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 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 int switchdev_port_obj_add_now(struct net_device *dev, 246 const struct switchdev_obj *obj, 247 struct netlink_ext_ack *extack) 248 { 249 ASSERT_RTNL(); 250 251 return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_ADD, 252 dev, obj, extack); 253 } 254 255 static void switchdev_port_obj_add_deferred(struct net_device *dev, 256 const void *data) 257 { 258 const struct switchdev_obj *obj = data; 259 int err; 260 261 err = switchdev_port_obj_add_now(dev, obj, NULL); 262 if (err && err != -EOPNOTSUPP) 263 netdev_err(dev, "failed (err=%d) to add object (id=%d)\n", 264 err, obj->id); 265 if (obj->complete) 266 obj->complete(dev, err, obj->complete_priv); 267 } 268 269 static int switchdev_port_obj_add_defer(struct net_device *dev, 270 const struct switchdev_obj *obj) 271 { 272 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj), 273 switchdev_port_obj_add_deferred); 274 } 275 276 /** 277 * switchdev_port_obj_add - Add port object 278 * 279 * @dev: port device 280 * @obj: object to add 281 * @extack: netlink extended ack 282 * 283 * rtnl_lock must be held and must not be in atomic section, 284 * in case SWITCHDEV_F_DEFER flag is not set. 285 */ 286 int switchdev_port_obj_add(struct net_device *dev, 287 const struct switchdev_obj *obj, 288 struct netlink_ext_ack *extack) 289 { 290 if (obj->flags & SWITCHDEV_F_DEFER) 291 return switchdev_port_obj_add_defer(dev, obj); 292 ASSERT_RTNL(); 293 return switchdev_port_obj_add_now(dev, obj, extack); 294 } 295 EXPORT_SYMBOL_GPL(switchdev_port_obj_add); 296 297 static int switchdev_port_obj_del_now(struct net_device *dev, 298 const struct switchdev_obj *obj) 299 { 300 return switchdev_port_obj_notify(SWITCHDEV_PORT_OBJ_DEL, 301 dev, obj, NULL); 302 } 303 304 static void switchdev_port_obj_del_deferred(struct net_device *dev, 305 const void *data) 306 { 307 const struct switchdev_obj *obj = data; 308 int err; 309 310 err = switchdev_port_obj_del_now(dev, obj); 311 if (err && err != -EOPNOTSUPP) 312 netdev_err(dev, "failed (err=%d) to del object (id=%d)\n", 313 err, obj->id); 314 if (obj->complete) 315 obj->complete(dev, err, obj->complete_priv); 316 } 317 318 static int switchdev_port_obj_del_defer(struct net_device *dev, 319 const struct switchdev_obj *obj) 320 { 321 return switchdev_deferred_enqueue(dev, obj, switchdev_obj_size(obj), 322 switchdev_port_obj_del_deferred); 323 } 324 325 /** 326 * switchdev_port_obj_del - Delete port object 327 * 328 * @dev: port device 329 * @obj: object to delete 330 * 331 * rtnl_lock must be held and must not be in atomic section, 332 * in case SWITCHDEV_F_DEFER flag is not set. 333 */ 334 int switchdev_port_obj_del(struct net_device *dev, 335 const struct switchdev_obj *obj) 336 { 337 if (obj->flags & SWITCHDEV_F_DEFER) 338 return switchdev_port_obj_del_defer(dev, obj); 339 ASSERT_RTNL(); 340 return switchdev_port_obj_del_now(dev, obj); 341 } 342 EXPORT_SYMBOL_GPL(switchdev_port_obj_del); 343 344 static ATOMIC_NOTIFIER_HEAD(switchdev_notif_chain); 345 static BLOCKING_NOTIFIER_HEAD(switchdev_blocking_notif_chain); 346 347 /** 348 * register_switchdev_notifier - Register notifier 349 * @nb: notifier_block 350 * 351 * Register switch device notifier. 352 */ 353 int register_switchdev_notifier(struct notifier_block *nb) 354 { 355 return atomic_notifier_chain_register(&switchdev_notif_chain, nb); 356 } 357 EXPORT_SYMBOL_GPL(register_switchdev_notifier); 358 359 /** 360 * unregister_switchdev_notifier - Unregister notifier 361 * @nb: notifier_block 362 * 363 * Unregister switch device notifier. 364 */ 365 int unregister_switchdev_notifier(struct notifier_block *nb) 366 { 367 return atomic_notifier_chain_unregister(&switchdev_notif_chain, nb); 368 } 369 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier); 370 371 /** 372 * call_switchdev_notifiers - Call notifiers 373 * @val: value passed unmodified to notifier function 374 * @dev: port device 375 * @info: notifier information data 376 * @extack: netlink extended ack 377 * Call all network notifier blocks. 378 */ 379 int call_switchdev_notifiers(unsigned long val, struct net_device *dev, 380 struct switchdev_notifier_info *info, 381 struct netlink_ext_ack *extack) 382 { 383 info->dev = dev; 384 info->extack = extack; 385 return atomic_notifier_call_chain(&switchdev_notif_chain, val, info); 386 } 387 EXPORT_SYMBOL_GPL(call_switchdev_notifiers); 388 389 int register_switchdev_blocking_notifier(struct notifier_block *nb) 390 { 391 struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain; 392 393 return blocking_notifier_chain_register(chain, nb); 394 } 395 EXPORT_SYMBOL_GPL(register_switchdev_blocking_notifier); 396 397 int unregister_switchdev_blocking_notifier(struct notifier_block *nb) 398 { 399 struct blocking_notifier_head *chain = &switchdev_blocking_notif_chain; 400 401 return blocking_notifier_chain_unregister(chain, nb); 402 } 403 EXPORT_SYMBOL_GPL(unregister_switchdev_blocking_notifier); 404 405 int call_switchdev_blocking_notifiers(unsigned long val, struct net_device *dev, 406 struct switchdev_notifier_info *info, 407 struct netlink_ext_ack *extack) 408 { 409 info->dev = dev; 410 info->extack = extack; 411 return blocking_notifier_call_chain(&switchdev_blocking_notif_chain, 412 val, info); 413 } 414 EXPORT_SYMBOL_GPL(call_switchdev_blocking_notifiers); 415 416 static int __switchdev_handle_port_obj_add(struct net_device *dev, 417 struct switchdev_notifier_port_obj_info *port_obj_info, 418 bool (*check_cb)(const struct net_device *dev), 419 int (*add_cb)(struct net_device *dev, 420 const struct switchdev_obj *obj, 421 struct netlink_ext_ack *extack)) 422 { 423 struct netlink_ext_ack *extack; 424 struct net_device *lower_dev; 425 struct list_head *iter; 426 int err = -EOPNOTSUPP; 427 428 extack = switchdev_notifier_info_to_extack(&port_obj_info->info); 429 430 if (check_cb(dev)) { 431 /* This flag is only checked if the return value is success. */ 432 port_obj_info->handled = true; 433 return add_cb(dev, port_obj_info->obj, extack); 434 } 435 436 /* Switch ports might be stacked under e.g. a LAG. Ignore the 437 * unsupported devices, another driver might be able to handle them. But 438 * propagate to the callers any hard errors. 439 * 440 * If the driver does its own bookkeeping of stacked ports, it's not 441 * necessary to go through this helper. 442 */ 443 netdev_for_each_lower_dev(dev, lower_dev, iter) { 444 if (netif_is_bridge_master(lower_dev)) 445 continue; 446 447 err = __switchdev_handle_port_obj_add(lower_dev, port_obj_info, 448 check_cb, add_cb); 449 if (err && err != -EOPNOTSUPP) 450 return err; 451 } 452 453 return err; 454 } 455 456 int switchdev_handle_port_obj_add(struct net_device *dev, 457 struct switchdev_notifier_port_obj_info *port_obj_info, 458 bool (*check_cb)(const struct net_device *dev), 459 int (*add_cb)(struct net_device *dev, 460 const struct switchdev_obj *obj, 461 struct netlink_ext_ack *extack)) 462 { 463 int err; 464 465 err = __switchdev_handle_port_obj_add(dev, port_obj_info, check_cb, 466 add_cb); 467 if (err == -EOPNOTSUPP) 468 err = 0; 469 return err; 470 } 471 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_add); 472 473 static int __switchdev_handle_port_obj_del(struct net_device *dev, 474 struct switchdev_notifier_port_obj_info *port_obj_info, 475 bool (*check_cb)(const struct net_device *dev), 476 int (*del_cb)(struct net_device *dev, 477 const struct switchdev_obj *obj)) 478 { 479 struct net_device *lower_dev; 480 struct list_head *iter; 481 int err = -EOPNOTSUPP; 482 483 if (check_cb(dev)) { 484 /* This flag is only checked if the return value is success. */ 485 port_obj_info->handled = true; 486 return del_cb(dev, port_obj_info->obj); 487 } 488 489 /* Switch ports might be stacked under e.g. a LAG. Ignore the 490 * unsupported devices, another driver might be able to handle them. But 491 * propagate to the callers any hard errors. 492 * 493 * If the driver does its own bookkeeping of stacked ports, it's not 494 * necessary to go through this helper. 495 */ 496 netdev_for_each_lower_dev(dev, lower_dev, iter) { 497 if (netif_is_bridge_master(lower_dev)) 498 continue; 499 500 err = __switchdev_handle_port_obj_del(lower_dev, port_obj_info, 501 check_cb, del_cb); 502 if (err && err != -EOPNOTSUPP) 503 return err; 504 } 505 506 return err; 507 } 508 509 int switchdev_handle_port_obj_del(struct net_device *dev, 510 struct switchdev_notifier_port_obj_info *port_obj_info, 511 bool (*check_cb)(const struct net_device *dev), 512 int (*del_cb)(struct net_device *dev, 513 const struct switchdev_obj *obj)) 514 { 515 int err; 516 517 err = __switchdev_handle_port_obj_del(dev, port_obj_info, check_cb, 518 del_cb); 519 if (err == -EOPNOTSUPP) 520 err = 0; 521 return err; 522 } 523 EXPORT_SYMBOL_GPL(switchdev_handle_port_obj_del); 524 525 static int __switchdev_handle_port_attr_set(struct net_device *dev, 526 struct switchdev_notifier_port_attr_info *port_attr_info, 527 bool (*check_cb)(const struct net_device *dev), 528 int (*set_cb)(struct net_device *dev, 529 const struct switchdev_attr *attr, 530 struct switchdev_trans *trans)) 531 { 532 struct net_device *lower_dev; 533 struct list_head *iter; 534 int err = -EOPNOTSUPP; 535 536 if (check_cb(dev)) { 537 port_attr_info->handled = true; 538 return set_cb(dev, port_attr_info->attr, 539 port_attr_info->trans); 540 } 541 542 /* Switch ports might be stacked under e.g. a LAG. Ignore the 543 * unsupported devices, another driver might be able to handle them. But 544 * propagate to the callers any hard errors. 545 * 546 * If the driver does its own bookkeeping of stacked ports, it's not 547 * necessary to go through this helper. 548 */ 549 netdev_for_each_lower_dev(dev, lower_dev, iter) { 550 if (netif_is_bridge_master(lower_dev)) 551 continue; 552 553 err = __switchdev_handle_port_attr_set(lower_dev, port_attr_info, 554 check_cb, set_cb); 555 if (err && err != -EOPNOTSUPP) 556 return err; 557 } 558 559 return err; 560 } 561 562 int switchdev_handle_port_attr_set(struct net_device *dev, 563 struct switchdev_notifier_port_attr_info *port_attr_info, 564 bool (*check_cb)(const struct net_device *dev), 565 int (*set_cb)(struct net_device *dev, 566 const struct switchdev_attr *attr, 567 struct switchdev_trans *trans)) 568 { 569 int err; 570 571 err = __switchdev_handle_port_attr_set(dev, port_attr_info, check_cb, 572 set_cb); 573 if (err == -EOPNOTSUPP) 574 err = 0; 575 return err; 576 } 577 EXPORT_SYMBOL_GPL(switchdev_handle_port_attr_set); 578