[Keeper]

Is there any way at run time to know if srcds_i486 or srcds_i686 was run? It matters in the EP1 games when you try to reach into the engine_i486/i686.so.

Or am I going about it wrong?

TIA,
Keeper

[BAILOPAN]

Use dlinfo() and pass the address of something you know is in engine_mumble_grumble.so

[Keeper]

Ok, I said "worked like a charm" ... it showed what I thought it should show. For EP1 games, all I get for a response on engine_i486.so is:

Found gametypes for Half-Life 2 Deathmatch                        
1: ./bin/engine_i486.so                                           
Linux game binary @ ./hl2mp/bin/server_i486.so                    
Linux engine binary @ ./bin/engine_i686.so

Basically since it doesn't show any other libraries I consider it unused. Here's what it shows for engine_i686.so:

Found gametypes for Half-Life 2 Deathmatch                                      
1: bin/engine_i686.so                                                           
2: bin/libsteamvalidateuseridtickets_i486.so                                    
3: bin/steam_api_i486.so                                                        
4: bin/shaderapiempty_i486.so                                                   
5: /plugins/srcds_1/hl2mp/bin/server_i486.so                                    
6: /plugins/srcds_1/bin/scenefilecache_i486.so                                  
7: /plugins/srcds_1/bin/../hl2mp/addons/mani_admin_plugin_i486.so               
8: /lib/tls/i686/cmov/libnss_files.so.2                                         
Linux game binary @ ./hl2mp/bin/server_i486.so                                  
Linux engine binary @ ./bin/engine_i686.so

What's strange is if I test with the i486 then on the systems that run i686 the server crashes because a hook switches it back to _i486:

 (gdb) bt
bt
#0  0xb148e82c in Host_Client_Printf () from ./bin/engine_i486.so
#1  0xb14907d9 in Host_Status_f () from ./bin/engine_i486.so
#2  0xb1905ca8 in __SourceHook_FHCls_ConCommandDispatch0::Func (this=0xb16f16a0) at mani_sourcehook.cpp:125
#3  0xb72e626b in Cmd_ExecuteString () from bin/engine_i686.so
#4  0xb72e6576 in Cbuf_Execute () from bin/engine_i686.so
#5  0xb7349b2e in _Host_RunFrame () from bin/engine_i686.so
#6  0xb734a2b2 in Host_RunFrame () from bin/engine_i686.so
#7  0xb73543bc in CHostState::State_Run () from bin/engine_i686.so
#8  0xb7354661 in CHostState::FrameUpdate () from bin/engine_i686.so
#9  0xb73547e7 in HostState_Frame () from bin/engine_i686.so
#10 0xb73e9f94 in CEngine::Frame () from bin/engine_i686.so
#11 0xb73e7dae in CDedicatedServerAPI::RunFrame () from bin/engine_i686.so
#12 0xb7e6f0fd in RunServer () from bin/dedicated_i686.so
#13 0xb73e778e in CModAppSystemGroup::Main () from bin/engine_i686.so
#14 0xb74eebc3 in CAppSystemGroup::Run () from bin/engine_i686.so
#15 0xb73e89df in CDedicatedServerAPI::ModInit () from bin/engine_i686.so
#16 0xb7e6f34a in CDedicatedAppSystemGroup::Main () from bin/dedicated_i686.so
#17 0xb7ea6713 in CAppSystemGroup::Run () from bin/dedicated_i686.so
#18 0xb7ea6713 in CAppSystemGroup::Run () from bin/dedicated_i686.so
#19 0xb7e6f758 in main () from bin/dedicated_i686.so
#20 0x0804909e in main ()

Why would it do this?

[Keeper]

Ok, figured it out. It's actually easier than I thought. All you have to do is a dlopen with the RTLD_NOLOAD flag and if the handle is null, then it's not loaded.

void * handle;
handle = dlopen ( "./bin/engine_i486.so", RTLD_NOW | RTLD_NOLOAD );

If handle is null then you are using i686 otherwise it's i486.

Thanks!
Keeper

[BAILOPAN]

I don't know why you say that's necessary. It shouldn't "switch" to i486 and I've not had that happen in my code before.

Regardless it's in poor form to re-dlopen() the executable. A few thoughts:

1. dlinfo() is a GNU-specific extension, it should get you the handle (I think).
2. I originally meant dladdr(), but I'm not sure if ::dli_fbase is a handle or the memory base of the library (I think the former).
3. We just parse /proc/<pid>/maps which is safest - but we use #2 for sanity checks. See sourcemod-central/core/MemoryUtils.cpp. But that's for signature scanning - if you need to use dlsym() then I guess you need a Handle.

[Keeper]

The newer version of the libraries support RTLD_SELF. Unfortunately it's not available to me, or else I'd use just dlinfo. I was using dlopen without the RTLD_NOLOAD which was causing the non loaded binary to be loaded which is why it switched and subsequently crashed.

Using RTLD_NOLOAD seems to work, but I will look at #3 to see if it would better suit our needs. If opening the binary even without loading it is bad form, then I don't want to do it. I'm using it to detour the ConnectClient in the engine. But I don't want to attempt to slap a detour on a non loaded binary, which is a possibility with EP1 games.

Thanks!

EDIT: I just looked in another of my older plugins. The code for the /proc/<pid>/maps is in there. I'll just use this since it will give me the info I need.