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