blob: 178f98482e13e217c16d447978a855a900808bc0 [file] [log] [blame]
Matan Barak03db3a22015-07-30 18:33:26 +03001/*
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>
Matan Barak238fdf42015-07-30 18:33:27 +030040#include <net/bonding.h>
Matan Barak03db3a22015-07-30 18:33:26 +030041
42#include <rdma/ib_cache.h>
43#include <rdma/ib_addr.h>
44
45enum gid_op_type {
46 GID_DEL = 0,
47 GID_ADD
48};
49
50struct update_gid_event_work {
51 struct work_struct work;
52 union ib_gid gid;
53 struct ib_gid_attr gid_attr;
54 enum gid_op_type gid_op;
55};
56
Matan Barak238fdf42015-07-30 18:33:27 +030057#define ROCE_NETDEV_CALLBACK_SZ 3
Matan Barak03db3a22015-07-30 18:33:26 +030058struct netdev_event_work_cmd {
59 roce_netdev_callback cb;
60 roce_netdev_filter filter;
Matan Barak238fdf42015-07-30 18:33:27 +030061 struct net_device *ndev;
62 struct net_device *filter_ndev;
Matan Barak03db3a22015-07-30 18:33:26 +030063};
64
65struct netdev_event_work {
66 struct work_struct work;
67 struct netdev_event_work_cmd cmds[ROCE_NETDEV_CALLBACK_SZ];
Matan Barak03db3a22015-07-30 18:33:26 +030068};
69
70static void update_gid(enum gid_op_type gid_op, struct ib_device *ib_dev,
71 u8 port, union ib_gid *gid,
72 struct ib_gid_attr *gid_attr)
73{
74 switch (gid_op) {
75 case GID_ADD:
76 ib_cache_gid_add(ib_dev, port, gid, gid_attr);
77 break;
78 case GID_DEL:
79 ib_cache_gid_del(ib_dev, port, gid, gid_attr);
80 break;
81 }
82}
83
Matan Barak238fdf42015-07-30 18:33:27 +030084enum bonding_slave_state {
85 BONDING_SLAVE_STATE_ACTIVE = 1UL << 0,
86 BONDING_SLAVE_STATE_INACTIVE = 1UL << 1,
87 /* No primary slave or the device isn't a slave in bonding */
88 BONDING_SLAVE_STATE_NA = 1UL << 2,
89};
90
91static enum bonding_slave_state is_eth_active_slave_of_bonding_rcu(struct net_device *dev,
92 struct net_device *upper)
93{
94 if (upper && netif_is_bond_master(upper)) {
95 struct net_device *pdev =
96 bond_option_active_slave_get_rcu(netdev_priv(upper));
97
98 if (pdev)
99 return dev == pdev ? BONDING_SLAVE_STATE_ACTIVE :
100 BONDING_SLAVE_STATE_INACTIVE;
101 }
102
103 return BONDING_SLAVE_STATE_NA;
104}
105
106static bool is_upper_dev_rcu(struct net_device *dev, struct net_device *upper)
107{
108 struct net_device *_upper = NULL;
109 struct list_head *iter;
110
111 netdev_for_each_all_upper_dev_rcu(dev, _upper, iter)
112 if (_upper == upper)
113 break;
114
115 return _upper == upper;
116}
117
118#define REQUIRED_BOND_STATES (BONDING_SLAVE_STATE_ACTIVE | \
119 BONDING_SLAVE_STATE_NA)
Matan Barak03db3a22015-07-30 18:33:26 +0300120static int is_eth_port_of_netdev(struct ib_device *ib_dev, u8 port,
121 struct net_device *rdma_ndev, void *cookie)
122{
Matan Barak03db3a22015-07-30 18:33:26 +0300123 struct net_device *event_ndev = (struct net_device *)cookie;
Matan Barak238fdf42015-07-30 18:33:27 +0300124 struct net_device *real_dev;
125 int res;
126
127 if (!rdma_ndev)
128 return 0;
129
130 rcu_read_lock();
131 real_dev = rdma_vlan_dev_real_dev(event_ndev);
132 if (!real_dev)
133 real_dev = event_ndev;
134
135 res = ((is_upper_dev_rcu(rdma_ndev, event_ndev) &&
136 (is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev) &
137 REQUIRED_BOND_STATES)) ||
138 real_dev == rdma_ndev);
139
140 rcu_read_unlock();
141 return res;
142}
143
144static int is_eth_port_inactive_slave(struct ib_device *ib_dev, u8 port,
145 struct net_device *rdma_ndev, void *cookie)
146{
147 struct net_device *master_dev;
Matan Barak03db3a22015-07-30 18:33:26 +0300148 int res;
149
150 if (!rdma_ndev)
151 return 0;
152
153 rcu_read_lock();
154 master_dev = netdev_master_upper_dev_get_rcu(rdma_ndev);
Matan Barak238fdf42015-07-30 18:33:27 +0300155 res = is_eth_active_slave_of_bonding_rcu(rdma_ndev, master_dev) ==
156 BONDING_SLAVE_STATE_INACTIVE;
Matan Barak03db3a22015-07-30 18:33:26 +0300157 rcu_read_unlock();
158
159 return res;
160}
161
162static int pass_all_filter(struct ib_device *ib_dev, u8 port,
163 struct net_device *rdma_ndev, void *cookie)
164{
165 return 1;
166}
167
Matan Barak238fdf42015-07-30 18:33:27 +0300168static int upper_device_filter(struct ib_device *ib_dev, u8 port,
169 struct net_device *rdma_ndev, void *cookie)
170{
171 struct net_device *event_ndev = (struct net_device *)cookie;
172 int res;
173
174 if (!rdma_ndev)
175 return 0;
176
177 if (rdma_ndev == event_ndev)
178 return 1;
179
180 rcu_read_lock();
181 res = is_upper_dev_rcu(rdma_ndev, event_ndev);
182 rcu_read_unlock();
183
184 return res;
185}
186
Matan Barak03db3a22015-07-30 18:33:26 +0300187static void update_gid_ip(enum gid_op_type gid_op,
188 struct ib_device *ib_dev,
189 u8 port, struct net_device *ndev,
190 struct sockaddr *addr)
191{
192 union ib_gid gid;
193 struct ib_gid_attr gid_attr;
194
195 rdma_ip2gid(addr, &gid);
196 memset(&gid_attr, 0, sizeof(gid_attr));
197 gid_attr.ndev = ndev;
198
199 update_gid(gid_op, ib_dev, port, &gid, &gid_attr);
200}
201
202static void enum_netdev_default_gids(struct ib_device *ib_dev,
203 u8 port, struct net_device *event_ndev,
204 struct net_device *rdma_ndev)
205{
Matan Barak238fdf42015-07-30 18:33:27 +0300206 rcu_read_lock();
207 if (!rdma_ndev ||
208 ((rdma_ndev != event_ndev &&
209 !is_upper_dev_rcu(rdma_ndev, event_ndev)) ||
210 is_eth_active_slave_of_bonding_rcu(rdma_ndev,
211 netdev_master_upper_dev_get_rcu(rdma_ndev)) ==
212 BONDING_SLAVE_STATE_INACTIVE)) {
213 rcu_read_unlock();
Matan Barak03db3a22015-07-30 18:33:26 +0300214 return;
Matan Barak238fdf42015-07-30 18:33:27 +0300215 }
216 rcu_read_unlock();
Matan Barak03db3a22015-07-30 18:33:26 +0300217
218 ib_cache_gid_set_default_gid(ib_dev, port, rdma_ndev,
219 IB_CACHE_GID_DEFAULT_MODE_SET);
220}
221
Matan Barak238fdf42015-07-30 18:33:27 +0300222static void bond_delete_netdev_default_gids(struct ib_device *ib_dev,
223 u8 port,
224 struct net_device *event_ndev,
225 struct net_device *rdma_ndev)
226{
227 struct net_device *real_dev = rdma_vlan_dev_real_dev(event_ndev);
228
229 if (!rdma_ndev)
230 return;
231
232 if (!real_dev)
233 real_dev = event_ndev;
234
235 rcu_read_lock();
236
237 if (is_upper_dev_rcu(rdma_ndev, event_ndev) &&
238 is_eth_active_slave_of_bonding_rcu(rdma_ndev, real_dev) ==
239 BONDING_SLAVE_STATE_INACTIVE) {
240 rcu_read_unlock();
241
242 ib_cache_gid_set_default_gid(ib_dev, port, rdma_ndev,
243 IB_CACHE_GID_DEFAULT_MODE_DELETE);
244 } else {
245 rcu_read_unlock();
246 }
247}
248
Matan Barak03db3a22015-07-30 18:33:26 +0300249static void enum_netdev_ipv4_ips(struct ib_device *ib_dev,
250 u8 port, struct net_device *ndev)
251{
252 struct in_device *in_dev;
Matan Barak39096422015-10-15 15:01:03 +0300253 struct sin_list {
254 struct list_head list;
255 struct sockaddr_in ip;
256 };
257 struct sin_list *sin_iter;
258 struct sin_list *sin_temp;
Matan Barak03db3a22015-07-30 18:33:26 +0300259
Matan Barak39096422015-10-15 15:01:03 +0300260 LIST_HEAD(sin_list);
Matan Barak03db3a22015-07-30 18:33:26 +0300261 if (ndev->reg_state >= NETREG_UNREGISTERING)
262 return;
263
Matan Barak39096422015-10-15 15:01:03 +0300264 rcu_read_lock();
265 in_dev = __in_dev_get_rcu(ndev);
266 if (!in_dev) {
267 rcu_read_unlock();
Matan Barak03db3a22015-07-30 18:33:26 +0300268 return;
Matan Barak39096422015-10-15 15:01:03 +0300269 }
Matan Barak03db3a22015-07-30 18:33:26 +0300270
271 for_ifa(in_dev) {
Matan Barak39096422015-10-15 15:01:03 +0300272 struct sin_list *entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
Matan Barak03db3a22015-07-30 18:33:26 +0300273
Matan Barak39096422015-10-15 15:01:03 +0300274 if (!entry) {
275 pr_warn("roce_gid_mgmt: couldn't allocate entry for IPv4 update\n");
276 continue;
277 }
278 entry->ip.sin_family = AF_INET;
279 entry->ip.sin_addr.s_addr = ifa->ifa_address;
280 list_add_tail(&entry->list, &sin_list);
Matan Barak03db3a22015-07-30 18:33:26 +0300281 }
282 endfor_ifa(in_dev);
Matan Barak39096422015-10-15 15:01:03 +0300283 rcu_read_unlock();
Matan Barak03db3a22015-07-30 18:33:26 +0300284
Matan Barak39096422015-10-15 15:01:03 +0300285 list_for_each_entry_safe(sin_iter, sin_temp, &sin_list, list) {
286 update_gid_ip(GID_ADD, ib_dev, port, ndev,
287 (struct sockaddr *)&sin_iter->ip);
288 list_del(&sin_iter->list);
289 kfree(sin_iter);
290 }
Matan Barak03db3a22015-07-30 18:33:26 +0300291}
292
293static void enum_netdev_ipv6_ips(struct ib_device *ib_dev,
294 u8 port, struct net_device *ndev)
295{
296 struct inet6_ifaddr *ifp;
297 struct inet6_dev *in6_dev;
298 struct sin6_list {
299 struct list_head list;
300 struct sockaddr_in6 sin6;
301 };
302 struct sin6_list *sin6_iter;
303 struct sin6_list *sin6_temp;
304 struct ib_gid_attr gid_attr = {.ndev = ndev};
305 LIST_HEAD(sin6_list);
306
307 if (ndev->reg_state >= NETREG_UNREGISTERING)
308 return;
309
310 in6_dev = in6_dev_get(ndev);
311 if (!in6_dev)
312 return;
313
314 read_lock_bh(&in6_dev->lock);
315 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
316 struct sin6_list *entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
317
318 if (!entry) {
319 pr_warn("roce_gid_mgmt: couldn't allocate entry for IPv6 update\n");
320 continue;
321 }
322
323 entry->sin6.sin6_family = AF_INET6;
324 entry->sin6.sin6_addr = ifp->addr;
325 list_add_tail(&entry->list, &sin6_list);
326 }
327 read_unlock_bh(&in6_dev->lock);
328
329 in6_dev_put(in6_dev);
330
331 list_for_each_entry_safe(sin6_iter, sin6_temp, &sin6_list, list) {
332 union ib_gid gid;
333
334 rdma_ip2gid((struct sockaddr *)&sin6_iter->sin6, &gid);
335 update_gid(GID_ADD, ib_dev, port, &gid, &gid_attr);
336 list_del(&sin6_iter->list);
337 kfree(sin6_iter);
338 }
339}
340
Matan Barak238fdf42015-07-30 18:33:27 +0300341static void _add_netdev_ips(struct ib_device *ib_dev, u8 port,
342 struct net_device *ndev)
343{
344 enum_netdev_ipv4_ips(ib_dev, port, ndev);
345 if (IS_ENABLED(CONFIG_IPV6))
346 enum_netdev_ipv6_ips(ib_dev, port, ndev);
347}
348
Matan Barak03db3a22015-07-30 18:33:26 +0300349static void add_netdev_ips(struct ib_device *ib_dev, u8 port,
350 struct net_device *rdma_ndev, void *cookie)
351{
352 struct net_device *event_ndev = (struct net_device *)cookie;
353
354 enum_netdev_default_gids(ib_dev, port, event_ndev, rdma_ndev);
Matan Barak238fdf42015-07-30 18:33:27 +0300355 _add_netdev_ips(ib_dev, port, event_ndev);
Matan Barak03db3a22015-07-30 18:33:26 +0300356}
357
358static void del_netdev_ips(struct ib_device *ib_dev, u8 port,
359 struct net_device *rdma_ndev, void *cookie)
360{
361 struct net_device *event_ndev = (struct net_device *)cookie;
362
363 ib_cache_gid_del_all_netdev_gids(ib_dev, port, event_ndev);
364}
365
366static void enum_all_gids_of_dev_cb(struct ib_device *ib_dev,
367 u8 port,
368 struct net_device *rdma_ndev,
369 void *cookie)
370{
371 struct net *net;
372 struct net_device *ndev;
373
374 /* Lock the rtnl to make sure the netdevs does not move under
375 * our feet
376 */
377 rtnl_lock();
378 for_each_net(net)
379 for_each_netdev(net, ndev)
380 if (is_eth_port_of_netdev(ib_dev, port, rdma_ndev, ndev))
381 add_netdev_ips(ib_dev, port, rdma_ndev, ndev);
382 rtnl_unlock();
383}
384
385/* This function will rescan all of the network devices in the system
386 * and add their gids, as needed, to the relevant RoCE devices. */
387int roce_rescan_device(struct ib_device *ib_dev)
388{
389 ib_enum_roce_netdev(ib_dev, pass_all_filter, NULL,
390 enum_all_gids_of_dev_cb, NULL);
391
392 return 0;
393}
394
395static void callback_for_addr_gid_device_scan(struct ib_device *device,
396 u8 port,
397 struct net_device *rdma_ndev,
398 void *cookie)
399{
400 struct update_gid_event_work *parsed = cookie;
401
402 return update_gid(parsed->gid_op, device,
403 port, &parsed->gid,
404 &parsed->gid_attr);
405}
406
Matan Barak238fdf42015-07-30 18:33:27 +0300407static void handle_netdev_upper(struct ib_device *ib_dev, u8 port,
408 void *cookie,
409 void (*handle_netdev)(struct ib_device *ib_dev,
410 u8 port,
411 struct net_device *ndev))
412{
413 struct net_device *ndev = (struct net_device *)cookie;
414 struct upper_list {
415 struct list_head list;
416 struct net_device *upper;
417 };
418 struct net_device *upper;
419 struct list_head *iter;
420 struct upper_list *upper_iter;
421 struct upper_list *upper_temp;
422 LIST_HEAD(upper_list);
423
424 rcu_read_lock();
425 netdev_for_each_all_upper_dev_rcu(ndev, upper, iter) {
426 struct upper_list *entry = kmalloc(sizeof(*entry),
427 GFP_ATOMIC);
428
429 if (!entry) {
430 pr_info("roce_gid_mgmt: couldn't allocate entry to delete ndev\n");
431 continue;
432 }
433
434 list_add_tail(&entry->list, &upper_list);
435 dev_hold(upper);
436 entry->upper = upper;
437 }
438 rcu_read_unlock();
439
440 handle_netdev(ib_dev, port, ndev);
441 list_for_each_entry_safe(upper_iter, upper_temp, &upper_list,
442 list) {
443 handle_netdev(ib_dev, port, upper_iter->upper);
444 dev_put(upper_iter->upper);
445 list_del(&upper_iter->list);
446 kfree(upper_iter);
447 }
448}
449
450static void _roce_del_all_netdev_gids(struct ib_device *ib_dev, u8 port,
451 struct net_device *event_ndev)
452{
453 ib_cache_gid_del_all_netdev_gids(ib_dev, port, event_ndev);
454}
455
456static void del_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
457 struct net_device *rdma_ndev, void *cookie)
458{
459 handle_netdev_upper(ib_dev, port, cookie, _roce_del_all_netdev_gids);
460}
461
462static void add_netdev_upper_ips(struct ib_device *ib_dev, u8 port,
463 struct net_device *rdma_ndev, void *cookie)
464{
465 handle_netdev_upper(ib_dev, port, cookie, _add_netdev_ips);
466}
467
468static void del_netdev_default_ips_join(struct ib_device *ib_dev, u8 port,
469 struct net_device *rdma_ndev,
470 void *cookie)
471{
472 struct net_device *master_ndev;
473
474 rcu_read_lock();
475 master_ndev = netdev_master_upper_dev_get_rcu(rdma_ndev);
476 if (master_ndev)
477 dev_hold(master_ndev);
478 rcu_read_unlock();
479
480 if (master_ndev) {
481 bond_delete_netdev_default_gids(ib_dev, port, master_ndev,
482 rdma_ndev);
483 dev_put(master_ndev);
484 }
485}
486
487static void del_netdev_default_ips(struct ib_device *ib_dev, u8 port,
488 struct net_device *rdma_ndev, void *cookie)
489{
490 struct net_device *event_ndev = (struct net_device *)cookie;
491
492 bond_delete_netdev_default_gids(ib_dev, port, event_ndev, rdma_ndev);
493}
494
Matan Barak03db3a22015-07-30 18:33:26 +0300495/* The following functions operate on all IB devices. netdevice_event and
496 * addr_event execute ib_enum_all_roce_netdevs through a work.
497 * ib_enum_all_roce_netdevs iterates through all IB devices.
498 */
499
500static void netdevice_event_work_handler(struct work_struct *_work)
501{
502 struct netdev_event_work *work =
503 container_of(_work, struct netdev_event_work, work);
504 unsigned int i;
505
Matan Barak238fdf42015-07-30 18:33:27 +0300506 for (i = 0; i < ARRAY_SIZE(work->cmds) && work->cmds[i].cb; i++) {
507 ib_enum_all_roce_netdevs(work->cmds[i].filter,
508 work->cmds[i].filter_ndev,
509 work->cmds[i].cb,
510 work->cmds[i].ndev);
511 dev_put(work->cmds[i].ndev);
512 dev_put(work->cmds[i].filter_ndev);
513 }
Matan Barak03db3a22015-07-30 18:33:26 +0300514
Matan Barak03db3a22015-07-30 18:33:26 +0300515 kfree(work);
516}
517
518static int netdevice_queue_work(struct netdev_event_work_cmd *cmds,
519 struct net_device *ndev)
520{
Matan Barak238fdf42015-07-30 18:33:27 +0300521 unsigned int i;
Matan Barak03db3a22015-07-30 18:33:26 +0300522 struct netdev_event_work *ndev_work =
523 kmalloc(sizeof(*ndev_work), GFP_KERNEL);
524
525 if (!ndev_work) {
526 pr_warn("roce_gid_mgmt: can't allocate work for netdevice_event\n");
527 return NOTIFY_DONE;
528 }
529
530 memcpy(ndev_work->cmds, cmds, sizeof(ndev_work->cmds));
Matan Barak238fdf42015-07-30 18:33:27 +0300531 for (i = 0; i < ARRAY_SIZE(ndev_work->cmds) && ndev_work->cmds[i].cb; i++) {
532 if (!ndev_work->cmds[i].ndev)
533 ndev_work->cmds[i].ndev = ndev;
534 if (!ndev_work->cmds[i].filter_ndev)
535 ndev_work->cmds[i].filter_ndev = ndev;
536 dev_hold(ndev_work->cmds[i].ndev);
537 dev_hold(ndev_work->cmds[i].filter_ndev);
538 }
Matan Barak03db3a22015-07-30 18:33:26 +0300539 INIT_WORK(&ndev_work->work, netdevice_event_work_handler);
540
541 queue_work(ib_wq, &ndev_work->work);
542
543 return NOTIFY_DONE;
544}
545
Matan Barak238fdf42015-07-30 18:33:27 +0300546static const struct netdev_event_work_cmd add_cmd = {
547 .cb = add_netdev_ips, .filter = is_eth_port_of_netdev};
548static const struct netdev_event_work_cmd add_cmd_upper_ips = {
549 .cb = add_netdev_upper_ips, .filter = is_eth_port_of_netdev};
550
Linus Torvalds26d21772015-09-09 08:33:31 -0700551static void netdevice_event_changeupper(struct netdev_notifier_changeupper_info *changeupper_info,
Matan Barak238fdf42015-07-30 18:33:27 +0300552 struct netdev_event_work_cmd *cmds)
553{
554 static const struct netdev_event_work_cmd upper_ips_del_cmd = {
555 .cb = del_netdev_upper_ips, .filter = upper_device_filter};
556 static const struct netdev_event_work_cmd bonding_default_del_cmd = {
557 .cb = del_netdev_default_ips, .filter = is_eth_port_inactive_slave};
558
Linus Torvalds26d21772015-09-09 08:33:31 -0700559 if (changeupper_info->linking == false) {
Matan Barak238fdf42015-07-30 18:33:27 +0300560 cmds[0] = upper_ips_del_cmd;
Linus Torvalds26d21772015-09-09 08:33:31 -0700561 cmds[0].ndev = changeupper_info->upper_dev;
Matan Barak238fdf42015-07-30 18:33:27 +0300562 cmds[1] = add_cmd;
Linus Torvalds26d21772015-09-09 08:33:31 -0700563 } else {
Matan Barak238fdf42015-07-30 18:33:27 +0300564 cmds[0] = bonding_default_del_cmd;
Linus Torvalds26d21772015-09-09 08:33:31 -0700565 cmds[0].ndev = changeupper_info->upper_dev;
Matan Barak238fdf42015-07-30 18:33:27 +0300566 cmds[1] = add_cmd_upper_ips;
Linus Torvalds26d21772015-09-09 08:33:31 -0700567 cmds[1].ndev = changeupper_info->upper_dev;
568 cmds[1].filter_ndev = changeupper_info->upper_dev;
Matan Barak238fdf42015-07-30 18:33:27 +0300569 }
570}
571
Matan Barak03db3a22015-07-30 18:33:26 +0300572static int netdevice_event(struct notifier_block *this, unsigned long event,
573 void *ptr)
574{
Matan Barak03db3a22015-07-30 18:33:26 +0300575 static const struct netdev_event_work_cmd del_cmd = {
576 .cb = del_netdev_ips, .filter = pass_all_filter};
Matan Barak238fdf42015-07-30 18:33:27 +0300577 static const struct netdev_event_work_cmd bonding_default_del_cmd_join = {
578 .cb = del_netdev_default_ips_join, .filter = is_eth_port_inactive_slave};
579 static const struct netdev_event_work_cmd default_del_cmd = {
580 .cb = del_netdev_default_ips, .filter = pass_all_filter};
581 static const struct netdev_event_work_cmd bonding_event_ips_del_cmd = {
582 .cb = del_netdev_upper_ips, .filter = upper_device_filter};
Matan Barak03db3a22015-07-30 18:33:26 +0300583 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
584 struct netdev_event_work_cmd cmds[ROCE_NETDEV_CALLBACK_SZ] = { {NULL} };
585
586 if (ndev->type != ARPHRD_ETHER)
587 return NOTIFY_DONE;
588
589 switch (event) {
590 case NETDEV_REGISTER:
591 case NETDEV_UP:
Matan Barak238fdf42015-07-30 18:33:27 +0300592 cmds[0] = bonding_default_del_cmd_join;
593 cmds[1] = add_cmd;
Matan Barak03db3a22015-07-30 18:33:26 +0300594 break;
595
596 case NETDEV_UNREGISTER:
597 if (ndev->reg_state < NETREG_UNREGISTERED)
598 cmds[0] = del_cmd;
599 else
600 return NOTIFY_DONE;
601 break;
602
603 case NETDEV_CHANGEADDR:
Matan Barak238fdf42015-07-30 18:33:27 +0300604 cmds[0] = default_del_cmd;
Matan Barak03db3a22015-07-30 18:33:26 +0300605 cmds[1] = add_cmd;
606 break;
Matan Barak238fdf42015-07-30 18:33:27 +0300607
608 case NETDEV_CHANGEUPPER:
609 netdevice_event_changeupper(
Linus Torvalds26d21772015-09-09 08:33:31 -0700610 container_of(ptr, struct netdev_notifier_changeupper_info, info),
Matan Barak238fdf42015-07-30 18:33:27 +0300611 cmds);
612 break;
613
614 case NETDEV_BONDING_FAILOVER:
615 cmds[0] = bonding_event_ips_del_cmd;
616 cmds[1] = bonding_default_del_cmd_join;
617 cmds[2] = add_cmd_upper_ips;
618 break;
619
Matan Barak03db3a22015-07-30 18:33:26 +0300620 default:
621 return NOTIFY_DONE;
622 }
623
624 return netdevice_queue_work(cmds, ndev);
625}
626
627static void update_gid_event_work_handler(struct work_struct *_work)
628{
629 struct update_gid_event_work *work =
630 container_of(_work, struct update_gid_event_work, work);
631
632 ib_enum_all_roce_netdevs(is_eth_port_of_netdev, work->gid_attr.ndev,
633 callback_for_addr_gid_device_scan, work);
634
635 dev_put(work->gid_attr.ndev);
636 kfree(work);
637}
638
639static int addr_event(struct notifier_block *this, unsigned long event,
640 struct sockaddr *sa, struct net_device *ndev)
641{
642 struct update_gid_event_work *work;
643 enum gid_op_type gid_op;
644
645 if (ndev->type != ARPHRD_ETHER)
646 return NOTIFY_DONE;
647
648 switch (event) {
649 case NETDEV_UP:
650 gid_op = GID_ADD;
651 break;
652
653 case NETDEV_DOWN:
654 gid_op = GID_DEL;
655 break;
656
657 default:
658 return NOTIFY_DONE;
659 }
660
661 work = kmalloc(sizeof(*work), GFP_ATOMIC);
662 if (!work) {
663 pr_warn("roce_gid_mgmt: Couldn't allocate work for addr_event\n");
664 return NOTIFY_DONE;
665 }
666
667 INIT_WORK(&work->work, update_gid_event_work_handler);
668
669 rdma_ip2gid(sa, &work->gid);
670 work->gid_op = gid_op;
671
672 memset(&work->gid_attr, 0, sizeof(work->gid_attr));
673 dev_hold(ndev);
674 work->gid_attr.ndev = ndev;
675
676 queue_work(ib_wq, &work->work);
677
678 return NOTIFY_DONE;
679}
680
681static int inetaddr_event(struct notifier_block *this, unsigned long event,
682 void *ptr)
683{
684 struct sockaddr_in in;
685 struct net_device *ndev;
686 struct in_ifaddr *ifa = ptr;
687
688 in.sin_family = AF_INET;
689 in.sin_addr.s_addr = ifa->ifa_address;
690 ndev = ifa->ifa_dev->dev;
691
692 return addr_event(this, event, (struct sockaddr *)&in, ndev);
693}
694
695static int inet6addr_event(struct notifier_block *this, unsigned long event,
696 void *ptr)
697{
698 struct sockaddr_in6 in6;
699 struct net_device *ndev;
700 struct inet6_ifaddr *ifa6 = ptr;
701
702 in6.sin6_family = AF_INET6;
703 in6.sin6_addr = ifa6->addr;
704 ndev = ifa6->idev->dev;
705
706 return addr_event(this, event, (struct sockaddr *)&in6, ndev);
707}
708
709static struct notifier_block nb_netdevice = {
710 .notifier_call = netdevice_event
711};
712
713static struct notifier_block nb_inetaddr = {
714 .notifier_call = inetaddr_event
715};
716
717static struct notifier_block nb_inet6addr = {
718 .notifier_call = inet6addr_event
719};
720
721int __init roce_gid_mgmt_init(void)
722{
723 register_inetaddr_notifier(&nb_inetaddr);
724 if (IS_ENABLED(CONFIG_IPV6))
725 register_inet6addr_notifier(&nb_inet6addr);
726 /* We relay on the netdevice notifier to enumerate all
727 * existing devices in the system. Register to this notifier
728 * last to make sure we will not miss any IP add/del
729 * callbacks.
730 */
731 register_netdevice_notifier(&nb_netdevice);
732
733 return 0;
734}
735
736void __exit roce_gid_mgmt_cleanup(void)
737{
738 if (IS_ENABLED(CONFIG_IPV6))
739 unregister_inet6addr_notifier(&nb_inet6addr);
740 unregister_inetaddr_notifier(&nb_inetaddr);
741 unregister_netdevice_notifier(&nb_netdevice);
742 /* Ensure all gid deletion tasks complete before we go down,
743 * to avoid any reference to free'd memory. By the time
744 * ib-core is removed, all physical devices have been removed,
745 * so no issue with remaining hardware contexts.
746 */
747}