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 <[email protected]>
   5  * Copyright (c) 2014-2015 Scott Feldman <[email protected]>
   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 netlink_ext_ack *extack)
 105 {
 106         int err;
 107         int rc;
 108
 109         struct switchdev_notifier_port_attr_info attr_info = {
 110                 .attr = attr,
 111                 .handled = false,
 112         };
 113
 114         rc = call_switchdev_blocking_notifiers(nt, dev,
 115                                                &attr_info.info, extack);
 116         err = notifier_to_errno(rc);
 117         if (err) {
 118                 WARN_ON(!attr_info.handled);
 119                 return err;
 120         }
 121
 122         if (!attr_info.handled)
 123                 return -EOPNOTSUPP;
 124
 125         return 0;
 126 }
 127
 128 static int switchdev_port_attr_set_now(struct net_device *dev,
 129                                        const struct switchdev_attr *attr,
 130                                        struct netlink_ext_ack *extack)
 131 {
 132         return switchdev_port_attr_notify(SWITCHDEV_PORT_ATTR_SET, dev, attr,
 133                                           extack);
 134 }
 135
 136 static void switchdev_port_attr_set_deferred(struct net_device *dev,
 137                                              const void *data)
 138 {
 139         const struct switchdev_attr *attr = data;
 140         int err;
 141
 142         err = switchdev_port_attr_set_now(dev, attr, NULL);
 143         if (err && err != -EOPNOTSUPP)
 144                 netdev_err(dev, "failed (err=%d) to set attribute (id=%d)\n",
 145                            err, attr->id);
 146         if (attr->complete)
 147                 attr->complete(dev, err, attr->complete_priv);
 148 }
 149
 150 static int switchdev_port_attr_set_defer(struct net_device *dev,
 151                                          const struct switchdev_attr *attr)
 152 {
 153         return switchdev_deferred_enqueue(dev, attr, sizeof(*attr),
 154                                           switchdev_port_attr_set_deferred);
 155 }
 156
 157 /**
 158  *      switchdev_port_attr_set - Set port attribute
 159  *
 160  *      @dev: port device
 161  *      @attr: attribute to set
 162  *      @extack: netlink extended ack, for error message propagation
 163  *
 164  *      rtnl_lock must be held and must not be in atomic section,
 165  *      in case SWITCHDEV_F_DEFER flag is not set.
 166  */
 167 int switchdev_port_attr_set(struct net_device *dev,
 168                             const struct switchdev_attr *attr,
 169                             struct netlink_ext_ack *extack)
 170 {
 171         if (attr->flags & SWITCHDEV_F_DEFER)
 172                 return switchdev_port_attr_set_defer(dev, attr);
 173         ASSERT_RTNL();
 174         return switchdev_port_attr_set_now(dev, attr, extack);
 175 }
 176 EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
 177
 178 static size_t switchdev_obj_size(const struct switchdev_obj *obj)
 179 {
 180         switch (obj->id) {
 181         case SWITCHDEV_OBJ_ID_PORT_VLAN:
 182                 return sizeof(struct switchdev_obj_port_vlan);
 183         case SWITCHDEV_OBJ_ID_PORT_MDB:
 184                 return sizeof(struct switchdev_obj_port_mdb);
 185         case SWITCHDEV_OBJ_ID_HOST_MDB:
 186                 return sizeof(struct switchdev_obj_port_mdb);
 187         default:
 188                 BUG();
 189         }
 190         return 0;
 191 }
 192
 193 static int switchdev_port_obj_notify(enum switchdev_notifier_type nt,
 194                                      struct net_device *dev,
 195                                      const struct switchdev_obj *obj,
 196                                      struct netlink_ext_ack *extack)
 197 {
 198         int rc;
 199         int err;
 200
 201         struct switchdev_notifier_port_obj_info obj_info = {
 202                 .obj = obj,
 203                 .handled = false,
 204         };
 205
 206         rc = call_switchdev_blocking_notifiers(nt, dev, &obj_info.info, extack);
 207         err = notifier_to_errno(rc);
 208         if (err) {
 209                 WARN_ON(!obj_info.handled);
 210                 return err;
 211         }
 212         if (!obj_info.handled)
 213                 return -EOPNOTSUPP;
 214         return 0;
 215 }
 216
 217 static void switchdev_port_obj_add_deferred(struct net_device *dev,
 218                                             const void *data)
 219 {
 220         const struct switchdev_obj *obj = data;
 221         int err;
 222
 223         ASSERT_RTNL();
 224         err = switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD,
 225                                         dev, obj, NULL);
 226         if (err && err != -EOPNOTSUPP)
 227                 netdev_err(dev, "failed (err=%d) to add object (id=%d)\n",
 228                            err, obj->id);
 229         if (obj->complete)
 230                 obj->complete(dev, err, obj->complete_priv);
 231 }
 232
 233 static int switchdev_port_obj_add_defer(struct net_device *dev,
 234                                         const struct switchdev_obj *obj)
 235 {
 236         return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
 237                                           switchdev_port_obj_add_deferred);
 238 }
 239
 240 /**
 241  *      switchdev_port_obj_add - Add port object
 242  *
 243  *      @dev: port device
 244  *      @obj: object to add
 245  *      @extack: netlink extended ack
 246  *
 247  *      rtnl_lock must be held and must not be in atomic section,
 248  *      in case SWITCHDEV_F_DEFER flag is not set.
 249  */
 250 int switchdev_port_obj_add(struct net_device *dev,
 251                            const struct switchdev_obj *obj,
 252                            struct netlink_ext_ack *extack)
 253 {
 254         if (obj->flags & SWITCHDEV_F_DEFER)
 255                 return switchdev_port_obj_add_defer(dev, obj);
 256         ASSERT_RTNL();
 257         return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD,
 258                                          dev, obj, extack);
 259 }
 260 EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
 261
 262 static int switchdev_port_obj_del_now(struct net_device *dev,
 263                                       const struct switchdev_obj *obj)
 264 {
 265         return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_DEL,
 266                                          dev, obj, NULL);
 267 }
 268
 269 static void switchdev_port_obj_del_deferred(struct net_device *dev,
 270                                             const void *data)
 271 {
 272         const struct switchdev_obj *obj = data;
 273         int err;
 274
 275         err = switchdev_port_obj_del_now(dev, obj);
 276         if (err && err != -EOPNOTSUPP)
 277                 netdev_err(dev, "failed (err=%d) to del object (id=%d)\n",
 278                            err, obj->id);
 279         if (obj->complete)
 280                 obj->complete(dev, err, obj->complete_priv);
 281 }
 282
 283 static int switchdev_port_obj_del_defer(struct net_device *dev,
 284                                         const struct switchdev_obj *obj)
 285 {
 286         return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj),
 287                                           switchdev_port_obj_del_deferred);
 288 }
 289
 290 /**
 291  *      switchdev_port_obj_del - Delete port object
 292  *
 293  *      @dev: port device
 294  *      @obj: object to delete
 295  *
 296  *      rtnl_lock must be held and must not be in atomic section,
 297  *      in case SWITCHDEV_F_DEFER flag is not set.
 298  */
 299 int switchdev_port_obj_del(struct net_device *dev,
 300                            const struct switchdev_obj *obj)
 301 {
 302         if (obj->flags & SWITCHDEV_F_DEFER)
 303                 return switchdev_port_obj_del_defer(dev, obj);
 304         ASSERT_RTNL();
 305         return switchdev_port_obj_del_now(dev, obj);
 306 }
 307 EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
 308
 309 static ATOMIC_NOTIFIER_HEAD(switchdev_notif_chain);
 310 static BLOCKING_NOTIFIER_HEAD(switchdev_blocking_notif_chain);
 311
 312 /**
 313  *      register_switchdev_notifier - Register notifier
 314  *      @nb: notifier_block
 315  *
 316  *      Register switch device notifier.
 317  */
 318 int register_switchdev_notifier(struct notifier_block *nb)
 319 {
 320         return atomic_notifier_chain_register(&switchdev_notif_chain, nb);
 321 }
 322 EXPORT_SYMBOL_GPL(register_switchdev_notifier);
 323
 324 /**
 325  *      unregister_switchdev_notifier - Unregister notifier
 326  *      @nb: notifier_block
 327  *
 328  *      Unregister switch device notifier.
 329  */
 330 int unregister_switchdev_notifier(struct notifier_block *nb)
 331 {
 332         return atomic_notifier_chain_unregister(&switchdev_notif_chain, nb);
 333 }
 334 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
 335
 336 /**
 337  *      call_switchdev_notifiers - Call notifiers
 338  *      @val: value passed unmodified to notifier function
 339  *      @dev: port device
 340  *      @info: notifier information data
 341  *      @extack: netlink extended ack
 342  *      Call all network notifier blocks.
 343  */
 344 int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
 345                              struct switchdev_notifier_info *info,
 346                              struct netlink_ext_ack *extack)
 347 {
 348         info->dev = dev;
 349         info->extack = extack;
 350         return atomic_notifier_call_chain(&switchdev_notif_chain, val, info);
 351 }
 352 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
 353
 354 int register_switchdev_blocking_notifier(struct notifier_block *nb)
 355 {
 356         struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
 357
 358         return blocking_notifier_chain_register(chain, nb);
 359 }
 360 EXPORT_SYMBOL_GPL(register_switchdev_blocking_notifier);
 361
 362 int unregister_switchdev_blocking_notifier(struct notifier_block *nb)
 363 {
 364         struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
 365
 366         return blocking_notifier_chain_unregister(chain, nb);
 367 }
 368 EXPORT_SYMBOL_GPL(unregister_switchdev_blocking_notifier);
 369
 370 int call_switchdev_blocking_notifiers(unsigned long val, struct net_device *dev,
 371                                       struct switchdev_notifier_info *info,
 372                                       struct netlink_ext_ack *extack)
 373 {
 374         info->dev = dev;
 375         info->extack = extack;
 376         return blocking_notifier_call_chain(&switchdev_blocking_notif_chain,
 377                                             val, info);
 378 }
 379 EXPORT_SYMBOL_GPL(call_switchdev_blocking_notifiers);
 380
 381 static int __switchdev_handle_port_obj_add(struct net_device *dev,
 382                         struct switchdev_notifier_port_obj_info *port_obj_info,
 383                         bool (*check_cb)(const struct net_device *dev),
 384                         int (*add_cb)(struct net_device *dev,
 385                                       const struct switchdev_obj *obj,
 386                                       struct netlink_ext_ack *extack))
 387 {
 388         struct netlink_ext_ack *extack;
 389         struct net_device *lower_dev;
 390         struct list_head *iter;
 391         int err = -EOPNOTSUPP;
 392
 393         extack = switchdev_notifier_info_to_extack(&port_obj_info->info);
 394
 395         if (check_cb(dev)) {
 396                 err = add_cb(dev, port_obj_info->obj, extack);
 397                 if (err != -EOPNOTSUPP)
 398                         port_obj_info->handled = true;
 399                 return err;
 400         }
 401
 402         /* Switch ports might be stacked under e.g. a LAG. Ignore the
 403          * unsupported devices, another driver might be able to handle them. But
 404          * propagate to the callers any hard errors.
 405          *
 406          * If the driver does its own bookkeeping of stacked ports, it's not
 407          * necessary to go through this helper.
 408          */
 409         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 410                 if (netif_is_bridge_master(lower_dev))
 411                         continue;
 412
 413                 err = __switchdev_handle_port_obj_add(lower_dev, port_obj_info,
 414                                                       check_cb, add_cb);
 415                 if (err && err != -EOPNOTSUPP)
 416                         return err;
 417         }
 418
 419         return err;
 420 }
 421
 422 int switchdev_handle_port_obj_add(struct net_device *dev,
 423                         struct switchdev_notifier_port_obj_info *port_obj_info,
 424                         bool (*check_cb)(const struct net_device *dev),
 425                         int (*add_cb)(struct net_device *dev,
 426                                       const struct switchdev_obj *obj,
 427                                       struct netlink_ext_ack *extack))
 428 {
 429         int err;
 430
 431         err = __switchdev_handle_port_obj_add(dev, port_obj_info, check_cb,
 432                                               add_cb);
 433         if (err == -EOPNOTSUPP)
 434                 err = 0;
 435         return err;
 436 }
 437 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_add);
 438
 439 static int __switchdev_handle_port_obj_del(struct net_device *dev,
 440                         struct switchdev_notifier_port_obj_info *port_obj_info,
 441                         bool (*check_cb)(const struct net_device *dev),
 442                         int (*del_cb)(struct net_device *dev,
 443                                       const struct switchdev_obj *obj))
 444 {
 445         struct net_device *lower_dev;
 446         struct list_head *iter;
 447         int err = -EOPNOTSUPP;
 448
 449         if (check_cb(dev)) {
 450                 err = del_cb(dev, port_obj_info->obj);
 451                 if (err != -EOPNOTSUPP)
 452                         port_obj_info->handled = true;
 453                 return err;
 454         }
 455
 456         /* Switch ports might be stacked under e.g. a LAG. Ignore the
 457          * unsupported devices, another driver might be able to handle them. But
 458          * propagate to the callers any hard errors.
 459          *
 460          * If the driver does its own bookkeeping of stacked ports, it's not
 461          * necessary to go through this helper.
 462          */
 463         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 464                 if (netif_is_bridge_master(lower_dev))
 465                         continue;
 466
 467                 err = __switchdev_handle_port_obj_del(lower_dev, port_obj_info,
 468                                                       check_cb, del_cb);
 469                 if (err && err != -EOPNOTSUPP)
 470                         return err;
 471         }
 472
 473         return err;
 474 }
 475
 476 int switchdev_handle_port_obj_del(struct net_device *dev,
 477                         struct switchdev_notifier_port_obj_info *port_obj_info,
 478                         bool (*check_cb)(const struct net_device *dev),
 479                         int (*del_cb)(struct net_device *dev,
 480                                       const struct switchdev_obj *obj))
 481 {
 482         int err;
 483
 484         err = __switchdev_handle_port_obj_del(dev, port_obj_info, check_cb,
 485                                               del_cb);
 486         if (err == -EOPNOTSUPP)
 487                 err = 0;
 488         return err;
 489 }
 490 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_del);
 491
 492 static int __switchdev_handle_port_attr_set(struct net_device *dev,
 493                         struct switchdev_notifier_port_attr_info *port_attr_info,
 494                         bool (*check_cb)(const struct net_device *dev),
 495                         int (*set_cb)(struct net_device *dev,
 496                                       const struct switchdev_attr *attr,
 497                                       struct netlink_ext_ack *extack))
 498 {
 499         struct netlink_ext_ack *extack;
 500         struct net_device *lower_dev;
 501         struct list_head *iter;
 502         int err = -EOPNOTSUPP;
 503
 504         extack = switchdev_notifier_info_to_extack(&port_attr_info->info);
 505
 506         if (check_cb(dev)) {
 507                 err = set_cb(dev, port_attr_info->attr, extack);
 508                 if (err != -EOPNOTSUPP)
 509                         port_attr_info->handled = true;
 510                 return err;
 511         }
 512
 513         /* Switch ports might be stacked under e.g. a LAG. Ignore the
 514          * unsupported devices, another driver might be able to handle them. But
 515          * propagate to the callers any hard errors.
 516          *
 517          * If the driver does its own bookkeeping of stacked ports, it's not
 518          * necessary to go through this helper.
 519          */
 520         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 521                 if (netif_is_bridge_master(lower_dev))
 522                         continue;
 523
 524                 err = __switchdev_handle_port_attr_set(lower_dev, port_attr_info,
 525                                                        check_cb, set_cb);
 526                 if (err && err != -EOPNOTSUPP)
 527                         return err;
 528         }
 529
 530         return err;
 531 }
 532
 533 int switchdev_handle_port_attr_set(struct net_device *dev,
 534                         struct switchdev_notifier_port_attr_info *port_attr_info,
 535                         bool (*check_cb)(const struct net_device *dev),
 536                         int (*set_cb)(struct net_device *dev,
 537                                       const struct switchdev_attr *attr,
 538                                       struct netlink_ext_ack *extack))
 539 {
 540         int err;
 541
 542         err = __switchdev_handle_port_attr_set(dev, port_attr_info, check_cb,
 543                                                set_cb);
 544         if (err == -EOPNOTSUPP)
 545                 err = 0;
 546         return err;
 547 }
 548 EXPORT_SYMBOL_GPL(switchdev_handle_port_attr_set);