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 static bool is_upper_dev_rcu(struct net_device *dev, struct net_device *upper)
144 struct net_device *_upper = NULL;
145 struct list_head *iter;
147 netdev_for_each_all_upper_dev_rcu(dev, _upper, iter)
151 return _upper == upper;
154 #define REQUIRED_BOND_STATES (BONDING_SLAVE_STATE_ACTIVE | \
155 BONDING_SLAVE_STATE_NA)
156 static int is_eth_port_of_netdev(struct ib_device *ib_dev, u8 port,
157 struct net_device *rdma_ndev, void *cookie)
159 struct net_device *event_ndev = (struct net_device *)cookie;
160 struct net_device *real_dev;
167 real_dev = rdma_vlan_dev_real_dev(event_ndev);
169 real_dev = event_ndev;
171 res = ((is_upper_dev_rcu(rdma_ndev, event_ndev) &&
172 (is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev) &
173 REQUIRED_BOND_STATES)) ||
174 real_dev == rdma_ndev);
180 static int is_eth_port_inactive_slave(struct ib_device *ib_dev, u8 port,
181 struct net_device *rdma_ndev, void *cookie)
183 struct net_device *master_dev;
190 master_dev = netdev_master_upper_dev_get_rcu(rdma_ndev);
191 res = is_eth_active_slave_of_bonding_rcu(rdma_ndev, master_dev) ==
192 BONDING_SLAVE_STATE_INACTIVE;
198 static int pass_all_filter(struct ib_device *ib_dev, u8 port,
199 struct net_device *rdma_ndev, void *cookie)
204 static int upper_device_filter(struct ib_device *ib_dev, u8 port,
205 struct net_device *rdma_ndev, void *cookie)
207 struct net_device *event_ndev = (struct net_device *)cookie;
213 if (rdma_ndev == event_ndev)
217 res = is_upper_dev_rcu(rdma_ndev, event_ndev);
223 static void update_gid_ip(enum gid_op_type gid_op,
224 struct ib_device *ib_dev,
225 u8 port, struct net_device *ndev,
226 struct sockaddr *addr)
229 struct ib_gid_attr gid_attr;
231 rdma_ip2gid(addr, &gid);
232 memset(&gid_attr, 0, sizeof(gid_attr));
233 gid_attr.ndev = ndev;
235 update_gid(gid_op, ib_dev, port, &gid, &gid_attr);
238 static void enum_netdev_default_gids(struct ib_device *ib_dev,
239 u8 port, struct net_device *event_ndev,
240 struct net_device *rdma_ndev)
242 unsigned long gid_type_mask;
246 ((rdma_ndev != event_ndev &&
247 !is_upper_dev_rcu(rdma_ndev, event_ndev)) ||
248 is_eth_active_slave_of_bonding_rcu(rdma_ndev,
249 netdev_master_upper_dev_get_rcu(rdma_ndev)) ==
250 BONDING_SLAVE_STATE_INACTIVE)) {
256 gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
258 ib_cache_gid_set_default_gid(ib_dev, port, rdma_ndev, gid_type_mask,
259 IB_CACHE_GID_DEFAULT_MODE_SET);
262 static void bond_delete_netdev_default_gids(struct ib_device *ib_dev,
264 struct net_device *event_ndev,
265 struct net_device *rdma_ndev)
267 struct net_device *real_dev = rdma_vlan_dev_real_dev(event_ndev);
273 real_dev = event_ndev;
277 if (is_upper_dev_rcu(rdma_ndev, event_ndev) &&
278 is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev) ==
279 BONDING_SLAVE_STATE_INACTIVE) {
280 unsigned long gid_type_mask;
284 gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
286 ib_cache_gid_set_default_gid(ib_dev, port, rdma_ndev,
288 IB_CACHE_GID_DEFAULT_MODE_DELETE);
294 static void enum_netdev_ipv4_ips(struct ib_device *ib_dev,
295 u8 port, struct net_device *ndev)
297 struct in_device *in_dev;
299 struct list_head list;
300 struct sockaddr_in ip;
302 struct sin_list *sin_iter;
303 struct sin_list *sin_temp;
306 if (ndev->reg_state >= NETREG_UNREGISTERING)
310 in_dev = __in_dev_get_rcu(ndev);
317 struct sin_list *entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
320 pr_warn("roce_gid_mgmt: couldn't allocate entry for IPv4 update\n");
323 entry->ip.sin_family = AF_INET;
324 entry->ip.sin_addr.s_addr = ifa->ifa_address;
325 list_add_tail(&entry->list, &sin_list);
330 list_for_each_entry_safe(sin_iter, sin_temp, &sin_list, list) {
331 update_gid_ip(GID_ADD, ib_dev, port, ndev,
332 (struct sockaddr *)&sin_iter->ip);
333 list_del(&sin_iter->list);
338 static void enum_netdev_ipv6_ips(struct ib_device *ib_dev,
339 u8 port, struct net_device *ndev)
341 struct inet6_ifaddr *ifp;
342 struct inet6_dev *in6_dev;
344 struct list_head list;
345 struct sockaddr_in6 sin6;
347 struct sin6_list *sin6_iter;
348 struct sin6_list *sin6_temp;
349 struct ib_gid_attr gid_attr = {.ndev = ndev};
350 LIST_HEAD(sin6_list);
352 if (ndev->reg_state >= NETREG_UNREGISTERING)
355 in6_dev = in6_dev_get(ndev);
359 read_lock_bh(&in6_dev->lock);
360 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
361 struct sin6_list *entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
364 pr_warn("roce_gid_mgmt: couldn't allocate entry for IPv6 update\n");
368 entry->sin6.sin6_family = AF_INET6;
369 entry->sin6.sin6_addr = ifp->addr;
370 list_add_tail(&entry->list, &sin6_list);
372 read_unlock_bh(&in6_dev->lock);
374 in6_dev_put(in6_dev);
376 list_for_each_entry_safe(sin6_iter, sin6_temp, &sin6_list, list) {
379 rdma_ip2gid((struct sockaddr *)&sin6_iter->sin6, &gid);
380 update_gid(GID_ADD, ib_dev, port, &gid, &gid_attr);
381 list_del(&sin6_iter->list);
386 static void _add_netdev_ips(struct ib_device *ib_dev, u8 port,
387 struct net_device *ndev)
389 enum_netdev_ipv4_ips(ib_dev, port, ndev);
390 if (IS_ENABLED(CONFIG_IPV6))
391 enum_netdev_ipv6_ips(ib_dev, port, ndev);
394 static void add_netdev_ips(struct ib_device *ib_dev, u8 port,
395 struct net_device *rdma_ndev, void *cookie)
397 struct net_device *event_ndev = (struct net_device *)cookie;
399 enum_netdev_default_gids(ib_dev, port, event_ndev, rdma_ndev);
400 _add_netdev_ips(ib_dev, port, event_ndev);
403 static void del_netdev_ips(struct ib_device *ib_dev, u8 port,
404 struct net_device *rdma_ndev, void *cookie)
406 struct net_device *event_ndev = (struct net_device *)cookie;
408 ib_cache_gid_del_all_netdev_gids(ib_dev, port, event_ndev);
411 static void enum_all_gids_of_dev_cb(struct ib_device *ib_dev,
413 struct net_device *rdma_ndev,
417 struct net_device *ndev;
419 /* Lock the rtnl to make sure the netdevs does not move under
424 for_each_netdev(net, ndev)
425 if (is_eth_port_of_netdev(ib_dev, port, rdma_ndev, ndev))
426 add_netdev_ips(ib_dev, port, rdma_ndev, ndev);
430 /* This function will rescan all of the network devices in the system
431 * and add their gids, as needed, to the relevant RoCE devices. */
432 int roce_rescan_device(struct ib_device *ib_dev)
434 ib_enum_roce_netdev(ib_dev, pass_all_filter, NULL,
435 enum_all_gids_of_dev_cb, NULL);
440 static void callback_for_addr_gid_device_scan(struct ib_device *device,
442 struct net_device *rdma_ndev,
445 struct update_gid_event_work *parsed = cookie;
447 return update_gid(parsed->gid_op, device,
452 static void handle_netdev_upper(struct ib_device *ib_dev, u8 port,
454 void (*handle_netdev)(struct ib_device *ib_dev,
456 struct net_device *ndev))
458 struct net_device *ndev = (struct net_device *)cookie;
460 struct list_head list;
461 struct net_device *upper;
463 struct net_device *upper;
464 struct list_head *iter;
465 struct upper_list *upper_iter;
466 struct upper_list *upper_temp;
467 LIST_HEAD(upper_list);
470 netdev_for_each_all_upper_dev_rcu(ndev, upper, iter) {
471 struct upper_list *entry = kmalloc(sizeof(*entry),
475 pr_info("roce_gid_mgmt: couldn't allocate entry to delete ndev\n");
479 list_add_tail(&entry->list, &upper_list);
481 entry->upper = upper;
485 handle_netdev(ib_dev, port, ndev);
486 list_for_each_entry_safe(upper_iter, upper_temp, &upper_list,
488 handle_netdev(ib_dev, port, upper_iter->upper);
489 dev_put(upper_iter->upper);
490 list_del(&upper_iter->list);
495 static void _roce_del_all_netdev_gids(struct ib_device *ib_dev, u8 port,
496 struct net_device *event_ndev)
498 ib_cache_gid_del_all_netdev_gids(ib_dev, port, event_ndev);
501 static void del_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
502 struct net_device *rdma_ndev, void *cookie)
504 handle_netdev_upper(ib_dev, port, cookie, _roce_del_all_netdev_gids);
507 static void add_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
508 struct net_device *rdma_ndev, void *cookie)
510 handle_netdev_upper(ib_dev, port, cookie, _add_netdev_ips);
513 static void del_netdev_default_ips_join(struct ib_device *ib_dev, u8 port,
514 struct net_device *rdma_ndev,
517 struct net_device *master_ndev;
520 master_ndev = netdev_master_upper_dev_get_rcu(rdma_ndev);
522 dev_hold(master_ndev);
526 bond_delete_netdev_default_gids(ib_dev, port, master_ndev,
528 dev_put(master_ndev);
532 static void del_netdev_default_ips(struct ib_device *ib_dev, u8 port,
533 struct net_device *rdma_ndev, void *cookie)
535 struct net_device *event_ndev = (struct net_device *)cookie;
537 bond_delete_netdev_default_gids(ib_dev, port, event_ndev, rdma_ndev);
540 /* The following functions operate on all IB devices. netdevice_event and
541 * addr_event execute ib_enum_all_roce_netdevs through a work.
542 * ib_enum_all_roce_netdevs iterates through all IB devices.
545 static void netdevice_event_work_handler(struct work_struct *_work)
547 struct netdev_event_work *work =
548 container_of(_work, struct netdev_event_work, work);
551 for (i = 0; i < ARRAY_SIZE(work->cmds) && work->cmds[i].cb; i++) {
552 ib_enum_all_roce_netdevs(work->cmds[i].filter,
553 work->cmds[i].filter_ndev,
556 dev_put(work->cmds[i].ndev);
557 dev_put(work->cmds[i].filter_ndev);
563 static int netdevice_queue_work(struct netdev_event_work_cmd *cmds,
564 struct net_device *ndev)
567 struct netdev_event_work *ndev_work =
568 kmalloc(sizeof(*ndev_work), GFP_KERNEL);
571 pr_warn("roce_gid_mgmt: can't allocate work for netdevice_event\n");
575 memcpy(ndev_work->cmds, cmds, sizeof(ndev_work->cmds));
576 for (i = 0; i < ARRAY_SIZE(ndev_work->cmds) && ndev_work->cmds[i].cb; i++) {
577 if (!ndev_work->cmds[i].ndev)
578 ndev_work->cmds[i].ndev = ndev;
579 if (!ndev_work->cmds[i].filter_ndev)
580 ndev_work->cmds[i].filter_ndev = ndev;
581 dev_hold(ndev_work->cmds[i].ndev);
582 dev_hold(ndev_work->cmds[i].filter_ndev);
584 INIT_WORK(&ndev_work->work, netdevice_event_work_handler);
586 queue_work(ib_wq, &ndev_work->work);
591 static const struct netdev_event_work_cmd add_cmd = {
592 .cb = add_netdev_ips, .filter = is_eth_port_of_netdev};
593 static const struct netdev_event_work_cmd add_cmd_upper_ips = {
594 .cb = add_netdev_upper_ips, .filter = is_eth_port_of_netdev};
596 static void netdevice_event_changeupper(struct netdev_notifier_changeupper_info *changeupper_info,
597 struct netdev_event_work_cmd *cmds)
599 static const struct netdev_event_work_cmd upper_ips_del_cmd = {
600 .cb = del_netdev_upper_ips, .filter = upper_device_filter};
601 static const struct netdev_event_work_cmd bonding_default_del_cmd = {
602 .cb = del_netdev_default_ips, .filter = is_eth_port_inactive_slave};
604 if (changeupper_info->linking == false) {
605 cmds[0] = upper_ips_del_cmd;
606 cmds[0].ndev = changeupper_info->upper_dev;
609 cmds[0] = bonding_default_del_cmd;
610 cmds[0].ndev = changeupper_info->upper_dev;
611 cmds[1] = add_cmd_upper_ips;
612 cmds[1].ndev = changeupper_info->upper_dev;
613 cmds[1].filter_ndev = changeupper_info->upper_dev;
617 static int netdevice_event(struct notifier_block *this, unsigned long event,
620 static const struct netdev_event_work_cmd del_cmd = {
621 .cb = del_netdev_ips, .filter = pass_all_filter};
622 static const struct netdev_event_work_cmd bonding_default_del_cmd_join = {
623 .cb = del_netdev_default_ips_join, .filter = is_eth_port_inactive_slave};
624 static const struct netdev_event_work_cmd default_del_cmd = {
625 .cb = del_netdev_default_ips, .filter = pass_all_filter};
626 static const struct netdev_event_work_cmd bonding_event_ips_del_cmd = {
627 .cb = del_netdev_upper_ips, .filter = upper_device_filter};
628 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
629 struct netdev_event_work_cmd cmds[ROCE_NETDEV_CALLBACK_SZ] = { {NULL} };
631 if (ndev->type != ARPHRD_ETHER)
635 case NETDEV_REGISTER:
637 cmds[0] = bonding_default_del_cmd_join;
641 case NETDEV_UNREGISTER:
642 if (ndev->reg_state < NETREG_UNREGISTERED)
648 case NETDEV_CHANGEADDR:
649 cmds[0] = default_del_cmd;
653 case NETDEV_CHANGEUPPER:
654 netdevice_event_changeupper(
655 container_of(ptr, struct netdev_notifier_changeupper_info, info),
659 case NETDEV_BONDING_FAILOVER:
660 cmds[0] = bonding_event_ips_del_cmd;
661 cmds[1] = bonding_default_del_cmd_join;
662 cmds[2] = add_cmd_upper_ips;
669 return netdevice_queue_work(cmds, ndev);
672 static void update_gid_event_work_handler(struct work_struct *_work)
674 struct update_gid_event_work *work =
675 container_of(_work, struct update_gid_event_work, work);
677 ib_enum_all_roce_netdevs(is_eth_port_of_netdev, work->gid_attr.ndev,
678 callback_for_addr_gid_device_scan, work);
680 dev_put(work->gid_attr.ndev);
684 static int addr_event(struct notifier_block *this, unsigned long event,
685 struct sockaddr *sa, struct net_device *ndev)
687 struct update_gid_event_work *work;
688 enum gid_op_type gid_op;
690 if (ndev->type != ARPHRD_ETHER)
706 work = kmalloc(sizeof(*work), GFP_ATOMIC);
708 pr_warn("roce_gid_mgmt: Couldn't allocate work for addr_event\n");
712 INIT_WORK(&work->work, update_gid_event_work_handler);
714 rdma_ip2gid(sa, &work->gid);
715 work->gid_op = gid_op;
717 memset(&work->gid_attr, 0, sizeof(work->gid_attr));
719 work->gid_attr.ndev = ndev;
721 queue_work(ib_wq, &work->work);
726 static int inetaddr_event(struct notifier_block *this, unsigned long event,
729 struct sockaddr_in in;
730 struct net_device *ndev;
731 struct in_ifaddr *ifa = ptr;
733 in.sin_family = AF_INET;
734 in.sin_addr.s_addr = ifa->ifa_address;
735 ndev = ifa->ifa_dev->dev;
737 return addr_event(this, event, (struct sockaddr *)&in, ndev);
740 static int inet6addr_event(struct notifier_block *this, unsigned long event,
743 struct sockaddr_in6 in6;
744 struct net_device *ndev;
745 struct inet6_ifaddr *ifa6 = ptr;
747 in6.sin6_family = AF_INET6;
748 in6.sin6_addr = ifa6->addr;
749 ndev = ifa6->idev->dev;
751 return addr_event(this, event, (struct sockaddr *)&in6, ndev);
754 static struct notifier_block nb_netdevice = {
755 .notifier_call = netdevice_event
758 static struct notifier_block nb_inetaddr = {
759 .notifier_call = inetaddr_event
762 static struct notifier_block nb_inet6addr = {
763 .notifier_call = inet6addr_event
766 int __init roce_gid_mgmt_init(void)
768 register_inetaddr_notifier(&nb_inetaddr);
769 if (IS_ENABLED(CONFIG_IPV6))
770 register_inet6addr_notifier(&nb_inet6addr);
771 /* We relay on the netdevice notifier to enumerate all
772 * existing devices in the system. Register to this notifier
773 * last to make sure we will not miss any IP add/del
776 register_netdevice_notifier(&nb_netdevice);
781 void __exit roce_gid_mgmt_cleanup(void)
783 if (IS_ENABLED(CONFIG_IPV6))
784 unregister_inet6addr_notifier(&nb_inet6addr);
785 unregister_inetaddr_notifier(&nb_inetaddr);
786 unregister_netdevice_notifier(&nb_netdevice);
787 /* Ensure all gid deletion tasks complete before we go down,
788 * to avoid any reference to free'd memory. By the time
789 * ib-core is removed, all physical devices have been removed,
790 * so no issue with remaining hardware contexts.