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>
45 static struct workqueue_struct
*gid_cache_wq
;
52 struct update_gid_event_work
{
53 struct work_struct work
;
55 struct ib_gid_attr gid_attr
;
56 enum gid_op_type gid_op
;
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
;
67 struct netdev_event_work
{
68 struct work_struct work
;
69 struct netdev_event_work_cmd cmds
[ROCE_NETDEV_CALLBACK_SZ
];
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
},
80 #define CAP_TO_GID_TABLE_SIZE ARRAY_SIZE(PORT_CAP_TO_GID_TYPE)
82 unsigned long roce_gid_type_mask_support(struct ib_device
*ib_dev
, u8 port
)
85 unsigned int ret_flags
= 0;
87 if (!rdma_protocol_roce(ib_dev
, port
))
88 return 1UL << IB_GID_TYPE_IB
;
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
;
96 EXPORT_SYMBOL(roce_gid_type_mask_support
);
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
)
103 unsigned long gid_type_mask
= roce_gid_type_mask_support(ib_dev
, port
);
105 for (i
= 0; i
< IB_GID_TYPE_SIZE
; i
++) {
106 if ((1UL << i
) & gid_type_mask
) {
107 gid_attr
->gid_type
= i
;
110 ib_cache_gid_add(ib_dev
, port
,
114 ib_cache_gid_del(ib_dev
, port
,
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,
129 static enum bonding_slave_state
is_eth_active_slave_of_bonding_rcu(struct net_device
*dev
,
130 struct net_device
*upper
)
132 if (upper
&& netif_is_bond_master(upper
)) {
133 struct net_device
*pdev
=
134 bond_option_active_slave_get_rcu(netdev_priv(upper
));
137 return dev
== pdev
? BONDING_SLAVE_STATE_ACTIVE
:
138 BONDING_SLAVE_STATE_INACTIVE
;
141 return BONDING_SLAVE_STATE_NA
;
144 #define REQUIRED_BOND_STATES (BONDING_SLAVE_STATE_ACTIVE | \
145 BONDING_SLAVE_STATE_NA)
146 static int is_eth_port_of_netdev(struct ib_device
*ib_dev
, u8 port
,
147 struct net_device
*rdma_ndev
, void *cookie
)
149 struct net_device
*real_dev
;
156 real_dev
= rdma_vlan_dev_real_dev(cookie
);
160 res
= ((rdma_is_upper_dev_rcu(rdma_ndev
, cookie
) &&
161 (is_eth_active_slave_of_bonding_rcu(rdma_ndev
, real_dev
) &
162 REQUIRED_BOND_STATES
)) ||
163 real_dev
== rdma_ndev
);
169 static int is_eth_port_inactive_slave(struct ib_device
*ib_dev
, u8 port
,
170 struct net_device
*rdma_ndev
, void *cookie
)
172 struct net_device
*master_dev
;
179 master_dev
= netdev_master_upper_dev_get_rcu(rdma_ndev
);
180 res
= is_eth_active_slave_of_bonding_rcu(rdma_ndev
, master_dev
) ==
181 BONDING_SLAVE_STATE_INACTIVE
;
187 static int pass_all_filter(struct ib_device
*ib_dev
, u8 port
,
188 struct net_device
*rdma_ndev
, void *cookie
)
193 static int upper_device_filter(struct ib_device
*ib_dev
, u8 port
,
194 struct net_device
*rdma_ndev
, void *cookie
)
201 if (rdma_ndev
== cookie
)
205 res
= rdma_is_upper_dev_rcu(rdma_ndev
, cookie
);
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
);
256 unsigned long gid_type_mask
;
262 real_dev
= event_ndev
;
266 if (((rdma_ndev
!= event_ndev
&&
267 !rdma_is_upper_dev_rcu(rdma_ndev
, event_ndev
)) ||
268 is_eth_active_slave_of_bonding_rcu(rdma_ndev
, real_dev
)
270 BONDING_SLAVE_STATE_INACTIVE
)) {
277 gid_type_mask
= roce_gid_type_mask_support(ib_dev
, port
);
279 ib_cache_gid_set_default_gid(ib_dev
, port
, rdma_ndev
,
281 IB_CACHE_GID_DEFAULT_MODE_DELETE
);
284 static void enum_netdev_ipv4_ips(struct ib_device
*ib_dev
,
285 u8 port
, struct net_device
*ndev
)
287 struct in_device
*in_dev
;
289 struct list_head list
;
290 struct sockaddr_in ip
;
292 struct sin_list
*sin_iter
;
293 struct sin_list
*sin_temp
;
296 if (ndev
->reg_state
>= NETREG_UNREGISTERING
)
300 in_dev
= __in_dev_get_rcu(ndev
);
307 struct sin_list
*entry
= kzalloc(sizeof(*entry
), GFP_ATOMIC
);
312 entry
->ip
.sin_family
= AF_INET
;
313 entry
->ip
.sin_addr
.s_addr
= ifa
->ifa_address
;
314 list_add_tail(&entry
->list
, &sin_list
);
319 list_for_each_entry_safe(sin_iter
, sin_temp
, &sin_list
, list
) {
320 update_gid_ip(GID_ADD
, ib_dev
, port
, ndev
,
321 (struct sockaddr
*)&sin_iter
->ip
);
322 list_del(&sin_iter
->list
);
327 static void enum_netdev_ipv6_ips(struct ib_device
*ib_dev
,
328 u8 port
, struct net_device
*ndev
)
330 struct inet6_ifaddr
*ifp
;
331 struct inet6_dev
*in6_dev
;
333 struct list_head list
;
334 struct sockaddr_in6 sin6
;
336 struct sin6_list
*sin6_iter
;
337 struct sin6_list
*sin6_temp
;
338 struct ib_gid_attr gid_attr
= {.ndev
= ndev
};
339 LIST_HEAD(sin6_list
);
341 if (ndev
->reg_state
>= NETREG_UNREGISTERING
)
344 in6_dev
= in6_dev_get(ndev
);
348 read_lock_bh(&in6_dev
->lock
);
349 list_for_each_entry(ifp
, &in6_dev
->addr_list
, if_list
) {
350 struct sin6_list
*entry
= kzalloc(sizeof(*entry
), GFP_ATOMIC
);
355 entry
->sin6
.sin6_family
= AF_INET6
;
356 entry
->sin6
.sin6_addr
= ifp
->addr
;
357 list_add_tail(&entry
->list
, &sin6_list
);
359 read_unlock_bh(&in6_dev
->lock
);
361 in6_dev_put(in6_dev
);
363 list_for_each_entry_safe(sin6_iter
, sin6_temp
, &sin6_list
, list
) {
366 rdma_ip2gid((struct sockaddr
*)&sin6_iter
->sin6
, &gid
);
367 update_gid(GID_ADD
, ib_dev
, port
, &gid
, &gid_attr
);
368 list_del(&sin6_iter
->list
);
373 static void _add_netdev_ips(struct ib_device
*ib_dev
, u8 port
,
374 struct net_device
*ndev
)
376 enum_netdev_ipv4_ips(ib_dev
, port
, ndev
);
377 if (IS_ENABLED(CONFIG_IPV6
))
378 enum_netdev_ipv6_ips(ib_dev
, port
, ndev
);
381 static void add_netdev_ips(struct ib_device
*ib_dev
, u8 port
,
382 struct net_device
*rdma_ndev
, void *cookie
)
384 enum_netdev_default_gids(ib_dev
, port
, cookie
, rdma_ndev
);
385 _add_netdev_ips(ib_dev
, port
, cookie
);
388 static void del_netdev_ips(struct ib_device
*ib_dev
, u8 port
,
389 struct net_device
*rdma_ndev
, void *cookie
)
391 ib_cache_gid_del_all_netdev_gids(ib_dev
, port
, cookie
);
394 static void enum_all_gids_of_dev_cb(struct ib_device
*ib_dev
,
396 struct net_device
*rdma_ndev
,
400 struct net_device
*ndev
;
402 /* Lock the rtnl to make sure the netdevs does not move under
406 down_read(&net_rwsem
);
408 for_each_netdev(net
, ndev
)
409 if (is_eth_port_of_netdev(ib_dev
, port
, rdma_ndev
, ndev
))
410 add_netdev_ips(ib_dev
, port
, rdma_ndev
, ndev
);
416 * rdma_roce_rescan_device - Rescan all of the network devices in the system
417 * and add their gids, as needed, to the relevant RoCE devices.
419 * @device: the rdma device
421 void rdma_roce_rescan_device(struct ib_device
*ib_dev
)
423 ib_enum_roce_netdev(ib_dev
, pass_all_filter
, NULL
,
424 enum_all_gids_of_dev_cb
, NULL
);
426 EXPORT_SYMBOL(rdma_roce_rescan_device
);
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
,
441 struct list_head list
;
442 struct net_device
*upper
;
445 static int netdev_upper_walk(struct net_device
*upper
, void *data
)
447 struct upper_list
*entry
= kmalloc(sizeof(*entry
), GFP_ATOMIC
);
448 struct list_head
*upper_list
= data
;
453 list_add_tail(&entry
->list
, upper_list
);
455 entry
->upper
= upper
;
460 static void handle_netdev_upper(struct ib_device
*ib_dev
, u8 port
,
462 void (*handle_netdev
)(struct ib_device
*ib_dev
,
464 struct net_device
*ndev
))
466 struct net_device
*ndev
= cookie
;
467 struct upper_list
*upper_iter
;
468 struct upper_list
*upper_temp
;
469 LIST_HEAD(upper_list
);
472 netdev_walk_all_upper_dev_rcu(ndev
, netdev_upper_walk
, &upper_list
);
475 handle_netdev(ib_dev
, port
, ndev
);
476 list_for_each_entry_safe(upper_iter
, upper_temp
, &upper_list
,
478 handle_netdev(ib_dev
, port
, upper_iter
->upper
);
479 dev_put(upper_iter
->upper
);
480 list_del(&upper_iter
->list
);
485 static void _roce_del_all_netdev_gids(struct ib_device
*ib_dev
, u8 port
,
486 struct net_device
*event_ndev
)
488 ib_cache_gid_del_all_netdev_gids(ib_dev
, port
, event_ndev
);
491 static void del_netdev_upper_ips(struct ib_device
*ib_dev
, u8 port
,
492 struct net_device
*rdma_ndev
, void *cookie
)
494 handle_netdev_upper(ib_dev
, port
, cookie
, _roce_del_all_netdev_gids
);
497 static void add_netdev_upper_ips(struct ib_device
*ib_dev
, u8 port
,
498 struct net_device
*rdma_ndev
, void *cookie
)
500 handle_netdev_upper(ib_dev
, port
, cookie
, _add_netdev_ips
);
503 static void del_netdev_default_ips_join(struct ib_device
*ib_dev
, u8 port
,
504 struct net_device
*rdma_ndev
,
507 struct net_device
*master_ndev
;
510 master_ndev
= netdev_master_upper_dev_get_rcu(rdma_ndev
);
512 dev_hold(master_ndev
);
516 bond_delete_netdev_default_gids(ib_dev
, port
, master_ndev
,
518 dev_put(master_ndev
);
522 static void del_netdev_default_ips(struct ib_device
*ib_dev
, u8 port
,
523 struct net_device
*rdma_ndev
, void *cookie
)
525 bond_delete_netdev_default_gids(ib_dev
, port
, cookie
, 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
);
561 memcpy(ndev_work
->cmds
, cmds
, sizeof(ndev_work
->cmds
));
562 for (i
= 0; i
< ARRAY_SIZE(ndev_work
->cmds
) && ndev_work
->cmds
[i
].cb
; i
++) {
563 if (!ndev_work
->cmds
[i
].ndev
)
564 ndev_work
->cmds
[i
].ndev
= ndev
;
565 if (!ndev_work
->cmds
[i
].filter_ndev
)
566 ndev_work
->cmds
[i
].filter_ndev
= ndev
;
567 dev_hold(ndev_work
->cmds
[i
].ndev
);
568 dev_hold(ndev_work
->cmds
[i
].filter_ndev
);
570 INIT_WORK(&ndev_work
->work
, netdevice_event_work_handler
);
572 queue_work(gid_cache_wq
, &ndev_work
->work
);
577 static const struct netdev_event_work_cmd add_cmd
= {
578 .cb
= add_netdev_ips
, .filter
= is_eth_port_of_netdev
};
579 static const struct netdev_event_work_cmd add_cmd_upper_ips
= {
580 .cb
= add_netdev_upper_ips
, .filter
= is_eth_port_of_netdev
};
582 static void netdevice_event_changeupper(struct netdev_notifier_changeupper_info
*changeupper_info
,
583 struct netdev_event_work_cmd
*cmds
)
585 static const struct netdev_event_work_cmd upper_ips_del_cmd
= {
586 .cb
= del_netdev_upper_ips
, .filter
= upper_device_filter
};
587 static const struct netdev_event_work_cmd bonding_default_del_cmd
= {
588 .cb
= del_netdev_default_ips
, .filter
= is_eth_port_inactive_slave
};
590 if (changeupper_info
->linking
== false) {
591 cmds
[0] = upper_ips_del_cmd
;
592 cmds
[0].ndev
= changeupper_info
->upper_dev
;
595 cmds
[0] = bonding_default_del_cmd
;
596 cmds
[0].ndev
= changeupper_info
->upper_dev
;
597 cmds
[1] = add_cmd_upper_ips
;
598 cmds
[1].ndev
= changeupper_info
->upper_dev
;
599 cmds
[1].filter_ndev
= changeupper_info
->upper_dev
;
603 static int netdevice_event(struct notifier_block
*this, unsigned long event
,
606 static const struct netdev_event_work_cmd del_cmd
= {
607 .cb
= del_netdev_ips
, .filter
= pass_all_filter
};
608 static const struct netdev_event_work_cmd bonding_default_del_cmd_join
= {
609 .cb
= del_netdev_default_ips_join
, .filter
= is_eth_port_inactive_slave
};
610 static const struct netdev_event_work_cmd default_del_cmd
= {
611 .cb
= del_netdev_default_ips
, .filter
= pass_all_filter
};
612 static const struct netdev_event_work_cmd bonding_event_ips_del_cmd
= {
613 .cb
= del_netdev_upper_ips
, .filter
= upper_device_filter
};
614 struct net_device
*ndev
= netdev_notifier_info_to_dev(ptr
);
615 struct netdev_event_work_cmd cmds
[ROCE_NETDEV_CALLBACK_SZ
] = { {NULL
} };
617 if (ndev
->type
!= ARPHRD_ETHER
)
621 case NETDEV_REGISTER
:
623 cmds
[0] = bonding_default_del_cmd_join
;
627 case NETDEV_UNREGISTER
:
628 if (ndev
->reg_state
< NETREG_UNREGISTERED
)
634 case NETDEV_CHANGEADDR
:
635 cmds
[0] = default_del_cmd
;
639 case NETDEV_CHANGEUPPER
:
640 netdevice_event_changeupper(
641 container_of(ptr
, struct netdev_notifier_changeupper_info
, info
),
645 case NETDEV_BONDING_FAILOVER
:
646 cmds
[0] = bonding_event_ips_del_cmd
;
647 cmds
[1] = bonding_default_del_cmd_join
;
648 cmds
[2] = add_cmd_upper_ips
;
655 return netdevice_queue_work(cmds
, ndev
);
658 static void update_gid_event_work_handler(struct work_struct
*_work
)
660 struct update_gid_event_work
*work
=
661 container_of(_work
, struct update_gid_event_work
, work
);
663 ib_enum_all_roce_netdevs(is_eth_port_of_netdev
, work
->gid_attr
.ndev
,
664 callback_for_addr_gid_device_scan
, work
);
666 dev_put(work
->gid_attr
.ndev
);
670 static int addr_event(struct notifier_block
*this, unsigned long event
,
671 struct sockaddr
*sa
, struct net_device
*ndev
)
673 struct update_gid_event_work
*work
;
674 enum gid_op_type gid_op
;
676 if (ndev
->type
!= ARPHRD_ETHER
)
692 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
696 INIT_WORK(&work
->work
, update_gid_event_work_handler
);
698 rdma_ip2gid(sa
, &work
->gid
);
699 work
->gid_op
= gid_op
;
701 memset(&work
->gid_attr
, 0, sizeof(work
->gid_attr
));
703 work
->gid_attr
.ndev
= ndev
;
705 queue_work(gid_cache_wq
, &work
->work
);
710 static int inetaddr_event(struct notifier_block
*this, unsigned long event
,
713 struct sockaddr_in in
;
714 struct net_device
*ndev
;
715 struct in_ifaddr
*ifa
= ptr
;
717 in
.sin_family
= AF_INET
;
718 in
.sin_addr
.s_addr
= ifa
->ifa_address
;
719 ndev
= ifa
->ifa_dev
->dev
;
721 return addr_event(this, event
, (struct sockaddr
*)&in
, ndev
);
724 static int inet6addr_event(struct notifier_block
*this, unsigned long event
,
727 struct sockaddr_in6 in6
;
728 struct net_device
*ndev
;
729 struct inet6_ifaddr
*ifa6
= ptr
;
731 in6
.sin6_family
= AF_INET6
;
732 in6
.sin6_addr
= ifa6
->addr
;
733 ndev
= ifa6
->idev
->dev
;
735 return addr_event(this, event
, (struct sockaddr
*)&in6
, ndev
);
738 static struct notifier_block nb_netdevice
= {
739 .notifier_call
= netdevice_event
742 static struct notifier_block nb_inetaddr
= {
743 .notifier_call
= inetaddr_event
746 static struct notifier_block nb_inet6addr
= {
747 .notifier_call
= inet6addr_event
750 int __init
roce_gid_mgmt_init(void)
752 gid_cache_wq
= alloc_ordered_workqueue("gid-cache-wq", 0);
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.
780 destroy_workqueue(gid_cache_wq
);