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 2017 OmniTI Computer Consulting, Inc. All rights reserved.
  25  */
  26 
  27 #include <sys/types.h>
  28 #include <sys/conf.h>
  29 #include <sys/id_space.h>
  30 #include <sys/esunddi.h>
  31 #include <sys/stat.h>
  32 #include <sys/mkdev.h>
  33 #include <sys/stream.h>
  34 #include <sys/strsubr.h>
  35 #include <sys/dlpi.h>
  36 #include <sys/modhash.h>
  37 #include <sys/mac.h>
  38 #include <sys/mac_provider.h>
  39 #include <sys/mac_impl.h>
  40 #include <sys/mac_client_impl.h>
  41 #include <sys/mac_client_priv.h>
  42 #include <sys/mac_soft_ring.h>
  43 #include <sys/mac_stat.h>
  44 #include <sys/dld.h>
  45 #include <sys/modctl.h>
  46 #include <sys/fs/dv_node.h>
  47 #include <sys/thread.h>
  48 #include <sys/proc.h>
  49 #include <sys/callb.h>
  50 #include <sys/cpuvar.h>
  51 #include <sys/atomic.h>
  52 #include <sys/sdt.h>
  53 #include <sys/mac_flow.h>
  54 #include <sys/ddi_intr_impl.h>
  55 #include <sys/disp.h>
  56 #include <sys/sdt.h>
  57 #include <sys/pattr.h>
  58 #include <sys/strsun.h>
  59 
  60 /*
  61  * MAC Provider Interface.
  62  *
  63  * Interface for GLDv3 compatible NIC drivers.
  64  */
  65 
  66 static void i_mac_notify_thread(void *);
  67 
  68 typedef void (*mac_notify_default_cb_fn_t)(mac_impl_t *);
  69 
  70 static const mac_notify_default_cb_fn_t mac_notify_cb_list[MAC_NNOTE] = {
  71         mac_fanout_recompute,   /* MAC_NOTE_LINK */
  72         NULL,           /* MAC_NOTE_UNICST */
  73         NULL,           /* MAC_NOTE_TX */
  74         NULL,           /* MAC_NOTE_DEVPROMISC */
  75         NULL,           /* MAC_NOTE_FASTPATH_FLUSH */
  76         NULL,           /* MAC_NOTE_SDU_SIZE */
  77         NULL,           /* MAC_NOTE_MARGIN */
  78         NULL,           /* MAC_NOTE_CAPAB_CHG */
  79         NULL            /* MAC_NOTE_LOWLINK */
  80 };
  81 
  82 /*
  83  * Driver support functions.
  84  */
  85 
  86 /* REGISTRATION */
  87 
  88 mac_register_t *
  89 mac_alloc(uint_t mac_version)
  90 {
  91         mac_register_t *mregp;
  92 
  93         /*
  94          * Make sure there isn't a version mismatch between the driver and
  95          * the framework.  In the future, if multiple versions are
  96          * supported, this check could become more sophisticated.
  97          */
  98         if (mac_version != MAC_VERSION)
  99                 return (NULL);
 100 
 101         mregp = kmem_zalloc(sizeof (mac_register_t), KM_SLEEP);
 102         mregp->m_version = mac_version;
 103         return (mregp);
 104 }
 105 
 106 void
 107 mac_free(mac_register_t *mregp)
 108 {
 109         kmem_free(mregp, sizeof (mac_register_t));
 110 }
 111 
 112 /*
 113  * mac_register() is how drivers register new MACs with the GLDv3
 114  * framework.  The mregp argument is allocated by drivers using the
 115  * mac_alloc() function, and can be freed using mac_free() immediately upon
 116  * return from mac_register().  Upon success (0 return value), the mhp
 117  * opaque pointer becomes the driver's handle to its MAC interface, and is
 118  * the argument to all other mac module entry points.
 119  */
 120 /* ARGSUSED */
 121 int
 122 mac_register(mac_register_t *mregp, mac_handle_t *mhp)
 123 {
 124         mac_impl_t              *mip;
 125         mactype_t               *mtype;
 126         int                     err = EINVAL;
 127         struct devnames         *dnp = NULL;
 128         uint_t                  instance;
 129         boolean_t               style1_created = B_FALSE;
 130         boolean_t               style2_created = B_FALSE;
 131         char                    *driver;
 132         minor_t                 minor = 0;
 133 
 134         /* A successful call to mac_init_ops() sets the DN_GLDV3_DRIVER flag. */
 135         if (!GLDV3_DRV(ddi_driver_major(mregp->m_dip)))
 136                 return (EINVAL);
 137 
 138         /* Find the required MAC-Type plugin. */
 139         if ((mtype = mactype_getplugin(mregp->m_type_ident)) == NULL)
 140                 return (EINVAL);
 141 
 142         /* Create a mac_impl_t to represent this MAC. */
 143         mip = kmem_cache_alloc(i_mac_impl_cachep, KM_SLEEP);
 144 
 145         /*
 146          * The mac is not ready for open yet.
 147          */
 148         mip->mi_state_flags |= MIS_DISABLED;
 149 
 150         /*
 151          * When a mac is registered, the m_instance field can be set to:
 152          *
 153          *  0:  Get the mac's instance number from m_dip.
 154          *      This is usually used for physical device dips.
 155          *
 156          *  [1 .. MAC_MAX_MINOR-1]: Use the value as the mac's instance number.
 157          *      For example, when an aggregation is created with the key option,
 158          *      "key" will be used as the instance number.
 159          *
 160          *  -1: Assign an instance number from [MAC_MAX_MINOR .. MAXMIN-1].
 161          *      This is often used when a MAC of a virtual link is registered
 162          *      (e.g., aggregation when "key" is not specified, or vnic).
 163          *
 164          * Note that the instance number is used to derive the mi_minor field
 165          * of mac_impl_t, which will then be used to derive the name of kstats
 166          * and the devfs nodes.  The first 2 cases are needed to preserve
 167          * backward compatibility.
 168          */
 169         switch (mregp->m_instance) {
 170         case 0:
 171                 instance = ddi_get_instance(mregp->m_dip);
 172                 break;
 173         case ((uint_t)-1):
 174                 minor = mac_minor_hold(B_TRUE);
 175                 if (minor == 0) {
 176                         err = ENOSPC;
 177                         goto fail;
 178                 }
 179                 instance = minor - 1;
 180                 break;
 181         default:
 182                 instance = mregp->m_instance;
 183                 if (instance >= MAC_MAX_MINOR) {
 184                         err = EINVAL;
 185                         goto fail;
 186                 }
 187                 break;
 188         }
 189 
 190         mip->mi_minor = (minor_t)(instance + 1);
 191         mip->mi_dip = mregp->m_dip;
 192         mip->mi_clients_list = NULL;
 193         mip->mi_nclients = 0;
 194 
 195         /* Set the default IEEE Port VLAN Identifier */
 196         mip->mi_pvid = 1;
 197 
 198         /* Default bridge link learning protection values */
 199         mip->mi_llimit = 1000;
 200         mip->mi_ldecay = 200;
 201 
 202         driver = (char *)ddi_driver_name(mip->mi_dip);
 203 
 204         /* Construct the MAC name as <drvname><instance> */
 205         (void) snprintf(mip->mi_name, sizeof (mip->mi_name), "%s%d",
 206             driver, instance);
 207 
 208         mip->mi_driver = mregp->m_driver;
 209 
 210         mip->mi_type = mtype;
 211         mip->mi_margin = mregp->m_margin;
 212         mip->mi_info.mi_media = mtype->mt_type;
 213         mip->mi_info.mi_nativemedia = mtype->mt_nativetype;
 214         if (mregp->m_max_sdu <= mregp->m_min_sdu)
 215                 goto fail;
 216         if (mregp->m_multicast_sdu == 0)
 217                 mregp->m_multicast_sdu = mregp->m_max_sdu;
 218         if (mregp->m_multicast_sdu < mregp->m_min_sdu ||
 219             mregp->m_multicast_sdu > mregp->m_max_sdu)
 220                 goto fail;
 221         mip->mi_sdu_min = mregp->m_min_sdu;
 222         mip->mi_sdu_max = mregp->m_max_sdu;
 223         mip->mi_sdu_multicast = mregp->m_multicast_sdu;
 224         mip->mi_info.mi_addr_length = mip->mi_type->mt_addr_length;
 225         /*
 226          * If the media supports a broadcast address, cache a pointer to it
 227          * in the mac_info_t so that upper layers can use it.
 228          */
 229         mip->mi_info.mi_brdcst_addr = mip->mi_type->mt_brdcst_addr;
 230 
 231         mip->mi_v12n_level = mregp->m_v12n;
 232 
 233         /*
 234          * Copy the unicast source address into the mac_info_t, but only if
 235          * the MAC-Type defines a non-zero address length.  We need to
 236          * handle MAC-Types that have an address length of 0
 237          * (point-to-point protocol MACs for example).
 238          */
 239         if (mip->mi_type->mt_addr_length > 0) {
 240                 if (mregp->m_src_addr == NULL)
 241                         goto fail;
 242                 mip->mi_info.mi_unicst_addr =
 243                     kmem_alloc(mip->mi_type->mt_addr_length, KM_SLEEP);
 244                 bcopy(mregp->m_src_addr, mip->mi_info.mi_unicst_addr,
 245                     mip->mi_type->mt_addr_length);
 246 
 247                 /*
 248                  * Copy the fixed 'factory' MAC address from the immutable
 249                  * info.  This is taken to be the MAC address currently in
 250                  * use.
 251                  */
 252                 bcopy(mip->mi_info.mi_unicst_addr, mip->mi_addr,
 253                     mip->mi_type->mt_addr_length);
 254 
 255                 /*
 256                  * At this point, we should set up the classification
 257                  * rules etc but we delay it till mac_open() so that
 258                  * the resource discovery has taken place and we
 259                  * know someone wants to use the device. Otherwise
 260                  * memory gets allocated for Rx ring structures even
 261                  * during probe.
 262                  */
 263 
 264                 /* Copy the destination address if one is provided. */
 265                 if (mregp->m_dst_addr != NULL) {
 266                         bcopy(mregp->m_dst_addr, mip->mi_dstaddr,
 267                             mip->mi_type->mt_addr_length);
 268                         mip->mi_dstaddr_set = B_TRUE;
 269                 }
 270         } else if (mregp->m_src_addr != NULL) {
 271                 goto fail;
 272         }
 273 
 274         /*
 275          * The format of the m_pdata is specific to the plugin.  It is
 276          * passed in as an argument to all of the plugin callbacks.  The
 277          * driver can update this information by calling
 278          * mac_pdata_update().
 279          */
 280         if (mip->mi_type->mt_ops.mtops_ops & MTOPS_PDATA_VERIFY) {
 281                 /*
 282                  * Verify if the supplied plugin data is valid.  Note that
 283                  * even if the caller passed in a NULL pointer as plugin data,
 284                  * we still need to verify if that's valid as the plugin may
 285                  * require plugin data to function.
 286                  */
 287                 if (!mip->mi_type->mt_ops.mtops_pdata_verify(mregp->m_pdata,
 288                     mregp->m_pdata_size)) {
 289                         goto fail;
 290                 }
 291                 if (mregp->m_pdata != NULL) {
 292                         mip->mi_pdata =
 293                             kmem_alloc(mregp->m_pdata_size, KM_SLEEP);
 294                         bcopy(mregp->m_pdata, mip->mi_pdata,
 295                             mregp->m_pdata_size);
 296                         mip->mi_pdata_size = mregp->m_pdata_size;
 297                 }
 298         } else if (mregp->m_pdata != NULL) {
 299                 /*
 300                  * The caller supplied non-NULL plugin data, but the plugin
 301                  * does not recognize plugin data.
 302                  */
 303                 err = EINVAL;
 304                 goto fail;
 305         }
 306 
 307         /*
 308          * Register the private properties.
 309          */
 310         mac_register_priv_prop(mip, mregp->m_priv_props);
 311 
 312         /*
 313          * Stash the driver callbacks into the mac_impl_t, but first sanity
 314          * check to make sure all mandatory callbacks are set.
 315          */
 316         if (mregp->m_callbacks->mc_getstat == NULL ||
 317             mregp->m_callbacks->mc_start == NULL ||
 318             mregp->m_callbacks->mc_stop == NULL ||
 319             mregp->m_callbacks->mc_setpromisc == NULL ||
 320             mregp->m_callbacks->mc_multicst == NULL) {
 321                 goto fail;
 322         }
 323         mip->mi_callbacks = mregp->m_callbacks;
 324 
 325         if (mac_capab_get((mac_handle_t)mip, MAC_CAPAB_LEGACY,
 326             &mip->mi_capab_legacy)) {
 327                 mip->mi_state_flags |= MIS_LEGACY;
 328                 mip->mi_phy_dev = mip->mi_capab_legacy.ml_dev;
 329         } else {
 330                 mip->mi_phy_dev = makedevice(ddi_driver_major(mip->mi_dip),
 331                     mip->mi_minor);
 332         }
 333 
 334         /*
 335          * Allocate a notification thread. thread_create blocks for memory
 336          * if needed, it never fails.
 337          */
 338         mip->mi_notify_thread = thread_create(NULL, 0, i_mac_notify_thread,
 339             mip, 0, &p0, TS_RUN, minclsyspri);
 340 
 341         /*
 342          * Initialize the capabilities
 343          */
 344 
 345         bzero(&mip->mi_rx_rings_cap, sizeof (mac_capab_rings_t));
 346         bzero(&mip->mi_tx_rings_cap, sizeof (mac_capab_rings_t));
 347 
 348         if (i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, NULL))
 349                 mip->mi_state_flags |= MIS_IS_VNIC;
 350 
 351         if (i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_AGGR, NULL))
 352                 mip->mi_state_flags |= MIS_IS_AGGR;
 353 
 354         mac_addr_factory_init(mip);
 355 
 356         /*
 357          * Enforce the virtrualization level registered.
 358          */
 359         if (mip->mi_v12n_level & MAC_VIRT_LEVEL1) {
 360                 if (mac_init_rings(mip, MAC_RING_TYPE_RX) != 0 ||
 361                     mac_init_rings(mip, MAC_RING_TYPE_TX) != 0)
 362                         goto fail;
 363 
 364                 /*
 365                  * The driver needs to register at least rx rings for this
 366                  * virtualization level.
 367                  */
 368                 if (mip->mi_rx_groups == NULL)
 369                         goto fail;
 370         }
 371 
 372         /*
 373          * The driver must set mc_unicst entry point to NULL when it advertises
 374          * CAP_RINGS for rx groups.
 375          */
 376         if (mip->mi_rx_groups != NULL) {
 377                 if (mregp->m_callbacks->mc_unicst != NULL)
 378                         goto fail;
 379         } else {
 380                 if (mregp->m_callbacks->mc_unicst == NULL)
 381                         goto fail;
 382         }
 383 
 384         /*
 385          * Initialize MAC addresses. Must be called after mac_init_rings().
 386          */
 387         mac_init_macaddr(mip);
 388 
 389         mip->mi_share_capab.ms_snum = 0;
 390         if (mip->mi_v12n_level & MAC_VIRT_HIO) {
 391                 (void) mac_capab_get((mac_handle_t)mip, MAC_CAPAB_SHARES,
 392                     &mip->mi_share_capab);
 393         }
 394 
 395         /*
 396          * Initialize the kstats for this device.
 397          */
 398         mac_driver_stat_create(mip);
 399 
 400         /* Zero out any properties. */
 401         bzero(&mip->mi_resource_props, sizeof (mac_resource_props_t));
 402 
 403         if (mip->mi_minor <= MAC_MAX_MINOR) {
 404                 /* Create a style-2 DLPI device */
 405                 if (ddi_create_minor_node(mip->mi_dip, driver, S_IFCHR, 0,
 406                     DDI_NT_NET, CLONE_DEV) != DDI_SUCCESS)
 407                         goto fail;
 408                 style2_created = B_TRUE;
 409 
 410                 /* Create a style-1 DLPI device */
 411                 if (ddi_create_minor_node(mip->mi_dip, mip->mi_name, S_IFCHR,
 412                     mip->mi_minor, DDI_NT_NET, 0) != DDI_SUCCESS)
 413                         goto fail;
 414                 style1_created = B_TRUE;
 415         }
 416 
 417         mac_flow_l2tab_create(mip, &mip->mi_flow_tab);
 418 
 419         rw_enter(&i_mac_impl_lock, RW_WRITER);
 420         if (mod_hash_insert(i_mac_impl_hash,
 421             (mod_hash_key_t)mip->mi_name, (mod_hash_val_t)mip) != 0) {
 422                 rw_exit(&i_mac_impl_lock);
 423                 err = EEXIST;
 424                 goto fail;
 425         }
 426 
 427         DTRACE_PROBE2(mac__register, struct devnames *, dnp,
 428             (mac_impl_t *), mip);
 429 
 430         /*
 431          * Mark the MAC to be ready for open.
 432          */
 433         mip->mi_state_flags &= ~MIS_DISABLED;
 434         rw_exit(&i_mac_impl_lock);
 435 
 436         atomic_inc_32(&i_mac_impl_count);
 437 
 438         cmn_err(CE_NOTE, "!%s registered", mip->mi_name);
 439         *mhp = (mac_handle_t)mip;
 440         return (0);
 441 
 442 fail:
 443         if (style1_created)
 444                 ddi_remove_minor_node(mip->mi_dip, mip->mi_name);
 445 
 446         if (style2_created)
 447                 ddi_remove_minor_node(mip->mi_dip, driver);
 448 
 449         mac_addr_factory_fini(mip);
 450 
 451         /* Clean up registered MAC addresses */
 452         mac_fini_macaddr(mip);
 453 
 454         /* Clean up registered rings */
 455         mac_free_rings(mip, MAC_RING_TYPE_RX);
 456         mac_free_rings(mip, MAC_RING_TYPE_TX);
 457 
 458         /* Clean up notification thread */
 459         if (mip->mi_notify_thread != NULL)
 460                 i_mac_notify_exit(mip);
 461 
 462         if (mip->mi_info.mi_unicst_addr != NULL) {
 463                 kmem_free(mip->mi_info.mi_unicst_addr,
 464                     mip->mi_type->mt_addr_length);
 465                 mip->mi_info.mi_unicst_addr = NULL;
 466         }
 467 
 468         mac_driver_stat_delete(mip);
 469 
 470         if (mip->mi_type != NULL) {
 471                 atomic_dec_32(&mip->mi_type->mt_ref);
 472                 mip->mi_type = NULL;
 473         }
 474 
 475         if (mip->mi_pdata != NULL) {
 476                 kmem_free(mip->mi_pdata, mip->mi_pdata_size);
 477                 mip->mi_pdata = NULL;
 478                 mip->mi_pdata_size = 0;
 479         }
 480 
 481         if (minor != 0) {
 482                 ASSERT(minor > MAC_MAX_MINOR);
 483                 mac_minor_rele(minor);
 484         }
 485 
 486         mip->mi_state_flags = 0;
 487         mac_unregister_priv_prop(mip);
 488 
 489         /*
 490          * Clear the state before destroying the mac_impl_t
 491          */
 492         mip->mi_state_flags = 0;
 493 
 494         kmem_cache_free(i_mac_impl_cachep, mip);
 495         return (err);
 496 }
 497 
 498 /*
 499  * Unregister from the GLDv3 framework
 500  */
 501 int
 502 mac_unregister(mac_handle_t mh)
 503 {
 504         int                     err;
 505         mac_impl_t              *mip = (mac_impl_t *)mh;
 506         mod_hash_val_t          val;
 507         mac_margin_req_t        *mmr, *nextmmr;
 508 
 509         /* Fail the unregister if there are any open references to this mac. */
 510         if ((err = mac_disable_nowait(mh)) != 0)
 511                 return (err);
 512 
 513         /*
 514          * Clean up notification thread and wait for it to exit.
 515          */
 516         i_mac_notify_exit(mip);
 517 
 518         /*
 519          * Prior to acquiring the MAC perimeter, remove the MAC instance from
 520          * the internal hash table. Such removal means table-walkers that
 521          * acquire the perimeter will not do so on behalf of what we are
 522          * unregistering, which prevents a deadlock.
 523          */
 524         rw_enter(&i_mac_impl_lock, RW_WRITER);
 525         (void) mod_hash_remove(i_mac_impl_hash,
 526             (mod_hash_key_t)mip->mi_name, &val);
 527         rw_exit(&i_mac_impl_lock);
 528         ASSERT(mip == (mac_impl_t *)val);
 529 
 530         i_mac_perim_enter(mip);
 531 
 532         /*
 533          * There is still resource properties configured over this mac.
 534          */
 535         if (mip->mi_resource_props.mrp_mask != 0)
 536                 mac_fastpath_enable((mac_handle_t)mip);
 537 
 538         if (mip->mi_minor < MAC_MAX_MINOR + 1) {
 539                 ddi_remove_minor_node(mip->mi_dip, mip->mi_name);
 540                 ddi_remove_minor_node(mip->mi_dip,
 541                     (char *)ddi_driver_name(mip->mi_dip));
 542         }
 543 
 544         ASSERT(mip->mi_nactiveclients == 0 && !(mip->mi_state_flags &
 545             MIS_EXCLUSIVE));
 546 
 547         mac_driver_stat_delete(mip);
 548 
 549         ASSERT(i_mac_impl_count > 0);
 550         atomic_dec_32(&i_mac_impl_count);
 551 
 552         if (mip->mi_pdata != NULL)
 553                 kmem_free(mip->mi_pdata, mip->mi_pdata_size);
 554         mip->mi_pdata = NULL;
 555         mip->mi_pdata_size = 0;
 556 
 557         /*
 558          * Free the list of margin request.
 559          */
 560         for (mmr = mip->mi_mmrp; mmr != NULL; mmr = nextmmr) {
 561                 nextmmr = mmr->mmr_nextp;
 562                 kmem_free(mmr, sizeof (mac_margin_req_t));
 563         }
 564         mip->mi_mmrp = NULL;
 565 
 566         mip->mi_linkstate = mip->mi_lowlinkstate = LINK_STATE_UNKNOWN;
 567         kmem_free(mip->mi_info.mi_unicst_addr, mip->mi_type->mt_addr_length);
 568         mip->mi_info.mi_unicst_addr = NULL;
 569 
 570         atomic_dec_32(&mip->mi_type->mt_ref);
 571         mip->mi_type = NULL;
 572 
 573         /*
 574          * Free the primary MAC address.
 575          */
 576         mac_fini_macaddr(mip);
 577 
 578         /*
 579          * free all rings
 580          */
 581         mac_free_rings(mip, MAC_RING_TYPE_RX);
 582         mac_free_rings(mip, MAC_RING_TYPE_TX);
 583 
 584         mac_addr_factory_fini(mip);
 585 
 586         bzero(mip->mi_addr, MAXMACADDRLEN);
 587         bzero(mip->mi_dstaddr, MAXMACADDRLEN);
 588         mip->mi_dstaddr_set = B_FALSE;
 589 
 590         /* and the flows */
 591         mac_flow_tab_destroy(mip->mi_flow_tab);
 592         mip->mi_flow_tab = NULL;
 593 
 594         if (mip->mi_minor > MAC_MAX_MINOR)
 595                 mac_minor_rele(mip->mi_minor);
 596 
 597         cmn_err(CE_NOTE, "!%s unregistered", mip->mi_name);
 598 
 599         /*
 600          * Reset the perim related fields to default values before
 601          * kmem_cache_free
 602          */
 603         i_mac_perim_exit(mip);
 604         mip->mi_state_flags = 0;
 605 
 606         mac_unregister_priv_prop(mip);
 607 
 608         ASSERT(mip->mi_bridge_link == NULL);
 609         kmem_cache_free(i_mac_impl_cachep, mip);
 610 
 611         return (0);
 612 }
 613 
 614 /* DATA RECEPTION */
 615 
 616 /*
 617  * This function is invoked for packets received by the MAC driver in
 618  * interrupt context. The ring generation number provided by the driver
 619  * is matched with the ring generation number held in MAC. If they do not
 620  * match, received packets are considered stale packets coming from an older
 621  * assignment of the ring. Drop them.
 622  */
 623 void
 624 mac_rx_ring(mac_handle_t mh, mac_ring_handle_t mrh, mblk_t *mp_chain,
 625     uint64_t mr_gen_num)
 626 {
 627         mac_ring_t              *mr = (mac_ring_t *)mrh;
 628 
 629         if ((mr != NULL) && (mr->mr_gen_num != mr_gen_num)) {
 630                 DTRACE_PROBE2(mac__rx__rings__stale__packet, uint64_t,
 631                     mr->mr_gen_num, uint64_t, mr_gen_num);
 632                 freemsgchain(mp_chain);
 633                 return;
 634         }
 635         mac_rx(mh, (mac_resource_handle_t)mrh, mp_chain);
 636 }
 637 
 638 /*
 639  * This function is invoked for each packet received by the underlying driver.
 640  */
 641 void
 642 mac_rx(mac_handle_t mh, mac_resource_handle_t mrh, mblk_t *mp_chain)
 643 {
 644         mac_impl_t *mip = (mac_impl_t *)mh;
 645 
 646         /*
 647          * Check if the link is part of a bridge.  If not, then we don't need
 648          * to take the lock to remain consistent.  Make this common case
 649          * lock-free and tail-call optimized.
 650          */
 651         if (mip->mi_bridge_link == NULL) {
 652                 mac_rx_common(mh, mrh, mp_chain);
 653         } else {
 654                 /*
 655                  * Once we take a reference on the bridge link, the bridge
 656                  * module itself can't unload, so the callback pointers are
 657                  * stable.
 658                  */
 659                 mutex_enter(&mip->mi_bridge_lock);
 660                 if ((mh = mip->mi_bridge_link) != NULL)
 661                         mac_bridge_ref_cb(mh, B_TRUE);
 662                 mutex_exit(&mip->mi_bridge_lock);
 663                 if (mh == NULL) {
 664                         mac_rx_common((mac_handle_t)mip, mrh, mp_chain);
 665                 } else {
 666                         mac_bridge_rx_cb(mh, mrh, mp_chain);
 667                         mac_bridge_ref_cb(mh, B_FALSE);
 668                 }
 669         }
 670 }
 671 
 672 /*
 673  * Special case function: this allows snooping of packets transmitted and
 674  * received by TRILL. By design, they go directly into the TRILL module.
 675  */
 676 void
 677 mac_trill_snoop(mac_handle_t mh, mblk_t *mp)
 678 {
 679         mac_impl_t *mip = (mac_impl_t *)mh;
 680 
 681         if (mip->mi_promisc_list != NULL)
 682                 mac_promisc_dispatch(mip, mp, NULL);
 683 }
 684 
 685 /*
 686  * This is the upward reentry point for packets arriving from the bridging
 687  * module and from mac_rx for links not part of a bridge.
 688  */
 689 void
 690 mac_rx_common(mac_handle_t mh, mac_resource_handle_t mrh, mblk_t *mp_chain)
 691 {
 692         mac_impl_t              *mip = (mac_impl_t *)mh;
 693         mac_ring_t              *mr = (mac_ring_t *)mrh;
 694         mac_soft_ring_set_t     *mac_srs;
 695         mblk_t                  *bp = mp_chain;
 696         boolean_t               hw_classified = B_FALSE;
 697 
 698         /*
 699          * If there are any promiscuous mode callbacks defined for
 700          * this MAC, pass them a copy if appropriate.
 701          */
 702         if (mip->mi_promisc_list != NULL)
 703                 mac_promisc_dispatch(mip, mp_chain, NULL);
 704 
 705         if (mr != NULL) {
 706                 /*
 707                  * If the SRS teardown has started, just return. The 'mr'
 708                  * continues to be valid until the driver unregisters the mac.
 709                  * Hardware classified packets will not make their way up
 710                  * beyond this point once the teardown has started. The driver
 711                  * is never passed a pointer to a flow entry or SRS or any
 712                  * structure that can be freed much before mac_unregister.
 713                  */
 714                 mutex_enter(&mr->mr_lock);
 715                 if ((mr->mr_state != MR_INUSE) || (mr->mr_flag &
 716                     (MR_INCIPIENT | MR_CONDEMNED | MR_QUIESCE))) {
 717                         mutex_exit(&mr->mr_lock);
 718                         freemsgchain(mp_chain);
 719                         return;
 720                 }
 721                 if (mr->mr_classify_type == MAC_HW_CLASSIFIER) {
 722                         hw_classified = B_TRUE;
 723                         MR_REFHOLD_LOCKED(mr);
 724                 }
 725                 mutex_exit(&mr->mr_lock);
 726 
 727                 /*
 728                  * We check if an SRS is controlling this ring.
 729                  * If so, we can directly call the srs_lower_proc
 730                  * routine otherwise we need to go through mac_rx_classify
 731                  * to reach the right place.
 732                  */
 733                 if (hw_classified) {
 734                         mac_srs = mr->mr_srs;
 735                         /*
 736                          * This is supposed to be the fast path.
 737                          * All packets received though here were steered by
 738                          * the hardware classifier, and share the same
 739                          * MAC header info.
 740                          */
 741                         mac_srs->srs_rx.sr_lower_proc(mh,
 742                             (mac_resource_handle_t)mac_srs, mp_chain, B_FALSE);
 743                         MR_REFRELE(mr);
 744                         return;
 745                 }
 746                 /* We'll fall through to software classification */
 747         } else {
 748                 flow_entry_t *flent;
 749                 int err;
 750 
 751                 rw_enter(&mip->mi_rw_lock, RW_READER);
 752                 if (mip->mi_single_active_client != NULL) {
 753                         flent = mip->mi_single_active_client->mci_flent_list;
 754                         FLOW_TRY_REFHOLD(flent, err);
 755                         rw_exit(&mip->mi_rw_lock);
 756                         if (err == 0) {
 757                                 (flent->fe_cb_fn)(flent->fe_cb_arg1,
 758                                     flent->fe_cb_arg2, mp_chain, B_FALSE);
 759                                 FLOW_REFRELE(flent);
 760                                 return;
 761                         }
 762                 } else {
 763                         rw_exit(&mip->mi_rw_lock);
 764                 }
 765         }
 766 
 767         if (!FLOW_TAB_EMPTY(mip->mi_flow_tab)) {
 768                 if ((bp = mac_rx_flow(mh, mrh, bp)) == NULL)
 769                         return;
 770         }
 771 
 772         freemsgchain(bp);
 773 }
 774 
 775 /* DATA TRANSMISSION */
 776 
 777 /*
 778  * A driver's notification to resume transmission, in case of a provider
 779  * without TX rings.
 780  */
 781 void
 782 mac_tx_update(mac_handle_t mh)
 783 {
 784         mac_tx_ring_update(mh, NULL);
 785 }
 786 
 787 /*
 788  * A driver's notification to resume transmission on the specified TX ring.
 789  */
 790 void
 791 mac_tx_ring_update(mac_handle_t mh, mac_ring_handle_t rh)
 792 {
 793         i_mac_tx_srs_notify((mac_impl_t *)mh, rh);
 794 }
 795 
 796 /* LINK STATE */
 797 /*
 798  * Notify the MAC layer about a link state change
 799  */
 800 void
 801 mac_link_update(mac_handle_t mh, link_state_t link)
 802 {
 803         mac_impl_t      *mip = (mac_impl_t *)mh;
 804 
 805         /*
 806          * Save the link state.
 807          */
 808         mip->mi_lowlinkstate = link;
 809 
 810         /*
 811          * Send a MAC_NOTE_LOWLINK notification.  This tells the notification
 812          * thread to deliver both lower and upper notifications.
 813          */
 814         i_mac_notify(mip, MAC_NOTE_LOWLINK);
 815 }
 816 
 817 /*
 818  * Notify the MAC layer about a link state change due to bridging.
 819  */
 820 void
 821 mac_link_redo(mac_handle_t mh, link_state_t link)
 822 {
 823         mac_impl_t      *mip = (mac_impl_t *)mh;
 824 
 825         /*
 826          * Save the link state.
 827          */
 828         mip->mi_linkstate = link;
 829 
 830         /*
 831          * Send a MAC_NOTE_LINK notification.  Only upper notifications are
 832          * made.
 833          */
 834         i_mac_notify(mip, MAC_NOTE_LINK);
 835 }
 836 
 837 /* MINOR NODE HANDLING */
 838 
 839 /*
 840  * Given a dev_t, return the instance number (PPA) associated with it.
 841  * Drivers can use this in their getinfo(9e) implementation to lookup
 842  * the instance number (i.e. PPA) of the device, to use as an index to
 843  * their own array of soft state structures.
 844  *
 845  * Returns -1 on error.
 846  */
 847 int
 848 mac_devt_to_instance(dev_t devt)
 849 {
 850         return (dld_devt_to_instance(devt));
 851 }
 852 
 853 /*
 854  * This function returns the first minor number that is available for
 855  * driver private use.  All minor numbers smaller than this are
 856  * reserved for GLDv3 use.
 857  */
 858 minor_t
 859 mac_private_minor(void)
 860 {
 861         return (MAC_PRIVATE_MINOR);
 862 }
 863 
 864 /* OTHER CONTROL INFORMATION */
 865 
 866 /*
 867  * A driver notified us that its primary MAC address has changed.
 868  */
 869 void
 870 mac_unicst_update(mac_handle_t mh, const uint8_t *addr)
 871 {
 872         mac_impl_t      *mip = (mac_impl_t *)mh;
 873 
 874         if (mip->mi_type->mt_addr_length == 0)
 875                 return;
 876 
 877         i_mac_perim_enter(mip);
 878 
 879         /*
 880          * If address changes, freshen the MAC address value and update
 881          * all MAC clients that share this MAC address.
 882          */
 883         if (bcmp(addr, mip->mi_addr, mip->mi_type->mt_addr_length) != 0) {
 884                 mac_freshen_macaddr(mac_find_macaddr(mip, mip->mi_addr),
 885                     (uint8_t *)addr);
 886         }
 887 
 888         i_mac_perim_exit(mip);
 889 
 890         /*
 891          * Send a MAC_NOTE_UNICST notification.
 892          */
 893         i_mac_notify(mip, MAC_NOTE_UNICST);
 894 }
 895 
 896 void
 897 mac_dst_update(mac_handle_t mh, const uint8_t *addr)
 898 {
 899         mac_impl_t      *mip = (mac_impl_t *)mh;
 900 
 901         if (mip->mi_type->mt_addr_length == 0)
 902                 return;
 903 
 904         i_mac_perim_enter(mip);
 905         bcopy(addr, mip->mi_dstaddr, mip->mi_type->mt_addr_length);
 906         i_mac_perim_exit(mip);
 907         i_mac_notify(mip, MAC_NOTE_DEST);
 908 }
 909 
 910 /*
 911  * MAC plugin information changed.
 912  */
 913 int
 914 mac_pdata_update(mac_handle_t mh, void *mac_pdata, size_t dsize)
 915 {
 916         mac_impl_t      *mip = (mac_impl_t *)mh;
 917 
 918         /*
 919          * Verify that the plugin supports MAC plugin data and that the
 920          * supplied data is valid.
 921          */
 922         if (!(mip->mi_type->mt_ops.mtops_ops & MTOPS_PDATA_VERIFY))
 923                 return (EINVAL);
 924         if (!mip->mi_type->mt_ops.mtops_pdata_verify(mac_pdata, dsize))
 925                 return (EINVAL);
 926 
 927         if (mip->mi_pdata != NULL)
 928                 kmem_free(mip->mi_pdata, mip->mi_pdata_size);
 929 
 930         mip->mi_pdata = kmem_alloc(dsize, KM_SLEEP);
 931         bcopy(mac_pdata, mip->mi_pdata, dsize);
 932         mip->mi_pdata_size = dsize;
 933 
 934         /*
 935          * Since the MAC plugin data is used to construct MAC headers that
 936          * were cached in fast-path headers, we need to flush fast-path
 937          * information for links associated with this mac.
 938          */
 939         i_mac_notify(mip, MAC_NOTE_FASTPATH_FLUSH);
 940         return (0);
 941 }
 942 
 943 /*
 944  * Invoked by driver as well as the framework to notify its capability change.
 945  */
 946 void
 947 mac_capab_update(mac_handle_t mh)
 948 {
 949         /* Send MAC_NOTE_CAPAB_CHG notification */
 950         i_mac_notify((mac_impl_t *)mh, MAC_NOTE_CAPAB_CHG);
 951 }
 952 
 953 /*
 954  * Used by normal drivers to update the max sdu size.
 955  * We need to handle the case of a smaller mi_sdu_multicast
 956  * since this is called by mac_set_mtu() even for drivers that
 957  * have differing unicast and multicast mtu and we don't want to
 958  * increase the multicast mtu by accident in that case.
 959  */
 960 int
 961 mac_maxsdu_update(mac_handle_t mh, uint_t sdu_max)
 962 {
 963         mac_impl_t      *mip = (mac_impl_t *)mh;
 964 
 965         if (sdu_max == 0 || sdu_max < mip->mi_sdu_min)
 966                 return (EINVAL);
 967         mip->mi_sdu_max = sdu_max;
 968         if (mip->mi_sdu_multicast > mip->mi_sdu_max)
 969                 mip->mi_sdu_multicast = mip->mi_sdu_max;
 970 
 971         /* Send a MAC_NOTE_SDU_SIZE notification. */
 972         i_mac_notify(mip, MAC_NOTE_SDU_SIZE);
 973         return (0);
 974 }
 975 
 976 /*
 977  * Version of the above function that is used by drivers that have a different
 978  * max sdu size for multicast/broadcast vs. unicast.
 979  */
 980 int
 981 mac_maxsdu_update2(mac_handle_t mh, uint_t sdu_max, uint_t sdu_multicast)
 982 {
 983         mac_impl_t      *mip = (mac_impl_t *)mh;
 984 
 985         if (sdu_max == 0 || sdu_max < mip->mi_sdu_min)
 986                 return (EINVAL);
 987         if (sdu_multicast == 0)
 988                 sdu_multicast = sdu_max;
 989         if (sdu_multicast > sdu_max || sdu_multicast < mip->mi_sdu_min)
 990                 return (EINVAL);
 991         mip->mi_sdu_max = sdu_max;
 992         mip->mi_sdu_multicast = sdu_multicast;
 993 
 994         /* Send a MAC_NOTE_SDU_SIZE notification. */
 995         i_mac_notify(mip, MAC_NOTE_SDU_SIZE);
 996         return (0);
 997 }
 998 
 999 static void
1000 mac_ring_intr_retarget(mac_group_t *group, mac_ring_t *ring)
1001 {
1002         mac_client_impl_t *mcip;
1003         flow_entry_t *flent;
1004         mac_soft_ring_set_t *mac_rx_srs;
1005         mac_cpus_t *srs_cpu;
1006         int i;
1007 
1008         if (((mcip = MAC_GROUP_ONLY_CLIENT(group)) != NULL) &&
1009             (!ring->mr_info.mri_intr.mi_ddi_shared)) {
1010                 /* interrupt can be re-targeted */
1011                 ASSERT(group->mrg_state == MAC_GROUP_STATE_RESERVED);
1012                 flent = mcip->mci_flent;
1013                 if (ring->mr_type == MAC_RING_TYPE_RX) {
1014                         for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
1015                                 mac_rx_srs = flent->fe_rx_srs[i];
1016                                 if (mac_rx_srs->srs_ring != ring)
1017                                         continue;
1018                                 srs_cpu = &mac_rx_srs->srs_cpu;
1019                                 mutex_enter(&cpu_lock);
1020                                 mac_rx_srs_retarget_intr(mac_rx_srs,
1021                                     srs_cpu->mc_rx_intr_cpu);
1022                                 mutex_exit(&cpu_lock);
1023                                 break;
1024                         }
1025                 } else {
1026                         if (flent->fe_tx_srs != NULL) {
1027                                 mutex_enter(&cpu_lock);
1028                                 mac_tx_srs_retarget_intr(
1029                                     flent->fe_tx_srs);
1030                                 mutex_exit(&cpu_lock);
1031                         }
1032                 }
1033         }
1034 }
1035 
1036 /*
1037  * Clients like aggr create pseudo rings (mac_ring_t) and expose them to
1038  * their clients. There is a 1-1 mapping pseudo ring and the hardware
1039  * ring. ddi interrupt handles are exported from the hardware ring to
1040  * the pseudo ring. Thus when the interrupt handle changes, clients of
1041  * aggr that are using the handle need to use the new handle and
1042  * re-target their interrupts.
1043  */
1044 static void
1045 mac_pseudo_ring_intr_retarget(mac_impl_t *mip, mac_ring_t *ring,
1046     ddi_intr_handle_t ddh)
1047 {
1048         mac_ring_t *pring;
1049         mac_group_t *pgroup;
1050         mac_impl_t *pmip;
1051         char macname[MAXNAMELEN];
1052         mac_perim_handle_t p_mph;
1053         uint64_t saved_gen_num;
1054 
1055 again:
1056         pring = (mac_ring_t *)ring->mr_prh;
1057         pgroup = (mac_group_t *)pring->mr_gh;
1058         pmip = (mac_impl_t *)pgroup->mrg_mh;
1059         saved_gen_num = ring->mr_gen_num;
1060         (void) strlcpy(macname, pmip->mi_name, MAXNAMELEN);
1061         /*
1062          * We need to enter aggr's perimeter. The locking hierarchy
1063          * dictates that aggr's perimeter should be entered first
1064          * and then the port's perimeter. So drop the port's
1065          * perimeter, enter aggr's and then re-enter port's
1066          * perimeter.
1067          */
1068         i_mac_perim_exit(mip);
1069         /*
1070          * While we know pmip is the aggr's mip, there is a
1071          * possibility that aggr could have unregistered by
1072          * the time we exit port's perimeter (mip) and
1073          * enter aggr's perimeter (pmip). To avoid that
1074          * scenario, enter aggr's perimeter using its name.
1075          */
1076         if (mac_perim_enter_by_macname(macname, &p_mph) != 0)
1077                 return;
1078         i_mac_perim_enter(mip);
1079         /*
1080          * Check if the ring got assigned to another aggregation before
1081          * be could enter aggr's and the port's perimeter. When a ring
1082          * gets deleted from an aggregation, it calls mac_stop_ring()
1083          * which increments the generation number. So checking
1084          * generation number will be enough.
1085          */
1086         if (ring->mr_gen_num != saved_gen_num && ring->mr_prh != NULL) {
1087                 i_mac_perim_exit(mip);
1088                 mac_perim_exit(p_mph);
1089                 i_mac_perim_enter(mip);
1090                 goto again;
1091         }
1092 
1093         /* Check if pseudo ring is still present */
1094         if (ring->mr_prh != NULL) {
1095                 pring->mr_info.mri_intr.mi_ddi_handle = ddh;
1096                 pring->mr_info.mri_intr.mi_ddi_shared =
1097                     ring->mr_info.mri_intr.mi_ddi_shared;
1098                 if (ddh != NULL)
1099                         mac_ring_intr_retarget(pgroup, pring);
1100         }
1101         i_mac_perim_exit(mip);
1102         mac_perim_exit(p_mph);
1103 }
1104 /*
1105  * API called by driver to provide new interrupt handle for TX/RX rings.
1106  * This usually happens when IRM (Interrupt Resource Manangement)
1107  * framework either gives the driver more MSI-x interrupts or takes
1108  * away MSI-x interrupts from the driver.
1109  */
1110 void
1111 mac_ring_intr_set(mac_ring_handle_t mrh, ddi_intr_handle_t ddh)
1112 {
1113         mac_ring_t      *ring = (mac_ring_t *)mrh;
1114         mac_group_t     *group = (mac_group_t *)ring->mr_gh;
1115         mac_impl_t      *mip = (mac_impl_t *)group->mrg_mh;
1116 
1117         i_mac_perim_enter(mip);
1118         ring->mr_info.mri_intr.mi_ddi_handle = ddh;
1119         if (ddh == NULL) {
1120                 /* Interrupts being reset */
1121                 ring->mr_info.mri_intr.mi_ddi_shared = B_FALSE;
1122                 if (ring->mr_prh != NULL) {
1123                         mac_pseudo_ring_intr_retarget(mip, ring, ddh);
1124                         return;
1125                 }
1126         } else {
1127                 /* New interrupt handle */
1128                 mac_compare_ddi_handle(mip->mi_rx_groups,
1129                     mip->mi_rx_group_count, ring);
1130                 if (!ring->mr_info.mri_intr.mi_ddi_shared) {
1131                         mac_compare_ddi_handle(mip->mi_tx_groups,
1132                             mip->mi_tx_group_count, ring);
1133                 }
1134                 if (ring->mr_prh != NULL) {
1135                         mac_pseudo_ring_intr_retarget(mip, ring, ddh);
1136                         return;
1137                 } else {
1138                         mac_ring_intr_retarget(group, ring);
1139                 }
1140         }
1141         i_mac_perim_exit(mip);
1142 }
1143 
1144 /* PRIVATE FUNCTIONS, FOR INTERNAL USE ONLY */
1145 
1146 /*
1147  * Updates the mac_impl structure with the current state of the link
1148  */
1149 static void
1150 i_mac_log_link_state(mac_impl_t *mip)
1151 {
1152         /*
1153          * If no change, then it is not interesting.
1154          */
1155         if (mip->mi_lastlowlinkstate == mip->mi_lowlinkstate)
1156                 return;
1157 
1158         switch (mip->mi_lowlinkstate) {
1159         case LINK_STATE_UP:
1160                 if (mip->mi_type->mt_ops.mtops_ops & MTOPS_LINK_DETAILS) {
1161                         char det[200];
1162 
1163                         mip->mi_type->mt_ops.mtops_link_details(det,
1164                             sizeof (det), (mac_handle_t)mip, mip->mi_pdata);
1165 
1166                         cmn_err(CE_NOTE, "!%s link up, %s", mip->mi_name, det);
1167                 } else {
1168                         cmn_err(CE_NOTE, "!%s link up", mip->mi_name);
1169                 }
1170                 break;
1171 
1172         case LINK_STATE_DOWN:
1173                 /*
1174                  * Only transitions from UP to DOWN are interesting
1175                  */
1176                 if (mip->mi_lastlowlinkstate != LINK_STATE_UNKNOWN)
1177                         cmn_err(CE_NOTE, "!%s link down", mip->mi_name);
1178                 break;
1179 
1180         case LINK_STATE_UNKNOWN:
1181                 /*
1182                  * This case is normally not interesting.
1183                  */
1184                 break;
1185         }
1186         mip->mi_lastlowlinkstate = mip->mi_lowlinkstate;
1187 }
1188 
1189 /*
1190  * Main routine for the callbacks notifications thread
1191  */
1192 static void
1193 i_mac_notify_thread(void *arg)
1194 {
1195         mac_impl_t      *mip = arg;
1196         callb_cpr_t     cprinfo;
1197         mac_cb_t        *mcb;
1198         mac_cb_info_t   *mcbi;
1199         mac_notify_cb_t *mncb;
1200 
1201         mcbi = &mip->mi_notify_cb_info;
1202         CALLB_CPR_INIT(&cprinfo, mcbi->mcbi_lockp, callb_generic_cpr,
1203             "i_mac_notify_thread");
1204 
1205         mutex_enter(mcbi->mcbi_lockp);
1206 
1207         for (;;) {
1208                 uint32_t        bits;
1209                 uint32_t        type;
1210 
1211                 bits = mip->mi_notify_bits;
1212                 if (bits == 0) {
1213                         CALLB_CPR_SAFE_BEGIN(&cprinfo);
1214                         cv_wait(&mcbi->mcbi_cv, mcbi->mcbi_lockp);
1215                         CALLB_CPR_SAFE_END(&cprinfo, mcbi->mcbi_lockp);
1216                         continue;
1217                 }
1218                 mip->mi_notify_bits = 0;
1219                 if ((bits & (1 << MAC_NNOTE)) != 0) {
1220                         /* request to quit */
1221                         ASSERT(mip->mi_state_flags & MIS_DISABLED);
1222                         break;
1223                 }
1224 
1225                 mutex_exit(mcbi->mcbi_lockp);
1226 
1227                 /*
1228                  * Log link changes on the actual link, but then do reports on
1229                  * synthetic state (if part of a bridge).
1230                  */
1231                 if ((bits & (1 << MAC_NOTE_LOWLINK)) != 0) {
1232                         link_state_t newstate;
1233                         mac_handle_t mh;
1234 
1235                         i_mac_log_link_state(mip);
1236                         newstate = mip->mi_lowlinkstate;
1237                         if (mip->mi_bridge_link != NULL) {
1238                                 mutex_enter(&mip->mi_bridge_lock);
1239                                 if ((mh = mip->mi_bridge_link) != NULL) {
1240                                         newstate = mac_bridge_ls_cb(mh,
1241                                             newstate);
1242                                 }
1243                                 mutex_exit(&mip->mi_bridge_lock);
1244                         }
1245                         if (newstate != mip->mi_linkstate) {
1246                                 mip->mi_linkstate = newstate;
1247                                 bits |= 1 << MAC_NOTE_LINK;
1248                         }
1249                 }
1250 
1251                 /*
1252                  * Do notification callbacks for each notification type.
1253                  */
1254                 for (type = 0; type < MAC_NNOTE; type++) {
1255                         if ((bits & (1 << type)) == 0) {
1256                                 continue;
1257                         }
1258 
1259                         if (mac_notify_cb_list[type] != NULL)
1260                                 (*mac_notify_cb_list[type])(mip);
1261 
1262                         /*
1263                          * Walk the list of notifications.
1264                          */
1265                         MAC_CALLBACK_WALKER_INC(&mip->mi_notify_cb_info);
1266                         for (mcb = mip->mi_notify_cb_list; mcb != NULL;
1267                             mcb = mcb->mcb_nextp) {
1268                                 mncb = (mac_notify_cb_t *)mcb->mcb_objp;
1269                                 mncb->mncb_fn(mncb->mncb_arg, type);
1270                         }
1271                         MAC_CALLBACK_WALKER_DCR(&mip->mi_notify_cb_info,
1272                             &mip->mi_notify_cb_list);
1273                 }
1274 
1275                 mutex_enter(mcbi->mcbi_lockp);
1276         }
1277 
1278         mip->mi_state_flags |= MIS_NOTIFY_DONE;
1279         cv_broadcast(&mcbi->mcbi_cv);
1280 
1281         /* CALLB_CPR_EXIT drops the lock */
1282         CALLB_CPR_EXIT(&cprinfo);
1283         thread_exit();
1284 }
1285 
1286 /*
1287  * Signal the i_mac_notify_thread asking it to quit.
1288  * Then wait till it is done.
1289  */
1290 void
1291 i_mac_notify_exit(mac_impl_t *mip)
1292 {
1293         mac_cb_info_t   *mcbi;
1294 
1295         mcbi = &mip->mi_notify_cb_info;
1296 
1297         mutex_enter(mcbi->mcbi_lockp);
1298         mip->mi_notify_bits = (1 << MAC_NNOTE);
1299         cv_broadcast(&mcbi->mcbi_cv);
1300 
1301 
1302         while ((mip->mi_notify_thread != NULL) &&
1303             !(mip->mi_state_flags & MIS_NOTIFY_DONE)) {
1304                 cv_wait(&mcbi->mcbi_cv, mcbi->mcbi_lockp);
1305         }
1306 
1307         /* Necessary clean up before doing kmem_cache_free */
1308         mip->mi_state_flags &= ~MIS_NOTIFY_DONE;
1309         mip->mi_notify_bits = 0;
1310         mip->mi_notify_thread = NULL;
1311         mutex_exit(mcbi->mcbi_lockp);
1312 }
1313 
1314 /*
1315  * Entry point invoked by drivers to dynamically add a ring to an
1316  * existing group.
1317  */
1318 int
1319 mac_group_add_ring(mac_group_handle_t gh, int index)
1320 {
1321         mac_group_t *group = (mac_group_t *)gh;
1322         mac_impl_t *mip = (mac_impl_t *)group->mrg_mh;
1323         int ret;
1324 
1325         i_mac_perim_enter(mip);
1326         ret = i_mac_group_add_ring(group, NULL, index);
1327         i_mac_perim_exit(mip);
1328         return (ret);
1329 }
1330 
1331 /*
1332  * Entry point invoked by drivers to dynamically remove a ring
1333  * from an existing group. The specified ring handle must no longer
1334  * be used by the driver after a call to this function.
1335  */
1336 void
1337 mac_group_rem_ring(mac_group_handle_t gh, mac_ring_handle_t rh)
1338 {
1339         mac_group_t *group = (mac_group_t *)gh;
1340         mac_impl_t *mip = (mac_impl_t *)group->mrg_mh;
1341 
1342         i_mac_perim_enter(mip);
1343         i_mac_group_rem_ring(group, (mac_ring_t *)rh, B_TRUE);
1344         i_mac_perim_exit(mip);
1345 }
1346 
1347 /*
1348  * mac_prop_info_*() callbacks called from the driver's prefix_propinfo()
1349  * entry points.
1350  */
1351 
1352 void
1353 mac_prop_info_set_default_uint8(mac_prop_info_handle_t ph, uint8_t val)
1354 {
1355         mac_prop_info_state_t *pr = (mac_prop_info_state_t *)ph;
1356 
1357         /* nothing to do if the caller doesn't want the default value */
1358         if (pr->pr_default == NULL)
1359                 return;
1360 
1361         ASSERT(pr->pr_default_size >= sizeof (uint8_t));
1362 
1363         *(uint8_t *)(pr->pr_default) = val;
1364         pr->pr_flags |= MAC_PROP_INFO_DEFAULT;
1365 }
1366 
1367 void
1368 mac_prop_info_set_default_uint64(mac_prop_info_handle_t ph, uint64_t val)
1369 {
1370         mac_prop_info_state_t *pr = (mac_prop_info_state_t *)ph;
1371 
1372         /* nothing to do if the caller doesn't want the default value */
1373         if (pr->pr_default == NULL)
1374                 return;
1375 
1376         ASSERT(pr->pr_default_size >= sizeof (uint64_t));
1377 
1378         bcopy(&val, pr->pr_default, sizeof (val));
1379 
1380         pr->pr_flags |= MAC_PROP_INFO_DEFAULT;
1381 }
1382 
1383 void
1384 mac_prop_info_set_default_uint32(mac_prop_info_handle_t ph, uint32_t val)
1385 {
1386         mac_prop_info_state_t *pr = (mac_prop_info_state_t *)ph;
1387 
1388         /* nothing to do if the caller doesn't want the default value */
1389         if (pr->pr_default == NULL)
1390                 return;
1391 
1392         ASSERT(pr->pr_default_size >= sizeof (uint32_t));
1393 
1394         bcopy(&val, pr->pr_default, sizeof (val));
1395 
1396         pr->pr_flags |= MAC_PROP_INFO_DEFAULT;
1397 }
1398 
1399 void
1400 mac_prop_info_set_default_str(mac_prop_info_handle_t ph, const char *str)
1401 {
1402         mac_prop_info_state_t *pr = (mac_prop_info_state_t *)ph;
1403 
1404         /* nothing to do if the caller doesn't want the default value */
1405         if (pr->pr_default == NULL)
1406                 return;
1407 
1408         if (strlen(str) >= pr->pr_default_size)
1409                 pr->pr_errno = ENOBUFS;
1410         else
1411                 (void) strlcpy(pr->pr_default, str, pr->pr_default_size);
1412         pr->pr_flags |= MAC_PROP_INFO_DEFAULT;
1413 }
1414 
1415 void
1416 mac_prop_info_set_default_link_flowctrl(mac_prop_info_handle_t ph,
1417     link_flowctrl_t val)
1418 {
1419         mac_prop_info_state_t *pr = (mac_prop_info_state_t *)ph;
1420 
1421         /* nothing to do if the caller doesn't want the default value */
1422         if (pr->pr_default == NULL)
1423                 return;
1424 
1425         ASSERT(pr->pr_default_size >= sizeof (link_flowctrl_t));
1426 
1427         bcopy(&val, pr->pr_default, sizeof (val));
1428 
1429         pr->pr_flags |= MAC_PROP_INFO_DEFAULT;
1430 }
1431 
1432 void
1433 mac_prop_info_set_range_uint32(mac_prop_info_handle_t ph, uint32_t min,
1434     uint32_t max)
1435 {
1436         mac_prop_info_state_t *pr = (mac_prop_info_state_t *)ph;
1437         mac_propval_range_t *range = pr->pr_range;
1438         mac_propval_uint32_range_t *range32;
1439 
1440         /* nothing to do if the caller doesn't want the range info */
1441         if (range == NULL)
1442                 return;
1443 
1444         if (pr->pr_range_cur_count++ == 0) {
1445                 /* first range */
1446                 pr->pr_flags |= MAC_PROP_INFO_RANGE;
1447                 range->mpr_type = MAC_PROPVAL_UINT32;
1448         } else {
1449                 /* all ranges of a property should be of the same type */
1450                 ASSERT(range->mpr_type == MAC_PROPVAL_UINT32);
1451                 if (pr->pr_range_cur_count > range->mpr_count) {
1452                         pr->pr_errno = ENOSPC;
1453                         return;
1454                 }
1455         }
1456 
1457         range32 = range->mpr_range_uint32;
1458         range32[pr->pr_range_cur_count - 1].mpur_min = min;
1459         range32[pr->pr_range_cur_count - 1].mpur_max = max;
1460 }
1461 
1462 void
1463 mac_prop_info_set_perm(mac_prop_info_handle_t ph, uint8_t perm)
1464 {
1465         mac_prop_info_state_t *pr = (mac_prop_info_state_t *)ph;
1466 
1467         pr->pr_perm = perm;
1468         pr->pr_flags |= MAC_PROP_INFO_PERM;
1469 }
1470 
1471 void mac_hcksum_get(mblk_t *mp, uint32_t *start, uint32_t *stuff,
1472     uint32_t *end, uint32_t *value, uint32_t *flags_ptr)
1473 {
1474         uint32_t flags;
1475 
1476         ASSERT(DB_TYPE(mp) == M_DATA);
1477 
1478         flags = DB_CKSUMFLAGS(mp) & HCK_FLAGS;
1479         if ((flags & (HCK_PARTIALCKSUM | HCK_FULLCKSUM)) != 0) {
1480                 if (value != NULL)
1481                         *value = (uint32_t)DB_CKSUM16(mp);
1482                 if ((flags & HCK_PARTIALCKSUM) != 0) {
1483                         if (start != NULL)
1484                                 *start = (uint32_t)DB_CKSUMSTART(mp);
1485                         if (stuff != NULL)
1486                                 *stuff = (uint32_t)DB_CKSUMSTUFF(mp);
1487                         if (end != NULL)
1488                                 *end = (uint32_t)DB_CKSUMEND(mp);
1489                 }
1490         }
1491 
1492         if (flags_ptr != NULL)
1493                 *flags_ptr = flags;
1494 }
1495 
1496 void mac_hcksum_set(mblk_t *mp, uint32_t start, uint32_t stuff,
1497     uint32_t end, uint32_t value, uint32_t flags)
1498 {
1499         ASSERT(DB_TYPE(mp) == M_DATA);
1500 
1501         DB_CKSUMSTART(mp) = (intptr_t)start;
1502         DB_CKSUMSTUFF(mp) = (intptr_t)stuff;
1503         DB_CKSUMEND(mp) = (intptr_t)end;
1504         DB_CKSUMFLAGS(mp) = (uint16_t)flags;
1505         DB_CKSUM16(mp) = (uint16_t)value;
1506 }
1507 
1508 void
1509 mac_lso_get(mblk_t *mp, uint32_t *mss, uint32_t *flags)
1510 {
1511         ASSERT(DB_TYPE(mp) == M_DATA);
1512 
1513         if (flags != NULL) {
1514                 *flags = DB_CKSUMFLAGS(mp) & HW_LSO;
1515                 if ((*flags != 0) && (mss != NULL))
1516                         *mss = (uint32_t)DB_LSOMSS(mp);
1517         }
1518 }