kernel/net/switchdev/switchdev.c

   1 /*
   2  * net/switchdev/switchdev.c - Switch device API
   3  * Copyright (c) 2014-2015 Jiri Pirko <jiri@resnulli.us>
   4  * Copyright (c) 2014-2015 Scott Feldman <sfeldma@gmail.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  */
  11
  12 #include <linux/kernel.h>
  13 #include <linux/types.h>
  14 #include <linux/init.h>
  15 #include <linux/mutex.h>
  16 #include <linux/notifier.h>
  17 #include <linux/netdevice.h>
  18 #include <linux/etherdevice.h>
  19 #include <linux/if_bridge.h>
  20 #include <linux/list.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/if_vlan.h>
  23 #include <linux/rtnetlink.h>
  24 #include <net/ip_fib.h>
  25 #include <net/switchdev.h>
  26
  27 /**
  28  *      switchdev_trans_item_enqueue - Enqueue data item to transaction queue
  29  *
  30  *      @trans: transaction
  31  *      @data: pointer to data being queued
  32  *      @destructor: data destructor
  33  *      @tritem: transaction item being queued
  34  *
  35  *      Enqeueue data item to transaction queue. tritem is typically placed in
  36  *      cointainter pointed at by data pointer. Destructor is called on
  37  *      transaction abort and after successful commit phase in case
  38  *      the caller did not dequeue the item before.
  39  */
  40 void switchdev_trans_item_enqueue(struct switchdev_trans *trans,
  41                                   void *data, void (*destructor)(void const *),
  42                                   struct switchdev_trans_item *tritem)
  43 {
  44         tritem->data = data;
  45         tritem->destructor = destructor;
  46         list_add_tail(&tritem->list, &trans->item_list);
  47 }
  48 EXPORT_SYMBOL_GPL(switchdev_trans_item_enqueue);
  49
  50 static struct switchdev_trans_item *
  51 __switchdev_trans_item_dequeue(struct switchdev_trans *trans)
  52 {
  53         struct switchdev_trans_item *tritem;
  54
  55         if (list_empty(&trans->item_list))
  56                 return NULL;
  57         tritem = list_first_entry(&trans->item_list,
  58                                   struct switchdev_trans_item, list);
  59         list_del(&tritem->list);
  60         return tritem;
  61 }
  62
  63 /**
  64  *      switchdev_trans_item_dequeue - Dequeue data item from transaction queue
  65  *
  66  *      @trans: transaction
  67  */
  68 void *switchdev_trans_item_dequeue(struct switchdev_trans *trans)
  69 {
  70         struct switchdev_trans_item *tritem;
  71
  72         tritem = __switchdev_trans_item_dequeue(trans);
  73         BUG_ON(!tritem);
  74         return tritem->data;
  75 }
  76 EXPORT_SYMBOL_GPL(switchdev_trans_item_dequeue);
  77
  78 static void switchdev_trans_init(struct switchdev_trans *trans)
  79 {
  80         INIT_LIST_HEAD(&trans->item_list);
  81 }
  82
  83 static void switchdev_trans_items_destroy(struct switchdev_trans *trans)
  84 {
  85         struct switchdev_trans_item *tritem;
  86
  87         while ((tritem = __switchdev_trans_item_dequeue(trans)))
  88                 tritem->destructor(tritem->data);
  89 }
  90
  91 static void switchdev_trans_items_warn_destroy(struct net_device *dev,
  92                                                struct switchdev_trans *trans)
  93 {
  94         WARN(!list_empty(&trans->item_list), "%s: transaction item queue is not empty.\n",
  95              dev->name);
  96         switchdev_trans_items_destroy(trans);
  97 }
  98
  99 static LIST_HEAD(deferred);
 100 static DEFINE_SPINLOCK(deferred_lock);
 101
 102 typedef void switchdev_deferred_func_t(struct net_device *dev,
 103                                        const void *data);
 104
 105 struct switchdev_deferred_item {
 106         struct list_head list;
 107         struct net_device *dev;
 108         switchdev_deferred_func_t *func;
 109         unsigned long data[0];
 110 };
 111
 112 static struct switchdev_deferred_item *switchdev_deferred_dequeue(void)
 113 {
 114         struct switchdev_deferred_item *dfitem;
 115
 116         spin_lock_bh(&deferred_lock);
 117         if (list_empty(&deferred)) {
 118                 dfitem = NULL;
 119                 goto unlock;
 120         }
 121         dfitem = list_first_entry(&deferred,
 122                                   struct switchdev_deferred_item, list);
 123         list_del(&dfitem->list);
 124 unlock:
 125         spin_unlock_bh(&deferred_lock);
 126         return dfitem;
 127 }
 128
 129 /**
 130  *      switchdev_deferred_process - Process ops in deferred queue
 131  *
 132  *      Called to flush the ops currently queued in deferred ops queue.
 133  *      rtnl_lock must be held.
 134  */
 135 void switchdev_deferred_process(void)
 136 {
 137         struct switchdev_deferred_item *dfitem;
 138
 139         ASSERT_RTNL();
 140
 141         while ((dfitem = switchdev_deferred_dequeue())) {
 142                 dfitem->func(dfitem->dev, dfitem->data);
 143                 dev_put(dfitem->dev);
 144                 kfree(dfitem);
 145         }
 146 }
 147 EXPORT_SYMBOL_GPL(switchdev_deferred_process);
 148
 149 static void switchdev_deferred_process_work(struct work_struct *work)
 150 {
 151         rtnl_lock();
 152         switchdev_deferred_process();
 153         rtnl_unlock();
 154 }
 155
 156 static DECLARE_WORK(deferred_process_work, switchdev_deferred_process_work);
 157
 158 static int switchdev_deferred_enqueue(struct net_device *dev,
 159                                       const void *data, size_t data_len,
 160                                       switchdev_deferred_func_t *func)
 161 {
 162         struct switchdev_deferred_item *dfitem;
 163
 164         dfitem = kmalloc(sizeof(*dfitem) + data_len, GFP_ATOMIC);
 165         if (!dfitem)
 166                 return -ENOMEM;
 167         dfitem->dev = dev;
 168         dfitem->func = func;
 169         memcpy(dfitem->data, data, data_len);
 170         dev_hold(dev);
 171         spin_lock_bh(&deferred_lock);
 172         list_add_tail(&dfitem->list, &deferred);
 173         spin_unlock_bh(&deferred_lock);
 174         schedule_work(&deferred_process_work);
 175         return 0;
 176 }
 177
 178 /**
 179  *      switchdev_port_attr_get - Get port attribute
 180  *
 181  *      @dev: port device
 182  *      @attr: attribute to get
 183  */
 184 int switchdev_port_attr_get(struct net_device *dev, struct switchdev_attr *attr)
 185 {
 186         const struct switchdev_ops *ops = dev->switchdev_ops;
 187         struct net_device *lower_dev;
 188         struct list_head *iter;
 189         struct switchdev_attr first = {
 190                 .id = SWITCHDEV_ATTR_ID_UNDEFINED
 191         };
 192         int err = -EOPNOTSUPP;
 193
 194         if (ops && ops->switchdev_port_attr_get)
 195                 return ops->switchdev_port_attr_get(dev, attr);
 196
 197         if (attr->flags & SWITCHDEV_F_NO_RECURSE)
 198                 return err;
 199
 200         /* Switch device port(s) may be stacked under
 201          * bond/team/vlan dev, so recurse down to get attr on
 202          * each port.  Return -ENODATA if attr values don't
 203          * compare across ports.
 204          */
 205
 206         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 207                 err = switchdev_port_attr_get(lower_dev, attr);
 208                 if (err)
 209                         break;
 210                 if (first.id == SWITCHDEV_ATTR_ID_UNDEFINED)
 211                         first = *attr;
 212                 else if (memcmp(&first, attr, sizeof(*attr)))
 213                         return -ENODATA;
 214         }
 215
 216         return err;
 217 }
 218 EXPORT_SYMBOL_GPL(switchdev_port_attr_get);
 219
 220 static int __switchdev_port_attr_set(struct net_device *dev,
 221                                      const struct switchdev_attr *attr,
 222                                      struct switchdev_trans *trans)
 223 {
 224         const struct switchdev_ops *ops = dev->switchdev_ops;
 225         struct net_device *lower_dev;
 226         struct list_head *iter;
 227         int err = -EOPNOTSUPP;
 228
 229         if (ops && ops->switchdev_port_attr_set) {
 230                 err = ops->switchdev_port_attr_set(dev, attr, trans);
 231                 goto done;
 232         }
 233
 234         if (attr->flags & SWITCHDEV_F_NO_RECURSE)
 235                 goto done;
 236
 237         /* Switch device port(s) may be stacked under
 238          * bond/team/vlan dev, so recurse down to set attr on
 239          * each port.
 240          */
 241
 242         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 243                 err = __switchdev_port_attr_set(lower_dev, attr, trans);
 244                 if (err)
 245                         break;
 246         }
 247
 248 done:
 249         if (err == -EOPNOTSUPP && attr->flags & SWITCHDEV_F_SKIP_EOPNOTSUPP)
 250                 err = 0;
 251
 252         return err;
 253 }
 254
 255 static int switchdev_port_attr_set_now(struct net_device *dev,
 256                                        const struct switchdev_attr *attr)
 257 {
 258         struct switchdev_trans trans;
 259         int err;
 260
 261         switchdev_trans_init(&trans);
 262
 263         /* Phase I: prepare for attr set. Driver/device should fail
 264          * here if there are going to be issues in the commit phase,
 265          * such as lack of resources or support.  The driver/device
 266          * should reserve resources needed for the commit phase here,
 267          * but should not commit the attr.
 268          */
 269
 270         trans.ph_prepare = true;
 271         err = __switchdev_port_attr_set(dev, attr, &trans);
 272         if (err) {
 273                 /* Prepare phase failed: abort the transaction.  Any
 274                  * resources reserved in the prepare phase are
 275                  * released.
 276                  */
 277
 278                 if (err != -EOPNOTSUPP)
 279                         switchdev_trans_items_destroy(&trans);
 280
 281                 return err;
 282         }
 283
 284         /* Phase II: commit attr set.  This cannot fail as a fault
 285          * of driver/device.  If it does, it's a bug in the driver/device
 286          * because the driver said everythings was OK in phase I.
 287          */
 288
 289         trans.ph_prepare = false;
 290         err = __switchdev_port_attr_set(dev, attr, &trans);
 291         WARN(err, "%s: Commit of attribute (id=%d) failed.\n",
 292              dev->name, attr->id);
 293         switchdev_trans_items_warn_destroy(dev, &trans);
 294
 295         return err;
 296 }
 297
 298 static void switchdev_port_attr_set_deferred(struct net_device *dev,
 299                                              const void *data)
 300 {
 301         const struct switchdev_attr *attr = data;
 302         int err;
 303
 304         err = switchdev_port_attr_set_now(dev, attr);
 305         if (err && err != -EOPNOTSUPP)
 306                 netdev_err(dev, "failed (err=%d) to set attribute (id=%d)\n",
 307                            err, attr->id);
 308 }
 309
 310 static int switchdev_port_attr_set_defer(struct net_device *dev,
 311                                          const struct switchdev_attr *attr)
 312 {
 313         return switchdev_deferred_enqueue(dev, attr, sizeof(*attr),
 314                                           switchdev_port_attr_set_deferred);
 315 }
 316
 317 /**
 318  *      switchdev_port_attr_set - Set port attribute
 319  *
 320  *      @dev: port device
 321  *      @attr: attribute to set
 322  *
 323  *      Use a 2-phase prepare-commit transaction model to ensure
 324  *      system is not left in a partially updated state due to
 325  *      failure from driver/device.
 326  *
 327  *      rtnl_lock must be held and must not be in atomic section,
 328  *      in case SWITCHDEV_F_DEFER flag is not set.
 329  */
 330 int switchdev_port_attr_set(struct net_device *dev,
 331                             const struct switchdev_attr *attr)
 332 {
 333         if (attr->flags & SWITCHDEV_F_DEFER)
 334                 return switchdev_port_attr_set_defer(dev, attr);
 335         ASSERT_RTNL();
 336         return switchdev_port_attr_set_now(dev, attr);
 337 }
 338 EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
 339
 340 static size_t switchdev_obj_size(const struct switchdev_obj *obj)
 341 {
 342         switch (obj->id) {
 343         case SWITCHDEV_OBJ_ID_PORT_VLAN:
 344                 return sizeof(struct switchdev_obj_port_vlan);
 345         case SWITCHDEV_OBJ_ID_IPV4_FIB:
 346                 return sizeof(struct switchdev_obj_ipv4_fib);
 347         case SWITCHDEV_OBJ_ID_PORT_FDB:
 348                 return sizeof(struct switchdev_obj_port_fdb);
 349         default:
 350                 BUG();
 351         }
 352         return 0;
 353 }
 354
 355 static int __switchdev_port_obj_add(struct net_device *dev,
 356                                     const struct switchdev_obj *obj,
 357                                     struct switchdev_trans *trans)
 358 {
 359         const struct switchdev_ops *ops = dev->switchdev_ops;
 360         struct net_device *lower_dev;
 361         struct list_head *iter;
 362         int err = -EOPNOTSUPP;
 363
 364         if (ops && ops->switchdev_port_obj_add)
 365                 return ops->switchdev_port_obj_add(dev, obj, trans);
 366
 367         /* Switch device port(s) may be stacked under
 368          * bond/team/vlan dev, so recurse down to add object on
 369          * each port.
 370          */
 371
 372         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 373                 err = __switchdev_port_obj_add(lower_dev, obj, trans);
 374                 if (err)
 375                         break;
 376         }
 377
 378         return err;
 379 }
 380
 381 static int switchdev_port_obj_add_now(struct net_device *dev,
 382                                       const struct switchdev_obj *obj)
 383 {
 384         struct switchdev_trans trans;
 385         int err;
 386
 387         ASSERT_RTNL();
 388
 389         switchdev_trans_init(&trans);
 390
 391         /* Phase I: prepare for obj add. Driver/device should fail
 392          * here if there are going to be issues in the commit phase,
 393          * such as lack of resources or support.  The driver/device
 394          * should reserve resources needed for the commit phase here,
 395          * but should not commit the obj.
 396          */
 397
 398         trans.ph_prepare = true;
 399         err = __switchdev_port_obj_add(dev, obj, &trans);
 400         if (err) {
 401                 /* Prepare phase failed: abort the transaction.  Any
 402                  * resources reserved in the prepare phase are
 403                  * released.
 404                  */
 405
 406                 if (err != -EOPNOTSUPP)
 407                         switchdev_trans_items_destroy(&trans);
 408
 409                 return err;
 410         }
 411
 412         /* Phase II: commit obj add.  This cannot fail as a fault
 413          * of driver/device.  If it does, it's a bug in the driver/device
 414          * because the driver said everythings was OK in phase I.
 415          */
 416
 417         trans.ph_prepare = false;
 418         err = __switchdev_port_obj_add(dev, obj, &trans);
 419         WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->id);
 420         switchdev_trans_items_warn_destroy(dev, &trans);
 421
 422         return err;
 423 }
 424
 425 static void switchdev_port_obj_add_deferred(struct net_device *dev,
 426                                             const void *data)
 427 {
 428         const struct switchdev_obj *obj = data;
 429         int err;
 430
 431         err = switchdev_port_obj_add_now(dev, obj);
 432         if (err && err != -EOPNOTSUPP)
 433                 netdev_err(dev, "failed (err=%d) to add object (id=%d)\n",
 434                            err, obj->id);
 435 }
 436
 437 static int switchdev_port_obj_add_defer(struct net_device *dev,
 438                                         const struct switchdev_obj *obj)
 439 {
 440         return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
 441                                           switchdev_port_obj_add_deferred);
 442 }
 443
 444 /**
 445  *      switchdev_port_obj_add - Add port object
 446  *
 447  *      @dev: port device
 448  *      @id: object ID
 449  *      @obj: object to add
 450  *
 451  *      Use a 2-phase prepare-commit transaction model to ensure
 452  *      system is not left in a partially updated state due to
 453  *      failure from driver/device.
 454  *
 455  *      rtnl_lock must be held and must not be in atomic section,
 456  *      in case SWITCHDEV_F_DEFER flag is not set.
 457  */
 458 int switchdev_port_obj_add(struct net_device *dev,
 459                            const struct switchdev_obj *obj)
 460 {
 461         if (obj->flags & SWITCHDEV_F_DEFER)
 462                 return switchdev_port_obj_add_defer(dev, obj);
 463         ASSERT_RTNL();
 464         return switchdev_port_obj_add_now(dev, obj);
 465 }
 466 EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
 467
 468 static int switchdev_port_obj_del_now(struct net_device *dev,
 469                                       const struct switchdev_obj *obj)
 470 {
 471         const struct switchdev_ops *ops = dev->switchdev_ops;
 472         struct net_device *lower_dev;
 473         struct list_head *iter;
 474         int err = -EOPNOTSUPP;
 475
 476         if (ops && ops->switchdev_port_obj_del)
 477                 return ops->switchdev_port_obj_del(dev, obj);
 478
 479         /* Switch device port(s) may be stacked under
 480          * bond/team/vlan dev, so recurse down to delete object on
 481          * each port.
 482          */
 483
 484         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 485                 err = switchdev_port_obj_del_now(lower_dev, obj);
 486                 if (err)
 487                         break;
 488         }
 489
 490         return err;
 491 }
 492
 493 static void switchdev_port_obj_del_deferred(struct net_device *dev,
 494                                             const void *data)
 495 {
 496         const struct switchdev_obj *obj = data;
 497         int err;
 498
 499         err = switchdev_port_obj_del_now(dev, obj);
 500         if (err && err != -EOPNOTSUPP)
 501                 netdev_err(dev, "failed (err=%d) to del object (id=%d)\n",
 502                            err, obj->id);
 503 }
 504
 505 static int switchdev_port_obj_del_defer(struct net_device *dev,
 506                                         const struct switchdev_obj *obj)
 507 {
 508         return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
 509                                           switchdev_port_obj_del_deferred);
 510 }
 511
 512 /**
 513  *      switchdev_port_obj_del - Delete port object
 514  *
 515  *      @dev: port device
 516  *      @id: object ID
 517  *      @obj: object to delete
 518  *
 519  *      rtnl_lock must be held and must not be in atomic section,
 520  *      in case SWITCHDEV_F_DEFER flag is not set.
 521  */
 522 int switchdev_port_obj_del(struct net_device *dev,
 523                            const struct switchdev_obj *obj)
 524 {
 525         if (obj->flags & SWITCHDEV_F_DEFER)
 526                 return switchdev_port_obj_del_defer(dev, obj);
 527         ASSERT_RTNL();
 528         return switchdev_port_obj_del_now(dev, obj);
 529 }
 530 EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
 531
 532 /**
 533  *      switchdev_port_obj_dump - Dump port objects
 534  *
 535  *      @dev: port device
 536  *      @id: object ID
 537  *      @obj: object to dump
 538  *      @cb: function to call with a filled object
 539  *
 540  *      rtnl_lock must be held.
 541  */
 542 int switchdev_port_obj_dump(struct net_device *dev, struct switchdev_obj *obj,
 543                             switchdev_obj_dump_cb_t *cb)
 544 {
 545         const struct switchdev_ops *ops = dev->switchdev_ops;
 546         struct net_device *lower_dev;
 547         struct list_head *iter;
 548         int err = -EOPNOTSUPP;
 549
 550         ASSERT_RTNL();
 551
 552         if (ops && ops->switchdev_port_obj_dump)
 553                 return ops->switchdev_port_obj_dump(dev, obj, cb);
 554
 555         /* Switch device port(s) may be stacked under
 556          * bond/team/vlan dev, so recurse down to dump objects on
 557          * first port at bottom of stack.
 558          */
 559
 560         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 561                 err = switchdev_port_obj_dump(lower_dev, obj, cb);
 562                 break;
 563         }
 564
 565         return err;
 566 }
 567 EXPORT_SYMBOL_GPL(switchdev_port_obj_dump);
 568
 569 static RAW_NOTIFIER_HEAD(switchdev_notif_chain);
 570
 571 /**
 572  *      register_switchdev_notifier - Register notifier
 573  *      @nb: notifier_block
 574  *
 575  *      Register switch device notifier. This should be used by code
 576  *      which needs to monitor events happening in particular device.
 577  *      Return values are same as for atomic_notifier_chain_register().
 578  */
 579 int register_switchdev_notifier(struct notifier_block *nb)
 580 {
 581         int err;
 582
 583         rtnl_lock();
 584         err = raw_notifier_chain_register(&switchdev_notif_chain, nb);
 585         rtnl_unlock();
 586         return err;
 587 }
 588 EXPORT_SYMBOL_GPL(register_switchdev_notifier);
 589
 590 /**
 591  *      unregister_switchdev_notifier - Unregister notifier
 592  *      @nb: notifier_block
 593  *
 594  *      Unregister switch device notifier.
 595  *      Return values are same as for atomic_notifier_chain_unregister().
 596  */
 597 int unregister_switchdev_notifier(struct notifier_block *nb)
 598 {
 599         int err;
 600
 601         rtnl_lock();
 602         err = raw_notifier_chain_unregister(&switchdev_notif_chain, nb);
 603         rtnl_unlock();
 604         return err;
 605 }
 606 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
 607
 608 /**
 609  *      call_switchdev_notifiers - Call notifiers
 610  *      @val: value passed unmodified to notifier function
 611  *      @dev: port device
 612  *      @info: notifier information data
 613  *
 614  *      Call all network notifier blocks. This should be called by driver
 615  *      when it needs to propagate hardware event.
 616  *      Return values are same as for atomic_notifier_call_chain().
 617  *      rtnl_lock must be held.
 618  */
 619 int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
 620                              struct switchdev_notifier_info *info)
 621 {
 622         int err;
 623
 624         ASSERT_RTNL();
 625
 626         info->dev = dev;
 627         err = raw_notifier_call_chain(&switchdev_notif_chain, val, info);
 628         return err;
 629 }
 630 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
 631
 632 struct switchdev_vlan_dump {
 633         struct switchdev_obj_port_vlan vlan;
 634         struct sk_buff *skb;
 635         u32 filter_mask;
 636         u16 flags;
 637         u16 begin;
 638         u16 end;
 639 };
 640
 641 static int switchdev_port_vlan_dump_put(struct switchdev_vlan_dump *dump)
 642 {
 643         struct bridge_vlan_info vinfo;
 644
 645         vinfo.flags = dump->flags;
 646
 647         if (dump->begin == 0 && dump->end == 0) {
 648                 return 0;
 649         } else if (dump->begin == dump->end) {
 650                 vinfo.vid = dump->begin;
 651                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 652                             sizeof(vinfo), &vinfo))
 653                         return -EMSGSIZE;
 654         } else {
 655                 vinfo.vid = dump->begin;
 656                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_BEGIN;
 657                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 658                             sizeof(vinfo), &vinfo))
 659                         return -EMSGSIZE;
 660                 vinfo.vid = dump->end;
 661                 vinfo.flags &= ~BRIDGE_VLAN_INFO_RANGE_BEGIN;
 662                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_END;
 663                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 664                             sizeof(vinfo), &vinfo))
 665                         return -EMSGSIZE;
 666         }
 667
 668         return 0;
 669 }
 670
 671 static int switchdev_port_vlan_dump_cb(struct switchdev_obj *obj)
 672 {
 673         struct switchdev_obj_port_vlan *vlan = SWITCHDEV_OBJ_PORT_VLAN(obj);
 674         struct switchdev_vlan_dump *dump =
 675                 container_of(vlan, struct switchdev_vlan_dump, vlan);
 676         int err = 0;
 677
 678         if (vlan->vid_begin > vlan->vid_end)
 679                 return -EINVAL;
 680
 681         if (dump->filter_mask & RTEXT_FILTER_BRVLAN) {
 682                 dump->flags = vlan->flags;
 683                 for (dump->begin = dump->end = vlan->vid_begin;
 684                      dump->begin <= vlan->vid_end;
 685                      dump->begin++, dump->end++) {
 686                         err = switchdev_port_vlan_dump_put(dump);
 687                         if (err)
 688                                 return err;
 689                 }
 690         } else if (dump->filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED) {
 691                 if (dump->begin > vlan->vid_begin &&
 692                     dump->begin >= vlan->vid_end) {
 693                         if ((dump->begin - 1) == vlan->vid_end &&
 694                             dump->flags == vlan->flags) {
 695                                 /* prepend */
 696                                 dump->begin = vlan->vid_begin;
 697                         } else {
 698                                 err = switchdev_port_vlan_dump_put(dump);
 699                                 dump->flags = vlan->flags;
 700                                 dump->begin = vlan->vid_begin;
 701                                 dump->end = vlan->vid_end;
 702                         }
 703                 } else if (dump->end <= vlan->vid_begin &&
 704                            dump->end < vlan->vid_end) {
 705                         if ((dump->end  + 1) == vlan->vid_begin &&
 706                             dump->flags == vlan->flags) {
 707                                 /* append */
 708                                 dump->end = vlan->vid_end;
 709                         } else {
 710                                 err = switchdev_port_vlan_dump_put(dump);
 711                                 dump->flags = vlan->flags;
 712                                 dump->begin = vlan->vid_begin;
 713                                 dump->end = vlan->vid_end;
 714                         }
 715                 } else {
 716                         err = -EINVAL;
 717                 }
 718         }
 719
 720         return err;
 721 }
 722
 723 static int switchdev_port_vlan_fill(struct sk_buff *skb, struct net_device *dev,
 724                                     u32 filter_mask)
 725 {
 726         struct switchdev_vlan_dump dump = {
 727                 .vlan.obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
 728                 .skb = skb,
 729                 .filter_mask = filter_mask,
 730         };
 731         int err = 0;
 732
 733         if ((filter_mask & RTEXT_FILTER_BRVLAN) ||
 734             (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)) {
 735                 err = switchdev_port_obj_dump(dev, &dump.vlan.obj,
 736                                               switchdev_port_vlan_dump_cb);
 737                 if (err)
 738                         goto err_out;
 739                 if (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)
 740                         /* last one */
 741                         err = switchdev_port_vlan_dump_put(&dump);
 742         }
 743
 744 err_out:
 745         return err == -EOPNOTSUPP ? 0 : err;
 746 }
 747
 748 /**
 749  *      switchdev_port_bridge_getlink - Get bridge port attributes
 750  *
 751  *      @dev: port device
 752  *
 753  *      Called for SELF on rtnl_bridge_getlink to get bridge port
 754  *      attributes.
 755  */
 756 int switchdev_port_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
 757                                   struct net_device *dev, u32 filter_mask,
 758                                   int nlflags)
 759 {
 760         struct switchdev_attr attr = {
 761                 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
 762         };
 763         u16 mode = BRIDGE_MODE_UNDEF;
 764         u32 mask = BR_LEARNING | BR_LEARNING_SYNC | BR_FLOOD;
 765         int err;
 766
 767         err = switchdev_port_attr_get(dev, &attr);
 768         if (err && err != -EOPNOTSUPP)
 769                 return err;
 770
 771         return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode,
 772                                        attr.u.brport_flags, mask, nlflags,
 773                                        filter_mask, switchdev_port_vlan_fill);
 774 }
 775 EXPORT_SYMBOL_GPL(switchdev_port_bridge_getlink);
 776
 777 static int switchdev_port_br_setflag(struct net_device *dev,
 778                                      struct nlattr *nlattr,
 779                                      unsigned long brport_flag)
 780 {
 781         struct switchdev_attr attr = {
 782                 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
 783         };
 784         u8 flag = nla_get_u8(nlattr);
 785         int err;
 786
 787         err = switchdev_port_attr_get(dev, &attr);
 788         if (err)
 789                 return err;
 790
 791         if (flag)
 792                 attr.u.brport_flags |= brport_flag;
 793         else
 794                 attr.u.brport_flags &= ~brport_flag;
 795
 796         return switchdev_port_attr_set(dev, &attr);
 797 }
 798
 799 static const struct nla_policy
 800 switchdev_port_bridge_policy[IFLA_BRPORT_MAX + 1] = {
 801         [IFLA_BRPORT_STATE]             = { .type = NLA_U8 },
 802         [IFLA_BRPORT_COST]              = { .type = NLA_U32 },
 803         [IFLA_BRPORT_PRIORITY]          = { .type = NLA_U16 },
 804         [IFLA_BRPORT_MODE]              = { .type = NLA_U8 },
 805         [IFLA_BRPORT_GUARD]             = { .type = NLA_U8 },
 806         [IFLA_BRPORT_PROTECT]           = { .type = NLA_U8 },
 807         [IFLA_BRPORT_FAST_LEAVE]        = { .type = NLA_U8 },
 808         [IFLA_BRPORT_LEARNING]          = { .type = NLA_U8 },
 809         [IFLA_BRPORT_LEARNING_SYNC]     = { .type = NLA_U8 },
 810         [IFLA_BRPORT_UNICAST_FLOOD]     = { .type = NLA_U8 },
 811 };
 812
 813 static int switchdev_port_br_setlink_protinfo(struct net_device *dev,
 814                                               struct nlattr *protinfo)
 815 {
 816         struct nlattr *attr;
 817         int rem;
 818         int err;
 819
 820         err = nla_validate_nested(protinfo, IFLA_BRPORT_MAX,
 821                                   switchdev_port_bridge_policy);
 822         if (err)
 823                 return err;
 824
 825         nla_for_each_nested(attr, protinfo, rem) {
 826                 switch (nla_type(attr)) {
 827                 case IFLA_BRPORT_LEARNING:
 828                         err = switchdev_port_br_setflag(dev, attr,
 829                                                         BR_LEARNING);
 830                         break;
 831                 case IFLA_BRPORT_LEARNING_SYNC:
 832                         err = switchdev_port_br_setflag(dev, attr,
 833                                                         BR_LEARNING_SYNC);
 834                         break;
 835                 case IFLA_BRPORT_UNICAST_FLOOD:
 836                         err = switchdev_port_br_setflag(dev, attr, BR_FLOOD);
 837                         break;
 838                 default:
 839                         err = -EOPNOTSUPP;
 840                         break;
 841                 }
 842                 if (err)
 843                         return err;
 844         }
 845
 846         return 0;
 847 }
 848
 849 static int switchdev_port_br_afspec(struct net_device *dev,
 850                                     struct nlattr *afspec,
 851                                     int (*f)(struct net_device *dev,
 852                                              const struct switchdev_obj *obj))
 853 {
 854         struct nlattr *attr;
 855         struct bridge_vlan_info *vinfo;
 856         struct switchdev_obj_port_vlan vlan = {
 857                 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
 858         };
 859         int rem;
 860         int err;
 861
 862         nla_for_each_nested(attr, afspec, rem) {
 863                 if (nla_type(attr) != IFLA_BRIDGE_VLAN_INFO)
 864                         continue;
 865                 if (nla_len(attr) != sizeof(struct bridge_vlan_info))
 866                         return -EINVAL;
 867                 vinfo = nla_data(attr);
 868                 if (!vinfo->vid || vinfo->vid >= VLAN_VID_MASK)
 869                         return -EINVAL;
 870                 vlan.flags = vinfo->flags;
 871                 if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_BEGIN) {
 872                         if (vlan.vid_begin)
 873                                 return -EINVAL;
 874                         vlan.vid_begin = vinfo->vid;
 875                         /* don't allow range of pvids */
 876                         if (vlan.flags & BRIDGE_VLAN_INFO_PVID)
 877                                 return -EINVAL;
 878                 } else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
 879                         if (!vlan.vid_begin)
 880                                 return -EINVAL;
 881                         vlan.vid_end = vinfo->vid;
 882                         if (vlan.vid_end <= vlan.vid_begin)
 883                                 return -EINVAL;
 884                         err = f(dev, &vlan.obj);
 885                         if (err)
 886                                 return err;
 887                         vlan.vid_begin = 0;
 888                 } else {
 889                         if (vlan.vid_begin)
 890                                 return -EINVAL;
 891                         vlan.vid_begin = vinfo->vid;
 892                         vlan.vid_end = vinfo->vid;
 893                         err = f(dev, &vlan.obj);
 894                         if (err)
 895                                 return err;
 896                         vlan.vid_begin = 0;
 897                 }
 898         }
 899
 900         return 0;
 901 }
 902
 903 /**
 904  *      switchdev_port_bridge_setlink - Set bridge port attributes
 905  *
 906  *      @dev: port device
 907  *      @nlh: netlink header
 908  *      @flags: netlink flags
 909  *
 910  *      Called for SELF on rtnl_bridge_setlink to set bridge port
 911  *      attributes.
 912  */
 913 int switchdev_port_bridge_setlink(struct net_device *dev,
 914                                   struct nlmsghdr *nlh, u16 flags)
 915 {
 916         struct nlattr *protinfo;
 917         struct nlattr *afspec;
 918         int err = 0;
 919
 920         protinfo = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 921                                    IFLA_PROTINFO);
 922         if (protinfo) {
 923                 err = switchdev_port_br_setlink_protinfo(dev, protinfo);
 924                 if (err)
 925                         return err;
 926         }
 927
 928         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 929                                  IFLA_AF_SPEC);
 930         if (afspec)
 931                 err = switchdev_port_br_afspec(dev, afspec,
 932                                                switchdev_port_obj_add);
 933
 934         return err;
 935 }
 936 EXPORT_SYMBOL_GPL(switchdev_port_bridge_setlink);
 937
 938 /**
 939  *      switchdev_port_bridge_dellink - Set bridge port attributes
 940  *
 941  *      @dev: port device
 942  *      @nlh: netlink header
 943  *      @flags: netlink flags
 944  *
 945  *      Called for SELF on rtnl_bridge_dellink to set bridge port
 946  *      attributes.
 947  */
 948 int switchdev_port_bridge_dellink(struct net_device *dev,
 949                                   struct nlmsghdr *nlh, u16 flags)
 950 {
 951         struct nlattr *afspec;
 952
 953         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 954                                  IFLA_AF_SPEC);
 955         if (afspec)
 956                 return switchdev_port_br_afspec(dev, afspec,
 957                                                 switchdev_port_obj_del);
 958
 959         return 0;
 960 }
 961 EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
 962
 963 /**
 964  *      switchdev_port_fdb_add - Add FDB (MAC/VLAN) entry to port
 965  *
 966  *      @ndmsg: netlink hdr
 967  *      @nlattr: netlink attributes
 968  *      @dev: port device
 969  *      @addr: MAC address to add
 970  *      @vid: VLAN to add
 971  *
 972  *      Add FDB entry to switch device.
 973  */
 974 int switchdev_port_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 975                            struct net_device *dev, const unsigned char *addr,
 976                            u16 vid, u16 nlm_flags)
 977 {
 978         struct switchdev_obj_port_fdb fdb = {
 979                 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
 980                 .vid = vid,
 981         };
 982
 983         ether_addr_copy(fdb.addr, addr);
 984         return switchdev_port_obj_add(dev, &fdb.obj);
 985 }
 986 EXPORT_SYMBOL_GPL(switchdev_port_fdb_add);
 987
 988 /**
 989  *      switchdev_port_fdb_del - Delete FDB (MAC/VLAN) entry from port
 990  *
 991  *      @ndmsg: netlink hdr
 992  *      @nlattr: netlink attributes
 993  *      @dev: port device
 994  *      @addr: MAC address to delete
 995  *      @vid: VLAN to delete
 996  *
 997  *      Delete FDB entry from switch device.
 998  */
 999 int switchdev_port_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
1000                            struct net_device *dev, const unsigned char *addr,
1001                            u16 vid)
1002 {
1003         struct switchdev_obj_port_fdb fdb = {
1004                 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
1005                 .vid = vid,
1006         };
1007
1008         ether_addr_copy(fdb.addr, addr);
1009         return switchdev_port_obj_del(dev, &fdb.obj);
1010 }
1011 EXPORT_SYMBOL_GPL(switchdev_port_fdb_del);
1012
1013 struct switchdev_fdb_dump {
1014         struct switchdev_obj_port_fdb fdb;
1015         struct net_device *dev;
1016         struct sk_buff *skb;
1017         struct netlink_callback *cb;
1018         int idx;
1019 };
1020
1021 static int switchdev_port_fdb_dump_cb(struct switchdev_obj *obj)
1022 {
1023         struct switchdev_obj_port_fdb *fdb = SWITCHDEV_OBJ_PORT_FDB(obj);
1024         struct switchdev_fdb_dump *dump =
1025                 container_of(fdb, struct switchdev_fdb_dump, fdb);
1026         u32 portid = NETLINK_CB(dump->cb->skb).portid;
1027         u32 seq = dump->cb->nlh->nlmsg_seq;
1028         struct nlmsghdr *nlh;
1029         struct ndmsg *ndm;
1030
1031         if (dump->idx < dump->cb->args[0])
1032                 goto skip;
1033
1034         nlh = nlmsg_put(dump->skb, portid, seq, RTM_NEWNEIGH,
1035                         sizeof(*ndm), NLM_F_MULTI);
1036         if (!nlh)
1037                 return -EMSGSIZE;
1038
1039         ndm = nlmsg_data(nlh);
1040         ndm->ndm_family  = AF_BRIDGE;
1041         ndm->ndm_pad1    = 0;
1042         ndm->ndm_pad2    = 0;
1043         ndm->ndm_flags   = NTF_SELF;
1044         ndm->ndm_type    = 0;
1045         ndm->ndm_ifindex = dump->dev->ifindex;
1046         ndm->ndm_state   = fdb->ndm_state;
1047
1048         if (nla_put(dump->skb, NDA_LLADDR, ETH_ALEN, fdb->addr))
1049                 goto nla_put_failure;
1050
1051         if (fdb->vid && nla_put_u16(dump->skb, NDA_VLAN, fdb->vid))
1052                 goto nla_put_failure;
1053
1054         nlmsg_end(dump->skb, nlh);
1055
1056 skip:
1057         dump->idx++;
1058         return 0;
1059
1060 nla_put_failure:
1061         nlmsg_cancel(dump->skb, nlh);
1062         return -EMSGSIZE;
1063 }
1064
1065 /**
1066  *      switchdev_port_fdb_dump - Dump port FDB (MAC/VLAN) entries
1067  *
1068  *      @skb: netlink skb
1069  *      @cb: netlink callback
1070  *      @dev: port device
1071  *      @filter_dev: filter device
1072  *      @idx:
1073  *
1074  *      Delete FDB entry from switch device.
1075  */
1076 int switchdev_port_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1077                             struct net_device *dev,
1078                             struct net_device *filter_dev, int idx)
1079 {
1080         struct switchdev_fdb_dump dump = {
1081                 .fdb.obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
1082                 .dev = dev,
1083                 .skb = skb,
1084                 .cb = cb,
1085                 .idx = idx,
1086         };
1087
1088         switchdev_port_obj_dump(dev, &dump.fdb.obj, switchdev_port_fdb_dump_cb);
1089         return dump.idx;
1090 }
1091 EXPORT_SYMBOL_GPL(switchdev_port_fdb_dump);
1092
1093 static struct net_device *switchdev_get_lowest_dev(struct net_device *dev)
1094 {
1095         const struct switchdev_ops *ops = dev->switchdev_ops;
1096         struct net_device *lower_dev;
1097         struct net_device *port_dev;
1098         struct list_head *iter;
1099
1100         /* Recusively search down until we find a sw port dev.
1101          * (A sw port dev supports switchdev_port_attr_get).
1102          */
1103
1104         if (ops && ops->switchdev_port_attr_get)
1105                 return dev;
1106
1107         netdev_for_each_lower_dev(dev, lower_dev, iter) {
1108                 port_dev = switchdev_get_lowest_dev(lower_dev);
1109                 if (port_dev)
1110                         return port_dev;
1111         }
1112
1113         return NULL;
1114 }
1115
1116 static struct net_device *switchdev_get_dev_by_nhs(struct fib_info *fi)
1117 {
1118         struct switchdev_attr attr = {
1119                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1120         };
1121         struct switchdev_attr prev_attr;
1122         struct net_device *dev = NULL;
1123         int nhsel;
1124
1125         ASSERT_RTNL();
1126
1127         /* For this route, all nexthop devs must be on the same switch. */
1128
1129         for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1130                 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1131
1132                 if (!nh->nh_dev)
1133                         return NULL;
1134
1135                 dev = switchdev_get_lowest_dev(nh->nh_dev);
1136                 if (!dev)
1137                         return NULL;
1138
1139                 if (switchdev_port_attr_get(dev, &attr))
1140                         return NULL;
1141
1142                 if (nhsel > 0 &&
1143                     !netdev_phys_item_id_same(&prev_attr.u.ppid, &attr.u.ppid))
1144                                 return NULL;
1145
1146                 prev_attr = attr;
1147         }
1148
1149         return dev;
1150 }
1151
1152 /**
1153  *      switchdev_fib_ipv4_add - Add/modify switch IPv4 route entry
1154  *
1155  *      @dst: route's IPv4 destination address
1156  *      @dst_len: destination address length (prefix length)
1157  *      @fi: route FIB info structure
1158  *      @tos: route TOS
1159  *      @type: route type
1160  *      @nlflags: netlink flags passed in (NLM_F_*)
1161  *      @tb_id: route table ID
1162  *
1163  *      Add/modify switch IPv4 route entry.
1164  */
1165 int switchdev_fib_ipv4_add(u32 dst, int dst_len, struct fib_info *fi,
1166                            u8 tos, u8 type, u32 nlflags, u32 tb_id)
1167 {
1168         struct switchdev_obj_ipv4_fib ipv4_fib = {
1169                 .obj.id = SWITCHDEV_OBJ_ID_IPV4_FIB,
1170                 .dst = dst,
1171                 .dst_len = dst_len,
1172                 .tos = tos,
1173                 .type = type,
1174                 .nlflags = nlflags,
1175                 .tb_id = tb_id,
1176         };
1177         struct net_device *dev;
1178         int err = 0;
1179
1180         memcpy(&ipv4_fib.fi, fi, sizeof(ipv4_fib.fi));
1181
1182         /* Don't offload route if using custom ip rules or if
1183          * IPv4 FIB offloading has been disabled completely.
1184          */
1185
1186 #ifdef CONFIG_IP_MULTIPLE_TABLES
1187         if (fi->fib_net->ipv4.fib_has_custom_rules)
1188                 return 0;
1189 #endif
1190
1191         if (fi->fib_net->ipv4.fib_offload_disabled)
1192                 return 0;
1193
1194         dev = switchdev_get_dev_by_nhs(fi);
1195         if (!dev)
1196                 return 0;
1197
1198         err = switchdev_port_obj_add(dev, &ipv4_fib.obj);
1199         if (!err)
1200                 fi->fib_flags |= RTNH_F_OFFLOAD;
1201
1202         return err == -EOPNOTSUPP ? 0 : err;
1203 }
1204 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_add);
1205
1206 /**
1207  *      switchdev_fib_ipv4_del - Delete IPv4 route entry from switch
1208  *
1209  *      @dst: route's IPv4 destination address
1210  *      @dst_len: destination address length (prefix length)
1211  *      @fi: route FIB info structure
1212  *      @tos: route TOS
1213  *      @type: route type
1214  *      @tb_id: route table ID
1215  *
1216  *      Delete IPv4 route entry from switch device.
1217  */
1218 int switchdev_fib_ipv4_del(u32 dst, int dst_len, struct fib_info *fi,
1219                            u8 tos, u8 type, u32 tb_id)
1220 {
1221         struct switchdev_obj_ipv4_fib ipv4_fib = {
1222                 .obj.id = SWITCHDEV_OBJ_ID_IPV4_FIB,
1223                 .dst = dst,
1224                 .dst_len = dst_len,
1225                 .tos = tos,
1226                 .type = type,
1227                 .nlflags = 0,
1228                 .tb_id = tb_id,
1229         };
1230         struct net_device *dev;
1231         int err = 0;
1232
1233         memcpy(&ipv4_fib.fi, fi, sizeof(ipv4_fib.fi));
1234
1235         if (!(fi->fib_flags & RTNH_F_OFFLOAD))
1236                 return 0;
1237
1238         dev = switchdev_get_dev_by_nhs(fi);
1239         if (!dev)
1240                 return 0;
1241
1242         err = switchdev_port_obj_del(dev, &ipv4_fib.obj);
1243         if (!err)
1244                 fi->fib_flags &= ~RTNH_F_OFFLOAD;
1245
1246         return err == -EOPNOTSUPP ? 0 : err;
1247 }
1248 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_del);
1249
1250 /**
1251  *      switchdev_fib_ipv4_abort - Abort an IPv4 FIB operation
1252  *
1253  *      @fi: route FIB info structure
1254  */
1255 void switchdev_fib_ipv4_abort(struct fib_info *fi)
1256 {
1257         /* There was a problem installing this route to the offload
1258          * device.  For now, until we come up with more refined
1259          * policy handling, abruptly end IPv4 fib offloading for
1260          * for entire net by flushing offload device(s) of all
1261          * IPv4 routes, and mark IPv4 fib offloading broken from
1262          * this point forward.
1263          */
1264
1265         fib_flush_external(fi->fib_net);
1266         fi->fib_net->ipv4.fib_offload_disabled = true;
1267 }
1268 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_abort);
1269
1270 static bool switchdev_port_same_parent_id(struct net_device *a,
1271                                           struct net_device *b)
1272 {
1273         struct switchdev_attr a_attr = {
1274                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1275                 .flags = SWITCHDEV_F_NO_RECURSE,
1276         };
1277         struct switchdev_attr b_attr = {
1278                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1279                 .flags = SWITCHDEV_F_NO_RECURSE,
1280         };
1281
1282         if (switchdev_port_attr_get(a, &a_attr) ||
1283             switchdev_port_attr_get(b, &b_attr))
1284                 return false;
1285
1286         return netdev_phys_item_id_same(&a_attr.u.ppid, &b_attr.u.ppid);
1287 }
1288
1289 static u32 switchdev_port_fwd_mark_get(struct net_device *dev,
1290                                        struct net_device *group_dev)
1291 {
1292         struct net_device *lower_dev;
1293         struct list_head *iter;
1294
1295         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1296                 if (lower_dev == dev)
1297                         continue;
1298                 if (switchdev_port_same_parent_id(dev, lower_dev))
1299                         return lower_dev->offload_fwd_mark;
1300                 return switchdev_port_fwd_mark_get(dev, lower_dev);
1301         }
1302
1303         return dev->ifindex;
1304 }
1305
1306 static void switchdev_port_fwd_mark_reset(struct net_device *group_dev,
1307                                           u32 old_mark, u32 *reset_mark)
1308 {
1309         struct net_device *lower_dev;
1310         struct list_head *iter;
1311
1312         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1313                 if (lower_dev->offload_fwd_mark == old_mark) {
1314                         if (!*reset_mark)
1315                                 *reset_mark = lower_dev->ifindex;
1316                         lower_dev->offload_fwd_mark = *reset_mark;
1317                 }
1318                 switchdev_port_fwd_mark_reset(lower_dev, old_mark, reset_mark);
1319         }
1320 }
1321
1322 /**
1323  *      switchdev_port_fwd_mark_set - Set port offload forwarding mark
1324  *
1325  *      @dev: port device
1326  *      @group_dev: containing device
1327  *      @joining: true if dev is joining group; false if leaving group
1328  *
1329  *      An ungrouped port's offload mark is just its ifindex.  A grouped
1330  *      port's (member of a bridge, for example) offload mark is the ifindex
1331  *      of one of the ports in the group with the same parent (switch) ID.
1332  *      Ports on the same device in the same group will have the same mark.
1333  *
1334  *      Example:
1335  *
1336  *              br0             ifindex=9
1337  *                sw1p1         ifindex=2       mark=2
1338  *                sw1p2         ifindex=3       mark=2
1339  *                sw2p1         ifindex=4       mark=5
1340  *                sw2p2         ifindex=5       mark=5
1341  *
1342  *      If sw2p2 leaves the bridge, we'll have:
1343  *
1344  *              br0             ifindex=9
1345  *                sw1p1         ifindex=2       mark=2
1346  *                sw1p2         ifindex=3       mark=2
1347  *                sw2p1         ifindex=4       mark=4
1348  *              sw2p2           ifindex=5       mark=5
1349  */
1350 void switchdev_port_fwd_mark_set(struct net_device *dev,
1351                                  struct net_device *group_dev,
1352                                  bool joining)
1353 {
1354         u32 mark = dev->ifindex;
1355         u32 reset_mark = 0;
1356
1357         if (group_dev) {
1358                 ASSERT_RTNL();
1359                 if (joining)
1360                         mark = switchdev_port_fwd_mark_get(dev, group_dev);
1361                 else if (dev->offload_fwd_mark == mark)
1362                         /* Ohoh, this port was the mark reference port,
1363                          * but it's leaving the group, so reset the
1364                          * mark for the remaining ports in the group.
1365                          */
1366                         switchdev_port_fwd_mark_reset(group_dev, mark,
1367                                                       &reset_mark);
1368         }
1369
1370         dev->offload_fwd_mark = mark;
1371 }
1372 EXPORT_SYMBOL_GPL(switchdev_port_fwd_mark_set);