net: revert "net: get rid of an signed integer overflow in ip_idents_reserve()"
authorYuqi Jin <jinyuqi@huawei.com>
Sat, 16 May 2020 03:46:49 +0000 (11:46 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 3 Jun 2020 06:17:33 +0000 (08:17 +0200)
[ Upstream commit a6211caa634da39d861a47437ffcda8b38ef421b ]

Commit adb03115f459 ("net: get rid of an signed integer overflow in ip_idents_reserve()")
used atomic_cmpxchg to replace "atomic_add_return" inside the function
"ip_idents_reserve". The reason was to avoid UBSAN warning.
However, this change has caused performance degrade and in GCC-8,
fno-strict-overflow is now mapped to -fwrapv -fwrapv-pointer
and signed integer overflow is now undefined by default at all
optimization levels[1]. Moreover, it was a bug in UBSAN vs -fwrapv
/-fno-strict-overflow, so Let's revert it safely.

[1] https://gcc.gnu.org/gcc-8/changes.html

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Suggested-by: Eric Dumazet <edumazet@google.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: Arvind Sankar <nivedita@alum.mit.edu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Jiong Wang <jiongwang@huawei.com>
Signed-off-by: Yuqi Jin <jinyuqi@huawei.com>
Signed-off-by: Shaokun Zhang <zhangshaokun@hisilicon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/ipv4/route.c

index 967acf2844ef1e46c9b44cce75e1ead3e30fa29a..a894adbb6c9b5584b6a1f924d49b2a8323ee8040 100644 (file)
@@ -499,18 +499,16 @@ u32 ip_idents_reserve(u32 hash, int segs)
        atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
        u32 old = ACCESS_ONCE(*p_tstamp);
        u32 now = (u32)jiffies;
-       u32 new, delta = 0;
+       u32 delta = 0;
 
        if (old != now && cmpxchg(p_tstamp, old, now) == old)
                delta = prandom_u32_max(now - old);
 
-       /* Do not use atomic_add_return() as it makes UBSAN unhappy */
-       do {
-               old = (u32)atomic_read(p_id);
-               new = old + delta + segs;
-       } while (atomic_cmpxchg(p_id, old, new) != old);
-
-       return new - segs;
+       /* If UBSAN reports an error there, please make sure your compiler
+        * supports -fno-strict-overflow before reporting it that was a bug
+        * in UBSAN, and it has been fixed in GCC-8.
+        */
+       return atomic_add_return(segs + delta, p_id) - segs;
 }
 EXPORT_SYMBOL(ip_idents_reserve);