Skip to contentSkip to conversation

why is the sys_close symbol exported ?

DISCUSSION jmfriedt 2 participants 2 messages Nov 18, 2016

lore
Share & tools
lore

jmfriedt

author
#0 lore raw

Following the various rootkit and system call redirection developments, the current way of identifying the location of the system call table seems to be brute force scanning the memory for the location of one of the system calls. This is only possible from a module if the symbol is exported: I see that only one system call symbol is still exported, that is sys_close. Removing this symbol export would hinder one of the ways of finding the systam call table: I have not been able to find the reason for exporting this particular symbol (while sys_open for example is not exported). Can anyone justify why that is ?

Thank you, Jean-Michel

Signature · 1 line
JM Friedt, FEMTO-ST Time & Frequency/SENSeOR, 26 rue de l'Epitaphe, 25000 Besancon, France
#1 lore raw
Quoted 10 lines · On Fri, 18 Nov 2016 09:56:52 +0100 jmfriedt <[email protected]> wrote:

Following the various rootkit and system call redirection developments, the current way of identifying the location of the system call table seems to be brute force scanning the memory for the location of one of the system calls. This is only possible from a module if the symbol is exported: I see that only one system call symbol is still exported, that is sys_close. Removing this symbol export would hinder one of the ways of finding the systam call table: I have not been able to find the reason for exporting this particular symbol (while sys_open for example is not exported). Can anyone justify why that is ?

Thank you, Jean-Michel

find . -name "*.[ch]" -exec grep -H sys_close {} \;

So currently it is needed by autofs4, binfmt_misc, net/kcm/kcmsock and those look like legitimate use cases.

It might be worth changing sys_close to just wrap a call to do_sys_close() which is the existing code. That would make it slightly harder.

That said anyone doing syscall table scanning that would can do it at least two other pretty reliable ways by working form the system call entry point, which is trivially discoverable.

It's one reason I'd really like to see kvm/qemu provide 'read only until the virtual machine exits' memory range so that you can irrevocably protect page ranges (like the syscalls and much of the kernel code) within a VM.

Alan