Re: 0887a7ebc9 ("ubsan: add trap instrumentation option"): BUG: kernel hang in early-boot stage, last printk: early console in setup code

[Kees Cook <keescook@chromium.org>]

On Mon, Jun 08, 2020 at 02:04:08PM +0800, kernel test robot wrote:
> The issue seems due to the lack of "-fsanitize-undefined-trap-on-error" in clang.

Hm? No, that's supported in Clang (at least as far back as Clang 9.)

> Greetings,
> 
> 0day kernel testing robot got the below dmesg and the first bad commit is
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
> 
> commit 0887a7ebc97770c7870abf3075a2e8cd502a7f52
> Author:     Kees Cook <keescook@chromium.org>
> AuthorDate: Mon Apr 6 20:12:27 2020 -0700
> Commit:     Linus Torvalds <torvalds@linux-foundation.org>
> CommitDate: Tue Apr 7 10:43:44 2020 -0700
> 
>     ubsan: add trap instrumentation option

In the randconfig, I see CONFIG_UBSAN_TRAP is enabled with lots of other
UBSAN options. If you're not expecting the results, it's very likely the
false positives in UBSAN are going to do bad things. :) This is "working
as expected", as noted in the commit log quoted below.

>     
>     Patch series "ubsan: Split out bounds checker", v5.
>     
>     This splits out the bounds checker so it can be individually used.  This
>     is enabled in Android and hopefully for syzbot.  Includes LKDTM tests for
>     behavioral corner-cases (beyond just the bounds checker), and adjusts
>     ubsan and kasan slightly for correct panic handling.
>     
>     This patch (of 6):
>     
>     The Undefined Behavior Sanitizer can operate in two modes: warning
>     reporting mode via lib/ubsan.c handler calls, or trap mode, which uses
>     __builtin_trap() as the handler.  Using lib/ubsan.c means the kernel image
>     is about 5% larger (due to all the debugging text and reporting structures
>     to capture details about the warning conditions).  Using the trap mode,
>     the image size changes are much smaller, though at the loss of the
>     "warning only" mode.
>     
>     In order to give greater flexibility to system builders that want minimal
>     changes to image size and are prepared to deal with kernel code being
>     aborted and potentially destabilizing the system, this introduces
>     CONFIG_UBSAN_TRAP.  The resulting image sizes comparison:
>     
>        text    data     bss       dec       hex     filename
>     19533663   6183037  18554956  44271656  2a38828 vmlinux.stock
>     19991849   7618513  18874448  46484810  2c54d4a vmlinux.ubsan
>     19712181   6284181  18366540  44362902  2a4ec96 vmlinux.ubsan-trap
>     
>     CONFIG_UBSAN=y:      image +4.8% (text +2.3%, data +18.9%)
>     CONFIG_UBSAN_TRAP=y: image +0.2% (text +0.9%, data +1.6%)
>     
>     Additionally adjusts the CONFIG_UBSAN Kconfig help for clarity and removes
>     the mention of non-existing boot param "ubsan_handle".

If you're trying to _boot_ a randconfig, I suspect there are going to be
a lot of surprises with UBSAN (in any mode) enabled. Right now, likely the
least noisy of them all is UBSAN_BOUNDS, which was split out for fuzzers.

FWIW, the dmesg appears to be catching a NULL pointer dereference
(enabled via CONFIG_UBSAN_MISC):

[    0.047646] UBSAN: Undefined behaviour in drivers/acpi/acpica/tbfadt.c:459:37
[    0.047650] member access within null pointer of type 'struct acpi_table_fadt'
[    0.047655] CPU: 0 PID: 0 Comm: swapper Not tainted 5.6.0-11597-g7baf219982281 #1
[    0.047659] Call Trace:
[    0.047676]  dump_stack+0x88/0xb9
[    0.047684]  ? ubsan_prologue+0x21/0x46
[    0.047689]  ? ubsan_type_mismatch_common+0x188/0x19e
[    0.047695]  ? __ubsan_handle_type_mismatch_v1+0x45/0x4a
[    0.047701]  ? acpi_tb_create_local_fadt+0xaa/0x435
[    0.047706]  ? acpi_tb_parse_fadt+0x54/0xd4
[    0.047712]  ? acpi_tb_parse_root_table+0x192/0x1bf
[    0.047717]  ? acpi_table_init+0x3b/0x56
[    0.047721]  ? acpi_boot_table_init+0xf/0x6e
[    0.047726]  ? setup_arch+0x459/0x520
[    0.047732]  ? start_kernel+0x5e/0x3ba
[    0.047737]  ? secondary_startup_64+0xa4/0xb0

I'm not sure how ACPI defines acpi_gbl_FADT though? There's no
dereference...

459:         if (acpi_gbl_FADT.header.length <= ACPI_FADT_V2_SIZE) {


BTW, this report only contained 1 actual dmesg. There were two files with
dmesg file names, but one of them was the gzipped reproduction steps again.

-Kees

-- 
Kees Cook