Hi, Sergey,

thanks for review!

On 10.12.25, Sergey Bronnikov wrote:

From: Mike Pall <mike>

Thanks to Peter Cawley.

(cherry picked from commit d1a2fef8a8f53b0055ee041f7f63d83a27444ffa)

Stack overflow can cause a segmentation fault in vararg

Typo: s/vararg/a vararg/

function on ARM64 and MIPS64 in LJ_FR2 mode. This happen

Typo: s/happen/happens/

because stack check in BC_IFUNCV is off by one on these

Typo: s/stack/the stack/

platforms without the patch. The original stack check
for ARM64 and MIPS64 was incorrect:

| RA == BASE + (RD=NARGS)*8 + framesize * 8 >= maxstack

while stack check on x86_64 is correct and therefore is

Typo: s/stack/the stack/

not affected by the problem:

| RA == BASE + (RD=NARGS+1)*8 + framesize * 8 +8 > maxstack

The patch partially fixes aforementioned issue by

Typo: s/aforementioned/the aforementioned/

bumping LJ_STACK_EXTRA by 1 to give a space to write
the entire frame link and fixing a number of last

Typo: s/a number/the number/

I'm not get this part. I suggest rephrasing it like the following:

| The patch partially fixes the aforementioned issue by bumping
| LJ_STACK_EXTRA by 1 to give a space to the entire frame link for a
| vararg function as the __newindex metamethod.

free slot in the stack (LJ_FR2 summand adjustment).

A fixup for a number of required slots in `call_init()` was added
for consistency with non-gc64 flavor.

Typo: s/gc64/GC64/

I would also add: "The check is too strict, so this can't lead to any
crash."

Sergey Bronnikov:
* added the description and the test for the problem

Part of tarantool/tarantool#12134
---
 src/lj_def.h                                  |  2 +-
 src/lj_dispatch.c                             |  2 +-
 src/vm_arm64.dasc                             |  1 +
 src/vm_mips64.dasc                            |  1 +
 ...048-fix-stack-checks-vararg-calls.test.lua | 53 +++++++++++++++++++
 5 files changed, 57 insertions(+), 2 deletions(-)
 create mode 100644 test/tarantool-tests/lj-1048-fix-stack-checks-vararg-calls.test.lua

diff --git a/src/lj_def.h b/src/lj_def.h
index a5bca6b0..7e4f251e 100644
--- a/src/lj_def.h
+++ b/src/lj_def.h
@@ -69,7 +69,7 @@ typedef unsigned int uintptr_t;
 #define LJ_MAX_UPVAL	60		/* Max. # of upvalues. */
 
 #define LJ_MAX_IDXCHAIN	100		/* __index/__newindex chain limit. */
-#define LJ_STACK_EXTRA	(5+2*LJ_FR2)	/* Extra stack space (metamethods). */
+#define LJ_STACK_EXTRA	(5+3*LJ_FR2)	/* Extra stack space (metamethods). */
 
 #define LJ_NUM_CBPAGE	1		/* Number of FFI callback pages. */
 
diff --git a/src/lj_dispatch.c b/src/lj_dispatch.c
index a44a5adf..431cb3c2 100644
--- a/src/lj_dispatch.c
+++ b/src/lj_dispatch.c
@@ -453,7 +453,7 @@ static int call_init(lua_State *L, GCfunc *fn)
     int numparams = pt->numparams;
     int gotparams = (int)(L->top - L->base);
     int need = pt->framesize;
-    if ((pt->flags & PROTO_VARARG)) need += 1+gotparams;
+    if ((pt->flags & PROTO_VARARG)) need += 1+LJ_FR2+gotparams;
     lj_state_checkstack(L, (MSize)need);
     numparams -= gotparams;
     return numparams >= 0 ? numparams : 0;

Let's add an additional test for this part of code (since we don't have
any). It may be taken from [1]. It doesn't fail now, but we may cover
this branch more precise.

Don't get what do you mean. 

true branch in gc32 is covered by the following tests:

test/LuaJIT-tests
test/PUC-Rio-Lua-5.1-tests
test/tarantool-c-tests/lj-1087-vm-handler-call.c_test
test/tarantool-tests/fix-ff-select-recording.test.lua
test/tarantool-tests/fix-mips64-spare-side-exit-patching.test.lua
test/tarantool-tests/fix-slot-check-for-mm-record.test.lua
test/tarantool-tests/fix-slots-overflow-for-varg-record.test.lua
test/tarantool-tests/gh-6098-fix-side-exit-patching-on-arm64.test.lua
test/tarantool-tests/lj-1024-varg-maxslot.test.lua
test/tarantool-tests/lj-1025-tsetm-maxslot.test.lua
test/tarantool-tests/lj-1026-arm64-invalid-hrefk-offset-check.test.lua
test/tarantool-tests/lj-1046-fix-bc-varg-recording.test.lua
test/tarantool-tests/lj-1164-record-meta-concat-varg-pcall.test.lua
test/tarantool-tests/lj-1295-bad-renames-for-sunk-values.test.lua
test/tarantool-tests/lj-584-bad-renames-for-sunk-values.test.lua
test/tarantool-tests/lj-704-bc-varg-use-def.test.lua

diff --git a/src/vm_arm64.dasc b/src/vm_arm64.dasc
index c5f0a7a7..cf8e575a 100644
--- a/src/vm_arm64.dasc
+++ b/src/vm_arm64.dasc

<snipped>

diff --git a/src/vm_mips64.dasc b/src/vm_mips64.dasc
index da187a7a..6c2975b4 100644
--- a/src/vm_mips64.dasc
+++ b/src/vm_mips64.dasc

<snipped>

diff --git a/test/tarantool-tests/lj-1048-fix-stack-checks-vararg-calls.test.lua b/test/tarantool-tests/lj-1048-fix-stack-checks-vararg-calls.test.lua
new file mode 100644
index 00000000..d471d41e
--- /dev/null
+++ b/test/tarantool-tests/lj-1048-fix-stack-checks-vararg-calls.test.lua
@@ -0,0 +1,53 @@
+local tap = require('tap')
+
+-- A test file to demonstrate a stack overflow in `pcall()` in

I would rephrase it like "demonstrate a crash due to Lua stack
out-of-bounds access".

Updated:

--- a/test/tarantool-tests/lj-1048-fix-stack-checks-vararg-calls.test.lua
+++ b/test/tarantool-tests/lj-1048-fix-stack-checks-vararg-calls.test.lua
@@ -1,7 +1,7 @@
 local tap = require('tap')
 
--- A test file to demonstrate a stack overflow in `pcall()` in
--- some cases, see below testcase descriptions.
+-- A test file to demonstrate a crash due to Lua stack
+-- out-of-bounds access, see below testcase descriptions.
 -- See also https://github.com/LuaJIT/LuaJIT/issues/1048.
 local test = tap.test('lj-1048-fix-stack-checks-vararg-calls'):skipcond({
   ['Test requires JIT enabled'] = not jit.status(),

+-- some cases, see below testcase descriptions.
+-- See also https://github.com/LuaJIT/LuaJIT/issues/1048.
+local test = tap.test('lj-1048-fix-stack-checks-vararg-calls'):skipcond({
+  ['Test requires JIT enabled'] = not jit.status(),
+})

I suppose there is no need in JIT here. This skipcond may be removed.

+
+test:plan(2)
+
+-- The testcase demonstrate a segmentation fault due to stack

Typo: s/testcase demonstrate/test case demonstrates/

+-- overflow by recursive calling `pcall()`. The functions are
+-- vararg because stack check in BC_IFUNCV is off by one on ARM64

Typo: s/stack/the stack/

+-- and MIPS64 without the patch.
+local function prober_1(...) -- luacheck: no unused
+  -- Any fast function can be used as metamethod, but `type` is
+  -- convenient here because it works fast and can be used with
+  -- any data type. Lua function cannot be used since it
+  -- will check the stack on each invocation.

Please add the comment, that we need to check using of the correct
value LJ_STACK_EXTRA slots (5+3*LJ_FR2) = 8 for GC64 mode.

Updated:

 local function prober_1(...) -- luacheck: no unused
   -- Any fast function can be used as metamethod, but `type` is
   -- convenient here because it works fast and can be used with
   -- any data type. Lua function cannot be used since it
-  -- will check the stack on each invocation.
+  -- will check the stack on each invocation. We need to check
+  -- using of the correct value LJ_STACK_EXTRA slots
+  -- (5+3*LJ_FR2) = 8 for GC64 mode.
   pcall(pcall, pcall, pcall, pcall, pcall, pcall, pcall, pcall, type, 0)
 end
 

+  pcall(pcall, pcall, pcall, pcall, pcall, pcall, pcall, pcall, type, 0)
+end
+
+local function looper(prober, n, ...)
+  prober(...)
+  return looper(prober, n + 1, n, ...)
+end
+
+pcall(coroutine.wrap(looper), prober_1, 0)
+
+test:ok(true, 'no stack overflow with recursive pcall')
+
+-- The testcase demonstrate a segmentation fault due to stack

Typo: s/testcase demonstrate/test case demonstrates/

+-- overflow when `pcall()` is used as `__newindex` metamethod.
+-- The function is vararg because stack check in BC_IFUNCV is off

Typo: s/stack/the stack/

+-- by one on ARM64 and MIPS64 without the patch.
+
+-- Any fast function can be used as metamethod, but `type` is

Typo: s/metamethod/a metamethod/

+-- convenient here because it works fast and can be used with
+-- any data type. Lua function cannot be used since it

Typo: s/Lua/The Lua/

+-- will check the stack on each invocation.
+local t = setmetatable({}, { __newindex = pcall, __call = type })
+
+local function prober_2(...) -- luacheck: no unused
+  -- Invokes `pcall(t, t, t)`.
+  t[t] = t
+end
+
+pcall(coroutine.wrap(looper), prober_2, 0)
+
+test:ok(true, 'no stack overflow with metamethod')
+
+test:done(true)
-- 
2.43.0

[1]: https://github.com/LuaJIT/LuaJIT/issues/1402#issue-3569942423