antirez / antirez/lua-cmsgpack
Lua stack corruption bug
- Vorherrschende Sprache
- C
- Sterne
- 383
- Forks
- 122
- PR-Merge-Kennzahlen
- Keine gemergten PRs in 30 T.
Beschreibung
On a "mp_pack()" call with a large depth of nested tables cmsgpack will crash Lua.
The function "mp_encode_lua_table()" grows the Lua stack by two per call. In a recursive situation (reaching the max allowed table depth) such as set in motion by the "test.lua" "Regression test for issue #4, cyclic references in tables.", the Lua stack space can and will usually get exhausted resulting in a crash/segfault/exception from Lua heap corruption.
It's particularly noticeable with a higher value than the default value of 16 for "LUACMSGPACK_MAX_NESTING" is set. Double that to 32 and "test.lua" will crash your Lua.
It doesn't crash currently at 16 because the default lua stack is 40 if you have a single argument or two.
```
#define LUA_MINSTACK 20
#define BASIC_STACK_SIZE (2*LUA_MINSTACK)
```
Where to catch it:
In "mp_encode_lua_type()" put a breakpoint on the "t = LUA_TNIL;" line.
When hit look at the Lua stack size and you will see it equals (LUACMSGPACK_MAX_NESTING x 2) + arg size. Where arg is the current working stack size on entry to "mp_pack()".
A solution that fixes the problem, add to the top of "mp_pack()":
`luaL_checkstack(L, (LUACMSGPACK_MAX_NESTING * 2), "Can't grow stack.");`
This will ensure enough stack space when hitting the nested table limit.
Also I'm sure you could add a similar line (like you have in "mp_decode_to_lua_type()", except growing by two) at the top of "mp_encode_lua_table()" et al.
Update:
What you can do (and what I should have had) was to have "LUA_USE_APICHECK" defined for Lua debug builds. Doing so enables an internal stack overflow check in Lua that asserts on hit.
Beitragsleitfaden
Für dieses Repository ist kein Beitragsleitfaden indexiert
Bewertung
Dieses Issue wurde noch nicht bewertet.