From: Sergey Kaplun via Tarantool-patches <tarantool-patches@dev.tarantool.org>
To: Sergey Bronnikov <estetus@gmail.com>
Cc: tarantool-patches@dev.tarantool.org
Subject: Re: [Tarantool-patches] [PATCH luajit 3/3][v3] Add stack check to pcall/xpcall.
Date: Thu, 12 Mar 2026 13:16:48 +0300 [thread overview]
Message-ID: <abKSkAwu0v9QGt35@root> (raw)
In-Reply-To: <f09ea0fa305bcf5607e100d4771ba5d52ef2d05f.1773300611.git.sergeyb@tarantool.org>
Hi, Sergey!
Thanks for the patch!
Please, fix my comments below.
Don't forget to add the corresponding iterative changes.
On 12.03.26, Sergey Bronnikov wrote:
> From: Mike Pall <mike>
>
> Analyzed by Peter Cawley.
>
> (cherry picked from commit a4c1640432a9d8a60624cdc8065b15078c228e36)
>
> The patch adds the stack check to fast functions `pcall()` and
> `xpcall()`.
Please add more verbose description:
| (cherry picked from commit a4c1640432a9d8a60624cdc8065b15078c228e36)
|
| The `pcall()` and `xpcall()` calls in GC64 mode require 2 slots. This
| means that all arguments should be moved up during emitting of the frame
| link to the stack. Hence, this may cause stack overflow without the
| corresponding check.
|
| This patch adds the corresponding checks to the VM. Non-GC64 VMs are
| updated as well for the consistency.
>
> Sergey Bronnikov:
> * added the description and the test for the problem
>
> Part of tarantool/tarantool#12134
> ---
> src/vm_arm.dasc | 7 ++++
> src/vm_arm64.dasc | 8 +++++
> src/vm_mips.dasc | 10 +++++-
> src/vm_mips64.dasc | 14 ++++++--
> src/vm_ppc.dasc | 9 +++++
> src/vm_x64.dasc | 6 ++++
> src/vm_x86.dasc | 6 ++++
> ...048-fix-stack-checks-vararg-calls.test.lua | 35 ++++++++++++++++++-
> 8 files changed, 90 insertions(+), 5 deletions(-)
>
> diff --git a/src/vm_arm.dasc b/src/vm_arm.dasc
> index 7095e660..efe9dcb2 100644
> --- a/src/vm_arm.dasc
> +++ b/src/vm_arm.dasc
<snipped>
> diff --git a/src/vm_arm64.dasc b/src/vm_arm64.dasc
> index 5ef37243..074c1f31 100644
> --- a/src/vm_arm64.dasc
> +++ b/src/vm_arm64.dasc
<snipped>
> diff --git a/src/vm_mips.dasc b/src/vm_mips.dasc
> index 32caabf7..69d09d52 100644
> --- a/src/vm_mips.dasc
> +++ b/src/vm_mips.dasc
<snipped>
> diff --git a/src/vm_mips64.dasc b/src/vm_mips64.dasc
> index 6c2975b4..4e60ee07 100644
> --- a/src/vm_mips64.dasc
> +++ b/src/vm_mips64.dasc
> @@ -1418,8 +1418,12 @@ static void build_subroutines(BuildCtx *ctx)
> |//-- Base library: catch errors ----------------------------------------
> |
> |.ffunc pcall
> + | ld TMP1, L->maxstack
> + | daddu TMP2, BASE, NARGS8:RC
> + | sltu AT, TMP1, TMP2
> + | bnez AT, ->fff_fallback
> + |. lbu TMP3, DISPATCH_GL(hookmask)(DISPATCH)
> | daddiu NARGS8:RC, NARGS8:RC, -8
> - | lbu TMP3, DISPATCH_GL(hookmask)(DISPATCH)
> | bltz NARGS8:RC, ->fff_fallback
> |. move TMP2, BASE
> | daddiu BASE, BASE, 16
> @@ -1440,8 +1444,12 @@ static void build_subroutines(BuildCtx *ctx)
> |. nop
> |
> |.ffunc xpcall
> - | daddiu NARGS8:TMP0, NARGS8:RC, -16
This neglets the first patch in the series. See the comment below.
> - | ld CARG1, 0(BASE)
> + | ld TMP1, L->maxstack
> + | daddu TMP2, BASE, NARGS8:RC
> + | sltu AT, TMP1, TMP2
> + | bnez AT, ->fff_fallback
> + |. ld CARG1, 0(BASE)
> + | daddiu NARGS8:RC, NARGS8:RC, -16
This line is incorrect. This neglets the 1st patch in the series.
It should be
| | daddiu NARGS8:TMP0, NARGS8:RC, -16
> | ld CARG2, 8(BASE)
> | bltz NARGS8:TMP0, ->fff_fallback
> |. lbu TMP1, DISPATCH_GL(hookmask)(DISPATCH)
> diff --git a/src/vm_ppc.dasc b/src/vm_ppc.dasc
> index 980ad897..f2ea933b 100644
> --- a/src/vm_ppc.dasc
> +++ b/src/vm_ppc.dasc
<snipped>
> diff --git a/src/vm_x64.dasc b/src/vm_x64.dasc
> index 8b6781a6..c57b76b7 100644
> --- a/src/vm_x64.dasc
> +++ b/src/vm_x64.dasc
<snipped>
> diff --git a/src/vm_x86.dasc b/src/vm_x86.dasc
> index 7c11c78e..36804d11 100644
> --- a/src/vm_x86.dasc
> +++ b/src/vm_x86.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
> index 3a8ad63d..ad8b151b 100644
> --- 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
> @@ -5,7 +5,7 @@ local tap = require('tap')
> -- See also https://github.com/LuaJIT/LuaJIT/issues/1048.
> local test = tap.test('lj-1048-fix-stack-checks-vararg-calls')
>
> -test:plan(2)
> +test:plan(5)
>
> -- The test case demonstrates a segmentation fault due to stack
> -- overflow by recursive calling `pcall()`. The functions are
> @@ -50,4 +50,37 @@ pcall(coroutine.wrap(looper), prober_2, 0)
>
> test:ok(true, 'no stack overflow with metamethod')
>
> +-- The testcases demonstrates a stack overflow in
> +-- `pcall()`/xpcall()` triggered using metamethod `__call`.
> +
> +t = coroutine.wrap(setmetatable)({}, { __call = pcall })
I've meant the following:
| t = setmetatable({}, { __call = pcall })
| coroutine.wrap(function() t() end)()
> +
> +test:ok(true, 'no stack overflow with metamethod __call with pcall()')
> +
> +t = coroutine.wrap(setmetatable)({}, { __call = xpcall })
I've meant the following:
| t = setmetatable({}, { __call = xpcall })
| coroutine.wrap(function() t() end)()
But this won't work since the second amount of xpcall must be the
function. So, this test case is invalid. We must to duplicate the second
approach with `xpcall()`
This works fine.
| LUA_PATH="src/?.lua;;" gdb --args src/luajit -e '
| local t = {}
| local function xpcall_wrapper()
| return xpcall(unpack(t))
| end
|
| local N_ITERATIONS = 200
|
| for i = 1, N_ITERATIONS do
| t[i], t[i + 1], t[i + 2] = xpcall, type, {}
| coroutine.wrap(xpcall_wrapper)()
| end
| '
> +
> +test:ok(true, 'no stack overflow with metamethod __call with xpcall()')
> +
> +-- The testcase demonstrates a stack overflow in
> +-- `pcall()`/`xpcall()` similar to the first testcase, but it is
> +-- triggered using `unpack()`.
> +
> +t = {}
> +local function f()
Lets name it `pcall_wrapper()`
> + return pcall(unpack(t))
> +end
> +
> +-- The problem is only reproduced on LuaJIT GC64 and is best
Typo: s/best/better/
> +-- reproduced under Valgrind than AddressSanitizer. The chosen
> +-- value was found experimentally and always results in an attempt
> +-- to write beyond the allocated memory.
> +local N_ITERATIONS = 200
> +
> +for i = 1, N_ITERATIONS do
> + t[i], t[i + 1], t[i + 2] = pcall, type, {}
> + coroutine.wrap(f)()
> +end
> +
> +test:ok(true, 'no stack overflow with unpacked pcalls')
> +
> test:done(true)
> --
> 2.43.0
>
--
Best regards,
Sergey Kaplun
prev parent reply other threads:[~2026-03-12 10:15 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-03-12 9:05 [Tarantool-patches] [PATCH luajit 0/3][v3] Fix stack overflow in pcall/xpcall Sergey Bronnikov via Tarantool-patches
2026-03-12 8:49 ` [Tarantool-patches] [PATCH luajit 1/3][v3] MIPS64: Fix xpcall() error case Sergey Bronnikov via Tarantool-patches
2026-03-12 8:49 ` [Tarantool-patches] [PATCH luajit 2/3][v3] LJ_FR2: Fix stack checks in vararg calls Sergey Bronnikov via Tarantool-patches
2026-03-12 9:36 ` Sergey Kaplun via Tarantool-patches
2026-03-12 12:25 ` Sergey Bronnikov via Tarantool-patches
2026-03-12 12:47 ` Sergey Kaplun via Tarantool-patches
2026-03-12 8:49 ` [Tarantool-patches] [PATCH luajit 3/3][v3] Add stack check to pcall/xpcall Sergey Bronnikov via Tarantool-patches
2026-03-12 10:16 ` Sergey Kaplun via Tarantool-patches [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=abKSkAwu0v9QGt35@root \
--to=tarantool-patches@dev.tarantool.org \
--cc=estetus@gmail.com \
--cc=skaplun@tarantool.org \
--subject='Re: [Tarantool-patches] [PATCH luajit 3/3][v3] Add stack check to pcall/xpcall.' \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox