1 /*
2 * Copyright (c) 2015, Mellanox Technologies inc. All rights reserved.
3 *
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:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
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.
22 *
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
30 * SOFTWARE.
31 */
32
33 #include "core_priv.h"
34
35 #include <linux/in.h>
36 #include <linux/in6.h>
37
38 /* For in6_dev_get/in6_dev_put */
39 #include <net/addrconf.h>
40 #include <net/bonding.h>
41
42 #include <rdma/ib_cache.h>
43 #include <rdma/ib_addr.h>
44
45 static struct workqueue_struct *gid_cache_wq;
46
47 static struct workqueue_struct *gid_cache_wq;
48
49 enum gid_op_type {
50 GID_DEL = 0,
51 GID_ADD
52 };
53
54 struct update_gid_event_work {
55 struct work_struct work;
56 union ib_gid gid;
57 struct ib_gid_attr gid_attr;
58 enum gid_op_type gid_op;
59 };
60
61 #define ROCE_NETDEV_CALLBACK_SZ 3
62 struct netdev_event_work_cmd {
63 roce_netdev_callback cb;
64 roce_netdev_filter filter;
65 struct net_device *ndev;
66 struct net_device *filter_ndev;
67 };
68
69 struct netdev_event_work {
70 struct work_struct work;
71 struct netdev_event_work_cmd cmds[ROCE_NETDEV_CALLBACK_SZ];
72 };
73
74 static const struct {
75 bool (*is_supported)(const struct ib_device *device, u8 port_num);
76 enum ib_gid_type gid_type;
77 } PORT_CAP_TO_GID_TYPE[] = {
78 {rdma_protocol_roce_eth_encap, IB_GID_TYPE_ROCE},
79 {rdma_protocol_roce_udp_encap, IB_GID_TYPE_ROCE_UDP_ENCAP},
80 };
81
82 #define CAP_TO_GID_TABLE_SIZE ARRAY_SIZE(PORT_CAP_TO_GID_TYPE)
83
roce_gid_type_mask_support(struct ib_device * ib_dev,u8 port)84 unsigned long roce_gid_type_mask_support(struct ib_device *ib_dev, u8 port)
85 {
86 int i;
87 unsigned int ret_flags = 0;
88
89 if (!rdma_protocol_roce(ib_dev, port))
90 return 1UL << IB_GID_TYPE_IB;
91
92 for (i = 0; i < CAP_TO_GID_TABLE_SIZE; i++)
93 if (PORT_CAP_TO_GID_TYPE[i].is_supported(ib_dev, port))
94 ret_flags |= 1UL << PORT_CAP_TO_GID_TYPE[i].gid_type;
95
96 return ret_flags;
97 }
98 EXPORT_SYMBOL(roce_gid_type_mask_support);
99
update_gid(enum gid_op_type gid_op,struct ib_device * ib_dev,u8 port,union ib_gid * gid,struct ib_gid_attr * gid_attr)100 static void update_gid(enum gid_op_type gid_op, struct ib_device *ib_dev,
101 u8 port, union ib_gid *gid,
102 struct ib_gid_attr *gid_attr)
103 {
104 int i;
105 unsigned long gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
106
107 for (i = 0; i < IB_GID_TYPE_SIZE; i++) {
108 if ((1UL << i) & gid_type_mask) {
109 gid_attr->gid_type = i;
110 switch (gid_op) {
111 case GID_ADD:
112 ib_cache_gid_add(ib_dev, port,
113 gid, gid_attr);
114 break;
115 case GID_DEL:
116 ib_cache_gid_del(ib_dev, port,
117 gid, gid_attr);
118 break;
119 }
120 }
121 }
122 }
123
124 enum bonding_slave_state {
125 BONDING_SLAVE_STATE_ACTIVE = 1UL << 0,
126 BONDING_SLAVE_STATE_INACTIVE = 1UL << 1,
127 /* No primary slave or the device isn't a slave in bonding */
128 BONDING_SLAVE_STATE_NA = 1UL << 2,
129 };
130
is_eth_active_slave_of_bonding_rcu(struct net_device * dev,struct net_device * upper)131 static enum bonding_slave_state is_eth_active_slave_of_bonding_rcu(struct net_device *dev,
132 struct net_device *upper)
133 {
134 if (upper && netif_is_bond_master(upper)) {
135 struct net_device *pdev =
136 bond_option_active_slave_get_rcu(netdev_priv(upper));
137
138 if (pdev)
139 return dev == pdev ? BONDING_SLAVE_STATE_ACTIVE :
140 BONDING_SLAVE_STATE_INACTIVE;
141 }
142
143 return BONDING_SLAVE_STATE_NA;
144 }
145
146 #define REQUIRED_BOND_STATES (BONDING_SLAVE_STATE_ACTIVE | \
147 BONDING_SLAVE_STATE_NA)
is_eth_port_of_netdev(struct ib_device * ib_dev,u8 port,struct net_device * rdma_ndev,void * cookie)148 static int is_eth_port_of_netdev(struct ib_device *ib_dev, u8 port,
149 struct net_device *rdma_ndev, void *cookie)
150 {
151 struct net_device *real_dev;
152 int res;
153
154 if (!rdma_ndev)
155 return 0;
156
157 rcu_read_lock();
158 real_dev = rdma_vlan_dev_real_dev(cookie);
159 if (!real_dev)
160 real_dev = cookie;
161
162 res = ((rdma_is_upper_dev_rcu(rdma_ndev, cookie) &&
163 (is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev) &
164 REQUIRED_BOND_STATES)) ||
165 real_dev == rdma_ndev);
166
167 rcu_read_unlock();
168 return res;
169 }
170
is_eth_port_inactive_slave(struct ib_device * ib_dev,u8 port,struct net_device * rdma_ndev,void * cookie)171 static int is_eth_port_inactive_slave(struct ib_device *ib_dev, u8 port,
172 struct net_device *rdma_ndev, void *cookie)
173 {
174 struct net_device *master_dev;
175 int res;
176
177 if (!rdma_ndev)
178 return 0;
179
180 rcu_read_lock();
181 master_dev = netdev_master_upper_dev_get_rcu(rdma_ndev);
182 res = is_eth_active_slave_of_bonding_rcu(rdma_ndev, master_dev) ==
183 BONDING_SLAVE_STATE_INACTIVE;
184 rcu_read_unlock();
185
186 return res;
187 }
188
pass_all_filter(struct ib_device * ib_dev,u8 port,struct net_device * rdma_ndev,void * cookie)189 static int pass_all_filter(struct ib_device *ib_dev, u8 port,
190 struct net_device *rdma_ndev, void *cookie)
191 {
192 return 1;
193 }
194
upper_device_filter(struct ib_device * ib_dev,u8 port,struct net_device * rdma_ndev,void * cookie)195 static int upper_device_filter(struct ib_device *ib_dev, u8 port,
196 struct net_device *rdma_ndev, void *cookie)
197 {
198 int res;
199
200 if (!rdma_ndev)
201 return 0;
202
203 if (rdma_ndev == cookie)
204 return 1;
205
206 rcu_read_lock();
207 res = rdma_is_upper_dev_rcu(rdma_ndev, cookie);
208 rcu_read_unlock();
209
210 return res;
211 }
212
update_gid_ip(enum gid_op_type gid_op,struct ib_device * ib_dev,u8 port,struct net_device * ndev,struct sockaddr * addr)213 static void update_gid_ip(enum gid_op_type gid_op,
214 struct ib_device *ib_dev,
215 u8 port, struct net_device *ndev,
216 struct sockaddr *addr)
217 {
218 union ib_gid gid;
219 struct ib_gid_attr gid_attr;
220
221 rdma_ip2gid(addr, &gid);
222 memset(&gid_attr, 0, sizeof(gid_attr));
223 gid_attr.ndev = ndev;
224
225 update_gid(gid_op, ib_dev, port, &gid, &gid_attr);
226 }
227
enum_netdev_default_gids(struct ib_device * ib_dev,u8 port,struct net_device * event_ndev,struct net_device * rdma_ndev)228 static void enum_netdev_default_gids(struct ib_device *ib_dev,
229 u8 port, struct net_device *event_ndev,
230 struct net_device *rdma_ndev)
231 {
232 unsigned long gid_type_mask;
233
234 rcu_read_lock();
235 if (!rdma_ndev ||
236 ((rdma_ndev != event_ndev &&
237 !rdma_is_upper_dev_rcu(rdma_ndev, event_ndev)) ||
238 is_eth_active_slave_of_bonding_rcu(rdma_ndev,
239 netdev_master_upper_dev_get_rcu(rdma_ndev)) ==
240 BONDING_SLAVE_STATE_INACTIVE)) {
241 rcu_read_unlock();
242 return;
243 }
244 rcu_read_unlock();
245
246 gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
247
248 ib_cache_gid_set_default_gid(ib_dev, port, rdma_ndev, gid_type_mask,
249 IB_CACHE_GID_DEFAULT_MODE_SET);
250 }
251
bond_delete_netdev_default_gids(struct ib_device * ib_dev,u8 port,struct net_device * event_ndev,struct net_device * rdma_ndev)252 static void bond_delete_netdev_default_gids(struct ib_device *ib_dev,
253 u8 port,
254 struct net_device *event_ndev,
255 struct net_device *rdma_ndev)
256 {
257 struct net_device *real_dev = rdma_vlan_dev_real_dev(event_ndev);
258
259 if (!rdma_ndev)
260 return;
261
262 if (!real_dev)
263 real_dev = event_ndev;
264
265 rcu_read_lock();
266
267 if (rdma_is_upper_dev_rcu(rdma_ndev, event_ndev) &&
268 is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev) ==
269 BONDING_SLAVE_STATE_INACTIVE) {
270 unsigned long gid_type_mask;
271
272 rcu_read_unlock();
273
274 gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
275
276 ib_cache_gid_set_default_gid(ib_dev, port, rdma_ndev,
277 gid_type_mask,
278 IB_CACHE_GID_DEFAULT_MODE_DELETE);
279 } else {
280 rcu_read_unlock();
281 }
282 }
283
enum_netdev_ipv4_ips(struct ib_device * ib_dev,u8 port,struct net_device * ndev)284 static void enum_netdev_ipv4_ips(struct ib_device *ib_dev,
285 u8 port, struct net_device *ndev)
286 {
287 struct in_device *in_dev;
288 struct sin_list {
289 struct list_head list;
290 struct sockaddr_in ip;
291 };
292 struct sin_list *sin_iter;
293 struct sin_list *sin_temp;
294
295 LIST_HEAD(sin_list);
296 if (ndev->reg_state >= NETREG_UNREGISTERING)
297 return;
298
299 rcu_read_lock();
300 in_dev = __in_dev_get_rcu(ndev);
301 if (!in_dev) {
302 rcu_read_unlock();
303 return;
304 }
305
306 for_ifa(in_dev) {
307 struct sin_list *entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
308
309 if (!entry)
310 continue;
311
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);
315 }
316 endfor_ifa(in_dev);
317 rcu_read_unlock();
318
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);
323 kfree(sin_iter);
324 }
325 }
326
enum_netdev_ipv6_ips(struct ib_device * ib_dev,u8 port,struct net_device * ndev)327 static void enum_netdev_ipv6_ips(struct ib_device *ib_dev,
328 u8 port, struct net_device *ndev)
329 {
330 struct inet6_ifaddr *ifp;
331 struct inet6_dev *in6_dev;
332 struct sin6_list {
333 struct list_head list;
334 struct sockaddr_in6 sin6;
335 };
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);
340
341 if (ndev->reg_state >= NETREG_UNREGISTERING)
342 return;
343
344 in6_dev = in6_dev_get(ndev);
345 if (!in6_dev)
346 return;
347
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);
351
352 if (!entry)
353 continue;
354
355 entry->sin6.sin6_family = AF_INET6;
356 entry->sin6.sin6_addr = ifp->addr;
357 list_add_tail(&entry->list, &sin6_list);
358 }
359 read_unlock_bh(&in6_dev->lock);
360
361 in6_dev_put(in6_dev);
362
363 list_for_each_entry_safe(sin6_iter, sin6_temp, &sin6_list, list) {
364 union ib_gid gid;
365
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);
369 kfree(sin6_iter);
370 }
371 }
372
_add_netdev_ips(struct ib_device * ib_dev,u8 port,struct net_device * ndev)373 static void _add_netdev_ips(struct ib_device *ib_dev, u8 port,
374 struct net_device *ndev)
375 {
376 enum_netdev_ipv4_ips(ib_dev, port, ndev);
377 if (IS_ENABLED(CONFIG_IPV6))
378 enum_netdev_ipv6_ips(ib_dev, port, ndev);
379 }
380
add_netdev_ips(struct ib_device * ib_dev,u8 port,struct net_device * rdma_ndev,void * cookie)381 static void add_netdev_ips(struct ib_device *ib_dev, u8 port,
382 struct net_device *rdma_ndev, void *cookie)
383 {
384 enum_netdev_default_gids(ib_dev, port, cookie, rdma_ndev);
385 _add_netdev_ips(ib_dev, port, cookie);
386 }
387
del_netdev_ips(struct ib_device * ib_dev,u8 port,struct net_device * rdma_ndev,void * cookie)388 static void del_netdev_ips(struct ib_device *ib_dev, u8 port,
389 struct net_device *rdma_ndev, void *cookie)
390 {
391 ib_cache_gid_del_all_netdev_gids(ib_dev, port, cookie);
392 }
393
enum_all_gids_of_dev_cb(struct ib_device * ib_dev,u8 port,struct net_device * rdma_ndev,void * cookie)394 static void enum_all_gids_of_dev_cb(struct ib_device *ib_dev,
395 u8 port,
396 struct net_device *rdma_ndev,
397 void *cookie)
398 {
399 struct net *net;
400 struct net_device *ndev;
401
402 /* Lock the rtnl to make sure the netdevs does not move under
403 * our feet
404 */
405 rtnl_lock();
406 for_each_net(net)
407 for_each_netdev(net, ndev)
408 if (is_eth_port_of_netdev(ib_dev, port, rdma_ndev, ndev))
409 add_netdev_ips(ib_dev, port, rdma_ndev, ndev);
410 rtnl_unlock();
411 }
412
413 /* This function will rescan all of the network devices in the system
414 * and add their gids, as needed, to the relevant RoCE devices. */
roce_rescan_device(struct ib_device * ib_dev)415 int roce_rescan_device(struct ib_device *ib_dev)
416 {
417 ib_enum_roce_netdev(ib_dev, pass_all_filter, NULL,
418 enum_all_gids_of_dev_cb, NULL);
419
420 return 0;
421 }
422
callback_for_addr_gid_device_scan(struct ib_device * device,u8 port,struct net_device * rdma_ndev,void * cookie)423 static void callback_for_addr_gid_device_scan(struct ib_device *device,
424 u8 port,
425 struct net_device *rdma_ndev,
426 void *cookie)
427 {
428 struct update_gid_event_work *parsed = cookie;
429
430 return update_gid(parsed->gid_op, device,
431 port, &parsed->gid,
432 &parsed->gid_attr);
433 }
434
435 struct upper_list {
436 struct list_head list;
437 struct net_device *upper;
438 };
439
netdev_upper_walk(struct net_device * upper,void * data)440 static int netdev_upper_walk(struct net_device *upper, void *data)
441 {
442 struct upper_list *entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
443 struct list_head *upper_list = data;
444
445 if (!entry)
446 return 0;
447
448 list_add_tail(&entry->list, upper_list);
449 dev_hold(upper);
450 entry->upper = upper;
451
452 return 0;
453 }
454
handle_netdev_upper(struct ib_device * ib_dev,u8 port,void * cookie,void (* handle_netdev)(struct ib_device * ib_dev,u8 port,struct net_device * ndev))455 static void handle_netdev_upper(struct ib_device *ib_dev, u8 port,
456 void *cookie,
457 void (*handle_netdev)(struct ib_device *ib_dev,
458 u8 port,
459 struct net_device *ndev))
460 {
461 struct net_device *ndev = cookie;
462 struct upper_list *upper_iter;
463 struct upper_list *upper_temp;
464 LIST_HEAD(upper_list);
465
466 rcu_read_lock();
467 netdev_walk_all_upper_dev_rcu(ndev, netdev_upper_walk, &upper_list);
468 rcu_read_unlock();
469
470 handle_netdev(ib_dev, port, ndev);
471 list_for_each_entry_safe(upper_iter, upper_temp, &upper_list,
472 list) {
473 handle_netdev(ib_dev, port, upper_iter->upper);
474 dev_put(upper_iter->upper);
475 list_del(&upper_iter->list);
476 kfree(upper_iter);
477 }
478 }
479
_roce_del_all_netdev_gids(struct ib_device * ib_dev,u8 port,struct net_device * event_ndev)480 static void _roce_del_all_netdev_gids(struct ib_device *ib_dev, u8 port,
481 struct net_device *event_ndev)
482 {
483 ib_cache_gid_del_all_netdev_gids(ib_dev, port, event_ndev);
484 }
485
del_netdev_upper_ips(struct ib_device * ib_dev,u8 port,struct net_device * rdma_ndev,void * cookie)486 static void del_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
487 struct net_device *rdma_ndev, void *cookie)
488 {
489 handle_netdev_upper(ib_dev, port, cookie, _roce_del_all_netdev_gids);
490 }
491
add_netdev_upper_ips(struct ib_device * ib_dev,u8 port,struct net_device * rdma_ndev,void * cookie)492 static void add_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
493 struct net_device *rdma_ndev, void *cookie)
494 {
495 handle_netdev_upper(ib_dev, port, cookie, _add_netdev_ips);
496 }
497
del_netdev_default_ips_join(struct ib_device * ib_dev,u8 port,struct net_device * rdma_ndev,void * cookie)498 static void del_netdev_default_ips_join(struct ib_device *ib_dev, u8 port,
499 struct net_device *rdma_ndev,
500 void *cookie)
501 {
502 struct net_device *master_ndev;
503
504 rcu_read_lock();
505 master_ndev = netdev_master_upper_dev_get_rcu(rdma_ndev);
506 if (master_ndev)
507 dev_hold(master_ndev);
508 rcu_read_unlock();
509
510 if (master_ndev) {
511 bond_delete_netdev_default_gids(ib_dev, port, master_ndev,
512 rdma_ndev);
513 dev_put(master_ndev);
514 }
515 }
516
del_netdev_default_ips(struct ib_device * ib_dev,u8 port,struct net_device * rdma_ndev,void * cookie)517 static void del_netdev_default_ips(struct ib_device *ib_dev, u8 port,
518 struct net_device *rdma_ndev, void *cookie)
519 {
520 bond_delete_netdev_default_gids(ib_dev, port, cookie, rdma_ndev);
521 }
522
523 /* The following functions operate on all IB devices. netdevice_event and
524 * addr_event execute ib_enum_all_roce_netdevs through a work.
525 * ib_enum_all_roce_netdevs iterates through all IB devices.
526 */
527
netdevice_event_work_handler(struct work_struct * _work)528 static void netdevice_event_work_handler(struct work_struct *_work)
529 {
530 struct netdev_event_work *work =
531 container_of(_work, struct netdev_event_work, work);
532 unsigned int i;
533
534 for (i = 0; i < ARRAY_SIZE(work->cmds) && work->cmds[i].cb; i++) {
535 ib_enum_all_roce_netdevs(work->cmds[i].filter,
536 work->cmds[i].filter_ndev,
537 work->cmds[i].cb,
538 work->cmds[i].ndev);
539 dev_put(work->cmds[i].ndev);
540 dev_put(work->cmds[i].filter_ndev);
541 }
542
543 kfree(work);
544 }
545
netdevice_queue_work(struct netdev_event_work_cmd * cmds,struct net_device * ndev)546 static int netdevice_queue_work(struct netdev_event_work_cmd *cmds,
547 struct net_device *ndev)
548 {
549 unsigned int i;
550 struct netdev_event_work *ndev_work =
551 kmalloc(sizeof(*ndev_work), GFP_KERNEL);
552
553 if (!ndev_work)
554 return NOTIFY_DONE;
555
556 memcpy(ndev_work->cmds, cmds, sizeof(ndev_work->cmds));
557 for (i = 0; i < ARRAY_SIZE(ndev_work->cmds) && ndev_work->cmds[i].cb; i++) {
558 if (!ndev_work->cmds[i].ndev)
559 ndev_work->cmds[i].ndev = ndev;
560 if (!ndev_work->cmds[i].filter_ndev)
561 ndev_work->cmds[i].filter_ndev = ndev;
562 dev_hold(ndev_work->cmds[i].ndev);
563 dev_hold(ndev_work->cmds[i].filter_ndev);
564 }
565 INIT_WORK(&ndev_work->work, netdevice_event_work_handler);
566
567 queue_work(gid_cache_wq, &ndev_work->work);
568
569 return NOTIFY_DONE;
570 }
571
572 static const struct netdev_event_work_cmd add_cmd = {
573 .cb = add_netdev_ips, .filter = is_eth_port_of_netdev};
574 static const struct netdev_event_work_cmd add_cmd_upper_ips = {
575 .cb = add_netdev_upper_ips, .filter = is_eth_port_of_netdev};
576
netdevice_event_changeupper(struct netdev_notifier_changeupper_info * changeupper_info,struct netdev_event_work_cmd * cmds)577 static void netdevice_event_changeupper(struct netdev_notifier_changeupper_info *changeupper_info,
578 struct netdev_event_work_cmd *cmds)
579 {
580 static const struct netdev_event_work_cmd upper_ips_del_cmd = {
581 .cb = del_netdev_upper_ips, .filter = upper_device_filter};
582 static const struct netdev_event_work_cmd bonding_default_del_cmd = {
583 .cb = del_netdev_default_ips, .filter = is_eth_port_inactive_slave};
584
585 if (changeupper_info->linking == false) {
586 cmds[0] = upper_ips_del_cmd;
587 cmds[0].ndev = changeupper_info->upper_dev;
588 cmds[1] = add_cmd;
589 } else {
590 cmds[0] = bonding_default_del_cmd;
591 cmds[0].ndev = changeupper_info->upper_dev;
592 cmds[1] = add_cmd_upper_ips;
593 cmds[1].ndev = changeupper_info->upper_dev;
594 cmds[1].filter_ndev = changeupper_info->upper_dev;
595 }
596 }
597
netdevice_event(struct notifier_block * this,unsigned long event,void * ptr)598 static int netdevice_event(struct notifier_block *this, unsigned long event,
599 void *ptr)
600 {
601 static const struct netdev_event_work_cmd del_cmd = {
602 .cb = del_netdev_ips, .filter = pass_all_filter};
603 static const struct netdev_event_work_cmd bonding_default_del_cmd_join = {
604 .cb = del_netdev_default_ips_join, .filter = is_eth_port_inactive_slave};
605 static const struct netdev_event_work_cmd default_del_cmd = {
606 .cb = del_netdev_default_ips, .filter = pass_all_filter};
607 static const struct netdev_event_work_cmd bonding_event_ips_del_cmd = {
608 .cb = del_netdev_upper_ips, .filter = upper_device_filter};
609 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
610 struct netdev_event_work_cmd cmds[ROCE_NETDEV_CALLBACK_SZ] = { {NULL} };
611
612 if (ndev->type != ARPHRD_ETHER)
613 return NOTIFY_DONE;
614
615 switch (event) {
616 case NETDEV_REGISTER:
617 case NETDEV_UP:
618 cmds[0] = bonding_default_del_cmd_join;
619 cmds[1] = add_cmd;
620 break;
621
622 case NETDEV_UNREGISTER:
623 if (ndev->reg_state < NETREG_UNREGISTERED)
624 cmds[0] = del_cmd;
625 else
626 return NOTIFY_DONE;
627 break;
628
629 case NETDEV_CHANGEADDR:
630 cmds[0] = default_del_cmd;
631 cmds[1] = add_cmd;
632 break;
633
634 case NETDEV_CHANGEUPPER:
635 netdevice_event_changeupper(
636 container_of(ptr, struct netdev_notifier_changeupper_info, info),
637 cmds);
638 break;
639
640 case NETDEV_BONDING_FAILOVER:
641 cmds[0] = bonding_event_ips_del_cmd;
642 cmds[1] = bonding_default_del_cmd_join;
643 cmds[2] = add_cmd_upper_ips;
644 break;
645
646 default:
647 return NOTIFY_DONE;
648 }
649
650 return netdevice_queue_work(cmds, ndev);
651 }
652
update_gid_event_work_handler(struct work_struct * _work)653 static void update_gid_event_work_handler(struct work_struct *_work)
654 {
655 struct update_gid_event_work *work =
656 container_of(_work, struct update_gid_event_work, work);
657
658 ib_enum_all_roce_netdevs(is_eth_port_of_netdev, work->gid_attr.ndev,
659 callback_for_addr_gid_device_scan, work);
660
661 dev_put(work->gid_attr.ndev);
662 kfree(work);
663 }
664
addr_event(struct notifier_block * this,unsigned long event,struct sockaddr * sa,struct net_device * ndev)665 static int addr_event(struct notifier_block *this, unsigned long event,
666 struct sockaddr *sa, struct net_device *ndev)
667 {
668 struct update_gid_event_work *work;
669 enum gid_op_type gid_op;
670
671 if (ndev->type != ARPHRD_ETHER)
672 return NOTIFY_DONE;
673
674 switch (event) {
675 case NETDEV_UP:
676 gid_op = GID_ADD;
677 break;
678
679 case NETDEV_DOWN:
680 gid_op = GID_DEL;
681 break;
682
683 default:
684 return NOTIFY_DONE;
685 }
686
687 work = kmalloc(sizeof(*work), GFP_ATOMIC);
688 if (!work)
689 return NOTIFY_DONE;
690
691 INIT_WORK(&work->work, update_gid_event_work_handler);
692
693 rdma_ip2gid(sa, &work->gid);
694 work->gid_op = gid_op;
695
696 memset(&work->gid_attr, 0, sizeof(work->gid_attr));
697 dev_hold(ndev);
698 work->gid_attr.ndev = ndev;
699
700 queue_work(gid_cache_wq, &work->work);
701
702 return NOTIFY_DONE;
703 }
704
inetaddr_event(struct notifier_block * this,unsigned long event,void * ptr)705 static int inetaddr_event(struct notifier_block *this, unsigned long event,
706 void *ptr)
707 {
708 struct sockaddr_in in;
709 struct net_device *ndev;
710 struct in_ifaddr *ifa = ptr;
711
712 in.sin_family = AF_INET;
713 in.sin_addr.s_addr = ifa->ifa_address;
714 ndev = ifa->ifa_dev->dev;
715
716 return addr_event(this, event, (struct sockaddr *)&in, ndev);
717 }
718
inet6addr_event(struct notifier_block * this,unsigned long event,void * ptr)719 static int inet6addr_event(struct notifier_block *this, unsigned long event,
720 void *ptr)
721 {
722 struct sockaddr_in6 in6;
723 struct net_device *ndev;
724 struct inet6_ifaddr *ifa6 = ptr;
725
726 in6.sin6_family = AF_INET6;
727 in6.sin6_addr = ifa6->addr;
728 ndev = ifa6->idev->dev;
729
730 return addr_event(this, event, (struct sockaddr *)&in6, ndev);
731 }
732
733 static struct notifier_block nb_netdevice = {
734 .notifier_call = netdevice_event
735 };
736
737 static struct notifier_block nb_inetaddr = {
738 .notifier_call = inetaddr_event
739 };
740
741 static struct notifier_block nb_inet6addr = {
742 .notifier_call = inet6addr_event
743 };
744
roce_gid_mgmt_init(void)745 int __init roce_gid_mgmt_init(void)
746 {
747 gid_cache_wq = alloc_ordered_workqueue("gid-cache-wq", 0);
748 if (!gid_cache_wq)
749 return -ENOMEM;
750
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
757 * callbacks.
758 */
759 register_netdevice_notifier(&nb_netdevice);
760
761 return 0;
762 }
763
roce_gid_mgmt_cleanup(void)764 void __exit roce_gid_mgmt_cleanup(void)
765 {
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.
774 */
775 destroy_workqueue(gid_cache_wq);
776 }
777