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 netlink_ext_ack *extack) 225 { 226 int rc; 227 int err; 228 229 struct switchdev_notifier_port_obj_info obj_info = { 230 .obj = obj, 231 .handled = false, 232 }; 233 234 rc = call_switchdev_blocking_notifiers(nt, dev, &obj_info.info, extack); 235 err = notifier_to_errno(rc); 236 if (err) { 237 WARN_ON(!obj_info.handled); 238 return err; 239 } 240 if (!obj_info.handled) 241 return -EOPNOTSUPP; 242 return 0; 243 } 244 245 static void switchdev_port_obj_add_deferred(struct net_device *dev, 246 const void *data) 247 { 248 const struct switchdev_obj *obj = data; 249 int err; 250 251 ASSERT_RTNL(); 252 err = switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD, 253 dev, obj, NULL); 254 if (err && err != -EOPNOTSUPP) 255 netdev_err(dev, "failed (err=%d) to add object (id=%d)\n", 256 err, obj->id); 257 if (obj->complete) 258 obj->complete(dev, err, obj->complete_priv); 259 } 260 261 static int switchdev_port_obj_add_defer(struct net_device *dev, 262 const struct switchdev_obj *obj) 263 { 264 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj), 265 switchdev_port_obj_add_deferred); 266 } 267 268 /** 269 * switchdev_port_obj_add - Add port object 270 * 271 * @dev: port device 272 * @obj: object to add 273 * @extack: netlink extended ack 274 * 275 * rtnl_lock must be held and must not be in atomic section, 276 * in case SWITCHDEV_F_DEFER flag is not set. 277 */ 278 int switchdev_port_obj_add(struct net_device *dev, 279 const struct switchdev_obj *obj, 280 struct netlink_ext_ack *extack) 281 { 282 if (obj->flags & SWITCHDEV_F_DEFER) 283 return switchdev_port_obj_add_defer(dev, obj); 284 ASSERT_RTNL(); 285 return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD, 286 dev, obj, extack); 287 } 288 EXPORT_SYMBOL_GPL(switchdev_port_obj_add); 289 290 static int switchdev_port_obj_del_now(struct net_device *dev, 291 const struct switchdev_obj *obj) 292 { 293 return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_DEL, 294 dev, obj, NULL); 295 } 296 297 static void switchdev_port_obj_del_deferred(struct net_device *dev, 298 const void *data) 299 { 300 const struct switchdev_obj *obj = data; 301 int err; 302 303 err = switchdev_port_obj_del_now(dev, obj); 304 if (err && err != -EOPNOTSUPP) 305 netdev_err(dev, "failed (err=%d) to del object (id=%d)\n", 306 err, obj->id); 307 if (obj->complete) 308 obj->complete(dev, err, obj->complete_priv); 309 } 310 311 static int switchdev_port_obj_del_defer(struct net_device *dev, 312 const struct switchdev_obj *obj) 313 { 314 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj), 315 switchdev_port_obj_del_deferred); 316 } 317 318 /** 319 * switchdev_port_obj_del - Delete port object 320 * 321 * @dev: port device 322 * @obj: object to delete 323 * 324 * rtnl_lock must be held and must not be in atomic section, 325 * in case SWITCHDEV_F_DEFER flag is not set. 326 */ 327 int switchdev_port_obj_del(struct net_device *dev, 328 const struct switchdev_obj *obj) 329 { 330 if (obj->flags & SWITCHDEV_F_DEFER) 331 return switchdev_port_obj_del_defer(dev, obj); 332 ASSERT_RTNL(); 333 return switchdev_port_obj_del_now(dev, obj); 334 } 335 EXPORT_SYMBOL_GPL(switchdev_port_obj_del); 336 337 static ATOMIC_NOTIFIER_HEAD(switchdev_notif_chain); 338 static BLOCKING_NOTIFIER_HEAD(switchdev_blocking_notif_chain); 339 340 /** 341 * register_switchdev_notifier - Register notifier 342 * @nb: notifier_block 343 * 344 * Register switch device notifier. 345 */ 346 int register_switchdev_notifier(struct notifier_block *nb) 347 { 348 return atomic_notifier_chain_register(&switchdev_notif_chain, nb); 349 } 350 EXPORT_SYMBOL_GPL(register_switchdev_notifier); 351 352 /** 353 * unregister_switchdev_notifier - Unregister notifier 354 * @nb: notifier_block 355 * 356 * Unregister switch device notifier. 357 */ 358 int unregister_switchdev_notifier(struct notifier_block *nb) 359 { 360 return atomic_notifier_chain_unregister(&switchdev_notif_chain, nb); 361 } 362 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier); 363 364 /** 365 * call_switchdev_notifiers - Call notifiers 366 * @val: value passed unmodified to notifier function 367 * @dev: port device 368 * @info: notifier information data 369 * @extack: netlink extended ack 370 * Call all network notifier blocks. 371 */ 372 int call_switchdev_notifiers(unsigned long val, struct net_device *dev, 373 struct switchdev_notifier_info *info, 374 struct netlink_ext_ack *extack) 375 { 376 info->dev = dev; 377 info->extack = extack; 378 return atomic_notifier_call_chain(&switchdev_notif_chain, val, info); 379 } 380 EXPORT_SYMBOL_GPL(call_switchdev_notifiers); 381 382 int register_switchdev_blocking_notifier(struct notifier_block *nb) 383 { 384 struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain; 385 386 return blocking_notifier_chain_register(chain, nb); 387 } 388 EXPORT_SYMBOL_GPL(register_switchdev_blocking_notifier); 389 390 int unregister_switchdev_blocking_notifier(struct notifier_block *nb) 391 { 392 struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain; 393 394 return blocking_notifier_chain_unregister(chain, nb); 395 } 396 EXPORT_SYMBOL_GPL(unregister_switchdev_blocking_notifier); 397 398 int call_switchdev_blocking_notifiers(unsigned long val, struct net_device *dev, 399 struct switchdev_notifier_info *info, 400 struct netlink_ext_ack *extack) 401 { 402 info->dev = dev; 403 info->extack = extack; 404 return blocking_notifier_call_chain(&switchdev_blocking_notif_chain, 405 val, info); 406 } 407 EXPORT_SYMBOL_GPL(call_switchdev_blocking_notifiers); 408 409 static int __switchdev_handle_port_obj_add(struct net_device *dev, 410 struct switchdev_notifier_port_obj_info *port_obj_info, 411 bool (*check_cb)(const struct net_device *dev), 412 int (*add_cb)(struct net_device *dev, 413 const struct switchdev_obj *obj, 414 struct netlink_ext_ack *extack)) 415 { 416 struct netlink_ext_ack *extack; 417 struct net_device *lower_dev; 418 struct list_head *iter; 419 int err = -EOPNOTSUPP; 420 421 extack = switchdev_notifier_info_to_extack(&port_obj_info->info); 422 423 if (check_cb(dev)) { 424 /* This flag is only checked if the return value is success. */ 425 port_obj_info->handled = true; 426 return add_cb(dev, port_obj_info->obj, extack); 427 } 428 429 /* Switch ports might be stacked under e.g. a LAG. Ignore the 430 * unsupported devices, another driver might be able to handle them. But 431 * propagate to the callers any hard errors. 432 * 433 * If the driver does its own bookkeeping of stacked ports, it's not 434 * necessary to go through this helper. 435 */ 436 netdev_for_each_lower_dev(dev, lower_dev, iter) { 437 if (netif_is_bridge_master(lower_dev)) 438 continue; 439 440 err = __switchdev_handle_port_obj_add(lower_dev, port_obj_info, 441 check_cb, add_cb); 442 if (err && err != -EOPNOTSUPP) 443 return err; 444 } 445 446 return err; 447 } 448 449 int switchdev_handle_port_obj_add(struct net_device *dev, 450 struct switchdev_notifier_port_obj_info *port_obj_info, 451 bool (*check_cb)(const struct net_device *dev), 452 int (*add_cb)(struct net_device *dev, 453 const struct switchdev_obj *obj, 454 struct netlink_ext_ack *extack)) 455 { 456 int err; 457 458 err = __switchdev_handle_port_obj_add(dev, port_obj_info, check_cb, 459 add_cb); 460 if (err == -EOPNOTSUPP) 461 err = 0; 462 return err; 463 } 464 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_add); 465 466 static int __switchdev_handle_port_obj_del(struct net_device *dev, 467 struct switchdev_notifier_port_obj_info *port_obj_info, 468 bool (*check_cb)(const struct net_device *dev), 469 int (*del_cb)(struct net_device *dev, 470 const struct switchdev_obj *obj)) 471 { 472 struct net_device *lower_dev; 473 struct list_head *iter; 474 int err = -EOPNOTSUPP; 475 476 if (check_cb(dev)) { 477 /* This flag is only checked if the return value is success. */ 478 port_obj_info->handled = true; 479 return del_cb(dev, port_obj_info->obj); 480 } 481 482 /* Switch ports might be stacked under e.g. a LAG. Ignore the 483 * unsupported devices, another driver might be able to handle them. But 484 * propagate to the callers any hard errors. 485 * 486 * If the driver does its own bookkeeping of stacked ports, it's not 487 * necessary to go through this helper. 488 */ 489 netdev_for_each_lower_dev(dev, lower_dev, iter) { 490 if (netif_is_bridge_master(lower_dev)) 491 continue; 492 493 err = __switchdev_handle_port_obj_del(lower_dev, port_obj_info, 494 check_cb, del_cb); 495 if (err && err != -EOPNOTSUPP) 496 return err; 497 } 498 499 return err; 500 } 501 502 int switchdev_handle_port_obj_del(struct net_device *dev, 503 struct switchdev_notifier_port_obj_info *port_obj_info, 504 bool (*check_cb)(const struct net_device *dev), 505 int (*del_cb)(struct net_device *dev, 506 const struct switchdev_obj *obj)) 507 { 508 int err; 509 510 err = __switchdev_handle_port_obj_del(dev, port_obj_info, check_cb, 511 del_cb); 512 if (err == -EOPNOTSUPP) 513 err = 0; 514 return err; 515 } 516 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_del); 517 518 static int __switchdev_handle_port_attr_set(struct net_device *dev, 519 struct switchdev_notifier_port_attr_info *port_attr_info, 520 bool (*check_cb)(const struct net_device *dev), 521 int (*set_cb)(struct net_device *dev, 522 const struct switchdev_attr *attr, 523 struct switchdev_trans *trans)) 524 { 525 struct net_device *lower_dev; 526 struct list_head *iter; 527 int err = -EOPNOTSUPP; 528 529 if (check_cb(dev)) { 530 port_attr_info->handled = true; 531 return set_cb(dev, port_attr_info->attr, 532 port_attr_info->trans); 533 } 534 535 /* Switch ports might be stacked under e.g. a LAG. Ignore the 536 * unsupported devices, another driver might be able to handle them. But 537 * propagate to the callers any hard errors. 538 * 539 * If the driver does its own bookkeeping of stacked ports, it's not 540 * necessary to go through this helper. 541 */ 542 netdev_for_each_lower_dev(dev, lower_dev, iter) { 543 if (netif_is_bridge_master(lower_dev)) 544 continue; 545 546 err = __switchdev_handle_port_attr_set(lower_dev, port_attr_info, 547 check_cb, set_cb); 548 if (err && err != -EOPNOTSUPP) 549 return err; 550 } 551 552 return err; 553 } 554 555 int switchdev_handle_port_attr_set(struct net_device *dev, 556 struct switchdev_notifier_port_attr_info *port_attr_info, 557 bool (*check_cb)(const struct net_device *dev), 558 int (*set_cb)(struct net_device *dev, 559 const struct switchdev_attr *attr, 560 struct switchdev_trans *trans)) 561 { 562 int err; 563 564 err = __switchdev_handle_port_attr_set(dev, port_attr_info, check_cb, 565 set_cb); 566 if (err == -EOPNOTSUPP) 567 err = 0; 568 return err; 569 } 570 EXPORT_SYMBOL_GPL(switchdev_handle_port_attr_set); 571