Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge branch 'ipv6-more-data-race-annotations'
Eric Dumazet says:
====================
ipv6: more data-race annotations
Inspired by one unrelated syzbot report.
This series adds missing (and boring) data-race annotations in IPv6.
Only the first patch adds sysctl_ipv6_flowlabel group
to speedup ip6_make_flowlabel() a bit.
====================
Link: https://patch.msgid.link/20260115094141.3124990-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
+57
-43
+19
-15
include/net/ipv6.h
+19
-15
include/net/ipv6.h
···
931
931
932
932
#if IS_ENABLED(CONFIG_IPV6)
933
933
934
-
static inline bool ipv6_can_nonlocal_bind(struct net *net,
935
-
struct inet_sock *inet)
934
+
static inline bool ipv6_can_nonlocal_bind(const struct net *net,
935
+
const struct inet_sock *inet)
936
936
{
937
-
return net->ipv6.sysctl.ip_nonlocal_bind ||
937
+
return READ_ONCE(net->ipv6.sysctl.ip_nonlocal_bind) ||
938
938
test_bit(INET_FLAGS_FREEBIND, &inet->inet_flags) ||
939
939
test_bit(INET_FLAGS_TRANSPARENT, &inet->inet_flags);
940
940
}
···
949
949
950
950
#define IP6_DEFAULT_AUTO_FLOW_LABELS IP6_AUTO_FLOW_LABEL_OPTOUT
951
951
952
-
static inline __be32 ip6_make_flowlabel(struct net *net, struct sk_buff *skb,
952
+
static inline __be32 ip6_make_flowlabel(const struct net *net,
953
+
struct sk_buff *skb,
953
954
__be32 flowlabel, bool autolabel,
954
955
struct flowi6 *fl6)
955
956
{
957
+
u8 auto_flowlabels;
956
958
u32 hash;
957
959
958
960
/* @flowlabel may include more than a flow label, eg, the traffic class.
···
962
960
*/
963
961
flowlabel &= IPV6_FLOWLABEL_MASK;
964
962
965
-
if (flowlabel ||
966
-
net->ipv6.sysctl.auto_flowlabels == IP6_AUTO_FLOW_LABEL_OFF ||
967
-
(!autolabel &&
968
-
net->ipv6.sysctl.auto_flowlabels != IP6_AUTO_FLOW_LABEL_FORCED))
963
+
if (flowlabel)
964
+
return flowlabel;
965
+
966
+
auto_flowlabels = READ_ONCE(net->ipv6.sysctl.auto_flowlabels);
967
+
if (auto_flowlabels == IP6_AUTO_FLOW_LABEL_OFF ||
968
+
(!autolabel && auto_flowlabels != IP6_AUTO_FLOW_LABEL_FORCED))
969
969
return flowlabel;
970
970
971
971
hash = skb_get_hash_flowi6(skb, fl6);
···
980
976
981
977
flowlabel = (__force __be32)hash & IPV6_FLOWLABEL_MASK;
982
978
983
-
if (net->ipv6.sysctl.flowlabel_state_ranges)
979
+
if (READ_ONCE(net->ipv6.sysctl.flowlabel_state_ranges))
984
980
flowlabel |= IPV6_FLOWLABEL_STATELESS_FLAG;
985
981
986
982
return flowlabel;
987
983
}
988
984
989
-
static inline int ip6_default_np_autolabel(struct net *net)
985
+
static inline int ip6_default_np_autolabel(const struct net *net)
990
986
{
991
-
switch (net->ipv6.sysctl.auto_flowlabels) {
987
+
switch (READ_ONCE(net->ipv6.sysctl.auto_flowlabels)) {
992
988
case IP6_AUTO_FLOW_LABEL_OFF:
993
989
case IP6_AUTO_FLOW_LABEL_OPTIN:
994
990
default:
···
999
995
}
1000
996
}
1001
997
#else
1002
-
static inline __be32 ip6_make_flowlabel(struct net *net, struct sk_buff *skb,
998
+
static inline __be32 ip6_make_flowlabel(const struct net *net, struct sk_buff *skb,
1003
999
__be32 flowlabel, bool autolabel,
1004
1000
struct flowi6 *fl6)
1005
1001
{
1006
1002
return flowlabel;
1007
1003
}
1008
-
static inline int ip6_default_np_autolabel(struct net *net)
1004
+
static inline int ip6_default_np_autolabel(const struct net *net)
1009
1005
{
1010
1006
return 0;
1011
1007
}
···
1014
1010
#if IS_ENABLED(CONFIG_IPV6)
1015
1011
static inline int ip6_multipath_hash_policy(const struct net *net)
1016
1012
{
1017
-
return net->ipv6.sysctl.multipath_hash_policy;
1013
+
return READ_ONCE(net->ipv6.sysctl.multipath_hash_policy);
1018
1014
}
1019
1015
static inline u32 ip6_multipath_hash_fields(const struct net *net)
1020
1016
{
1021
-
return net->ipv6.sysctl.multipath_hash_fields;
1017
+
return READ_ONCE(net->ipv6.sysctl.multipath_hash_fields);
1022
1018
}
1023
1019
#else
1024
1020
static inline int ip6_multipath_hash_policy(const struct net *net)
+7
-3
include/net/netns/ipv6.h
+7
-3
include/net/netns/ipv6.h
···
30
30
int ip6_rt_min_advmss;
31
31
u32 multipath_hash_fields;
32
32
u8 multipath_hash_policy;
33
-
u8 bindv6only;
33
+
34
+
__cacheline_group_begin(sysctl_ipv6_flowlabel);
34
35
u8 flowlabel_consistency;
35
36
u8 auto_flowlabels;
36
-
int icmpv6_time;
37
+
u8 flowlabel_state_ranges;
38
+
__cacheline_group_end(sysctl_ipv6_flowlabel);
39
+
37
40
u8 icmpv6_echo_ignore_all;
38
41
u8 icmpv6_echo_ignore_multicast;
39
42
u8 icmpv6_echo_ignore_anycast;
43
+
int icmpv6_time;
40
44
DECLARE_BITMAP(icmpv6_ratemask, ICMPV6_MSG_MAX + 1);
41
45
unsigned long *icmpv6_ratemask_ptr;
42
46
u8 anycast_src_echo_reply;
47
+
u8 bindv6only;
43
48
u8 ip_nonlocal_bind;
44
49
u8 fwmark_reflect;
45
-
u8 flowlabel_state_ranges;
46
50
int idgen_retries;
47
51
int idgen_delay;
48
52
int flowlabel_reflect;
+2
-2
net/ipv6/af_inet6.c
+2
-2
net/ipv6/af_inet6.c
···
224
224
inet6_set_bit(MC6_LOOP, sk);
225
225
inet6_set_bit(MC6_ALL, sk);
226
226
np->pmtudisc = IPV6_PMTUDISC_WANT;
227
-
inet6_assign_bit(REPFLOW, sk, net->ipv6.sysctl.flowlabel_reflect &
228
-
FLOWLABEL_REFLECT_ESTABLISHED);
227
+
inet6_assign_bit(REPFLOW, sk, READ_ONCE(net->ipv6.sysctl.flowlabel_reflect) &
228
+
FLOWLABEL_REFLECT_ESTABLISHED);
229
229
sk->sk_ipv6only = net->ipv6.sysctl.bindv6only;
230
230
sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
231
231
+6
-4
net/ipv6/exthdrs.c
+6
-4
net/ipv6/exthdrs.c
···
314
314
}
315
315
316
316
extlen = (skb_transport_header(skb)[1] + 1) << 3;
317
-
if (extlen > net->ipv6.sysctl.max_dst_opts_len)
317
+
if (extlen > READ_ONCE(net->ipv6.sysctl.max_dst_opts_len))
318
318
goto fail_and_free;
319
319
320
320
opt->lastopt = opt->dst1 = skb_network_header_len(skb);
···
322
322
dstbuf = opt->dst1;
323
323
#endif
324
324
325
-
if (ip6_parse_tlv(false, skb, net->ipv6.sysctl.max_dst_opts_cnt)) {
325
+
if (ip6_parse_tlv(false, skb,
326
+
READ_ONCE(net->ipv6.sysctl.max_dst_opts_cnt))) {
326
327
skb->transport_header += extlen;
327
328
opt = IP6CB(skb);
328
329
#if IS_ENABLED(CONFIG_IPV6_MIP6)
···
1050
1049
}
1051
1050
1052
1051
extlen = (skb_transport_header(skb)[1] + 1) << 3;
1053
-
if (extlen > net->ipv6.sysctl.max_hbh_opts_len)
1052
+
if (extlen > READ_ONCE(net->ipv6.sysctl.max_hbh_opts_len))
1054
1053
goto fail_and_free;
1055
1054
1056
1055
opt->flags |= IP6SKB_HOPBYHOP;
1057
-
if (ip6_parse_tlv(true, skb, net->ipv6.sysctl.max_hbh_opts_cnt)) {
1056
+
if (ip6_parse_tlv(true, skb,
1057
+
READ_ONCE(net->ipv6.sysctl.max_hbh_opts_cnt))) {
1058
1058
skb->transport_header += extlen;
1059
1059
opt = IP6CB(skb);
1060
1060
opt->nhoff = sizeof(struct ipv6hdr);
+2
-1
net/ipv6/icmp.c
+2
-1
net/ipv6/icmp.c
···
958
958
tmp_hdr.icmp6_type = type;
959
959
960
960
memset(&fl6, 0, sizeof(fl6));
961
-
if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_ICMPV6_ECHO_REPLIES)
961
+
if (READ_ONCE(net->ipv6.sysctl.flowlabel_reflect) &
962
+
FLOWLABEL_REFLECT_ICMPV6_ECHO_REPLIES)
962
963
fl6.flowlabel = ip6_flowlabel(ipv6_hdr(skb));
963
964
964
965
fl6.flowi6_proto = IPPROTO_ICMPV6;
+6
-6
net/ipv6/ip6_fib.c
+6
-6
net/ipv6/ip6_fib.c
···
1374
1374
if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1375
1375
(rt->fib6_flags & RTF_EXPIRES))
1376
1376
mod_timer(&net->ipv6.ip6_fib_timer,
1377
-
jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1377
+
jiffies + READ_ONCE(net->ipv6.sysctl.ip6_rt_gc_interval));
1378
1378
}
1379
1379
1380
1380
void fib6_force_start_gc(struct net *net)
1381
1381
{
1382
1382
if (!timer_pending(&net->ipv6.ip6_fib_timer))
1383
1383
mod_timer(&net->ipv6.ip6_fib_timer,
1384
-
jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1384
+
jiffies + READ_ONCE(net->ipv6.sysctl.ip6_rt_gc_interval));
1385
1385
}
1386
1386
1387
1387
static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
···
2413
2413
void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2414
2414
{
2415
2415
struct fib6_gc_args gc_args;
2416
+
int ip6_rt_gc_interval;
2416
2417
unsigned long now;
2417
2418
2418
2419
if (force) {
···
2422
2421
mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2423
2422
return;
2424
2423
}
2425
-
gc_args.timeout = expires ? (int)expires :
2426
-
net->ipv6.sysctl.ip6_rt_gc_interval;
2424
+
ip6_rt_gc_interval = READ_ONCE(net->ipv6.sysctl.ip6_rt_gc_interval);
2425
+
gc_args.timeout = expires ? (int)expires : ip6_rt_gc_interval;
2427
2426
gc_args.more = 0;
2428
2427
2429
2428
fib6_gc_all(net, &gc_args);
···
2432
2431
2433
2432
if (gc_args.more)
2434
2433
mod_timer(&net->ipv6.ip6_fib_timer,
2435
-
round_jiffies(now
2436
-
+ net->ipv6.sysctl.ip6_rt_gc_interval));
2434
+
round_jiffies(now + ip6_rt_gc_interval));
2437
2435
else
2438
2436
timer_delete(&net->ipv6.ip6_fib_timer);
2439
2437
spin_unlock_bh(&net->ipv6.fib6_gc_lock);
+13
-11
net/ipv6/route.c
+13
-11
net/ipv6/route.c
···
2895
2895
2896
2896
dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2897
2897
rt->rt6i_flags |= RTF_MODIFIED;
2898
-
rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2898
+
rt6_update_expires(rt, READ_ONCE(net->ipv6.sysctl.ip6_rt_mtu_expires));
2899
2899
}
2900
2900
2901
2901
static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
···
3256
3256
rcu_read_lock();
3257
3257
3258
3258
net = dst_dev_net_rcu(dst);
3259
-
if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3260
-
mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3259
+
mtu = max_t(unsigned int, mtu,
3260
+
READ_ONCE(net->ipv6.sysctl.ip6_rt_min_advmss));
3261
3261
3262
3262
rcu_read_unlock();
3263
3263
···
3359
3359
static void ip6_dst_gc(struct dst_ops *ops)
3360
3360
{
3361
3361
struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3362
-
int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3363
-
int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3364
-
int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3365
-
unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3362
+
int rt_min_interval = READ_ONCE(net->ipv6.sysctl.ip6_rt_gc_min_interval);
3363
+
int rt_elasticity = READ_ONCE(net->ipv6.sysctl.ip6_rt_gc_elasticity);
3364
+
int rt_gc_timeout = READ_ONCE(net->ipv6.sysctl.ip6_rt_gc_timeout);
3365
+
unsigned long rt_last_gc = READ_ONCE(net->ipv6.ip6_rt_last_gc);
3366
3366
unsigned int val;
3367
3367
int entries;
3368
3368
···
5005
5005
};
5006
5006
struct net *net = dev_net(dev);
5007
5007
5008
-
if (net->ipv6.sysctl.skip_notify_on_dev_down)
5008
+
if (READ_ONCE(net->ipv6.sysctl.skip_notify_on_dev_down))
5009
5009
fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
5010
5010
else
5011
5011
fib6_clean_all(net, fib6_ifdown, &arg);
···
6405
6405
void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6406
6406
bool offload, bool trap, bool offload_failed)
6407
6407
{
6408
+
u8 fib_notify_on_flag_change;
6408
6409
struct sk_buff *skb;
6409
6410
int err;
6410
6411
···
6417
6416
WRITE_ONCE(f6i->offload, offload);
6418
6417
WRITE_ONCE(f6i->trap, trap);
6419
6418
6419
+
fib_notify_on_flag_change = READ_ONCE(net->ipv6.sysctl.fib_notify_on_flag_change);
6420
6420
/* 2 means send notifications only if offload_failed was changed. */
6421
-
if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6421
+
if (fib_notify_on_flag_change == 2 &&
6422
6422
READ_ONCE(f6i->offload_failed) == offload_failed)
6423
6423
return;
6424
6424
···
6431
6429
*/
6432
6430
return;
6433
6431
6434
-
if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6432
+
if (!fib_notify_on_flag_change)
6435
6433
return;
6436
6434
6437
6435
skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
···
6528
6526
return ret;
6529
6527
6530
6528
net = (struct net *)ctl->extra1;
6531
-
delay = net->ipv6.sysctl.flush_delay;
6529
+
delay = READ_ONCE(net->ipv6.sysctl.flush_delay);
6532
6530
fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6533
6531
return 0;
6534
6532
}
+2
-1
net/ipv6/tcp_ipv6.c
+2
-1
net/ipv6/tcp_ipv6.c
···
1085
1085
txhash = inet_twsk(sk)->tw_txhash;
1086
1086
}
1087
1087
} else {
1088
-
if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1088
+
if (READ_ONCE(net->ipv6.sysctl.flowlabel_reflect) &
1089
+
FLOWLABEL_REFLECT_TCP_RESET)
1089
1090
label = ip6_flowlabel(ipv6h);
1090
1091
}
1091
1092