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 *event_ndev = (struct net_device *)cookie;
148 struct net_device *real_dev;
155 real_dev = rdma_vlan_dev_real_dev(event_ndev);
157 real_dev = event_ndev;
159 res = ((rdma_is_upper_dev_rcu(rdma_ndev, event_ndev) &&
160 (is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev) &
161 REQUIRED_BOND_STATES)) ||
162 real_dev == rdma_ndev);
168 static int is_eth_port_inactive_slave(struct ib_device *ib_dev, u8 port,
169 struct net_device *rdma_ndev, void *cookie)
171 struct net_device *master_dev;
178 master_dev = netdev_master_upper_dev_get_rcu(rdma_ndev);
179 res = is_eth_active_slave_of_bonding_rcu(rdma_ndev, master_dev) ==
180 BONDING_SLAVE_STATE_INACTIVE;
186 static int pass_all_filter(struct ib_device *ib_dev, u8 port,
187 struct net_device *rdma_ndev, void *cookie)
192 static int upper_device_filter(struct ib_device *ib_dev, u8 port,
193 struct net_device *rdma_ndev, void *cookie)
195 struct net_device *event_ndev = (struct net_device *)cookie;
201 if (rdma_ndev == event_ndev)
205 res = rdma_is_upper_dev_rcu(rdma_ndev, event_ndev);
211 static void update_gid_ip(enum gid_op_type gid_op,
212 struct ib_device *ib_dev,
213 u8 port, struct net_device *ndev,
214 struct sockaddr *addr)
217 struct ib_gid_attr gid_attr;
219 rdma_ip2gid(addr, &gid);
220 memset(&gid_attr, 0, sizeof(gid_attr));
221 gid_attr.ndev = ndev;
223 update_gid(gid_op, ib_dev, port, &gid, &gid_attr);
226 static void enum_netdev_default_gids(struct ib_device *ib_dev,
227 u8 port, struct net_device *event_ndev,
228 struct net_device *rdma_ndev)
230 unsigned long gid_type_mask;
234 ((rdma_ndev != event_ndev &&
235 !rdma_is_upper_dev_rcu(rdma_ndev, event_ndev)) ||
236 is_eth_active_slave_of_bonding_rcu(rdma_ndev,
237 netdev_master_upper_dev_get_rcu(rdma_ndev)) ==
238 BONDING_SLAVE_STATE_INACTIVE)) {
244 gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
246 ib_cache_gid_set_default_gid(ib_dev, port, rdma_ndev, gid_type_mask,
247 IB_CACHE_GID_DEFAULT_MODE_SET);
250 static void bond_delete_netdev_default_gids(struct ib_device *ib_dev,
252 struct net_device *event_ndev,
253 struct net_device *rdma_ndev)
255 struct net_device *real_dev = rdma_vlan_dev_real_dev(event_ndev);
261 real_dev = event_ndev;
265 if (rdma_is_upper_dev_rcu(rdma_ndev, event_ndev) &&
266 is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev) ==
267 BONDING_SLAVE_STATE_INACTIVE) {
268 unsigned long gid_type_mask;
272 gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
274 ib_cache_gid_set_default_gid(ib_dev, port, rdma_ndev,
276 IB_CACHE_GID_DEFAULT_MODE_DELETE);
282 static void enum_netdev_ipv4_ips(struct ib_device *ib_dev,
283 u8 port, struct net_device *ndev)
285 struct in_device *in_dev;
287 struct list_head list;
288 struct sockaddr_in ip;
290 struct sin_list *sin_iter;
291 struct sin_list *sin_temp;
294 if (ndev->reg_state >= NETREG_UNREGISTERING)
298 in_dev = __in_dev_get_rcu(ndev);
305 struct sin_list *entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
308 pr_warn("roce_gid_mgmt: couldn't allocate entry for IPv4 update\n");
311 entry->ip.sin_family = AF_INET;
312 entry->ip.sin_addr.s_addr = ifa->ifa_address;
313 list_add_tail(&entry->list, &sin_list);
318 list_for_each_entry_safe(sin_iter, sin_temp, &sin_list, list) {
319 update_gid_ip(GID_ADD, ib_dev, port, ndev,
320 (struct sockaddr *)&sin_iter->ip);
321 list_del(&sin_iter->list);
326 static void enum_netdev_ipv6_ips(struct ib_device *ib_dev,
327 u8 port, struct net_device *ndev)
329 struct inet6_ifaddr *ifp;
330 struct inet6_dev *in6_dev;
332 struct list_head list;
333 struct sockaddr_in6 sin6;
335 struct sin6_list *sin6_iter;
336 struct sin6_list *sin6_temp;
337 struct ib_gid_attr gid_attr = {.ndev = ndev};
338 LIST_HEAD(sin6_list);
340 if (ndev->reg_state >= NETREG_UNREGISTERING)
343 in6_dev = in6_dev_get(ndev);
347 read_lock_bh(&in6_dev->lock);
348 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
349 struct sin6_list *entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
352 pr_warn("roce_gid_mgmt: couldn't allocate entry for IPv6 update\n");
356 entry->sin6.sin6_family = AF_INET6;
357 entry->sin6.sin6_addr = ifp->addr;
358 list_add_tail(&entry->list, &sin6_list);
360 read_unlock_bh(&in6_dev->lock);
362 in6_dev_put(in6_dev);
364 list_for_each_entry_safe(sin6_iter, sin6_temp, &sin6_list, list) {
367 rdma_ip2gid((struct sockaddr *)&sin6_iter->sin6, &gid);
368 update_gid(GID_ADD, ib_dev, port, &gid, &gid_attr);
369 list_del(&sin6_iter->list);
374 static void _add_netdev_ips(struct ib_device *ib_dev, u8 port,
375 struct net_device *ndev)
377 enum_netdev_ipv4_ips(ib_dev, port, ndev);
378 if (IS_ENABLED(CONFIG_IPV6))
379 enum_netdev_ipv6_ips(ib_dev, port, ndev);
382 static void add_netdev_ips(struct ib_device *ib_dev, u8 port,
383 struct net_device *rdma_ndev, void *cookie)
385 struct net_device *event_ndev = (struct net_device *)cookie;
387 enum_netdev_default_gids(ib_dev, port, event_ndev, rdma_ndev);
388 _add_netdev_ips(ib_dev, port, event_ndev);
391 static void del_netdev_ips(struct ib_device *ib_dev, u8 port,
392 struct net_device *rdma_ndev, void *cookie)
394 struct net_device *event_ndev = (struct net_device *)cookie;
396 ib_cache_gid_del_all_netdev_gids(ib_dev, port, event_ndev);
399 static void enum_all_gids_of_dev_cb(struct ib_device *ib_dev,
401 struct net_device *rdma_ndev,
405 struct net_device *ndev;
407 /* Lock the rtnl to make sure the netdevs does not move under
412 for_each_netdev(net, ndev)
413 if (is_eth_port_of_netdev(ib_dev, port, rdma_ndev, ndev))
414 add_netdev_ips(ib_dev, port, rdma_ndev, ndev);
418 /* This function will rescan all of the network devices in the system
419 * and add their gids, as needed, to the relevant RoCE devices. */
420 int roce_rescan_device(struct ib_device *ib_dev)
422 ib_enum_roce_netdev(ib_dev, pass_all_filter, NULL,
423 enum_all_gids_of_dev_cb, NULL);
428 static void callback_for_addr_gid_device_scan(struct ib_device *device,
430 struct net_device *rdma_ndev,
433 struct update_gid_event_work *parsed = cookie;
435 return update_gid(parsed->gid_op, device,
440 static void handle_netdev_upper(struct ib_device *ib_dev, u8 port,
442 void (*handle_netdev)(struct ib_device *ib_dev,
444 struct net_device *ndev))
446 struct net_device *ndev = (struct net_device *)cookie;
448 struct list_head list;
449 struct net_device *upper;
451 struct net_device *upper;
452 struct list_head *iter;
453 struct upper_list *upper_iter;
454 struct upper_list *upper_temp;
455 LIST_HEAD(upper_list);
458 netdev_for_each_all_upper_dev_rcu(ndev, upper, iter) {
459 struct upper_list *entry = kmalloc(sizeof(*entry),
463 pr_info("roce_gid_mgmt: couldn't allocate entry to delete ndev\n");
467 list_add_tail(&entry->list, &upper_list);
469 entry->upper = upper;
473 handle_netdev(ib_dev, port, ndev);
474 list_for_each_entry_safe(upper_iter, upper_temp, &upper_list,
476 handle_netdev(ib_dev, port, upper_iter->upper);
477 dev_put(upper_iter->upper);
478 list_del(&upper_iter->list);
483 static void _roce_del_all_netdev_gids(struct ib_device *ib_dev, u8 port,
484 struct net_device *event_ndev)
486 ib_cache_gid_del_all_netdev_gids(ib_dev, port, event_ndev);
489 static void del_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
490 struct net_device *rdma_ndev, void *cookie)
492 handle_netdev_upper(ib_dev, port, cookie, _roce_del_all_netdev_gids);
495 static void add_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
496 struct net_device *rdma_ndev, void *cookie)
498 handle_netdev_upper(ib_dev, port, cookie, _add_netdev_ips);
501 static void del_netdev_default_ips_join(struct ib_device *ib_dev, u8 port,
502 struct net_device *rdma_ndev,
505 struct net_device *master_ndev;
508 master_ndev = netdev_master_upper_dev_get_rcu(rdma_ndev);
510 dev_hold(master_ndev);
514 bond_delete_netdev_default_gids(ib_dev, port, master_ndev,
516 dev_put(master_ndev);
520 static void del_netdev_default_ips(struct ib_device *ib_dev, u8 port,
521 struct net_device *rdma_ndev, void *cookie)
523 struct net_device *event_ndev = (struct net_device *)cookie;
525 bond_delete_netdev_default_gids(ib_dev, port, event_ndev, rdma_ndev);
528 /* The following functions operate on all IB devices. netdevice_event and
529 * addr_event execute ib_enum_all_roce_netdevs through a work.
530 * ib_enum_all_roce_netdevs iterates through all IB devices.
533 static void netdevice_event_work_handler(struct work_struct *_work)
535 struct netdev_event_work *work =
536 container_of(_work, struct netdev_event_work, work);
539 for (i = 0; i < ARRAY_SIZE(work->cmds) && work->cmds[i].cb; i++) {
540 ib_enum_all_roce_netdevs(work->cmds[i].filter,
541 work->cmds[i].filter_ndev,
544 dev_put(work->cmds[i].ndev);
545 dev_put(work->cmds[i].filter_ndev);
551 static int netdevice_queue_work(struct netdev_event_work_cmd *cmds,
552 struct net_device *ndev)
555 struct netdev_event_work *ndev_work =
556 kmalloc(sizeof(*ndev_work), GFP_KERNEL);
559 pr_warn("roce_gid_mgmt: can't allocate work for netdevice_event\n");
563 memcpy(ndev_work->cmds, cmds, sizeof(ndev_work->cmds));
564 for (i = 0; i < ARRAY_SIZE(ndev_work->cmds) && ndev_work->cmds[i].cb; i++) {
565 if (!ndev_work->cmds[i].ndev)
566 ndev_work->cmds[i].ndev = ndev;
567 if (!ndev_work->cmds[i].filter_ndev)
568 ndev_work->cmds[i].filter_ndev = ndev;
569 dev_hold(ndev_work->cmds[i].ndev);
570 dev_hold(ndev_work->cmds[i].filter_ndev);
572 INIT_WORK(&ndev_work->work, netdevice_event_work_handler);
574 queue_work(ib_wq, &ndev_work->work);
579 static const struct netdev_event_work_cmd add_cmd = {
580 .cb = add_netdev_ips, .filter = is_eth_port_of_netdev};
581 static const struct netdev_event_work_cmd add_cmd_upper_ips = {
582 .cb = add_netdev_upper_ips, .filter = is_eth_port_of_netdev};
584 static void netdevice_event_changeupper(struct netdev_notifier_changeupper_info *changeupper_info,
585 struct netdev_event_work_cmd *cmds)
587 static const struct netdev_event_work_cmd upper_ips_del_cmd = {
588 .cb = del_netdev_upper_ips, .filter = upper_device_filter};
589 static const struct netdev_event_work_cmd bonding_default_del_cmd = {
590 .cb = del_netdev_default_ips, .filter = is_eth_port_inactive_slave};
592 if (changeupper_info->linking == false) {
593 cmds[0] = upper_ips_del_cmd;
594 cmds[0].ndev = changeupper_info->upper_dev;
597 cmds[0] = bonding_default_del_cmd;
598 cmds[0].ndev = changeupper_info->upper_dev;
599 cmds[1] = add_cmd_upper_ips;
600 cmds[1].ndev = changeupper_info->upper_dev;
601 cmds[1].filter_ndev = changeupper_info->upper_dev;
605 static int netdevice_event(struct notifier_block *this, unsigned long event,
608 static const struct netdev_event_work_cmd del_cmd = {
609 .cb = del_netdev_ips, .filter = pass_all_filter};
610 static const struct netdev_event_work_cmd bonding_default_del_cmd_join = {
611 .cb = del_netdev_default_ips_join, .filter = is_eth_port_inactive_slave};
612 static const struct netdev_event_work_cmd default_del_cmd = {
613 .cb = del_netdev_default_ips, .filter = pass_all_filter};
614 static const struct netdev_event_work_cmd bonding_event_ips_del_cmd = {
615 .cb = del_netdev_upper_ips, .filter = upper_device_filter};
616 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
617 struct netdev_event_work_cmd cmds[ROCE_NETDEV_CALLBACK_SZ] = { {NULL} };
619 if (ndev->type != ARPHRD_ETHER)
623 case NETDEV_REGISTER:
625 cmds[0] = bonding_default_del_cmd_join;
629 case NETDEV_UNREGISTER:
630 if (ndev->reg_state < NETREG_UNREGISTERED)
636 case NETDEV_CHANGEADDR:
637 cmds[0] = default_del_cmd;
641 case NETDEV_CHANGEUPPER:
642 netdevice_event_changeupper(
643 container_of(ptr, struct netdev_notifier_changeupper_info, info),
647 case NETDEV_BONDING_FAILOVER:
648 cmds[0] = bonding_event_ips_del_cmd;
649 cmds[1] = bonding_default_del_cmd_join;
650 cmds[2] = add_cmd_upper_ips;
657 return netdevice_queue_work(cmds, ndev);
660 static void update_gid_event_work_handler(struct work_struct *_work)
662 struct update_gid_event_work *work =
663 container_of(_work, struct update_gid_event_work, work);
665 ib_enum_all_roce_netdevs(is_eth_port_of_netdev, work->gid_attr.ndev,
666 callback_for_addr_gid_device_scan, work);
668 dev_put(work->gid_attr.ndev);
672 static int addr_event(struct notifier_block *this, unsigned long event,
673 struct sockaddr *sa, struct net_device *ndev)
675 struct update_gid_event_work *work;
676 enum gid_op_type gid_op;
678 if (ndev->type != ARPHRD_ETHER)
694 work = kmalloc(sizeof(*work), GFP_ATOMIC);
696 pr_warn("roce_gid_mgmt: Couldn't allocate work for addr_event\n");
700 INIT_WORK(&work->work, update_gid_event_work_handler);
702 rdma_ip2gid(sa, &work->gid);
703 work->gid_op = gid_op;
705 memset(&work->gid_attr, 0, sizeof(work->gid_attr));
707 work->gid_attr.ndev = ndev;
709 queue_work(ib_wq, &work->work);
714 static int inetaddr_event(struct notifier_block *this, unsigned long event,
717 struct sockaddr_in in;
718 struct net_device *ndev;
719 struct in_ifaddr *ifa = ptr;
721 in.sin_family = AF_INET;
722 in.sin_addr.s_addr = ifa->ifa_address;
723 ndev = ifa->ifa_dev->dev;
725 return addr_event(this, event, (struct sockaddr *)&in, ndev);
728 static int inet6addr_event(struct notifier_block *this, unsigned long event,
731 struct sockaddr_in6 in6;
732 struct net_device *ndev;
733 struct inet6_ifaddr *ifa6 = ptr;
735 in6.sin6_family = AF_INET6;
736 in6.sin6_addr = ifa6->addr;
737 ndev = ifa6->idev->dev;
739 return addr_event(this, event, (struct sockaddr *)&in6, ndev);
742 static struct notifier_block nb_netdevice = {
743 .notifier_call = netdevice_event
746 static struct notifier_block nb_inetaddr = {
747 .notifier_call = inetaddr_event
750 static struct notifier_block nb_inet6addr = {
751 .notifier_call = inet6addr_event
754 int __init roce_gid_mgmt_init(void)
756 register_inetaddr_notifier(&nb_inetaddr);
757 if (IS_ENABLED(CONFIG_IPV6))
758 register_inet6addr_notifier(&nb_inet6addr);
759 /* We relay on the netdevice notifier to enumerate all
760 * existing devices in the system. Register to this notifier
761 * last to make sure we will not miss any IP add/del
764 register_netdevice_notifier(&nb_netdevice);
769 void __exit roce_gid_mgmt_cleanup(void)
771 if (IS_ENABLED(CONFIG_IPV6))
772 unregister_inet6addr_notifier(&nb_inet6addr);
773 unregister_inetaddr_notifier(&nb_inetaddr);
774 unregister_netdevice_notifier(&nb_netdevice);
775 /* Ensure all gid deletion tasks complete before we go down,
776 * to avoid any reference to free'd memory. By the time
777 * ib-core is removed, all physical devices have been removed,
778 * so no issue with remaining hardware contexts.