]> xenbits.xensource.com Git - people/aperard/linux.git/commitdiff
ipv6: annotate data-races around cnf.forwarding
authorEric Dumazet <edumazet@google.com>
Wed, 28 Feb 2024 13:54:30 +0000 (13:54 +0000)
committerDavid S. Miller <davem@davemloft.net>
Fri, 1 Mar 2024 08:42:31 +0000 (08:42 +0000)
idev->cnf.forwarding and net->ipv6.devconf_all->forwarding
might be read locklessly, add appropriate READ_ONCE()
and WRITE_ONCE() annotations.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/usb/cdc_mbim.c
include/net/ipv6.h
net/core/filter.c
net/ipv6/addrconf.c
net/ipv6/ip6_output.c
net/ipv6/ndisc.c
net/ipv6/route.c

index cd4083e0b3b9e6bf8c8fa08ce5b6006d33d9bedd..e13e4920ee9b2e814c4f5ff5df139dc07d739abd 100644 (file)
@@ -339,7 +339,7 @@ static void do_neigh_solicit(struct usbnet *dev, u8 *buf, u16 tci)
        in6_dev = in6_dev_get(netdev);
        if (!in6_dev)
                goto out;
-       is_router = !!in6_dev->cnf.forwarding;
+       is_router = !!READ_ONCE(in6_dev->cnf.forwarding);
        in6_dev_put(in6_dev);
 
        /* ipv6_stub != NULL if in6_dev_get returned an inet6_dev */
index cf25ea21d770d4fe1b235bf2d1ec0088b4b0ff45..88a8e554f7a126a1d817c0cc3bb947c7a43c5cdf 100644 (file)
@@ -534,13 +534,15 @@ static inline int ipv6_hopopt_jumbo_remove(struct sk_buff *skb)
        return 0;
 }
 
-static inline bool ipv6_accept_ra(struct inet6_dev *idev)
+static inline bool ipv6_accept_ra(const struct inet6_dev *idev)
 {
+       s32 accept_ra = READ_ONCE(idev->cnf.accept_ra);
+
        /* If forwarding is enabled, RA are not accepted unless the special
         * hybrid mode (accept_ra=2) is enabled.
         */
-       return idev->cnf.forwarding ? idev->cnf.accept_ra == 2 :
-           idev->cnf.accept_ra;
+       return READ_ONCE(idev->cnf.forwarding) ? accept_ra == 2 :
+               accept_ra;
 }
 
 #define IPV6_FRAG_HIGH_THRESH  (4 * 1024*1024) /* 4194304 */
index 358870408a51e61f3cbc552736806e4dfee1ec39..58e8e1a70aa752a2c045117e00d8797478da4738 100644 (file)
@@ -5988,7 +5988,7 @@ static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
                return -ENODEV;
 
        idev = __in6_dev_get_safely(dev);
-       if (unlikely(!idev || !idev->cnf.forwarding))
+       if (unlikely(!idev || !READ_ONCE(idev->cnf.forwarding)))
                return BPF_FIB_LKUP_RET_FWD_DISABLED;
 
        if (flags & BPF_FIB_LOOKUP_OUTPUT) {
index 1b534edc7ca2c461da19f4dca2040b94eaa4581b..ec99e393f6a91ec202dcc9f2dbb3517c6eb81e2a 100644 (file)
@@ -551,7 +551,8 @@ static int inet6_netconf_fill_devconf(struct sk_buff *skb, int ifindex,
                goto out;
 
        if ((all || type == NETCONFA_FORWARDING) &&
-           nla_put_s32(skb, NETCONFA_FORWARDING, devconf->forwarding) < 0)
+           nla_put_s32(skb, NETCONFA_FORWARDING,
+                       READ_ONCE(devconf->forwarding)) < 0)
                goto nla_put_failure;
 #ifdef CONFIG_IPV6_MROUTE
        if ((all || type == NETCONFA_MC_FORWARDING) &&
@@ -869,7 +870,8 @@ static void addrconf_forward_change(struct net *net, __s32 newf)
                idev = __in6_dev_get(dev);
                if (idev) {
                        int changed = (!idev->cnf.forwarding) ^ (!newf);
-                       idev->cnf.forwarding = newf;
+
+                       WRITE_ONCE(idev->cnf.forwarding, newf);
                        if (changed)
                                dev_forward_change(idev);
                }
@@ -886,7 +888,7 @@ static int addrconf_fixup_forwarding(struct ctl_table *table, int *p, int newf)
 
        net = (struct net *)table->extra2;
        old = *p;
-       *p = newf;
+       WRITE_ONCE(*p, newf);
 
        if (p == &net->ipv6.devconf_dflt->forwarding) {
                if ((!newf) ^ (!old))
@@ -901,7 +903,7 @@ static int addrconf_fixup_forwarding(struct ctl_table *table, int *p, int newf)
        if (p == &net->ipv6.devconf_all->forwarding) {
                int old_dflt = net->ipv6.devconf_dflt->forwarding;
 
-               net->ipv6.devconf_dflt->forwarding = newf;
+               WRITE_ONCE(net->ipv6.devconf_dflt->forwarding, newf);
                if ((!newf) ^ (!old_dflt))
                        inet6_netconf_notify_devconf(net, RTM_NEWNETCONF,
                                                     NETCONFA_FORWARDING,
index 0559bd0005858631f88c706f98c625ad0bfff278..444be8c84cc579bf32b2950e0261ffe7c1d265a8 100644 (file)
@@ -501,7 +501,7 @@ int ip6_forward(struct sk_buff *skb)
        u32 mtu;
 
        idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
-       if (net->ipv6.devconf_all->forwarding == 0)
+       if (READ_ONCE(net->ipv6.devconf_all->forwarding) == 0)
                goto error;
 
        if (skb->pkt_type != PACKET_HOST)
index 9c9c31268432ee58c1a381d0333d85a558a602e1..1fb5e37bc78be54c71b49506e833a53edff3fa0e 100644 (file)
@@ -903,7 +903,7 @@ have_ifp:
                }
 
                if (ipv6_chk_acast_addr(net, dev, &msg->target) ||
-                   (idev->cnf.forwarding &&
+                   (READ_ONCE(idev->cnf.forwarding) &&
                     (net->ipv6.devconf_all->proxy_ndp || idev->cnf.proxy_ndp) &&
                     (is_router = pndisc_is_router(&msg->target, dev)) >= 0)) {
                        if (!(NEIGH_CB(skb)->flags & LOCALLY_ENQUEUED) &&
@@ -929,7 +929,7 @@ have_ifp:
        }
 
        if (is_router < 0)
-               is_router = idev->cnf.forwarding;
+               is_router = READ_ONCE(idev->cnf.forwarding);
 
        if (dad) {
                ndisc_send_na(dev, &in6addr_linklocal_allnodes, &msg->target,
@@ -1080,7 +1080,7 @@ static enum skb_drop_reason ndisc_recv_na(struct sk_buff *skb)
         * Note that we don't do a (daddr == all-routers-mcast) check.
         */
        new_state = msg->icmph.icmp6_solicited ? NUD_REACHABLE : NUD_STALE;
-       if (!neigh && lladdr && idev && idev->cnf.forwarding) {
+       if (!neigh && lladdr && idev && READ_ONCE(idev->cnf.forwarding)) {
                if (accept_untracked_na(dev, saddr)) {
                        neigh = neigh_create(&nd_tbl, &msg->target, dev);
                        new_state = NUD_STALE;
@@ -1100,7 +1100,8 @@ static enum skb_drop_reason ndisc_recv_na(struct sk_buff *skb)
                 * has already sent a NA to us.
                 */
                if (lladdr && !memcmp(lladdr, dev->dev_addr, dev->addr_len) &&
-                   net->ipv6.devconf_all->forwarding && net->ipv6.devconf_all->proxy_ndp &&
+                   READ_ONCE(net->ipv6.devconf_all->forwarding) &&
+                   net->ipv6.devconf_all->proxy_ndp &&
                    pneigh_lookup(&nd_tbl, net, &msg->target, dev, 0)) {
                        /* XXX: idev->cnf.proxy_ndp */
                        goto out;
@@ -1148,7 +1149,7 @@ static enum skb_drop_reason ndisc_recv_rs(struct sk_buff *skb)
        }
 
        /* Don't accept RS if we're not in router mode */
-       if (!idev->cnf.forwarding)
+       if (!READ_ONCE(idev->cnf.forwarding))
                goto out;
 
        /*
index 66c685b0b6199fee3bc39768eab5a6fb831bd2f5..6a2b53de48182525a923e62ba3fbd13cba60a48a 100644 (file)
@@ -2220,7 +2220,7 @@ struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
 
        strict |= flags & RT6_LOOKUP_F_IFACE;
        strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
-       if (net->ipv6.devconf_all->forwarding == 0)
+       if (READ_ONCE(net->ipv6.devconf_all->forwarding) == 0)
                strict |= RT6_LOOKUP_F_REACHABLE;
 
        rcu_read_lock();
@@ -4149,7 +4149,7 @@ static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_bu
        in6_dev = __in6_dev_get(skb->dev);
        if (!in6_dev)
                return;
-       if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
+       if (READ_ONCE(in6_dev->cnf.forwarding) || !in6_dev->cnf.accept_redirects)
                return;
 
        /* RFC2461 8.1: