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_ID_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_ID_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                 goto done;
 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 == -EOPNOTSUPP &&
 160                     attr->flags & SWITCHDEV_F_SKIP_EOPNOTSUPP)
 161                         continue;
 162                 if (err)
 163                         break;
 164         }
 165
 166 done:
 167         if (err == -EOPNOTSUPP && attr->flags & SWITCHDEV_F_SKIP_EOPNOTSUPP)
 168                 err = 0;
 169
 170         return err;
 171 }
 172
 173 struct switchdev_attr_set_work {
 174         struct work_struct work;
 175         struct net_device *dev;
 176         struct switchdev_attr attr;
 177 };
 178
 179 static void switchdev_port_attr_set_work(struct work_struct *work)
 180 {
 181         struct switchdev_attr_set_work *asw =
 182                 container_of(work, struct switchdev_attr_set_work, work);
 183         int err;
 184
 185         rtnl_lock();
 186         err = switchdev_port_attr_set(asw->dev, &asw->attr);
 187         if (err && err != -EOPNOTSUPP)
 188                 netdev_err(asw->dev, "failed (err=%d) to set attribute (id=%d)\n",
 189                            err, asw->attr.id);
 190         rtnl_unlock();
 191
 192         dev_put(asw->dev);
 193         kfree(work);
 194 }
 195
 196 static int switchdev_port_attr_set_defer(struct net_device *dev,
 197                                          struct switchdev_attr *attr)
 198 {
 199         struct switchdev_attr_set_work *asw;
 200
 201         asw = kmalloc(sizeof(*asw), GFP_ATOMIC);
 202         if (!asw)
 203                 return -ENOMEM;
 204
 205         INIT_WORK(&asw->work, switchdev_port_attr_set_work);
 206
 207         dev_hold(dev);
 208         asw->dev = dev;
 209         memcpy(&asw->attr, attr, sizeof(asw->attr));
 210
 211         schedule_work(&asw->work);
 212
 213         return 0;
 214 }
 215
 216 /**
 217  *      switchdev_port_attr_set - Set port attribute
 218  *
 219  *      @dev: port device
 220  *      @attr: attribute to set
 221  *
 222  *      Use a 2-phase prepare-commit transaction model to ensure
 223  *      system is not left in a partially updated state due to
 224  *      failure from driver/device.
 225  */
 226 int switchdev_port_attr_set(struct net_device *dev, struct switchdev_attr *attr)
 227 {
 228         struct switchdev_trans trans;
 229         int err;
 230
 231         if (!rtnl_is_locked()) {
 232                 /* Running prepare-commit transaction across stacked
 233                  * devices requires nothing moves, so if rtnl_lock is
 234                  * not held, schedule a worker thread to hold rtnl_lock
 235                  * while setting attr.
 236                  */
 237
 238                 return switchdev_port_attr_set_defer(dev, attr);
 239         }
 240
 241         switchdev_trans_init(&trans);
 242
 243         /* Phase I: prepare for attr set. Driver/device should fail
 244          * here if there are going to be issues in the commit phase,
 245          * such as lack of resources or support.  The driver/device
 246          * should reserve resources needed for the commit phase here,
 247          * but should not commit the attr.
 248          */
 249
 250         trans.ph_prepare = true;
 251         err = __switchdev_port_attr_set(dev, attr, &trans);
 252         if (err) {
 253                 /* Prepare phase failed: abort the transaction.  Any
 254                  * resources reserved in the prepare phase are
 255                  * released.
 256                  */
 257
 258                 if (err != -EOPNOTSUPP)
 259                         switchdev_trans_items_destroy(&trans);
 260
 261                 return err;
 262         }
 263
 264         /* Phase II: commit attr set.  This cannot fail as a fault
 265          * of driver/device.  If it does, it's a bug in the driver/device
 266          * because the driver said everythings was OK in phase I.
 267          */
 268
 269         trans.ph_prepare = false;
 270         err = __switchdev_port_attr_set(dev, attr, &trans);
 271         WARN(err, "%s: Commit of attribute (id=%d) failed.\n",
 272              dev->name, attr->id);
 273         switchdev_trans_items_warn_destroy(dev, &trans);
 274
 275         return err;
 276 }
 277 EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
 278
 279 static int __switchdev_port_obj_add(struct net_device *dev,
 280                                     const struct switchdev_obj *obj,
 281                                     struct switchdev_trans *trans)
 282 {
 283         const struct switchdev_ops *ops = dev->switchdev_ops;
 284         struct net_device *lower_dev;
 285         struct list_head *iter;
 286         int err = -EOPNOTSUPP;
 287
 288         if (ops && ops->switchdev_port_obj_add)
 289                 return ops->switchdev_port_obj_add(dev, obj, trans);
 290
 291         /* Switch device port(s) may be stacked under
 292          * bond/team/vlan dev, so recurse down to add object on
 293          * each port.
 294          */
 295
 296         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 297                 err = __switchdev_port_obj_add(lower_dev, obj, trans);
 298                 if (err)
 299                         break;
 300         }
 301
 302         return err;
 303 }
 304
 305 /**
 306  *      switchdev_port_obj_add - Add port object
 307  *
 308  *      @dev: port device
 309  *      @id: object ID
 310  *      @obj: object to add
 311  *
 312  *      Use a 2-phase prepare-commit transaction model to ensure
 313  *      system is not left in a partially updated state due to
 314  *      failure from driver/device.
 315  *
 316  *      rtnl_lock must be held.
 317  */
 318 int switchdev_port_obj_add(struct net_device *dev,
 319                            const struct switchdev_obj *obj)
 320 {
 321         struct switchdev_trans trans;
 322         int err;
 323
 324         ASSERT_RTNL();
 325
 326         switchdev_trans_init(&trans);
 327
 328         /* Phase I: prepare for obj add. Driver/device should fail
 329          * here if there are going to be issues in the commit phase,
 330          * such as lack of resources or support.  The driver/device
 331          * should reserve resources needed for the commit phase here,
 332          * but should not commit the obj.
 333          */
 334
 335         trans.ph_prepare = true;
 336         err = __switchdev_port_obj_add(dev, obj, &trans);
 337         if (err) {
 338                 /* Prepare phase failed: abort the transaction.  Any
 339                  * resources reserved in the prepare phase are
 340                  * released.
 341                  */
 342
 343                 if (err != -EOPNOTSUPP)
 344                         switchdev_trans_items_destroy(&trans);
 345
 346                 return err;
 347         }
 348
 349         /* Phase II: commit obj add.  This cannot fail as a fault
 350          * of driver/device.  If it does, it's a bug in the driver/device
 351          * because the driver said everythings was OK in phase I.
 352          */
 353
 354         trans.ph_prepare = false;
 355         err = __switchdev_port_obj_add(dev, obj, &trans);
 356         WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->id);
 357         switchdev_trans_items_warn_destroy(dev, &trans);
 358
 359         return err;
 360 }
 361 EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
 362
 363 /**
 364  *      switchdev_port_obj_del - Delete port object
 365  *
 366  *      @dev: port device
 367  *      @id: object ID
 368  *      @obj: object to delete
 369  */
 370 int switchdev_port_obj_del(struct net_device *dev,
 371                            const struct switchdev_obj *obj)
 372 {
 373         const struct switchdev_ops *ops = dev->switchdev_ops;
 374         struct net_device *lower_dev;
 375         struct list_head *iter;
 376         int err = -EOPNOTSUPP;
 377
 378         if (ops && ops->switchdev_port_obj_del)
 379                 return ops->switchdev_port_obj_del(dev, obj);
 380
 381         /* Switch device port(s) may be stacked under
 382          * bond/team/vlan dev, so recurse down to delete object on
 383          * each port.
 384          */
 385
 386         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 387                 err = switchdev_port_obj_del(lower_dev, obj);
 388                 if (err)
 389                         break;
 390         }
 391
 392         return err;
 393 }
 394 EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
 395
 396 /**
 397  *      switchdev_port_obj_dump - Dump port objects
 398  *
 399  *      @dev: port device
 400  *      @id: object ID
 401  *      @obj: object to dump
 402  *      @cb: function to call with a filled object
 403  */
 404 int switchdev_port_obj_dump(struct net_device *dev, struct switchdev_obj *obj,
 405                             switchdev_obj_dump_cb_t *cb)
 406 {
 407         const struct switchdev_ops *ops = dev->switchdev_ops;
 408         struct net_device *lower_dev;
 409         struct list_head *iter;
 410         int err = -EOPNOTSUPP;
 411
 412         if (ops && ops->switchdev_port_obj_dump)
 413                 return ops->switchdev_port_obj_dump(dev, obj, cb);
 414
 415         /* Switch device port(s) may be stacked under
 416          * bond/team/vlan dev, so recurse down to dump objects on
 417          * first port at bottom of stack.
 418          */
 419
 420         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 421                 err = switchdev_port_obj_dump(lower_dev, obj, cb);
 422                 break;
 423         }
 424
 425         return err;
 426 }
 427 EXPORT_SYMBOL_GPL(switchdev_port_obj_dump);
 428
 429 static DEFINE_MUTEX(switchdev_mutex);
 430 static RAW_NOTIFIER_HEAD(switchdev_notif_chain);
 431
 432 /**
 433  *      register_switchdev_notifier - Register notifier
 434  *      @nb: notifier_block
 435  *
 436  *      Register switch device notifier. This should be used by code
 437  *      which needs to monitor events happening in particular device.
 438  *      Return values are same as for atomic_notifier_chain_register().
 439  */
 440 int register_switchdev_notifier(struct notifier_block *nb)
 441 {
 442         int err;
 443
 444         mutex_lock(&switchdev_mutex);
 445         err = raw_notifier_chain_register(&switchdev_notif_chain, nb);
 446         mutex_unlock(&switchdev_mutex);
 447         return err;
 448 }
 449 EXPORT_SYMBOL_GPL(register_switchdev_notifier);
 450
 451 /**
 452  *      unregister_switchdev_notifier - Unregister notifier
 453  *      @nb: notifier_block
 454  *
 455  *      Unregister switch device notifier.
 456  *      Return values are same as for atomic_notifier_chain_unregister().
 457  */
 458 int unregister_switchdev_notifier(struct notifier_block *nb)
 459 {
 460         int err;
 461
 462         mutex_lock(&switchdev_mutex);
 463         err = raw_notifier_chain_unregister(&switchdev_notif_chain, nb);
 464         mutex_unlock(&switchdev_mutex);
 465         return err;
 466 }
 467 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
 468
 469 /**
 470  *      call_switchdev_notifiers - Call notifiers
 471  *      @val: value passed unmodified to notifier function
 472  *      @dev: port device
 473  *      @info: notifier information data
 474  *
 475  *      Call all network notifier blocks. This should be called by driver
 476  *      when it needs to propagate hardware event.
 477  *      Return values are same as for atomic_notifier_call_chain().
 478  */
 479 int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
 480                              struct switchdev_notifier_info *info)
 481 {
 482         int err;
 483
 484         info->dev = dev;
 485         mutex_lock(&switchdev_mutex);
 486         err = raw_notifier_call_chain(&switchdev_notif_chain, val, info);
 487         mutex_unlock(&switchdev_mutex);
 488         return err;
 489 }
 490 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
 491
 492 struct switchdev_vlan_dump {
 493         struct switchdev_obj_port_vlan vlan;
 494         struct sk_buff *skb;
 495         u32 filter_mask;
 496         u16 flags;
 497         u16 begin;
 498         u16 end;
 499 };
 500
 501 static int switchdev_port_vlan_dump_put(struct switchdev_vlan_dump *dump)
 502 {
 503         struct bridge_vlan_info vinfo;
 504
 505         vinfo.flags = dump->flags;
 506
 507         if (dump->begin == 0 && dump->end == 0) {
 508                 return 0;
 509         } else if (dump->begin == dump->end) {
 510                 vinfo.vid = dump->begin;
 511                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 512                             sizeof(vinfo), &vinfo))
 513                         return -EMSGSIZE;
 514         } else {
 515                 vinfo.vid = dump->begin;
 516                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_BEGIN;
 517                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 518                             sizeof(vinfo), &vinfo))
 519                         return -EMSGSIZE;
 520                 vinfo.vid = dump->end;
 521                 vinfo.flags &= ~BRIDGE_VLAN_INFO_RANGE_BEGIN;
 522                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_END;
 523                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 524                             sizeof(vinfo), &vinfo))
 525                         return -EMSGSIZE;
 526         }
 527
 528         return 0;
 529 }
 530
 531 static int switchdev_port_vlan_dump_cb(struct switchdev_obj *obj)
 532 {
 533         struct switchdev_obj_port_vlan *vlan = SWITCHDEV_OBJ_PORT_VLAN(obj);
 534         struct switchdev_vlan_dump *dump =
 535                 container_of(vlan, struct switchdev_vlan_dump, vlan);
 536         int err = 0;
 537
 538         if (vlan->vid_begin > vlan->vid_end)
 539                 return -EINVAL;
 540
 541         if (dump->filter_mask & RTEXT_FILTER_BRVLAN) {
 542                 dump->flags = vlan->flags;
 543                 for (dump->begin = dump->end = vlan->vid_begin;
 544                      dump->begin <= vlan->vid_end;
 545                      dump->begin++, dump->end++) {
 546                         err = switchdev_port_vlan_dump_put(dump);
 547                         if (err)
 548                                 return err;
 549                 }
 550         } else if (dump->filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED) {
 551                 if (dump->begin > vlan->vid_begin &&
 552                     dump->begin >= vlan->vid_end) {
 553                         if ((dump->begin - 1) == vlan->vid_end &&
 554                             dump->flags == vlan->flags) {
 555                                 /* prepend */
 556                                 dump->begin = vlan->vid_begin;
 557                         } else {
 558                                 err = switchdev_port_vlan_dump_put(dump);
 559                                 dump->flags = vlan->flags;
 560                                 dump->begin = vlan->vid_begin;
 561                                 dump->end = vlan->vid_end;
 562                         }
 563                 } else if (dump->end <= vlan->vid_begin &&
 564                            dump->end < vlan->vid_end) {
 565                         if ((dump->end  + 1) == vlan->vid_begin &&
 566                             dump->flags == vlan->flags) {
 567                                 /* append */
 568                                 dump->end = vlan->vid_end;
 569                         } else {
 570                                 err = switchdev_port_vlan_dump_put(dump);
 571                                 dump->flags = vlan->flags;
 572                                 dump->begin = vlan->vid_begin;
 573                                 dump->end = vlan->vid_end;
 574                         }
 575                 } else {
 576                         err = -EINVAL;
 577                 }
 578         }
 579
 580         return err;
 581 }
 582
 583 static int switchdev_port_vlan_fill(struct sk_buff *skb, struct net_device *dev,
 584                                     u32 filter_mask)
 585 {
 586         struct switchdev_vlan_dump dump = {
 587                 .vlan.obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
 588                 .skb = skb,
 589                 .filter_mask = filter_mask,
 590         };
 591         int err = 0;
 592
 593         if ((filter_mask & RTEXT_FILTER_BRVLAN) ||
 594             (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)) {
 595                 err = switchdev_port_obj_dump(dev, &dump.vlan.obj,
 596                                               switchdev_port_vlan_dump_cb);
 597                 if (err)
 598                         goto err_out;
 599                 if (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)
 600                         /* last one */
 601                         err = switchdev_port_vlan_dump_put(&dump);
 602         }
 603
 604 err_out:
 605         return err == -EOPNOTSUPP ? 0 : err;
 606 }
 607
 608 /**
 609  *      switchdev_port_bridge_getlink - Get bridge port attributes
 610  *
 611  *      @dev: port device
 612  *
 613  *      Called for SELF on rtnl_bridge_getlink to get bridge port
 614  *      attributes.
 615  */
 616 int switchdev_port_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
 617                                   struct net_device *dev, u32 filter_mask,
 618                                   int nlflags)
 619 {
 620         struct switchdev_attr attr = {
 621                 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
 622         };
 623         u16 mode = BRIDGE_MODE_UNDEF;
 624         u32 mask = BR_LEARNING | BR_LEARNING_SYNC;
 625         int err;
 626
 627         err = switchdev_port_attr_get(dev, &attr);
 628         if (err && err != -EOPNOTSUPP)
 629                 return err;
 630
 631         return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode,
 632                                        attr.u.brport_flags, mask, nlflags,
 633                                        filter_mask, switchdev_port_vlan_fill);
 634 }
 635 EXPORT_SYMBOL_GPL(switchdev_port_bridge_getlink);
 636
 637 static int switchdev_port_br_setflag(struct net_device *dev,
 638                                      struct nlattr *nlattr,
 639                                      unsigned long brport_flag)
 640 {
 641         struct switchdev_attr attr = {
 642                 .id = SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS,
 643         };
 644         u8 flag = nla_get_u8(nlattr);
 645         int err;
 646
 647         err = switchdev_port_attr_get(dev, &attr);
 648         if (err)
 649                 return err;
 650
 651         if (flag)
 652                 attr.u.brport_flags |= brport_flag;
 653         else
 654                 attr.u.brport_flags &= ~brport_flag;
 655
 656         return switchdev_port_attr_set(dev, &attr);
 657 }
 658
 659 static const struct nla_policy
 660 switchdev_port_bridge_policy[IFLA_BRPORT_MAX + 1] = {
 661         [IFLA_BRPORT_STATE]             = { .type = NLA_U8 },
 662         [IFLA_BRPORT_COST]              = { .type = NLA_U32 },
 663         [IFLA_BRPORT_PRIORITY]          = { .type = NLA_U16 },
 664         [IFLA_BRPORT_MODE]              = { .type = NLA_U8 },
 665         [IFLA_BRPORT_GUARD]             = { .type = NLA_U8 },
 666         [IFLA_BRPORT_PROTECT]           = { .type = NLA_U8 },
 667         [IFLA_BRPORT_FAST_LEAVE]        = { .type = NLA_U8 },
 668         [IFLA_BRPORT_LEARNING]          = { .type = NLA_U8 },
 669         [IFLA_BRPORT_LEARNING_SYNC]     = { .type = NLA_U8 },
 670         [IFLA_BRPORT_UNICAST_FLOOD]     = { .type = NLA_U8 },
 671 };
 672
 673 static int switchdev_port_br_setlink_protinfo(struct net_device *dev,
 674                                               struct nlattr *protinfo)
 675 {
 676         struct nlattr *attr;
 677         int rem;
 678         int err;
 679
 680         err = nla_validate_nested(protinfo, IFLA_BRPORT_MAX,
 681                                   switchdev_port_bridge_policy);
 682         if (err)
 683                 return err;
 684
 685         nla_for_each_nested(attr, protinfo, rem) {
 686                 switch (nla_type(attr)) {
 687                 case IFLA_BRPORT_LEARNING:
 688                         err = switchdev_port_br_setflag(dev, attr,
 689                                                         BR_LEARNING);
 690                         break;
 691                 case IFLA_BRPORT_LEARNING_SYNC:
 692                         err = switchdev_port_br_setflag(dev, attr,
 693                                                         BR_LEARNING_SYNC);
 694                         break;
 695                 default:
 696                         err = -EOPNOTSUPP;
 697                         break;
 698                 }
 699                 if (err)
 700                         return err;
 701         }
 702
 703         return 0;
 704 }
 705
 706 static int switchdev_port_br_afspec(struct net_device *dev,
 707                                     struct nlattr *afspec,
 708                                     int (*f)(struct net_device *dev,
 709                                              const struct switchdev_obj *obj))
 710 {
 711         struct nlattr *attr;
 712         struct bridge_vlan_info *vinfo;
 713         struct switchdev_obj_port_vlan vlan = {
 714                 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
 715         };
 716         int rem;
 717         int err;
 718
 719         nla_for_each_nested(attr, afspec, rem) {
 720                 if (nla_type(attr) != IFLA_BRIDGE_VLAN_INFO)
 721                         continue;
 722                 if (nla_len(attr) != sizeof(struct bridge_vlan_info))
 723                         return -EINVAL;
 724                 vinfo = nla_data(attr);
 725                 vlan.flags = vinfo->flags;
 726                 if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_BEGIN) {
 727                         if (vlan.vid_begin)
 728                                 return -EINVAL;
 729                         vlan.vid_begin = vinfo->vid;
 730                 } else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
 731                         if (!vlan.vid_begin)
 732                                 return -EINVAL;
 733                         vlan.vid_end = vinfo->vid;
 734                         if (vlan.vid_end <= vlan.vid_begin)
 735                                 return -EINVAL;
 736                         err = f(dev, &vlan.obj);
 737                         if (err)
 738                                 return err;
 739                         memset(&vlan, 0, sizeof(vlan));
 740                 } else {
 741                         if (vlan.vid_begin)
 742                                 return -EINVAL;
 743                         vlan.vid_begin = vinfo->vid;
 744                         vlan.vid_end = vinfo->vid;
 745                         err = f(dev, &vlan.obj);
 746                         if (err)
 747                                 return err;
 748                         memset(&vlan, 0, sizeof(vlan));
 749                 }
 750         }
 751
 752         return 0;
 753 }
 754
 755 /**
 756  *      switchdev_port_bridge_setlink - Set bridge port attributes
 757  *
 758  *      @dev: port device
 759  *      @nlh: netlink header
 760  *      @flags: netlink flags
 761  *
 762  *      Called for SELF on rtnl_bridge_setlink to set bridge port
 763  *      attributes.
 764  */
 765 int switchdev_port_bridge_setlink(struct net_device *dev,
 766                                   struct nlmsghdr *nlh, u16 flags)
 767 {
 768         struct nlattr *protinfo;
 769         struct nlattr *afspec;
 770         int err = 0;
 771
 772         protinfo = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 773                                    IFLA_PROTINFO);
 774         if (protinfo) {
 775                 err = switchdev_port_br_setlink_protinfo(dev, protinfo);
 776                 if (err)
 777                         return err;
 778         }
 779
 780         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 781                                  IFLA_AF_SPEC);
 782         if (afspec)
 783                 err = switchdev_port_br_afspec(dev, afspec,
 784                                                switchdev_port_obj_add);
 785
 786         return err;
 787 }
 788 EXPORT_SYMBOL_GPL(switchdev_port_bridge_setlink);
 789
 790 /**
 791  *      switchdev_port_bridge_dellink - Set bridge port attributes
 792  *
 793  *      @dev: port device
 794  *      @nlh: netlink header
 795  *      @flags: netlink flags
 796  *
 797  *      Called for SELF on rtnl_bridge_dellink to set bridge port
 798  *      attributes.
 799  */
 800 int switchdev_port_bridge_dellink(struct net_device *dev,
 801                                   struct nlmsghdr *nlh, u16 flags)
 802 {
 803         struct nlattr *afspec;
 804
 805         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 806                                  IFLA_AF_SPEC);
 807         if (afspec)
 808                 return switchdev_port_br_afspec(dev, afspec,
 809                                                 switchdev_port_obj_del);
 810
 811         return 0;
 812 }
 813 EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
 814
 815 /**
 816  *      switchdev_port_fdb_add - Add FDB (MAC/VLAN) entry to port
 817  *
 818  *      @ndmsg: netlink hdr
 819  *      @nlattr: netlink attributes
 820  *      @dev: port device
 821  *      @addr: MAC address to add
 822  *      @vid: VLAN to add
 823  *
 824  *      Add FDB entry to switch device.
 825  */
 826 int switchdev_port_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 827                            struct net_device *dev, const unsigned char *addr,
 828                            u16 vid, u16 nlm_flags)
 829 {
 830         struct switchdev_obj_port_fdb fdb = {
 831                 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
 832                 .addr = addr,
 833                 .vid = vid,
 834         };
 835
 836         return switchdev_port_obj_add(dev, &fdb.obj);
 837 }
 838 EXPORT_SYMBOL_GPL(switchdev_port_fdb_add);
 839
 840 /**
 841  *      switchdev_port_fdb_del - Delete FDB (MAC/VLAN) entry from port
 842  *
 843  *      @ndmsg: netlink hdr
 844  *      @nlattr: netlink attributes
 845  *      @dev: port device
 846  *      @addr: MAC address to delete
 847  *      @vid: VLAN to delete
 848  *
 849  *      Delete FDB entry from switch device.
 850  */
 851 int switchdev_port_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 852                            struct net_device *dev, const unsigned char *addr,
 853                            u16 vid)
 854 {
 855         struct switchdev_obj_port_fdb fdb = {
 856                 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
 857                 .addr = addr,
 858                 .vid = vid,
 859         };
 860
 861         return switchdev_port_obj_del(dev, &fdb.obj);
 862 }
 863 EXPORT_SYMBOL_GPL(switchdev_port_fdb_del);
 864
 865 struct switchdev_fdb_dump {
 866         struct switchdev_obj_port_fdb fdb;
 867         struct net_device *dev;
 868         struct sk_buff *skb;
 869         struct netlink_callback *cb;
 870         int idx;
 871 };
 872
 873 static int switchdev_port_fdb_dump_cb(struct switchdev_obj *obj)
 874 {
 875         struct switchdev_obj_port_fdb *fdb = SWITCHDEV_OBJ_PORT_FDB(obj);
 876         struct switchdev_fdb_dump *dump =
 877                 container_of(fdb, struct switchdev_fdb_dump, fdb);
 878         u32 portid = NETLINK_CB(dump->cb->skb).portid;
 879         u32 seq = dump->cb->nlh->nlmsg_seq;
 880         struct nlmsghdr *nlh;
 881         struct ndmsg *ndm;
 882
 883         if (dump->idx < dump->cb->args[0])
 884                 goto skip;
 885
 886         nlh = nlmsg_put(dump->skb, portid, seq, RTM_NEWNEIGH,
 887                         sizeof(*ndm), NLM_F_MULTI);
 888         if (!nlh)
 889                 return -EMSGSIZE;
 890
 891         ndm = nlmsg_data(nlh);
 892         ndm->ndm_family  = AF_BRIDGE;
 893         ndm->ndm_pad1    = 0;
 894         ndm->ndm_pad2    = 0;
 895         ndm->ndm_flags   = NTF_SELF;
 896         ndm->ndm_type    = 0;
 897         ndm->ndm_ifindex = dump->dev->ifindex;
 898         ndm->ndm_state   = fdb->ndm_state;
 899
 900         if (nla_put(dump->skb, NDA_LLADDR, ETH_ALEN, fdb->addr))
 901                 goto nla_put_failure;
 902
 903         if (fdb->vid && nla_put_u16(dump->skb, NDA_VLAN, fdb->vid))
 904                 goto nla_put_failure;
 905
 906         nlmsg_end(dump->skb, nlh);
 907
 908 skip:
 909         dump->idx++;
 910         return 0;
 911
 912 nla_put_failure:
 913         nlmsg_cancel(dump->skb, nlh);
 914         return -EMSGSIZE;
 915 }
 916
 917 /**
 918  *      switchdev_port_fdb_dump - Dump port FDB (MAC/VLAN) entries
 919  *
 920  *      @skb: netlink skb
 921  *      @cb: netlink callback
 922  *      @dev: port device
 923  *      @filter_dev: filter device
 924  *      @idx:
 925  *
 926  *      Delete FDB entry from switch device.
 927  */
 928 int switchdev_port_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 929                             struct net_device *dev,
 930                             struct net_device *filter_dev, int idx)
 931 {
 932         struct switchdev_fdb_dump dump = {
 933                 .fdb.obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
 934                 .dev = dev,
 935                 .skb = skb,
 936                 .cb = cb,
 937                 .idx = idx,
 938         };
 939
 940         switchdev_port_obj_dump(dev, &dump.fdb.obj, switchdev_port_fdb_dump_cb);
 941         return dump.idx;
 942 }
 943 EXPORT_SYMBOL_GPL(switchdev_port_fdb_dump);
 944
 945 static struct net_device *switchdev_get_lowest_dev(struct net_device *dev)
 946 {
 947         const struct switchdev_ops *ops = dev->switchdev_ops;
 948         struct net_device *lower_dev;
 949         struct net_device *port_dev;
 950         struct list_head *iter;
 951
 952         /* Recusively search down until we find a sw port dev.
 953          * (A sw port dev supports switchdev_port_attr_get).
 954          */
 955
 956         if (ops && ops->switchdev_port_attr_get)
 957                 return dev;
 958
 959         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 960                 port_dev = switchdev_get_lowest_dev(lower_dev);
 961                 if (port_dev)
 962                         return port_dev;
 963         }
 964
 965         return NULL;
 966 }
 967
 968 static struct net_device *switchdev_get_dev_by_nhs(struct fib_info *fi)
 969 {
 970         struct switchdev_attr attr = {
 971                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
 972         };
 973         struct switchdev_attr prev_attr;
 974         struct net_device *dev = NULL;
 975         int nhsel;
 976
 977         /* For this route, all nexthop devs must be on the same switch. */
 978
 979         for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
 980                 const struct fib_nh *nh = &fi->fib_nh[nhsel];
 981
 982                 if (!nh->nh_dev)
 983                         return NULL;
 984
 985                 dev = switchdev_get_lowest_dev(nh->nh_dev);
 986                 if (!dev)
 987                         return NULL;
 988
 989                 if (switchdev_port_attr_get(dev, &attr))
 990                         return NULL;
 991
 992                 if (nhsel > 0 &&
 993                     !netdev_phys_item_id_same(&prev_attr.u.ppid, &attr.u.ppid))
 994                                 return NULL;
 995
 996                 prev_attr = attr;
 997         }
 998
 999         return dev;
1000 }
1001
1002 /**
1003  *      switchdev_fib_ipv4_add - Add/modify switch IPv4 route entry
1004  *
1005  *      @dst: route's IPv4 destination address
1006  *      @dst_len: destination address length (prefix length)
1007  *      @fi: route FIB info structure
1008  *      @tos: route TOS
1009  *      @type: route type
1010  *      @nlflags: netlink flags passed in (NLM_F_*)
1011  *      @tb_id: route table ID
1012  *
1013  *      Add/modify switch IPv4 route entry.
1014  */
1015 int switchdev_fib_ipv4_add(u32 dst, int dst_len, struct fib_info *fi,
1016                            u8 tos, u8 type, u32 nlflags, u32 tb_id)
1017 {
1018         struct switchdev_obj_ipv4_fib ipv4_fib = {
1019                 .obj.id = SWITCHDEV_OBJ_ID_IPV4_FIB,
1020                 .dst = dst,
1021                 .dst_len = dst_len,
1022                 .fi = fi,
1023                 .tos = tos,
1024                 .type = type,
1025                 .nlflags = nlflags,
1026                 .tb_id = tb_id,
1027         };
1028         struct net_device *dev;
1029         int err = 0;
1030
1031         /* Don't offload route if using custom ip rules or if
1032          * IPv4 FIB offloading has been disabled completely.
1033          */
1034
1035 #ifdef CONFIG_IP_MULTIPLE_TABLES
1036         if (fi->fib_net->ipv4.fib_has_custom_rules)
1037                 return 0;
1038 #endif
1039
1040         if (fi->fib_net->ipv4.fib_offload_disabled)
1041                 return 0;
1042
1043         dev = switchdev_get_dev_by_nhs(fi);
1044         if (!dev)
1045                 return 0;
1046
1047         err = switchdev_port_obj_add(dev, &ipv4_fib.obj);
1048         if (!err)
1049                 fi->fib_flags |= RTNH_F_OFFLOAD;
1050
1051         return err == -EOPNOTSUPP ? 0 : err;
1052 }
1053 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_add);
1054
1055 /**
1056  *      switchdev_fib_ipv4_del - Delete IPv4 route entry from switch
1057  *
1058  *      @dst: route's IPv4 destination address
1059  *      @dst_len: destination address length (prefix length)
1060  *      @fi: route FIB info structure
1061  *      @tos: route TOS
1062  *      @type: route type
1063  *      @tb_id: route table ID
1064  *
1065  *      Delete IPv4 route entry from switch device.
1066  */
1067 int switchdev_fib_ipv4_del(u32 dst, int dst_len, struct fib_info *fi,
1068                            u8 tos, u8 type, u32 tb_id)
1069 {
1070         struct switchdev_obj_ipv4_fib ipv4_fib = {
1071                 .obj.id = SWITCHDEV_OBJ_ID_IPV4_FIB,
1072                 .dst = dst,
1073                 .dst_len = dst_len,
1074                 .fi = fi,
1075                 .tos = tos,
1076                 .type = type,
1077                 .nlflags = 0,
1078                 .tb_id = tb_id,
1079         };
1080         struct net_device *dev;
1081         int err = 0;
1082
1083         if (!(fi->fib_flags & RTNH_F_OFFLOAD))
1084                 return 0;
1085
1086         dev = switchdev_get_dev_by_nhs(fi);
1087         if (!dev)
1088                 return 0;
1089
1090         err = switchdev_port_obj_del(dev, &ipv4_fib.obj);
1091         if (!err)
1092                 fi->fib_flags &= ~RTNH_F_OFFLOAD;
1093
1094         return err == -EOPNOTSUPP ? 0 : err;
1095 }
1096 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_del);
1097
1098 /**
1099  *      switchdev_fib_ipv4_abort - Abort an IPv4 FIB operation
1100  *
1101  *      @fi: route FIB info structure
1102  */
1103 void switchdev_fib_ipv4_abort(struct fib_info *fi)
1104 {
1105         /* There was a problem installing this route to the offload
1106          * device.  For now, until we come up with more refined
1107          * policy handling, abruptly end IPv4 fib offloading for
1108          * for entire net by flushing offload device(s) of all
1109          * IPv4 routes, and mark IPv4 fib offloading broken from
1110          * this point forward.
1111          */
1112
1113         fib_flush_external(fi->fib_net);
1114         fi->fib_net->ipv4.fib_offload_disabled = true;
1115 }
1116 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_abort);
1117
1118 static bool switchdev_port_same_parent_id(struct net_device *a,
1119                                           struct net_device *b)
1120 {
1121         struct switchdev_attr a_attr = {
1122                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1123                 .flags = SWITCHDEV_F_NO_RECURSE,
1124         };
1125         struct switchdev_attr b_attr = {
1126                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1127                 .flags = SWITCHDEV_F_NO_RECURSE,
1128         };
1129
1130         if (switchdev_port_attr_get(a, &a_attr) ||
1131             switchdev_port_attr_get(b, &b_attr))
1132                 return false;
1133
1134         return netdev_phys_item_id_same(&a_attr.u.ppid, &b_attr.u.ppid);
1135 }
1136
1137 static u32 switchdev_port_fwd_mark_get(struct net_device *dev,
1138                                        struct net_device *group_dev)
1139 {
1140         struct net_device *lower_dev;
1141         struct list_head *iter;
1142
1143         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1144                 if (lower_dev == dev)
1145                         continue;
1146                 if (switchdev_port_same_parent_id(dev, lower_dev))
1147                         return lower_dev->offload_fwd_mark;
1148                 return switchdev_port_fwd_mark_get(dev, lower_dev);
1149         }
1150
1151         return dev->ifindex;
1152 }
1153
1154 static void switchdev_port_fwd_mark_reset(struct net_device *group_dev,
1155                                           u32 old_mark, u32 *reset_mark)
1156 {
1157         struct net_device *lower_dev;
1158         struct list_head *iter;
1159
1160         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1161                 if (lower_dev->offload_fwd_mark == old_mark) {
1162                         if (!*reset_mark)
1163                                 *reset_mark = lower_dev->ifindex;
1164                         lower_dev->offload_fwd_mark = *reset_mark;
1165                 }
1166                 switchdev_port_fwd_mark_reset(lower_dev, old_mark, reset_mark);
1167         }
1168 }
1169
1170 /**
1171  *      switchdev_port_fwd_mark_set - Set port offload forwarding mark
1172  *
1173  *      @dev: port device
1174  *      @group_dev: containing device
1175  *      @joining: true if dev is joining group; false if leaving group
1176  *
1177  *      An ungrouped port's offload mark is just its ifindex.  A grouped
1178  *      port's (member of a bridge, for example) offload mark is the ifindex
1179  *      of one of the ports in the group with the same parent (switch) ID.
1180  *      Ports on the same device in the same group will have the same mark.
1181  *
1182  *      Example:
1183  *
1184  *              br0             ifindex=9
1185  *                sw1p1         ifindex=2       mark=2
1186  *                sw1p2         ifindex=3       mark=2
1187  *                sw2p1         ifindex=4       mark=5
1188  *                sw2p2         ifindex=5       mark=5
1189  *
1190  *      If sw2p2 leaves the bridge, we'll have:
1191  *
1192  *              br0             ifindex=9
1193  *                sw1p1         ifindex=2       mark=2
1194  *                sw1p2         ifindex=3       mark=2
1195  *                sw2p1         ifindex=4       mark=4
1196  *              sw2p2           ifindex=5       mark=5
1197  */
1198 void switchdev_port_fwd_mark_set(struct net_device *dev,
1199                                  struct net_device *group_dev,
1200                                  bool joining)
1201 {
1202         u32 mark = dev->ifindex;
1203         u32 reset_mark = 0;
1204
1205         if (group_dev && joining) {
1206                 mark = switchdev_port_fwd_mark_get(dev, group_dev);
1207         } else if (group_dev && !joining) {
1208                 if (dev->offload_fwd_mark == mark)
1209                         /* Ohoh, this port was the mark reference port,
1210                          * but it's leaving the group, so reset the
1211                          * mark for the remaining ports in the group.
1212                          */
1213                         switchdev_port_fwd_mark_reset(group_dev, mark,
1214                                                       &reset_mark);
1215         }
1216
1217         dev->offload_fwd_mark = mark;
1218 }
1219 EXPORT_SYMBOL_GPL(switchdev_port_fwd_mark_set);