Recent capa version don't work with sample dumped from memory

[Ana06]

Description

I have dumped the UPX packed sample a27afd3ec5b3f2e705347b2c39847fcd659e430396abd081109db0e7f96d1d3e from memory using pd64. capa 7.3-8.0 does not detect any capabilities in the dumped sample, while capa 7.0-7.2 does.

I wonder why capa does not detect the capabilities of this sample in the newer versions and if we can fix it.

capa 7.3-8.0 does detect capabilities on the UPX unpacked version (unpacked using upx -d after renaming the sections).

Steps to Reproduce

1. Execute a27afd3ec5b3f2e705347b2c39847fcd659e430396abd081109db0e7f96d1d3e
2. Dump the executed process with pd64 -pid <PID>
3. Run capa 7.3 and 7.2 on the dumped sample

Versions

capa 7.3-8.0 standalone tool.

[mike-hunhoff]

It appears that something broke when we upgraded Vivisect from version 1.1.1 to >= 1.2.0.

Vivisect 1.1.1:

$ python scripts/show-features.py dumped.exe_  | grep api
  insn: 0x401070: api(__set_app_type)
  insn: 0x4010C7: api(__set_app_type)
  insn: 0x40119E: api(__getmainargs)
  insn: 0x401212: api(Sleep)
  insn: 0x401284: api(SetUnhandledExceptionFilter)
  insn: 0x40134C: api(malloc)
  insn: 0x401374: api(strlen)
  insn: 0x401380: api(malloc)
  insn: 0x401398: api(memcpy)
  insn: 0x401406: api(_cexit)
  insn: 0x40144C: api(_amsg_exit)
  insn: 0x40146A: api(_initterm)
  insn: 0x401493: api(GetStartupInfoA)
  insn: 0x401493: api(GetStartupInfo)
  insn: 0x4014B4: api(_initterm)
  insn: 0x401559: api(GetModuleHandleA)
  insn: 0x401559: api(GetModuleHandle)
  insn: 0x40156E: api(GetProcAddress)
[...]
  insn: 0x402E02: api(RtlDeleteCriticalSection)
  insn: 0x402E37: api(RtlInitializeCriticalSection)
  insn: 0x402EFA: api(strlen)
  insn: 0x402F77: api(strncmp)
  insn: 0x401559: api(GetModuleHandleA)
  insn: 0x401559: api(GetModuleHandle)
  insn: 0x40156E: api(GetProcAddress)
  insn: 0x412712: api(VirtualProtect)
  insn: 0x412730: api(VirtualProtect)
$

Vivisect 1.2.1:

$ python scripts/show-features.py dumped.exe_  | grep api
  insn: 0x412712: api(VirtualProtect)
  insn: 0x412730: api(VirtualProtect)
$

I'm not sure of the issue source. It could be a change to Vivisect between 1.1.1 and 1.2.0 or something that we didn't change in capa and/or viv_utils to accommodate the Vivisect changes. Running show-features.py for the function 0x401635 tells us that capa isn't seeing a majority of the instructions/basic blocks when using Vivisect 1.2.1 versus Vivisect 1.1.1:

Vivisect 1.1.1:

$ python scripts/show-features.py -F 0x401635 dumped.exe_
global: global: format(pe)
global: global: os(windows)
global: global: arch(amd64)
func: 0x401635
 func: 0x401635: characteristic(calls to) -> 0x401C9C
 func: 0x401635: characteristic(calls to) -> 0x401C9C
 bb: 0x401635: basic block
  insn: 0x401635: mnemonic(push)
 [...]
  insn: 0x401696: mnemonic(call)
  insn: 0x401696: characteristic(indirect call)
 bb: 0x401698: basic block
  insn: 0x401698: mnemonic(nop)
  insn: 0x401699: mnemonic(add)
  insn: 0x40169D: mnemonic(pop)
  insn: 0x40169E: mnemonic(ret)
$

Vivisect 1.2.1:

$ python scripts/show-features.py -F 0x401635 dumped.exe_
global: global: format(pe)
global: global: os(windows)
global: global: arch(amd64)
func: 0x401635
 func: 0x401635: characteristic(calls to) -> 0x401C9C
 bb: 0x401635: basic block
  insn: 0x401635: mnemonic(push)
$ 

[williballenthin]

PE generally looks ok, but imports aren't reconstructed:

[williballenthin]

the same number of functions are recovered using viv 1.1.1 and viv 1.2.1, but the older version recovers many more instruction features:

[williballenthin]

the new version seems to be recovering only a single instruction from each function:

[williballenthin]

there are lots of warnings during viv workspace creation, and the addresses seem to line up with functions/instructions above:

the logging message is here:
https://github.com/vivisect/vivisect/blob/5531bdae6f6511028eb3692f410ee763c77356f0/vivisect/tools/graphutil.py#L522
but this code hasn't been touched in 5 years or so. So the bug is probably in a different routine.

[williballenthin]

by manually bisecting the changes, here's the offending viv commit:

vivisect/vivisect@6d6a5d9

[williballenthin]

this was added in vivisect/vivisect#622