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  *      @id: object ID
 308  *      @obj: object to add
 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  *      @id: object ID
 361  *      @obj: object to delete
 362  *
 363  *      rtnl_lock must be held and must not be in atomic section,
 364  *      in case SWITCHDEV_F_DEFER flag is not set.
 365  */
 366 int switchdev_port_obj_del(struct net_device *dev,
 367                            const struct switchdev_obj *obj)
 368 {
 369         if (obj->flags & SWITCHDEV_F_DEFER)
 370                 return switchdev_port_obj_del_defer(dev, obj);
 371         ASSERT_RTNL();
 372         return switchdev_port_obj_del_now(dev, obj);
 373 }
 374 EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
 375
 376 static ATOMIC_NOTIFIER_HEAD(switchdev_notif_chain);
 377 static BLOCKING_NOTIFIER_HEAD(switchdev_blocking_notif_chain);
 378
 379 /**
 380  *      register_switchdev_notifier - Register notifier
 381  *      @nb: notifier_block
 382  *
 383  *      Register switch device notifier.
 384  */
 385 int register_switchdev_notifier(struct notifier_block *nb)
 386 {
 387         return atomic_notifier_chain_register(&switchdev_notif_chain, nb);
 388 }
 389 EXPORT_SYMBOL_GPL(register_switchdev_notifier);
 390
 391 /**
 392  *      unregister_switchdev_notifier - Unregister notifier
 393  *      @nb: notifier_block
 394  *
 395  *      Unregister switch device notifier.
 396  */
 397 int unregister_switchdev_notifier(struct notifier_block *nb)
 398 {
 399         return atomic_notifier_chain_unregister(&switchdev_notif_chain, nb);
 400 }
 401 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
 402
 403 /**
 404  *      call_switchdev_notifiers - Call notifiers
 405  *      @val: value passed unmodified to notifier function
 406  *      @dev: port device
 407  *      @info: notifier information data
 408  *
 409  *      Call all network notifier blocks.
 410  */
 411 int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
 412                              struct switchdev_notifier_info *info,
 413                              struct netlink_ext_ack *extack)
 414 {
 415         info->dev = dev;
 416         info->extack = extack;
 417         return atomic_notifier_call_chain(&switchdev_notif_chain, val, info);
 418 }
 419 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
 420
 421 int register_switchdev_blocking_notifier(struct notifier_block *nb)
 422 {
 423         struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
 424
 425         return blocking_notifier_chain_register(chain, nb);
 426 }
 427 EXPORT_SYMBOL_GPL(register_switchdev_blocking_notifier);
 428
 429 int unregister_switchdev_blocking_notifier(struct notifier_block *nb)
 430 {
 431         struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain;
 432
 433         return blocking_notifier_chain_unregister(chain, nb);
 434 }
 435 EXPORT_SYMBOL_GPL(unregister_switchdev_blocking_notifier);
 436
 437 int call_switchdev_blocking_notifiers(unsigned long val, struct net_device *dev,
 438                                       struct switchdev_notifier_info *info,
 439                                       struct netlink_ext_ack *extack)
 440 {
 441         info->dev = dev;
 442         info->extack = extack;
 443         return blocking_notifier_call_chain(&switchdev_blocking_notif_chain,
 444                                             val, info);
 445 }
 446 EXPORT_SYMBOL_GPL(call_switchdev_blocking_notifiers);
 447
 448 static int __switchdev_handle_port_obj_add(struct net_device *dev,
 449                         struct switchdev_notifier_port_obj_info *port_obj_info,
 450                         bool (*check_cb)(const struct net_device *dev),
 451                         int (*add_cb)(struct net_device *dev,
 452                                       const struct switchdev_obj *obj,
 453                                       struct switchdev_trans *trans,
 454                                       struct netlink_ext_ack *extack))
 455 {
 456         struct netlink_ext_ack *extack;
 457         struct net_device *lower_dev;
 458         struct list_head *iter;
 459         int err = -EOPNOTSUPP;
 460
 461         extack = switchdev_notifier_info_to_extack(&port_obj_info->info);
 462
 463         if (check_cb(dev)) {
 464                 /* This flag is only checked if the return value is success. */
 465                 port_obj_info->handled = true;
 466                 return add_cb(dev, port_obj_info->obj, port_obj_info->trans,
 467                               extack);
 468         }
 469
 470         /* Switch ports might be stacked under e.g. a LAG. Ignore the
 471          * unsupported devices, another driver might be able to handle them. But
 472          * propagate to the callers any hard errors.
 473          *
 474          * If the driver does its own bookkeeping of stacked ports, it's not
 475          * necessary to go through this helper.
 476          */
 477         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 478                 if (netif_is_bridge_master(lower_dev))
 479                         continue;
 480
 481                 err = __switchdev_handle_port_obj_add(lower_dev, port_obj_info,
 482                                                       check_cb, add_cb);
 483                 if (err && err != -EOPNOTSUPP)
 484                         return err;
 485         }
 486
 487         return err;
 488 }
 489
 490 int switchdev_handle_port_obj_add(struct net_device *dev,
 491                         struct switchdev_notifier_port_obj_info *port_obj_info,
 492                         bool (*check_cb)(const struct net_device *dev),
 493                         int (*add_cb)(struct net_device *dev,
 494                                       const struct switchdev_obj *obj,
 495                                       struct switchdev_trans *trans,
 496                                       struct netlink_ext_ack *extack))
 497 {
 498         int err;
 499
 500         err = __switchdev_handle_port_obj_add(dev, port_obj_info, check_cb,
 501                                               add_cb);
 502         if (err == -EOPNOTSUPP)
 503                 err = 0;
 504         return err;
 505 }
 506 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_add);
 507
 508 static int __switchdev_handle_port_obj_del(struct net_device *dev,
 509                         struct switchdev_notifier_port_obj_info *port_obj_info,
 510                         bool (*check_cb)(const struct net_device *dev),
 511                         int (*del_cb)(struct net_device *dev,
 512                                       const struct switchdev_obj *obj))
 513 {
 514         struct net_device *lower_dev;
 515         struct list_head *iter;
 516         int err = -EOPNOTSUPP;
 517
 518         if (check_cb(dev)) {
 519                 /* This flag is only checked if the return value is success. */
 520                 port_obj_info->handled = true;
 521                 return del_cb(dev, port_obj_info->obj);
 522         }
 523
 524         /* Switch ports might be stacked under e.g. a LAG. Ignore the
 525          * unsupported devices, another driver might be able to handle them. But
 526          * propagate to the callers any hard errors.
 527          *
 528          * If the driver does its own bookkeeping of stacked ports, it's not
 529          * necessary to go through this helper.
 530          */
 531         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 532                 if (netif_is_bridge_master(lower_dev))
 533                         continue;
 534
 535                 err = __switchdev_handle_port_obj_del(lower_dev, port_obj_info,
 536                                                       check_cb, del_cb);
 537                 if (err && err != -EOPNOTSUPP)
 538                         return err;
 539         }
 540
 541         return err;
 542 }
 543
 544 int switchdev_handle_port_obj_del(struct net_device *dev,
 545                         struct switchdev_notifier_port_obj_info *port_obj_info,
 546                         bool (*check_cb)(const struct net_device *dev),
 547                         int (*del_cb)(struct net_device *dev,
 548                                       const struct switchdev_obj *obj))
 549 {
 550         int err;
 551
 552         err = __switchdev_handle_port_obj_del(dev, port_obj_info, check_cb,
 553                                               del_cb);
 554         if (err == -EOPNOTSUPP)
 555                 err = 0;
 556         return err;
 557 }
 558 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_del);
 559
 560 static int __switchdev_handle_port_attr_set(struct net_device *dev,
 561                         struct switchdev_notifier_port_attr_info *port_attr_info,
 562                         bool (*check_cb)(const struct net_device *dev),
 563                         int (*set_cb)(struct net_device *dev,
 564                                       const struct switchdev_attr *attr,
 565                                       struct switchdev_trans *trans))
 566 {
 567         struct net_device *lower_dev;
 568         struct list_head *iter;
 569         int err = -EOPNOTSUPP;
 570
 571         if (check_cb(dev)) {
 572                 port_attr_info->handled = true;
 573                 return set_cb(dev, port_attr_info->attr,
 574                               port_attr_info->trans);
 575         }
 576
 577         /* Switch ports might be stacked under e.g. a LAG. Ignore the
 578          * unsupported devices, another driver might be able to handle them. But
 579          * propagate to the callers any hard errors.
 580          *
 581          * If the driver does its own bookkeeping of stacked ports, it's not
 582          * necessary to go through this helper.
 583          */
 584         netdev_for_each_lower_dev(dev, lower_dev, iter) {
 585                 if (netif_is_bridge_master(lower_dev))
 586                         continue;
 587
 588                 err = __switchdev_handle_port_attr_set(lower_dev, port_attr_info,
 589                                                        check_cb, set_cb);
 590                 if (err && err != -EOPNOTSUPP)
 591                         return err;
 592         }
 593
 594         return err;
 595 }
 596
 597 int switchdev_handle_port_attr_set(struct net_device *dev,
 598                         struct switchdev_notifier_port_attr_info *port_attr_info,
 599                         bool (*check_cb)(const struct net_device *dev),
 600                         int (*set_cb)(struct net_device *dev,
 601                                       const struct switchdev_attr *attr,
 602                                       struct switchdev_trans *trans))
 603 {
 604         int err;
 605
 606         err = __switchdev_handle_port_attr_set(dev, port_attr_info, check_cb,
 607                                                set_cb);
 608         if (err == -EOPNOTSUPP)
 609                 err = 0;
 610         return err;
 611 }
 612 EXPORT_SYMBOL_GPL(switchdev_handle_port_attr_set);