ipv6: use fib6_info_hold_safe() when necessary
authorWei Wang <weiwan@google.com>
Sun, 22 Jul 2018 03:56:32 +0000 (20:56 -0700)
committerDavid S. Miller <davem@davemloft.net>
Mon, 23 Jul 2018 18:19:02 +0000 (11:19 -0700)
In the code path where only rcu read lock is held, e.g. in the route
lookup code path, it is not safe to directly call fib6_info_hold()
because the fib6_info may already have been deleted but still exists
in the rcu grace period. Holding reference to it could cause double
free and crash the kernel.

This patch adds a new function fib6_info_hold_safe() and replace
fib6_info_hold() in all necessary places.

Syzbot reported 3 crash traces because of this. One of them is:
8021q: adding VLAN 0 to HW filter on device team0
IPv6: ADDRCONF(NETDEV_CHANGE): team0: link becomes ready
dst_release: dst:(____ptrval____) refcnt:-1
dst_release: dst:(____ptrval____) refcnt:-2
WARNING: CPU: 1 PID: 4845 at include/net/dst.h:239 dst_hold include/net/dst.h:239 [inline]
WARNING: CPU: 1 PID: 4845 at include/net/dst.h:239 ip6_setup_cork+0xd66/0x1830 net/ipv6/ip6_output.c:1204
dst_release: dst:(____ptrval____) refcnt:-1
Kernel panic - not syncing: panic_on_warn set ...

CPU: 1 PID: 4845 Comm: syz-executor493 Not tainted 4.18.0-rc3+ #10
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113
 panic+0x238/0x4e7 kernel/panic.c:184
dst_release: dst:(____ptrval____) refcnt:-2
dst_release: dst:(____ptrval____) refcnt:-3
 __warn.cold.8+0x163/0x1ba kernel/panic.c:536
dst_release: dst:(____ptrval____) refcnt:-4
 report_bug+0x252/0x2d0 lib/bug.c:186
 fixup_bug arch/x86/kernel/traps.c:178 [inline]
 do_error_trap+0x1fc/0x4d0 arch/x86/kernel/traps.c:296
dst_release: dst:(____ptrval____) refcnt:-5
 do_invalid_op+0x1b/0x20 arch/x86/kernel/traps.c:316
 invalid_op+0x14/0x20 arch/x86/entry/entry_64.S:992
RIP: 0010:dst_hold include/net/dst.h:239 [inline]
RIP: 0010:ip6_setup_cork+0xd66/0x1830 net/ipv6/ip6_output.c:1204
Code: c1 ed 03 89 9d 18 ff ff ff 48 b8 00 00 00 00 00 fc ff df 41 c6 44 05 00 f8 e9 2d 01 00 00 4c 8b a5 c8 fe ff ff e8 1a f6 e6 fa <0f> 0b e9 6a fc ff ff e8 0e f6 e6 fa 48 8b 85 d0 fe ff ff 48 8d 78
RSP: 0018:ffff8801a8fcf178 EFLAGS: 00010293
RAX: ffff8801a8eba5c0 RBX: 0000000000000000 RCX: ffffffff869511e6
RDX: 0000000000000000 RSI: ffffffff869515b6 RDI: 0000000000000005
RBP: ffff8801a8fcf2c8 R08: ffff8801a8eba5c0 R09: ffffed0035ac8338
R10: ffffed0035ac8338 R11: ffff8801ad6419c3 R12: ffff8801a8fcf720
R13: ffff8801a8fcf6a0 R14: ffff8801ad6419c0 R15: ffff8801ad641980
 ip6_make_skb+0x2c8/0x600 net/ipv6/ip6_output.c:1768
 udpv6_sendmsg+0x2c90/0x35f0 net/ipv6/udp.c:1376
 inet_sendmsg+0x1a1/0x690 net/ipv4/af_inet.c:798
 sock_sendmsg_nosec net/socket.c:641 [inline]
 sock_sendmsg+0xd5/0x120 net/socket.c:651
 ___sys_sendmsg+0x51d/0x930 net/socket.c:2125
 __sys_sendmmsg+0x240/0x6f0 net/socket.c:2220
 __do_sys_sendmmsg net/socket.c:2249 [inline]
 __se_sys_sendmmsg net/socket.c:2246 [inline]
 __x64_sys_sendmmsg+0x9d/0x100 net/socket.c:2246
 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290
 entry_SYSCALL_64_after_hwframe+0x49/0xbe
RIP: 0033:0x446ba9
Code: e8 cc bb 02 00 48 83 c4 18 c3 0f 1f 80 00 00 00 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 eb 08 fc ff c3 66 2e 0f 1f 84 00 00 00 00
RSP: 002b:00007fb39a469da8 EFLAGS: 00000246 ORIG_RAX: 0000000000000133
RAX: ffffffffffffffda RBX: 00000000006dcc54 RCX: 0000000000446ba9
RDX: 00000000000000b8 RSI: 0000000020001b00 RDI: 0000000000000003
RBP: 00000000006dcc50 R08: 00007fb39a46a700 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 45c828efc7a64843
R13: e6eeb815b9d8a477 R14: 5068caf6f713c6fc R15: 0000000000000001
Dumping ftrace buffer:
   (ftrace buffer empty)
Kernel Offset: disabled
Rebooting in 86400 seconds..

Fixes: 93531c674315 ("net/ipv6: separate handling of FIB entries from dst based routes")
Reported-by: syzbot+902e2a1bcd4f7808cef5@syzkaller.appspotmail.com
Reported-by: syzbot+8ae62d67f647abeeceb9@syzkaller.appspotmail.com
Reported-by: syzbot+3f08feb14086930677d0@syzkaller.appspotmail.com
Signed-off-by: Wei Wang <weiwan@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/ip6_fib.h
net/ipv6/addrconf.c
net/ipv6/route.c

index 71b9043..3d49305 100644 (file)
@@ -281,6 +281,11 @@ static inline void fib6_info_hold(struct fib6_info *f6i)
        atomic_inc(&f6i->fib6_ref);
 }
 
+static inline bool fib6_info_hold_safe(struct fib6_info *f6i)
+{
+       return atomic_inc_not_zero(&f6i->fib6_ref);
+}
+
 static inline void fib6_info_release(struct fib6_info *f6i)
 {
        if (f6i && atomic_dec_and_test(&f6i->fib6_ref))
index 91580c6..f66a1ca 100644 (file)
@@ -2374,7 +2374,8 @@ static struct fib6_info *addrconf_get_prefix_route(const struct in6_addr *pfx,
                        continue;
                if ((rt->fib6_flags & noflags) != 0)
                        continue;
-               fib6_info_hold(rt);
+               if (!fib6_info_hold_safe(rt))
+                       continue;
                break;
        }
 out:
index 2ce0bd1..ec18b3c 100644 (file)
@@ -972,10 +972,10 @@ static void ip6_rt_init_dst(struct rt6_info *rt, struct fib6_info *ort)
        rt->dst.lastuse = jiffies;
 }
 
+/* Caller must already hold reference to @from */
 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
 {
        rt->rt6i_flags &= ~RTF_EXPIRES;
-       fib6_info_hold(from);
        rcu_assign_pointer(rt->from, from);
        dst_init_metrics(&rt->dst, from->fib6_metrics->metrics, true);
        if (from->fib6_metrics != &dst_default_metrics) {
@@ -984,6 +984,7 @@ static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
        }
 }
 
+/* Caller must already hold reference to @ort */
 static void ip6_rt_copy_init(struct rt6_info *rt, struct fib6_info *ort)
 {
        struct net_device *dev = fib6_info_nh_dev(ort);
@@ -1044,9 +1045,14 @@ static struct rt6_info *ip6_create_rt_rcu(struct fib6_info *rt)
        struct net_device *dev = rt->fib6_nh.nh_dev;
        struct rt6_info *nrt;
 
+       if (!fib6_info_hold_safe(rt))
+               return NULL;
+
        nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
        if (nrt)
                ip6_rt_copy_init(nrt, rt);
+       else
+               fib6_info_release(rt);
 
        return nrt;
 }
@@ -1178,10 +1184,15 @@ static struct rt6_info *ip6_rt_cache_alloc(struct fib6_info *ort,
         *      Clone the route.
         */
 
+       if (!fib6_info_hold_safe(ort))
+               return NULL;
+
        dev = ip6_rt_get_dev_rcu(ort);
        rt = ip6_dst_alloc(dev_net(dev), dev, 0);
-       if (!rt)
+       if (!rt) {
+               fib6_info_release(ort);
                return NULL;
+       }
 
        ip6_rt_copy_init(rt, ort);
        rt->rt6i_flags |= RTF_CACHE;
@@ -1210,12 +1221,17 @@ static struct rt6_info *ip6_rt_pcpu_alloc(struct fib6_info *rt)
        struct net_device *dev;
        struct rt6_info *pcpu_rt;
 
+       if (!fib6_info_hold_safe(rt))
+               return NULL;
+
        rcu_read_lock();
        dev = ip6_rt_get_dev_rcu(rt);
        pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags);
        rcu_read_unlock();
-       if (!pcpu_rt)
+       if (!pcpu_rt) {
+               fib6_info_release(rt);
                return NULL;
+       }
        ip6_rt_copy_init(pcpu_rt, rt);
        pcpu_rt->rt6i_flags |= RTF_PCPU;
        return pcpu_rt;
@@ -2486,7 +2502,7 @@ restart:
 
 out:
        if (ret)
-               dst_hold(&ret->dst);
+               ip6_hold_safe(net, &ret, true);
        else
                ret = ip6_create_rt_rcu(rt);
 
@@ -3303,7 +3319,8 @@ static int ip6_route_del(struct fib6_config *cfg,
                                continue;
                        if (cfg->fc_protocol && cfg->fc_protocol != rt->fib6_protocol)
                                continue;
-                       fib6_info_hold(rt);
+                       if (!fib6_info_hold_safe(rt))
+                               continue;
                        rcu_read_unlock();
 
                        /* if gateway was specified only delete the one hop */
@@ -3409,6 +3426,9 @@ static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_bu
 
        rcu_read_lock();
        from = rcu_dereference(rt->from);
+       /* This fib6_info_hold() is safe here because we hold reference to rt
+        * and rt already holds reference to fib6_info.
+        */
        fib6_info_hold(from);
        rcu_read_unlock();
 
@@ -3470,7 +3490,8 @@ static struct fib6_info *rt6_get_route_info(struct net *net,
                        continue;
                if (!ipv6_addr_equal(&rt->fib6_nh.nh_gw, gwaddr))
                        continue;
-               fib6_info_hold(rt);
+               if (!fib6_info_hold_safe(rt))
+                       continue;
                break;
        }
 out:
@@ -3530,8 +3551,8 @@ struct fib6_info *rt6_get_dflt_router(struct net *net,
                    ipv6_addr_equal(&rt->fib6_nh.nh_gw, addr))
                        break;
        }
-       if (rt)
-               fib6_info_hold(rt);
+       if (rt && !fib6_info_hold_safe(rt))
+               rt = NULL;
        rcu_read_unlock();
        return rt;
 }
@@ -3579,8 +3600,8 @@ restart:
                struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
 
                if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
-                   (!idev || idev->cnf.accept_ra != 2)) {
-                       fib6_info_hold(rt);
+                   (!idev || idev->cnf.accept_ra != 2) &&
+                   fib6_info_hold_safe(rt)) {
                        rcu_read_unlock();
                        ip6_del_rt(net, rt);
                        goto restart;