nsjail/kafel for other architectures about nsjail HOT 8 CLOSED

Here's import for aarch64 - google/kafel@2ae8e11

But generally, yes, argument sizes might be different between architectures - but generally the same for the same size of bitness, so it should be technically ok to use existing arm template for mips32 syscalls, and aarch64 template for mips64 sysacalls. The argument sizes are typically the same (e.g. integers, masks), just things like addresses have different bitness.

from nsjail.

Hm.. the whole syscall table is architecture-specific, I don't think these should be ignored, but simply added to the array?

{"sysmips",
 4149 ,
 NORMAL,
 {
         [ARG_0] = .....
 }},

Or, am I misunderstanding here? If you did mips syscall table on the basis of aarch64 or some other syscall table, then the way to go is to simply add them (at the end, or sorted, I guess it doesn't matter). We don't have any canonical syscall tables.

from nsjail.

I was (still am) a bit confused about few differences between naming in this table and the one used in the nsjail. Which one is correct ? or does it even matter!?
I will send a pull request soon, but please be critical about it and check it properly before merge ;) as I only have a little experience with these details. :)
Thanks

from nsjail.

I just noticed that my method will not work as the size of input arguments for the same syscall is different in different architectures:
accept4 in arm_syscall.c

    {"accept4",
     366,
     NORMAL,
     {
             [ARG_0] = {"fd", 4}, [ARG_1] = {"upeer_sockaddr", 4},
             [ARG_2] = {"upeer_addrlen", 4}, [ARG_3] = {"flags", 4},
     }},

accept4 in amd64_syscall.c

    {"accept4",
     288,
     NORMAL,
     {
             [ARG_0] = {"fd", 4}, [ARG_1] = {"upeer_sockaddr", 8},
             [ARG_2] = {"upeer_addrlen", 8}, [ARG_3] = {"flags", 4},
     }},

I also noticed that some syscalls that are available on all architecture are missing in kafel such as statx and pkey_mprotect. Please take a look at this: https://fedora.juszkiewicz.com.pl/syscalls.html

from nsjail.

In other words, if you're trying to create syscall table for mips32, please use arm_syscall template, if for mips64 please try to use aarch64 template. Then please try to remove syscalls which are not present in mips, and add syscalls which are specific to mips.

from nsjail.

The number of undefined syscalls in the templates are high. I should think about a better solution to generate this list!
https://github.com/ebadi/syscalls-table/blob/master/parser.py

python3 parser.py 32 mipso32
Syscall found: 294
Syscall not found 63

python3 parser.py 64 mips64
Syscall found: 239
Syscall not found 85

from nsjail.

What about manually checking (at least some rudumentaty check) with

https://github.com/torvalds/linux/blob/master/arch/mips/kernel/scall64-64.S

and

https://github.com/torvalds/linux/blob/master/arch/mips/kernel/scall32-o32.S

Given that we don't have tables for mips yet, I think we could marge your tables as-is, given that it passes this rudimentary/manual check (total number of syscalls, a few examples of syscalls and their numbers in your table), and fix any eventual problems later on?

from nsjail.

I cleaned up my code here and did a quick check to make sure that syscall numbers match with the links you provided. Please take a look at the result

from nsjail.