net/switchdev/switchdev.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/switchdev/switchdev.c - Switch device API
   4  * Copyright (c) 2014-2015 Jiri Pirko <jiri@resnulli.us>
   5  * Copyright (c) 2014-2015 Scott Feldman <sfeldma@gmail.com>
   6  */
   7
   8 #include <linux/kernel.h>
   9 #include <linux/types.h>
  10 #include <linux/init.h>
  11 #include <linux/mutex.h>
  12 #include <linux/notifier.h>
  13 #include <linux/netdevice.h>
  14 #include <linux/etherdevice.h>
  15 #include <linux/if_bridge.h>
  16 #include <linux/list.h>
  17 #include <linux/workqueue.h>
  18 #include <linux/if_vlan.h>
  19 #include <linux/rtnetlink.h>
  20 #include <net/switchdev.h>
  21
  22 static LIST_HEAD(deferred);
  23 static DEFINE_SPINLOCK(deferred_lock);
  24
  25 typedef void switchdev_deferred_func_t(struct net_device *dev,
  26                                        const void *data);
  27
  28 struct switchdev_deferred_item {
  29         struct list_head list;
  30         struct net_device *dev;
  31         switchdev_deferred_func_t *func;
  32         unsigned long data[];
  33 };
  34
  35 static struct switchdev_deferred_item *switchdev_deferred_dequeue(void)
  36 {
  37         struct switchdev_deferred_item *dfitem;
  38
  39         spin_lock_bh(&deferred_lock);
  40         if (list_empty(&deferred)) {
  41                 dfitem = NULL;
  42                 goto unlock;
  43         }
  44         dfitem = list_first_entry(&deferred,
  45                                   struct switchdev_deferred_item, list);
  46         list_del(&dfitem->list);
  47 unlock:
  48         spin_unlock_bh(&deferred_lock);
  49         return dfitem;
  50 }
  51
  52 /**
  53  *      switchdev_deferred_process - Process ops in deferred queue
  54  *
  55  *      Called to flush the ops currently queued in deferred ops queue.
  56  *      rtnl_lock must be held.
  57  */
  58 void switchdev_deferred_process(void)
  59 {
  60         struct switchdev_deferred_item *dfitem;
  61
  62         ASSERT_RTNL();
  63
  64         while ((dfitem = switchdev_deferred_dequeue())) {
  65                 dfitem->func(dfitem->dev, dfitem->data);
  66                 dev_put(dfitem->dev);
  67                 kfree(dfitem);
  68         }
  69 }
  70 EXPORT_SYMBOL_GPL(switchdev_deferred_process);
  71
  72 static void switchdev_deferred_process_work(struct work_struct *work)
  73 {
  74         rtnl_lock();
  75         switchdev_deferred_process();
  76         rtnl_unlock();
  77 }
  78
  79 static DECLARE_WORK(deferred_process_work, switchdev_deferred_process_work);
  80
  81 static int switchdev_deferred_enqueue(struct net_device *dev,
  82                                       const void *data, size_t data_len,
  83                                       switchdev_deferred_func_t *func)
  84 {
  85         struct switchdev_deferred_item *dfitem;
  86
  87         dfitem = kmalloc(sizeof(*dfitem) + data_len, GFP_ATOMIC);
  88         if (!dfitem)
  89                 return -ENOMEM;
  90         dfitem->dev = dev;
  91         dfitem->func = func;
  92         memcpy(dfitem->data, data, data_len);
  93         dev_hold(dev);
  94         spin_lock_bh(&deferred_lock);
  95         list_add_tail(&dfitem->list, &deferred);
  96         spin_unlock_bh(&deferred_lock);
  97         schedule_work(&deferred_process_work);
  98         return 0;
  99 }
 100
 101 static int switchdev_port_attr_notify(enum switchdev_notifier_type nt,
 102                                       struct net_device *dev,
 103                                       const struct switchdev_attr *attr,
 104                                       struct switchdev_trans *trans)
 105 {
 106         int err;
 107         int rc;
 108
 109         struct switchdev_notifier_port_attr_info attr_info = {
 110                 .attr = attr,
 111                 .trans = trans,
 112                 .handled = false,
 113         };
 114
 115         rc = call_switchdev_blocking_notifiers(nt, dev,
 116                                                &attr_info.info, NULL);
 117         err = notifier_to_errno(rc);
 118         if (err) {
 119                 WARN_ON(!attr_info.handled);
 120                 return err;
 121         }
 122
 123         if (!attr_info.handled)
 124                 return -EOPNOTSUPP;
 125
 126         return 0;
 127 }
 128
 129 static int switchdev_port_attr_set_now(struct net_device *dev,
 130                                        const struct switchdev_attr *attr)
 131 {
 132         struct switchdev_trans trans;
 133         int err;
 134
 135         /* Phase I: prepare for attr set. Driver/device should fail
 136          * here if there are going to be issues in the commit phase,
 137          * such as lack of resources or support.  The driver/device
 138          * should reserve resources needed for the commit phase here,
 139          * but should not commit the attr.
 140          */
 141
 142         trans.ph_prepare = true;
 143         err = switchdev_port_attr_notify(SWITCHDEV_PORT_ATTR_SET, dev, attr,
 144                                          &trans);
 145         if (err)
 146                 return err;
 147
 148         /* Phase II: commit attr set.  This cannot fail as a fault
 149          * of driver/device.  If it does, it's a bug in the driver/device
 150          * because the driver said everythings was OK in phase I.
 151          */
 152
 153         trans.ph_prepare = false;
 154         err = switchdev_port_attr_notify(SWITCHDEV_PORT_ATTR_SET, dev, attr,
 155                                          &trans);
 156         WARN(err, "%s: Commit of attribute (id=%d) failed.\n",
 157              dev->name, attr->id);
 158
 159         return err;
 160 }
 161
 162 static void switchdev_port_attr_set_deferred(struct net_device *dev,
 163                                              const void *data)
 164 {
 165         const struct switchdev_attr *attr = data;
 166         int err;
 167
 168         err = switchdev_port_attr_set_now(dev, attr);
 169         if (err && err != -EOPNOTSUPP)
 170                 netdev_err(dev, "failed (err=%d) to set attribute (id=%d)\n",
 171                            err, attr->id);
 172         if (attr->complete)
 173                 attr->complete(dev, err, attr->complete_priv);
 174 }
 175
 176 static int switchdev_port_attr_set_defer(struct net_device *dev,
 177                                          const struct switchdev_attr *attr)
 178 {
 179         return switchdev_deferred_enqueue(dev, attr, sizeof(*attr),
 180                                           switchdev_port_attr_set_deferred);
 181 }
 182
 183 /**
 184  *      switchdev_port_attr_set - Set port attribute
 185  *
 186  *      @dev: port device
 187  *      @attr: attribute to set
 188  *
 189  *      Use a 2-phase prepare-commit transaction model to ensure
 190  *      system is not left in a partially updated state due to
 191  *      failure from driver/device.
 192  *
 193  *      rtnl_lock must be held and must not be in atomic section,
 194  *      in case SWITCHDEV_F_DEFER flag is not set.
 195  */
 196 int switchdev_port_attr_set(struct net_device *dev,
 197                             const struct switchdev_attr *attr)
 198 {
 199         if (attr->flags & SWITCHDEV_F_DEFER)
 200                 return switchdev_port_attr_set_defer(dev, attr);
 201         ASSERT_RTNL();
 202         return switchdev_port_attr_set_now(dev, attr);
 203 }
 204 EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
 205
 206 static size_t switchdev_obj_size(const struct switchdev_obj *obj)
 207 {
 208         switch (obj->id) {
 209         case SWITCHDEV_OBJ_ID_PORT_VLAN:
 210                 return sizeof(struct switchdev_obj_port_vlan);
 211         case SWITCHDEV_OBJ_ID_PORT_MDB:
 212                 return sizeof(struct switchdev_obj_port_mdb);
 213         case SWITCHDEV_OBJ_ID_HOST_MDB:
 214                 return sizeof(struct switchdev_obj_port_mdb);
 215         default:
 216                 BUG();
 217         }
 218         return 0;
 219 }
 220
 221 static int switchdev_port_obj_notify(enum switchdev_notifier_type nt,
 222                                      struct net_device *dev,
 223                                      const struct switchdev_obj *obj,
 224                                      struct switchdev_trans *trans,
 225                                      struct netlink_ext_ack *extack)
 226 {
 227         int rc;
 228         int err;
 229
 230         struct switchdev_notifier_port_obj_info obj_info = {
 231                 .obj = obj,
 232                 .trans = trans,
 233                 .handled = false,
 234         };
 235
 236         rc = call_switchdev_blocking_notifiers(nt, dev, &obj_info.info, extack);
 237         err = notifier_to_errno(rc);
 238         if (err) {
 239                 WARN_ON(!obj_info.handled);
 240                 return err;
 241         }
 242         if (!obj_info.handled)
 243                 return -EOPNOTSUPP;
 244         return 0;
 245 }
 246
 247 static int switchdev_port_obj_add_now(struct net_device *dev,
 248                                       const struct switchdev_obj *obj,
 249                                       struct netlink_ext_ack *extack)
 250 {
 251         struct switchdev_trans trans;
 252         int err;
 253
 254         ASSERT_RTNL();
 255
 256         /* Phase I: prepare for obj add. Driver/device should fail
 257          * here if there are going to be issues in the commit phase,
 258          * such as lack of resources or support.  The driver/device
 259          * should reserve resources needed for the commit phase here,
 260          * but should not commit the obj.
 261          */
 262
 263         trans.ph_prepare = true;
 264         err = switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD,
 265                                         dev, obj, &trans, extack);
 266         if (err)
 267                 return err;
 268
 269         /* Phase II: commit obj add.  This cannot fail as a fault
 270          * of driver/device.  If it does, it's a bug in the driver/device
 271          * because the driver said everythings was OK in phase I.
 272          */
 273
 274         trans.ph_prepare = false;
 275         err = switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD,
 276                                         dev, obj, &trans, extack);
 277         WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->id);
 278
 279         return err;
 280 }
 281
 282 static void switchdev_port_obj_add_deferred(struct net_device *dev,
 283                                             const void *data)
 284 {
 285         const struct switchdev_obj *obj = data;
 286         int err;
 287
 288         err = switchdev_port_obj_add_now(dev, obj, NULL);
 289         if (err && err != -EOPNOTSUPP)
 290                 netdev_err(dev, "failed (err=%d) to add object (id=%d)\n",
 291                            err, obj->id);
 292         if (obj->complete)
 293                 obj->complete(dev, err, obj->complete_priv);
 294 }
 295
 296 static int switchdev_port_obj_add_defer(struct net_device *dev,
 297                                         const struct switchdev_obj *obj)
 298 {
 299         return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
 300                                           switchdev_port_obj_add_deferred);
 301 }
 302
 303 /**
 304  *      switchdev_port_obj_add - Add port object
 305  *
 306  *      @dev: port device
 307  *      @obj: object to add
 308  *      @extack: netlink extended ack
 309  *
 310  *      Use a 2-phase prepare-commit transaction model to ensure
 311  *      system is not left in a partially updated state due to
 312  *      failure from driver/device.
 313  *
 314  *      rtnl_lock must be held and must not be in atomic section,
 315  *      in case SWITCHDEV_F_DEFER flag is not set.
 316  */
 317 int switchdev_port_obj_add(struct net_device *dev,
 318                            const struct switchdev_obj *obj,
 319                            struct netlink_ext_ack *extack)
 320 {
 321         if (obj->flags & SWITCHDEV_F_DEFER)
 322                 return switchdev_port_obj_add_defer(dev, obj);
 323         ASSERT_RTNL();
 324         return switchdev_port_obj_add_now(dev, obj, extack);
 325 }
 326 EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
 327
 328 static int switchdev_port_obj_del_now(struct net_device *dev,
 329                                       const struct switchdev_obj *obj)
 330 {
 331         return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_DEL,
 332                                          dev, obj, NULL, NULL);
 333 }
 334
 335 static void switchdev_port_obj_del_deferred(struct net_device *dev,
 336                                             const void *data)
 337 {
 338         const struct switchdev_obj *obj = data;
 339         int err;
 340
 341         err = switchdev_port_obj_del_now(dev, obj);
 342         if (err && err != -EOPNOTSUPP)
 343                 netdev_err(dev, "failed (err=%d) to del object (id=%d)\n",
 344                            err, obj->id);
 345         if (obj->complete)
 346                 obj->complete(dev, err, obj->complete_priv);
 347 }
 348
 349 static int switchdev_port_obj_del_defer(struct net_device *dev,
 350                                         const struct switchdev_obj *obj)
 351 {
 352         return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
 353                                           switchdev_port_obj_del_deferred);
 354 }
 355
 356 /**
 357  *      switchdev_port_obj_del - Delete port object
 358  *
 359  *      @dev: port device
 360  *      @obj: object to delete
 361  *
 362  *      rtnl_lock must be held and must not be in atomic section,
 363  *      in case SWITCHDEV_F_DEFER flag is not set.
 364  */
 365 int switchdev_port_obj_del(struct net_device *dev,
 366                            const struct switchdev_obj *obj)
 367 {
 368         if (obj->flags & SWITCHDEV_F_DEFER)
 369                 return switchdev_port_obj_del_defer(dev, obj);
 370         ASSERT_RTNL();
 371         return switchdev_port_obj_del_now(dev, obj);
 372 }
 373 EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
 374
 375 static ATOMIC_NOTIFIER_HEAD(switchdev_notif_chain);
 376 static BLOCKING_NOTIFIER_HEAD(switchdev_blocking_notif_chain);
 377
 378 /**
 379  *      register_switchdev_notifier - Register notifier
 380  *      @nb: notifier_block
 381  *
 382  *      Register switch device notifier.
 383  */
 384 int register_switchdev_notifier(struct notifier_block *nb)
 385 {
 386         return atomic_notifier_chain_register(&switchdev_notif_chain, nb);
 387 }
 388 EXPORT_SYMBOL_GPL(register_switchdev_notifier);
 389
 390 /**
 391  *      unregister_switchdev_notifier - Unregister notifier
 392  *      @nb: notifier_block
 393  *
 394  *      Unregister switch device notifier.
 395  */
 396 int unregister_switchdev_notifier(struct notifier_block *nb)
 397 {
 398         return atomic_notifier_chain_unregister(&switchdev_notif_chain, nb);
 399 }
 400 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
 401
 402 /**
 403  *      call_switchdev_notifiers - Call notifiers
 404  *      @val: value passed unmodified to notifier function
 405  *      @dev: port device
 406  *      @info: notifier information data
 407  *      @extack: netlink extended ack
 408  *      Call all network notifier blocks.
 409  */
 410 int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
 411                              struct switchdev_notifier_info *info,
 412                              struct netlink_ext_ack *extack)
 413 {
 414         info->dev = dev;
 415         info->extack = extack;
 416         return atomic_notifier_call_chain(&switchdev_notif_chain, val, info);
 417 }
 418 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
 419
 420 int register_switchdev_blocking_notifier(struct notifier_block *nb)
 421 {
 422         struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
 423
 424         return blocking_notifier_chain_register(chain, nb);
 425 }
 426 EXPORT_SYMBOL_GPL(register_switchdev_blocking_notifier);
 427
 428 int unregister_switchdev_blocking_notifier(struct notifier_block *nb)
 429 {
 430         struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
 431
 432         return blocking_notifier_chain_unregister(chain, nb);
 433 }
 434 EXPORT_SYMBOL_GPL(unregister_switchdev_blocking_notifier);
 435
 436 int call_switchdev_blocking_notifiers(unsigned long val, struct net_device *dev,
 437                                       struct switchdev_notifier_info *info,
 438                                       struct netlink_ext_ack *extack)
 439 {
 440         info->dev = dev;
 441         info->extack = extack;
 442         return blocking_notifier_call_chain(&switchdev_blocking_notif_chain,
 443                                             val, info);
 444 }
 445 EXPORT_SYMBOL_GPL(call_switchdev_blocking_notifiers);
 446
 447 static int __switchdev_handle_port_obj_add(struct net_device *dev,
 448                         struct switchdev_notifier_port_obj_info *port_obj_info,
 449                         bool (*check_cb)(const struct net_device *dev),
 450                         int (*add_cb)(struct net_device *dev,
 451                                       const struct switchdev_obj *obj,
 452                                       struct switchdev_trans *trans,
 453                                       struct netlink_ext_ack *extack))
 454 {
 455         struct netlink_ext_ack *extack;
 456         struct net_device *lower_dev;
 457         struct list_head *iter;
 458         int err = -EOPNOTSUPP;
 459
 460         extack = switchdev_notifier_info_to_extack(&port_obj_info->info);
 461
 462         if (check_cb(dev)) {
 463                 /* This flag is only checked if the return value is success. */
 464                 port_obj_info->handled = true;
 465                 return add_cb(dev, port_obj_info->obj, port_obj_info->trans,
 466                               extack);
 467         }
 468
 469         /* Switch ports might be stacked under e.g. a LAG. Ignore the
 470          * unsupported devices, another driver might be able to handle them. But
 471          * propagate to the callers any hard errors.
 472          *
 473          * If the driver does its own bookkeeping of stacked ports, it's not
 474          * necessary to go through this helper.
 475          */
 476         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 477                 if (netif_is_bridge_master(lower_dev))
 478                         continue;
 479
 480                 err = __switchdev_handle_port_obj_add(lower_dev, port_obj_info,
 481                                                       check_cb, add_cb);
 482                 if (err && err != -EOPNOTSUPP)
 483                         return err;
 484         }
 485
 486         return err;
 487 }
 488
 489 int switchdev_handle_port_obj_add(struct net_device *dev,
 490                         struct switchdev_notifier_port_obj_info *port_obj_info,
 491                         bool (*check_cb)(const struct net_device *dev),
 492                         int (*add_cb)(struct net_device *dev,
 493                                       const struct switchdev_obj *obj,
 494                                       struct switchdev_trans *trans,
 495                                       struct netlink_ext_ack *extack))
 496 {
 497         int err;
 498
 499         err = __switchdev_handle_port_obj_add(dev, port_obj_info, check_cb,
 500                                               add_cb);
 501         if (err == -EOPNOTSUPP)
 502                 err = 0;
 503         return err;
 504 }
 505 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_add);
 506
 507 static int __switchdev_handle_port_obj_del(struct net_device *dev,
 508                         struct switchdev_notifier_port_obj_info *port_obj_info,
 509                         bool (*check_cb)(const struct net_device *dev),
 510                         int (*del_cb)(struct net_device *dev,
 511                                       const struct switchdev_obj *obj))
 512 {
 513         struct net_device *lower_dev;
 514         struct list_head *iter;
 515         int err = -EOPNOTSUPP;
 516
 517         if (check_cb(dev)) {
 518                 /* This flag is only checked if the return value is success. */
 519                 port_obj_info->handled = true;
 520                 return del_cb(dev, port_obj_info->obj);
 521         }
 522
 523         /* Switch ports might be stacked under e.g. a LAG. Ignore the
 524          * unsupported devices, another driver might be able to handle them. But
 525          * propagate to the callers any hard errors.
 526          *
 527          * If the driver does its own bookkeeping of stacked ports, it's not
 528          * necessary to go through this helper.
 529          */
 530         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 531                 if (netif_is_bridge_master(lower_dev))
 532                         continue;
 533
 534                 err = __switchdev_handle_port_obj_del(lower_dev, port_obj_info,
 535                                                       check_cb, del_cb);
 536                 if (err && err != -EOPNOTSUPP)
 537                         return err;
 538         }
 539
 540         return err;
 541 }
 542
 543 int switchdev_handle_port_obj_del(struct net_device *dev,
 544                         struct switchdev_notifier_port_obj_info *port_obj_info,
 545                         bool (*check_cb)(const struct net_device *dev),
 546                         int (*del_cb)(struct net_device *dev,
 547                                       const struct switchdev_obj *obj))
 548 {
 549         int err;
 550
 551         err = __switchdev_handle_port_obj_del(dev, port_obj_info, check_cb,
 552                                               del_cb);
 553         if (err == -EOPNOTSUPP)
 554                 err = 0;
 555         return err;
 556 }
 557 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_del);
 558
 559 static int __switchdev_handle_port_attr_set(struct net_device *dev,
 560                         struct switchdev_notifier_port_attr_info *port_attr_info,
 561                         bool (*check_cb)(const struct net_device *dev),
 562                         int (*set_cb)(struct net_device *dev,
 563                                       const struct switchdev_attr *attr,
 564                                       struct switchdev_trans *trans))
 565 {
 566         struct net_device *lower_dev;
 567         struct list_head *iter;
 568         int err = -EOPNOTSUPP;
 569
 570         if (check_cb(dev)) {
 571                 port_attr_info->handled = true;
 572                 return set_cb(dev, port_attr_info->attr,
 573                               port_attr_info->trans);
 574         }
 575
 576         /* Switch ports might be stacked under e.g. a LAG. Ignore the
 577          * unsupported devices, another driver might be able to handle them. But
 578          * propagate to the callers any hard errors.
 579          *
 580          * If the driver does its own bookkeeping of stacked ports, it's not
 581          * necessary to go through this helper.
 582          */
 583         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 584                 if (netif_is_bridge_master(lower_dev))
 585                         continue;
 586
 587                 err = __switchdev_handle_port_attr_set(lower_dev, port_attr_info,
 588                                                        check_cb, set_cb);
 589                 if (err && err != -EOPNOTSUPP)
 590                         return err;
 591         }
 592
 593         return err;
 594 }
 595
 596 int switchdev_handle_port_attr_set(struct net_device *dev,
 597                         struct switchdev_notifier_port_attr_info *port_attr_info,
 598                         bool (*check_cb)(const struct net_device *dev),
 599                         int (*set_cb)(struct net_device *dev,
 600                                       const struct switchdev_attr *attr,
 601                                       struct switchdev_trans *trans))
 602 {
 603         int err;
 604
 605         err = __switchdev_handle_port_attr_set(dev, port_attr_info, check_cb,
 606                                                set_cb);
 607         if (err == -EOPNOTSUPP)
 608                 err = 0;
 609         return err;
 610 }
 611 EXPORT_SYMBOL_GPL(switchdev_handle_port_attr_set);