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/if_bridge.h>
  19 #include <linux/list.h>
  20 #include <net/ip_fib.h>
  21 #include <net/switchdev.h>
  22
  23 /**
  24  *      switchdev_trans_item_enqueue - Enqueue data item to transaction queue
  25  *
  26  *      @trans: transaction
  27  *      @data: pointer to data being queued
  28  *      @destructor: data destructor
  29  *      @tritem: transaction item being queued
  30  *
  31  *      Enqeueue data item to transaction queue. tritem is typically placed in
  32  *      cointainter pointed at by data pointer. Destructor is called on
  33  *      transaction abort and after successful commit phase in case
  34  *      the caller did not dequeue the item before.
  35  */
  36 void switchdev_trans_item_enqueue(struct switchdev_trans *trans,
  37                                   void *data, void (*destructor)(void const *),
  38                                   struct switchdev_trans_item *tritem)
  39 {
  40         tritem->data = data;
  41         tritem->destructor = destructor;
  42         list_add_tail(&tritem->list, &trans->item_list);
  43 }
  44 EXPORT_SYMBOL_GPL(switchdev_trans_item_enqueue);
  45
  46 static struct switchdev_trans_item *
  47 __switchdev_trans_item_dequeue(struct switchdev_trans *trans)
  48 {
  49         struct switchdev_trans_item *tritem;
  50
  51         if (list_empty(&trans->item_list))
  52                 return NULL;
  53         tritem = list_first_entry(&trans->item_list,
  54                                   struct switchdev_trans_item, list);
  55         list_del(&tritem->list);
  56         return tritem;
  57 }
  58
  59 /**
  60  *      switchdev_trans_item_dequeue - Dequeue data item from transaction queue
  61  *
  62  *      @trans: transaction
  63  */
  64 void *switchdev_trans_item_dequeue(struct switchdev_trans *trans)
  65 {
  66         struct switchdev_trans_item *tritem;
  67
  68         tritem = __switchdev_trans_item_dequeue(trans);
  69         BUG_ON(!tritem);
  70         return tritem->data;
  71 }
  72 EXPORT_SYMBOL_GPL(switchdev_trans_item_dequeue);
  73
  74 static void switchdev_trans_init(struct switchdev_trans *trans)
  75 {
  76         INIT_LIST_HEAD(&trans->item_list);
  77 }
  78
  79 static void switchdev_trans_items_destroy(struct switchdev_trans *trans)
  80 {
  81         struct switchdev_trans_item *tritem;
  82
  83         while ((tritem = __switchdev_trans_item_dequeue(trans)))
  84                 tritem->destructor(tritem->data);
  85 }
  86
  87 static void switchdev_trans_items_warn_destroy(struct net_device *dev,
  88                                                struct switchdev_trans *trans)
  89 {
  90         WARN(!list_empty(&trans->item_list), "%s: transaction item queue is not empty.\n",
  91              dev->name);
  92         switchdev_trans_items_destroy(trans);
  93 }
  94
  95 /**
  96  *      switchdev_port_attr_get - Get port attribute
  97  *
  98  *      @dev: port device
  99  *      @attr: attribute to get
 100  */
 101 int switchdev_port_attr_get(struct net_device *dev, struct switchdev_attr *attr)
 102 {
 103         const struct switchdev_ops *ops = dev->switchdev_ops;
 104         struct net_device *lower_dev;
 105         struct list_head *iter;
 106         struct switchdev_attr first = {
 107                 .id = SWITCHDEV_ATTR_UNDEFINED
 108         };
 109         int err = -EOPNOTSUPP;
 110
 111         if (ops && ops->switchdev_port_attr_get)
 112                 return ops->switchdev_port_attr_get(dev, attr);
 113
 114         if (attr->flags & SWITCHDEV_F_NO_RECURSE)
 115                 return err;
 116
 117         /* Switch device port(s) may be stacked under
 118          * bond/team/vlan dev, so recurse down to get attr on
 119          * each port.  Return -ENODATA if attr values don't
 120          * compare across ports.
 121          */
 122
 123         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 124                 err = switchdev_port_attr_get(lower_dev, attr);
 125                 if (err)
 126                         break;
 127                 if (first.id == SWITCHDEV_ATTR_UNDEFINED)
 128                         first = *attr;
 129                 else if (memcmp(&first, attr, sizeof(*attr)))
 130                         return -ENODATA;
 131         }
 132
 133         return err;
 134 }
 135 EXPORT_SYMBOL_GPL(switchdev_port_attr_get);
 136
 137 static int __switchdev_port_attr_set(struct net_device *dev,
 138                                      struct switchdev_attr *attr,
 139                                      struct switchdev_trans *trans)
 140 {
 141         const struct switchdev_ops *ops = dev->switchdev_ops;
 142         struct net_device *lower_dev;
 143         struct list_head *iter;
 144         int err = -EOPNOTSUPP;
 145
 146         if (ops && ops->switchdev_port_attr_set)
 147                 return ops->switchdev_port_attr_set(dev, attr, trans);
 148
 149         if (attr->flags & SWITCHDEV_F_NO_RECURSE)
 150                 return err;
 151
 152         /* Switch device port(s) may be stacked under
 153          * bond/team/vlan dev, so recurse down to set attr on
 154          * each port.
 155          */
 156
 157         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 158                 err = __switchdev_port_attr_set(lower_dev, attr, trans);
 159                 if (err)
 160                         break;
 161         }
 162
 163         return err;
 164 }
 165
 166 struct switchdev_attr_set_work {
 167         struct work_struct work;
 168         struct net_device *dev;
 169         struct switchdev_attr attr;
 170 };
 171
 172 static void switchdev_port_attr_set_work(struct work_struct *work)
 173 {
 174         struct switchdev_attr_set_work *asw =
 175                 container_of(work, struct switchdev_attr_set_work, work);
 176         int err;
 177
 178         rtnl_lock();
 179         err = switchdev_port_attr_set(asw->dev, &asw->attr);
 180         if (err && err != -EOPNOTSUPP)
 181                 netdev_err(asw->dev, "failed (err=%d) to set attribute (id=%d)\n",
 182                            err, asw->attr.id);
 183         rtnl_unlock();
 184
 185         dev_put(asw->dev);
 186         kfree(work);
 187 }
 188
 189 static int switchdev_port_attr_set_defer(struct net_device *dev,
 190                                          struct switchdev_attr *attr)
 191 {
 192         struct switchdev_attr_set_work *asw;
 193
 194         asw = kmalloc(sizeof(*asw), GFP_ATOMIC);
 195         if (!asw)
 196                 return -ENOMEM;
 197
 198         INIT_WORK(&asw->work, switchdev_port_attr_set_work);
 199
 200         dev_hold(dev);
 201         asw->dev = dev;
 202         memcpy(&asw->attr, attr, sizeof(asw->attr));
 203
 204         schedule_work(&asw->work);
 205
 206         return 0;
 207 }
 208
 209 /**
 210  *      switchdev_port_attr_set - Set port attribute
 211  *
 212  *      @dev: port device
 213  *      @attr: attribute to set
 214  *
 215  *      Use a 2-phase prepare-commit transaction model to ensure
 216  *      system is not left in a partially updated state due to
 217  *      failure from driver/device.
 218  */
 219 int switchdev_port_attr_set(struct net_device *dev, struct switchdev_attr *attr)
 220 {
 221         struct switchdev_trans trans;
 222         int err;
 223
 224         if (!rtnl_is_locked()) {
 225                 /* Running prepare-commit transaction across stacked
 226                  * devices requires nothing moves, so if rtnl_lock is
 227                  * not held, schedule a worker thread to hold rtnl_lock
 228                  * while setting attr.
 229                  */
 230
 231                 return switchdev_port_attr_set_defer(dev, attr);
 232         }
 233
 234         switchdev_trans_init(&trans);
 235
 236         /* Phase I: prepare for attr set. Driver/device should fail
 237          * here if there are going to be issues in the commit phase,
 238          * such as lack of resources or support.  The driver/device
 239          * should reserve resources needed for the commit phase here,
 240          * but should not commit the attr.
 241          */
 242
 243         trans.ph_prepare = true;
 244         err = __switchdev_port_attr_set(dev, attr, &trans);
 245         if (err) {
 246                 /* Prepare phase failed: abort the transaction.  Any
 247                  * resources reserved in the prepare phase are
 248                  * released.
 249                  */
 250
 251                 if (err != -EOPNOTSUPP)
 252                         switchdev_trans_items_destroy(&trans);
 253
 254                 return err;
 255         }
 256
 257         /* Phase II: commit attr set.  This cannot fail as a fault
 258          * of driver/device.  If it does, it's a bug in the driver/device
 259          * because the driver said everythings was OK in phase I.
 260          */
 261
 262         trans.ph_prepare = false;
 263         err = __switchdev_port_attr_set(dev, attr, &trans);
 264         WARN(err, "%s: Commit of attribute (id=%d) failed.\n",
 265              dev->name, attr->id);
 266         switchdev_trans_items_warn_destroy(dev, &trans);
 267
 268         return err;
 269 }
 270 EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
 271
 272 static int __switchdev_port_obj_add(struct net_device *dev,
 273                                     struct switchdev_obj *obj,
 274                                     struct switchdev_trans *trans)
 275 {
 276         const struct switchdev_ops *ops = dev->switchdev_ops;
 277         struct net_device *lower_dev;
 278         struct list_head *iter;
 279         int err = -EOPNOTSUPP;
 280
 281         if (ops && ops->switchdev_port_obj_add)
 282                 return ops->switchdev_port_obj_add(dev, obj, trans);
 283
 284         /* Switch device port(s) may be stacked under
 285          * bond/team/vlan dev, so recurse down to add object on
 286          * each port.
 287          */
 288
 289         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 290                 err = __switchdev_port_obj_add(lower_dev, obj, trans);
 291                 if (err)
 292                         break;
 293         }
 294
 295         return err;
 296 }
 297
 298 /**
 299  *      switchdev_port_obj_add - Add port object
 300  *
 301  *      @dev: port device
 302  *      @obj: object to add
 303  *
 304  *      Use a 2-phase prepare-commit transaction model to ensure
 305  *      system is not left in a partially updated state due to
 306  *      failure from driver/device.
 307  *
 308  *      rtnl_lock must be held.
 309  */
 310 int switchdev_port_obj_add(struct net_device *dev, struct switchdev_obj *obj)
 311 {
 312         struct switchdev_trans trans;
 313         int err;
 314
 315         ASSERT_RTNL();
 316
 317         switchdev_trans_init(&trans);
 318
 319         /* Phase I: prepare for obj add. Driver/device should fail
 320          * here if there are going to be issues in the commit phase,
 321          * such as lack of resources or support.  The driver/device
 322          * should reserve resources needed for the commit phase here,
 323          * but should not commit the obj.
 324          */
 325
 326         trans.ph_prepare = true;
 327         err = __switchdev_port_obj_add(dev, obj, &trans);
 328         if (err) {
 329                 /* Prepare phase failed: abort the transaction.  Any
 330                  * resources reserved in the prepare phase are
 331                  * released.
 332                  */
 333
 334                 if (err != -EOPNOTSUPP)
 335                         switchdev_trans_items_destroy(&trans);
 336
 337                 return err;
 338         }
 339
 340         /* Phase II: commit obj add.  This cannot fail as a fault
 341          * of driver/device.  If it does, it's a bug in the driver/device
 342          * because the driver said everythings was OK in phase I.
 343          */
 344
 345         trans.ph_prepare = false;
 346         err = __switchdev_port_obj_add(dev, obj, &trans);
 347         WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->id);
 348         switchdev_trans_items_warn_destroy(dev, &trans);
 349
 350         return err;
 351 }
 352 EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
 353
 354 /**
 355  *      switchdev_port_obj_del - Delete port object
 356  *
 357  *      @dev: port device
 358  *      @obj: object to delete
 359  */
 360 int switchdev_port_obj_del(struct net_device *dev, struct switchdev_obj *obj)
 361 {
 362         const struct switchdev_ops *ops = dev->switchdev_ops;
 363         struct net_device *lower_dev;
 364         struct list_head *iter;
 365         int err = -EOPNOTSUPP;
 366
 367         if (ops && ops->switchdev_port_obj_del)
 368                 return ops->switchdev_port_obj_del(dev, obj);
 369
 370         /* Switch device port(s) may be stacked under
 371          * bond/team/vlan dev, so recurse down to delete object on
 372          * each port.
 373          */
 374
 375         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 376                 err = switchdev_port_obj_del(lower_dev, obj);
 377                 if (err)
 378                         break;
 379         }
 380
 381         return err;
 382 }
 383 EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
 384
 385 /**
 386  *      switchdev_port_obj_dump - Dump port objects
 387  *
 388  *      @dev: port device
 389  *      @obj: object to dump
 390  */
 391 int switchdev_port_obj_dump(struct net_device *dev, struct switchdev_obj *obj)
 392 {
 393         const struct switchdev_ops *ops = dev->switchdev_ops;
 394         struct net_device *lower_dev;
 395         struct list_head *iter;
 396         int err = -EOPNOTSUPP;
 397
 398         if (ops && ops->switchdev_port_obj_dump)
 399                 return ops->switchdev_port_obj_dump(dev, obj);
 400
 401         /* Switch device port(s) may be stacked under
 402          * bond/team/vlan dev, so recurse down to dump objects on
 403          * first port at bottom of stack.
 404          */
 405
 406         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 407                 err = switchdev_port_obj_dump(lower_dev, obj);
 408                 break;
 409         }
 410
 411         return err;
 412 }
 413 EXPORT_SYMBOL_GPL(switchdev_port_obj_dump);
 414
 415 static DEFINE_MUTEX(switchdev_mutex);
 416 static RAW_NOTIFIER_HEAD(switchdev_notif_chain);
 417
 418 /**
 419  *      register_switchdev_notifier - Register notifier
 420  *      @nb: notifier_block
 421  *
 422  *      Register switch device notifier. This should be used by code
 423  *      which needs to monitor events happening in particular device.
 424  *      Return values are same as for atomic_notifier_chain_register().
 425  */
 426 int register_switchdev_notifier(struct notifier_block *nb)
 427 {
 428         int err;
 429
 430         mutex_lock(&switchdev_mutex);
 431         err = raw_notifier_chain_register(&switchdev_notif_chain, nb);
 432         mutex_unlock(&switchdev_mutex);
 433         return err;
 434 }
 435 EXPORT_SYMBOL_GPL(register_switchdev_notifier);
 436
 437 /**
 438  *      unregister_switchdev_notifier - Unregister notifier
 439  *      @nb: notifier_block
 440  *
 441  *      Unregister switch device notifier.
 442  *      Return values are same as for atomic_notifier_chain_unregister().
 443  */
 444 int unregister_switchdev_notifier(struct notifier_block *nb)
 445 {
 446         int err;
 447
 448         mutex_lock(&switchdev_mutex);
 449         err = raw_notifier_chain_unregister(&switchdev_notif_chain, nb);
 450         mutex_unlock(&switchdev_mutex);
 451         return err;
 452 }
 453 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
 454
 455 /**
 456  *      call_switchdev_notifiers - Call notifiers
 457  *      @val: value passed unmodified to notifier function
 458  *      @dev: port device
 459  *      @info: notifier information data
 460  *
 461  *      Call all network notifier blocks. This should be called by driver
 462  *      when it needs to propagate hardware event.
 463  *      Return values are same as for atomic_notifier_call_chain().
 464  */
 465 int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
 466                              struct switchdev_notifier_info *info)
 467 {
 468         int err;
 469
 470         info->dev = dev;
 471         mutex_lock(&switchdev_mutex);
 472         err = raw_notifier_call_chain(&switchdev_notif_chain, val, info);
 473         mutex_unlock(&switchdev_mutex);
 474         return err;
 475 }
 476 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
 477
 478 struct switchdev_vlan_dump {
 479         struct switchdev_obj obj;
 480         struct sk_buff *skb;
 481         u32 filter_mask;
 482         u16 flags;
 483         u16 begin;
 484         u16 end;
 485 };
 486
 487 static int switchdev_port_vlan_dump_put(struct switchdev_vlan_dump *dump)
 488 {
 489         struct bridge_vlan_info vinfo;
 490
 491         vinfo.flags = dump->flags;
 492
 493         if (dump->begin == 0 && dump->end == 0) {
 494                 return 0;
 495         } else if (dump->begin == dump->end) {
 496                 vinfo.vid = dump->begin;
 497                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 498                             sizeof(vinfo), &vinfo))
 499                         return -EMSGSIZE;
 500         } else {
 501                 vinfo.vid = dump->begin;
 502                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_BEGIN;
 503                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 504                             sizeof(vinfo), &vinfo))
 505                         return -EMSGSIZE;
 506                 vinfo.vid = dump->end;
 507                 vinfo.flags &= ~BRIDGE_VLAN_INFO_RANGE_BEGIN;
 508                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_END;
 509                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 510                             sizeof(vinfo), &vinfo))
 511                         return -EMSGSIZE;
 512         }
 513
 514         return 0;
 515 }
 516
 517 static int switchdev_port_vlan_dump_cb(struct net_device *dev,
 518                                        struct switchdev_obj *obj)
 519 {
 520         struct switchdev_vlan_dump *dump =
 521                 container_of(obj, struct switchdev_vlan_dump, obj);
 522         struct switchdev_obj_vlan *vlan = &dump->obj.u.vlan;
 523         int err = 0;
 524
 525         if (vlan->vid_begin > vlan->vid_end)
 526                 return -EINVAL;
 527
 528         if (dump->filter_mask & RTEXT_FILTER_BRVLAN) {
 529                 dump->flags = vlan->flags;
 530                 for (dump->begin = dump->end = vlan->vid_begin;
 531                      dump->begin <= vlan->vid_end;
 532                      dump->begin++, dump->end++) {
 533                         err = switchdev_port_vlan_dump_put(dump);
 534                         if (err)
 535                                 return err;
 536                 }
 537         } else if (dump->filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED) {
 538                 if (dump->begin > vlan->vid_begin &&
 539                     dump->begin >= vlan->vid_end) {
 540                         if ((dump->begin - 1) == vlan->vid_end &&
 541                             dump->flags == vlan->flags) {
 542                                 /* prepend */
 543                                 dump->begin = vlan->vid_begin;
 544                         } else {
 545                                 err = switchdev_port_vlan_dump_put(dump);
 546                                 dump->flags = vlan->flags;
 547                                 dump->begin = vlan->vid_begin;
 548                                 dump->end = vlan->vid_end;
 549                         }
 550                 } else if (dump->end <= vlan->vid_begin &&
 551                            dump->end < vlan->vid_end) {
 552                         if ((dump->end  + 1) == vlan->vid_begin &&
 553                             dump->flags == vlan->flags) {
 554                                 /* append */
 555                                 dump->end = vlan->vid_end;
 556                         } else {
 557                                 err = switchdev_port_vlan_dump_put(dump);
 558                                 dump->flags = vlan->flags;
 559                                 dump->begin = vlan->vid_begin;
 560                                 dump->end = vlan->vid_end;
 561                         }
 562                 } else {
 563                         err = -EINVAL;
 564                 }
 565         }
 566
 567         return err;
 568 }
 569
 570 static int switchdev_port_vlan_fill(struct sk_buff *skb, struct net_device *dev,
 571                                     u32 filter_mask)
 572 {
 573         struct switchdev_vlan_dump dump = {
 574                 .obj = {
 575                         .id = SWITCHDEV_OBJ_PORT_VLAN,
 576                         .cb = switchdev_port_vlan_dump_cb,
 577                 },
 578                 .skb = skb,
 579                 .filter_mask = filter_mask,
 580         };
 581         int err = 0;
 582
 583         if ((filter_mask & RTEXT_FILTER_BRVLAN) ||
 584             (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)) {
 585                 err = switchdev_port_obj_dump(dev, &dump.obj);
 586                 if (err)
 587                         goto err_out;
 588                 if (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)
 589                         /* last one */
 590                         err = switchdev_port_vlan_dump_put(&dump);
 591         }
 592
 593 err_out:
 594         return err == -EOPNOTSUPP ? 0 : err;
 595 }
 596
 597 /**
 598  *      switchdev_port_bridge_getlink - Get bridge port attributes
 599  *
 600  *      @dev: port device
 601  *
 602  *      Called for SELF on rtnl_bridge_getlink to get bridge port
 603  *      attributes.
 604  */
 605 int switchdev_port_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
 606                                   struct net_device *dev, u32 filter_mask,
 607                                   int nlflags)
 608 {
 609         struct switchdev_attr attr = {
 610                 .id = SWITCHDEV_ATTR_PORT_BRIDGE_FLAGS,
 611         };
 612         u16 mode = BRIDGE_MODE_UNDEF;
 613         u32 mask = BR_LEARNING | BR_LEARNING_SYNC;
 614         int err;
 615
 616         err = switchdev_port_attr_get(dev, &attr);
 617         if (err && err != -EOPNOTSUPP)
 618                 return err;
 619
 620         return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode,
 621                                        attr.u.brport_flags, mask, nlflags,
 622                                        filter_mask, switchdev_port_vlan_fill);
 623 }
 624 EXPORT_SYMBOL_GPL(switchdev_port_bridge_getlink);
 625
 626 static int switchdev_port_br_setflag(struct net_device *dev,
 627                                      struct nlattr *nlattr,
 628                                      unsigned long brport_flag)
 629 {
 630         struct switchdev_attr attr = {
 631                 .id = SWITCHDEV_ATTR_PORT_BRIDGE_FLAGS,
 632         };
 633         u8 flag = nla_get_u8(nlattr);
 634         int err;
 635
 636         err = switchdev_port_attr_get(dev, &attr);
 637         if (err)
 638                 return err;
 639
 640         if (flag)
 641                 attr.u.brport_flags |= brport_flag;
 642         else
 643                 attr.u.brport_flags &= ~brport_flag;
 644
 645         return switchdev_port_attr_set(dev, &attr);
 646 }
 647
 648 static const struct nla_policy
 649 switchdev_port_bridge_policy[IFLA_BRPORT_MAX + 1] = {
 650         [IFLA_BRPORT_STATE]             = { .type = NLA_U8 },
 651         [IFLA_BRPORT_COST]              = { .type = NLA_U32 },
 652         [IFLA_BRPORT_PRIORITY]          = { .type = NLA_U16 },
 653         [IFLA_BRPORT_MODE]              = { .type = NLA_U8 },
 654         [IFLA_BRPORT_GUARD]             = { .type = NLA_U8 },
 655         [IFLA_BRPORT_PROTECT]           = { .type = NLA_U8 },
 656         [IFLA_BRPORT_FAST_LEAVE]        = { .type = NLA_U8 },
 657         [IFLA_BRPORT_LEARNING]          = { .type = NLA_U8 },
 658         [IFLA_BRPORT_LEARNING_SYNC]     = { .type = NLA_U8 },
 659         [IFLA_BRPORT_UNICAST_FLOOD]     = { .type = NLA_U8 },
 660 };
 661
 662 static int switchdev_port_br_setlink_protinfo(struct net_device *dev,
 663                                               struct nlattr *protinfo)
 664 {
 665         struct nlattr *attr;
 666         int rem;
 667         int err;
 668
 669         err = nla_validate_nested(protinfo, IFLA_BRPORT_MAX,
 670                                   switchdev_port_bridge_policy);
 671         if (err)
 672                 return err;
 673
 674         nla_for_each_nested(attr, protinfo, rem) {
 675                 switch (nla_type(attr)) {
 676                 case IFLA_BRPORT_LEARNING:
 677                         err = switchdev_port_br_setflag(dev, attr,
 678                                                         BR_LEARNING);
 679                         break;
 680                 case IFLA_BRPORT_LEARNING_SYNC:
 681                         err = switchdev_port_br_setflag(dev, attr,
 682                                                         BR_LEARNING_SYNC);
 683                         break;
 684                 default:
 685                         err = -EOPNOTSUPP;
 686                         break;
 687                 }
 688                 if (err)
 689                         return err;
 690         }
 691
 692         return 0;
 693 }
 694
 695 static int switchdev_port_br_afspec(struct net_device *dev,
 696                                     struct nlattr *afspec,
 697                                     int (*f)(struct net_device *dev,
 698                                              struct switchdev_obj *obj))
 699 {
 700         struct nlattr *attr;
 701         struct bridge_vlan_info *vinfo;
 702         struct switchdev_obj obj = {
 703                 .id = SWITCHDEV_OBJ_PORT_VLAN,
 704         };
 705         struct switchdev_obj_vlan *vlan = &obj.u.vlan;
 706         int rem;
 707         int err;
 708
 709         nla_for_each_nested(attr, afspec, rem) {
 710                 if (nla_type(attr) != IFLA_BRIDGE_VLAN_INFO)
 711                         continue;
 712                 if (nla_len(attr) != sizeof(struct bridge_vlan_info))
 713                         return -EINVAL;
 714                 vinfo = nla_data(attr);
 715                 vlan->flags = vinfo->flags;
 716                 if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_BEGIN) {
 717                         if (vlan->vid_begin)
 718                                 return -EINVAL;
 719                         vlan->vid_begin = vinfo->vid;
 720                 } else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
 721                         if (!vlan->vid_begin)
 722                                 return -EINVAL;
 723                         vlan->vid_end = vinfo->vid;
 724                         if (vlan->vid_end <= vlan->vid_begin)
 725                                 return -EINVAL;
 726                         err = f(dev, &obj);
 727                         if (err)
 728                                 return err;
 729                         memset(vlan, 0, sizeof(*vlan));
 730                 } else {
 731                         if (vlan->vid_begin)
 732                                 return -EINVAL;
 733                         vlan->vid_begin = vinfo->vid;
 734                         vlan->vid_end = vinfo->vid;
 735                         err = f(dev, &obj);
 736                         if (err)
 737                                 return err;
 738                         memset(vlan, 0, sizeof(*vlan));
 739                 }
 740         }
 741
 742         return 0;
 743 }
 744
 745 /**
 746  *      switchdev_port_bridge_setlink - Set bridge port attributes
 747  *
 748  *      @dev: port device
 749  *      @nlh: netlink header
 750  *      @flags: netlink flags
 751  *
 752  *      Called for SELF on rtnl_bridge_setlink to set bridge port
 753  *      attributes.
 754  */
 755 int switchdev_port_bridge_setlink(struct net_device *dev,
 756                                   struct nlmsghdr *nlh, u16 flags)
 757 {
 758         struct nlattr *protinfo;
 759         struct nlattr *afspec;
 760         int err = 0;
 761
 762         protinfo = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 763                                    IFLA_PROTINFO);
 764         if (protinfo) {
 765                 err = switchdev_port_br_setlink_protinfo(dev, protinfo);
 766                 if (err)
 767                         return err;
 768         }
 769
 770         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 771                                  IFLA_AF_SPEC);
 772         if (afspec)
 773                 err = switchdev_port_br_afspec(dev, afspec,
 774                                                switchdev_port_obj_add);
 775
 776         return err;
 777 }
 778 EXPORT_SYMBOL_GPL(switchdev_port_bridge_setlink);
 779
 780 /**
 781  *      switchdev_port_bridge_dellink - Set bridge port attributes
 782  *
 783  *      @dev: port device
 784  *      @nlh: netlink header
 785  *      @flags: netlink flags
 786  *
 787  *      Called for SELF on rtnl_bridge_dellink to set bridge port
 788  *      attributes.
 789  */
 790 int switchdev_port_bridge_dellink(struct net_device *dev,
 791                                   struct nlmsghdr *nlh, u16 flags)
 792 {
 793         struct nlattr *afspec;
 794
 795         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 796                                  IFLA_AF_SPEC);
 797         if (afspec)
 798                 return switchdev_port_br_afspec(dev, afspec,
 799                                                 switchdev_port_obj_del);
 800
 801         return 0;
 802 }
 803 EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
 804
 805 /**
 806  *      switchdev_port_fdb_add - Add FDB (MAC/VLAN) entry to port
 807  *
 808  *      @ndmsg: netlink hdr
 809  *      @nlattr: netlink attributes
 810  *      @dev: port device
 811  *      @addr: MAC address to add
 812  *      @vid: VLAN to add
 813  *
 814  *      Add FDB entry to switch device.
 815  */
 816 int switchdev_port_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 817                            struct net_device *dev, const unsigned char *addr,
 818                            u16 vid, u16 nlm_flags)
 819 {
 820         struct switchdev_obj obj = {
 821                 .id = SWITCHDEV_OBJ_PORT_FDB,
 822                 .u.fdb = {
 823                         .addr = addr,
 824                         .vid = vid,
 825                 },
 826         };
 827
 828         return switchdev_port_obj_add(dev, &obj);
 829 }
 830 EXPORT_SYMBOL_GPL(switchdev_port_fdb_add);
 831
 832 /**
 833  *      switchdev_port_fdb_del - Delete FDB (MAC/VLAN) entry from port
 834  *
 835  *      @ndmsg: netlink hdr
 836  *      @nlattr: netlink attributes
 837  *      @dev: port device
 838  *      @addr: MAC address to delete
 839  *      @vid: VLAN to delete
 840  *
 841  *      Delete FDB entry from switch device.
 842  */
 843 int switchdev_port_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 844                            struct net_device *dev, const unsigned char *addr,
 845                            u16 vid)
 846 {
 847         struct switchdev_obj obj = {
 848                 .id = SWITCHDEV_OBJ_PORT_FDB,
 849                 .u.fdb = {
 850                         .addr = addr,
 851                         .vid = vid,
 852                 },
 853         };
 854
 855         return switchdev_port_obj_del(dev, &obj);
 856 }
 857 EXPORT_SYMBOL_GPL(switchdev_port_fdb_del);
 858
 859 struct switchdev_fdb_dump {
 860         struct switchdev_obj obj;
 861         struct sk_buff *skb;
 862         struct netlink_callback *cb;
 863         int idx;
 864 };
 865
 866 static int switchdev_port_fdb_dump_cb(struct net_device *dev,
 867                                       struct switchdev_obj *obj)
 868 {
 869         struct switchdev_fdb_dump *dump =
 870                 container_of(obj, struct switchdev_fdb_dump, obj);
 871         u32 portid = NETLINK_CB(dump->cb->skb).portid;
 872         u32 seq = dump->cb->nlh->nlmsg_seq;
 873         struct nlmsghdr *nlh;
 874         struct ndmsg *ndm;
 875
 876         if (dump->idx < dump->cb->args[0])
 877                 goto skip;
 878
 879         nlh = nlmsg_put(dump->skb, portid, seq, RTM_NEWNEIGH,
 880                         sizeof(*ndm), NLM_F_MULTI);
 881         if (!nlh)
 882                 return -EMSGSIZE;
 883
 884         ndm = nlmsg_data(nlh);
 885         ndm->ndm_family  = AF_BRIDGE;
 886         ndm->ndm_pad1    = 0;
 887         ndm->ndm_pad2    = 0;
 888         ndm->ndm_flags   = NTF_SELF;
 889         ndm->ndm_type    = 0;
 890         ndm->ndm_ifindex = dev->ifindex;
 891         ndm->ndm_state   = obj->u.fdb.ndm_state;
 892
 893         if (nla_put(dump->skb, NDA_LLADDR, ETH_ALEN, obj->u.fdb.addr))
 894                 goto nla_put_failure;
 895
 896         if (obj->u.fdb.vid && nla_put_u16(dump->skb, NDA_VLAN, obj->u.fdb.vid))
 897                 goto nla_put_failure;
 898
 899         nlmsg_end(dump->skb, nlh);
 900
 901 skip:
 902         dump->idx++;
 903         return 0;
 904
 905 nla_put_failure:
 906         nlmsg_cancel(dump->skb, nlh);
 907         return -EMSGSIZE;
 908 }
 909
 910 /**
 911  *      switchdev_port_fdb_dump - Dump port FDB (MAC/VLAN) entries
 912  *
 913  *      @skb: netlink skb
 914  *      @cb: netlink callback
 915  *      @dev: port device
 916  *      @filter_dev: filter device
 917  *      @idx:
 918  *
 919  *      Delete FDB entry from switch device.
 920  */
 921 int switchdev_port_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 922                             struct net_device *dev,
 923                             struct net_device *filter_dev, int idx)
 924 {
 925         struct switchdev_fdb_dump dump = {
 926                 .obj = {
 927                         .id = SWITCHDEV_OBJ_PORT_FDB,
 928                         .cb = switchdev_port_fdb_dump_cb,
 929                 },
 930                 .skb = skb,
 931                 .cb = cb,
 932                 .idx = idx,
 933         };
 934
 935         switchdev_port_obj_dump(dev, &dump.obj);
 936         return dump.idx;
 937 }
 938 EXPORT_SYMBOL_GPL(switchdev_port_fdb_dump);
 939
 940 static struct net_device *switchdev_get_lowest_dev(struct net_device *dev)
 941 {
 942         const struct switchdev_ops *ops = dev->switchdev_ops;
 943         struct net_device *lower_dev;
 944         struct net_device *port_dev;
 945         struct list_head *iter;
 946
 947         /* Recusively search down until we find a sw port dev.
 948          * (A sw port dev supports switchdev_port_attr_get).
 949          */
 950
 951         if (ops && ops->switchdev_port_attr_get)
 952                 return dev;
 953
 954         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 955                 port_dev = switchdev_get_lowest_dev(lower_dev);
 956                 if (port_dev)
 957                         return port_dev;
 958         }
 959
 960         return NULL;
 961 }
 962
 963 static struct net_device *switchdev_get_dev_by_nhs(struct fib_info *fi)
 964 {
 965         struct switchdev_attr attr = {
 966                 .id = SWITCHDEV_ATTR_PORT_PARENT_ID,
 967         };
 968         struct switchdev_attr prev_attr;
 969         struct net_device *dev = NULL;
 970         int nhsel;
 971
 972         /* For this route, all nexthop devs must be on the same switch. */
 973
 974         for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
 975                 const struct fib_nh *nh = &fi->fib_nh[nhsel];
 976
 977                 if (!nh->nh_dev)
 978                         return NULL;
 979
 980                 dev = switchdev_get_lowest_dev(nh->nh_dev);
 981                 if (!dev)
 982                         return NULL;
 983
 984                 if (switchdev_port_attr_get(dev, &attr))
 985                         return NULL;
 986
 987                 if (nhsel > 0 &&
 988                     !netdev_phys_item_id_same(&prev_attr.u.ppid, &attr.u.ppid))
 989                                 return NULL;
 990
 991                 prev_attr = attr;
 992         }
 993
 994         return dev;
 995 }
 996
 997 /**
 998  *      switchdev_fib_ipv4_add - Add/modify switch IPv4 route entry
 999  *
1000  *      @dst: route's IPv4 destination address
1001  *      @dst_len: destination address length (prefix length)
1002  *      @fi: route FIB info structure
1003  *      @tos: route TOS
1004  *      @type: route type
1005  *      @nlflags: netlink flags passed in (NLM_F_*)
1006  *      @tb_id: route table ID
1007  *
1008  *      Add/modify switch IPv4 route entry.
1009  */
1010 int switchdev_fib_ipv4_add(u32 dst, int dst_len, struct fib_info *fi,
1011                            u8 tos, u8 type, u32 nlflags, u32 tb_id)
1012 {
1013         struct switchdev_obj fib_obj = {
1014                 .id = SWITCHDEV_OBJ_IPV4_FIB,
1015                 .u.ipv4_fib = {
1016                         .dst = dst,
1017                         .dst_len = dst_len,
1018                         .fi = fi,
1019                         .tos = tos,
1020                         .type = type,
1021                         .nlflags = nlflags,
1022                         .tb_id = tb_id,
1023                 },
1024         };
1025         struct net_device *dev;
1026         int err = 0;
1027
1028         /* Don't offload route if using custom ip rules or if
1029          * IPv4 FIB offloading has been disabled completely.
1030          */
1031
1032 #ifdef CONFIG_IP_MULTIPLE_TABLES
1033         if (fi->fib_net->ipv4.fib_has_custom_rules)
1034                 return 0;
1035 #endif
1036
1037         if (fi->fib_net->ipv4.fib_offload_disabled)
1038                 return 0;
1039
1040         dev = switchdev_get_dev_by_nhs(fi);
1041         if (!dev)
1042                 return 0;
1043
1044         err = switchdev_port_obj_add(dev, &fib_obj);
1045         if (!err)
1046                 fi->fib_flags |= RTNH_F_OFFLOAD;
1047
1048         return err == -EOPNOTSUPP ? 0 : err;
1049 }
1050 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_add);
1051
1052 /**
1053  *      switchdev_fib_ipv4_del - Delete IPv4 route entry from switch
1054  *
1055  *      @dst: route's IPv4 destination address
1056  *      @dst_len: destination address length (prefix length)
1057  *      @fi: route FIB info structure
1058  *      @tos: route TOS
1059  *      @type: route type
1060  *      @tb_id: route table ID
1061  *
1062  *      Delete IPv4 route entry from switch device.
1063  */
1064 int switchdev_fib_ipv4_del(u32 dst, int dst_len, struct fib_info *fi,
1065                            u8 tos, u8 type, u32 tb_id)
1066 {
1067         struct switchdev_obj fib_obj = {
1068                 .id = SWITCHDEV_OBJ_IPV4_FIB,
1069                 .u.ipv4_fib = {
1070                         .dst = dst,
1071                         .dst_len = dst_len,
1072                         .fi = fi,
1073                         .tos = tos,
1074                         .type = type,
1075                         .nlflags = 0,
1076                         .tb_id = tb_id,
1077                 },
1078         };
1079         struct net_device *dev;
1080         int err = 0;
1081
1082         if (!(fi->fib_flags & RTNH_F_OFFLOAD))
1083                 return 0;
1084
1085         dev = switchdev_get_dev_by_nhs(fi);
1086         if (!dev)
1087                 return 0;
1088
1089         err = switchdev_port_obj_del(dev, &fib_obj);
1090         if (!err)
1091                 fi->fib_flags &= ~RTNH_F_OFFLOAD;
1092
1093         return err == -EOPNOTSUPP ? 0 : err;
1094 }
1095 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_del);
1096
1097 /**
1098  *      switchdev_fib_ipv4_abort - Abort an IPv4 FIB operation
1099  *
1100  *      @fi: route FIB info structure
1101  */
1102 void switchdev_fib_ipv4_abort(struct fib_info *fi)
1103 {
1104         /* There was a problem installing this route to the offload
1105          * device.  For now, until we come up with more refined
1106          * policy handling, abruptly end IPv4 fib offloading for
1107          * for entire net by flushing offload device(s) of all
1108          * IPv4 routes, and mark IPv4 fib offloading broken from
1109          * this point forward.
1110          */
1111
1112         fib_flush_external(fi->fib_net);
1113         fi->fib_net->ipv4.fib_offload_disabled = true;
1114 }
1115 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_abort);
1116
1117 static bool switchdev_port_same_parent_id(struct net_device *a,
1118                                           struct net_device *b)
1119 {
1120         struct switchdev_attr a_attr = {
1121                 .id = SWITCHDEV_ATTR_PORT_PARENT_ID,
1122                 .flags = SWITCHDEV_F_NO_RECURSE,
1123         };
1124         struct switchdev_attr b_attr = {
1125                 .id = SWITCHDEV_ATTR_PORT_PARENT_ID,
1126                 .flags = SWITCHDEV_F_NO_RECURSE,
1127         };
1128
1129         if (switchdev_port_attr_get(a, &a_attr) ||
1130             switchdev_port_attr_get(b, &b_attr))
1131                 return false;
1132
1133         return netdev_phys_item_id_same(&a_attr.u.ppid, &b_attr.u.ppid);
1134 }
1135
1136 static u32 switchdev_port_fwd_mark_get(struct net_device *dev,
1137                                        struct net_device *group_dev)
1138 {
1139         struct net_device *lower_dev;
1140         struct list_head *iter;
1141
1142         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1143                 if (lower_dev == dev)
1144                         continue;
1145                 if (switchdev_port_same_parent_id(dev, lower_dev))
1146                         return lower_dev->offload_fwd_mark;
1147                 return switchdev_port_fwd_mark_get(dev, lower_dev);
1148         }
1149
1150         return dev->ifindex;
1151 }
1152
1153 static void switchdev_port_fwd_mark_reset(struct net_device *group_dev,
1154                                           u32 old_mark, u32 *reset_mark)
1155 {
1156         struct net_device *lower_dev;
1157         struct list_head *iter;
1158
1159         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1160                 if (lower_dev->offload_fwd_mark == old_mark) {
1161                         if (!*reset_mark)
1162                                 *reset_mark = lower_dev->ifindex;
1163                         lower_dev->offload_fwd_mark = *reset_mark;
1164                 }
1165                 switchdev_port_fwd_mark_reset(lower_dev, old_mark, reset_mark);
1166         }
1167 }
1168
1169 /**
1170  *      switchdev_port_fwd_mark_set - Set port offload forwarding mark
1171  *
1172  *      @dev: port device
1173  *      @group_dev: containing device
1174  *      @joining: true if dev is joining group; false if leaving group
1175  *
1176  *      An ungrouped port's offload mark is just its ifindex.  A grouped
1177  *      port's (member of a bridge, for example) offload mark is the ifindex
1178  *      of one of the ports in the group with the same parent (switch) ID.
1179  *      Ports on the same device in the same group will have the same mark.
1180  *
1181  *      Example:
1182  *
1183  *              br0             ifindex=9
1184  *                sw1p1         ifindex=2       mark=2
1185  *                sw1p2         ifindex=3       mark=2
1186  *                sw2p1         ifindex=4       mark=5
1187  *                sw2p2         ifindex=5       mark=5
1188  *
1189  *      If sw2p2 leaves the bridge, we'll have:
1190  *
1191  *              br0             ifindex=9
1192  *                sw1p1         ifindex=2       mark=2
1193  *                sw1p2         ifindex=3       mark=2
1194  *                sw2p1         ifindex=4       mark=4
1195  *              sw2p2           ifindex=5       mark=5
1196  */
1197 void switchdev_port_fwd_mark_set(struct net_device *dev,
1198                                  struct net_device *group_dev,
1199                                  bool joining)
1200 {
1201         u32 mark = dev->ifindex;
1202         u32 reset_mark = 0;
1203
1204         if (group_dev && joining) {
1205                 mark = switchdev_port_fwd_mark_get(dev, group_dev);
1206         } else if (group_dev && !joining) {
1207                 if (dev->offload_fwd_mark == mark)
1208                         /* Ohoh, this port was the mark reference port,
1209                          * but it's leaving the group, so reset the
1210                          * mark for the remaining ports in the group.
1211                          */
1212                         switchdev_port_fwd_mark_reset(group_dev, mark,
1213                                                       &reset_mark);
1214         }
1215
1216         dev->offload_fwd_mark = mark;
1217 }
1218 EXPORT_SYMBOL_GPL(switchdev_port_fwd_mark_set);