[OpenWrt-Devel] [RFC] netifd: Crash when netifd reload is handled during netifd startup

[Felix Fietkau]

Hi Hans,

thanks for testing. I uploaded a new patch (same URL), which uses a
uloop timer to defer processing of incoming invoke msgs.
Note that this changes the ubus context data structure and thus affects
everything that depends on ubus, so it's better to reflash after rebuilding.

- Felix

On 2014-06-24 16:11, Hans Dedecker wrote:
> Hi,
> 
> Applied the ubus patch but netifd_reload is still called while
> config_init_all is processing the config and thus leading to a crash
> when netifd_reload is done
> 
> Added extra traces in netifd which confirms this :
> un 24 16:00:44 OpenWrt daemon.notice netifd: config_init_all : Enter
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Create interface
> 'loopback'
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Create simple device
> 'lo'
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Initialize device 'lo'
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Remove a route from
> device lo
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Remove a route from
> device lo
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Remove a route from
> device lo
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Network device 'lo' is
> now present
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Add user for device
> 'lo', refcount=1
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Interface 'loopback',
> available=1
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Add user for device
> 'lo', refcount=2
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: netifd_reload : Enter
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: config_init_all : Enter
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Update interface
> 'loopback'
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Create new device
> 'br-lan' (Bridge)
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Initialize device
> 'br-lan'
> Jun 24 16:00:44 OpenWrt daemon.notice netifd: Create interface 'lan'
> .....
> Jun 24 16:00:45 OpenWrt daemon.notice netifd: config_init_all : Exit
> Jun 24 16:00:45 OpenWrt daemon.notice netifd: netifd_reload : Exit
> 
> 
> Hans
> On Tue, Jun 24, 2014 at 2:05 PM, Felix Fietkau <nbd at openwrt.org> wrote:
>> On 2014-06-24 12:46, Hans Dedecker wrote:
>>> Netifd is crashing when when a network reload (ubus call network reload) is handled during the parsing of the network config in the function config_init_all (called from main) at startup.
>>> As an ubus_invoke function call is issued when the interfaces are created; ubus will also process the pending ubus calls in this case the network reload during the invoke.
>>> As netifd_reload calls again config_init_all network config will be parsed again; on return from netifd_reload the original config_init_all function call will continue but will crash as references hold to interface/device/etc ... lists are not correct anymore.
>>> This potential problem has always been present but due to netifd_reload timing behavior change in netifd commit 5db02763d61785529bef538f196c180e968b7c26 this problem can easily be triggered.
>>> To solve the issue I was thinking about deferring the network reload when the function config_init_all is parsing the config.
>>> Any opinion if this is the correct way to go or any other alternatives ?
>> Please try applying this patch to ubus:
>> http://nbd.name/libubus-req-defer.patch
>>
>> It should ensure that no invoke will be processed while netifd is busy
>> with registering/unregistering objects or sending notify calls.
>>
>> - Felix
> 
_______________________________________________
openwrt-devel mailing list
openwrt-devel at lists.openwrt.org
https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel