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 <net/ip_fib.h>
  24 #include <net/switchdev.h>
  25
  26 /**
  27  *      switchdev_trans_item_enqueue - Enqueue data item to transaction queue
  28  *
  29  *      @trans: transaction
  30  *      @data: pointer to data being queued
  31  *      @destructor: data destructor
  32  *      @tritem: transaction item being queued
  33  *
  34  *      Enqeueue data item to transaction queue. tritem is typically placed in
  35  *      cointainter pointed at by data pointer. Destructor is called on
  36  *      transaction abort and after successful commit phase in case
  37  *      the caller did not dequeue the item before.
  38  */
  39 void switchdev_trans_item_enqueue(struct switchdev_trans *trans,
  40                                   void *data, void (*destructor)(void const *),
  41                                   struct switchdev_trans_item *tritem)
  42 {
  43         tritem->data = data;
  44         tritem->destructor = destructor;
  45         list_add_tail(&tritem->list, &trans->item_list);
  46 }
  47 EXPORT_SYMBOL_GPL(switchdev_trans_item_enqueue);
  48
  49 static struct switchdev_trans_item *
  50 __switchdev_trans_item_dequeue(struct switchdev_trans *trans)
  51 {
  52         struct switchdev_trans_item *tritem;
  53
  54         if (list_empty(&trans->item_list))
  55                 return NULL;
  56         tritem = list_first_entry(&trans->item_list,
  57                                   struct switchdev_trans_item, list);
  58         list_del(&tritem->list);
  59         return tritem;
  60 }
  61
  62 /**
  63  *      switchdev_trans_item_dequeue - Dequeue data item from transaction queue
  64  *
  65  *      @trans: transaction
  66  */
  67 void *switchdev_trans_item_dequeue(struct switchdev_trans *trans)
  68 {
  69         struct switchdev_trans_item *tritem;
  70
  71         tritem = __switchdev_trans_item_dequeue(trans);
  72         BUG_ON(!tritem);
  73         return tritem->data;
  74 }
  75 EXPORT_SYMBOL_GPL(switchdev_trans_item_dequeue);
  76
  77 static void switchdev_trans_init(struct switchdev_trans *trans)
  78 {
  79         INIT_LIST_HEAD(&trans->item_list);
  80 }
  81
  82 static void switchdev_trans_items_destroy(struct switchdev_trans *trans)
  83 {
  84         struct switchdev_trans_item *tritem;
  85
  86         while ((tritem = __switchdev_trans_item_dequeue(trans)))
  87                 tritem->destructor(tritem->data);
  88 }
  89
  90 static void switchdev_trans_items_warn_destroy(struct net_device *dev,
  91                                                struct switchdev_trans *trans)
  92 {
  93         WARN(!list_empty(&trans->item_list), "%s: transaction item queue is not empty.\n",
  94              dev->name);
  95         switchdev_trans_items_destroy(trans);
  96 }
  97
  98 static LIST_HEAD(deferred);
  99 static DEFINE_SPINLOCK(deferred_lock);
 100
 101 typedef void switchdev_deferred_func_t(struct net_device *dev,
 102                                        const void *data);
 103
 104 struct switchdev_deferred_item {
 105         struct list_head list;
 106         struct net_device *dev;
 107         switchdev_deferred_func_t *func;
 108         unsigned long data[0];
 109 };
 110
 111 static struct switchdev_deferred_item *switchdev_deferred_dequeue(void)
 112 {
 113         struct switchdev_deferred_item *dfitem;
 114
 115         spin_lock_bh(&deferred_lock);
 116         if (list_empty(&deferred)) {
 117                 dfitem = NULL;
 118                 goto unlock;
 119         }
 120         dfitem = list_first_entry(&deferred,
 121                                   struct switchdev_deferred_item, list);
 122         list_del(&dfitem->list);
 123 unlock:
 124         spin_unlock_bh(&deferred_lock);
 125         return dfitem;
 126 }
 127
 128 /**
 129  *      switchdev_deferred_process - Process ops in deferred queue
 130  *
 131  *      Called to flush the ops currently queued in deferred ops queue.
 132  *      rtnl_lock must be held.
 133  */
 134 void switchdev_deferred_process(void)
 135 {
 136         struct switchdev_deferred_item *dfitem;
 137
 138         ASSERT_RTNL();
 139
 140         while ((dfitem = switchdev_deferred_dequeue())) {
 141                 dfitem->func(dfitem->dev, dfitem->data);
 142                 dev_put(dfitem->dev);
 143                 kfree(dfitem);
 144         }
 145 }
 146 EXPORT_SYMBOL_GPL(switchdev_deferred_process);
 147
 148 static void switchdev_deferred_process_work(struct work_struct *work)
 149 {
 150         rtnl_lock();
 151         switchdev_deferred_process();
 152         rtnl_unlock();
 153 }
 154
 155 static DECLARE_WORK(deferred_process_work, switchdev_deferred_process_work);
 156
 157 static int switchdev_deferred_enqueue(struct net_device *dev,
 158                                       const void *data, size_t data_len,
 159                                       switchdev_deferred_func_t *func)
 160 {
 161         struct switchdev_deferred_item *dfitem;
 162
 163         dfitem = kmalloc(sizeof(*dfitem) + data_len, GFP_ATOMIC);
 164         if (!dfitem)
 165                 return -ENOMEM;
 166         dfitem->dev = dev;
 167         dfitem->func = func;
 168         memcpy(dfitem->data, data, data_len);
 169         dev_hold(dev);
 170         spin_lock_bh(&deferred_lock);
 171         list_add_tail(&dfitem->list, &deferred);
 172         spin_unlock_bh(&deferred_lock);
 173         schedule_work(&deferred_process_work);
 174         return 0;
 175 }
 176
 177 /**
 178  *      switchdev_port_attr_get - Get port attribute
 179  *
 180  *      @dev: port device
 181  *      @attr: attribute to get
 182  */
 183 int switchdev_port_attr_get(struct net_device *dev, struct switchdev_attr *attr)
 184 {
 185         const struct switchdev_ops *ops = dev->switchdev_ops;
 186         struct net_device *lower_dev;
 187         struct list_head *iter;
 188         struct switchdev_attr first = {
 189                 .id = SWITCHDEV_ATTR_ID_UNDEFINED
 190         };
 191         int err = -EOPNOTSUPP;
 192
 193         if (ops && ops->switchdev_port_attr_get)
 194                 return ops->switchdev_port_attr_get(dev, attr);
 195
 196         if (attr->flags & SWITCHDEV_F_NO_RECURSE)
 197                 return err;
 198
 199         /* Switch device port(s) may be stacked under
 200          * bond/team/vlan dev, so recurse down to get attr on
 201          * each port.  Return -ENODATA if attr values don't
 202          * compare across ports.
 203          */
 204
 205         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 206                 err = switchdev_port_attr_get(lower_dev, attr);
 207                 if (err)
 208                         break;
 209                 if (first.id == SWITCHDEV_ATTR_ID_UNDEFINED)
 210                         first = *attr;
 211                 else if (memcmp(&first, attr, sizeof(*attr)))
 212                         return -ENODATA;
 213         }
 214
 215         return err;
 216 }
 217 EXPORT_SYMBOL_GPL(switchdev_port_attr_get);
 218
 219 static int __switchdev_port_attr_set(struct net_device *dev,
 220                                      const struct switchdev_attr *attr,
 221                                      struct switchdev_trans *trans)
 222 {
 223         const struct switchdev_ops *ops = dev->switchdev_ops;
 224         struct net_device *lower_dev;
 225         struct list_head *iter;
 226         int err = -EOPNOTSUPP;
 227
 228         if (ops && ops->switchdev_port_attr_set) {
 229                 err = ops->switchdev_port_attr_set(dev, attr, trans);
 230                 goto done;
 231         }
 232
 233         if (attr->flags & SWITCHDEV_F_NO_RECURSE)
 234                 goto done;
 235
 236         /* Switch device port(s) may be stacked under
 237          * bond/team/vlan dev, so recurse down to set attr on
 238          * each port.
 239          */
 240
 241         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 242                 err = __switchdev_port_attr_set(lower_dev, attr, trans);
 243                 if (err)
 244                         break;
 245         }
 246
 247 done:
 248         if (err == -EOPNOTSUPP && attr->flags & SWITCHDEV_F_SKIP_EOPNOTSUPP)
 249                 err = 0;
 250
 251         return err;
 252 }
 253
 254 static int switchdev_port_attr_set_now(struct net_device *dev,
 255                                        const struct switchdev_attr *attr)
 256 {
 257         struct switchdev_trans trans;
 258         int err;
 259
 260         switchdev_trans_init(&trans);
 261
 262         /* Phase I: prepare for attr set. Driver/device should fail
 263          * here if there are going to be issues in the commit phase,
 264          * such as lack of resources or support.  The driver/device
 265          * should reserve resources needed for the commit phase here,
 266          * but should not commit the attr.
 267          */
 268
 269         trans.ph_prepare = true;
 270         err = __switchdev_port_attr_set(dev, attr, &trans);
 271         if (err) {
 272                 /* Prepare phase failed: abort the transaction.  Any
 273                  * resources reserved in the prepare phase are
 274                  * released.
 275                  */
 276
 277                 if (err != -EOPNOTSUPP)
 278                         switchdev_trans_items_destroy(&trans);
 279
 280                 return err;
 281         }
 282
 283         /* Phase II: commit attr set.  This cannot fail as a fault
 284          * of driver/device.  If it does, it's a bug in the driver/device
 285          * because the driver said everythings was OK in phase I.
 286          */
 287
 288         trans.ph_prepare = false;
 289         err = __switchdev_port_attr_set(dev, attr, &trans);
 290         WARN(err, "%s: Commit of attribute (id=%d) failed.\n",
 291              dev->name, attr->id);
 292         switchdev_trans_items_warn_destroy(dev, &trans);
 293
 294         return err;
 295 }
 296
 297 static void switchdev_port_attr_set_deferred(struct net_device *dev,
 298                                              const void *data)
 299 {
 300         const struct switchdev_attr *attr = data;
 301         int err;
 302
 303         err = switchdev_port_attr_set_now(dev, attr);
 304         if (err && err != -EOPNOTSUPP)
 305                 netdev_err(dev, "failed (err=%d) to set attribute (id=%d)\n",
 306                            err, attr->id);
 307 }
 308
 309 static int switchdev_port_attr_set_defer(struct net_device *dev,
 310                                          const struct switchdev_attr *attr)
 311 {
 312         return switchdev_deferred_enqueue(dev, attr, sizeof(*attr),
 313                                           switchdev_port_attr_set_deferred);
 314 }
 315
 316 /**
 317  *      switchdev_port_attr_set - Set port attribute
 318  *
 319  *      @dev: port device
 320  *      @attr: attribute to set
 321  *
 322  *      Use a 2-phase prepare-commit transaction model to ensure
 323  *      system is not left in a partially updated state due to
 324  *      failure from driver/device.
 325  *
 326  *      rtnl_lock must be held and must not be in atomic section,
 327  *      in case SWITCHDEV_F_DEFER flag is not set.
 328  */
 329 int switchdev_port_attr_set(struct net_device *dev,
 330                             const struct switchdev_attr *attr)
 331 {
 332         if (attr->flags & SWITCHDEV_F_DEFER)
 333                 return switchdev_port_attr_set_defer(dev, attr);
 334         ASSERT_RTNL();
 335         return switchdev_port_attr_set_now(dev, attr);
 336 }
 337 EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
 338
 339 static size_t switchdev_obj_size(const struct switchdev_obj *obj)
 340 {
 341         switch (obj->id) {
 342         case SWITCHDEV_OBJ_ID_PORT_VLAN:
 343                 return sizeof(struct switchdev_obj_port_vlan);
 344         case SWITCHDEV_OBJ_ID_IPV4_FIB:
 345                 return sizeof(struct switchdev_obj_ipv4_fib);
 346         case SWITCHDEV_OBJ_ID_PORT_FDB:
 347                 return sizeof(struct switchdev_obj_port_fdb);
 348         default:
 349                 BUG();
 350         }
 351         return 0;
 352 }
 353
 354 static int __switchdev_port_obj_add(struct net_device *dev,
 355                                     const struct switchdev_obj *obj,
 356                                     struct switchdev_trans *trans)
 357 {
 358         const struct switchdev_ops *ops = dev->switchdev_ops;
 359         struct net_device *lower_dev;
 360         struct list_head *iter;
 361         int err = -EOPNOTSUPP;
 362
 363         if (ops && ops->switchdev_port_obj_add)
 364                 return ops->switchdev_port_obj_add(dev, obj, trans);
 365
 366         /* Switch device port(s) may be stacked under
 367          * bond/team/vlan dev, so recurse down to add object on
 368          * each port.
 369          */
 370
 371         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 372                 err = __switchdev_port_obj_add(lower_dev, obj, trans);
 373                 if (err)
 374                         break;
 375         }
 376
 377         return err;
 378 }
 379
 380 static int switchdev_port_obj_add_now(struct net_device *dev,
 381                                       const struct switchdev_obj *obj)
 382 {
 383         struct switchdev_trans trans;
 384         int err;
 385
 386         ASSERT_RTNL();
 387
 388         switchdev_trans_init(&trans);
 389
 390         /* Phase I: prepare for obj add. Driver/device should fail
 391          * here if there are going to be issues in the commit phase,
 392          * such as lack of resources or support.  The driver/device
 393          * should reserve resources needed for the commit phase here,
 394          * but should not commit the obj.
 395          */
 396
 397         trans.ph_prepare = true;
 398         err = __switchdev_port_obj_add(dev, obj, &trans);
 399         if (err) {
 400                 /* Prepare phase failed: abort the transaction.  Any
 401                  * resources reserved in the prepare phase are
 402                  * released.
 403                  */
 404
 405                 if (err != -EOPNOTSUPP)
 406                         switchdev_trans_items_destroy(&trans);
 407
 408                 return err;
 409         }
 410
 411         /* Phase II: commit obj add.  This cannot fail as a fault
 412          * of driver/device.  If it does, it's a bug in the driver/device
 413          * because the driver said everythings was OK in phase I.
 414          */
 415
 416         trans.ph_prepare = false;
 417         err = __switchdev_port_obj_add(dev, obj, &trans);
 418         WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->id);
 419         switchdev_trans_items_warn_destroy(dev, &trans);
 420
 421         return err;
 422 }
 423
 424 static void switchdev_port_obj_add_deferred(struct net_device *dev,
 425                                             const void *data)
 426 {
 427         const struct switchdev_obj *obj = data;
 428         int err;
 429
 430         err = switchdev_port_obj_add_now(dev, obj);
 431         if (err && err != -EOPNOTSUPP)
 432                 netdev_err(dev, "failed (err=%d) to add object (id=%d)\n",
 433                            err, obj->id);
 434 }
 435
 436 static int switchdev_port_obj_add_defer(struct net_device *dev,
 437                                         const struct switchdev_obj *obj)
 438 {
 439         return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
 440                                           switchdev_port_obj_add_deferred);
 441 }
 442
 443 /**
 444  *      switchdev_port_obj_add - Add port object
 445  *
 446  *      @dev: port device
 447  *      @id: object ID
 448  *      @obj: object to add
 449  *
 450  *      Use a 2-phase prepare-commit transaction model to ensure
 451  *      system is not left in a partially updated state due to
 452  *      failure from driver/device.
 453  *
 454  *      rtnl_lock must be held and must not be in atomic section,
 455  *      in case SWITCHDEV_F_DEFER flag is not set.
 456  */
 457 int switchdev_port_obj_add(struct net_device *dev,
 458                            const struct switchdev_obj *obj)
 459 {
 460         if (obj->flags & SWITCHDEV_F_DEFER)
 461                 return switchdev_port_obj_add_defer(dev, obj);
 462         ASSERT_RTNL();
 463         return switchdev_port_obj_add_now(dev, obj);
 464 }
 465 EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
 466
 467 static int switchdev_port_obj_del_now(struct net_device *dev,
 468                                       const struct switchdev_obj *obj)
 469 {
 470         const struct switchdev_ops *ops = dev->switchdev_ops;
 471         struct net_device *lower_dev;
 472         struct list_head *iter;
 473         int err = -EOPNOTSUPP;
 474
 475         if (ops && ops->switchdev_port_obj_del)
 476                 return ops->switchdev_port_obj_del(dev, obj);
 477
 478         /* Switch device port(s) may be stacked under
 479          * bond/team/vlan dev, so recurse down to delete object on
 480          * each port.
 481          */
 482
 483         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 484                 err = switchdev_port_obj_del_now(lower_dev, obj);
 485                 if (err)
 486                         break;
 487         }
 488
 489         return err;
 490 }
 491
 492 static void switchdev_port_obj_del_deferred(struct net_device *dev,
 493                                             const void *data)
 494 {
 495         const struct switchdev_obj *obj = data;
 496         int err;
 497
 498         err = switchdev_port_obj_del_now(dev, obj);
 499         if (err && err != -EOPNOTSUPP)
 500                 netdev_err(dev, "failed (err=%d) to del object (id=%d)\n",
 501                            err, obj->id);
 502 }
 503
 504 static int switchdev_port_obj_del_defer(struct net_device *dev,
 505                                         const struct switchdev_obj *obj)
 506 {
 507         return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
 508                                           switchdev_port_obj_del_deferred);
 509 }
 510
 511 /**
 512  *      switchdev_port_obj_del - Delete port object
 513  *
 514  *      @dev: port device
 515  *      @id: object ID
 516  *      @obj: object to delete
 517  *
 518  *      rtnl_lock must be held and must not be in atomic section,
 519  *      in case SWITCHDEV_F_DEFER flag is not set.
 520  */
 521 int switchdev_port_obj_del(struct net_device *dev,
 522                            const struct switchdev_obj *obj)
 523 {
 524         if (obj->flags & SWITCHDEV_F_DEFER)
 525                 return switchdev_port_obj_del_defer(dev, obj);
 526         ASSERT_RTNL();
 527         return switchdev_port_obj_del_now(dev, obj);
 528 }
 529 EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
 530
 531 /**
 532  *      switchdev_port_obj_dump - Dump port objects
 533  *
 534  *      @dev: port device
 535  *      @id: object ID
 536  *      @obj: object to dump
 537  *      @cb: function to call with a filled object
 538  *
 539  *      rtnl_lock must be held.
 540  */
 541 int switchdev_port_obj_dump(struct net_device *dev, struct switchdev_obj *obj,
 542                             switchdev_obj_dump_cb_t *cb)
 543 {
 544         const struct switchdev_ops *ops = dev->switchdev_ops;
 545         struct net_device *lower_dev;
 546         struct list_head *iter;
 547         int err = -EOPNOTSUPP;
 548
 549         ASSERT_RTNL();
 550
 551         if (ops && ops->switchdev_port_obj_dump)
 552                 return ops->switchdev_port_obj_dump(dev, obj, cb);
 553
 554         /* Switch device port(s) may be stacked under
 555          * bond/team/vlan dev, so recurse down to dump objects on
 556          * first port at bottom of stack.
 557          */
 558
 559         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 560                 err = switchdev_port_obj_dump(lower_dev, obj, cb);
 561                 break;
 562         }
 563
 564         return err;
 565 }
 566 EXPORT_SYMBOL_GPL(switchdev_port_obj_dump);
 567
 568 static DEFINE_MUTEX(switchdev_mutex);
 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         mutex_lock(&switchdev_mutex);
 584         err = raw_notifier_chain_register(&switchdev_notif_chain, nb);
 585         mutex_unlock(&switchdev_mutex);
 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         mutex_lock(&switchdev_mutex);
 602         err = raw_notifier_chain_unregister(&switchdev_notif_chain, nb);
 603         mutex_unlock(&switchdev_mutex);
 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  */
 618 int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
 619                              struct switchdev_notifier_info *info)
 620 {
 621         int err;
 622
 623         info->dev = dev;
 624         mutex_lock(&switchdev_mutex);
 625         err = raw_notifier_call_chain(&switchdev_notif_chain, val, info);
 626         mutex_unlock(&switchdev_mutex);
 627         return err;
 628 }
 629 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
 630
 631 struct switchdev_vlan_dump {
 632         struct switchdev_obj_port_vlan vlan;
 633         struct sk_buff *skb;
 634         u32 filter_mask;
 635         u16 flags;
 636         u16 begin;
 637         u16 end;
 638 };
 639
 640 static int switchdev_port_vlan_dump_put(struct switchdev_vlan_dump *dump)
 641 {
 642         struct bridge_vlan_info vinfo;
 643
 644         vinfo.flags = dump->flags;
 645
 646         if (dump->begin == 0 && dump->end == 0) {
 647                 return 0;
 648         } else if (dump->begin == dump->end) {
 649                 vinfo.vid = dump->begin;
 650                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 651                             sizeof(vinfo), &vinfo))
 652                         return -EMSGSIZE;
 653         } else {
 654                 vinfo.vid = dump->begin;
 655                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_BEGIN;
 656                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 657                             sizeof(vinfo), &vinfo))
 658                         return -EMSGSIZE;
 659                 vinfo.vid = dump->end;
 660                 vinfo.flags &= ~BRIDGE_VLAN_INFO_RANGE_BEGIN;
 661                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_END;
 662                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 663                             sizeof(vinfo), &vinfo))
 664                         return -EMSGSIZE;
 665         }
 666
 667         return 0;
 668 }
 669
 670 static int switchdev_port_vlan_dump_cb(struct switchdev_obj *obj)
 671 {
 672         struct switchdev_obj_port_vlan *vlan = SWITCHDEV_OBJ_PORT_VLAN(obj);
 673         struct switchdev_vlan_dump *dump =
 674                 container_of(vlan, struct switchdev_vlan_dump, vlan);
 675         int err = 0;
 676
 677         if (vlan->vid_begin > vlan->vid_end)
 678                 return -EINVAL;
 679
 680         if (dump->filter_mask & RTEXT_FILTER_BRVLAN) {
 681                 dump->flags = vlan->flags;
 682                 for (dump->begin = dump->end = vlan->vid_begin;
 683                      dump->begin <= vlan->vid_end;
 684                      dump->begin++, dump->end++) {
 685                         err = switchdev_port_vlan_dump_put(dump);
 686                         if (err)
 687                                 return err;
 688                 }
 689         } else if (dump->filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED) {
 690                 if (dump->begin > vlan->vid_begin &&
 691                     dump->begin >= vlan->vid_end) {
 692                         if ((dump->begin - 1) == vlan->vid_end &&
 693                             dump->flags == vlan->flags) {
 694                                 /* prepend */
 695                                 dump->begin = vlan->vid_begin;
 696                         } else {
 697                                 err = switchdev_port_vlan_dump_put(dump);
 698                                 dump->flags = vlan->flags;
 699                                 dump->begin = vlan->vid_begin;
 700                                 dump->end = vlan->vid_end;
 701                         }
 702                 } else if (dump->end <= vlan->vid_begin &&
 703                            dump->end < vlan->vid_end) {
 704                         if ((dump->end  + 1) == vlan->vid_begin &&
 705                             dump->flags == vlan->flags) {
 706                                 /* append */
 707                                 dump->end = vlan->vid_end;
 708                         } else {
 709                                 err = switchdev_port_vlan_dump_put(dump);
 710                                 dump->flags = vlan->flags;
 711                                 dump->begin = vlan->vid_begin;
 712                                 dump->end = vlan->vid_end;
 713                         }
 714                 } else {
 715                         err = -EINVAL;
 716                 }
 717         }
 718
 719         return err;
 720 }
 721
 722 static int switchdev_port_vlan_fill(struct sk_buff *skb, struct net_device *dev,
 723                                     u32 filter_mask)
 724 {
 725         struct switchdev_vlan_dump dump = {
 726                 .vlan.obj.orig_dev = dev,
 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                 .orig_dev = dev,
 762                 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
 763         };
 764         u16 mode = BRIDGE_MODE_UNDEF;
 765         u32 mask = BR_LEARNING | BR_LEARNING_SYNC | BR_FLOOD;
 766         int err;
 767
 768         err = switchdev_port_attr_get(dev, &attr);
 769         if (err && err != -EOPNOTSUPP)
 770                 return err;
 771
 772         return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode,
 773                                        attr.u.brport_flags, mask, nlflags,
 774                                        filter_mask, switchdev_port_vlan_fill);
 775 }
 776 EXPORT_SYMBOL_GPL(switchdev_port_bridge_getlink);
 777
 778 static int switchdev_port_br_setflag(struct net_device *dev,
 779                                      struct nlattr *nlattr,
 780                                      unsigned long brport_flag)
 781 {
 782         struct switchdev_attr attr = {
 783                 .orig_dev = dev,
 784                 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
 785         };
 786         u8 flag = nla_get_u8(nlattr);
 787         int err;
 788
 789         err = switchdev_port_attr_get(dev, &attr);
 790         if (err)
 791                 return err;
 792
 793         if (flag)
 794                 attr.u.brport_flags |= brport_flag;
 795         else
 796                 attr.u.brport_flags &= ~brport_flag;
 797
 798         return switchdev_port_attr_set(dev, &attr);
 799 }
 800
 801 static const struct nla_policy
 802 switchdev_port_bridge_policy[IFLA_BRPORT_MAX + 1] = {
 803         [IFLA_BRPORT_STATE]             = { .type = NLA_U8 },
 804         [IFLA_BRPORT_COST]              = { .type = NLA_U32 },
 805         [IFLA_BRPORT_PRIORITY]          = { .type = NLA_U16 },
 806         [IFLA_BRPORT_MODE]              = { .type = NLA_U8 },
 807         [IFLA_BRPORT_GUARD]             = { .type = NLA_U8 },
 808         [IFLA_BRPORT_PROTECT]           = { .type = NLA_U8 },
 809         [IFLA_BRPORT_FAST_LEAVE]        = { .type = NLA_U8 },
 810         [IFLA_BRPORT_LEARNING]          = { .type = NLA_U8 },
 811         [IFLA_BRPORT_LEARNING_SYNC]     = { .type = NLA_U8 },
 812         [IFLA_BRPORT_UNICAST_FLOOD]     = { .type = NLA_U8 },
 813 };
 814
 815 static int switchdev_port_br_setlink_protinfo(struct net_device *dev,
 816                                               struct nlattr *protinfo)
 817 {
 818         struct nlattr *attr;
 819         int rem;
 820         int err;
 821
 822         err = nla_validate_nested(protinfo, IFLA_BRPORT_MAX,
 823                                   switchdev_port_bridge_policy);
 824         if (err)
 825                 return err;
 826
 827         nla_for_each_nested(attr, protinfo, rem) {
 828                 switch (nla_type(attr)) {
 829                 case IFLA_BRPORT_LEARNING:
 830                         err = switchdev_port_br_setflag(dev, attr,
 831                                                         BR_LEARNING);
 832                         break;
 833                 case IFLA_BRPORT_LEARNING_SYNC:
 834                         err = switchdev_port_br_setflag(dev, attr,
 835                                                         BR_LEARNING_SYNC);
 836                         break;
 837                 case IFLA_BRPORT_UNICAST_FLOOD:
 838                         err = switchdev_port_br_setflag(dev, attr, BR_FLOOD);
 839                         break;
 840                 default:
 841                         err = -EOPNOTSUPP;
 842                         break;
 843                 }
 844                 if (err)
 845                         return err;
 846         }
 847
 848         return 0;
 849 }
 850
 851 static int switchdev_port_br_afspec(struct net_device *dev,
 852                                     struct nlattr *afspec,
 853                                     int (*f)(struct net_device *dev,
 854                                              const struct switchdev_obj *obj))
 855 {
 856         struct nlattr *attr;
 857         struct bridge_vlan_info *vinfo;
 858         struct switchdev_obj_port_vlan vlan = {
 859                 .obj.orig_dev = dev,
 860                 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
 861         };
 862         int rem;
 863         int err;
 864
 865         nla_for_each_nested(attr, afspec, rem) {
 866                 if (nla_type(attr) != IFLA_BRIDGE_VLAN_INFO)
 867                         continue;
 868                 if (nla_len(attr) != sizeof(struct bridge_vlan_info))
 869                         return -EINVAL;
 870                 vinfo = nla_data(attr);
 871                 if (!vinfo->vid || vinfo->vid >= VLAN_VID_MASK)
 872                         return -EINVAL;
 873                 vlan.flags = vinfo->flags;
 874                 if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_BEGIN) {
 875                         if (vlan.vid_begin)
 876                                 return -EINVAL;
 877                         vlan.vid_begin = vinfo->vid;
 878                         /* don't allow range of pvids */
 879                         if (vlan.flags & BRIDGE_VLAN_INFO_PVID)
 880                                 return -EINVAL;
 881                 } else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
 882                         if (!vlan.vid_begin)
 883                                 return -EINVAL;
 884                         vlan.vid_end = vinfo->vid;
 885                         if (vlan.vid_end <= vlan.vid_begin)
 886                                 return -EINVAL;
 887                         err = f(dev, &vlan.obj);
 888                         if (err)
 889                                 return err;
 890                         vlan.vid_begin = 0;
 891                 } else {
 892                         if (vlan.vid_begin)
 893                                 return -EINVAL;
 894                         vlan.vid_begin = vinfo->vid;
 895                         vlan.vid_end = vinfo->vid;
 896                         err = f(dev, &vlan.obj);
 897                         if (err)
 898                                 return err;
 899                         vlan.vid_begin = 0;
 900                 }
 901         }
 902
 903         return 0;
 904 }
 905
 906 /**
 907  *      switchdev_port_bridge_setlink - Set bridge port attributes
 908  *
 909  *      @dev: port device
 910  *      @nlh: netlink header
 911  *      @flags: netlink flags
 912  *
 913  *      Called for SELF on rtnl_bridge_setlink to set bridge port
 914  *      attributes.
 915  */
 916 int switchdev_port_bridge_setlink(struct net_device *dev,
 917                                   struct nlmsghdr *nlh, u16 flags)
 918 {
 919         struct nlattr *protinfo;
 920         struct nlattr *afspec;
 921         int err = 0;
 922
 923         protinfo = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 924                                    IFLA_PROTINFO);
 925         if (protinfo) {
 926                 err = switchdev_port_br_setlink_protinfo(dev, protinfo);
 927                 if (err)
 928                         return err;
 929         }
 930
 931         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 932                                  IFLA_AF_SPEC);
 933         if (afspec)
 934                 err = switchdev_port_br_afspec(dev, afspec,
 935                                                switchdev_port_obj_add);
 936
 937         return err;
 938 }
 939 EXPORT_SYMBOL_GPL(switchdev_port_bridge_setlink);
 940
 941 /**
 942  *      switchdev_port_bridge_dellink - Set bridge port attributes
 943  *
 944  *      @dev: port device
 945  *      @nlh: netlink header
 946  *      @flags: netlink flags
 947  *
 948  *      Called for SELF on rtnl_bridge_dellink to set bridge port
 949  *      attributes.
 950  */
 951 int switchdev_port_bridge_dellink(struct net_device *dev,
 952                                   struct nlmsghdr *nlh, u16 flags)
 953 {
 954         struct nlattr *afspec;
 955
 956         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 957                                  IFLA_AF_SPEC);
 958         if (afspec)
 959                 return switchdev_port_br_afspec(dev, afspec,
 960                                                 switchdev_port_obj_del);
 961
 962         return 0;
 963 }
 964 EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
 965
 966 /**
 967  *      switchdev_port_fdb_add - Add FDB (MAC/VLAN) entry to port
 968  *
 969  *      @ndmsg: netlink hdr
 970  *      @nlattr: netlink attributes
 971  *      @dev: port device
 972  *      @addr: MAC address to add
 973  *      @vid: VLAN to add
 974  *
 975  *      Add FDB entry to switch device.
 976  */
 977 int switchdev_port_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 978                            struct net_device *dev, const unsigned char *addr,
 979                            u16 vid, u16 nlm_flags)
 980 {
 981         struct switchdev_obj_port_fdb fdb = {
 982                 .obj.orig_dev = dev,
 983                 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
 984                 .vid = vid,
 985         };
 986
 987         ether_addr_copy(fdb.addr, addr);
 988         return switchdev_port_obj_add(dev, &fdb.obj);
 989 }
 990 EXPORT_SYMBOL_GPL(switchdev_port_fdb_add);
 991
 992 /**
 993  *      switchdev_port_fdb_del - Delete FDB (MAC/VLAN) entry from port
 994  *
 995  *      @ndmsg: netlink hdr
 996  *      @nlattr: netlink attributes
 997  *      @dev: port device
 998  *      @addr: MAC address to delete
 999  *      @vid: VLAN to delete
1000  *
1001  *      Delete FDB entry from switch device.
1002  */
1003 int switchdev_port_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
1004                            struct net_device *dev, const unsigned char *addr,
1005                            u16 vid)
1006 {
1007         struct switchdev_obj_port_fdb fdb = {
1008                 .obj.orig_dev = dev,
1009                 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
1010                 .vid = vid,
1011         };
1012
1013         ether_addr_copy(fdb.addr, addr);
1014         return switchdev_port_obj_del(dev, &fdb.obj);
1015 }
1016 EXPORT_SYMBOL_GPL(switchdev_port_fdb_del);
1017
1018 struct switchdev_fdb_dump {
1019         struct switchdev_obj_port_fdb fdb;
1020         struct net_device *dev;
1021         struct sk_buff *skb;
1022         struct netlink_callback *cb;
1023         int idx;
1024 };
1025
1026 static int switchdev_port_fdb_dump_cb(struct switchdev_obj *obj)
1027 {
1028         struct switchdev_obj_port_fdb *fdb = SWITCHDEV_OBJ_PORT_FDB(obj);
1029         struct switchdev_fdb_dump *dump =
1030                 container_of(fdb, struct switchdev_fdb_dump, fdb);
1031         u32 portid = NETLINK_CB(dump->cb->skb).portid;
1032         u32 seq = dump->cb->nlh->nlmsg_seq;
1033         struct nlmsghdr *nlh;
1034         struct ndmsg *ndm;
1035
1036         if (dump->idx < dump->cb->args[0])
1037                 goto skip;
1038
1039         nlh = nlmsg_put(dump->skb, portid, seq, RTM_NEWNEIGH,
1040                         sizeof(*ndm), NLM_F_MULTI);
1041         if (!nlh)
1042                 return -EMSGSIZE;
1043
1044         ndm = nlmsg_data(nlh);
1045         ndm->ndm_family  = AF_BRIDGE;
1046         ndm->ndm_pad1    = 0;
1047         ndm->ndm_pad2    = 0;
1048         ndm->ndm_flags   = NTF_SELF;
1049         ndm->ndm_type    = 0;
1050         ndm->ndm_ifindex = dump->dev->ifindex;
1051         ndm->ndm_state   = fdb->ndm_state;
1052
1053         if (nla_put(dump->skb, NDA_LLADDR, ETH_ALEN, fdb->addr))
1054                 goto nla_put_failure;
1055
1056         if (fdb->vid && nla_put_u16(dump->skb, NDA_VLAN, fdb->vid))
1057                 goto nla_put_failure;
1058
1059         nlmsg_end(dump->skb, nlh);
1060
1061 skip:
1062         dump->idx++;
1063         return 0;
1064
1065 nla_put_failure:
1066         nlmsg_cancel(dump->skb, nlh);
1067         return -EMSGSIZE;
1068 }
1069
1070 /**
1071  *      switchdev_port_fdb_dump - Dump port FDB (MAC/VLAN) entries
1072  *
1073  *      @skb: netlink skb
1074  *      @cb: netlink callback
1075  *      @dev: port device
1076  *      @filter_dev: filter device
1077  *      @idx:
1078  *
1079  *      Delete FDB entry from switch device.
1080  */
1081 int switchdev_port_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1082                             struct net_device *dev,
1083                             struct net_device *filter_dev, int idx)
1084 {
1085         struct switchdev_fdb_dump dump = {
1086                 .fdb.obj.orig_dev = dev,
1087                 .fdb.obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
1088                 .dev = dev,
1089                 .skb = skb,
1090                 .cb = cb,
1091                 .idx = idx,
1092         };
1093
1094         switchdev_port_obj_dump(dev, &dump.fdb.obj, switchdev_port_fdb_dump_cb);
1095         return dump.idx;
1096 }
1097 EXPORT_SYMBOL_GPL(switchdev_port_fdb_dump);
1098
1099 static struct net_device *switchdev_get_lowest_dev(struct net_device *dev)
1100 {
1101         const struct switchdev_ops *ops = dev->switchdev_ops;
1102         struct net_device *lower_dev;
1103         struct net_device *port_dev;
1104         struct list_head *iter;
1105
1106         /* Recusively search down until we find a sw port dev.
1107          * (A sw port dev supports switchdev_port_attr_get).
1108          */
1109
1110         if (ops && ops->switchdev_port_attr_get)
1111                 return dev;
1112
1113         netdev_for_each_lower_dev(dev, lower_dev, iter) {
1114                 port_dev = switchdev_get_lowest_dev(lower_dev);
1115                 if (port_dev)
1116                         return port_dev;
1117         }
1118
1119         return NULL;
1120 }
1121
1122 static struct net_device *switchdev_get_dev_by_nhs(struct fib_info *fi)
1123 {
1124         struct switchdev_attr attr = {
1125                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1126         };
1127         struct switchdev_attr prev_attr;
1128         struct net_device *dev = NULL;
1129         int nhsel;
1130
1131         ASSERT_RTNL();
1132
1133         /* For this route, all nexthop devs must be on the same switch. */
1134
1135         for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1136                 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1137
1138                 if (!nh->nh_dev)
1139                         return NULL;
1140
1141                 dev = switchdev_get_lowest_dev(nh->nh_dev);
1142                 if (!dev)
1143                         return NULL;
1144
1145                 attr.orig_dev = dev;
1146                 if (switchdev_port_attr_get(dev, &attr))
1147                         return NULL;
1148
1149                 if (nhsel > 0 &&
1150                     !netdev_phys_item_id_same(&prev_attr.u.ppid, &attr.u.ppid))
1151                                 return NULL;
1152
1153                 prev_attr = attr;
1154         }
1155
1156         return dev;
1157 }
1158
1159 /**
1160  *      switchdev_fib_ipv4_add - Add/modify switch IPv4 route entry
1161  *
1162  *      @dst: route's IPv4 destination address
1163  *      @dst_len: destination address length (prefix length)
1164  *      @fi: route FIB info structure
1165  *      @tos: route TOS
1166  *      @type: route type
1167  *      @nlflags: netlink flags passed in (NLM_F_*)
1168  *      @tb_id: route table ID
1169  *
1170  *      Add/modify switch IPv4 route entry.
1171  */
1172 int switchdev_fib_ipv4_add(u32 dst, int dst_len, struct fib_info *fi,
1173                            u8 tos, u8 type, u32 nlflags, u32 tb_id)
1174 {
1175         struct switchdev_obj_ipv4_fib ipv4_fib = {
1176                 .obj.id = SWITCHDEV_OBJ_ID_IPV4_FIB,
1177                 .dst = dst,
1178                 .dst_len = dst_len,
1179                 .tos = tos,
1180                 .type = type,
1181                 .nlflags = nlflags,
1182                 .tb_id = tb_id,
1183         };
1184         struct net_device *dev;
1185         int err = 0;
1186
1187         memcpy(&ipv4_fib.fi, fi, sizeof(ipv4_fib.fi));
1188
1189         /* Don't offload route if using custom ip rules or if
1190          * IPv4 FIB offloading has been disabled completely.
1191          */
1192
1193 #ifdef CONFIG_IP_MULTIPLE_TABLES
1194         if (fi->fib_net->ipv4.fib_has_custom_rules)
1195                 return 0;
1196 #endif
1197
1198         if (fi->fib_net->ipv4.fib_offload_disabled)
1199                 return 0;
1200
1201         dev = switchdev_get_dev_by_nhs(fi);
1202         if (!dev)
1203                 return 0;
1204
1205         ipv4_fib.obj.orig_dev = dev;
1206         err = switchdev_port_obj_add(dev, &ipv4_fib.obj);
1207         if (!err)
1208                 fi->fib_flags |= RTNH_F_OFFLOAD;
1209
1210         return err == -EOPNOTSUPP ? 0 : err;
1211 }
1212 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_add);
1213
1214 /**
1215  *      switchdev_fib_ipv4_del - Delete IPv4 route entry from switch
1216  *
1217  *      @dst: route's IPv4 destination address
1218  *      @dst_len: destination address length (prefix length)
1219  *      @fi: route FIB info structure
1220  *      @tos: route TOS
1221  *      @type: route type
1222  *      @tb_id: route table ID
1223  *
1224  *      Delete IPv4 route entry from switch device.
1225  */
1226 int switchdev_fib_ipv4_del(u32 dst, int dst_len, struct fib_info *fi,
1227                            u8 tos, u8 type, u32 tb_id)
1228 {
1229         struct switchdev_obj_ipv4_fib ipv4_fib = {
1230                 .obj.id = SWITCHDEV_OBJ_ID_IPV4_FIB,
1231                 .dst = dst,
1232                 .dst_len = dst_len,
1233                 .tos = tos,
1234                 .type = type,
1235                 .nlflags = 0,
1236                 .tb_id = tb_id,
1237         };
1238         struct net_device *dev;
1239         int err = 0;
1240
1241         memcpy(&ipv4_fib.fi, fi, sizeof(ipv4_fib.fi));
1242
1243         if (!(fi->fib_flags & RTNH_F_OFFLOAD))
1244                 return 0;
1245
1246         dev = switchdev_get_dev_by_nhs(fi);
1247         if (!dev)
1248                 return 0;
1249
1250         ipv4_fib.obj.orig_dev = dev;
1251         err = switchdev_port_obj_del(dev, &ipv4_fib.obj);
1252         if (!err)
1253                 fi->fib_flags &= ~RTNH_F_OFFLOAD;
1254
1255         return err == -EOPNOTSUPP ? 0 : err;
1256 }
1257 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_del);
1258
1259 /**
1260  *      switchdev_fib_ipv4_abort - Abort an IPv4 FIB operation
1261  *
1262  *      @fi: route FIB info structure
1263  */
1264 void switchdev_fib_ipv4_abort(struct fib_info *fi)
1265 {
1266         /* There was a problem installing this route to the offload
1267          * device.  For now, until we come up with more refined
1268          * policy handling, abruptly end IPv4 fib offloading for
1269          * for entire net by flushing offload device(s) of all
1270          * IPv4 routes, and mark IPv4 fib offloading broken from
1271          * this point forward.
1272          */
1273
1274         fib_flush_external(fi->fib_net);
1275         fi->fib_net->ipv4.fib_offload_disabled = true;
1276 }
1277 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_abort);
1278
1279 static bool switchdev_port_same_parent_id(struct net_device *a,
1280                                           struct net_device *b)
1281 {
1282         struct switchdev_attr a_attr = {
1283                 .orig_dev = a,
1284                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1285                 .flags = SWITCHDEV_F_NO_RECURSE,
1286         };
1287         struct switchdev_attr b_attr = {
1288                 .orig_dev = b,
1289                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1290                 .flags = SWITCHDEV_F_NO_RECURSE,
1291         };
1292
1293         if (switchdev_port_attr_get(a, &a_attr) ||
1294             switchdev_port_attr_get(b, &b_attr))
1295                 return false;
1296
1297         return netdev_phys_item_id_same(&a_attr.u.ppid, &b_attr.u.ppid);
1298 }
1299
1300 static u32 switchdev_port_fwd_mark_get(struct net_device *dev,
1301                                        struct net_device *group_dev)
1302 {
1303         struct net_device *lower_dev;
1304         struct list_head *iter;
1305
1306         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1307                 if (lower_dev == dev)
1308                         continue;
1309                 if (switchdev_port_same_parent_id(dev, lower_dev))
1310                         return lower_dev->offload_fwd_mark;
1311                 return switchdev_port_fwd_mark_get(dev, lower_dev);
1312         }
1313
1314         return dev->ifindex;
1315 }
1316
1317 static void switchdev_port_fwd_mark_reset(struct net_device *group_dev,
1318                                           u32 old_mark, u32 *reset_mark)
1319 {
1320         struct net_device *lower_dev;
1321         struct list_head *iter;
1322
1323         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1324                 if (lower_dev->offload_fwd_mark == old_mark) {
1325                         if (!*reset_mark)
1326                                 *reset_mark = lower_dev->ifindex;
1327                         lower_dev->offload_fwd_mark = *reset_mark;
1328                 }
1329                 switchdev_port_fwd_mark_reset(lower_dev, old_mark, reset_mark);
1330         }
1331 }
1332
1333 /**
1334  *      switchdev_port_fwd_mark_set - Set port offload forwarding mark
1335  *
1336  *      @dev: port device
1337  *      @group_dev: containing device
1338  *      @joining: true if dev is joining group; false if leaving group
1339  *
1340  *      An ungrouped port's offload mark is just its ifindex.  A grouped
1341  *      port's (member of a bridge, for example) offload mark is the ifindex
1342  *      of one of the ports in the group with the same parent (switch) ID.
1343  *      Ports on the same device in the same group will have the same mark.
1344  *
1345  *      Example:
1346  *
1347  *              br0             ifindex=9
1348  *                sw1p1         ifindex=2       mark=2
1349  *                sw1p2         ifindex=3       mark=2
1350  *                sw2p1         ifindex=4       mark=5
1351  *                sw2p2         ifindex=5       mark=5
1352  *
1353  *      If sw2p2 leaves the bridge, we'll have:
1354  *
1355  *              br0             ifindex=9
1356  *                sw1p1         ifindex=2       mark=2
1357  *                sw1p2         ifindex=3       mark=2
1358  *                sw2p1         ifindex=4       mark=4
1359  *              sw2p2           ifindex=5       mark=5
1360  */
1361 void switchdev_port_fwd_mark_set(struct net_device *dev,
1362                                  struct net_device *group_dev,
1363                                  bool joining)
1364 {
1365         u32 mark = dev->ifindex;
1366         u32 reset_mark = 0;
1367
1368         if (group_dev) {
1369                 ASSERT_RTNL();
1370                 if (joining)
1371                         mark = switchdev_port_fwd_mark_get(dev, group_dev);
1372                 else if (dev->offload_fwd_mark == mark)
1373                         /* Ohoh, this port was the mark reference port,
1374                          * but it's leaving the group, so reset the
1375                          * mark for the remaining ports in the group.
1376                          */
1377                         switchdev_port_fwd_mark_reset(group_dev, mark,
1378                                                       &reset_mark);
1379         }
1380
1381         dev->offload_fwd_mark = mark;
1382 }
1383 EXPORT_SYMBOL_GPL(switchdev_port_fwd_mark_set);