drivers/infiniband/core/roce_gid_mgmt.c

   1 /*
   2  * Copyright (c) 2015, Mellanox Technologies inc.  All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32
  33 #include "core_priv.h"
  34
  35 #include <linux/in.h>
  36 #include <linux/in6.h>
  37
  38 /* For in6_dev_get/in6_dev_put */
  39 #include <net/addrconf.h>
  40 #include <net/bonding.h>
  41
  42 #include <rdma/ib_cache.h>
  43 #include <rdma/ib_addr.h>
  44
  45 static struct workqueue_struct *gid_cache_wq;
  46
  47 enum gid_op_type {
  48         GID_DEL = 0,
  49         GID_ADD
  50 };
  51
  52 struct update_gid_event_work {
  53         struct work_struct work;
  54         union ib_gid       gid;
  55         struct ib_gid_attr gid_attr;
  56         enum gid_op_type gid_op;
  57 };
  58
  59 #define ROCE_NETDEV_CALLBACK_SZ         3
  60 struct netdev_event_work_cmd {
  61         roce_netdev_callback    cb;
  62         roce_netdev_filter      filter;
  63         struct net_device       *ndev;
  64         struct net_device       *filter_ndev;
  65 };
  66
  67 struct netdev_event_work {
  68         struct work_struct              work;
  69         struct netdev_event_work_cmd    cmds[ROCE_NETDEV_CALLBACK_SZ];
  70 };
  71
  72 static const struct {
  73         bool (*is_supported)(const struct ib_device *device, u8 port_num);
  74         enum ib_gid_type gid_type;
  75 } PORT_CAP_TO_GID_TYPE[] = {
  76         {rdma_protocol_roce_eth_encap, IB_GID_TYPE_ROCE},
  77         {rdma_protocol_roce_udp_encap, IB_GID_TYPE_ROCE_UDP_ENCAP},
  78 };
  79
  80 #define CAP_TO_GID_TABLE_SIZE   ARRAY_SIZE(PORT_CAP_TO_GID_TYPE)
  81
  82 unsigned long roce_gid_type_mask_support(struct ib_device *ib_dev, u8 port)
  83 {
  84         int i;
  85         unsigned int ret_flags = 0;
  86
  87         if (!rdma_protocol_roce(ib_dev, port))
  88                 return 1UL << IB_GID_TYPE_IB;
  89
  90         for (i = 0; i < CAP_TO_GID_TABLE_SIZE; i++)
  91                 if (PORT_CAP_TO_GID_TYPE[i].is_supported(ib_dev, port))
  92                         ret_flags |= 1UL << PORT_CAP_TO_GID_TYPE[i].gid_type;
  93
  94         return ret_flags;
  95 }
  96 EXPORT_SYMBOL(roce_gid_type_mask_support);
  97
  98 static void update_gid(enum gid_op_type gid_op, struct ib_device *ib_dev,
  99                        u8 port, union ib_gid *gid,
 100                        struct ib_gid_attr *gid_attr)
 101 {
 102         int i;
 103         unsigned long gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
 104
 105         for (i = 0; i < IB_GID_TYPE_SIZE; i++) {
 106                 if ((1UL << i) & gid_type_mask) {
 107                         gid_attr->gid_type = i;
 108                         switch (gid_op) {
 109                         case GID_ADD:
 110                                 ib_cache_gid_add(ib_dev, port,
 111                                                  gid, gid_attr);
 112                                 break;
 113                         case GID_DEL:
 114                                 ib_cache_gid_del(ib_dev, port,
 115                                                  gid, gid_attr);
 116                                 break;
 117                         }
 118                 }
 119         }
 120 }
 121
 122 enum bonding_slave_state {
 123         BONDING_SLAVE_STATE_ACTIVE      = 1UL << 0,
 124         BONDING_SLAVE_STATE_INACTIVE    = 1UL << 1,
 125         /* No primary slave or the device isn't a slave in bonding */
 126         BONDING_SLAVE_STATE_NA          = 1UL << 2,
 127 };
 128
 129 static enum bonding_slave_state is_eth_active_slave_of_bonding_rcu(struct net_device *dev,
 130                                                                    struct net_device *upper)
 131 {
 132         if (upper && netif_is_bond_master(upper)) {
 133                 struct net_device *pdev =
 134                         bond_option_active_slave_get_rcu(netdev_priv(upper));
 135
 136                 if (pdev)
 137                         return dev == pdev ? BONDING_SLAVE_STATE_ACTIVE :
 138                                 BONDING_SLAVE_STATE_INACTIVE;
 139         }
 140
 141         return BONDING_SLAVE_STATE_NA;
 142 }
 143
 144 #define REQUIRED_BOND_STATES            (BONDING_SLAVE_STATE_ACTIVE |   \
 145                                          BONDING_SLAVE_STATE_NA)
 146 static bool
 147 is_eth_port_of_netdev_filter(struct ib_device *ib_dev, u8 port,
 148                              struct net_device *rdma_ndev, void *cookie)
 149 {
 150         struct net_device *real_dev;
 151         bool res;
 152
 153         if (!rdma_ndev)
 154                 return false;
 155
 156         rcu_read_lock();
 157         real_dev = rdma_vlan_dev_real_dev(cookie);
 158         if (!real_dev)
 159                 real_dev = cookie;
 160
 161         res = ((rdma_is_upper_dev_rcu(rdma_ndev, cookie) &&
 162                (is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev) &
 163                 REQUIRED_BOND_STATES)) ||
 164                real_dev == rdma_ndev);
 165
 166         rcu_read_unlock();
 167         return res;
 168 }
 169
 170 static bool
 171 is_eth_port_inactive_slave_filter(struct ib_device *ib_dev, u8 port,
 172                                   struct net_device *rdma_ndev, void *cookie)
 173 {
 174         struct net_device *master_dev;
 175         bool res;
 176
 177         if (!rdma_ndev)
 178                 return false;
 179
 180         rcu_read_lock();
 181         master_dev = netdev_master_upper_dev_get_rcu(rdma_ndev);
 182         res = is_eth_active_slave_of_bonding_rcu(rdma_ndev, master_dev) ==
 183                 BONDING_SLAVE_STATE_INACTIVE;
 184         rcu_read_unlock();
 185
 186         return res;
 187 }
 188
 189 /** is_ndev_for_default_gid_filter - Check if a given netdevice
 190  * can be considered for default GIDs or not.
 191  * @ib_dev:             IB device to check
 192  * @port:               Port to consider for adding default GID
 193  * @rdma_ndev:          rdma netdevice pointer
 194  * @cookie_ndev:        Netdevice to consider to form a default GID
 195  *
 196  * is_ndev_for_default_gid_filter() returns true if a given netdevice can be
 197  * considered for deriving default RoCE GID, returns false otherwise.
 198  */
 199 static bool
 200 is_ndev_for_default_gid_filter(struct ib_device *ib_dev, u8 port,
 201                                struct net_device *rdma_ndev, void *cookie)
 202 {
 203         struct net_device *cookie_ndev = cookie;
 204         bool res;
 205
 206         if (!rdma_ndev)
 207                 return false;
 208
 209         rcu_read_lock();
 210
 211         /*
 212          * When rdma netdevice is used in bonding, bonding master netdevice
 213          * should be considered for default GIDs. Therefore, ignore slave rdma
 214          * netdevices when bonding is considered.
 215          * Additionally when event(cookie) netdevice is bond master device,
 216          * make sure that it the upper netdevice of rdma netdevice.
 217          */
 218         res = ((cookie_ndev == rdma_ndev && !netif_is_bond_slave(rdma_ndev)) ||
 219                (netif_is_bond_master(cookie_ndev) &&
 220                 rdma_is_upper_dev_rcu(rdma_ndev, cookie_ndev)));
 221
 222         rcu_read_unlock();
 223         return res;
 224 }
 225
 226 static bool pass_all_filter(struct ib_device *ib_dev, u8 port,
 227                             struct net_device *rdma_ndev, void *cookie)
 228 {
 229         return true;
 230 }
 231
 232 static bool upper_device_filter(struct ib_device *ib_dev, u8 port,
 233                                 struct net_device *rdma_ndev, void *cookie)
 234 {
 235         bool res;
 236
 237         if (!rdma_ndev)
 238                 return false;
 239
 240         if (rdma_ndev == cookie)
 241                 return true;
 242
 243         rcu_read_lock();
 244         res = rdma_is_upper_dev_rcu(rdma_ndev, cookie);
 245         rcu_read_unlock();
 246
 247         return res;
 248 }
 249
 250 /**
 251  * is_upper_ndev_bond_master_filter - Check if a given netdevice
 252  * is bond master device of netdevice of the the RDMA device of port.
 253  * @ib_dev:             IB device to check
 254  * @port:               Port to consider for adding default GID
 255  * @rdma_ndev:          Pointer to rdma netdevice
 256  * @cookie:             Netdevice to consider to form a default GID
 257  *
 258  * is_upper_ndev_bond_master_filter() returns true if a cookie_netdev
 259  * is bond master device and rdma_ndev is its lower netdevice. It might
 260  * not have been established as slave device yet.
 261  */
 262 static bool
 263 is_upper_ndev_bond_master_filter(struct ib_device *ib_dev, u8 port,
 264                                  struct net_device *rdma_ndev,
 265                                  void *cookie)
 266 {
 267         struct net_device *cookie_ndev = cookie;
 268         bool match = false;
 269
 270         if (!rdma_ndev)
 271                 return false;
 272
 273         rcu_read_lock();
 274         if (netif_is_bond_master(cookie_ndev) &&
 275             rdma_is_upper_dev_rcu(rdma_ndev, cookie_ndev))
 276                 match = true;
 277         rcu_read_unlock();
 278         return match;
 279 }
 280
 281 static void update_gid_ip(enum gid_op_type gid_op,
 282                           struct ib_device *ib_dev,
 283                           u8 port, struct net_device *ndev,
 284                           struct sockaddr *addr)
 285 {
 286         union ib_gid gid;
 287         struct ib_gid_attr gid_attr;
 288
 289         rdma_ip2gid(addr, &gid);
 290         memset(&gid_attr, 0, sizeof(gid_attr));
 291         gid_attr.ndev = ndev;
 292
 293         update_gid(gid_op, ib_dev, port, &gid, &gid_attr);
 294 }
 295
 296 static void bond_delete_netdev_default_gids(struct ib_device *ib_dev,
 297                                             u8 port,
 298                                             struct net_device *rdma_ndev,
 299                                             struct net_device *event_ndev)
 300 {
 301         struct net_device *real_dev = rdma_vlan_dev_real_dev(event_ndev);
 302         unsigned long gid_type_mask;
 303
 304         if (!rdma_ndev)
 305                 return;
 306
 307         if (!real_dev)
 308                 real_dev = event_ndev;
 309
 310         rcu_read_lock();
 311
 312         if (((rdma_ndev != event_ndev &&
 313               !rdma_is_upper_dev_rcu(rdma_ndev, event_ndev)) ||
 314              is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev)
 315                                                  ==
 316              BONDING_SLAVE_STATE_INACTIVE)) {
 317                 rcu_read_unlock();
 318                 return;
 319         }
 320
 321         rcu_read_unlock();
 322
 323         gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
 324
 325         ib_cache_gid_set_default_gid(ib_dev, port, rdma_ndev,
 326                                      gid_type_mask,
 327                                      IB_CACHE_GID_DEFAULT_MODE_DELETE);
 328 }
 329
 330 static void enum_netdev_ipv4_ips(struct ib_device *ib_dev,
 331                                  u8 port, struct net_device *ndev)
 332 {
 333         const struct in_ifaddr *ifa;
 334         struct in_device *in_dev;
 335         struct sin_list {
 336                 struct list_head        list;
 337                 struct sockaddr_in      ip;
 338         };
 339         struct sin_list *sin_iter;
 340         struct sin_list *sin_temp;
 341
 342         LIST_HEAD(sin_list);
 343         if (ndev->reg_state >= NETREG_UNREGISTERING)
 344                 return;
 345
 346         rcu_read_lock();
 347         in_dev = __in_dev_get_rcu(ndev);
 348         if (!in_dev) {
 349                 rcu_read_unlock();
 350                 return;
 351         }
 352
 353         in_dev_for_each_ifa_rcu(ifa, in_dev) {
 354                 struct sin_list *entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
 355
 356                 if (!entry)
 357                         continue;
 358
 359                 entry->ip.sin_family = AF_INET;
 360                 entry->ip.sin_addr.s_addr = ifa->ifa_address;
 361                 list_add_tail(&entry->list, &sin_list);
 362         }
 363
 364         rcu_read_unlock();
 365
 366         list_for_each_entry_safe(sin_iter, sin_temp, &sin_list, list) {
 367                 update_gid_ip(GID_ADD, ib_dev, port, ndev,
 368                               (struct sockaddr *)&sin_iter->ip);
 369                 list_del(&sin_iter->list);
 370                 kfree(sin_iter);
 371         }
 372 }
 373
 374 static void enum_netdev_ipv6_ips(struct ib_device *ib_dev,
 375                                  u8 port, struct net_device *ndev)
 376 {
 377         struct inet6_ifaddr *ifp;
 378         struct inet6_dev *in6_dev;
 379         struct sin6_list {
 380                 struct list_head        list;
 381                 struct sockaddr_in6     sin6;
 382         };
 383         struct sin6_list *sin6_iter;
 384         struct sin6_list *sin6_temp;
 385         struct ib_gid_attr gid_attr = {.ndev = ndev};
 386         LIST_HEAD(sin6_list);
 387
 388         if (ndev->reg_state >= NETREG_UNREGISTERING)
 389                 return;
 390
 391         in6_dev = in6_dev_get(ndev);
 392         if (!in6_dev)
 393                 return;
 394
 395         read_lock_bh(&in6_dev->lock);
 396         list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
 397                 struct sin6_list *entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
 398
 399                 if (!entry)
 400                         continue;
 401
 402                 entry->sin6.sin6_family = AF_INET6;
 403                 entry->sin6.sin6_addr = ifp->addr;
 404                 list_add_tail(&entry->list, &sin6_list);
 405         }
 406         read_unlock_bh(&in6_dev->lock);
 407
 408         in6_dev_put(in6_dev);
 409
 410         list_for_each_entry_safe(sin6_iter, sin6_temp, &sin6_list, list) {
 411                 union ib_gid    gid;
 412
 413                 rdma_ip2gid((struct sockaddr *)&sin6_iter->sin6, &gid);
 414                 update_gid(GID_ADD, ib_dev, port, &gid, &gid_attr);
 415                 list_del(&sin6_iter->list);
 416                 kfree(sin6_iter);
 417         }
 418 }
 419
 420 static void _add_netdev_ips(struct ib_device *ib_dev, u8 port,
 421                             struct net_device *ndev)
 422 {
 423         enum_netdev_ipv4_ips(ib_dev, port, ndev);
 424         if (IS_ENABLED(CONFIG_IPV6))
 425                 enum_netdev_ipv6_ips(ib_dev, port, ndev);
 426 }
 427
 428 static void add_netdev_ips(struct ib_device *ib_dev, u8 port,
 429                            struct net_device *rdma_ndev, void *cookie)
 430 {
 431         _add_netdev_ips(ib_dev, port, cookie);
 432 }
 433
 434 static void del_netdev_ips(struct ib_device *ib_dev, u8 port,
 435                            struct net_device *rdma_ndev, void *cookie)
 436 {
 437         ib_cache_gid_del_all_netdev_gids(ib_dev, port, cookie);
 438 }
 439
 440 /**
 441  * del_default_gids - Delete default GIDs of the event/cookie netdevice
 442  * @ib_dev:     RDMA device pointer
 443  * @port:       Port of the RDMA device whose GID table to consider
 444  * @rdma_ndev:  Unused rdma netdevice
 445  * @cookie:     Pointer to event netdevice
 446  *
 447  * del_default_gids() deletes the default GIDs of the event/cookie netdevice.
 448  */
 449 static void del_default_gids(struct ib_device *ib_dev, u8 port,
 450                              struct net_device *rdma_ndev, void *cookie)
 451 {
 452         struct net_device *cookie_ndev = cookie;
 453         unsigned long gid_type_mask;
 454
 455         gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
 456
 457         ib_cache_gid_set_default_gid(ib_dev, port, cookie_ndev, gid_type_mask,
 458                                      IB_CACHE_GID_DEFAULT_MODE_DELETE);
 459 }
 460
 461 static void add_default_gids(struct ib_device *ib_dev, u8 port,
 462                              struct net_device *rdma_ndev, void *cookie)
 463 {
 464         struct net_device *event_ndev = cookie;
 465         unsigned long gid_type_mask;
 466
 467         gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
 468         ib_cache_gid_set_default_gid(ib_dev, port, event_ndev, gid_type_mask,
 469                                      IB_CACHE_GID_DEFAULT_MODE_SET);
 470 }
 471
 472 static void enum_all_gids_of_dev_cb(struct ib_device *ib_dev,
 473                                     u8 port,
 474                                     struct net_device *rdma_ndev,
 475                                     void *cookie)
 476 {
 477         struct net *net;
 478         struct net_device *ndev;
 479
 480         /* Lock the rtnl to make sure the netdevs does not move under
 481          * our feet
 482          */
 483         rtnl_lock();
 484         down_read(&net_rwsem);
 485         for_each_net(net)
 486                 for_each_netdev(net, ndev) {
 487                         /*
 488                          * Filter and add default GIDs of the primary netdevice
 489                          * when not in bonding mode, or add default GIDs
 490                          * of bond master device, when in bonding mode.
 491                          */
 492                         if (is_ndev_for_default_gid_filter(ib_dev, port,
 493                                                            rdma_ndev, ndev))
 494                                 add_default_gids(ib_dev, port, rdma_ndev, ndev);
 495
 496                         if (is_eth_port_of_netdev_filter(ib_dev, port,
 497                                                          rdma_ndev, ndev))
 498                                 _add_netdev_ips(ib_dev, port, ndev);
 499                 }
 500         up_read(&net_rwsem);
 501         rtnl_unlock();
 502 }
 503
 504 /**
 505  * rdma_roce_rescan_device - Rescan all of the network devices in the system
 506  * and add their gids, as needed, to the relevant RoCE devices.
 507  *
 508  * @device:         the rdma device
 509  */
 510 void rdma_roce_rescan_device(struct ib_device *ib_dev)
 511 {
 512         ib_enum_roce_netdev(ib_dev, pass_all_filter, NULL,
 513                             enum_all_gids_of_dev_cb, NULL);
 514 }
 515 EXPORT_SYMBOL(rdma_roce_rescan_device);
 516
 517 static void callback_for_addr_gid_device_scan(struct ib_device *device,
 518                                               u8 port,
 519                                               struct net_device *rdma_ndev,
 520                                               void *cookie)
 521 {
 522         struct update_gid_event_work *parsed = cookie;
 523
 524         return update_gid(parsed->gid_op, device,
 525                           port, &parsed->gid,
 526                           &parsed->gid_attr);
 527 }
 528
 529 struct upper_list {
 530         struct list_head list;
 531         struct net_device *upper;
 532 };
 533
 534 static int netdev_upper_walk(struct net_device *upper, void *data)
 535 {
 536         struct upper_list *entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
 537         struct list_head *upper_list = data;
 538
 539         if (!entry)
 540                 return 0;
 541
 542         list_add_tail(&entry->list, upper_list);
 543         dev_hold(upper);
 544         entry->upper = upper;
 545
 546         return 0;
 547 }
 548
 549 static void handle_netdev_upper(struct ib_device *ib_dev, u8 port,
 550                                 void *cookie,
 551                                 void (*handle_netdev)(struct ib_device *ib_dev,
 552                                                       u8 port,
 553                                                       struct net_device *ndev))
 554 {
 555         struct net_device *ndev = cookie;
 556         struct upper_list *upper_iter;
 557         struct upper_list *upper_temp;
 558         LIST_HEAD(upper_list);
 559
 560         rcu_read_lock();
 561         netdev_walk_all_upper_dev_rcu(ndev, netdev_upper_walk, &upper_list);
 562         rcu_read_unlock();
 563
 564         handle_netdev(ib_dev, port, ndev);
 565         list_for_each_entry_safe(upper_iter, upper_temp, &upper_list,
 566                                  list) {
 567                 handle_netdev(ib_dev, port, upper_iter->upper);
 568                 dev_put(upper_iter->upper);
 569                 list_del(&upper_iter->list);
 570                 kfree(upper_iter);
 571         }
 572 }
 573
 574 static void _roce_del_all_netdev_gids(struct ib_device *ib_dev, u8 port,
 575                                       struct net_device *event_ndev)
 576 {
 577         ib_cache_gid_del_all_netdev_gids(ib_dev, port, event_ndev);
 578 }
 579
 580 static void del_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
 581                                  struct net_device *rdma_ndev, void *cookie)
 582 {
 583         handle_netdev_upper(ib_dev, port, cookie, _roce_del_all_netdev_gids);
 584 }
 585
 586 static void add_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
 587                                  struct net_device *rdma_ndev, void *cookie)
 588 {
 589         handle_netdev_upper(ib_dev, port, cookie, _add_netdev_ips);
 590 }
 591
 592 static void del_netdev_default_ips_join(struct ib_device *ib_dev, u8 port,
 593                                         struct net_device *rdma_ndev,
 594                                         void *cookie)
 595 {
 596         struct net_device *master_ndev;
 597
 598         rcu_read_lock();
 599         master_ndev = netdev_master_upper_dev_get_rcu(rdma_ndev);
 600         if (master_ndev)
 601                 dev_hold(master_ndev);
 602         rcu_read_unlock();
 603
 604         if (master_ndev) {
 605                 bond_delete_netdev_default_gids(ib_dev, port, rdma_ndev,
 606                                                 master_ndev);
 607                 dev_put(master_ndev);
 608         }
 609 }
 610
 611 /* The following functions operate on all IB devices. netdevice_event and
 612  * addr_event execute ib_enum_all_roce_netdevs through a work.
 613  * ib_enum_all_roce_netdevs iterates through all IB devices.
 614  */
 615
 616 static void netdevice_event_work_handler(struct work_struct *_work)
 617 {
 618         struct netdev_event_work *work =
 619                 container_of(_work, struct netdev_event_work, work);
 620         unsigned int i;
 621
 622         for (i = 0; i < ARRAY_SIZE(work->cmds) && work->cmds[i].cb; i++) {
 623                 ib_enum_all_roce_netdevs(work->cmds[i].filter,
 624                                          work->cmds[i].filter_ndev,
 625                                          work->cmds[i].cb,
 626                                          work->cmds[i].ndev);
 627                 dev_put(work->cmds[i].ndev);
 628                 dev_put(work->cmds[i].filter_ndev);
 629         }
 630
 631         kfree(work);
 632 }
 633
 634 static int netdevice_queue_work(struct netdev_event_work_cmd *cmds,
 635                                 struct net_device *ndev)
 636 {
 637         unsigned int i;
 638         struct netdev_event_work *ndev_work =
 639                 kmalloc(sizeof(*ndev_work), GFP_KERNEL);
 640
 641         if (!ndev_work)
 642                 return NOTIFY_DONE;
 643
 644         memcpy(ndev_work->cmds, cmds, sizeof(ndev_work->cmds));
 645         for (i = 0; i < ARRAY_SIZE(ndev_work->cmds) && ndev_work->cmds[i].cb; i++) {
 646                 if (!ndev_work->cmds[i].ndev)
 647                         ndev_work->cmds[i].ndev = ndev;
 648                 if (!ndev_work->cmds[i].filter_ndev)
 649                         ndev_work->cmds[i].filter_ndev = ndev;
 650                 dev_hold(ndev_work->cmds[i].ndev);
 651                 dev_hold(ndev_work->cmds[i].filter_ndev);
 652         }
 653         INIT_WORK(&ndev_work->work, netdevice_event_work_handler);
 654
 655         queue_work(gid_cache_wq, &ndev_work->work);
 656
 657         return NOTIFY_DONE;
 658 }
 659
 660 static const struct netdev_event_work_cmd add_cmd = {
 661         .cb     = add_netdev_ips,
 662         .filter = is_eth_port_of_netdev_filter
 663 };
 664
 665 static const struct netdev_event_work_cmd add_cmd_upper_ips = {
 666         .cb     = add_netdev_upper_ips,
 667         .filter = is_eth_port_of_netdev_filter
 668 };
 669
 670 static void
 671 ndev_event_unlink(struct netdev_notifier_changeupper_info *changeupper_info,
 672                   struct netdev_event_work_cmd *cmds)
 673 {
 674         static const struct netdev_event_work_cmd
 675                         upper_ips_del_cmd = {
 676                                 .cb     = del_netdev_upper_ips,
 677                                 .filter = upper_device_filter
 678         };
 679
 680         cmds[0] = upper_ips_del_cmd;
 681         cmds[0].ndev = changeupper_info->upper_dev;
 682         cmds[1] = add_cmd;
 683 }
 684
 685 static const struct netdev_event_work_cmd bonding_default_add_cmd = {
 686         .cb     = add_default_gids,
 687         .filter = is_upper_ndev_bond_master_filter
 688 };
 689
 690 static void
 691 ndev_event_link(struct net_device *event_ndev,
 692                 struct netdev_notifier_changeupper_info *changeupper_info,
 693                 struct netdev_event_work_cmd *cmds)
 694 {
 695         static const struct netdev_event_work_cmd
 696                         bonding_default_del_cmd = {
 697                                 .cb     = del_default_gids,
 698                                 .filter = is_upper_ndev_bond_master_filter
 699                         };
 700         /*
 701          * When a lower netdev is linked to its upper bonding
 702          * netdev, delete lower slave netdev's default GIDs.
 703          */
 704         cmds[0] = bonding_default_del_cmd;
 705         cmds[0].ndev = event_ndev;
 706         cmds[0].filter_ndev = changeupper_info->upper_dev;
 707
 708         /* Now add bonding upper device default GIDs */
 709         cmds[1] = bonding_default_add_cmd;
 710         cmds[1].ndev = changeupper_info->upper_dev;
 711         cmds[1].filter_ndev = changeupper_info->upper_dev;
 712
 713         /* Now add bonding upper device IP based GIDs */
 714         cmds[2] = add_cmd_upper_ips;
 715         cmds[2].ndev = changeupper_info->upper_dev;
 716         cmds[2].filter_ndev = changeupper_info->upper_dev;
 717 }
 718
 719 static void netdevice_event_changeupper(struct net_device *event_ndev,
 720                 struct netdev_notifier_changeupper_info *changeupper_info,
 721                 struct netdev_event_work_cmd *cmds)
 722 {
 723         if (changeupper_info->linking)
 724                 ndev_event_link(event_ndev, changeupper_info, cmds);
 725         else
 726                 ndev_event_unlink(changeupper_info, cmds);
 727 }
 728
 729 static const struct netdev_event_work_cmd add_default_gid_cmd = {
 730         .cb     = add_default_gids,
 731         .filter = is_ndev_for_default_gid_filter,
 732 };
 733
 734 static int netdevice_event(struct notifier_block *this, unsigned long event,
 735                            void *ptr)
 736 {
 737         static const struct netdev_event_work_cmd del_cmd = {
 738                 .cb = del_netdev_ips, .filter = pass_all_filter};
 739         static const struct netdev_event_work_cmd
 740                         bonding_default_del_cmd_join = {
 741                                 .cb     = del_netdev_default_ips_join,
 742                                 .filter = is_eth_port_inactive_slave_filter
 743                         };
 744         static const struct netdev_event_work_cmd
 745                         netdev_del_cmd = {
 746                                 .cb     = del_netdev_ips,
 747                                 .filter = is_eth_port_of_netdev_filter
 748                         };
 749         static const struct netdev_event_work_cmd bonding_event_ips_del_cmd = {
 750                 .cb = del_netdev_upper_ips, .filter = upper_device_filter};
 751         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
 752         struct netdev_event_work_cmd cmds[ROCE_NETDEV_CALLBACK_SZ] = { {NULL} };
 753
 754         if (ndev->type != ARPHRD_ETHER)
 755                 return NOTIFY_DONE;
 756
 757         switch (event) {
 758         case NETDEV_REGISTER:
 759         case NETDEV_UP:
 760                 cmds[0] = bonding_default_del_cmd_join;
 761                 cmds[1] = add_default_gid_cmd;
 762                 cmds[2] = add_cmd;
 763                 break;
 764
 765         case NETDEV_UNREGISTER:
 766                 if (ndev->reg_state < NETREG_UNREGISTERED)
 767                         cmds[0] = del_cmd;
 768                 else
 769                         return NOTIFY_DONE;
 770                 break;
 771
 772         case NETDEV_CHANGEADDR:
 773                 cmds[0] = netdev_del_cmd;
 774                 if (ndev->reg_state == NETREG_REGISTERED) {
 775                         cmds[1] = add_default_gid_cmd;
 776                         cmds[2] = add_cmd;
 777                 }
 778                 break;
 779
 780         case NETDEV_CHANGEUPPER:
 781                 netdevice_event_changeupper(ndev,
 782                         container_of(ptr, struct netdev_notifier_changeupper_info, info),
 783                         cmds);
 784                 break;
 785
 786         case NETDEV_BONDING_FAILOVER:
 787                 cmds[0] = bonding_event_ips_del_cmd;
 788                 /* Add default GIDs of the bond device */
 789                 cmds[1] = bonding_default_add_cmd;
 790                 /* Add IP based GIDs of the bond device */
 791                 cmds[2] = add_cmd_upper_ips;
 792                 break;
 793
 794         default:
 795                 return NOTIFY_DONE;
 796         }
 797
 798         return netdevice_queue_work(cmds, ndev);
 799 }
 800
 801 static void update_gid_event_work_handler(struct work_struct *_work)
 802 {
 803         struct update_gid_event_work *work =
 804                 container_of(_work, struct update_gid_event_work, work);
 805
 806         ib_enum_all_roce_netdevs(is_eth_port_of_netdev_filter,
 807                                  work->gid_attr.ndev,
 808                                  callback_for_addr_gid_device_scan, work);
 809
 810         dev_put(work->gid_attr.ndev);
 811         kfree(work);
 812 }
 813
 814 static int addr_event(struct notifier_block *this, unsigned long event,
 815                       struct sockaddr *sa, struct net_device *ndev)
 816 {
 817         struct update_gid_event_work *work;
 818         enum gid_op_type gid_op;
 819
 820         if (ndev->type != ARPHRD_ETHER)
 821                 return NOTIFY_DONE;
 822
 823         switch (event) {
 824         case NETDEV_UP:
 825                 gid_op = GID_ADD;
 826                 break;
 827
 828         case NETDEV_DOWN:
 829                 gid_op = GID_DEL;
 830                 break;
 831
 832         default:
 833                 return NOTIFY_DONE;
 834         }
 835
 836         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 837         if (!work)
 838                 return NOTIFY_DONE;
 839
 840         INIT_WORK(&work->work, update_gid_event_work_handler);
 841
 842         rdma_ip2gid(sa, &work->gid);
 843         work->gid_op = gid_op;
 844
 845         memset(&work->gid_attr, 0, sizeof(work->gid_attr));
 846         dev_hold(ndev);
 847         work->gid_attr.ndev   = ndev;
 848
 849         queue_work(gid_cache_wq, &work->work);
 850
 851         return NOTIFY_DONE;
 852 }
 853
 854 static int inetaddr_event(struct notifier_block *this, unsigned long event,
 855                           void *ptr)
 856 {
 857         struct sockaddr_in      in;
 858         struct net_device       *ndev;
 859         struct in_ifaddr        *ifa = ptr;
 860
 861         in.sin_family = AF_INET;
 862         in.sin_addr.s_addr = ifa->ifa_address;
 863         ndev = ifa->ifa_dev->dev;
 864
 865         return addr_event(this, event, (struct sockaddr *)&in, ndev);
 866 }
 867
 868 static int inet6addr_event(struct notifier_block *this, unsigned long event,
 869                            void *ptr)
 870 {
 871         struct sockaddr_in6     in6;
 872         struct net_device       *ndev;
 873         struct inet6_ifaddr     *ifa6 = ptr;
 874
 875         in6.sin6_family = AF_INET6;
 876         in6.sin6_addr = ifa6->addr;
 877         ndev = ifa6->idev->dev;
 878
 879         return addr_event(this, event, (struct sockaddr *)&in6, ndev);
 880 }
 881
 882 static struct notifier_block nb_netdevice = {
 883         .notifier_call = netdevice_event
 884 };
 885
 886 static struct notifier_block nb_inetaddr = {
 887         .notifier_call = inetaddr_event
 888 };
 889
 890 static struct notifier_block nb_inet6addr = {
 891         .notifier_call = inet6addr_event
 892 };
 893
 894 int __init roce_gid_mgmt_init(void)
 895 {
 896         gid_cache_wq = alloc_ordered_workqueue("gid-cache-wq", 0);
 897         if (!gid_cache_wq)
 898                 return -ENOMEM;
 899
 900         register_inetaddr_notifier(&nb_inetaddr);
 901         if (IS_ENABLED(CONFIG_IPV6))
 902                 register_inet6addr_notifier(&nb_inet6addr);
 903         /* We relay on the netdevice notifier to enumerate all
 904          * existing devices in the system. Register to this notifier
 905          * last to make sure we will not miss any IP add/del
 906          * callbacks.
 907          */
 908         register_netdevice_notifier(&nb_netdevice);
 909
 910         return 0;
 911 }
 912
 913 void __exit roce_gid_mgmt_cleanup(void)
 914 {
 915         if (IS_ENABLED(CONFIG_IPV6))
 916                 unregister_inet6addr_notifier(&nb_inet6addr);
 917         unregister_inetaddr_notifier(&nb_inetaddr);
 918         unregister_netdevice_notifier(&nb_netdevice);
 919         /* Ensure all gid deletion tasks complete before we go down,
 920          * to avoid any reference to free'd memory. By the time
 921          * ib-core is removed, all physical devices have been removed,
 922          * so no issue with remaining hardware contexts.
 923          */
 924         destroy_workqueue(gid_cache_wq);
 925 }