Vector35 / Vector35/binaryninja-api
MLIL call-parameter recovery failure
Nobody has claimed this yet.
- Dominant language
- C++
- Stars
- 1.3k
- Forks
- 298
- Avg merge
- 5d 5h
- Merged PRs (30d)
- 19
Description
Version and Platform (required):
- Binary Ninja Version: 5.4.10257-dev, caa625a3
- OS: macos
- OS Version: 26.5.2
- CPU Architecture: arm64
Bug Description:
MLIL call-parameter recovery drops a stack argument when two adjacent 32-bit outgoing argument slots overlap a previously inferred 64-bit stack variable.
This occurs in the list function of the CGC challenge binary KPRCA_00024/cb_1. A call to sprintf has seven arguments, all of which are clearly present in LLIL:
[esp + 0x00] = edi // dst
[esp + 0x04] = 0x804bca3 // format
[esp + 0x08] = edi // prefix
[esp + 0x0c] = esi // name
[esp + 0x10] = edx // price_integer
[esp + 0x14] = eax // price_fraction
[esp + 0x18] = ebx // count
call(0x804b520) // sprintf
However, MLIL contains only six parameters:
0x804b520(
var_105c_1,
"%s%s (Price: %d.%02d, Count: %d)\n",
var_1050_1,
var_104c_1,
var_1048_1,
var_1044_1
)
The missing argument is the value stored at [esp + 0x08], which should be the first variadic argument corresponding to the first %s.
LLIL identifies the two adjacent slots as fields of the same eight-byte stack variable:
[esp + 8 {var_1058+0x4}].d = edi
[esp + 4 {var_1058}].d = 0x804bca3
The lower four-byte field at [esp + 4] is recovered as the format argument, but the upper four-byte field at [esp + 8] is not emitted as an independent MLIL call parameter.
A call-site type override does not resolve the issue. The following non-variadic seven-parameter type is correctly applied:
int32_t(
char* dst,
const char* format,
char* prefix,
char* name,
int32_t price_integer,
int32_t price_fraction,
int32_t count
)
The adjusted type reports seven parameters, cdecl, and has_variable_arguments == false, but both non-SSA MLIL and SSA MLIL still contain only six call parameters.
Because the missing parameter occurs in the middle of the argument list, the displayed parameter names become shifted. This makes the issue appear in the GUI as though the final count parameter were missing, although the value corresponding to count is present as var_1044_1.
Steps To Reproduce:
-
Open the CGC challenge binary
KPRCA_00024/cb_1in Binary Ninja. -
Navigate to the function
listat address0x08049040. -
Locate the call instruction at address
0x080490d9, whose destination issprintfat0x0804b520. -
Inspect the LLIL immediately before the call. The relevant instructions are:
0x080490ba [esp + 0x18 {var_1044_1}].d = ebx
0x080490be [esp + 0x0c {var_1050_1}].d = esi
0x080490c2 [esp + 0x08 {var_1058+0x4}].d = edi
0x080490c6 [esp + 0x00 {var_105c_1}].d = edi
0x080490c9 [esp + 0x10 {var_104c_1}].d = edx
0x080490cd [esp + 0x14 {var_1048_1}].d = eax
0x080490d1 [esp + 0x04 {var_1058}].d = 0x804bca3
0x080490d9 call(0x804b520)
- Inspect the corresponding non-SSA MLIL call. It contains six parameters:
0x804b520(
var_105c_1,
"%s%s (Price: %d.%02d, Count: %d)\n",
var_1050_1,
var_104c_1,
var_1048_1,
var_1044_1
)
- Inspect the corresponding SSA MLIL call. It also contains six parameters:
mem#4 = 0x804b520(
var_105c_1#3,
"%s%s (Price: %d.%02d, Count: %d)\n",
var_1050_1#2,
var_104c_1#2,
var_1048_1#2,
var_1044_1#2
) @ mem#3
- Apply the following call-site type override at
0x080490d9:
int32_t(
char* dst,
const char* format,
char* prefix,
char* name,
int32_t price_integer,
int32_t price_fraction,
int32_t count
)
-
Reanalyze the function and inspect the MLIL again.
-
Observe that the call type has seven non-variadic
cdeclparameters, but both MLIL forms still contain only six parameter expressions. The store at[esp + 0x08]remains absent from the call parameters.
Expected Behavior:
MLIL call-parameter recovery should treat each ABI argument slot independently, even when adjacent slots have been grouped into a larger stack variable for another use.
For this call, MLIL should recover all seven arguments:
0x804b520(
var_105c_1,
"%s%s (Price: %d.%02d, Count: %d)\n",
&var_1010,
var_1050_1,
var_104c_1,
var_1048_1,
var_1044_1
)
Equivalently, the argument stored at [esp + 0x08 {var_1058+0x4}] should be emitted as an independent parameter between the format string and var_1050_1.
Applying a seven-parameter call-site type override should also cause the outgoing stack slots to be reconsidered so that all seven parameter expressions are generated.
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the CGC challenge binary KPRCA_00024/cb_1, function list at 0x08049040, and the sprintf call at 0x080490d9; compare the preceding LLIL stores with the non-SSA and SSA MLIL calls. Done means MLIL emits the value from [esp + 0x08] as the third argument and a seven-parameter call-site override is reflected in both MLIL forms.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp
- Domain
- compilers, reverse-engineering
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Quiet
- Clarity
- Clearly specified
- Newbie friendliness
- 48/100