All Projects → stealth → call-graphs

stealth / call-graphs

Licence: other
No description, website, or topics provided.

PrivSep calling

-finstrument-functions past and present

I am instrumenting processes since more than 20y, mostly with very raw and basic methods. On the plus side, you dont need to install nodejs on your machine just to pop some memory contents off a process.

I started building call graphs for OpenSSH somewhere in 2005. The aim was to better understand the relations between the various sshd processes in advent of PrivilegeSeparation. Call graphs by itself are little helpful for code reviews, except for such cases where you struggle to find the IPC and privilege relations between all the forked processes. The code that I wrote 12y ago is not working anymore today. This document describes why and which hurdles you need to overcome in order to produce a valuable call graph for an OpenSSH sshd session today.

Here we go

  • sshd is not an ET_EXEC ELF file, but of type ET_DYN:
ELF Header:
  Magic:   7f 45 4c 46 02 01 01 00 00 00 00 00 00 00 00 00 
  Class:                             ELF64
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              DYN (Shared object file)
  Machine:                           Advanced Micro Devices X86-64
  Version:                           0x1
  Entry point address:               0x31337
...

This means the kernel ELF loader will apply ASLR also for the PT_LOAD segments from the ELF base image itself, not just for the DT_NEEDED libraries which are loaded as dependencies by ld.so. This is what you get when building your binary with -pie. It is meant as a hardening measure against the borrowed code chunks attack which some people wrongly call ROP.

  • Therefore, static processing of libs and binaries is not sufficient to resolve addresses to symbol names. One has to check at runtime whats loaded and at which addresses before one can parse symbol tables.
  • sshd is using a strange mix of processes, re-executions and threads inside these processes (let alone signal handlers) which must be synchronized to get a valuable instrumentation tracelog. This mix is due to the PrivilegeSeparation efforts of OpenSSH.
  • Some of the synchronization primitives one would normally use in the C/C++ runtime environment may be using the futex() syscall, which is not an allowed syscall inside the sshd seccomp sandbox.
  • sshd may be closing arbitrary file-descriptors at any time - most commonly when setting up new sessions - which is affecting instrumentation log-files. Re-opening files inside the seccomp sandbox is not allowed.
  • In order to cluster the graph in a way it makes sense, one wants the tracelog in a single file and not scattered across directories and jails. As a bonus, inside the ascii call graph you can see the scheduler switching between processes.

You can find a complete sshd session inside this repo as dot and ascii files. Use dotty and the bird view to navigate through the graph. Red nodes (functions) are running privileged (euid 0). Grey is for unprivileged calls. Euid 496 is the seccomp sandbox user. You will see five sub-graphs, each one denoting its own process. These are the offspring after the re-exec: accepting sshd, its child priv sshd, sandboxed net sshd, pam sshd and the user-session sshd holding the pty.

openssh-7.4p1-sshd-session.jpg

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].