net/switchdev/switchdev.c

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