, sizeof (mcip->mci_name)); 4971 rw_exit(&mip->mi_rw_lock); 4972 4973 mac_rename_flow(mcip->mci_flent, new_name); 4974 4975 if (mcip->mci_nflents == 1) 4976 return; 4977 4978 /* 4979 * We have to rename all the others too, no stats to destroy for 4980 * these. 4981 */ 4982 for (flent = mcip->mci_flent_list; flent != NULL; 4983 flent = flent->fe_client_next) { 4984 if (flent != mcip->mci_flent) { 4985 vid = i_mac_flow_vid(flent); 4986 (void) sprintf(flowname, "%s%u", new_name, vid); 4987 mac_flow_set_name(flent, flowname); 4988 } 4989 } 4990 } 4991 4992 4993 /* 4994 * Add a flow to the MAC client's flow list - i.e list of MAC/VID tuples 4995 * defined for the specified MAC client. 4996 */ 4997 static void 4998 mac_client_add_to_flow_list(mac_client_impl_t *mcip, flow_entry_t *flent) 4999 { 5000 ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip)); 5001 /* 5002 * The promisc Rx data path walks the mci_flent_list. Protect by 5003 * using mi_rw_lock 5004 */ 5005 rw_enter(&mcip->mci_rw_lock, RW_WRITER); 5006 5007 mcip->mci_vidcache = MCIP_VIDCACHE_INVALID; 5008 5009 /* Add it to the head */ 5010 flent->fe_client_next = mcip->mci_flent_list; 5011 mcip->mci_flent_list = flent; 5012 mcip->mci_nflents++; 5013 5014 /* 5015 * Keep track of the number of non-zero VIDs addresses per MAC 5016 * client to avoid figuring it out in the data-path. 5017 */ 5018 if (i_mac_flow_vid(flent) != VLAN_ID_NONE) 5019 mcip->mci_nvids++; 5020 5021 rw_exit(&mcip->mci_rw_lock); 5022 } 5023 5024 /* 5025 * Remove a flow entry from the MAC client's list. 5026 */ 5027 static void 5028 mac_client_remove_flow_from_list(mac_client_impl_t *mcip, flow_entry_t *flent) 5029 { 5030 flow_entry_t *fe = mcip->mci_flent_list; 5031 flow_entry_t *prev_fe = NULL; 5032 5033 ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip)); 5034 /* 5035 * The promisc Rx data path walks the mci_flent_list. Protect by 5036 * using mci_rw_lock 5037 */ 5038 rw_enter(&mcip->mci_rw_lock, RW_WRITER); 5039 mcip->mci_vidcache = MCIP_VIDCACHE_INVALID; 5040 5041 while ((fe != NULL) && (fe != flent)) { 5042 prev_fe = fe; 5043 fe = fe->fe_client_next; 5044 } 5045 5046 ASSERT(fe != NULL); 5047 if (prev_fe == NULL) { 5048 /* Deleting the first node */ 5049 mcip->mci_flent_list = fe->fe_client_next; 5050 } else { 5051 prev_fe->fe_client_next = fe->fe_client_next; 5052 } 5053 mcip->mci_nflents--; 5054 5055 if (i_mac_flow_vid(flent) != VLAN_ID_NONE) 5056 mcip->mci_nvids--; 5057 5058 rw_exit(&mcip->mci_rw_lock); 5059 } 5060 5061 /* 5062 * Check if the given VID belongs to this MAC client. 5063 */ 5064 boolean_t 5065 mac_client_check_flow_vid(mac_client_impl_t *mcip, uint16_t vid) 5066 { 5067 flow_entry_t *flent; 5068 uint16_t mci_vid; 5069 uint32_t cache = mcip->mci_vidcache; 5070 5071 /* 5072 * In hopes of not having to touch the mci_rw_lock, check to see if 5073 * this vid matches our cached result. 5074 */ 5075 if (MCIP_VIDCACHE_ISVALID(cache) && MCIP_VIDCACHE_VID(cache) == vid) 5076 return (MCIP_VIDCACHE_BOOL(cache) ? B_TRUE : B_FALSE); 5077 5078 /* The mci_flent_list is protected by mci_rw_lock */ 5079 rw_enter(&mcip->mci_rw_lock, RW_WRITER); 5080 for (flent = mcip->mci_flent_list; flent != NULL; 5081 flent = flent->fe_client_next) { 5082 mci_vid = i_mac_flow_vid(flent); 5083 if (vid == mci_vid) { 5084 mcip->mci_vidcache = MCIP_VIDCACHE_CACHE(vid, B_TRUE); 5085 rw_exit(&mcip->mci_rw_lock); 5086 return (B_TRUE); 5087 } 5088 } 5089 5090 mcip->mci_vidcache = MCIP_VIDCACHE_CACHE(vid, B_FALSE); 5091 rw_exit(&mcip->mci_rw_lock); 5092 return (B_FALSE); 5093 } 5094 5095 /* 5096 * Get the flow entry for the specified <MAC addr, VID> tuple. 5097 */ 5098 static flow_entry_t * 5099 mac_client_get_flow(mac_client_impl_t *mcip, mac_unicast_impl_t *muip) 5100 { 5101 mac_address_t *map = mcip->mci_unicast; 5102 flow_entry_t *flent; 5103 uint16_t vid; 5104 flow_desc_t flow_desc; 5105 5106 ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip)); 5107 5108 mac_flow_get_desc(mcip->mci_flent, &flow_desc); 5109 if (bcmp(flow_desc.fd_dst_mac, map->ma_addr, map->ma_len) != 0) 5110 return (NULL); 5111 5112 for (flent = mcip->mci_flent_list; flent != NULL; 5113 flent = flent->fe_client_next) { 5114 vid = i_mac_flow_vid(flent); 5115 if (vid == muip->mui_vid) { 5116 return (flent); 5117 } 5118 } 5119 5120 return (NULL); 5121 } 5122 5123 /* 5124 * Since mci_flent has the SRSs, when we want to remove it, we replace 5125 * the flow_desc_t in mci_flent with that of an existing flent and then 5126 * remove that flent instead of mci_flent. 5127 */ 5128 static flow_entry_t * 5129 mac_client_swap_mciflent(mac_client_impl_t *mcip) 5130 { 5131 flow_entry_t *flent = mcip->mci_flent; 5132 flow_tab_t *ft = flent->fe_flow_tab; 5133 flow_entry_t *flent1; 5134 flow_desc_t fl_desc; 5135 char fl_name[MAXFLOWNAMELEN]; 5136 int err; 5137 5138 ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip)); 5139 ASSERT(mcip->mci_nflents > 1); 5140 5141 /* get the next flent following the primary flent */ 5142 flent1 = mcip->mci_flent_list->fe_client_next; 5143 ASSERT(flent1 != NULL && flent1->fe_flow_tab == ft); 5144 5145 /* 5146 * Remove the flent from the flow table before updating the 5147 * flow descriptor as the hash depends on the flow descriptor. 5148 * This also helps incoming packet classification avoid having 5149 * to grab fe_lock. Access to fe_flow_desc of a flent not in the 5150 * flow table is done under the fe_lock so that log or stat functions 5151 * see a self-consistent fe_flow_desc. The name and desc are specific 5152 * to a flow, the rest are shared by all the clients, including 5153 * resource control etc. 5154 */ 5155 mac_flow_remove(ft, flent, B_TRUE); 5156 mac_flow_remove(ft, flent1, B_TRUE); 5157 5158 bcopy(&flent->fe_flow_desc, &fl_desc, sizeof (flow_desc_t)); 5159 bcopy(flent->fe_flow_name, fl_name, MAXFLOWNAMELEN); 5160 5161 /* update the primary flow entry */ 5162 mutex_enter(&flent->fe_lock); 5163 bcopy(&flent1->fe_flow_desc, &flent->fe_flow_desc, 5164 sizeof (flow_desc_t)); 5165 bcopy(&flent1->fe_flow_name, &flent->fe_flow_name, MAXFLOWNAMELEN); 5166 mutex_exit(&flent->fe_lock); 5167 5168 /* update the flow entry that is to be freed */ 5169 mutex_enter(&flent1->fe_lock); 5170 bcopy(&fl_desc, &flent1->fe_flow_desc, sizeof (flow_desc_t)); 5171 bcopy(fl_name, &flent1->fe_flow_name, MAXFLOWNAMELEN); 5172 mutex_exit(&flent1->fe_lock); 5173 5174 /* now reinsert the flow entries in the table */ 5175 err = mac_flow_add(ft, flent); 5176 ASSERT(err == 0); 5177 5178 err = mac_flow_add(ft, flent1); 5179 ASSERT(err == 0); 5180 5181 return (flent1); 5182 } 5183 5184 /* 5185 * Return whether there is only one flow entry associated with this 5186 * MAC client. 5187 */ 5188 static boolean_t 5189 mac_client_single_rcvr(mac_client_impl_t *mcip) 5190 { 5191 return (mcip->mci_nflents == 1); 5192 } 5193 5194 int 5195 mac_validate_props(mac_impl_t *mip, mac_resource_props_t *mrp) 5196 { 5197 boolean_t reset; 5198 uint32_t rings_needed; 5199 uint32_t rings_avail; 5200 mac_group_type_t gtype; 5201 mac_resource_props_t *mip_mrp; 5202 5203 if (mrp == NULL) 5204 return (0); 5205 5206 if (mrp->mrp_mask & MRP_PRIORITY) { 5207 mac_priority_level_t pri = mrp->mrp_priority; 5208 5209 if (pri < MPL_LOW || pri > MPL_RESET) 5210 return (EINVAL); 5211 } 5212 5213 if (mrp->mrp_mask & MRP_MAXBW) { 5214 uint64_t maxbw = mrp->mrp_maxbw; 5215 5216 if (maxbw < MRP_MAXBW_MINVAL && maxbw != 0) 5217 return (EINVAL); 5218 } 5219 if (mrp->mrp_mask & MRP_CPUS) { 5220 int i, j; 5221 mac_cpu_mode_t fanout; 5222 5223 if (mrp->mrp_ncpus > ncpus) 5224 return (EINVAL); 5225 5226 for (i = 0; i < mrp->mrp_ncpus; i++) { 5227 for (j = 0; j < mrp->mrp_ncpus; j++) { 5228 if (i != j && 5229 mrp->mrp_cpu[i] == mrp->mrp_cpu[j]) { 5230 return (EINVAL); 5231 } 5232 } 5233 } 5234 5235 for (i = 0; i < mrp->mrp_ncpus; i++) { 5236 cpu_t *cp; 5237 int rv; 5238 5239 mutex_enter(&cpu_lock); 5240 cp = cpu_get(mrp->mrp_cpu[i]); 5241 if (cp != NULL) 5242 rv = cpu_is_online(cp); 5243 else 5244 rv = 0; 5245 mutex_exit(&cpu_lock); 5246 if (rv == 0) 5247 return (EINVAL); 5248 } 5249 5250 fanout = mrp->mrp_fanout_mode; 5251 if (fanout < 0 || fanout > MCM_CPUS) 5252 return (EINVAL); 5253 } 5254 5255 if (mrp->mrp_mask & MRP_PROTECT) { 5256 int err = mac_protect_validate(mrp); 5257 if (err != 0) 5258 return (err); 5259 } 5260 5261 if (!(mrp->mrp_mask & MRP_RX_RINGS) && 5262 !(mrp->mrp_mask & MRP_TX_RINGS)) { 5263 return (0); 5264 } 5265 5266 /* 5267 * mip will be null when we come from mac_flow_create or 5268 * mac_link_flow_modify. In the latter case it is a user flow, 5269 * for which we don't support rings. In the former we would 5270 * have validated the props beforehand (i_mac_unicast_add -> 5271 * mac_client_set_resources -> validate for the primary and 5272 * vnic_dev_create -> mac_client_set_resources -> validate for 5273 * a vnic. 5274 */ 5275 if (mip == NULL) 5276 return (0); 5277 5278 /* 5279 * We don't support setting rings property for a VNIC that is using a 5280 * primary address (VLAN) 5281 */ 5282 if ((mip->mi_state_flags & MIS_IS_VNIC) && 5283 mac_is_vnic_primary((mac_handle_t)mip)) { 5284 return (ENOTSUP); 5285 } 5286 5287 mip_mrp = &mip->mi_resource_props; 5288 /* 5289 * The rings property should be validated against the NICs 5290 * resources 5291 */ 5292 if (mip->mi_state_flags & MIS_IS_VNIC) 5293 mip = (mac_impl_t *)mac_get_lower_mac_handle((mac_handle_t)mip); 5294 5295 reset = mrp->mrp_mask & MRP_RINGS_RESET; 5296 /* 5297 * If groups are not supported, return error. 5298 */ 5299 if (((mrp->mrp_mask & MRP_RX_RINGS) && mip->mi_rx_groups == NULL) || 5300 ((mrp->mrp_mask & MRP_TX_RINGS) && mip->mi_tx_groups == NULL)) { 5301 return (EINVAL); 5302 } 5303 /* 5304 * If we are just resetting, there is no validation needed. 5305 */ 5306 if (reset) 5307 return (0); 5308 5309 if (mrp->mrp_mask & MRP_RX_RINGS) { 5310 rings_needed = mrp->mrp_nrxrings; 5311 /* 5312 * We just want to check if the number of additional 5313 * rings requested is available. 5314 */ 5315 if (mip_mrp->mrp_mask & MRP_RX_RINGS) { 5316 if (mrp->mrp_nrxrings > mip_mrp->mrp_nrxrings) 5317 /* Just check for the additional rings */ 5318 rings_needed -= mip_mrp->mrp_nrxrings; 5319 else 5320 /* We are not asking for additional rings */ 5321 rings_needed = 0; 5322 } 5323 rings_avail = mip->mi_rxrings_avail; 5324 gtype = mip->mi_rx_group_type; 5325 } else { 5326 rings_needed = mrp->mrp_ntxrings; 5327 /* Similarly for the TX rings */ 5328 if (mip_mrp->mrp_mask & MRP_TX_RINGS) { 5329 if (mrp->mrp_ntxrings > mip_mrp->mrp_ntxrings) 5330 /* Just check for the additional rings */ 5331 rings_needed -= mip_mrp->mrp_ntxrings; 5332 else 5333 /* We are not asking for additional rings */ 5334 rings_needed = 0; 5335 } 5336 rings_avail = mip->mi_txrings_avail; 5337 gtype = mip->mi_tx_group_type; 5338 } 5339 5340 /* Error if the group is dynamic .. */ 5341 if (gtype == MAC_GROUP_TYPE_DYNAMIC) { 5342 /* 5343 * .. and rings specified are more than available. 5344 */ 5345 if (rings_needed > rings_avail) 5346 return (EINVAL); 5347 } else { 5348 /* 5349 * OR group is static and we have specified some rings. 5350 */ 5351 if (rings_needed > 0) 5352 return (EINVAL); 5353 } 5354 return (0); 5355 } 5356 5357 /* 5358 * Send a MAC_NOTE_LINK notification to all the MAC clients whenever the 5359 * underlying physical link is down. This is to allow MAC clients to 5360 * communicate with other clients. 5361 */ 5362 void 5363 mac_virtual_link_update(mac_impl_t *mip) 5364 { 5365 if (mip->mi_linkstate != LINK_STATE_UP) 5366 i_mac_notify(mip, MAC_NOTE_LINK); 5367 } 5368 5369 /* 5370 * For clients that have a pass-thru MAC, e.g. VNIC, we set the VNIC's 5371 * mac handle in the client. 5372 */ 5373 void 5374 mac_set_upper_mac(mac_client_handle_t mch, mac_handle_t mh, 5375 mac_resource_props_t *mrp) 5376 { 5377 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 5378 mac_impl_t *mip = (mac_impl_t *)mh; 5379 5380 mcip->mci_upper_mip = mip; 5381 /* If there are any properties, copy it over too */ 5382 if (mrp != NULL) { 5383 bcopy(mrp, &mip->mi_resource_props, 5384 sizeof (mac_resource_props_t)); 5385 } 5386 } 5387 5388 /* 5389 * Mark the mac as being used exclusively by the single mac client that is 5390 * doing some control operation on this mac. No further opens of this mac 5391 * will be allowed until this client calls mac_unmark_exclusive. The mac 5392 * client calling this function must already be in the mac perimeter 5393 */ 5394 int 5395 mac_mark_exclusive(mac_handle_t mh) 5396 { 5397 mac_impl_t *mip = (mac_impl_t *)mh; 5398 5399 ASSERT(MAC_PERIM_HELD(mh)); 5400 /* 5401 * Look up its entry in the global hash table. 5402 */ 5403 rw_enter(&i_mac_impl_lock, RW_WRITER); 5404 if (mip->mi_state_flags & MIS_DISABLED) { 5405 rw_exit(&i_mac_impl_lock); 5406 return (ENOENT); 5407 } 5408 5409 /* 5410 * A reference to mac is held even if the link is not plumbed. 5411 * In i_dls_link_create() we open the MAC interface and hold the 5412 * reference. There is an additional reference for the mac_open 5413 * done in acquiring the mac perimeter 5414 */ 5415 if (mip->mi_ref != 2) { 5416 rw_exit(&i_mac_impl_lock); 5417 return (EBUSY); 5418 } 5419 5420 ASSERT(!(mip->mi_state_flags & MIS_EXCLUSIVE_HELD)); 5421 mip->mi_state_flags |= MIS_EXCLUSIVE_HELD; 5422 rw_exit(&i_mac_impl_lock); 5423 return (0); 5424 } 5425 5426 void 5427 mac_unmark_exclusive(mac_handle_t mh) 5428 { 5429 mac_impl_t *mip = (mac_impl_t *)mh; 5430 5431 ASSERT(MAC_PERIM_HELD(mh)); 5432 5433 rw_enter(&i_mac_impl_lock, RW_WRITER); 5434 /* 1 for the creation and another for the perimeter */ 5435 ASSERT(mip->mi_ref == 2 && (mip->mi_state_flags & MIS_EXCLUSIVE_HELD)); 5436 mip->mi_state_flags &= ~MIS_EXCLUSIVE_HELD; 5437 rw_exit(&i_mac_impl_lock); 5438 } 5439 5440 /* 5441 * Set the MTU for the specified MAC. 5442 */ 5443 int 5444 mac_set_mtu(mac_handle_t mh, uint_t new_mtu, uint_t *old_mtu_arg) 5445 { 5446 mac_impl_t *mip = (mac_impl_t *)mh; 5447 uint_t old_mtu; 5448 int rv = 0; 5449 5450 i_mac_perim_enter(mip); 5451 5452 if (!(mip->mi_callbacks->mc_callbacks & (MC_SETPROP|MC_GETPROP))) { 5453 rv = ENOTSUP; 5454 goto bail; 5455 } 5456 5457 old_mtu = mip->mi_sdu_max; 5458 5459 if (new_mtu == 0 || new_mtu < mip->mi_sdu_min) { 5460 rv = EINVAL; 5461 goto bail; 5462 } 5463 5464 rw_enter(&mip->mi_rw_lock, RW_READER); 5465 if (mip->mi_mtrp != NULL && new_mtu < mip->mi_mtrp->mtr_mtu) { 5466 rv = EBUSY; 5467 rw_exit(&mip->mi_rw_lock); 5468 goto bail; 5469 } 5470 rw_exit(&mip->mi_rw_lock); 5471 5472 if (old_mtu != new_mtu) { 5473 rv = mip->mi_callbacks->mc_setprop(mip->mi_driver, 5474 "mtu", MAC_PROP_MTU, sizeof (uint_t), &new_mtu); 5475 if (rv != 0) 5476 goto bail; 5477 rv = mac_maxsdu_update(mh, new_mtu); 5478 ASSERT(rv == 0); 5479 } 5480 5481 bail: 5482 i_mac_perim_exit(mip); 5483 5484 if (rv == 0 && old_mtu_arg != NULL) 5485 *old_mtu_arg = old_mtu; 5486 return (rv); 5487 } 5488 5489 /* 5490 * Return the RX h/w information for the group indexed by grp_num. 5491 */ 5492 void 5493 mac_get_hwrxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num, 5494 uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts, 5495 char *clnts_name) 5496 { 5497 mac_impl_t *mip = (mac_impl_t *)mh; 5498 mac_grp_client_t *mcip; 5499 uint_t i = 0, index = 0; 5500 mac_ring_t *ring; 5501 5502 /* Revisit when we implement fully dynamic group allocation */ 5503 ASSERT(grp_index >= 0 && grp_index < mip->mi_rx_group_count); 5504 5505 rw_enter(&mip->mi_rw_lock, RW_READER); 5506 *grp_num = mip->mi_rx_groups[grp_index].mrg_index; 5507 *type = mip->mi_rx_groups[grp_index].mrg_type; 5508 *n_rings = mip->mi_rx_groups[grp_index].mrg_cur_count; 5509 ring = mip->mi_rx_groups[grp_index].mrg_rings; 5510 for (index = 0; index < mip->mi_rx_groups[grp_index].mrg_cur_count; 5511 index++) { 5512 rings[index] = ring->mr_index; 5513 ring = ring->mr_next; 5514 } 5515 /* Assuming the 1st is the default group */ 5516 index = 0; 5517 if (grp_index == 0) { 5518 (void) strlcpy(clnts_name, "<default,mcast>,", 5519 MAXCLIENTNAMELEN); 5520 index += strlen("<default,mcast>,"); 5521 } 5522 for (mcip = mip->mi_rx_groups[grp_index].mrg_clients; mcip != NULL; 5523 mcip = mcip->mgc_next) { 5524 int name_len = strlen(mcip->mgc_client->mci_name); 5525 5526 /* 5527 * MAXCLIENTNAMELEN is the buffer size reserved for client 5528 * names. 5529 * XXXX Formating the client name string needs to be moved 5530 * to user land when fixing the size of dhi_clnts in 5531 * dld_hwgrpinfo_t. We should use n_clients * client_name for 5532 * dhi_clntsin instead of MAXCLIENTNAMELEN 5533 */ 5534 if (index + name_len >= MAXCLIENTNAMELEN) { 5535 index = MAXCLIENTNAMELEN; 5536 break; 5537 } 5538 bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]), 5539 name_len); 5540 index += name_len; 5541 clnts_name[index++] = ','; 5542 i++; 5543 } 5544 5545 /* Get rid of the last , */ 5546 if (index > 0) 5547 clnts_name[index - 1] = '\0'; 5548 *n_clnts = i; 5549 rw_exit(&mip->mi_rw_lock); 5550 } 5551 5552 /* 5553 * Return the TX h/w information for the group indexed by grp_num. 5554 */ 5555 void 5556 mac_get_hwtxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num, 5557 uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts, 5558 char *clnts_name) 5559 { 5560 mac_impl_t *mip = (mac_impl_t *)mh; 5561 mac_grp_client_t *mcip; 5562 uint_t i = 0, index = 0; 5563 mac_ring_t *ring; 5564 5565 /* Revisit when we implement fully dynamic group allocation */ 5566 ASSERT(grp_index >= 0 && grp_index <= mip->mi_tx_group_count); 5567 5568 rw_enter(&mip->mi_rw_lock, RW_READER); 5569 *grp_num = mip->mi_tx_groups[grp_index].mrg_index > 0 ? 5570 mip->mi_tx_groups[grp_index].mrg_index : grp_index; 5571 *type = mip->mi_tx_groups[grp_index].mrg_type; 5572 *n_rings = mip->mi_tx_groups[grp_index].mrg_cur_count; 5573 ring = mip->mi_tx_groups[grp_index].mrg_rings; 5574 for (index = 0; index < mip->mi_tx_groups[grp_index].mrg_cur_count; 5575 index++) { 5576 rings[index] = ring->mr_index; 5577 ring = ring->mr_next; 5578 } 5579 index = 0; 5580 /* Default group has an index of -1 */ 5581 if (mip->mi_tx_groups[grp_index].mrg_index < 0) { 5582 (void) strlcpy(clnts_name, "<default>,", 5583 MAXCLIENTNAMELEN); 5584 index += strlen("<default>,"); 5585 } 5586 for (mcip = mip->mi_tx_groups[grp_index].mrg_clients; mcip != NULL; 5587 mcip = mcip->mgc_next) { 5588 int name_len = strlen(mcip->mgc_client->mci_name); 5589 5590 /* 5591 * MAXCLIENTNAMELEN is the buffer size reserved for client 5592 * names. 5593 * XXXX Formating the client name string needs to be moved 5594 * to user land when fixing the size of dhi_clnts in 5595 * dld_hwgrpinfo_t. We should use n_clients * client_name for 5596 * dhi_clntsin instead of MAXCLIENTNAMELEN 5597 */ 5598 if (index + name_len >= MAXCLIENTNAMELEN) { 5599 index = MAXCLIENTNAMELEN; 5600 break; 5601 } 5602 bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]), 5603 name_len); 5604 index += name_len; 5605 clnts_name[index++] = ','; 5606 i++; 5607 } 5608 5609 /* Get rid of the last , */ 5610 if (index > 0) 5611 clnts_name[index - 1] = '\0'; 5612 *n_clnts = i; 5613 rw_exit(&mip->mi_rw_lock); 5614 } 5615 5616 /* 5617 * Return the group count for RX or TX. 5618 */ 5619 uint_t 5620 mac_hwgrp_num(mac_handle_t mh, int type) 5621 { 5622 mac_impl_t *mip = (mac_impl_t *)mh; 5623 5624 /* 5625 * Return the Rx and Tx group count; for the Tx we need to 5626 * include the default too. 5627 */ 5628 return (type == MAC_RING_TYPE_RX ? mip->mi_rx_group_count : 5629 mip->mi_tx_groups != NULL ? mip->mi_tx_group_count + 1 : 0); 5630 } 5631 5632 /* 5633 * The total number of free TX rings for this MAC. 5634 */ 5635 uint_t 5636 mac_txavail_get(mac_handle_t mh) 5637 { 5638 mac_impl_t *mip = (mac_impl_t *)mh; 5639 5640 return (mip->mi_txrings_avail); 5641 } 5642 5643 /* 5644 * The total number of free RX rings for this MAC. 5645 */ 5646 uint_t 5647 mac_rxavail_get(mac_handle_t mh) 5648 { 5649 mac_impl_t *mip = (mac_impl_t *)mh; 5650 5651 return (mip->mi_rxrings_avail); 5652 } 5653 5654 /* 5655 * The total number of reserved RX rings on this MAC. 5656 */ 5657 uint_t 5658 mac_rxrsvd_get(mac_handle_t mh) 5659 { 5660 mac_impl_t *mip = (mac_impl_t *)mh; 5661 5662 return (mip->mi_rxrings_rsvd); 5663 } 5664 5665 /* 5666 * The total number of reserved TX rings on this MAC. 5667 */ 5668 uint_t 5669 mac_txrsvd_get(mac_handle_t mh) 5670 { 5671 mac_impl_t *mip = (mac_impl_t *)mh; 5672 5673 return (mip->mi_txrings_rsvd); 5674 } 5675 5676 /* 5677 * Total number of free RX groups on this MAC. 5678 */ 5679 uint_t 5680 mac_rxhwlnksavail_get(mac_handle_t mh) 5681 { 5682 mac_impl_t *mip = (mac_impl_t *)mh; 5683 5684 return (mip->mi_rxhwclnt_avail); 5685 } 5686 5687 /* 5688 * Total number of RX groups reserved on this MAC. 5689 */ 5690 uint_t 5691 mac_rxhwlnksrsvd_get(mac_handle_t mh) 5692 { 5693 mac_impl_t *mip = (mac_impl_t *)mh; 5694 5695 return (mip->mi_rxhwclnt_used); 5696 } 5697 5698 /* 5699 * Total number of free TX groups on this MAC. 5700 */ 5701 uint_t 5702 mac_txhwlnksavail_get(mac_handle_t mh) 5703 { 5704 mac_impl_t *mip = (mac_impl_t *)mh; 5705 5706 return (mip->mi_txhwclnt_avail); 5707 } 5708 5709 /* 5710 * Total number of TX groups reserved on this MAC. 5711 */ 5712 uint_t 5713 mac_txhwlnksrsvd_get(mac_handle_t mh) 5714 { 5715 mac_impl_t *mip = (mac_impl_t *)mh; 5716 5717 return (mip->mi_txhwclnt_used); 5718 } 5719 5720 /* 5721 * Initialize the rings property for a mac client. A non-0 value for 5722 * rxring or txring specifies the number of rings required, a value 5723 * of MAC_RXRINGS_NONE/MAC_TXRINGS_NONE specifies that it doesn't need 5724 * any RX/TX rings and a value of MAC_RXRINGS_DONTCARE/MAC_TXRINGS_DONTCARE 5725 * means the system can decide whether it can give any rings or not. 5726 */ 5727 void 5728 mac_client_set_rings(mac_client_handle_t mch, int rxrings, int txrings) 5729 { 5730 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 5731 mac_resource_props_t *mrp = MCIP_RESOURCE_PROPS(mcip); 5732 5733 if (rxrings != MAC_RXRINGS_DONTCARE) { 5734 mrp->mrp_mask |= MRP_RX_RINGS; 5735 mrp->mrp_nrxrings = rxrings; 5736 } 5737 5738 if (txrings != MAC_TXRINGS_DONTCARE) { 5739 mrp->mrp_mask |= MRP_TX_RINGS; 5740 mrp->mrp_ntxrings = txrings; 5741 } 5742 } 5743 5744 boolean_t 5745 mac_get_promisc_filtered(mac_client_handle_t mch) 5746 { 5747 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 5748 5749 return (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED); 5750 } 5751 5752 void 5753 mac_set_promisc_filtered(mac_client_handle_t mch, boolean_t enable) 5754 { 5755 mac_client_impl_t *mcip = (mac_client_impl_t *)mch; 5756 5757 ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip)); 5758 if (enable) 5759 mcip->mci_protect_flags |= MPT_FLAG_PROMISC_FILTERED; 5760 else 5761 mcip->mci_protect_flags &= ~MPT_FLAG_PROMISC_FILTERED; 5762 } 5763