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                                     const 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  *      @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,
 312                            const struct switchdev_obj *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, 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, obj, &trans);
 349         WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->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,
 364                            const struct switchdev_obj *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, 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, 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, struct switchdev_obj *obj,
 398                             switchdev_obj_dump_cb_t *cb)
 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, 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, 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_port_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(struct switchdev_obj *obj)
 525 {
 526         struct switchdev_obj_port_vlan *vlan = SWITCHDEV_OBJ_PORT_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                 .vlan.obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
 581                 .skb = skb,
 582                 .filter_mask = filter_mask,
 583         };
 584         int err = 0;
 585
 586         if ((filter_mask & RTEXT_FILTER_BRVLAN) ||
 587             (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)) {
 588                 err = switchdev_port_obj_dump(dev, &dump.vlan.obj,
 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                                              const struct switchdev_obj *obj))
 703 {
 704         struct nlattr *attr;
 705         struct bridge_vlan_info *vinfo;
 706         struct switchdev_obj_port_vlan vlan = {
 707                 .obj.id = SWITCHDEV_OBJ_ID_PORT_VLAN,
 708         };
 709         int rem;
 710         int err;
 711
 712         nla_for_each_nested(attr, afspec, rem) {
 713                 if (nla_type(attr) != IFLA_BRIDGE_VLAN_INFO)
 714                         continue;
 715                 if (nla_len(attr) != sizeof(struct bridge_vlan_info))
 716                         return -EINVAL;
 717                 vinfo = nla_data(attr);
 718                 vlan.flags = vinfo->flags;
 719                 if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_BEGIN) {
 720                         if (vlan.vid_begin)
 721                                 return -EINVAL;
 722                         vlan.vid_begin = vinfo->vid;
 723                 } else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
 724                         if (!vlan.vid_begin)
 725                                 return -EINVAL;
 726                         vlan.vid_end = vinfo->vid;
 727                         if (vlan.vid_end <= vlan.vid_begin)
 728                                 return -EINVAL;
 729                         err = f(dev, &vlan.obj);
 730                         if (err)
 731                                 return err;
 732                         memset(&vlan, 0, sizeof(vlan));
 733                 } else {
 734                         if (vlan.vid_begin)
 735                                 return -EINVAL;
 736                         vlan.vid_begin = vinfo->vid;
 737                         vlan.vid_end = vinfo->vid;
 738                         err = f(dev, &vlan.obj);
 739                         if (err)
 740                                 return err;
 741                         memset(&vlan, 0, sizeof(vlan));
 742                 }
 743         }
 744
 745         return 0;
 746 }
 747
 748 /**
 749  *      switchdev_port_bridge_setlink - Set bridge port attributes
 750  *
 751  *      @dev: port device
 752  *      @nlh: netlink header
 753  *      @flags: netlink flags
 754  *
 755  *      Called for SELF on rtnl_bridge_setlink to set bridge port
 756  *      attributes.
 757  */
 758 int switchdev_port_bridge_setlink(struct net_device *dev,
 759                                   struct nlmsghdr *nlh, u16 flags)
 760 {
 761         struct nlattr *protinfo;
 762         struct nlattr *afspec;
 763         int err = 0;
 764
 765         protinfo = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 766                                    IFLA_PROTINFO);
 767         if (protinfo) {
 768                 err = switchdev_port_br_setlink_protinfo(dev, protinfo);
 769                 if (err)
 770                         return err;
 771         }
 772
 773         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 774                                  IFLA_AF_SPEC);
 775         if (afspec)
 776                 err = switchdev_port_br_afspec(dev, afspec,
 777                                                switchdev_port_obj_add);
 778
 779         return err;
 780 }
 781 EXPORT_SYMBOL_GPL(switchdev_port_bridge_setlink);
 782
 783 /**
 784  *      switchdev_port_bridge_dellink - Set bridge port attributes
 785  *
 786  *      @dev: port device
 787  *      @nlh: netlink header
 788  *      @flags: netlink flags
 789  *
 790  *      Called for SELF on rtnl_bridge_dellink to set bridge port
 791  *      attributes.
 792  */
 793 int switchdev_port_bridge_dellink(struct net_device *dev,
 794                                   struct nlmsghdr *nlh, u16 flags)
 795 {
 796         struct nlattr *afspec;
 797
 798         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 799                                  IFLA_AF_SPEC);
 800         if (afspec)
 801                 return switchdev_port_br_afspec(dev, afspec,
 802                                                 switchdev_port_obj_del);
 803
 804         return 0;
 805 }
 806 EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
 807
 808 /**
 809  *      switchdev_port_fdb_add - Add FDB (MAC/VLAN) entry to port
 810  *
 811  *      @ndmsg: netlink hdr
 812  *      @nlattr: netlink attributes
 813  *      @dev: port device
 814  *      @addr: MAC address to add
 815  *      @vid: VLAN to add
 816  *
 817  *      Add FDB entry to switch device.
 818  */
 819 int switchdev_port_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 820                            struct net_device *dev, const unsigned char *addr,
 821                            u16 vid, u16 nlm_flags)
 822 {
 823         struct switchdev_obj_port_fdb fdb = {
 824                 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
 825                 .addr = addr,
 826                 .vid = vid,
 827         };
 828
 829         return switchdev_port_obj_add(dev, &fdb.obj);
 830 }
 831 EXPORT_SYMBOL_GPL(switchdev_port_fdb_add);
 832
 833 /**
 834  *      switchdev_port_fdb_del - Delete FDB (MAC/VLAN) entry from port
 835  *
 836  *      @ndmsg: netlink hdr
 837  *      @nlattr: netlink attributes
 838  *      @dev: port device
 839  *      @addr: MAC address to delete
 840  *      @vid: VLAN to delete
 841  *
 842  *      Delete FDB entry from switch device.
 843  */
 844 int switchdev_port_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 845                            struct net_device *dev, const unsigned char *addr,
 846                            u16 vid)
 847 {
 848         struct switchdev_obj_port_fdb fdb = {
 849                 .obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
 850                 .addr = addr,
 851                 .vid = vid,
 852         };
 853
 854         return switchdev_port_obj_del(dev, &fdb.obj);
 855 }
 856 EXPORT_SYMBOL_GPL(switchdev_port_fdb_del);
 857
 858 struct switchdev_fdb_dump {
 859         struct switchdev_obj_port_fdb fdb;
 860         struct net_device *dev;
 861         struct sk_buff *skb;
 862         struct netlink_callback *cb;
 863         int idx;
 864 };
 865
 866 static int switchdev_port_fdb_dump_cb(struct switchdev_obj *obj)
 867 {
 868         struct switchdev_obj_port_fdb *fdb = SWITCHDEV_OBJ_PORT_FDB(obj);
 869         struct switchdev_fdb_dump *dump =
 870                 container_of(fdb, struct switchdev_fdb_dump, fdb);
 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 = dump->dev->ifindex;
 891         ndm->ndm_state   = fdb->ndm_state;
 892
 893         if (nla_put(dump->skb, NDA_LLADDR, ETH_ALEN, fdb->addr))
 894                 goto nla_put_failure;
 895
 896         if (fdb->vid && nla_put_u16(dump->skb, NDA_VLAN, 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                 .fdb.obj.id = SWITCHDEV_OBJ_ID_PORT_FDB,
 927                 .dev = dev,
 928                 .skb = skb,
 929                 .cb = cb,
 930                 .idx = idx,
 931         };
 932
 933         switchdev_port_obj_dump(dev, &dump.fdb.obj, switchdev_port_fdb_dump_cb);
 934         return dump.idx;
 935 }
 936 EXPORT_SYMBOL_GPL(switchdev_port_fdb_dump);
 937
 938 static struct net_device *switchdev_get_lowest_dev(struct net_device *dev)
 939 {
 940         const struct switchdev_ops *ops = dev->switchdev_ops;
 941         struct net_device *lower_dev;
 942         struct net_device *port_dev;
 943         struct list_head *iter;
 944
 945         /* Recusively search down until we find a sw port dev.
 946          * (A sw port dev supports switchdev_port_attr_get).
 947          */
 948
 949         if (ops && ops->switchdev_port_attr_get)
 950                 return dev;
 951
 952         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 953                 port_dev = switchdev_get_lowest_dev(lower_dev);
 954                 if (port_dev)
 955                         return port_dev;
 956         }
 957
 958         return NULL;
 959 }
 960
 961 static struct net_device *switchdev_get_dev_by_nhs(struct fib_info *fi)
 962 {
 963         struct switchdev_attr attr = {
 964                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
 965         };
 966         struct switchdev_attr prev_attr;
 967         struct net_device *dev = NULL;
 968         int nhsel;
 969
 970         /* For this route, all nexthop devs must be on the same switch. */
 971
 972         for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
 973                 const struct fib_nh *nh = &fi->fib_nh[nhsel];
 974
 975                 if (!nh->nh_dev)
 976                         return NULL;
 977
 978                 dev = switchdev_get_lowest_dev(nh->nh_dev);
 979                 if (!dev)
 980                         return NULL;
 981
 982                 if (switchdev_port_attr_get(dev, &attr))
 983                         return NULL;
 984
 985                 if (nhsel > 0 &&
 986                     !netdev_phys_item_id_same(&prev_attr.u.ppid, &attr.u.ppid))
 987                                 return NULL;
 988
 989                 prev_attr = attr;
 990         }
 991
 992         return dev;
 993 }
 994
 995 /**
 996  *      switchdev_fib_ipv4_add - Add/modify switch IPv4 route entry
 997  *
 998  *      @dst: route's IPv4 destination address
 999  *      @dst_len: destination address length (prefix length)
1000  *      @fi: route FIB info structure
1001  *      @tos: route TOS
1002  *      @type: route type
1003  *      @nlflags: netlink flags passed in (NLM_F_*)
1004  *      @tb_id: route table ID
1005  *
1006  *      Add/modify switch IPv4 route entry.
1007  */
1008 int switchdev_fib_ipv4_add(u32 dst, int dst_len, struct fib_info *fi,
1009                            u8 tos, u8 type, u32 nlflags, u32 tb_id)
1010 {
1011         struct switchdev_obj_ipv4_fib ipv4_fib = {
1012                 .obj.id = SWITCHDEV_OBJ_ID_IPV4_FIB,
1013                 .dst = dst,
1014                 .dst_len = dst_len,
1015                 .fi = fi,
1016                 .tos = tos,
1017                 .type = type,
1018                 .nlflags = nlflags,
1019                 .tb_id = tb_id,
1020         };
1021         struct net_device *dev;
1022         int err = 0;
1023
1024         /* Don't offload route if using custom ip rules or if
1025          * IPv4 FIB offloading has been disabled completely.
1026          */
1027
1028 #ifdef CONFIG_IP_MULTIPLE_TABLES
1029         if (fi->fib_net->ipv4.fib_has_custom_rules)
1030                 return 0;
1031 #endif
1032
1033         if (fi->fib_net->ipv4.fib_offload_disabled)
1034                 return 0;
1035
1036         dev = switchdev_get_dev_by_nhs(fi);
1037         if (!dev)
1038                 return 0;
1039
1040         err = switchdev_port_obj_add(dev, &ipv4_fib.obj);
1041         if (!err)
1042                 fi->fib_flags |= RTNH_F_OFFLOAD;
1043
1044         return err == -EOPNOTSUPP ? 0 : err;
1045 }
1046 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_add);
1047
1048 /**
1049  *      switchdev_fib_ipv4_del - Delete IPv4 route entry from switch
1050  *
1051  *      @dst: route's IPv4 destination address
1052  *      @dst_len: destination address length (prefix length)
1053  *      @fi: route FIB info structure
1054  *      @tos: route TOS
1055  *      @type: route type
1056  *      @tb_id: route table ID
1057  *
1058  *      Delete IPv4 route entry from switch device.
1059  */
1060 int switchdev_fib_ipv4_del(u32 dst, int dst_len, struct fib_info *fi,
1061                            u8 tos, u8 type, u32 tb_id)
1062 {
1063         struct switchdev_obj_ipv4_fib ipv4_fib = {
1064                 .obj.id = SWITCHDEV_OBJ_ID_IPV4_FIB,
1065                 .dst = dst,
1066                 .dst_len = dst_len,
1067                 .fi = fi,
1068                 .tos = tos,
1069                 .type = type,
1070                 .nlflags = 0,
1071                 .tb_id = tb_id,
1072         };
1073         struct net_device *dev;
1074         int err = 0;
1075
1076         if (!(fi->fib_flags & RTNH_F_OFFLOAD))
1077                 return 0;
1078
1079         dev = switchdev_get_dev_by_nhs(fi);
1080         if (!dev)
1081                 return 0;
1082
1083         err = switchdev_port_obj_del(dev, &ipv4_fib.obj);
1084         if (!err)
1085                 fi->fib_flags &= ~RTNH_F_OFFLOAD;
1086
1087         return err == -EOPNOTSUPP ? 0 : err;
1088 }
1089 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_del);
1090
1091 /**
1092  *      switchdev_fib_ipv4_abort - Abort an IPv4 FIB operation
1093  *
1094  *      @fi: route FIB info structure
1095  */
1096 void switchdev_fib_ipv4_abort(struct fib_info *fi)
1097 {
1098         /* There was a problem installing this route to the offload
1099          * device.  For now, until we come up with more refined
1100          * policy handling, abruptly end IPv4 fib offloading for
1101          * for entire net by flushing offload device(s) of all
1102          * IPv4 routes, and mark IPv4 fib offloading broken from
1103          * this point forward.
1104          */
1105
1106         fib_flush_external(fi->fib_net);
1107         fi->fib_net->ipv4.fib_offload_disabled = true;
1108 }
1109 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_abort);
1110
1111 static bool switchdev_port_same_parent_id(struct net_device *a,
1112                                           struct net_device *b)
1113 {
1114         struct switchdev_attr a_attr = {
1115                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1116                 .flags = SWITCHDEV_F_NO_RECURSE,
1117         };
1118         struct switchdev_attr b_attr = {
1119                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
1120                 .flags = SWITCHDEV_F_NO_RECURSE,
1121         };
1122
1123         if (switchdev_port_attr_get(a, &a_attr) ||
1124             switchdev_port_attr_get(b, &b_attr))
1125                 return false;
1126
1127         return netdev_phys_item_id_same(&a_attr.u.ppid, &b_attr.u.ppid);
1128 }
1129
1130 static u32 switchdev_port_fwd_mark_get(struct net_device *dev,
1131                                        struct net_device *group_dev)
1132 {
1133         struct net_device *lower_dev;
1134         struct list_head *iter;
1135
1136         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1137                 if (lower_dev == dev)
1138                         continue;
1139                 if (switchdev_port_same_parent_id(dev, lower_dev))
1140                         return lower_dev->offload_fwd_mark;
1141                 return switchdev_port_fwd_mark_get(dev, lower_dev);
1142         }
1143
1144         return dev->ifindex;
1145 }
1146
1147 static void switchdev_port_fwd_mark_reset(struct net_device *group_dev,
1148                                           u32 old_mark, u32 *reset_mark)
1149 {
1150         struct net_device *lower_dev;
1151         struct list_head *iter;
1152
1153         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1154                 if (lower_dev->offload_fwd_mark == old_mark) {
1155                         if (!*reset_mark)
1156                                 *reset_mark = lower_dev->ifindex;
1157                         lower_dev->offload_fwd_mark = *reset_mark;
1158                 }
1159                 switchdev_port_fwd_mark_reset(lower_dev, old_mark, reset_mark);
1160         }
1161 }
1162
1163 /**
1164  *      switchdev_port_fwd_mark_set - Set port offload forwarding mark
1165  *
1166  *      @dev: port device
1167  *      @group_dev: containing device
1168  *      @joining: true if dev is joining group; false if leaving group
1169  *
1170  *      An ungrouped port's offload mark is just its ifindex.  A grouped
1171  *      port's (member of a bridge, for example) offload mark is the ifindex
1172  *      of one of the ports in the group with the same parent (switch) ID.
1173  *      Ports on the same device in the same group will have the same mark.
1174  *
1175  *      Example:
1176  *
1177  *              br0             ifindex=9
1178  *                sw1p1         ifindex=2       mark=2
1179  *                sw1p2         ifindex=3       mark=2
1180  *                sw2p1         ifindex=4       mark=5
1181  *                sw2p2         ifindex=5       mark=5
1182  *
1183  *      If sw2p2 leaves the bridge, we'll have:
1184  *
1185  *              br0             ifindex=9
1186  *                sw1p1         ifindex=2       mark=2
1187  *                sw1p2         ifindex=3       mark=2
1188  *                sw2p1         ifindex=4       mark=4
1189  *              sw2p2           ifindex=5       mark=5
1190  */
1191 void switchdev_port_fwd_mark_set(struct net_device *dev,
1192                                  struct net_device *group_dev,
1193                                  bool joining)
1194 {
1195         u32 mark = dev->ifindex;
1196         u32 reset_mark = 0;
1197
1198         if (group_dev && joining) {
1199                 mark = switchdev_port_fwd_mark_get(dev, group_dev);
1200         } else if (group_dev && !joining) {
1201                 if (dev->offload_fwd_mark == mark)
1202                         /* Ohoh, this port was the mark reference port,
1203                          * but it's leaving the group, so reset the
1204                          * mark for the remaining ports in the group.
1205                          */
1206                         switchdev_port_fwd_mark_reset(group_dev, mark,
1207                                                       &reset_mark);
1208         }
1209
1210         dev->offload_fwd_mark = mark;
1211 }
1212 EXPORT_SYMBOL_GPL(switchdev_port_fwd_mark_set);