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
);
310 entry
->ip
.sin_family
= AF_INET
;
311 entry
->ip
.sin_addr
.s_addr
= ifa
->ifa_address
;
312 list_add_tail(&entry
->list
, &sin_list
);
317 list_for_each_entry_safe(sin_iter
, sin_temp
, &sin_list
, list
) {
318 update_gid_ip(GID_ADD
, ib_dev
, port
, ndev
,
319 (struct sockaddr
*)&sin_iter
->ip
);
320 list_del(&sin_iter
->list
);
325 static void enum_netdev_ipv6_ips(struct ib_device
*ib_dev
,
326 u8 port
, struct net_device
*ndev
)
328 struct inet6_ifaddr
*ifp
;
329 struct inet6_dev
*in6_dev
;
331 struct list_head list
;
332 struct sockaddr_in6 sin6
;
334 struct sin6_list
*sin6_iter
;
335 struct sin6_list
*sin6_temp
;
336 struct ib_gid_attr gid_attr
= {.ndev
= ndev
};
337 LIST_HEAD(sin6_list
);
339 if (ndev
->reg_state
>= NETREG_UNREGISTERING
)
342 in6_dev
= in6_dev_get(ndev
);
346 read_lock_bh(&in6_dev
->lock
);
347 list_for_each_entry(ifp
, &in6_dev
->addr_list
, if_list
) {
348 struct sin6_list
*entry
= kzalloc(sizeof(*entry
), GFP_ATOMIC
);
353 entry
->sin6
.sin6_family
= AF_INET6
;
354 entry
->sin6
.sin6_addr
= ifp
->addr
;
355 list_add_tail(&entry
->list
, &sin6_list
);
357 read_unlock_bh(&in6_dev
->lock
);
359 in6_dev_put(in6_dev
);
361 list_for_each_entry_safe(sin6_iter
, sin6_temp
, &sin6_list
, list
) {
364 rdma_ip2gid((struct sockaddr
*)&sin6_iter
->sin6
, &gid
);
365 update_gid(GID_ADD
, ib_dev
, port
, &gid
, &gid_attr
);
366 list_del(&sin6_iter
->list
);
371 static void _add_netdev_ips(struct ib_device
*ib_dev
, u8 port
,
372 struct net_device
*ndev
)
374 enum_netdev_ipv4_ips(ib_dev
, port
, ndev
);
375 if (IS_ENABLED(CONFIG_IPV6
))
376 enum_netdev_ipv6_ips(ib_dev
, port
, ndev
);
379 static void add_netdev_ips(struct ib_device
*ib_dev
, u8 port
,
380 struct net_device
*rdma_ndev
, void *cookie
)
382 struct net_device
*event_ndev
= (struct net_device
*)cookie
;
384 enum_netdev_default_gids(ib_dev
, port
, event_ndev
, rdma_ndev
);
385 _add_netdev_ips(ib_dev
, port
, event_ndev
);
388 static void del_netdev_ips(struct ib_device
*ib_dev
, u8 port
,
389 struct net_device
*rdma_ndev
, void *cookie
)
391 struct net_device
*event_ndev
= (struct net_device
*)cookie
;
393 ib_cache_gid_del_all_netdev_gids(ib_dev
, port
, event_ndev
);
396 static void enum_all_gids_of_dev_cb(struct ib_device
*ib_dev
,
398 struct net_device
*rdma_ndev
,
402 struct net_device
*ndev
;
404 /* Lock the rtnl to make sure the netdevs does not move under
409 for_each_netdev(net
, ndev
)
410 if (is_eth_port_of_netdev(ib_dev
, port
, rdma_ndev
, ndev
))
411 add_netdev_ips(ib_dev
, port
, rdma_ndev
, ndev
);
415 /* This function will rescan all of the network devices in the system
416 * and add their gids, as needed, to the relevant RoCE devices. */
417 int roce_rescan_device(struct ib_device
*ib_dev
)
419 ib_enum_roce_netdev(ib_dev
, pass_all_filter
, NULL
,
420 enum_all_gids_of_dev_cb
, NULL
);
425 static void callback_for_addr_gid_device_scan(struct ib_device
*device
,
427 struct net_device
*rdma_ndev
,
430 struct update_gid_event_work
*parsed
= cookie
;
432 return update_gid(parsed
->gid_op
, device
,
438 struct list_head list
;
439 struct net_device
*upper
;
442 static int netdev_upper_walk(struct net_device
*upper
, void *data
)
444 struct upper_list
*entry
= kmalloc(sizeof(*entry
), GFP_ATOMIC
);
445 struct list_head
*upper_list
= data
;
450 list_add_tail(&entry
->list
, upper_list
);
452 entry
->upper
= upper
;
457 static void handle_netdev_upper(struct ib_device
*ib_dev
, u8 port
,
459 void (*handle_netdev
)(struct ib_device
*ib_dev
,
461 struct net_device
*ndev
))
463 struct net_device
*ndev
= (struct net_device
*)cookie
;
464 struct upper_list
*upper_iter
;
465 struct upper_list
*upper_temp
;
466 LIST_HEAD(upper_list
);
469 netdev_walk_all_upper_dev_rcu(ndev
, netdev_upper_walk
, &upper_list
);
472 handle_netdev(ib_dev
, port
, ndev
);
473 list_for_each_entry_safe(upper_iter
, upper_temp
, &upper_list
,
475 handle_netdev(ib_dev
, port
, upper_iter
->upper
);
476 dev_put(upper_iter
->upper
);
477 list_del(&upper_iter
->list
);
482 static void _roce_del_all_netdev_gids(struct ib_device
*ib_dev
, u8 port
,
483 struct net_device
*event_ndev
)
485 ib_cache_gid_del_all_netdev_gids(ib_dev
, port
, event_ndev
);
488 static void del_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
, _roce_del_all_netdev_gids
);
494 static void add_netdev_upper_ips(struct ib_device
*ib_dev
, u8 port
,
495 struct net_device
*rdma_ndev
, void *cookie
)
497 handle_netdev_upper(ib_dev
, port
, cookie
, _add_netdev_ips
);
500 static void del_netdev_default_ips_join(struct ib_device
*ib_dev
, u8 port
,
501 struct net_device
*rdma_ndev
,
504 struct net_device
*master_ndev
;
507 master_ndev
= netdev_master_upper_dev_get_rcu(rdma_ndev
);
509 dev_hold(master_ndev
);
513 bond_delete_netdev_default_gids(ib_dev
, port
, master_ndev
,
515 dev_put(master_ndev
);
519 static void del_netdev_default_ips(struct ib_device
*ib_dev
, u8 port
,
520 struct net_device
*rdma_ndev
, void *cookie
)
522 struct net_device
*event_ndev
= (struct net_device
*)cookie
;
524 bond_delete_netdev_default_gids(ib_dev
, port
, event_ndev
, rdma_ndev
);
527 /* The following functions operate on all IB devices. netdevice_event and
528 * addr_event execute ib_enum_all_roce_netdevs through a work.
529 * ib_enum_all_roce_netdevs iterates through all IB devices.
532 static void netdevice_event_work_handler(struct work_struct
*_work
)
534 struct netdev_event_work
*work
=
535 container_of(_work
, struct netdev_event_work
, work
);
538 for (i
= 0; i
< ARRAY_SIZE(work
->cmds
) && work
->cmds
[i
].cb
; i
++) {
539 ib_enum_all_roce_netdevs(work
->cmds
[i
].filter
,
540 work
->cmds
[i
].filter_ndev
,
543 dev_put(work
->cmds
[i
].ndev
);
544 dev_put(work
->cmds
[i
].filter_ndev
);
550 static int netdevice_queue_work(struct netdev_event_work_cmd
*cmds
,
551 struct net_device
*ndev
)
554 struct netdev_event_work
*ndev_work
=
555 kmalloc(sizeof(*ndev_work
), GFP_KERNEL
);
560 memcpy(ndev_work
->cmds
, cmds
, sizeof(ndev_work
->cmds
));
561 for (i
= 0; i
< ARRAY_SIZE(ndev_work
->cmds
) && ndev_work
->cmds
[i
].cb
; i
++) {
562 if (!ndev_work
->cmds
[i
].ndev
)
563 ndev_work
->cmds
[i
].ndev
= ndev
;
564 if (!ndev_work
->cmds
[i
].filter_ndev
)
565 ndev_work
->cmds
[i
].filter_ndev
= ndev
;
566 dev_hold(ndev_work
->cmds
[i
].ndev
);
567 dev_hold(ndev_work
->cmds
[i
].filter_ndev
);
569 INIT_WORK(&ndev_work
->work
, netdevice_event_work_handler
);
571 queue_work(ib_wq
, &ndev_work
->work
);
576 static const struct netdev_event_work_cmd add_cmd
= {
577 .cb
= add_netdev_ips
, .filter
= is_eth_port_of_netdev
};
578 static const struct netdev_event_work_cmd add_cmd_upper_ips
= {
579 .cb
= add_netdev_upper_ips
, .filter
= is_eth_port_of_netdev
};
581 static void netdevice_event_changeupper(struct netdev_notifier_changeupper_info
*changeupper_info
,
582 struct netdev_event_work_cmd
*cmds
)
584 static const struct netdev_event_work_cmd upper_ips_del_cmd
= {
585 .cb
= del_netdev_upper_ips
, .filter
= upper_device_filter
};
586 static const struct netdev_event_work_cmd bonding_default_del_cmd
= {
587 .cb
= del_netdev_default_ips
, .filter
= is_eth_port_inactive_slave
};
589 if (changeupper_info
->linking
== false) {
590 cmds
[0] = upper_ips_del_cmd
;
591 cmds
[0].ndev
= changeupper_info
->upper_dev
;
594 cmds
[0] = bonding_default_del_cmd
;
595 cmds
[0].ndev
= changeupper_info
->upper_dev
;
596 cmds
[1] = add_cmd_upper_ips
;
597 cmds
[1].ndev
= changeupper_info
->upper_dev
;
598 cmds
[1].filter_ndev
= changeupper_info
->upper_dev
;
602 static int netdevice_event(struct notifier_block
*this, unsigned long event
,
605 static const struct netdev_event_work_cmd del_cmd
= {
606 .cb
= del_netdev_ips
, .filter
= pass_all_filter
};
607 static const struct netdev_event_work_cmd bonding_default_del_cmd_join
= {
608 .cb
= del_netdev_default_ips_join
, .filter
= is_eth_port_inactive_slave
};
609 static const struct netdev_event_work_cmd default_del_cmd
= {
610 .cb
= del_netdev_default_ips
, .filter
= pass_all_filter
};
611 static const struct netdev_event_work_cmd bonding_event_ips_del_cmd
= {
612 .cb
= del_netdev_upper_ips
, .filter
= upper_device_filter
};
613 struct net_device
*ndev
= netdev_notifier_info_to_dev(ptr
);
614 struct netdev_event_work_cmd cmds
[ROCE_NETDEV_CALLBACK_SZ
] = { {NULL
} };
616 if (ndev
->type
!= ARPHRD_ETHER
)
620 case NETDEV_REGISTER
:
622 cmds
[0] = bonding_default_del_cmd_join
;
626 case NETDEV_UNREGISTER
:
627 if (ndev
->reg_state
< NETREG_UNREGISTERED
)
633 case NETDEV_CHANGEADDR
:
634 cmds
[0] = default_del_cmd
;
638 case NETDEV_CHANGEUPPER
:
639 netdevice_event_changeupper(
640 container_of(ptr
, struct netdev_notifier_changeupper_info
, info
),
644 case NETDEV_BONDING_FAILOVER
:
645 cmds
[0] = bonding_event_ips_del_cmd
;
646 cmds
[1] = bonding_default_del_cmd_join
;
647 cmds
[2] = add_cmd_upper_ips
;
654 return netdevice_queue_work(cmds
, ndev
);
657 static void update_gid_event_work_handler(struct work_struct
*_work
)
659 struct update_gid_event_work
*work
=
660 container_of(_work
, struct update_gid_event_work
, work
);
662 ib_enum_all_roce_netdevs(is_eth_port_of_netdev
, work
->gid_attr
.ndev
,
663 callback_for_addr_gid_device_scan
, work
);
665 dev_put(work
->gid_attr
.ndev
);
669 static int addr_event(struct notifier_block
*this, unsigned long event
,
670 struct sockaddr
*sa
, struct net_device
*ndev
)
672 struct update_gid_event_work
*work
;
673 enum gid_op_type gid_op
;
675 if (ndev
->type
!= ARPHRD_ETHER
)
691 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
695 INIT_WORK(&work
->work
, update_gid_event_work_handler
);
697 rdma_ip2gid(sa
, &work
->gid
);
698 work
->gid_op
= gid_op
;
700 memset(&work
->gid_attr
, 0, sizeof(work
->gid_attr
));
702 work
->gid_attr
.ndev
= ndev
;
704 queue_work(ib_wq
, &work
->work
);
709 static int inetaddr_event(struct notifier_block
*this, unsigned long event
,
712 struct sockaddr_in in
;
713 struct net_device
*ndev
;
714 struct in_ifaddr
*ifa
= ptr
;
716 in
.sin_family
= AF_INET
;
717 in
.sin_addr
.s_addr
= ifa
->ifa_address
;
718 ndev
= ifa
->ifa_dev
->dev
;
720 return addr_event(this, event
, (struct sockaddr
*)&in
, ndev
);
723 static int inet6addr_event(struct notifier_block
*this, unsigned long event
,
726 struct sockaddr_in6 in6
;
727 struct net_device
*ndev
;
728 struct inet6_ifaddr
*ifa6
= ptr
;
730 in6
.sin6_family
= AF_INET6
;
731 in6
.sin6_addr
= ifa6
->addr
;
732 ndev
= ifa6
->idev
->dev
;
734 return addr_event(this, event
, (struct sockaddr
*)&in6
, ndev
);
737 static struct notifier_block nb_netdevice
= {
738 .notifier_call
= netdevice_event
741 static struct notifier_block nb_inetaddr
= {
742 .notifier_call
= inetaddr_event
745 static struct notifier_block nb_inet6addr
= {
746 .notifier_call
= inet6addr_event
749 int __init
roce_gid_mgmt_init(void)
751 register_inetaddr_notifier(&nb_inetaddr
);
752 if (IS_ENABLED(CONFIG_IPV6
))
753 register_inet6addr_notifier(&nb_inet6addr
);
754 /* We relay on the netdevice notifier to enumerate all
755 * existing devices in the system. Register to this notifier
756 * last to make sure we will not miss any IP add/del
759 register_netdevice_notifier(&nb_netdevice
);
764 void __exit
roce_gid_mgmt_cleanup(void)
766 if (IS_ENABLED(CONFIG_IPV6
))
767 unregister_inet6addr_notifier(&nb_inet6addr
);
768 unregister_inetaddr_notifier(&nb_inetaddr
);
769 unregister_netdevice_notifier(&nb_netdevice
);
770 /* Ensure all gid deletion tasks complete before we go down,
771 * to avoid any reference to free'd memory. By the time
772 * ib-core is removed, all physical devices have been removed,
773 * so no issue with remaining hardware contexts.