Re: [PATCH v2 net 13/17] net: Fix data-races around sysctl_fb_tunnels_only_for_init_net.

[Kuniyuki Iwashima <kuniyu@amazon.com>]

From:   Nathan Chancellor <nathan@kernel.org>
Date:   Thu, 18 Aug 2022 09:23:19 -0700
> On Thu, Aug 18, 2022 at 08:01:54AM -0700, Kuniyuki Iwashima wrote:
> > From:   kernel test robot <lkp@intel.com>
> > Date:   Thu, 18 Aug 2022 16:51:37 +0800
> > > Hi Kuniyuki,
> > > 
> > > Thank you for the patch! Yet something to improve:
> > > 
> > > [auto build test ERROR on net/master]
> > > 
> > > url:    https://github.com/intel-lab-lkp/linux/commits/Kuniyuki-Iwashima/net-sysctl-Fix-data-races-around-net-core-XXX/20220818-115941
> > > base:   https://git.kernel.org/pub/scm/linux/kernel/git/davem/net.git fc4aaf9fb3c99bcb326d52f9d320ed5680bd1cee
> > > config: riscv-randconfig-r032-20220818 (https://download.01.org/0day-ci/archive/20220818/202208181615.Lu9xjiEv-lkp@intel.com/config)
> > > compiler: clang version 16.0.0 (https://github.com/llvm/llvm-project aed5e3bea138ce581d682158eb61c27b3cfdd6ec)
> > > reproduce (this is a W=1 build):
> > >         wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
> > >         chmod +x ~/bin/make.cross
> > >         # install riscv cross compiling tool for clang build
> > >         # apt-get install binutils-riscv64-linux-gnu
> > >         # https://github.com/intel-lab-lkp/linux/commit/6bc3dfb3dc4862f4e00ba93c45cd5c0251b85d5b
> > >         git remote add linux-review https://github.com/intel-lab-lkp/linux
> > >         git fetch --no-tags linux-review Kuniyuki-Iwashima/net-sysctl-Fix-data-races-around-net-core-XXX/20220818-115941
> > >         git checkout 6bc3dfb3dc4862f4e00ba93c45cd5c0251b85d5b
> > >         # save the config file
> > >         mkdir build_dir && cp config build_dir/.config
> > >         COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=riscv SHELL=/bin/bash
> > > 
> > > If you fix the issue, kindly add following tag where applicable
> > > Reported-by: kernel test robot <lkp@intel.com>
> > > 
> > > All errors (new ones prefixed by >>):
> > > 
> > > >> ld.lld: error: undefined symbol: sysctl_fb_tunnels_only_for_init_net
> > >    >>> referenced by ip_tunnel.c
> > >    >>>               ipv4/ip_tunnel.o:(ip_tunnel_init_net) in archive net/built-in.a
> > >    >>> referenced by ip_tunnel.c
> > >    >>>               ipv4/ip_tunnel.o:(ip_tunnel_init_net) in archive net/built-in.a
> > 
> > Hmm... I tested allmodconfig with x86_64 but it seems not enough...
> > 
> > I don't think just using READ_ONCE() causes regression.
> > Is this really regression or always-broken stuff in some arch,
> > or ... clang 16?
> > 
> > Anyway, I'll take a look.
> 
> You'll see the same error with the same configuration and GCC:
> 
> riscv64-linux-gnu-ld: net/ipv4/ip_tunnel.o: in function `.L536':
> ip_tunnel.c:(.text+0x1d4a): undefined reference to `sysctl_fb_tunnels_only_for_init_net'
> riscv64-linux-gnu-ld: ip_tunnel.c:(.text+0x1d4e): undefined reference to `sysctl_fb_tunnels_only_for_init_net'
> 
> $ scripts/config --file build/riscv/.config -s SYSCTL
> undef
> 
> Prior to your change, the '!IS_ENABLED(CONFIG_SYSCTL)' would cause
> net_has_fallback_tunnels() to unconditionally 'return 1' in the
> CONFIG_SYSCTL=n case,

Yes, you are right.
I've just noticed it and this fixed the error.
Also, I did the same mistake in the 14th patch...
Thank you, Nathan!

---8<---
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 2563d30736e9..78dd63a5c7c8 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -640,9 +640,14 @@ extern int sysctl_devconf_inherit_init_net;
  */
 static inline bool net_has_fallback_tunnels(const struct net *net)
 {
-	return !IS_ENABLED(CONFIG_SYSCTL) ||
-	       !sysctl_fb_tunnels_only_for_init_net ||
-	       (net == &init_net && sysctl_fb_tunnels_only_for_init_net == 1);
+#if IS_ENABLED(CONFIG_SYSCTL)
+	int fb_tunnels_only_for_init_net = READ_ONCE(sysctl_fb_tunnels_only_for_init_net);
+
+	return !fb_tunnels_only_for_init_net ||
+		(net_eq(net, &init_net) && fb_tunnels_only_for_init_net == 1);
+#else
+	return true;
+#endif
 }
 
 static inline int netdev_queue_numa_node_read(const struct netdev_queue *q)
---8<---


> meaning 'fb_tunnels_only_for_init_net' was never
> emitted in the final assembly so the linker would not complain about it
> never being defined (the kernel relies on this trick a lot, which is why
> you cannot build the kernel with -O0).
> 
> After your change, sysctl_fb_tunnels_only_for_init_net is
> unconditionally used but it is only defined in sysctl_net_core.c, which
> is only built when CONFIG_SYSCTL=y, hence the link error.
> 
> I suspect hoisting '!IS_ENABLED(CONFIG_SYSCTL)' out of the return into
> its own conditional would fix the error:
> 
>   static inline bool net_has_fallback_tunnels(const struct net *net)
>   {
>       int fb_tunnels_only_for_init_net;
> 
>       if (!IS_ENABLED(CONFIG_SYSCTL))
>           return true;
> 
>       fb_tunnels_only_for_init_net = READ_ONCE(sysctl_fb_tunnels_only_for_init_net);
> 
>       return !fb_tunnels_only_for_init_net ||
>              (net == &init_net && fb_tunnels_only_for_init_net == 1)
>   }
> 
> Cheers,
> Nathan