drivers/net/team/team.c

   1 /*
   2  * net/drivers/team/team.c - Network team device driver
   3  * Copyright (c) 2011 Jiri Pirko <jpirko@redhat.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 2 of the License, or
   8  * (at your option) any later version.
   9  */
  10
  11 #include <linux/kernel.h>
  12 #include <linux/types.h>
  13 #include <linux/module.h>
  14 #include <linux/init.h>
  15 #include <linux/slab.h>
  16 #include <linux/rcupdate.h>
  17 #include <linux/errno.h>
  18 #include <linux/ctype.h>
  19 #include <linux/notifier.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/if_vlan.h>
  22 #include <linux/if_arp.h>
  23 #include <linux/socket.h>
  24 #include <linux/etherdevice.h>
  25 #include <linux/rtnetlink.h>
  26 #include <net/rtnetlink.h>
  27 #include <net/genetlink.h>
  28 #include <net/netlink.h>
  29 #include <linux/if_team.h>
  30
  31 #define DRV_NAME "team"
  32
  33
  34 /**********
  35  * Helpers
  36  **********/
  37
  38 #define team_port_exists(dev) (dev->priv_flags & IFF_TEAM_PORT)
  39
  40 static struct team_port *team_port_get_rcu(const struct net_device *dev)
  41 {
  42         struct team_port *port = rcu_dereference(dev->rx_handler_data);
  43
  44         return team_port_exists(dev) ? port : NULL;
  45 }
  46
  47 static struct team_port *team_port_get_rtnl(const struct net_device *dev)
  48 {
  49         struct team_port *port = rtnl_dereference(dev->rx_handler_data);
  50
  51         return team_port_exists(dev) ? port : NULL;
  52 }
  53
  54 /*
  55  * Since the ability to change mac address for open port device is tested in
  56  * team_port_add, this function can be called without control of return value
  57  */
  58 static int __set_port_mac(struct net_device *port_dev,
  59                           const unsigned char *dev_addr)
  60 {
  61         struct sockaddr addr;
  62
  63         memcpy(addr.sa_data, dev_addr, ETH_ALEN);
  64         addr.sa_family = ARPHRD_ETHER;
  65         return dev_set_mac_address(port_dev, &addr);
  66 }
  67
  68 int team_port_set_orig_mac(struct team_port *port)
  69 {
  70         return __set_port_mac(port->dev, port->orig.dev_addr);
  71 }
  72
  73 int team_port_set_team_mac(struct team_port *port)
  74 {
  75         return __set_port_mac(port->dev, port->team->dev->dev_addr);
  76 }
  77 EXPORT_SYMBOL(team_port_set_team_mac);
  78
  79 static void team_refresh_port_linkup(struct team_port *port)
  80 {
  81         port->linkup = port->user.linkup_enabled ? port->user.linkup :
  82                                                    port->state.linkup;
  83 }
  84
  85 /*******************
  86  * Options handling
  87  *******************/
  88
  89 struct team_option_inst { /* One for each option instance */
  90         struct list_head list;
  91         struct team_option *option;
  92         struct team_port *port; /* != NULL if per-port */
  93         bool changed;
  94         bool removed;
  95 };
  96
  97 static struct team_option *__team_find_option(struct team *team,
  98                                               const char *opt_name)
  99 {
 100         struct team_option *option;
 101
 102         list_for_each_entry(option, &team->option_list, list) {
 103                 if (strcmp(option->name, opt_name) == 0)
 104                         return option;
 105         }
 106         return NULL;
 107 }
 108
 109 static int __team_option_inst_add(struct team *team, struct team_option *option,
 110                                   struct team_port *port)
 111 {
 112         struct team_option_inst *opt_inst;
 113
 114         opt_inst = kmalloc(sizeof(*opt_inst), GFP_KERNEL);
 115         if (!opt_inst)
 116                 return -ENOMEM;
 117         opt_inst->option = option;
 118         opt_inst->port = port;
 119         opt_inst->changed = true;
 120         opt_inst->removed = false;
 121         list_add_tail(&opt_inst->list, &team->option_inst_list);
 122         return 0;
 123 }
 124
 125 static void __team_option_inst_del(struct team_option_inst *opt_inst)
 126 {
 127         list_del(&opt_inst->list);
 128         kfree(opt_inst);
 129 }
 130
 131 static void __team_option_inst_del_option(struct team *team,
 132                                           struct team_option *option)
 133 {
 134         struct team_option_inst *opt_inst, *tmp;
 135
 136         list_for_each_entry_safe(opt_inst, tmp, &team->option_inst_list, list) {
 137                 if (opt_inst->option == option)
 138                         __team_option_inst_del(opt_inst);
 139         }
 140 }
 141
 142 static int __team_option_inst_add_option(struct team *team,
 143                                          struct team_option *option)
 144 {
 145         struct team_port *port;
 146         int err;
 147
 148         if (!option->per_port)
 149                 return __team_option_inst_add(team, option, 0);
 150
 151         list_for_each_entry(port, &team->port_list, list) {
 152                 err = __team_option_inst_add(team, option, port);
 153                 if (err)
 154                         goto inst_del_option;
 155         }
 156         return 0;
 157
 158 inst_del_option:
 159         __team_option_inst_del_option(team, option);
 160         return err;
 161 }
 162
 163 static void __team_option_inst_mark_removed_option(struct team *team,
 164                                                    struct team_option *option)
 165 {
 166         struct team_option_inst *opt_inst;
 167
 168         list_for_each_entry(opt_inst, &team->option_inst_list, list) {
 169                 if (opt_inst->option == option) {
 170                         opt_inst->changed = true;
 171                         opt_inst->removed = true;
 172                 }
 173         }
 174 }
 175
 176 static void __team_option_inst_del_port(struct team *team,
 177                                         struct team_port *port)
 178 {
 179         struct team_option_inst *opt_inst, *tmp;
 180
 181         list_for_each_entry_safe(opt_inst, tmp, &team->option_inst_list, list) {
 182                 if (opt_inst->option->per_port &&
 183                     opt_inst->port == port)
 184                         __team_option_inst_del(opt_inst);
 185         }
 186 }
 187
 188 static int __team_option_inst_add_port(struct team *team,
 189                                        struct team_port *port)
 190 {
 191         struct team_option *option;
 192         int err;
 193
 194         list_for_each_entry(option, &team->option_list, list) {
 195                 if (!option->per_port)
 196                         continue;
 197                 err = __team_option_inst_add(team, option, port);
 198                 if (err)
 199                         goto inst_del_port;
 200         }
 201         return 0;
 202
 203 inst_del_port:
 204         __team_option_inst_del_port(team, port);
 205         return err;
 206 }
 207
 208 static void __team_option_inst_mark_removed_port(struct team *team,
 209                                                  struct team_port *port)
 210 {
 211         struct team_option_inst *opt_inst;
 212
 213         list_for_each_entry(opt_inst, &team->option_inst_list, list) {
 214                 if (opt_inst->port == port) {
 215                         opt_inst->changed = true;
 216                         opt_inst->removed = true;
 217                 }
 218         }
 219 }
 220
 221 static int __team_options_register(struct team *team,
 222                                    const struct team_option *option,
 223                                    size_t option_count)
 224 {
 225         int i;
 226         struct team_option **dst_opts;
 227         int err;
 228
 229         dst_opts = kzalloc(sizeof(struct team_option *) * option_count,
 230                            GFP_KERNEL);
 231         if (!dst_opts)
 232                 return -ENOMEM;
 233         for (i = 0; i < option_count; i++, option++) {
 234                 if (__team_find_option(team, option->name)) {
 235                         err = -EEXIST;
 236                         goto alloc_rollback;
 237                 }
 238                 dst_opts[i] = kmemdup(option, sizeof(*option), GFP_KERNEL);
 239                 if (!dst_opts[i]) {
 240                         err = -ENOMEM;
 241                         goto alloc_rollback;
 242                 }
 243         }
 244
 245         for (i = 0; i < option_count; i++) {
 246                 err = __team_option_inst_add_option(team, dst_opts[i]);
 247                 if (err)
 248                         goto inst_rollback;
 249                 list_add_tail(&dst_opts[i]->list, &team->option_list);
 250         }
 251
 252         kfree(dst_opts);
 253         return 0;
 254
 255 inst_rollback:
 256         for (i--; i >= 0; i--)
 257                 __team_option_inst_del_option(team, dst_opts[i]);
 258
 259         i = option_count - 1;
 260 alloc_rollback:
 261         for (i--; i >= 0; i--)
 262                 kfree(dst_opts[i]);
 263
 264         kfree(dst_opts);
 265         return err;
 266 }
 267
 268 static void __team_options_mark_removed(struct team *team,
 269                                         const struct team_option *option,
 270                                         size_t option_count)
 271 {
 272         int i;
 273
 274         for (i = 0; i < option_count; i++, option++) {
 275                 struct team_option *del_opt;
 276
 277                 del_opt = __team_find_option(team, option->name);
 278                 if (del_opt)
 279                         __team_option_inst_mark_removed_option(team, del_opt);
 280         }
 281 }
 282
 283 static void __team_options_unregister(struct team *team,
 284                                       const struct team_option *option,
 285                                       size_t option_count)
 286 {
 287         int i;
 288
 289         for (i = 0; i < option_count; i++, option++) {
 290                 struct team_option *del_opt;
 291
 292                 del_opt = __team_find_option(team, option->name);
 293                 if (del_opt) {
 294                         __team_option_inst_del_option(team, del_opt);
 295                         list_del(&del_opt->list);
 296                         kfree(del_opt);
 297                 }
 298         }
 299 }
 300
 301 static void __team_options_change_check(struct team *team);
 302
 303 int team_options_register(struct team *team,
 304                           const struct team_option *option,
 305                           size_t option_count)
 306 {
 307         int err;
 308
 309         err = __team_options_register(team, option, option_count);
 310         if (err)
 311                 return err;
 312         __team_options_change_check(team);
 313         return 0;
 314 }
 315 EXPORT_SYMBOL(team_options_register);
 316
 317 void team_options_unregister(struct team *team,
 318                              const struct team_option *option,
 319                              size_t option_count)
 320 {
 321         __team_options_mark_removed(team, option, option_count);
 322         __team_options_change_check(team);
 323         __team_options_unregister(team, option, option_count);
 324 }
 325 EXPORT_SYMBOL(team_options_unregister);
 326
 327 static int team_option_port_add(struct team *team, struct team_port *port)
 328 {
 329         int err;
 330
 331         err = __team_option_inst_add_port(team, port);
 332         if (err)
 333                 return err;
 334         __team_options_change_check(team);
 335         return 0;
 336 }
 337
 338 static void team_option_port_del(struct team *team, struct team_port *port)
 339 {
 340         __team_option_inst_mark_removed_port(team, port);
 341         __team_options_change_check(team);
 342         __team_option_inst_del_port(team, port);
 343 }
 344
 345 static int team_option_get(struct team *team,
 346                            struct team_option_inst *opt_inst,
 347                            struct team_gsetter_ctx *ctx)
 348 {
 349         return opt_inst->option->getter(team, ctx);
 350 }
 351
 352 static int team_option_set(struct team *team,
 353                            struct team_option_inst *opt_inst,
 354                            struct team_gsetter_ctx *ctx)
 355 {
 356         int err;
 357
 358         err = opt_inst->option->setter(team, ctx);
 359         if (err)
 360                 return err;
 361
 362         opt_inst->changed = true;
 363         __team_options_change_check(team);
 364         return err;
 365 }
 366
 367 /****************
 368  * Mode handling
 369  ****************/
 370
 371 static LIST_HEAD(mode_list);
 372 static DEFINE_SPINLOCK(mode_list_lock);
 373
 374 static struct team_mode *__find_mode(const char *kind)
 375 {
 376         struct team_mode *mode;
 377
 378         list_for_each_entry(mode, &mode_list, list) {
 379                 if (strcmp(mode->kind, kind) == 0)
 380                         return mode;
 381         }
 382         return NULL;
 383 }
 384
 385 static bool is_good_mode_name(const char *name)
 386 {
 387         while (*name != '\0') {
 388                 if (!isalpha(*name) && !isdigit(*name) && *name != '_')
 389                         return false;
 390                 name++;
 391         }
 392         return true;
 393 }
 394
 395 int team_mode_register(struct team_mode *mode)
 396 {
 397         int err = 0;
 398
 399         if (!is_good_mode_name(mode->kind) ||
 400             mode->priv_size > TEAM_MODE_PRIV_SIZE)
 401                 return -EINVAL;
 402         spin_lock(&mode_list_lock);
 403         if (__find_mode(mode->kind)) {
 404                 err = -EEXIST;
 405                 goto unlock;
 406         }
 407         list_add_tail(&mode->list, &mode_list);
 408 unlock:
 409         spin_unlock(&mode_list_lock);
 410         return err;
 411 }
 412 EXPORT_SYMBOL(team_mode_register);
 413
 414 int team_mode_unregister(struct team_mode *mode)
 415 {
 416         spin_lock(&mode_list_lock);
 417         list_del_init(&mode->list);
 418         spin_unlock(&mode_list_lock);
 419         return 0;
 420 }
 421 EXPORT_SYMBOL(team_mode_unregister);
 422
 423 static struct team_mode *team_mode_get(const char *kind)
 424 {
 425         struct team_mode *mode;
 426
 427         spin_lock(&mode_list_lock);
 428         mode = __find_mode(kind);
 429         if (!mode) {
 430                 spin_unlock(&mode_list_lock);
 431                 request_module("team-mode-%s", kind);
 432                 spin_lock(&mode_list_lock);
 433                 mode = __find_mode(kind);
 434         }
 435         if (mode)
 436                 if (!try_module_get(mode->owner))
 437                         mode = NULL;
 438
 439         spin_unlock(&mode_list_lock);
 440         return mode;
 441 }
 442
 443 static void team_mode_put(const struct team_mode *mode)
 444 {
 445         module_put(mode->owner);
 446 }
 447
 448 static bool team_dummy_transmit(struct team *team, struct sk_buff *skb)
 449 {
 450         dev_kfree_skb_any(skb);
 451         return false;
 452 }
 453
 454 rx_handler_result_t team_dummy_receive(struct team *team,
 455                                        struct team_port *port,
 456                                        struct sk_buff *skb)
 457 {
 458         return RX_HANDLER_ANOTHER;
 459 }
 460
 461 static void team_adjust_ops(struct team *team)
 462 {
 463         /*
 464          * To avoid checks in rx/tx skb paths, ensure here that non-null and
 465          * correct ops are always set.
 466          */
 467
 468         if (list_empty(&team->port_list) ||
 469             !team->mode || !team->mode->ops->transmit)
 470                 team->ops.transmit = team_dummy_transmit;
 471         else
 472                 team->ops.transmit = team->mode->ops->transmit;
 473
 474         if (list_empty(&team->port_list) ||
 475             !team->mode || !team->mode->ops->receive)
 476                 team->ops.receive = team_dummy_receive;
 477         else
 478                 team->ops.receive = team->mode->ops->receive;
 479 }
 480
 481 /*
 482  * We can benefit from the fact that it's ensured no port is present
 483  * at the time of mode change. Therefore no packets are in fly so there's no
 484  * need to set mode operations in any special way.
 485  */
 486 static int __team_change_mode(struct team *team,
 487                               const struct team_mode *new_mode)
 488 {
 489         /* Check if mode was previously set and do cleanup if so */
 490         if (team->mode) {
 491                 void (*exit_op)(struct team *team) = team->ops.exit;
 492
 493                 /* Clear ops area so no callback is called any longer */
 494                 memset(&team->ops, 0, sizeof(struct team_mode_ops));
 495                 team_adjust_ops(team);
 496
 497                 if (exit_op)
 498                         exit_op(team);
 499                 team_mode_put(team->mode);
 500                 team->mode = NULL;
 501                 /* zero private data area */
 502                 memset(&team->mode_priv, 0,
 503                        sizeof(struct team) - offsetof(struct team, mode_priv));
 504         }
 505
 506         if (!new_mode)
 507                 return 0;
 508
 509         if (new_mode->ops->init) {
 510                 int err;
 511
 512                 err = new_mode->ops->init(team);
 513                 if (err)
 514                         return err;
 515         }
 516
 517         team->mode = new_mode;
 518         memcpy(&team->ops, new_mode->ops, sizeof(struct team_mode_ops));
 519         team_adjust_ops(team);
 520
 521         return 0;
 522 }
 523
 524 static int team_change_mode(struct team *team, const char *kind)
 525 {
 526         struct team_mode *new_mode;
 527         struct net_device *dev = team->dev;
 528         int err;
 529
 530         if (!list_empty(&team->port_list)) {
 531                 netdev_err(dev, "No ports can be present during mode change\n");
 532                 return -EBUSY;
 533         }
 534
 535         if (team->mode && strcmp(team->mode->kind, kind) == 0) {
 536                 netdev_err(dev, "Unable to change to the same mode the team is in\n");
 537                 return -EINVAL;
 538         }
 539
 540         new_mode = team_mode_get(kind);
 541         if (!new_mode) {
 542                 netdev_err(dev, "Mode \"%s\" not found\n", kind);
 543                 return -EINVAL;
 544         }
 545
 546         err = __team_change_mode(team, new_mode);
 547         if (err) {
 548                 netdev_err(dev, "Failed to change to mode \"%s\"\n", kind);
 549                 team_mode_put(new_mode);
 550                 return err;
 551         }
 552
 553         netdev_info(dev, "Mode changed to \"%s\"\n", kind);
 554         return 0;
 555 }
 556
 557
 558 /************************
 559  * Rx path frame handler
 560  ************************/
 561
 562 /* note: already called with rcu_read_lock */
 563 static rx_handler_result_t team_handle_frame(struct sk_buff **pskb)
 564 {
 565         struct sk_buff *skb = *pskb;
 566         struct team_port *port;
 567         struct team *team;
 568         rx_handler_result_t res;
 569
 570         skb = skb_share_check(skb, GFP_ATOMIC);
 571         if (!skb)
 572                 return RX_HANDLER_CONSUMED;
 573
 574         *pskb = skb;
 575
 576         port = team_port_get_rcu(skb->dev);
 577         team = port->team;
 578
 579         res = team->ops.receive(team, port, skb);
 580         if (res == RX_HANDLER_ANOTHER) {
 581                 struct team_pcpu_stats *pcpu_stats;
 582
 583                 pcpu_stats = this_cpu_ptr(team->pcpu_stats);
 584                 u64_stats_update_begin(&pcpu_stats->syncp);
 585                 pcpu_stats->rx_packets++;
 586                 pcpu_stats->rx_bytes += skb->len;
 587                 if (skb->pkt_type == PACKET_MULTICAST)
 588                         pcpu_stats->rx_multicast++;
 589                 u64_stats_update_end(&pcpu_stats->syncp);
 590
 591                 skb->dev = team->dev;
 592         } else {
 593                 this_cpu_inc(team->pcpu_stats->rx_dropped);
 594         }
 595
 596         return res;
 597 }
 598
 599
 600 /****************
 601  * Port handling
 602  ****************/
 603
 604 static bool team_port_find(const struct team *team,
 605                            const struct team_port *port)
 606 {
 607         struct team_port *cur;
 608
 609         list_for_each_entry(cur, &team->port_list, list)
 610                 if (cur == port)
 611                         return true;
 612         return false;
 613 }
 614
 615 /*
 616  * Add/delete port to the team port list. Write guarded by rtnl_lock.
 617  * Takes care of correct port->index setup (might be racy).
 618  */
 619 static void team_port_list_add_port(struct team *team,
 620                                     struct team_port *port)
 621 {
 622         port->index = team->port_count++;
 623         hlist_add_head_rcu(&port->hlist,
 624                            team_port_index_hash(team, port->index));
 625         list_add_tail_rcu(&port->list, &team->port_list);
 626 }
 627
 628 static void __reconstruct_port_hlist(struct team *team, int rm_index)
 629 {
 630         int i;
 631         struct team_port *port;
 632
 633         for (i = rm_index + 1; i < team->port_count; i++) {
 634                 port = team_get_port_by_index(team, i);
 635                 hlist_del_rcu(&port->hlist);
 636                 port->index--;
 637                 hlist_add_head_rcu(&port->hlist,
 638                                    team_port_index_hash(team, port->index));
 639         }
 640 }
 641
 642 static void team_port_list_del_port(struct team *team,
 643                                    struct team_port *port)
 644 {
 645         int rm_index = port->index;
 646
 647         hlist_del_rcu(&port->hlist);
 648         list_del_rcu(&port->list);
 649         __reconstruct_port_hlist(team, rm_index);
 650         team->port_count--;
 651 }
 652
 653 #define TEAM_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
 654                             NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
 655                             NETIF_F_HIGHDMA | NETIF_F_LRO)
 656
 657 static void __team_compute_features(struct team *team)
 658 {
 659         struct team_port *port;
 660         u32 vlan_features = TEAM_VLAN_FEATURES;
 661         unsigned short max_hard_header_len = ETH_HLEN;
 662
 663         list_for_each_entry(port, &team->port_list, list) {
 664                 vlan_features = netdev_increment_features(vlan_features,
 665                                         port->dev->vlan_features,
 666                                         TEAM_VLAN_FEATURES);
 667
 668                 if (port->dev->hard_header_len > max_hard_header_len)
 669                         max_hard_header_len = port->dev->hard_header_len;
 670         }
 671
 672         team->dev->vlan_features = vlan_features;
 673         team->dev->hard_header_len = max_hard_header_len;
 674
 675         netdev_change_features(team->dev);
 676 }
 677
 678 static void team_compute_features(struct team *team)
 679 {
 680         mutex_lock(&team->lock);
 681         __team_compute_features(team);
 682         mutex_unlock(&team->lock);
 683 }
 684
 685 static int team_port_enter(struct team *team, struct team_port *port)
 686 {
 687         int err = 0;
 688
 689         dev_hold(team->dev);
 690         port->dev->priv_flags |= IFF_TEAM_PORT;
 691         if (team->ops.port_enter) {
 692                 err = team->ops.port_enter(team, port);
 693                 if (err) {
 694                         netdev_err(team->dev, "Device %s failed to enter team mode\n",
 695                                    port->dev->name);
 696                         goto err_port_enter;
 697                 }
 698         }
 699
 700         return 0;
 701
 702 err_port_enter:
 703         port->dev->priv_flags &= ~IFF_TEAM_PORT;
 704         dev_put(team->dev);
 705
 706         return err;
 707 }
 708
 709 static void team_port_leave(struct team *team, struct team_port *port)
 710 {
 711         if (team->ops.port_leave)
 712                 team->ops.port_leave(team, port);
 713         port->dev->priv_flags &= ~IFF_TEAM_PORT;
 714         dev_put(team->dev);
 715 }
 716
 717 static void __team_port_change_check(struct team_port *port, bool linkup);
 718
 719 static int team_port_add(struct team *team, struct net_device *port_dev)
 720 {
 721         struct net_device *dev = team->dev;
 722         struct team_port *port;
 723         char *portname = port_dev->name;
 724         int err;
 725
 726         if (port_dev->flags & IFF_LOOPBACK ||
 727             port_dev->type != ARPHRD_ETHER) {
 728                 netdev_err(dev, "Device %s is of an unsupported type\n",
 729                            portname);
 730                 return -EINVAL;
 731         }
 732
 733         if (team_port_exists(port_dev)) {
 734                 netdev_err(dev, "Device %s is already a port "
 735                                 "of a team device\n", portname);
 736                 return -EBUSY;
 737         }
 738
 739         if (port_dev->flags & IFF_UP) {
 740                 netdev_err(dev, "Device %s is up. Set it down before adding it as a team port\n",
 741                            portname);
 742                 return -EBUSY;
 743         }
 744
 745         port = kzalloc(sizeof(struct team_port), GFP_KERNEL);
 746         if (!port)
 747                 return -ENOMEM;
 748
 749         port->dev = port_dev;
 750         port->team = team;
 751
 752         port->orig.mtu = port_dev->mtu;
 753         err = dev_set_mtu(port_dev, dev->mtu);
 754         if (err) {
 755                 netdev_dbg(dev, "Error %d calling dev_set_mtu\n", err);
 756                 goto err_set_mtu;
 757         }
 758
 759         memcpy(port->orig.dev_addr, port_dev->dev_addr, ETH_ALEN);
 760
 761         err = team_port_enter(team, port);
 762         if (err) {
 763                 netdev_err(dev, "Device %s failed to enter team mode\n",
 764                            portname);
 765                 goto err_port_enter;
 766         }
 767
 768         err = dev_open(port_dev);
 769         if (err) {
 770                 netdev_dbg(dev, "Device %s opening failed\n",
 771                            portname);
 772                 goto err_dev_open;
 773         }
 774
 775         err = vlan_vids_add_by_dev(port_dev, dev);
 776         if (err) {
 777                 netdev_err(dev, "Failed to add vlan ids to device %s\n",
 778                                 portname);
 779                 goto err_vids_add;
 780         }
 781
 782         err = netdev_set_master(port_dev, dev);
 783         if (err) {
 784                 netdev_err(dev, "Device %s failed to set master\n", portname);
 785                 goto err_set_master;
 786         }
 787
 788         err = netdev_rx_handler_register(port_dev, team_handle_frame,
 789                                          port);
 790         if (err) {
 791                 netdev_err(dev, "Device %s failed to register rx_handler\n",
 792                            portname);
 793                 goto err_handler_register;
 794         }
 795
 796         err = team_option_port_add(team, port);
 797         if (err) {
 798                 netdev_err(dev, "Device %s failed to add per-port options\n",
 799                            portname);
 800                 goto err_option_port_add;
 801         }
 802
 803         team_port_list_add_port(team, port);
 804         team_adjust_ops(team);
 805         __team_compute_features(team);
 806         __team_port_change_check(port, !!netif_carrier_ok(port_dev));
 807
 808         netdev_info(dev, "Port device %s added\n", portname);
 809
 810         return 0;
 811
 812 err_option_port_add:
 813         netdev_rx_handler_unregister(port_dev);
 814
 815 err_handler_register:
 816         netdev_set_master(port_dev, NULL);
 817
 818 err_set_master:
 819         vlan_vids_del_by_dev(port_dev, dev);
 820
 821 err_vids_add:
 822         dev_close(port_dev);
 823
 824 err_dev_open:
 825         team_port_leave(team, port);
 826         team_port_set_orig_mac(port);
 827
 828 err_port_enter:
 829         dev_set_mtu(port_dev, port->orig.mtu);
 830
 831 err_set_mtu:
 832         kfree(port);
 833
 834         return err;
 835 }
 836
 837 static int team_port_del(struct team *team, struct net_device *port_dev)
 838 {
 839         struct net_device *dev = team->dev;
 840         struct team_port *port;
 841         char *portname = port_dev->name;
 842
 843         port = team_port_get_rtnl(port_dev);
 844         if (!port || !team_port_find(team, port)) {
 845                 netdev_err(dev, "Device %s does not act as a port of this team\n",
 846                            portname);
 847                 return -ENOENT;
 848         }
 849
 850         port->removed = true;
 851         __team_port_change_check(port, false);
 852         team_port_list_del_port(team, port);
 853         team_adjust_ops(team);
 854         team_option_port_del(team, port);
 855         netdev_rx_handler_unregister(port_dev);
 856         netdev_set_master(port_dev, NULL);
 857         vlan_vids_del_by_dev(port_dev, dev);
 858         dev_close(port_dev);
 859         team_port_leave(team, port);
 860         team_port_set_orig_mac(port);
 861         dev_set_mtu(port_dev, port->orig.mtu);
 862         synchronize_rcu();
 863         kfree(port);
 864         netdev_info(dev, "Port device %s removed\n", portname);
 865         __team_compute_features(team);
 866
 867         return 0;
 868 }
 869
 870
 871 /*****************
 872  * Net device ops
 873  *****************/
 874
 875 static const char team_no_mode_kind[] = "*NOMODE*";
 876
 877 static int team_mode_option_get(struct team *team, struct team_gsetter_ctx *ctx)
 878 {
 879         ctx->data.str_val = team->mode ? team->mode->kind : team_no_mode_kind;
 880         return 0;
 881 }
 882
 883 static int team_mode_option_set(struct team *team, struct team_gsetter_ctx *ctx)
 884 {
 885         return team_change_mode(team, ctx->data.str_val);
 886 }
 887
 888 static int team_user_linkup_option_get(struct team *team,
 889                                        struct team_gsetter_ctx *ctx)
 890 {
 891         ctx->data.bool_val = ctx->port->user.linkup;
 892         return 0;
 893 }
 894
 895 static int team_user_linkup_option_set(struct team *team,
 896                                        struct team_gsetter_ctx *ctx)
 897 {
 898         ctx->port->user.linkup = ctx->data.bool_val;
 899         team_refresh_port_linkup(ctx->port);
 900         return 0;
 901 }
 902
 903 static int team_user_linkup_en_option_get(struct team *team,
 904                                           struct team_gsetter_ctx *ctx)
 905 {
 906         struct team_port *port = ctx->port;
 907
 908         ctx->data.bool_val = port->user.linkup_enabled;
 909         return 0;
 910 }
 911
 912 static int team_user_linkup_en_option_set(struct team *team,
 913                                           struct team_gsetter_ctx *ctx)
 914 {
 915         struct team_port *port = ctx->port;
 916
 917         port->user.linkup_enabled = ctx->data.bool_val;
 918         team_refresh_port_linkup(ctx->port);
 919         return 0;
 920 }
 921
 922 static const struct team_option team_options[] = {
 923         {
 924                 .name = "mode",
 925                 .type = TEAM_OPTION_TYPE_STRING,
 926                 .getter = team_mode_option_get,
 927                 .setter = team_mode_option_set,
 928         },
 929         {
 930                 .name = "user_linkup",
 931                 .type = TEAM_OPTION_TYPE_BOOL,
 932                 .per_port = true,
 933                 .getter = team_user_linkup_option_get,
 934                 .setter = team_user_linkup_option_set,
 935         },
 936         {
 937                 .name = "user_linkup_enabled",
 938                 .type = TEAM_OPTION_TYPE_BOOL,
 939                 .per_port = true,
 940                 .getter = team_user_linkup_en_option_get,
 941                 .setter = team_user_linkup_en_option_set,
 942         },
 943 };
 944
 945 static int team_init(struct net_device *dev)
 946 {
 947         struct team *team = netdev_priv(dev);
 948         int i;
 949         int err;
 950
 951         team->dev = dev;
 952         mutex_init(&team->lock);
 953
 954         team->pcpu_stats = alloc_percpu(struct team_pcpu_stats);
 955         if (!team->pcpu_stats)
 956                 return -ENOMEM;
 957
 958         for (i = 0; i < TEAM_PORT_HASHENTRIES; i++)
 959                 INIT_HLIST_HEAD(&team->port_hlist[i]);
 960         INIT_LIST_HEAD(&team->port_list);
 961
 962         team_adjust_ops(team);
 963
 964         INIT_LIST_HEAD(&team->option_list);
 965         INIT_LIST_HEAD(&team->option_inst_list);
 966         err = team_options_register(team, team_options, ARRAY_SIZE(team_options));
 967         if (err)
 968                 goto err_options_register;
 969         netif_carrier_off(dev);
 970
 971         return 0;
 972
 973 err_options_register:
 974         free_percpu(team->pcpu_stats);
 975
 976         return err;
 977 }
 978
 979 static void team_uninit(struct net_device *dev)
 980 {
 981         struct team *team = netdev_priv(dev);
 982         struct team_port *port;
 983         struct team_port *tmp;
 984
 985         mutex_lock(&team->lock);
 986         list_for_each_entry_safe(port, tmp, &team->port_list, list)
 987                 team_port_del(team, port->dev);
 988
 989         __team_change_mode(team, NULL); /* cleanup */
 990         __team_options_unregister(team, team_options, ARRAY_SIZE(team_options));
 991         mutex_unlock(&team->lock);
 992 }
 993
 994 static void team_destructor(struct net_device *dev)
 995 {
 996         struct team *team = netdev_priv(dev);
 997
 998         free_percpu(team->pcpu_stats);
 999         free_netdev(dev);
1000 }
1001
1002 static int team_open(struct net_device *dev)
1003 {
1004         netif_carrier_on(dev);
1005         return 0;
1006 }
1007
1008 static int team_close(struct net_device *dev)
1009 {
1010         netif_carrier_off(dev);
1011         return 0;
1012 }
1013
1014 /*
1015  * note: already called with rcu_read_lock
1016  */
1017 static netdev_tx_t team_xmit(struct sk_buff *skb, struct net_device *dev)
1018 {
1019         struct team *team = netdev_priv(dev);
1020         bool tx_success = false;
1021         unsigned int len = skb->len;
1022
1023         tx_success = team->ops.transmit(team, skb);
1024         if (tx_success) {
1025                 struct team_pcpu_stats *pcpu_stats;
1026
1027                 pcpu_stats = this_cpu_ptr(team->pcpu_stats);
1028                 u64_stats_update_begin(&pcpu_stats->syncp);
1029                 pcpu_stats->tx_packets++;
1030                 pcpu_stats->tx_bytes += len;
1031                 u64_stats_update_end(&pcpu_stats->syncp);
1032         } else {
1033                 this_cpu_inc(team->pcpu_stats->tx_dropped);
1034         }
1035
1036         return NETDEV_TX_OK;
1037 }
1038
1039 static void team_change_rx_flags(struct net_device *dev, int change)
1040 {
1041         struct team *team = netdev_priv(dev);
1042         struct team_port *port;
1043         int inc;
1044
1045         rcu_read_lock();
1046         list_for_each_entry_rcu(port, &team->port_list, list) {
1047                 if (change & IFF_PROMISC) {
1048                         inc = dev->flags & IFF_PROMISC ? 1 : -1;
1049                         dev_set_promiscuity(port->dev, inc);
1050                 }
1051                 if (change & IFF_ALLMULTI) {
1052                         inc = dev->flags & IFF_ALLMULTI ? 1 : -1;
1053                         dev_set_allmulti(port->dev, inc);
1054                 }
1055         }
1056         rcu_read_unlock();
1057 }
1058
1059 static void team_set_rx_mode(struct net_device *dev)
1060 {
1061         struct team *team = netdev_priv(dev);
1062         struct team_port *port;
1063
1064         rcu_read_lock();
1065         list_for_each_entry_rcu(port, &team->port_list, list) {
1066                 dev_uc_sync(port->dev, dev);
1067                 dev_mc_sync(port->dev, dev);
1068         }
1069         rcu_read_unlock();
1070 }
1071
1072 static int team_set_mac_address(struct net_device *dev, void *p)
1073 {
1074         struct team *team = netdev_priv(dev);
1075         struct team_port *port;
1076         struct sockaddr *addr = p;
1077
1078         dev->addr_assign_type &= ~NET_ADDR_RANDOM;
1079         memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
1080         rcu_read_lock();
1081         list_for_each_entry_rcu(port, &team->port_list, list)
1082                 if (team->ops.port_change_mac)
1083                         team->ops.port_change_mac(team, port);
1084         rcu_read_unlock();
1085         return 0;
1086 }
1087
1088 static int team_change_mtu(struct net_device *dev, int new_mtu)
1089 {
1090         struct team *team = netdev_priv(dev);
1091         struct team_port *port;
1092         int err;
1093
1094         /*
1095          * Alhough this is reader, it's guarded by team lock. It's not possible
1096          * to traverse list in reverse under rcu_read_lock
1097          */
1098         mutex_lock(&team->lock);
1099         list_for_each_entry(port, &team->port_list, list) {
1100                 err = dev_set_mtu(port->dev, new_mtu);
1101                 if (err) {
1102                         netdev_err(dev, "Device %s failed to change mtu",
1103                                    port->dev->name);
1104                         goto unwind;
1105                 }
1106         }
1107         mutex_unlock(&team->lock);
1108
1109         dev->mtu = new_mtu;
1110
1111         return 0;
1112
1113 unwind:
1114         list_for_each_entry_continue_reverse(port, &team->port_list, list)
1115                 dev_set_mtu(port->dev, dev->mtu);
1116         mutex_unlock(&team->lock);
1117
1118         return err;
1119 }
1120
1121 static struct rtnl_link_stats64 *
1122 team_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1123 {
1124         struct team *team = netdev_priv(dev);
1125         struct team_pcpu_stats *p;
1126         u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
1127         u32 rx_dropped = 0, tx_dropped = 0;
1128         unsigned int start;
1129         int i;
1130
1131         for_each_possible_cpu(i) {
1132                 p = per_cpu_ptr(team->pcpu_stats, i);
1133                 do {
1134                         start = u64_stats_fetch_begin_bh(&p->syncp);
1135                         rx_packets      = p->rx_packets;
1136                         rx_bytes        = p->rx_bytes;
1137                         rx_multicast    = p->rx_multicast;
1138                         tx_packets      = p->tx_packets;
1139                         tx_bytes        = p->tx_bytes;
1140                 } while (u64_stats_fetch_retry_bh(&p->syncp, start));
1141
1142                 stats->rx_packets       += rx_packets;
1143                 stats->rx_bytes         += rx_bytes;
1144                 stats->multicast        += rx_multicast;
1145                 stats->tx_packets       += tx_packets;
1146                 stats->tx_bytes         += tx_bytes;
1147                 /*
1148                  * rx_dropped & tx_dropped are u32, updated
1149                  * without syncp protection.
1150                  */
1151                 rx_dropped      += p->rx_dropped;
1152                 tx_dropped      += p->tx_dropped;
1153         }
1154         stats->rx_dropped       = rx_dropped;
1155         stats->tx_dropped       = tx_dropped;
1156         return stats;
1157 }
1158
1159 static int team_vlan_rx_add_vid(struct net_device *dev, uint16_t vid)
1160 {
1161         struct team *team = netdev_priv(dev);
1162         struct team_port *port;
1163         int err;
1164
1165         /*
1166          * Alhough this is reader, it's guarded by team lock. It's not possible
1167          * to traverse list in reverse under rcu_read_lock
1168          */
1169         mutex_lock(&team->lock);
1170         list_for_each_entry(port, &team->port_list, list) {
1171                 err = vlan_vid_add(port->dev, vid);
1172                 if (err)
1173                         goto unwind;
1174         }
1175         mutex_unlock(&team->lock);
1176
1177         return 0;
1178
1179 unwind:
1180         list_for_each_entry_continue_reverse(port, &team->port_list, list)
1181                 vlan_vid_del(port->dev, vid);
1182         mutex_unlock(&team->lock);
1183
1184         return err;
1185 }
1186
1187 static int team_vlan_rx_kill_vid(struct net_device *dev, uint16_t vid)
1188 {
1189         struct team *team = netdev_priv(dev);
1190         struct team_port *port;
1191
1192         rcu_read_lock();
1193         list_for_each_entry_rcu(port, &team->port_list, list)
1194                 vlan_vid_del(port->dev, vid);
1195         rcu_read_unlock();
1196
1197         return 0;
1198 }
1199
1200 static int team_add_slave(struct net_device *dev, struct net_device *port_dev)
1201 {
1202         struct team *team = netdev_priv(dev);
1203         int err;
1204
1205         mutex_lock(&team->lock);
1206         err = team_port_add(team, port_dev);
1207         mutex_unlock(&team->lock);
1208         return err;
1209 }
1210
1211 static int team_del_slave(struct net_device *dev, struct net_device *port_dev)
1212 {
1213         struct team *team = netdev_priv(dev);
1214         int err;
1215
1216         mutex_lock(&team->lock);
1217         err = team_port_del(team, port_dev);
1218         mutex_unlock(&team->lock);
1219         return err;
1220 }
1221
1222 static netdev_features_t team_fix_features(struct net_device *dev,
1223                                            netdev_features_t features)
1224 {
1225         struct team_port *port;
1226         struct team *team = netdev_priv(dev);
1227         netdev_features_t mask;
1228
1229         mask = features;
1230         features &= ~NETIF_F_ONE_FOR_ALL;
1231         features |= NETIF_F_ALL_FOR_ALL;
1232
1233         rcu_read_lock();
1234         list_for_each_entry_rcu(port, &team->port_list, list) {
1235                 features = netdev_increment_features(features,
1236                                                      port->dev->features,
1237                                                      mask);
1238         }
1239         rcu_read_unlock();
1240         return features;
1241 }
1242
1243 static const struct net_device_ops team_netdev_ops = {
1244         .ndo_init               = team_init,
1245         .ndo_uninit             = team_uninit,
1246         .ndo_open               = team_open,
1247         .ndo_stop               = team_close,
1248         .ndo_start_xmit         = team_xmit,
1249         .ndo_change_rx_flags    = team_change_rx_flags,
1250         .ndo_set_rx_mode        = team_set_rx_mode,
1251         .ndo_set_mac_address    = team_set_mac_address,
1252         .ndo_change_mtu         = team_change_mtu,
1253         .ndo_get_stats64        = team_get_stats64,
1254         .ndo_vlan_rx_add_vid    = team_vlan_rx_add_vid,
1255         .ndo_vlan_rx_kill_vid   = team_vlan_rx_kill_vid,
1256         .ndo_add_slave          = team_add_slave,
1257         .ndo_del_slave          = team_del_slave,
1258         .ndo_fix_features       = team_fix_features,
1259 };
1260
1261
1262 /***********************
1263  * rt netlink interface
1264  ***********************/
1265
1266 static void team_setup(struct net_device *dev)
1267 {
1268         ether_setup(dev);
1269
1270         dev->netdev_ops = &team_netdev_ops;
1271         dev->destructor = team_destructor;
1272         dev->tx_queue_len = 0;
1273         dev->flags |= IFF_MULTICAST;
1274         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
1275
1276         /*
1277          * Indicate we support unicast address filtering. That way core won't
1278          * bring us to promisc mode in case a unicast addr is added.
1279          * Let this up to underlay drivers.
1280          */
1281         dev->priv_flags |= IFF_UNICAST_FLT;
1282
1283         dev->features |= NETIF_F_LLTX;
1284         dev->features |= NETIF_F_GRO;
1285         dev->hw_features = NETIF_F_HW_VLAN_TX |
1286                            NETIF_F_HW_VLAN_RX |
1287                            NETIF_F_HW_VLAN_FILTER;
1288
1289         dev->features |= dev->hw_features;
1290 }
1291
1292 static int team_newlink(struct net *src_net, struct net_device *dev,
1293                         struct nlattr *tb[], struct nlattr *data[])
1294 {
1295         int err;
1296
1297         if (tb[IFLA_ADDRESS] == NULL)
1298                 eth_hw_addr_random(dev);
1299
1300         err = register_netdevice(dev);
1301         if (err)
1302                 return err;
1303
1304         return 0;
1305 }
1306
1307 static int team_validate(struct nlattr *tb[], struct nlattr *data[])
1308 {
1309         if (tb[IFLA_ADDRESS]) {
1310                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1311                         return -EINVAL;
1312                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1313                         return -EADDRNOTAVAIL;
1314         }
1315         return 0;
1316 }
1317
1318 static struct rtnl_link_ops team_link_ops __read_mostly = {
1319         .kind           = DRV_NAME,
1320         .priv_size      = sizeof(struct team),
1321         .setup          = team_setup,
1322         .newlink        = team_newlink,
1323         .validate       = team_validate,
1324 };
1325
1326
1327 /***********************************
1328  * Generic netlink custom interface
1329  ***********************************/
1330
1331 static struct genl_family team_nl_family = {
1332         .id             = GENL_ID_GENERATE,
1333         .name           = TEAM_GENL_NAME,
1334         .version        = TEAM_GENL_VERSION,
1335         .maxattr        = TEAM_ATTR_MAX,
1336         .netnsok        = true,
1337 };
1338
1339 static const struct nla_policy team_nl_policy[TEAM_ATTR_MAX + 1] = {
1340         [TEAM_ATTR_UNSPEC]                      = { .type = NLA_UNSPEC, },
1341         [TEAM_ATTR_TEAM_IFINDEX]                = { .type = NLA_U32 },
1342         [TEAM_ATTR_LIST_OPTION]                 = { .type = NLA_NESTED },
1343         [TEAM_ATTR_LIST_PORT]                   = { .type = NLA_NESTED },
1344 };
1345
1346 static const struct nla_policy
1347 team_nl_option_policy[TEAM_ATTR_OPTION_MAX + 1] = {
1348         [TEAM_ATTR_OPTION_UNSPEC]               = { .type = NLA_UNSPEC, },
1349         [TEAM_ATTR_OPTION_NAME] = {
1350                 .type = NLA_STRING,
1351                 .len = TEAM_STRING_MAX_LEN,
1352         },
1353         [TEAM_ATTR_OPTION_CHANGED]              = { .type = NLA_FLAG },
1354         [TEAM_ATTR_OPTION_TYPE]                 = { .type = NLA_U8 },
1355         [TEAM_ATTR_OPTION_DATA]                 = { .type = NLA_BINARY },
1356 };
1357
1358 static int team_nl_cmd_noop(struct sk_buff *skb, struct genl_info *info)
1359 {
1360         struct sk_buff *msg;
1361         void *hdr;
1362         int err;
1363
1364         msg = nlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1365         if (!msg)
1366                 return -ENOMEM;
1367
1368         hdr = genlmsg_put(msg, info->snd_pid, info->snd_seq,
1369                           &team_nl_family, 0, TEAM_CMD_NOOP);
1370         if (IS_ERR(hdr)) {
1371                 err = PTR_ERR(hdr);
1372                 goto err_msg_put;
1373         }
1374
1375         genlmsg_end(msg, hdr);
1376
1377         return genlmsg_unicast(genl_info_net(info), msg, info->snd_pid);
1378
1379 err_msg_put:
1380         nlmsg_free(msg);
1381
1382         return err;
1383 }
1384
1385 /*
1386  * Netlink cmd functions should be locked by following two functions.
1387  * Since dev gets held here, that ensures dev won't disappear in between.
1388  */
1389 static struct team *team_nl_team_get(struct genl_info *info)
1390 {
1391         struct net *net = genl_info_net(info);
1392         int ifindex;
1393         struct net_device *dev;
1394         struct team *team;
1395
1396         if (!info->attrs[TEAM_ATTR_TEAM_IFINDEX])
1397                 return NULL;
1398
1399         ifindex = nla_get_u32(info->attrs[TEAM_ATTR_TEAM_IFINDEX]);
1400         dev = dev_get_by_index(net, ifindex);
1401         if (!dev || dev->netdev_ops != &team_netdev_ops) {
1402                 if (dev)
1403                         dev_put(dev);
1404                 return NULL;
1405         }
1406
1407         team = netdev_priv(dev);
1408         mutex_lock(&team->lock);
1409         return team;
1410 }
1411
1412 static void team_nl_team_put(struct team *team)
1413 {
1414         mutex_unlock(&team->lock);
1415         dev_put(team->dev);
1416 }
1417
1418 static int team_nl_send_generic(struct genl_info *info, struct team *team,
1419                                 int (*fill_func)(struct sk_buff *skb,
1420                                                  struct genl_info *info,
1421                                                  int flags, struct team *team))
1422 {
1423         struct sk_buff *skb;
1424         int err;
1425
1426         skb = nlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1427         if (!skb)
1428                 return -ENOMEM;
1429
1430         err = fill_func(skb, info, NLM_F_ACK, team);
1431         if (err < 0)
1432                 goto err_fill;
1433
1434         err = genlmsg_unicast(genl_info_net(info), skb, info->snd_pid);
1435         return err;
1436
1437 err_fill:
1438         nlmsg_free(skb);
1439         return err;
1440 }
1441
1442 static int team_nl_fill_options_get(struct sk_buff *skb,
1443                                     u32 pid, u32 seq, int flags,
1444                                     struct team *team, bool fillall)
1445 {
1446         struct nlattr *option_list;
1447         void *hdr;
1448         struct team_option_inst *opt_inst;
1449         int err;
1450
1451         hdr = genlmsg_put(skb, pid, seq, &team_nl_family, flags,
1452                           TEAM_CMD_OPTIONS_GET);
1453         if (IS_ERR(hdr))
1454                 return PTR_ERR(hdr);
1455
1456         if (nla_put_u32(skb, TEAM_ATTR_TEAM_IFINDEX, team->dev->ifindex))
1457                 goto nla_put_failure;
1458         option_list = nla_nest_start(skb, TEAM_ATTR_LIST_OPTION);
1459         if (!option_list)
1460                 return -EMSGSIZE;
1461
1462         list_for_each_entry(opt_inst, &team->option_inst_list, list) {
1463                 struct nlattr *option_item;
1464                 struct team_option *option = opt_inst->option;
1465                 struct team_gsetter_ctx ctx;
1466
1467                 /* Include only changed options if fill all mode is not on */
1468                 if (!fillall && !opt_inst->changed)
1469                         continue;
1470                 option_item = nla_nest_start(skb, TEAM_ATTR_ITEM_OPTION);
1471                 if (!option_item)
1472                         goto nla_put_failure;
1473                 if (nla_put_string(skb, TEAM_ATTR_OPTION_NAME, option->name))
1474                         goto nla_put_failure;
1475                 if (opt_inst->changed) {
1476                         if (nla_put_flag(skb, TEAM_ATTR_OPTION_CHANGED))
1477                                 goto nla_put_failure;
1478                         opt_inst->changed = false;
1479                 }
1480                 if (opt_inst->removed &&
1481                     nla_put_flag(skb, TEAM_ATTR_OPTION_REMOVED))
1482                         goto nla_put_failure;
1483                 if (opt_inst->port &&
1484                     nla_put_u32(skb, TEAM_ATTR_OPTION_PORT_IFINDEX,
1485                                 opt_inst->port->dev->ifindex))
1486                         goto nla_put_failure;
1487                 ctx.port = opt_inst->port;
1488                 switch (option->type) {
1489                 case TEAM_OPTION_TYPE_U32:
1490                         if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_U32))
1491                                 goto nla_put_failure;
1492                         err = team_option_get(team, opt_inst, &ctx);
1493                         if (err)
1494                                 goto errout;
1495                         if (nla_put_u32(skb, TEAM_ATTR_OPTION_DATA,
1496                                         ctx.data.u32_val))
1497                                 goto nla_put_failure;
1498                         break;
1499                 case TEAM_OPTION_TYPE_STRING:
1500                         if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_STRING))
1501                                 goto nla_put_failure;
1502                         err = team_option_get(team, opt_inst, &ctx);
1503                         if (err)
1504                                 goto errout;
1505                         if (nla_put_string(skb, TEAM_ATTR_OPTION_DATA,
1506                                            ctx.data.str_val))
1507                                 goto nla_put_failure;
1508                         break;
1509                 case TEAM_OPTION_TYPE_BINARY:
1510                         if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_BINARY))
1511                                 goto nla_put_failure;
1512                         err = team_option_get(team, opt_inst, &ctx);
1513                         if (err)
1514                                 goto errout;
1515                         if (nla_put(skb, TEAM_ATTR_OPTION_DATA,
1516                                     ctx.data.bin_val.len, ctx.data.bin_val.ptr))
1517                                 goto nla_put_failure;
1518                         break;
1519                 case TEAM_OPTION_TYPE_BOOL:
1520                         if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_FLAG))
1521                                 goto nla_put_failure;
1522                         err = team_option_get(team, opt_inst, &ctx);
1523                         if (err)
1524                                 goto errout;
1525                         if (ctx.data.bool_val &&
1526                             nla_put_flag(skb, TEAM_ATTR_OPTION_DATA))
1527                                 goto nla_put_failure;
1528                         break;
1529                 default:
1530                         BUG();
1531                 }
1532                 nla_nest_end(skb, option_item);
1533         }
1534
1535         nla_nest_end(skb, option_list);
1536         return genlmsg_end(skb, hdr);
1537
1538 nla_put_failure:
1539         err = -EMSGSIZE;
1540 errout:
1541         genlmsg_cancel(skb, hdr);
1542         return err;
1543 }
1544
1545 static int team_nl_fill_options_get_all(struct sk_buff *skb,
1546                                         struct genl_info *info, int flags,
1547                                         struct team *team)
1548 {
1549         return team_nl_fill_options_get(skb, info->snd_pid,
1550                                         info->snd_seq, NLM_F_ACK,
1551                                         team, true);
1552 }
1553
1554 static int team_nl_cmd_options_get(struct sk_buff *skb, struct genl_info *info)
1555 {
1556         struct team *team;
1557         int err;
1558
1559         team = team_nl_team_get(info);
1560         if (!team)
1561                 return -EINVAL;
1562
1563         err = team_nl_send_generic(info, team, team_nl_fill_options_get_all);
1564
1565         team_nl_team_put(team);
1566
1567         return err;
1568 }
1569
1570 static int team_nl_cmd_options_set(struct sk_buff *skb, struct genl_info *info)
1571 {
1572         struct team *team;
1573         int err = 0;
1574         int i;
1575         struct nlattr *nl_option;
1576
1577         team = team_nl_team_get(info);
1578         if (!team)
1579                 return -EINVAL;
1580
1581         err = -EINVAL;
1582         if (!info->attrs[TEAM_ATTR_LIST_OPTION]) {
1583                 err = -EINVAL;
1584                 goto team_put;
1585         }
1586
1587         nla_for_each_nested(nl_option, info->attrs[TEAM_ATTR_LIST_OPTION], i) {
1588                 struct nlattr *opt_attrs[TEAM_ATTR_OPTION_MAX + 1];
1589                 struct nlattr *attr_port_ifindex;
1590                 struct nlattr *attr_data;
1591                 enum team_option_type opt_type;
1592                 int opt_port_ifindex = 0; /* != 0 for per-port options */
1593                 struct team_option_inst *opt_inst;
1594                 char *opt_name;
1595                 bool opt_found = false;
1596
1597                 if (nla_type(nl_option) != TEAM_ATTR_ITEM_OPTION) {
1598                         err = -EINVAL;
1599                         goto team_put;
1600                 }
1601                 err = nla_parse_nested(opt_attrs, TEAM_ATTR_OPTION_MAX,
1602                                        nl_option, team_nl_option_policy);
1603                 if (err)
1604                         goto team_put;
1605                 if (!opt_attrs[TEAM_ATTR_OPTION_NAME] ||
1606                     !opt_attrs[TEAM_ATTR_OPTION_TYPE]) {
1607                         err = -EINVAL;
1608                         goto team_put;
1609                 }
1610                 switch (nla_get_u8(opt_attrs[TEAM_ATTR_OPTION_TYPE])) {
1611                 case NLA_U32:
1612                         opt_type = TEAM_OPTION_TYPE_U32;
1613                         break;
1614                 case NLA_STRING:
1615                         opt_type = TEAM_OPTION_TYPE_STRING;
1616                         break;
1617                 case NLA_BINARY:
1618                         opt_type = TEAM_OPTION_TYPE_BINARY;
1619                         break;
1620                 case NLA_FLAG:
1621                         opt_type = TEAM_OPTION_TYPE_BOOL;
1622                         break;
1623                 default:
1624                         goto team_put;
1625                 }
1626
1627                 attr_data = opt_attrs[TEAM_ATTR_OPTION_DATA];
1628                 if (opt_type != TEAM_OPTION_TYPE_BOOL && !attr_data) {
1629                         err = -EINVAL;
1630                         goto team_put;
1631                 }
1632
1633                 opt_name = nla_data(opt_attrs[TEAM_ATTR_OPTION_NAME]);
1634                 attr_port_ifindex = opt_attrs[TEAM_ATTR_OPTION_PORT_IFINDEX];
1635                 if (attr_port_ifindex)
1636                         opt_port_ifindex = nla_get_u32(attr_port_ifindex);
1637
1638                 list_for_each_entry(opt_inst, &team->option_inst_list, list) {
1639                         struct team_option *option = opt_inst->option;
1640                         struct team_gsetter_ctx ctx;
1641                         int tmp_ifindex;
1642
1643                         tmp_ifindex = opt_inst->port ?
1644                                       opt_inst->port->dev->ifindex : 0;
1645                         if (option->type != opt_type ||
1646                             strcmp(option->name, opt_name) ||
1647                             tmp_ifindex != opt_port_ifindex)
1648                                 continue;
1649                         opt_found = true;
1650                         ctx.port = opt_inst->port;
1651                         switch (opt_type) {
1652                         case TEAM_OPTION_TYPE_U32:
1653                                 ctx.data.u32_val = nla_get_u32(attr_data);
1654                                 break;
1655                         case TEAM_OPTION_TYPE_STRING:
1656                                 if (nla_len(attr_data) > TEAM_STRING_MAX_LEN) {
1657                                         err = -EINVAL;
1658                                         goto team_put;
1659                                 }
1660                                 ctx.data.str_val = nla_data(attr_data);
1661                                 break;
1662                         case TEAM_OPTION_TYPE_BINARY:
1663                                 ctx.data.bin_val.len = nla_len(attr_data);
1664                                 ctx.data.bin_val.ptr = nla_data(attr_data);
1665                                 break;
1666                         case TEAM_OPTION_TYPE_BOOL:
1667                                 ctx.data.bool_val = attr_data ? true : false;
1668                                 break;
1669                         default:
1670                                 BUG();
1671                         }
1672                         err = team_option_set(team, opt_inst, &ctx);
1673                         if (err)
1674                                 goto team_put;
1675                 }
1676                 if (!opt_found) {
1677                         err = -ENOENT;
1678                         goto team_put;
1679                 }
1680         }
1681
1682 team_put:
1683         team_nl_team_put(team);
1684
1685         return err;
1686 }
1687
1688 static int team_nl_fill_port_list_get(struct sk_buff *skb,
1689                                       u32 pid, u32 seq, int flags,
1690                                       struct team *team,
1691                                       bool fillall)
1692 {
1693         struct nlattr *port_list;
1694         void *hdr;
1695         struct team_port *port;
1696
1697         hdr = genlmsg_put(skb, pid, seq, &team_nl_family, flags,
1698                           TEAM_CMD_PORT_LIST_GET);
1699         if (IS_ERR(hdr))
1700                 return PTR_ERR(hdr);
1701
1702         if (nla_put_u32(skb, TEAM_ATTR_TEAM_IFINDEX, team->dev->ifindex))
1703                 goto nla_put_failure;
1704         port_list = nla_nest_start(skb, TEAM_ATTR_LIST_PORT);
1705         if (!port_list)
1706                 return -EMSGSIZE;
1707
1708         list_for_each_entry(port, &team->port_list, list) {
1709                 struct nlattr *port_item;
1710
1711                 /* Include only changed ports if fill all mode is not on */
1712                 if (!fillall && !port->changed)
1713                         continue;
1714                 port_item = nla_nest_start(skb, TEAM_ATTR_ITEM_PORT);
1715                 if (!port_item)
1716                         goto nla_put_failure;
1717                 if (nla_put_u32(skb, TEAM_ATTR_PORT_IFINDEX, port->dev->ifindex))
1718                         goto nla_put_failure;
1719                 if (port->changed) {
1720                         if (nla_put_flag(skb, TEAM_ATTR_PORT_CHANGED))
1721                                 goto nla_put_failure;
1722                         port->changed = false;
1723                 }
1724                 if ((port->removed &&
1725                      nla_put_flag(skb, TEAM_ATTR_PORT_REMOVED)) ||
1726                     (port->state.linkup &&
1727                      nla_put_flag(skb, TEAM_ATTR_PORT_LINKUP)) ||
1728                     nla_put_u32(skb, TEAM_ATTR_PORT_SPEED, port->state.speed) ||
1729                     nla_put_u8(skb, TEAM_ATTR_PORT_DUPLEX, port->state.duplex))
1730                         goto nla_put_failure;
1731                 nla_nest_end(skb, port_item);
1732         }
1733
1734         nla_nest_end(skb, port_list);
1735         return genlmsg_end(skb, hdr);
1736
1737 nla_put_failure:
1738         genlmsg_cancel(skb, hdr);
1739         return -EMSGSIZE;
1740 }
1741
1742 static int team_nl_fill_port_list_get_all(struct sk_buff *skb,
1743                                           struct genl_info *info, int flags,
1744                                           struct team *team)
1745 {
1746         return team_nl_fill_port_list_get(skb, info->snd_pid,
1747                                           info->snd_seq, NLM_F_ACK,
1748                                           team, true);
1749 }
1750
1751 static int team_nl_cmd_port_list_get(struct sk_buff *skb,
1752                                      struct genl_info *info)
1753 {
1754         struct team *team;
1755         int err;
1756
1757         team = team_nl_team_get(info);
1758         if (!team)
1759                 return -EINVAL;
1760
1761         err = team_nl_send_generic(info, team, team_nl_fill_port_list_get_all);
1762
1763         team_nl_team_put(team);
1764
1765         return err;
1766 }
1767
1768 static struct genl_ops team_nl_ops[] = {
1769         {
1770                 .cmd = TEAM_CMD_NOOP,
1771                 .doit = team_nl_cmd_noop,
1772                 .policy = team_nl_policy,
1773         },
1774         {
1775                 .cmd = TEAM_CMD_OPTIONS_SET,
1776                 .doit = team_nl_cmd_options_set,
1777                 .policy = team_nl_policy,
1778                 .flags = GENL_ADMIN_PERM,
1779         },
1780         {
1781                 .cmd = TEAM_CMD_OPTIONS_GET,
1782                 .doit = team_nl_cmd_options_get,
1783                 .policy = team_nl_policy,
1784                 .flags = GENL_ADMIN_PERM,
1785         },
1786         {
1787                 .cmd = TEAM_CMD_PORT_LIST_GET,
1788                 .doit = team_nl_cmd_port_list_get,
1789                 .policy = team_nl_policy,
1790                 .flags = GENL_ADMIN_PERM,
1791         },
1792 };
1793
1794 static struct genl_multicast_group team_change_event_mcgrp = {
1795         .name = TEAM_GENL_CHANGE_EVENT_MC_GRP_NAME,
1796 };
1797
1798 static int team_nl_send_event_options_get(struct team *team)
1799 {
1800         struct sk_buff *skb;
1801         int err;
1802         struct net *net = dev_net(team->dev);
1803
1804         skb = nlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1805         if (!skb)
1806                 return -ENOMEM;
1807
1808         err = team_nl_fill_options_get(skb, 0, 0, 0, team, false);
1809         if (err < 0)
1810                 goto err_fill;
1811
1812         err = genlmsg_multicast_netns(net, skb, 0, team_change_event_mcgrp.id,
1813                                       GFP_KERNEL);
1814         return err;
1815
1816 err_fill:
1817         nlmsg_free(skb);
1818         return err;
1819 }
1820
1821 static int team_nl_send_event_port_list_get(struct team *team)
1822 {
1823         struct sk_buff *skb;
1824         int err;
1825         struct net *net = dev_net(team->dev);
1826
1827         skb = nlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1828         if (!skb)
1829                 return -ENOMEM;
1830
1831         err = team_nl_fill_port_list_get(skb, 0, 0, 0, team, false);
1832         if (err < 0)
1833                 goto err_fill;
1834
1835         err = genlmsg_multicast_netns(net, skb, 0, team_change_event_mcgrp.id,
1836                                       GFP_KERNEL);
1837         return err;
1838
1839 err_fill:
1840         nlmsg_free(skb);
1841         return err;
1842 }
1843
1844 static int team_nl_init(void)
1845 {
1846         int err;
1847
1848         err = genl_register_family_with_ops(&team_nl_family, team_nl_ops,
1849                                             ARRAY_SIZE(team_nl_ops));
1850         if (err)
1851                 return err;
1852
1853         err = genl_register_mc_group(&team_nl_family, &team_change_event_mcgrp);
1854         if (err)
1855                 goto err_change_event_grp_reg;
1856
1857         return 0;
1858
1859 err_change_event_grp_reg:
1860         genl_unregister_family(&team_nl_family);
1861
1862         return err;
1863 }
1864
1865 static void team_nl_fini(void)
1866 {
1867         genl_unregister_family(&team_nl_family);
1868 }
1869
1870
1871 /******************
1872  * Change checkers
1873  ******************/
1874
1875 static void __team_options_change_check(struct team *team)
1876 {
1877         int err;
1878
1879         err = team_nl_send_event_options_get(team);
1880         if (err)
1881                 netdev_warn(team->dev, "Failed to send options change via netlink\n");
1882 }
1883
1884 /* rtnl lock is held */
1885 static void __team_port_change_check(struct team_port *port, bool linkup)
1886 {
1887         int err;
1888
1889         if (!port->removed && port->state.linkup == linkup)
1890                 return;
1891
1892         port->changed = true;
1893         port->state.linkup = linkup;
1894         team_refresh_port_linkup(port);
1895         if (linkup) {
1896                 struct ethtool_cmd ecmd;
1897
1898                 err = __ethtool_get_settings(port->dev, &ecmd);
1899                 if (!err) {
1900                         port->state.speed = ethtool_cmd_speed(&ecmd);
1901                         port->state.duplex = ecmd.duplex;
1902                         goto send_event;
1903                 }
1904         }
1905         port->state.speed = 0;
1906         port->state.duplex = 0;
1907
1908 send_event:
1909         err = team_nl_send_event_port_list_get(port->team);
1910         if (err)
1911                 netdev_warn(port->team->dev, "Failed to send port change of device %s via netlink\n",
1912                             port->dev->name);
1913
1914 }
1915
1916 static void team_port_change_check(struct team_port *port, bool linkup)
1917 {
1918         struct team *team = port->team;
1919
1920         mutex_lock(&team->lock);
1921         __team_port_change_check(port, linkup);
1922         mutex_unlock(&team->lock);
1923 }
1924
1925 /************************************
1926  * Net device notifier event handler
1927  ************************************/
1928
1929 static int team_device_event(struct notifier_block *unused,
1930                              unsigned long event, void *ptr)
1931 {
1932         struct net_device *dev = (struct net_device *) ptr;
1933         struct team_port *port;
1934
1935         port = team_port_get_rtnl(dev);
1936         if (!port)
1937                 return NOTIFY_DONE;
1938
1939         switch (event) {
1940         case NETDEV_UP:
1941                 if (netif_carrier_ok(dev))
1942                         team_port_change_check(port, true);
1943         case NETDEV_DOWN:
1944                 team_port_change_check(port, false);
1945         case NETDEV_CHANGE:
1946                 if (netif_running(port->dev))
1947                         team_port_change_check(port,
1948                                                !!netif_carrier_ok(port->dev));
1949                 break;
1950         case NETDEV_UNREGISTER:
1951                 team_del_slave(port->team->dev, dev);
1952                 break;
1953         case NETDEV_FEAT_CHANGE:
1954                 team_compute_features(port->team);
1955                 break;
1956         case NETDEV_CHANGEMTU:
1957                 /* Forbid to change mtu of underlaying device */
1958                 return NOTIFY_BAD;
1959         case NETDEV_PRE_TYPE_CHANGE:
1960                 /* Forbid to change type of underlaying device */
1961                 return NOTIFY_BAD;
1962         }
1963         return NOTIFY_DONE;
1964 }
1965
1966 static struct notifier_block team_notifier_block __read_mostly = {
1967         .notifier_call = team_device_event,
1968 };
1969
1970
1971 /***********************
1972  * Module init and exit
1973  ***********************/
1974
1975 static int __init team_module_init(void)
1976 {
1977         int err;
1978
1979         register_netdevice_notifier(&team_notifier_block);
1980
1981         err = rtnl_link_register(&team_link_ops);
1982         if (err)
1983                 goto err_rtnl_reg;
1984
1985         err = team_nl_init();
1986         if (err)
1987                 goto err_nl_init;
1988
1989         return 0;
1990
1991 err_nl_init:
1992         rtnl_link_unregister(&team_link_ops);
1993
1994 err_rtnl_reg:
1995         unregister_netdevice_notifier(&team_notifier_block);
1996
1997         return err;
1998 }
1999
2000 static void __exit team_module_exit(void)
2001 {
2002         team_nl_fini();
2003         rtnl_link_unregister(&team_link_ops);
2004         unregister_netdevice_notifier(&team_notifier_block);
2005 }
2006
2007 module_init(team_module_init);
2008 module_exit(team_module_exit);
2009
2010 MODULE_LICENSE("GPL v2");
2011 MODULE_AUTHOR("Jiri Pirko <jpirko@redhat.com>");
2012 MODULE_DESCRIPTION("Ethernet team device driver");
2013 MODULE_ALIAS_RTNL_LINK(DRV_NAME);