vxlan: Do not assume RTNL is held in vxlan_fdb_info()
authorIdo Schimmel <idosch@mellanox.com>
Sun, 24 May 2020 21:38:56 +0000 (00:38 +0300)
committerDavid S. Miller <davem@davemloft.net>
Mon, 25 May 2020 02:34:11 +0000 (19:34 -0700)
vxlan_fdb_info() is not always called with RTNL held or from an RCU
read-side critical section. For example, in the following call path:

vxlan_cleanup()
  vxlan_fdb_destroy()
    vxlan_fdb_notify()
      __vxlan_fdb_notify()
        vxlan_fdb_info()

The use of rtnl_dereference() can therefore result in the following
splat [1].

Fix this by dereferencing the nexthop under RCU read-side critical
section.

[1]
[May24 22:56] =============================
[  +0.004676] WARNING: suspicious RCU usage
[  +0.004614] 5.7.0-rc5-custom-16219-g201392003491 #2772 Not tainted
[  +0.007116] -----------------------------
[  +0.004657] drivers/net/vxlan.c:276 suspicious rcu_dereference_check() usage!
[  +0.008164]
              other info that might help us debug this:

[  +0.009126]
              rcu_scheduler_active = 2, debug_locks = 1
[  +0.007504] 5 locks held by bash/6892:
[  +0.004392]  #0: ffff8881d47e3410 (&sig->cred_guard_mutex){+.+.}-{3:3}, at: __do_execve_file.isra.27+0x392/0x23c0
[  +0.011795]  #1: ffff8881d47e34b0 (&sig->exec_update_mutex){+.+.}-{3:3}, at: flush_old_exec+0x510/0x2030
[  +0.010947]  #2: ffff8881a141b0b0 (ptlock_ptr(page)#2){+.+.}-{2:2}, at: unmap_page_range+0x9c0/0x2590
[  +0.010585]  #3: ffff888230009d50 ((&vxlan->age_timer)){+.-.}-{0:0}, at: call_timer_fn+0xe8/0x800
[  +0.010192]  #4: ffff888183729bc8 (&vxlan->hash_lock[h]){+.-.}-{2:2}, at: vxlan_cleanup+0x133/0x4a0
[  +0.010382]
              stack backtrace:
[  +0.005103] CPU: 1 PID: 6892 Comm: bash Not tainted 5.7.0-rc5-custom-16219-g201392003491 #2772
[  +0.009675] Hardware name: Mellanox Technologies Ltd. MSN2100-CB2FO/SA001017, BIOS 5.6.5 06/07/2016
[  +0.010155] Call Trace:
[  +0.002775]  <IRQ>
[  +0.002313]  dump_stack+0xfd/0x178
[  +0.003895]  lockdep_rcu_suspicious+0x14a/0x153
[  +0.005157]  vxlan_fdb_info+0xe39/0x12a0
[  +0.004775]  __vxlan_fdb_notify+0xb8/0x160
[  +0.004672]  vxlan_fdb_notify+0x8e/0xe0
[  +0.004370]  vxlan_fdb_destroy+0x117/0x330
[  +0.004662]  vxlan_cleanup+0x1aa/0x4a0
[  +0.004329]  call_timer_fn+0x1c4/0x800
[  +0.004357]  run_timer_softirq+0x129d/0x17e0
[  +0.004762]  __do_softirq+0x24c/0xaef
[  +0.004232]  irq_exit+0x167/0x190
[  +0.003767]  smp_apic_timer_interrupt+0x1dd/0x6a0
[  +0.005340]  apic_timer_interrupt+0xf/0x20
[  +0.004620]  </IRQ>

Fixes: 1274e1cc4226 ("vxlan: ecmp support for mac fdb entries")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reported-by: Amit Cohen <amitc@mellanox.com>
Acked-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/vxlan.c

index 3e88fbe..a0015cd 100644 (file)
@@ -263,6 +263,8 @@ static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
        struct nlmsghdr *nlh;
        struct nexthop *nh;
        struct ndmsg *ndm;
+       int nh_family;
+       u32 nh_id;
 
        nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
        if (nlh == NULL)
@@ -273,13 +275,20 @@ static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 
        send_eth = send_ip = true;
 
-       nh = rcu_dereference_rtnl(fdb->nh);
+       rcu_read_lock();
+       nh = rcu_dereference(fdb->nh);
+       if (nh) {
+               nh_family = nexthop_get_family(nh);
+               nh_id = nh->id;
+       }
+       rcu_read_unlock();
+
        if (type == RTM_GETNEIGH) {
                if (rdst) {
                        send_ip = !vxlan_addr_any(&rdst->remote_ip);
                        ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
                } else if (nh) {
-                       ndm->ndm_family = nexthop_get_family(nh);
+                       ndm->ndm_family = nh_family;
                }
                send_eth = !is_zero_ether_addr(fdb->eth_addr);
        } else
@@ -299,7 +308,7 @@ static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
        if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
                goto nla_put_failure;
        if (nh) {
-               if (nla_put_u32(skb, NDA_NH_ID, nh->id))
+               if (nla_put_u32(skb, NDA_NH_ID, nh_id))
                        goto nla_put_failure;
        } else if (rdst) {
                if (send_ip && vxlan_nla_put_addr(skb, NDA_DST,