2 * Copyright (c) 2015, Mellanox Technologies inc. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #include "core_priv.h"
36 #include <linux/in6.h>
38 /* For in6_dev_get/in6_dev_put */
39 #include <net/addrconf.h>
40 #include <net/bonding.h>
42 #include <rdma/ib_cache.h>
43 #include <rdma/ib_addr.h>
50 struct update_gid_event_work {
51 struct work_struct work;
53 struct ib_gid_attr gid_attr;
54 enum gid_op_type gid_op;
57 #define ROCE_NETDEV_CALLBACK_SZ 3
58 struct netdev_event_work_cmd {
59 roce_netdev_callback cb;
60 roce_netdev_filter filter;
61 struct net_device *ndev;
62 struct net_device *filter_ndev;
65 struct netdev_event_work {
66 struct work_struct work;
67 struct netdev_event_work_cmd cmds[ROCE_NETDEV_CALLBACK_SZ];
71 bool (*is_supported)(const struct ib_device *device, u8 port_num);
72 enum ib_gid_type gid_type;
73 } PORT_CAP_TO_GID_TYPE[] = {
74 {rdma_protocol_roce_eth_encap, IB_GID_TYPE_ROCE},
75 {rdma_protocol_roce_udp_encap, IB_GID_TYPE_ROCE_UDP_ENCAP},
78 #define CAP_TO_GID_TABLE_SIZE ARRAY_SIZE(PORT_CAP_TO_GID_TYPE)
80 unsigned long roce_gid_type_mask_support(struct ib_device *ib_dev, u8 port)
83 unsigned int ret_flags = 0;
85 if (!rdma_protocol_roce(ib_dev, port))
86 return 1UL << IB_GID_TYPE_IB;
88 for (i = 0; i < CAP_TO_GID_TABLE_SIZE; i++)
89 if (PORT_CAP_TO_GID_TYPE[i].is_supported(ib_dev, port))
90 ret_flags |= 1UL << PORT_CAP_TO_GID_TYPE[i].gid_type;
94 EXPORT_SYMBOL(roce_gid_type_mask_support);
96 static void update_gid(enum gid_op_type gid_op, struct ib_device *ib_dev,
97 u8 port, union ib_gid *gid,
98 struct ib_gid_attr *gid_attr)
101 unsigned long gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
103 for (i = 0; i < IB_GID_TYPE_SIZE; i++) {
104 if ((1UL << i) & gid_type_mask) {
105 gid_attr->gid_type = i;
108 ib_cache_gid_add(ib_dev, port,
112 ib_cache_gid_del(ib_dev, port,
120 enum bonding_slave_state {
121 BONDING_SLAVE_STATE_ACTIVE = 1UL << 0,
122 BONDING_SLAVE_STATE_INACTIVE = 1UL << 1,
123 /* No primary slave or the device isn't a slave in bonding */
124 BONDING_SLAVE_STATE_NA = 1UL << 2,
127 static enum bonding_slave_state is_eth_active_slave_of_bonding_rcu(struct net_device *dev,
128 struct net_device *upper)
130 if (upper && netif_is_bond_master(upper)) {
131 struct net_device *pdev =
132 bond_option_active_slave_get_rcu(netdev_priv(upper));
135 return dev == pdev ? BONDING_SLAVE_STATE_ACTIVE :
136 BONDING_SLAVE_STATE_INACTIVE;
139 return BONDING_SLAVE_STATE_NA;
142 #define REQUIRED_BOND_STATES (BONDING_SLAVE_STATE_ACTIVE | \
143 BONDING_SLAVE_STATE_NA)
144 static int is_eth_port_of_netdev(struct ib_device *ib_dev, u8 port,
145 struct net_device *rdma_ndev, void *cookie)
147 struct net_device *real_dev;
154 real_dev = rdma_vlan_dev_real_dev(cookie);
158 res = ((rdma_is_upper_dev_rcu(rdma_ndev, cookie) &&
159 (is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev) &
160 REQUIRED_BOND_STATES)) ||
161 real_dev == rdma_ndev);
167 static int is_eth_port_inactive_slave(struct ib_device *ib_dev, u8 port,
168 struct net_device *rdma_ndev, void *cookie)
170 struct net_device *master_dev;
177 master_dev = netdev_master_upper_dev_get_rcu(rdma_ndev);
178 res = is_eth_active_slave_of_bonding_rcu(rdma_ndev, master_dev) ==
179 BONDING_SLAVE_STATE_INACTIVE;
185 static int pass_all_filter(struct ib_device *ib_dev, u8 port,
186 struct net_device *rdma_ndev, void *cookie)
191 static int upper_device_filter(struct ib_device *ib_dev, u8 port,
192 struct net_device *rdma_ndev, void *cookie)
199 if (rdma_ndev == cookie)
203 res = rdma_is_upper_dev_rcu(rdma_ndev, cookie);
209 static void update_gid_ip(enum gid_op_type gid_op,
210 struct ib_device *ib_dev,
211 u8 port, struct net_device *ndev,
212 struct sockaddr *addr)
215 struct ib_gid_attr gid_attr;
217 rdma_ip2gid(addr, &gid);
218 memset(&gid_attr, 0, sizeof(gid_attr));
219 gid_attr.ndev = ndev;
221 update_gid(gid_op, ib_dev, port, &gid, &gid_attr);
224 static void enum_netdev_default_gids(struct ib_device *ib_dev,
225 u8 port, struct net_device *event_ndev,
226 struct net_device *rdma_ndev)
228 unsigned long gid_type_mask;
232 ((rdma_ndev != event_ndev &&
233 !rdma_is_upper_dev_rcu(rdma_ndev, event_ndev)) ||
234 is_eth_active_slave_of_bonding_rcu(rdma_ndev,
235 netdev_master_upper_dev_get_rcu(rdma_ndev)) ==
236 BONDING_SLAVE_STATE_INACTIVE)) {
242 gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
244 ib_cache_gid_set_default_gid(ib_dev, port, rdma_ndev, gid_type_mask,
245 IB_CACHE_GID_DEFAULT_MODE_SET);
248 static void bond_delete_netdev_default_gids(struct ib_device *ib_dev,
250 struct net_device *event_ndev,
251 struct net_device *rdma_ndev)
253 struct net_device *real_dev = rdma_vlan_dev_real_dev(event_ndev);
259 real_dev = event_ndev;
263 if (rdma_is_upper_dev_rcu(rdma_ndev, event_ndev) &&
264 is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev) ==
265 BONDING_SLAVE_STATE_INACTIVE) {
266 unsigned long gid_type_mask;
270 gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
272 ib_cache_gid_set_default_gid(ib_dev, port, rdma_ndev,
274 IB_CACHE_GID_DEFAULT_MODE_DELETE);
280 static void enum_netdev_ipv4_ips(struct ib_device *ib_dev,
281 u8 port, struct net_device *ndev)
283 struct in_device *in_dev;
285 struct list_head list;
286 struct sockaddr_in ip;
288 struct sin_list *sin_iter;
289 struct sin_list *sin_temp;
292 if (ndev->reg_state >= NETREG_UNREGISTERING)
296 in_dev = __in_dev_get_rcu(ndev);
303 struct sin_list *entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
308 entry->ip.sin_family = AF_INET;
309 entry->ip.sin_addr.s_addr = ifa->ifa_address;
310 list_add_tail(&entry->list, &sin_list);
315 list_for_each_entry_safe(sin_iter, sin_temp, &sin_list, list) {
316 update_gid_ip(GID_ADD, ib_dev, port, ndev,
317 (struct sockaddr *)&sin_iter->ip);
318 list_del(&sin_iter->list);
323 static void enum_netdev_ipv6_ips(struct ib_device *ib_dev,
324 u8 port, struct net_device *ndev)
326 struct inet6_ifaddr *ifp;
327 struct inet6_dev *in6_dev;
329 struct list_head list;
330 struct sockaddr_in6 sin6;
332 struct sin6_list *sin6_iter;
333 struct sin6_list *sin6_temp;
334 struct ib_gid_attr gid_attr = {.ndev = ndev};
335 LIST_HEAD(sin6_list);
337 if (ndev->reg_state >= NETREG_UNREGISTERING)
340 in6_dev = in6_dev_get(ndev);
344 read_lock_bh(&in6_dev->lock);
345 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
346 struct sin6_list *entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
351 entry->sin6.sin6_family = AF_INET6;
352 entry->sin6.sin6_addr = ifp->addr;
353 list_add_tail(&entry->list, &sin6_list);
355 read_unlock_bh(&in6_dev->lock);
357 in6_dev_put(in6_dev);
359 list_for_each_entry_safe(sin6_iter, sin6_temp, &sin6_list, list) {
362 rdma_ip2gid((struct sockaddr *)&sin6_iter->sin6, &gid);
363 update_gid(GID_ADD, ib_dev, port, &gid, &gid_attr);
364 list_del(&sin6_iter->list);
369 static void _add_netdev_ips(struct ib_device *ib_dev, u8 port,
370 struct net_device *ndev)
372 enum_netdev_ipv4_ips(ib_dev, port, ndev);
373 if (IS_ENABLED(CONFIG_IPV6))
374 enum_netdev_ipv6_ips(ib_dev, port, ndev);
377 static void add_netdev_ips(struct ib_device *ib_dev, u8 port,
378 struct net_device *rdma_ndev, void *cookie)
380 enum_netdev_default_gids(ib_dev, port, cookie, rdma_ndev);
381 _add_netdev_ips(ib_dev, port, cookie);
384 static void del_netdev_ips(struct ib_device *ib_dev, u8 port,
385 struct net_device *rdma_ndev, void *cookie)
387 ib_cache_gid_del_all_netdev_gids(ib_dev, port, cookie);
390 static void enum_all_gids_of_dev_cb(struct ib_device *ib_dev,
392 struct net_device *rdma_ndev,
396 struct net_device *ndev;
398 /* Lock the rtnl to make sure the netdevs does not move under
403 for_each_netdev(net, ndev)
404 if (is_eth_port_of_netdev(ib_dev, port, rdma_ndev, ndev))
405 add_netdev_ips(ib_dev, port, rdma_ndev, ndev);
409 /* This function will rescan all of the network devices in the system
410 * and add their gids, as needed, to the relevant RoCE devices. */
411 int roce_rescan_device(struct ib_device *ib_dev)
413 ib_enum_roce_netdev(ib_dev, pass_all_filter, NULL,
414 enum_all_gids_of_dev_cb, NULL);
419 static void callback_for_addr_gid_device_scan(struct ib_device *device,
421 struct net_device *rdma_ndev,
424 struct update_gid_event_work *parsed = cookie;
426 return update_gid(parsed->gid_op, device,
432 struct list_head list;
433 struct net_device *upper;
436 static int netdev_upper_walk(struct net_device *upper, void *data)
438 struct upper_list *entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
439 struct list_head *upper_list = data;
444 list_add_tail(&entry->list, upper_list);
446 entry->upper = upper;
451 static void handle_netdev_upper(struct ib_device *ib_dev, u8 port,
453 void (*handle_netdev)(struct ib_device *ib_dev,
455 struct net_device *ndev))
457 struct net_device *ndev = cookie;
458 struct upper_list *upper_iter;
459 struct upper_list *upper_temp;
460 LIST_HEAD(upper_list);
463 netdev_walk_all_upper_dev_rcu(ndev, netdev_upper_walk, &upper_list);
466 handle_netdev(ib_dev, port, ndev);
467 list_for_each_entry_safe(upper_iter, upper_temp, &upper_list,
469 handle_netdev(ib_dev, port, upper_iter->upper);
470 dev_put(upper_iter->upper);
471 list_del(&upper_iter->list);
476 static void _roce_del_all_netdev_gids(struct ib_device *ib_dev, u8 port,
477 struct net_device *event_ndev)
479 ib_cache_gid_del_all_netdev_gids(ib_dev, port, event_ndev);
482 static void del_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
483 struct net_device *rdma_ndev, void *cookie)
485 handle_netdev_upper(ib_dev, port, cookie, _roce_del_all_netdev_gids);
488 static void add_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
489 struct net_device *rdma_ndev, void *cookie)
491 handle_netdev_upper(ib_dev, port, cookie, _add_netdev_ips);
494 static void del_netdev_default_ips_join(struct ib_device *ib_dev, u8 port,
495 struct net_device *rdma_ndev,
498 struct net_device *master_ndev;
501 master_ndev = netdev_master_upper_dev_get_rcu(rdma_ndev);
503 dev_hold(master_ndev);
507 bond_delete_netdev_default_gids(ib_dev, port, master_ndev,
509 dev_put(master_ndev);
513 static void del_netdev_default_ips(struct ib_device *ib_dev, u8 port,
514 struct net_device *rdma_ndev, void *cookie)
516 bond_delete_netdev_default_gids(ib_dev, port, cookie, rdma_ndev);
519 /* The following functions operate on all IB devices. netdevice_event and
520 * addr_event execute ib_enum_all_roce_netdevs through a work.
521 * ib_enum_all_roce_netdevs iterates through all IB devices.
524 static void netdevice_event_work_handler(struct work_struct *_work)
526 struct netdev_event_work *work =
527 container_of(_work, struct netdev_event_work, work);
530 for (i = 0; i < ARRAY_SIZE(work->cmds) && work->cmds[i].cb; i++) {
531 ib_enum_all_roce_netdevs(work->cmds[i].filter,
532 work->cmds[i].filter_ndev,
535 dev_put(work->cmds[i].ndev);
536 dev_put(work->cmds[i].filter_ndev);
542 static int netdevice_queue_work(struct netdev_event_work_cmd *cmds,
543 struct net_device *ndev)
546 struct netdev_event_work *ndev_work =
547 kmalloc(sizeof(*ndev_work), GFP_KERNEL);
552 memcpy(ndev_work->cmds, cmds, sizeof(ndev_work->cmds));
553 for (i = 0; i < ARRAY_SIZE(ndev_work->cmds) && ndev_work->cmds[i].cb; i++) {
554 if (!ndev_work->cmds[i].ndev)
555 ndev_work->cmds[i].ndev = ndev;
556 if (!ndev_work->cmds[i].filter_ndev)
557 ndev_work->cmds[i].filter_ndev = ndev;
558 dev_hold(ndev_work->cmds[i].ndev);
559 dev_hold(ndev_work->cmds[i].filter_ndev);
561 INIT_WORK(&ndev_work->work, netdevice_event_work_handler);
563 queue_work(ib_wq, &ndev_work->work);
568 static const struct netdev_event_work_cmd add_cmd = {
569 .cb = add_netdev_ips, .filter = is_eth_port_of_netdev};
570 static const struct netdev_event_work_cmd add_cmd_upper_ips = {
571 .cb = add_netdev_upper_ips, .filter = is_eth_port_of_netdev};
573 static void netdevice_event_changeupper(struct netdev_notifier_changeupper_info *changeupper_info,
574 struct netdev_event_work_cmd *cmds)
576 static const struct netdev_event_work_cmd upper_ips_del_cmd = {
577 .cb = del_netdev_upper_ips, .filter = upper_device_filter};
578 static const struct netdev_event_work_cmd bonding_default_del_cmd = {
579 .cb = del_netdev_default_ips, .filter = is_eth_port_inactive_slave};
581 if (changeupper_info->linking == false) {
582 cmds[0] = upper_ips_del_cmd;
583 cmds[0].ndev = changeupper_info->upper_dev;
586 cmds[0] = bonding_default_del_cmd;
587 cmds[0].ndev = changeupper_info->upper_dev;
588 cmds[1] = add_cmd_upper_ips;
589 cmds[1].ndev = changeupper_info->upper_dev;
590 cmds[1].filter_ndev = changeupper_info->upper_dev;
594 static int netdevice_event(struct notifier_block *this, unsigned long event,
597 static const struct netdev_event_work_cmd del_cmd = {
598 .cb = del_netdev_ips, .filter = pass_all_filter};
599 static const struct netdev_event_work_cmd bonding_default_del_cmd_join = {
600 .cb = del_netdev_default_ips_join, .filter = is_eth_port_inactive_slave};
601 static const struct netdev_event_work_cmd default_del_cmd = {
602 .cb = del_netdev_default_ips, .filter = pass_all_filter};
603 static const struct netdev_event_work_cmd bonding_event_ips_del_cmd = {
604 .cb = del_netdev_upper_ips, .filter = upper_device_filter};
605 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
606 struct netdev_event_work_cmd cmds[ROCE_NETDEV_CALLBACK_SZ] = { {NULL} };
608 if (ndev->type != ARPHRD_ETHER)
612 case NETDEV_REGISTER:
614 cmds[0] = bonding_default_del_cmd_join;
618 case NETDEV_UNREGISTER:
619 if (ndev->reg_state < NETREG_UNREGISTERED)
625 case NETDEV_CHANGEADDR:
626 cmds[0] = default_del_cmd;
630 case NETDEV_CHANGEUPPER:
631 netdevice_event_changeupper(
632 container_of(ptr, struct netdev_notifier_changeupper_info, info),
636 case NETDEV_BONDING_FAILOVER:
637 cmds[0] = bonding_event_ips_del_cmd;
638 cmds[1] = bonding_default_del_cmd_join;
639 cmds[2] = add_cmd_upper_ips;
646 return netdevice_queue_work(cmds, ndev);
649 static void update_gid_event_work_handler(struct work_struct *_work)
651 struct update_gid_event_work *work =
652 container_of(_work, struct update_gid_event_work, work);
654 ib_enum_all_roce_netdevs(is_eth_port_of_netdev, work->gid_attr.ndev,
655 callback_for_addr_gid_device_scan, work);
657 dev_put(work->gid_attr.ndev);
661 static int addr_event(struct notifier_block *this, unsigned long event,
662 struct sockaddr *sa, struct net_device *ndev)
664 struct update_gid_event_work *work;
665 enum gid_op_type gid_op;
667 if (ndev->type != ARPHRD_ETHER)
683 work = kmalloc(sizeof(*work), GFP_ATOMIC);
687 INIT_WORK(&work->work, update_gid_event_work_handler);
689 rdma_ip2gid(sa, &work->gid);
690 work->gid_op = gid_op;
692 memset(&work->gid_attr, 0, sizeof(work->gid_attr));
694 work->gid_attr.ndev = ndev;
696 queue_work(ib_wq, &work->work);
701 static int inetaddr_event(struct notifier_block *this, unsigned long event,
704 struct sockaddr_in in;
705 struct net_device *ndev;
706 struct in_ifaddr *ifa = ptr;
708 in.sin_family = AF_INET;
709 in.sin_addr.s_addr = ifa->ifa_address;
710 ndev = ifa->ifa_dev->dev;
712 return addr_event(this, event, (struct sockaddr *)&in, ndev);
715 static int inet6addr_event(struct notifier_block *this, unsigned long event,
718 struct sockaddr_in6 in6;
719 struct net_device *ndev;
720 struct inet6_ifaddr *ifa6 = ptr;
722 in6.sin6_family = AF_INET6;
723 in6.sin6_addr = ifa6->addr;
724 ndev = ifa6->idev->dev;
726 return addr_event(this, event, (struct sockaddr *)&in6, ndev);
729 static struct notifier_block nb_netdevice = {
730 .notifier_call = netdevice_event
733 static struct notifier_block nb_inetaddr = {
734 .notifier_call = inetaddr_event
737 static struct notifier_block nb_inet6addr = {
738 .notifier_call = inet6addr_event
741 int __init roce_gid_mgmt_init(void)
743 register_inetaddr_notifier(&nb_inetaddr);
744 if (IS_ENABLED(CONFIG_IPV6))
745 register_inet6addr_notifier(&nb_inet6addr);
746 /* We relay on the netdevice notifier to enumerate all
747 * existing devices in the system. Register to this notifier
748 * last to make sure we will not miss any IP add/del
751 register_netdevice_notifier(&nb_netdevice);
756 void __exit roce_gid_mgmt_cleanup(void)
758 if (IS_ENABLED(CONFIG_IPV6))
759 unregister_inet6addr_notifier(&nb_inet6addr);
760 unregister_inetaddr_notifier(&nb_inetaddr);
761 unregister_netdevice_notifier(&nb_netdevice);
762 /* Ensure all gid deletion tasks complete before we go down,
763 * to avoid any reference to free'd memory. By the time
764 * ib-core is removed, all physical devices have been removed,
765 * so no issue with remaining hardware contexts.