net/switchdev/switchdev.c

   1 /*
   2  * net/switchdev/switchdev.c - Switch device API
   3  * Copyright (c) 2014 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 <net/ip_fib.h>
  20 #include <net/switchdev.h>
  21
  22 /**
  23  *      switchdev_port_attr_get - Get port attribute
  24  *
  25  *      @dev: port device
  26  *      @attr: attribute to get
  27  */
  28 int switchdev_port_attr_get(struct net_device *dev, struct switchdev_attr *attr)
  29 {
  30         const struct switchdev_ops *ops = dev->switchdev_ops;
  31         struct net_device *lower_dev;
  32         struct list_head *iter;
  33         struct switchdev_attr first = {
  34                 .id = SWITCHDEV_ATTR_UNDEFINED
  35         };
  36         int err = -EOPNOTSUPP;
  37
  38         if (ops && ops->switchdev_port_attr_get)
  39                 return ops->switchdev_port_attr_get(dev, attr);
  40
  41         if (attr->flags & SWITCHDEV_F_NO_RECURSE)
  42                 return err;
  43
  44         /* Switch device port(s) may be stacked under
  45          * bond/team/vlan dev, so recurse down to get attr on
  46          * each port.  Return -ENODATA if attr values don't
  47          * compare across ports.
  48          */
  49
  50         netdev_for_each_lower_dev(dev, lower_dev, iter) {
  51                 err = switchdev_port_attr_get(lower_dev, attr);
  52                 if (err)
  53                         break;
  54                 if (first.id == SWITCHDEV_ATTR_UNDEFINED)
  55                         first = *attr;
  56                 else if (memcmp(&first, attr, sizeof(*attr)))
  57                         return -ENODATA;
  58         }
  59
  60         return err;
  61 }
  62 EXPORT_SYMBOL_GPL(switchdev_port_attr_get);
  63
  64 static int __switchdev_port_attr_set(struct net_device *dev,
  65                                      struct switchdev_attr *attr)
  66 {
  67         const struct switchdev_ops *ops = dev->switchdev_ops;
  68         struct net_device *lower_dev;
  69         struct list_head *iter;
  70         int err = -EOPNOTSUPP;
  71
  72         if (ops && ops->switchdev_port_attr_set)
  73                 return ops->switchdev_port_attr_set(dev, attr);
  74
  75         if (attr->flags & SWITCHDEV_F_NO_RECURSE)
  76                 return err;
  77
  78         /* Switch device port(s) may be stacked under
  79          * bond/team/vlan dev, so recurse down to set attr on
  80          * each port.
  81          */
  82
  83         netdev_for_each_lower_dev(dev, lower_dev, iter) {
  84                 err = __switchdev_port_attr_set(lower_dev, attr);
  85                 if (err)
  86                         break;
  87         }
  88
  89         return err;
  90 }
  91
  92 struct switchdev_attr_set_work {
  93         struct work_struct work;
  94         struct net_device *dev;
  95         struct switchdev_attr attr;
  96 };
  97
  98 static void switchdev_port_attr_set_work(struct work_struct *work)
  99 {
 100         struct switchdev_attr_set_work *asw =
 101                 container_of(work, struct switchdev_attr_set_work, work);
 102         int err;
 103
 104         rtnl_lock();
 105         err = switchdev_port_attr_set(asw->dev, &asw->attr);
 106         if (err && err != -EOPNOTSUPP)
 107                 netdev_err(asw->dev, "failed (err=%d) to set attribute (id=%d)\n",
 108                            err, asw->attr.id);
 109         rtnl_unlock();
 110
 111         dev_put(asw->dev);
 112         kfree(work);
 113 }
 114
 115 static int switchdev_port_attr_set_defer(struct net_device *dev,
 116                                          struct switchdev_attr *attr)
 117 {
 118         struct switchdev_attr_set_work *asw;
 119
 120         asw = kmalloc(sizeof(*asw), GFP_ATOMIC);
 121         if (!asw)
 122                 return -ENOMEM;
 123
 124         INIT_WORK(&asw->work, switchdev_port_attr_set_work);
 125
 126         dev_hold(dev);
 127         asw->dev = dev;
 128         memcpy(&asw->attr, attr, sizeof(asw->attr));
 129
 130         schedule_work(&asw->work);
 131
 132         return 0;
 133 }
 134
 135 /**
 136  *      switchdev_port_attr_set - Set port attribute
 137  *
 138  *      @dev: port device
 139  *      @attr: attribute to set
 140  *
 141  *      Use a 2-phase prepare-commit transaction model to ensure
 142  *      system is not left in a partially updated state due to
 143  *      failure from driver/device.
 144  */
 145 int switchdev_port_attr_set(struct net_device *dev, struct switchdev_attr *attr)
 146 {
 147         int err;
 148
 149         if (!rtnl_is_locked()) {
 150                 /* Running prepare-commit transaction across stacked
 151                  * devices requires nothing moves, so if rtnl_lock is
 152                  * not held, schedule a worker thread to hold rtnl_lock
 153                  * while setting attr.
 154                  */
 155
 156                 return switchdev_port_attr_set_defer(dev, attr);
 157         }
 158
 159         /* Phase I: prepare for attr set. Driver/device should fail
 160          * here if there are going to be issues in the commit phase,
 161          * such as lack of resources or support.  The driver/device
 162          * should reserve resources needed for the commit phase here,
 163          * but should not commit the attr.
 164          */
 165
 166         attr->trans = SWITCHDEV_TRANS_PREPARE;
 167         err = __switchdev_port_attr_set(dev, attr);
 168         if (err) {
 169                 /* Prepare phase failed: abort the transaction.  Any
 170                  * resources reserved in the prepare phase are
 171                  * released.
 172                  */
 173
 174                 if (err != -EOPNOTSUPP) {
 175                         attr->trans = SWITCHDEV_TRANS_ABORT;
 176                         __switchdev_port_attr_set(dev, attr);
 177                 }
 178
 179                 return err;
 180         }
 181
 182         /* Phase II: commit attr set.  This cannot fail as a fault
 183          * of driver/device.  If it does, it's a bug in the driver/device
 184          * because the driver said everythings was OK in phase I.
 185          */
 186
 187         attr->trans = SWITCHDEV_TRANS_COMMIT;
 188         err = __switchdev_port_attr_set(dev, attr);
 189         WARN(err, "%s: Commit of attribute (id=%d) failed.\n",
 190              dev->name, attr->id);
 191
 192         return err;
 193 }
 194 EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
 195
 196 static int __switchdev_port_obj_add(struct net_device *dev,
 197                                     struct switchdev_obj *obj)
 198 {
 199         const struct switchdev_ops *ops = dev->switchdev_ops;
 200         struct net_device *lower_dev;
 201         struct list_head *iter;
 202         int err = -EOPNOTSUPP;
 203
 204         if (ops && ops->switchdev_port_obj_add)
 205                 return ops->switchdev_port_obj_add(dev, obj);
 206
 207         /* Switch device port(s) may be stacked under
 208          * bond/team/vlan dev, so recurse down to add object on
 209          * each port.
 210          */
 211
 212         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 213                 err = __switchdev_port_obj_add(lower_dev, obj);
 214                 if (err)
 215                         break;
 216         }
 217
 218         return err;
 219 }
 220
 221 /**
 222  *      switchdev_port_obj_add - Add port object
 223  *
 224  *      @dev: port device
 225  *      @obj: object to add
 226  *
 227  *      Use a 2-phase prepare-commit transaction model to ensure
 228  *      system is not left in a partially updated state due to
 229  *      failure from driver/device.
 230  *
 231  *      rtnl_lock must be held.
 232  */
 233 int switchdev_port_obj_add(struct net_device *dev, struct switchdev_obj *obj)
 234 {
 235         int err;
 236
 237         ASSERT_RTNL();
 238
 239         /* Phase I: prepare for obj add. Driver/device should fail
 240          * here if there are going to be issues in the commit phase,
 241          * such as lack of resources or support.  The driver/device
 242          * should reserve resources needed for the commit phase here,
 243          * but should not commit the obj.
 244          */
 245
 246         obj->trans = SWITCHDEV_TRANS_PREPARE;
 247         err = __switchdev_port_obj_add(dev, obj);
 248         if (err) {
 249                 /* Prepare phase failed: abort the transaction.  Any
 250                  * resources reserved in the prepare phase are
 251                  * released.
 252                  */
 253
 254                 if (err != -EOPNOTSUPP) {
 255                         obj->trans = SWITCHDEV_TRANS_ABORT;
 256                         __switchdev_port_obj_add(dev, obj);
 257                 }
 258
 259                 return err;
 260         }
 261
 262         /* Phase II: commit obj add.  This cannot fail as a fault
 263          * of driver/device.  If it does, it's a bug in the driver/device
 264          * because the driver said everythings was OK in phase I.
 265          */
 266
 267         obj->trans = SWITCHDEV_TRANS_COMMIT;
 268         err = __switchdev_port_obj_add(dev, obj);
 269         WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->id);
 270
 271         return err;
 272 }
 273 EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
 274
 275 /**
 276  *      switchdev_port_obj_del - Delete port object
 277  *
 278  *      @dev: port device
 279  *      @obj: object to delete
 280  */
 281 int switchdev_port_obj_del(struct net_device *dev, struct switchdev_obj *obj)
 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_del)
 289                 return ops->switchdev_port_obj_del(dev, obj);
 290
 291         /* Switch device port(s) may be stacked under
 292          * bond/team/vlan dev, so recurse down to delete object on
 293          * each port.
 294          */
 295
 296         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 297                 err = switchdev_port_obj_del(lower_dev, obj);
 298                 if (err)
 299                         break;
 300         }
 301
 302         return err;
 303 }
 304 EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
 305
 306 /**
 307  *      switchdev_port_obj_dump - Dump port objects
 308  *
 309  *      @dev: port device
 310  *      @obj: object to dump
 311  */
 312 int switchdev_port_obj_dump(struct net_device *dev, struct switchdev_obj *obj)
 313 {
 314         const struct switchdev_ops *ops = dev->switchdev_ops;
 315         struct net_device *lower_dev;
 316         struct list_head *iter;
 317         int err = -EOPNOTSUPP;
 318
 319         if (ops && ops->switchdev_port_obj_dump)
 320                 return ops->switchdev_port_obj_dump(dev, obj);
 321
 322         /* Switch device port(s) may be stacked under
 323          * bond/team/vlan dev, so recurse down to dump objects on
 324          * first port at bottom of stack.
 325          */
 326
 327         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 328                 err = switchdev_port_obj_dump(lower_dev, obj);
 329                 break;
 330         }
 331
 332         return err;
 333 }
 334 EXPORT_SYMBOL_GPL(switchdev_port_obj_dump);
 335
 336 static DEFINE_MUTEX(switchdev_mutex);
 337 static RAW_NOTIFIER_HEAD(switchdev_notif_chain);
 338
 339 /**
 340  *      register_switchdev_notifier - Register notifier
 341  *      @nb: notifier_block
 342  *
 343  *      Register switch device notifier. This should be used by code
 344  *      which needs to monitor events happening in particular device.
 345  *      Return values are same as for atomic_notifier_chain_register().
 346  */
 347 int register_switchdev_notifier(struct notifier_block *nb)
 348 {
 349         int err;
 350
 351         mutex_lock(&switchdev_mutex);
 352         err = raw_notifier_chain_register(&switchdev_notif_chain, nb);
 353         mutex_unlock(&switchdev_mutex);
 354         return err;
 355 }
 356 EXPORT_SYMBOL_GPL(register_switchdev_notifier);
 357
 358 /**
 359  *      unregister_switchdev_notifier - Unregister notifier
 360  *      @nb: notifier_block
 361  *
 362  *      Unregister switch device notifier.
 363  *      Return values are same as for atomic_notifier_chain_unregister().
 364  */
 365 int unregister_switchdev_notifier(struct notifier_block *nb)
 366 {
 367         int err;
 368
 369         mutex_lock(&switchdev_mutex);
 370         err = raw_notifier_chain_unregister(&switchdev_notif_chain, nb);
 371         mutex_unlock(&switchdev_mutex);
 372         return err;
 373 }
 374 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
 375
 376 /**
 377  *      call_switchdev_notifiers - Call notifiers
 378  *      @val: value passed unmodified to notifier function
 379  *      @dev: port device
 380  *      @info: notifier information data
 381  *
 382  *      Call all network notifier blocks. This should be called by driver
 383  *      when it needs to propagate hardware event.
 384  *      Return values are same as for atomic_notifier_call_chain().
 385  */
 386 int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
 387                              struct switchdev_notifier_info *info)
 388 {
 389         int err;
 390
 391         info->dev = dev;
 392         mutex_lock(&switchdev_mutex);
 393         err = raw_notifier_call_chain(&switchdev_notif_chain, val, info);
 394         mutex_unlock(&switchdev_mutex);
 395         return err;
 396 }
 397 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
 398
 399 struct switchdev_vlan_dump {
 400         struct switchdev_obj obj;
 401         struct sk_buff *skb;
 402         u32 filter_mask;
 403         u16 flags;
 404         u16 begin;
 405         u16 end;
 406 };
 407
 408 static int switchdev_port_vlan_dump_put(struct net_device *dev,
 409                                         struct switchdev_vlan_dump *dump)
 410 {
 411         struct bridge_vlan_info vinfo;
 412
 413         vinfo.flags = dump->flags;
 414
 415         if (dump->begin == 0 && dump->end == 0) {
 416                 return 0;
 417         } else if (dump->begin == dump->end) {
 418                 vinfo.vid = dump->begin;
 419                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 420                             sizeof(vinfo), &vinfo))
 421                         return -EMSGSIZE;
 422         } else {
 423                 vinfo.vid = dump->begin;
 424                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_BEGIN;
 425                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 426                             sizeof(vinfo), &vinfo))
 427                         return -EMSGSIZE;
 428                 vinfo.vid = dump->end;
 429                 vinfo.flags &= ~BRIDGE_VLAN_INFO_RANGE_BEGIN;
 430                 vinfo.flags |= BRIDGE_VLAN_INFO_RANGE_END;
 431                 if (nla_put(dump->skb, IFLA_BRIDGE_VLAN_INFO,
 432                             sizeof(vinfo), &vinfo))
 433                         return -EMSGSIZE;
 434         }
 435
 436         return 0;
 437 }
 438
 439 static int switchdev_port_vlan_dump_cb(struct net_device *dev,
 440                                        struct switchdev_obj *obj)
 441 {
 442         struct switchdev_vlan_dump *dump =
 443                 container_of(obj, struct switchdev_vlan_dump, obj);
 444         struct switchdev_obj_vlan *vlan = &dump->obj.u.vlan;
 445         int err = 0;
 446
 447         if (vlan->vid_begin > vlan->vid_end)
 448                 return -EINVAL;
 449
 450         if (dump->filter_mask & RTEXT_FILTER_BRVLAN) {
 451                 dump->flags = vlan->flags;
 452                 for (dump->begin = dump->end = vlan->vid_begin;
 453                      dump->begin <= vlan->vid_end;
 454                      dump->begin++, dump->end++) {
 455                         err = switchdev_port_vlan_dump_put(dev, dump);
 456                         if (err)
 457                                 return err;
 458                 }
 459         } else if (dump->filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED) {
 460                 if (dump->begin > vlan->vid_begin &&
 461                     dump->begin >= vlan->vid_end) {
 462                         if ((dump->begin - 1) == vlan->vid_end &&
 463                             dump->flags == vlan->flags) {
 464                                 /* prepend */
 465                                 dump->begin = vlan->vid_begin;
 466                         } else {
 467                                 err = switchdev_port_vlan_dump_put(dev, dump);
 468                                 dump->flags = vlan->flags;
 469                                 dump->begin = vlan->vid_begin;
 470                                 dump->end = vlan->vid_end;
 471                         }
 472                 } else if (dump->end <= vlan->vid_begin &&
 473                            dump->end < vlan->vid_end) {
 474                         if ((dump->end  + 1) == vlan->vid_begin &&
 475                             dump->flags == vlan->flags) {
 476                                 /* append */
 477                                 dump->end = vlan->vid_end;
 478                         } else {
 479                                 err = switchdev_port_vlan_dump_put(dev, dump);
 480                                 dump->flags = vlan->flags;
 481                                 dump->begin = vlan->vid_begin;
 482                                 dump->end = vlan->vid_end;
 483                         }
 484                 } else {
 485                         err = -EINVAL;
 486                 }
 487         }
 488
 489         return err;
 490 }
 491
 492 static int switchdev_port_vlan_fill(struct sk_buff *skb, struct net_device *dev,
 493                                     u32 filter_mask)
 494 {
 495         struct switchdev_vlan_dump dump = {
 496                 .obj = {
 497                         .id = SWITCHDEV_OBJ_PORT_VLAN,
 498                         .cb = switchdev_port_vlan_dump_cb,
 499                 },
 500                 .skb = skb,
 501                 .filter_mask = filter_mask,
 502         };
 503         int err = 0;
 504
 505         if ((filter_mask & RTEXT_FILTER_BRVLAN) ||
 506             (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)) {
 507                 err = switchdev_port_obj_dump(dev, &dump.obj);
 508                 if (err)
 509                         goto err_out;
 510                 if (filter_mask & RTEXT_FILTER_BRVLAN_COMPRESSED)
 511                         /* last one */
 512                         err = switchdev_port_vlan_dump_put(dev, &dump);
 513         }
 514
 515 err_out:
 516         return err == -EOPNOTSUPP ? 0 : err;
 517 }
 518
 519 /**
 520  *      switchdev_port_bridge_getlink - Get bridge port attributes
 521  *
 522  *      @dev: port device
 523  *
 524  *      Called for SELF on rtnl_bridge_getlink to get bridge port
 525  *      attributes.
 526  */
 527 int switchdev_port_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
 528                                   struct net_device *dev, u32 filter_mask,
 529                                   int nlflags)
 530 {
 531         struct switchdev_attr attr = {
 532                 .id = SWITCHDEV_ATTR_PORT_BRIDGE_FLAGS,
 533         };
 534         u16 mode = BRIDGE_MODE_UNDEF;
 535         u32 mask = BR_LEARNING | BR_LEARNING_SYNC;
 536         int err;
 537
 538         err = switchdev_port_attr_get(dev, &attr);
 539         if (err && err != -EOPNOTSUPP)
 540                 return err;
 541
 542         return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode,
 543                                        attr.u.brport_flags, mask, nlflags,
 544                                        filter_mask, switchdev_port_vlan_fill);
 545 }
 546 EXPORT_SYMBOL_GPL(switchdev_port_bridge_getlink);
 547
 548 static int switchdev_port_br_setflag(struct net_device *dev,
 549                                      struct nlattr *nlattr,
 550                                      unsigned long brport_flag)
 551 {
 552         struct switchdev_attr attr = {
 553                 .id = SWITCHDEV_ATTR_PORT_BRIDGE_FLAGS,
 554         };
 555         u8 flag = nla_get_u8(nlattr);
 556         int err;
 557
 558         err = switchdev_port_attr_get(dev, &attr);
 559         if (err)
 560                 return err;
 561
 562         if (flag)
 563                 attr.u.brport_flags |= brport_flag;
 564         else
 565                 attr.u.brport_flags &= ~brport_flag;
 566
 567         return switchdev_port_attr_set(dev, &attr);
 568 }
 569
 570 static const struct nla_policy
 571 switchdev_port_bridge_policy[IFLA_BRPORT_MAX + 1] = {
 572         [IFLA_BRPORT_STATE]             = { .type = NLA_U8 },
 573         [IFLA_BRPORT_COST]              = { .type = NLA_U32 },
 574         [IFLA_BRPORT_PRIORITY]          = { .type = NLA_U16 },
 575         [IFLA_BRPORT_MODE]              = { .type = NLA_U8 },
 576         [IFLA_BRPORT_GUARD]             = { .type = NLA_U8 },
 577         [IFLA_BRPORT_PROTECT]           = { .type = NLA_U8 },
 578         [IFLA_BRPORT_FAST_LEAVE]        = { .type = NLA_U8 },
 579         [IFLA_BRPORT_LEARNING]          = { .type = NLA_U8 },
 580         [IFLA_BRPORT_LEARNING_SYNC]     = { .type = NLA_U8 },
 581         [IFLA_BRPORT_UNICAST_FLOOD]     = { .type = NLA_U8 },
 582 };
 583
 584 static int switchdev_port_br_setlink_protinfo(struct net_device *dev,
 585                                               struct nlattr *protinfo)
 586 {
 587         struct nlattr *attr;
 588         int rem;
 589         int err;
 590
 591         err = nla_validate_nested(protinfo, IFLA_BRPORT_MAX,
 592                                   switchdev_port_bridge_policy);
 593         if (err)
 594                 return err;
 595
 596         nla_for_each_nested(attr, protinfo, rem) {
 597                 switch (nla_type(attr)) {
 598                 case IFLA_BRPORT_LEARNING:
 599                         err = switchdev_port_br_setflag(dev, attr,
 600                                                         BR_LEARNING);
 601                         break;
 602                 case IFLA_BRPORT_LEARNING_SYNC:
 603                         err = switchdev_port_br_setflag(dev, attr,
 604                                                         BR_LEARNING_SYNC);
 605                         break;
 606                 default:
 607                         err = -EOPNOTSUPP;
 608                         break;
 609                 }
 610                 if (err)
 611                         return err;
 612         }
 613
 614         return 0;
 615 }
 616
 617 static int switchdev_port_br_afspec(struct net_device *dev,
 618                                     struct nlattr *afspec,
 619                                     int (*f)(struct net_device *dev,
 620                                              struct switchdev_obj *obj))
 621 {
 622         struct nlattr *attr;
 623         struct bridge_vlan_info *vinfo;
 624         struct switchdev_obj obj = {
 625                 .id = SWITCHDEV_OBJ_PORT_VLAN,
 626         };
 627         struct switchdev_obj_vlan *vlan = &obj.u.vlan;
 628         int rem;
 629         int err;
 630
 631         nla_for_each_nested(attr, afspec, rem) {
 632                 if (nla_type(attr) != IFLA_BRIDGE_VLAN_INFO)
 633                         continue;
 634                 if (nla_len(attr) != sizeof(struct bridge_vlan_info))
 635                         return -EINVAL;
 636                 vinfo = nla_data(attr);
 637                 vlan->flags = vinfo->flags;
 638                 if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_BEGIN) {
 639                         if (vlan->vid_begin)
 640                                 return -EINVAL;
 641                         vlan->vid_begin = vinfo->vid;
 642                 } else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
 643                         if (!vlan->vid_begin)
 644                                 return -EINVAL;
 645                         vlan->vid_end = vinfo->vid;
 646                         if (vlan->vid_end <= vlan->vid_begin)
 647                                 return -EINVAL;
 648                         err = f(dev, &obj);
 649                         if (err)
 650                                 return err;
 651                         memset(vlan, 0, sizeof(*vlan));
 652                 } else {
 653                         if (vlan->vid_begin)
 654                                 return -EINVAL;
 655                         vlan->vid_begin = vinfo->vid;
 656                         vlan->vid_end = vinfo->vid;
 657                         err = f(dev, &obj);
 658                         if (err)
 659                                 return err;
 660                         memset(vlan, 0, sizeof(*vlan));
 661                 }
 662         }
 663
 664         return 0;
 665 }
 666
 667 /**
 668  *      switchdev_port_bridge_setlink - Set bridge port attributes
 669  *
 670  *      @dev: port device
 671  *      @nlh: netlink header
 672  *      @flags: netlink flags
 673  *
 674  *      Called for SELF on rtnl_bridge_setlink to set bridge port
 675  *      attributes.
 676  */
 677 int switchdev_port_bridge_setlink(struct net_device *dev,
 678                                   struct nlmsghdr *nlh, u16 flags)
 679 {
 680         struct nlattr *protinfo;
 681         struct nlattr *afspec;
 682         int err = 0;
 683
 684         protinfo = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 685                                    IFLA_PROTINFO);
 686         if (protinfo) {
 687                 err = switchdev_port_br_setlink_protinfo(dev, protinfo);
 688                 if (err)
 689                         return err;
 690         }
 691
 692         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 693                                  IFLA_AF_SPEC);
 694         if (afspec)
 695                 err = switchdev_port_br_afspec(dev, afspec,
 696                                                switchdev_port_obj_add);
 697
 698         return err;
 699 }
 700 EXPORT_SYMBOL_GPL(switchdev_port_bridge_setlink);
 701
 702 /**
 703  *      switchdev_port_bridge_dellink - Set bridge port attributes
 704  *
 705  *      @dev: port device
 706  *      @nlh: netlink header
 707  *      @flags: netlink flags
 708  *
 709  *      Called for SELF on rtnl_bridge_dellink to set bridge port
 710  *      attributes.
 711  */
 712 int switchdev_port_bridge_dellink(struct net_device *dev,
 713                                   struct nlmsghdr *nlh, u16 flags)
 714 {
 715         struct nlattr *afspec;
 716
 717         afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
 718                                  IFLA_AF_SPEC);
 719         if (afspec)
 720                 return switchdev_port_br_afspec(dev, afspec,
 721                                                 switchdev_port_obj_del);
 722
 723         return 0;
 724 }
 725 EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
 726
 727 /**
 728  *      switchdev_port_fdb_add - Add FDB (MAC/VLAN) entry to port
 729  *
 730  *      @ndmsg: netlink hdr
 731  *      @nlattr: netlink attributes
 732  *      @dev: port device
 733  *      @addr: MAC address to add
 734  *      @vid: VLAN to add
 735  *
 736  *      Add FDB entry to switch device.
 737  */
 738 int switchdev_port_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 739                            struct net_device *dev, const unsigned char *addr,
 740                            u16 vid, u16 nlm_flags)
 741 {
 742         struct switchdev_obj obj = {
 743                 .id = SWITCHDEV_OBJ_PORT_FDB,
 744                 .u.fdb = {
 745                         .addr = addr,
 746                         .vid = vid,
 747                 },
 748         };
 749
 750         return switchdev_port_obj_add(dev, &obj);
 751 }
 752 EXPORT_SYMBOL_GPL(switchdev_port_fdb_add);
 753
 754 /**
 755  *      switchdev_port_fdb_del - Delete FDB (MAC/VLAN) entry from port
 756  *
 757  *      @ndmsg: netlink hdr
 758  *      @nlattr: netlink attributes
 759  *      @dev: port device
 760  *      @addr: MAC address to delete
 761  *      @vid: VLAN to delete
 762  *
 763  *      Delete FDB entry from switch device.
 764  */
 765 int switchdev_port_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 766                            struct net_device *dev, const unsigned char *addr,
 767                            u16 vid)
 768 {
 769         struct switchdev_obj obj = {
 770                 .id = SWITCHDEV_OBJ_PORT_FDB,
 771                 .u.fdb = {
 772                         .addr = addr,
 773                         .vid = vid,
 774                 },
 775         };
 776
 777         return switchdev_port_obj_del(dev, &obj);
 778 }
 779 EXPORT_SYMBOL_GPL(switchdev_port_fdb_del);
 780
 781 struct switchdev_fdb_dump {
 782         struct switchdev_obj obj;
 783         struct sk_buff *skb;
 784         struct netlink_callback *cb;
 785         int idx;
 786 };
 787
 788 static int switchdev_port_fdb_dump_cb(struct net_device *dev,
 789                                       struct switchdev_obj *obj)
 790 {
 791         struct switchdev_fdb_dump *dump =
 792                 container_of(obj, struct switchdev_fdb_dump, obj);
 793         u32 portid = NETLINK_CB(dump->cb->skb).portid;
 794         u32 seq = dump->cb->nlh->nlmsg_seq;
 795         struct nlmsghdr *nlh;
 796         struct ndmsg *ndm;
 797
 798         if (dump->idx < dump->cb->args[0])
 799                 goto skip;
 800
 801         nlh = nlmsg_put(dump->skb, portid, seq, RTM_NEWNEIGH,
 802                         sizeof(*ndm), NLM_F_MULTI);
 803         if (!nlh)
 804                 return -EMSGSIZE;
 805
 806         ndm = nlmsg_data(nlh);
 807         ndm->ndm_family  = AF_BRIDGE;
 808         ndm->ndm_pad1    = 0;
 809         ndm->ndm_pad2    = 0;
 810         ndm->ndm_flags   = NTF_SELF;
 811         ndm->ndm_type    = 0;
 812         ndm->ndm_ifindex = dev->ifindex;
 813         ndm->ndm_state   = obj->u.fdb.ndm_state;
 814
 815         if (nla_put(dump->skb, NDA_LLADDR, ETH_ALEN, obj->u.fdb.addr))
 816                 goto nla_put_failure;
 817
 818         if (obj->u.fdb.vid && nla_put_u16(dump->skb, NDA_VLAN, obj->u.fdb.vid))
 819                 goto nla_put_failure;
 820
 821         nlmsg_end(dump->skb, nlh);
 822
 823 skip:
 824         dump->idx++;
 825         return 0;
 826
 827 nla_put_failure:
 828         nlmsg_cancel(dump->skb, nlh);
 829         return -EMSGSIZE;
 830 }
 831
 832 /**
 833  *      switchdev_port_fdb_dump - Dump port FDB (MAC/VLAN) entries
 834  *
 835  *      @skb: netlink skb
 836  *      @cb: netlink callback
 837  *      @dev: port device
 838  *      @filter_dev: filter device
 839  *      @idx:
 840  *
 841  *      Delete FDB entry from switch device.
 842  */
 843 int switchdev_port_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 844                             struct net_device *dev,
 845                             struct net_device *filter_dev, int idx)
 846 {
 847         struct switchdev_fdb_dump dump = {
 848                 .obj = {
 849                         .id = SWITCHDEV_OBJ_PORT_FDB,
 850                         .cb = switchdev_port_fdb_dump_cb,
 851                 },
 852                 .skb = skb,
 853                 .cb = cb,
 854                 .idx = idx,
 855         };
 856
 857         switchdev_port_obj_dump(dev, &dump.obj);
 858         return dump.idx;
 859 }
 860 EXPORT_SYMBOL_GPL(switchdev_port_fdb_dump);
 861
 862 static struct net_device *switchdev_get_lowest_dev(struct net_device *dev)
 863 {
 864         const struct switchdev_ops *ops = dev->switchdev_ops;
 865         struct net_device *lower_dev;
 866         struct net_device *port_dev;
 867         struct list_head *iter;
 868
 869         /* Recusively search down until we find a sw port dev.
 870          * (A sw port dev supports switchdev_port_attr_get).
 871          */
 872
 873         if (ops && ops->switchdev_port_attr_get)
 874                 return dev;
 875
 876         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 877                 port_dev = switchdev_get_lowest_dev(lower_dev);
 878                 if (port_dev)
 879                         return port_dev;
 880         }
 881
 882         return NULL;
 883 }
 884
 885 static struct net_device *switchdev_get_dev_by_nhs(struct fib_info *fi)
 886 {
 887         struct switchdev_attr attr = {
 888                 .id = SWITCHDEV_ATTR_PORT_PARENT_ID,
 889         };
 890         struct switchdev_attr prev_attr;
 891         struct net_device *dev = NULL;
 892         int nhsel;
 893
 894         /* For this route, all nexthop devs must be on the same switch. */
 895
 896         for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
 897                 const struct fib_nh *nh = &fi->fib_nh[nhsel];
 898
 899                 if (!nh->nh_dev)
 900                         return NULL;
 901
 902                 dev = switchdev_get_lowest_dev(nh->nh_dev);
 903                 if (!dev)
 904                         return NULL;
 905
 906                 if (switchdev_port_attr_get(dev, &attr))
 907                         return NULL;
 908
 909                 if (nhsel > 0 &&
 910                     !netdev_phys_item_id_same(&prev_attr.u.ppid, &attr.u.ppid))
 911                                 return NULL;
 912
 913                 prev_attr = attr;
 914         }
 915
 916         return dev;
 917 }
 918
 919 /**
 920  *      switchdev_fib_ipv4_add - Add/modify switch IPv4 route entry
 921  *
 922  *      @dst: route's IPv4 destination address
 923  *      @dst_len: destination address length (prefix length)
 924  *      @fi: route FIB info structure
 925  *      @tos: route TOS
 926  *      @type: route type
 927  *      @nlflags: netlink flags passed in (NLM_F_*)
 928  *      @tb_id: route table ID
 929  *
 930  *      Add/modify switch IPv4 route entry.
 931  */
 932 int switchdev_fib_ipv4_add(u32 dst, int dst_len, struct fib_info *fi,
 933                            u8 tos, u8 type, u32 nlflags, u32 tb_id)
 934 {
 935         struct switchdev_obj fib_obj = {
 936                 .id = SWITCHDEV_OBJ_IPV4_FIB,
 937                 .u.ipv4_fib = {
 938                         .dst = dst,
 939                         .dst_len = dst_len,
 940                         .fi = fi,
 941                         .tos = tos,
 942                         .type = type,
 943                         .nlflags = nlflags,
 944                         .tb_id = tb_id,
 945                 },
 946         };
 947         struct net_device *dev;
 948         int err = 0;
 949
 950         /* Don't offload route if using custom ip rules or if
 951          * IPv4 FIB offloading has been disabled completely.
 952          */
 953
 954 #ifdef CONFIG_IP_MULTIPLE_TABLES
 955         if (fi->fib_net->ipv4.fib_has_custom_rules)
 956                 return 0;
 957 #endif
 958
 959         if (fi->fib_net->ipv4.fib_offload_disabled)
 960                 return 0;
 961
 962         dev = switchdev_get_dev_by_nhs(fi);
 963         if (!dev)
 964                 return 0;
 965
 966         err = switchdev_port_obj_add(dev, &fib_obj);
 967         if (!err)
 968                 fi->fib_flags |= RTNH_F_OFFLOAD;
 969
 970         return err == -EOPNOTSUPP ? 0 : err;
 971 }
 972 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_add);
 973
 974 /**
 975  *      switchdev_fib_ipv4_del - Delete IPv4 route entry from switch
 976  *
 977  *      @dst: route's IPv4 destination address
 978  *      @dst_len: destination address length (prefix length)
 979  *      @fi: route FIB info structure
 980  *      @tos: route TOS
 981  *      @type: route type
 982  *      @tb_id: route table ID
 983  *
 984  *      Delete IPv4 route entry from switch device.
 985  */
 986 int switchdev_fib_ipv4_del(u32 dst, int dst_len, struct fib_info *fi,
 987                            u8 tos, u8 type, u32 tb_id)
 988 {
 989         struct switchdev_obj fib_obj = {
 990                 .id = SWITCHDEV_OBJ_IPV4_FIB,
 991                 .u.ipv4_fib = {
 992                         .dst = dst,
 993                         .dst_len = dst_len,
 994                         .fi = fi,
 995                         .tos = tos,
 996                         .type = type,
 997                         .nlflags = 0,
 998                         .tb_id = tb_id,
 999                 },
1000         };
1001         struct net_device *dev;
1002         int err = 0;
1003
1004         if (!(fi->fib_flags & RTNH_F_OFFLOAD))
1005                 return 0;
1006
1007         dev = switchdev_get_dev_by_nhs(fi);
1008         if (!dev)
1009                 return 0;
1010
1011         err = switchdev_port_obj_del(dev, &fib_obj);
1012         if (!err)
1013                 fi->fib_flags &= ~RTNH_F_OFFLOAD;
1014
1015         return err == -EOPNOTSUPP ? 0 : err;
1016 }
1017 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_del);
1018
1019 /**
1020  *      switchdev_fib_ipv4_abort - Abort an IPv4 FIB operation
1021  *
1022  *      @fi: route FIB info structure
1023  */
1024 void switchdev_fib_ipv4_abort(struct fib_info *fi)
1025 {
1026         /* There was a problem installing this route to the offload
1027          * device.  For now, until we come up with more refined
1028          * policy handling, abruptly end IPv4 fib offloading for
1029          * for entire net by flushing offload device(s) of all
1030          * IPv4 routes, and mark IPv4 fib offloading broken from
1031          * this point forward.
1032          */
1033
1034         fib_flush_external(fi->fib_net);
1035         fi->fib_net->ipv4.fib_offload_disabled = true;
1036 }
1037 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_abort);
1038
1039 static bool switchdev_port_same_parent_id(struct net_device *a,
1040                                           struct net_device *b)
1041 {
1042         struct switchdev_attr a_attr = {
1043                 .id = SWITCHDEV_ATTR_PORT_PARENT_ID,
1044                 .flags = SWITCHDEV_F_NO_RECURSE,
1045         };
1046         struct switchdev_attr b_attr = {
1047                 .id = SWITCHDEV_ATTR_PORT_PARENT_ID,
1048                 .flags = SWITCHDEV_F_NO_RECURSE,
1049         };
1050
1051         if (switchdev_port_attr_get(a, &a_attr) ||
1052             switchdev_port_attr_get(b, &b_attr))
1053                 return false;
1054
1055         return netdev_phys_item_id_same(&a_attr.u.ppid, &b_attr.u.ppid);
1056 }
1057
1058 static u32 switchdev_port_fwd_mark_get(struct net_device *dev,
1059                                        struct net_device *group_dev)
1060 {
1061         struct net_device *lower_dev;
1062         struct list_head *iter;
1063
1064         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1065                 if (lower_dev == dev)
1066                         continue;
1067                 if (switchdev_port_same_parent_id(dev, lower_dev))
1068                         return lower_dev->offload_fwd_mark;
1069                 return switchdev_port_fwd_mark_get(dev, lower_dev);
1070         }
1071
1072         return dev->ifindex;
1073 }
1074
1075 static void switchdev_port_fwd_mark_reset(struct net_device *group_dev,
1076                                           u32 old_mark, u32 *reset_mark)
1077 {
1078         struct net_device *lower_dev;
1079         struct list_head *iter;
1080
1081         netdev_for_each_lower_dev(group_dev, lower_dev, iter) {
1082                 if (lower_dev->offload_fwd_mark == old_mark) {
1083                         if (!*reset_mark)
1084                                 *reset_mark = lower_dev->ifindex;
1085                         lower_dev->offload_fwd_mark = *reset_mark;
1086                 }
1087                 switchdev_port_fwd_mark_reset(lower_dev, old_mark, reset_mark);
1088         }
1089 }
1090
1091 /**
1092  *      switchdev_port_fwd_mark_set - Set port offload forwarding mark
1093  *
1094  *      @dev: port device
1095  *      @group_dev: containing device
1096  *      @joining: true if dev is joining group; false if leaving group
1097  *
1098  *      An ungrouped port's offload mark is just its ifindex.  A grouped
1099  *      port's (member of a bridge, for example) offload mark is the ifindex
1100  *      of one of the ports in the group with the same parent (switch) ID.
1101  *      Ports on the same device in the same group will have the same mark.
1102  *
1103  *      Example:
1104  *
1105  *              br0             ifindex=9
1106  *                sw1p1         ifindex=2       mark=2
1107  *                sw1p2         ifindex=3       mark=2
1108  *                sw2p1         ifindex=4       mark=5
1109  *                sw2p2         ifindex=5       mark=5
1110  *
1111  *      If sw2p2 leaves the bridge, we'll have:
1112  *
1113  *              br0             ifindex=9
1114  *                sw1p1         ifindex=2       mark=2
1115  *                sw1p2         ifindex=3       mark=2
1116  *                sw2p1         ifindex=4       mark=4
1117  *              sw2p2           ifindex=5       mark=5
1118  */
1119 void switchdev_port_fwd_mark_set(struct net_device *dev,
1120                                  struct net_device *group_dev,
1121                                  bool joining)
1122 {
1123         u32 mark = dev->ifindex;
1124         u32 reset_mark = 0;
1125
1126         if (group_dev && joining) {
1127                 mark = switchdev_port_fwd_mark_get(dev, group_dev);
1128         } else if (group_dev && !joining) {
1129                 if (dev->offload_fwd_mark == mark)
1130                         /* Ohoh, this port was the mark reference port,
1131                          * but it's leaving the group, so reset the
1132                          * mark for the remaining ports in the group.
1133                          */
1134                         switchdev_port_fwd_mark_reset(group_dev, mark,
1135                                                       &reset_mark);
1136         }
1137
1138         dev->offload_fwd_mark = mark;
1139 }
1140 EXPORT_SYMBOL_GPL(switchdev_port_fwd_mark_set);