Tarantool development patches archive
 help / color / mirror / Atom feed
From: Sergey Bronnikov via Tarantool-patches <tarantool-patches@dev.tarantool.org>
To: Sergey Kaplun <skaplun@tarantool.org>,
	Maxim Kokryashkin <m.kokryashkin@tarantool.org>
Cc: tarantool-patches@dev.tarantool.org
Subject: Re: [Tarantool-patches] [PATCH luajit] FFI: Fix various issues in recff_cdata_arith.
Date: Mon, 9 Sep 2024 18:37:36 +0300	[thread overview]
Message-ID: <5f51a9ad-3302-4e0c-8d91-c5a26c147fe2@tarantool.org> (raw)
In-Reply-To: <20240821165250.11087-1-skaplun@tarantool.org>

[-- Attachment #1: Type: text/plain, Size: 10062 bytes --]

Hi, Sergey,

thanks for the patch! see my comments below.

On 21.08.2024 19:52, Sergey Kaplun wrote:
> From: Mike Pall <mike>
>
> Thanks to Sergey Kaplun.
>
> (cherry picked from commit 7a608e4425ce0777f5c980dad9f4fdc1bcce0b8c)
>
> The aforementioned function doesn't handle gentle recording of the cdata
> addition to `nil` or some string, presuming that the interpreter will throw
> an error. This may lead to an assertion due to an uninitialized ctype
> state or an attempt to use in the fold engine the non-cdata summand (casted
> to `IR_KPTR`) as the (invalid) GC pointer.
>
> This patch handles such cases by:
> * Initializing the ctype state where it is needed.
> * Raising an error when the argument has a suspicious type. Since the
>    interpreter will throw the error anyway, these traces will abort
>    anyway.
>
> Sergey Kaplun:
> * added the description and the test for the problem
>
> Part of tarantool/tarantool#10199
> ---
>
> Branch:https://github.com/tarantool/luajit/tree/skaplun/lj-1224-fix-jit-cdata-arith
> Related issues:
> *https://github.com/tarantool/tarantool/issues/10199
> *https://github.com/LuaJIT/LuaJIT/issues/1224
>
>   src/lj_crecord.c                              | 10 ++--
>   .../lj-1224-fix-cdata-arith-ptr.test.lua      | 48 +++++++++++++++++++

This test does not fail without fix (but repro from the issue does):

[0] ~/sources/MRG/tarantool/third_party/luajit $ ./build/gc64/src/luajit 
-Ohotloop=1 -e "
repeat
   r = 1LL + nil
until true
"
LuaJIT ASSERT 
/home/sergeyb/sources/MRG/tarantool/third_party/luajit/src/lj_ctype.c:185: 
lj_ctype_intern: uninitialized cts->L
Aborted
[0] ~/sources/MRG/tarantool/third_party/luajit $ ./build/gc64/src/luajit 
test/tarantool-tests/lj-1224-fix-cdata-arith-ptr.test.lua
TAP version 13
1..2
     # cdata arithmetic with nil
     1..2
     ok - correct recording error with bad cdata arithmetic
     ok - correct error message
     # cdata arithmetic with nil: end
ok - cdata arithmetic with nil
     # cdata arithmetic with string
     1..2
     ok - correct recording error with bad cdata arithmetic
     ok - correct error message
     # cdata arithmetic with string: end
ok - cdata arithmetic with string
[0] ~/sources/MRG/tarantool/third_party/luajit $


>   .../lj-1224-fix-cdata-arith-side.test.lua     | 20 ++++++++
>   .../lj-1224-fix-cdata-arith-side/script.lua   | 16 +++++++
>   .../lj-1224-fix-cdata-arith.test.lua          | 20 ++++++++
>   .../lj-1224-fix-cdata-arith/script.lua        | 18 +++++++
>   6 files changed, 128 insertions(+), 4 deletions(-)
>   create mode 100644 test/tarantool-tests/lj-1224-fix-cdata-arith-ptr.test.lua
>   create mode 100644 test/tarantool-tests/lj-1224-fix-cdata-arith-side.test.lua
>   create mode 100644 test/tarantool-tests/lj-1224-fix-cdata-arith-side/script.lua
>   create mode 100644 test/tarantool-tests/lj-1224-fix-cdata-arith.test.lua
>   create mode 100644 test/tarantool-tests/lj-1224-fix-cdata-arith/script.lua
>
> diff --git a/src/lj_crecord.c b/src/lj_crecord.c
> index 255bfa45..81ae6dfa 100644
> --- a/src/lj_crecord.c
> +++ b/src/lj_crecord.c
> @@ -1473,7 +1473,8 @@ static TRef crec_arith_meta(jit_State *J, TRef *sp, CType **s, CTState *cts,
>   
>   void LJ_FASTCALL recff_cdata_arith(jit_State *J, RecordFFData *rd)
>   {
> -  CTState *cts = ctype_ctsG(J2G(J));
> +  CTState *cts = ctype_cts(J->L);
> +  MMS mm = (MMS)rd->data;
>     TRef sp[2];
>     CType *s[2];
>     MSize i;
> @@ -1523,6 +1524,8 @@ void LJ_FASTCALL recff_cdata_arith(jit_State *J, RecordFFData *rd)
>   	}
>         }
>       } else if (tref_isnil(tr)) {
> +      if (!(mm == MM_len || mm == MM_eq || mm == MM_lt || mm == MM_le))
> +	lj_trace_err(J, LJ_TRERR_BADTYPE);
>         tr = lj_ir_kptr(J, NULL);
>         ct = ctype_get(cts, CTID_P_VOID);
>       } else if (tref_isinteger(tr)) {
> @@ -1541,12 +1544,12 @@ void LJ_FASTCALL recff_cdata_arith(jit_State *J, RecordFFData *rd)
>   	  ct = ctype_child(cts, cct);
>   	  tr = lj_ir_kint(J, (int32_t)ofs);
>   	} else {  /* Interpreter will throw or return false. */
> -	  ct = ctype_get(cts, CTID_P_VOID);
> +	  lj_trace_err(J, LJ_TRERR_BADTYPE);
>   	}
>         } else if (ctype_isptr(ct->info)) {
>   	tr = emitir(IRT(IR_ADD, IRT_PTR), tr, lj_ir_kintp(J, sizeof(GCstr)));
>         } else {
> -	ct = ctype_get(cts, CTID_P_VOID);
> +	lj_trace_err(J, LJ_TRERR_BADTYPE);
>         }
>       } else if (!tref_isnum(tr)) {
>         tr = 0;
> @@ -1558,7 +1561,6 @@ void LJ_FASTCALL recff_cdata_arith(jit_State *J, RecordFFData *rd)
>     }
>     {
>       TRef tr;
> -    MMS mm = (MMS)rd->data;
>       if ((mm == MM_len || mm == MM_concat ||
>   	 (!(tr = crec_arith_int64(J, sp, s, mm)) &&
>   	  !(tr = crec_arith_ptr(J, sp, s, mm)))) &&
> diff --git a/test/tarantool-tests/lj-1224-fix-cdata-arith-ptr.test.lua b/test/tarantool-tests/lj-1224-fix-cdata-arith-ptr.test.lua
> new file mode 100644
> index 00000000..9de9586d
> --- /dev/null
> +++ b/test/tarantool-tests/lj-1224-fix-cdata-arith-ptr.test.lua
> @@ -0,0 +1,48 @@
> +local tap = require('tap')
> +
> +-- Test file to demonstrate LuaJIT's incorrect recording of cdata
> +-- arithmetic looks like pointer arithmetic, which raises the
> +-- error.
> +-- See also:https://github.com/LuaJIT/LuaJIT/issues/1224.
> +
> +local test = tap.test('lj-1224-fix-cdata-arith-ptr'):skipcond({
> +  ['Test requires JIT enabled'] = not jit.status(),
> +})
> +
> +test:plan(2)
> +
> +local function arith_nil_protected()
> +  local i = 1
> +  while i < 3 do
> +    -- Before the patch, `nil` was "cast" to `IR_KPTR(NULL)`
> +    -- during the recording of the cdata arith metamethod, and the
> +    -- fold engine tried to add this value to the `-1LL`. This
> +    -- obviously leads to the value being out-of-range for the GC
> +    -- pointer.
> +    local _ = -1LL + nil
> +    i = i + 1
> +  end
> +end
> +
> +-- The similar test but for the string value instead of nil.
> +local function arith_str_protected()
> +  local i = 1
> +  while i < 3 do
> +    local _ = 1LL + ''
> +    i = i + 1
> +  end
> +end
> +
> +local function check_error(subtest, test_f)
> +  subtest:plan(2)
> +  -- Reset counters.
> +  jit.opt.start('hotloop=1')
> +  local result, errmsg = pcall(test_f)
> +  subtest:ok(not result, 'correct recording error with bad cdata arithmetic')
> +  subtest:like(errmsg, 'attempt to perform arithmetic', 'correct error message')
> +end
> +
> +test:test('cdata arithmetic with nil',    check_error, arith_nil_protected)
> +test:test('cdata arithmetic with string', check_error, arith_str_protected)
> +
> +test:done(true)
> diff --git a/test/tarantool-tests/lj-1224-fix-cdata-arith-side.test.lua b/test/tarantool-tests/lj-1224-fix-cdata-arith-side.test.lua
> new file mode 100644
> index 00000000..3f3e6011
> --- /dev/null
> +++ b/test/tarantool-tests/lj-1224-fix-cdata-arith-side.test.lua
> @@ -0,0 +1,20 @@
> +local tap = require('tap')
> +
> +-- Test file to demonstrate LuaJIT's incorrect recording for the
> +-- side trace with cdata arithmetic, which raises the error.
> +-- See also:https://github.com/LuaJIT/LuaJIT/issues/1224.
> +
> +local test = tap.test('lj-1224-fix-cdata-arith-side'):skipcond({
> +  ['Test requires JIT enabled'] = not jit.status(),
> +})
> +
> +-- The loading of the 'tap' module initializes `cts->L` during
> +-- parsing. Run standalone script for testing.
> +local script = require('utils').exec.makecmd(arg)
> +
> +test:plan(1)
> +
> +local output = script()
> +test:is(output, 'OK', 'correct recording with uninitialized cts->L')
> +
> +test:done(true)
> diff --git a/test/tarantool-tests/lj-1224-fix-cdata-arith-side/script.lua b/test/tarantool-tests/lj-1224-fix-cdata-arith-side/script.lua
> new file mode 100644
> index 00000000..e96877ff
> --- /dev/null
> +++ b/test/tarantool-tests/lj-1224-fix-cdata-arith-side/script.lua
> @@ -0,0 +1,16 @@
> +local function test_record_protected()
> +  -- Start the root trace here.
> +  for _ = 1, 3 do end
> +  -- An exit from the root trace triggered the side trace
> +  -- recording.
> +  local _ = 1LL + ''
> +end
> +
> +jit.opt.start('hotloop=1', 'hotexit=1')
> +
> +local status, errmsg = pcall(test_record_protected)
> +
> +assert(not status, 'recoding correctly handling the error')
> +assert(errmsg:match('attempt to perform arithmetic'), 'correct error message')
> +
> +print('OK')
> diff --git a/test/tarantool-tests/lj-1224-fix-cdata-arith.test.lua b/test/tarantool-tests/lj-1224-fix-cdata-arith.test.lua
> new file mode 100644
> index 00000000..ba89b472
> --- /dev/null
> +++ b/test/tarantool-tests/lj-1224-fix-cdata-arith.test.lua
> @@ -0,0 +1,20 @@
> +local tap = require('tap')
> +
> +-- Test file to demonstrate LuaJIT's incorrect recording of cdata
> +-- arithmetic, which raises the error.
> +-- See also:https://github.com/LuaJIT/LuaJIT/issues/1224.
> +
> +local test = tap.test('lj-1224-fix-cdata-arith'):skipcond({
> +  ['Test requires JIT enabled'] = not jit.status(),
> +})
> +
> +-- The loading of the 'tap' module initializes `cts->L` during
> +-- parsing. Run standalone script for testing.
> +local script = require('utils').exec.makecmd(arg)
> +
> +test:plan(1)
> +
> +local output = script()
> +test:is(output, 'OK', 'correct recording with uninitialized cts->L')
> +
> +test:done(true)
> diff --git a/test/tarantool-tests/lj-1224-fix-cdata-arith/script.lua b/test/tarantool-tests/lj-1224-fix-cdata-arith/script.lua
> new file mode 100644
> index 00000000..df79e430
> --- /dev/null
> +++ b/test/tarantool-tests/lj-1224-fix-cdata-arith/script.lua
> @@ -0,0 +1,18 @@
> +local function test_record_protected()
> +  local i = 1
> +  while i < 3 do
> +    -- Use `while` to start compilation of the trace at the first
> +    -- iteration, before `cts->L` is uninitialized.
> +    local _ = 1LL + nil
> +    i = i + 1
> +  end
> +end
> +
> +jit.opt.start('hotloop=1')
> +
> +local status, errmsg = pcall(test_record_protected)
> +
> +assert(not status, 'recoding correctly handling the error')
> +assert(errmsg:match('attempt to perform arithmetic'), 'correct error message')
> +
> +print('OK')

[-- Attachment #2: Type: text/html, Size: 12523 bytes --]

  reply	other threads:[~2024-09-09 15:37 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-08-21 16:52 Sergey Kaplun via Tarantool-patches
2024-09-09 15:37 ` Sergey Bronnikov via Tarantool-patches [this message]
2024-09-10 10:04   ` Sergey Bronnikov via Tarantool-patches
2024-09-10 14:01     ` Sergey Kaplun via Tarantool-patches
2024-09-12 10:16       ` Sergey Kaplun via Tarantool-patches
2024-09-12 11:45         ` Sergey Bronnikov via Tarantool-patches

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=5f51a9ad-3302-4e0c-8d91-c5a26c147fe2@tarantool.org \
    --to=tarantool-patches@dev.tarantool.org \
    --cc=m.kokryashkin@tarantool.org \
    --cc=sergeyb@tarantool.org \
    --cc=skaplun@tarantool.org \
    --subject='Re: [Tarantool-patches] [PATCH luajit] FFI: Fix various issues in recff_cdata_arith.' \
    /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