Re: linux-next: Tree for July 28 ("copy_from_user_overflow" undefined)

[Sedat Dilek <sedat.dilek@googlemail.com>]

[-- Attachment #1: Type: text/plain, Size: 7218 bytes --]

On Fri, Jul 29, 2011 at 5:13 PM, Randy Dunlap <rdunlap@xenotime.net> wrote:
> On Fri, 29 Jul 2011 13:07:16 +0200 Sedat Dilek wrote:
>
>> On Fri, Jul 29, 2011 at 11:45 AM, Sedat Dilek
>> <sedat.dilek@googlemail.com> wrote:
>> > On Fri, Jul 29, 2011 at 11:02 AM, Sedat Dilek
>> > <sedat.dilek@googlemail.com> wrote:
>> >> On Fri, Jul 29, 2011 at 7:09 AM, Randy Dunlap <rdunlap@xenotime.net> wrote:
>> >>> On Fri, 29 Jul 2011 03:34:26 +0200 Sedat Dilek wrote:
>> >>>
>> >>>> On Fri, Jul 29, 2011 at 1:01 AM, Randy Dunlap <rdunlap@xenotime.net> wrote:
>> >>>> > On Thu, 28 Jul 2011 16:05:31 +0200 Sedat Dilek wrote:
>> >>>> >
>> >>>> >> Hi,
>> >>>> >>
>> >>>> >> just see this build-break on the last mile:
>> >>>> >>
>> >>>> >> [...]
>> >>>> >>   OBJCOPY arch/x86/boot/compressed/vmlinux.bin
>> >>>> >>   HOSTCC  arch/x86/boot/compressed/relocs
>> >>>> >>   HOSTCC  arch/x86/boot/compressed/mkpiggy
>> >>>> >> /mnt/sdb3/linux-kernel/linux-3.0/debian/build/source_i386_none/arch/x86/boot/compressed/relocs.c:
>> >>>> >> In function 'print_absolute_symbols':
>> >>>> >> /mnt/sdb3/linux-kernel/linux-3.0/debian/build/source_i386_none/arch/x86/boot/compressed/relocs.c:405:14:
>> >>>> >> warning: variable 'sh_symtab' set but not used
>> >>>> >> [-Wunused-but-set-variable]
>> >>>> >> ERROR: "copy_from_user_overflow" [fs/binfmt_misc.ko] undefined!
>> >>>> >> make[5]: *** [__modpost] Error 1
>> >>>> >> make[4]: *** [modules] Error 2
>> >>>> >> make[4]: *** Waiting for unfinished jobs....
>> >>>> >>   RELOCS  arch/x86/boot/compressed/vmlinux.relocs
>> >>>> >>   GZIP    arch/x86/boot/compressed/vmlinux.bin.gz
>> >>>> >>   MKPIGGY arch/x86/boot/compressed/piggy.S
>> >>>> >>   AS      arch/x86/boot/compressed/piggy.o
>> >>>> >>   LD      arch/x86/boot/compressed/vmlinux
>> >>>> >>   ZOFFSET arch/x86/boot/zoffset.h
>> >>>> >>   OBJCOPY arch/x86/boot/vmlinux.bin
>> >>>> >>   AS      arch/x86/boot/header.o
>> >>>> >>   LD      arch/x86/boot/setup.elf
>> >>>> >>   OBJCOPY arch/x86/boot/setup.bin
>> >>>> >>   BUILD   arch/x86/boot/bzImage
>> >>>> >> Setup is 15596 bytes (padded to 15872 bytes).
>> >>>> >> System is 2471 kB
>> >>>> >> CRC eb598167
>> >>>> >> Kernel: arch/x86/boot/bzImage is ready  (#1)
>> >>>> >> make[3]: *** [sub-make] Error 2
>> >>>> >> make[2]: *** [all] Error 2
>> >>>> >>
>> >>>> >> I have these binfmt kernel-config options set:
>> >>>> >>
>> >>>> >> CONFIG_BINFMT_ELF=y
>> >>>> >> # CONFIG_BINFMT_AOUT is not set
>> >>>> >> CONFIG_BINFMT_MISC=m
>> >>>> >
>> >>>> > How does someone turn off this error message?
>> >>>> >
>> >>>> > I have:
>> >>>> > # CONFIG_DEBUG_STRICT_USER_COPY_CHECKS is not set
>> >>>> > # CONFIG_STRICT_DEVMEM is not set
>> >>>> >
>> >>>> > and I still cannot get this error to go away (on x86_64).
>> >>>> >
>> >>>> >
>> >>>> > ---
>> >>>> > ~Randy
>> >>>> > *** Remember to use Documentation/SubmitChecklist when testing your code ***
>> >>>> >
>> >>>>
>> >>>> Can you say with which linux-next version you saw this first?
>> >>>> My last version was next-20110722 (which was OK).
>> >>>
>> >>> It looks like 2011-0727 has lots of these errors in my randconfig builds,
>> >>> but I can't say that they are all incorrect, so it's not very conclusive.
>> >>>
>> >>> ---
>> >>> ~Randy
>> >>> *** Remember to use Documentation/SubmitChecklist when testing your code ***
>> >>>
>> >>
>> >> I can confirm next-20110726 is fine.
>> >>
>> >> # egrep 'BINFMT|DEBUG_STRICT_USER_COPY_CHECKS|STRICT_DEVMEM'
>> >> /boot/config-3.0.0-next20110726.3-686-small
>> >> CONFIG_BINFMT_ELF=y
>> >> CONFIG_BINFMT_AOUT=m
>> >> CONFIG_BINFMT_MISC=m
>> >> CONFIG_STRICT_DEVMEM=y
>> >> # CONFIG_DEBUG_STRICT_USER_COPY_CHECKS is not set
>> >>
>> >> So, the binfmt_misc culprit was introduced between next-20110726
>> >> (good) and next-20110727 (bad).
>> >> I'll try to look into the diff, but can't promise.
>> >>
>> >> - Sedat -
>> >>
>> >
>> > [ CCing all people from 0416 patch ]
>> >
>> > OK, here a quick review and looking for suspicious commits:
>> >
>> > $ git format-patch next-20110726..next-20110727
>> >
>> > 487 patches in total.
>> >
>> > $ grep copy_from_user_overflow -nr *.patch
>> >
>> > 0389-x86_64-allmodconfig.patch
>> > 0414-Enabling-DEBUG_STRICT_USER_COPY_CHECKS-causes-the-fo.patch
>> > 0415-Strict-user-copy-checks-are-only-really-supported-on.patch
>> > 0416-The-help-text-for-this-config-is-duplicated-across-t.patch
>> >
>> > 0416 has thrown out...
>> >
>> > -void copy_from_user_overflow(void)
>> > -{
>> > -       WARN(1, "Buffer overflow detected!\n");
>> > -}
>> > -EXPORT_SYMBOL(copy_from_user_overflow);
>> >
>> > ...and consilidated that in...
>> >
>> > --- /dev/null
>> > +++ b/lib/usercopy.c
>> > @@ -0,0 +1,8 @@
>> > +#include <linux/module.h>
>> > +#include <linux/bug.h>
>> > +
>> > +void copy_from_user_overflow(void)
>> > +{
>> > +       WARN(1, "Buffer overflow detected!\n");
>> > +}
>> > +EXPORT_SYMBOL(copy_from_user_overflow);
>> >
>> > ...BUT builds usercopy.o only on
>> > CONFIG_ARCH_HAS_DEBUG_STRICT_USER_COPY_CHECKS...
>> >
>> > -- a/lib/Makefile
>> > +++ b/lib/Makefile
>> > @@ -14,6 +14,7 @@ lib-y := ctype.o string.o vsprintf.o cmdline.o \
>> >         proportions.o prio_heap.o ratelimit.o show_mem.o \
>> >         is_single_threaded.o plist.o decompress.o find_next_bit.o
>> >
>> > +lib-$(CONFIG_ARCH_HAS_DEBUG_STRICT_USER_COPY_CHECKS) += usercopy.o
>> >  lib-$(CONFIG_MMU) += ioremap.o
>> >  lib-$(CONFIG_SMP) += cpumask.o
>> >
>> > So, I can rebuild next-201107{27,28} with
>> > CONFIG_ARCH_HAS_DEBUG_STRICT_USER_COPY_CHECKS=y and look again if the
>> > build-breakage in binfmt_misc is gone.
>> >
>> > Any comments from x86 folk?
>> >
>> > - Sedat -
>> >
>>
>> I was able to compile next-20110728 by reverting:
>>
>> commit ab3b49d018924085e5d949c1af53fae0da9a8d94
>> "The help text for this config is duplicated across the x86, parisc,
>> and s390 Kconfig.debug files."
>>
>> The revert-patch is attached, also my kernel-config.
>>
>> Randy, can you test with it for x86_64, please?
>> Thanks in advance.
>
> Yes, that works for me.  Thanks for your work on this problem.
>
> ---
> ~Randy
> *** Remember to use Documentation/SubmitChecklist when testing your code ***
>

Cool, you tested so fast, thanks.

While watching DebConf11 live-stream, I played a bit with trying to
fix the problem.

I added back partially copy_from_user_overflow() back to
arch/x86/lib/usercopy.c, so both x86_32 and x86_64 benefit from the
change (initially copy_from_user_overflow() was in
arch/x86/lib/usercopy_{32,64}).
The attached patch works!

I am still not understanding why there are several usercopy.c files
below arch/ dir.
The one "remained" in x86 includes only copy_from_user_nmi() and could
IMHO be renamed to usercopy_nmi.c with adaptation in
arch/x86/lib/Makefile.

x86 folks?

- Sedat -

$ find ./ -name usercopy.c
./lib/usercopy.c
./arch/mn10300/lib/usercopy.c
./arch/x86/lib/usercopy.c
./arch/cris/arch-v10/lib/usercopy.c
./arch/cris/arch-v32/lib/usercopy.c
./arch/m32r/lib/usercopy.c

[-- Attachment #2: arch_x86_lib_usercopy.c.diff --]
[-- Type: plain/text, Size: 590 bytes --]