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