Rev Symbol

rev-symbol - Symbol Recovery

Analyze function code characteristics to recover/identify function symbols and names.

Adapted from P4nda0s/reverse-skills (MIT) for the Ghidra MCP toolchain (upstream targets IDA; the analysis methodology is unchanged).

Data Sources

The upstream skill reads an IDA-NO-MCP export directory (decompile/*.c with callers: / callees: headers, strings.txt, imports.txt, exports.txt). The equivalent sources here — repo scripts live under frida-mobile-security/scripts/utils/:

The offline tools need no Ghidra instance and are deterministic — use them for the static tables; use Ghidra MCP for pseudocode, xrefs and renaming. Single-function recovery should query on demand; build a full export only when the whole .so is in scope.

Pre-check

Ghidra MCP (ghidra_* tools):

1. Target .so not yet imported → adb pull it from the device, then ghidra_import_file (ELF auto-detects the language; explicit IDs: ARM64 AARCH64:LE:64:v8A, ARM32 ARM:LE:32:v7).
2. Query the loaded instance: decompile by address/name, get callers/callees, list strings/imports/exports, rename functions, apply comments.
3. Exact tool names depend on the loaded tool groups — list or search them with ghidra_list_tool_groups / ghidra_search_tools when unsure.

If no Ghidra instance is connected, check ghidra_list_instances and connect with ghidra_connect_instance, or ask the user to start one.

Symbol Recovery Steps

Step 1: Analyze Internal Characteristics

Carefully examine the target function for:

• String constants: Strings used in the function may reveal its purpose
• Numeric constants / Magic Numbers:
  • MD5: 0x67452301, 0xEFCDAB89, 0x98BADCFE, 0x10325476
  • CRC32: 0xEDB88320
  • Base64 charset: ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/
  • AES S-Box: 0x63, 0x7C, 0x77, 0x7B...
  • Zlib: 0x78, 0x9C (compression header)
  • other constants/magic numbers...
• Code structure: Loop patterns, bitwise operations, specific algorithm flows

If you can identify a known algorithm through constants/structure, tell the user directly.

Step 2: Analyze Cross-References

Analyze Callees (called functions):

• Get the callee list with ghidra_get_function_callees, or read the calls from the pseudocode (ghidra_decompile_function).
• Classify each callee against the import table — ghidra_list_imports / ghidra_list_external_locations, or elfinfo.py offline. A callee that resolves to an import (external location, thunk_*, undefined in .dynsym) is a library call; name it from the import symbol.
• Recognize call patterns even when symbols are missing:

Paired function patterns (identify by matching call pairs):

// malloc/free, new/delete, alloc/dealloc
xx = sub_A(0x100);        // alloc: takes size, returns pointer
...
sub_B(xx);                // free: takes the same pointer

// mutex_lock/mutex_unlock, pthread_mutex_lock/unlock
sub_A(lock_ptr);          // lock
...                       // critical section
sub_B(lock_ptr);          // unlock (same lock object)

// open/close, fopen/fclose, CreateFile/CloseHandle
fd = sub_A("/path", 0);   // open: path + flags, returns handle
...
sub_B(fd);                // close: takes the handle

// pthread_create/pthread_join
sub_A(&tid, 0, func, arg); // create: out param, attr, func, arg
...
sub_B(tid, &ret);          // join: tid, out param

Argument pattern recognition:

// socket(AF_INET, SOCK_STREAM, 0) - fixed constants
sub_XXX(2, 1, 0);         // socket: domain=2, type=1, protocol=0

// connect/bind(sockfd, addr, addrlen)
sub_XXX(fd, &var, 16);   // addr struct, len=16 for IPv4

// memcpy/memmove(dst, src, size)
sub_XXX(dst, src, n);     // 3 params: dst, src, count

// memset(ptr, value, size)
sub_XXX(ptr, 0, 0x100);   // 3 params: ptr, byte value, count

// read/write(fd, buf, count)
ret = sub_XXX(fd, buf, n); // returns bytes read/written

// strcmp/strncmp(s1, s2) or (s1, s2, n)
if (sub_XXX(s1, s2) == 0)  // returns 0 on equal

Return value patterns:

// file/socket operations: -1 on error
if ((fd = sub_XXX(...)) == -1) goto error;

// allocation: NULL on failure
if (!(ptr = sub_XXX(size))) goto error;

// success/error: 0 = success
if (sub_XXX(...) != 0) goto error;

// strlen: returns size_t
len = sub_XXX(str);
sub_YYY(dst, src, len);   // len used in memcpy

Analyze Callers (calling functions):

• Get the caller list with ghidra_get_function_callers, or find_branch_callers.py offline.
• A caller counts as named only when its symbol is not auto-generated (FUN_ / sub_ / thunk_ / LAB_ / j_ prefix) — i.e. it comes from the export table or has already been renamed.
• Bounded recursion: if the direct caller is unnamed, walk up its callers; stop at the first named caller or at depth 3. If nothing is named, fall back to internal characteristics (Step 1) and web search (Step 3).
• Analyze how the return value is used by callers.

Step 3: Information Gathering and Search

Collect the following information:

• Strings referenced by the function (ghidra_get_xrefs_to on the string address, or find_strref.py offline)
• Magic Numbers / constants
• Known imports in the call chain (ghidra_list_imports, or elfinfo.py offline)
• Caller/callee symbol names (ghidra_list_exports, elfinfo.py exports, already-renamed functions)
• Paired function patterns identified

Based on collected information:

1. First attempt local reasoning based on:

   • Function signature (number and types of parameters)
   • Paired call patterns (alloc/free, lock/unlock)
   • Known imports in the call chain
   • Code structure similarity to known algorithms

2. If uncertain, use Web Search to search:

   • Search Magic Numbers: 0x67452301 0xEFCDAB89 algorithm
   • Search code patterns: rotate left xor constant algorithm
   • Search unique strings found in the function
   • Search parameter patterns: function(int, int, 0) socket

Output Format

## Symbol Recovery Analysis: <function_address>

### Function Characteristics
- Strings: <list discovered strings>
- Constants: <list key constants>
- Called imports: <list>

### Cross-Reference Analysis
- Callers: <callers and their symbols>
- Callees: <callees and their symbols>

### Inference Result
- **Suggested symbol name**: <suggested_name>
- **Confidence**: High / Medium / Low
- **Reasoning**: <explain why this name is suggested>

### Similar Open Source Implementation
- <if similar open source code is found, provide link>