Re: qemu:metag image runtime failure in -next due to 'kthread: allow to cancel kthread work'

[Kees Cook]

On Mon, Sep 19, 2016 at 3:57 PM, James Hogan <james.hogan@xxxxxxxxxx> wrote:
> On Mon, Sep 19, 2016 at 02:51:54PM -0700, Kees Cook wrote:
>> On Mon, Sep 19, 2016 at 2:37 PM, James Hogan <james.hogan@xxxxxxxxxx> wrote:
>> > Okay, I just built x86_64 default defconfig (on ef98de028afd, half way
>> > through the mm patches on linux-next from the other day where metag
>> > stopped booting). Perhaps I'm missing some important config option to
>> > enable the memory protection (if so I appologise).
>> >
>> > For metag:
>> >
>> > $ readelf -S drivers/tty/pty.o
>> >   [Nr] Name              Type            Addr     Off    Size   ES Flg Lk Inf Al
>> >   [51] .data..ro_after_i PROGBITS        00000000 00f0c0 00007c 00  WA  0   0  4
>> >
>> > $ readelf -S vmlinux.bust:
>> >   [Nr] Name              Type            Addr     Off    Size   ES Flg Lk Inf Al
>> >   [ 4] .rodata           PROGBITS        40190000 194000 04c9c8 00  WA  0   0 64
>> >
>> > And x86_64:
>> >
>> > $ readelf -S drivers/tty/pty.o
>> >   [Nr] Name              Type             Address           Offset
>> >        Size              EntSize          Flags  Link  Info  Align
>> >   [18] .data..ro_after_i PROGBITS         0000000000000000  00001140
>> >        00000000000000f8  0000000000000000  WA       0     0     64
>> >
>> > $ readelf -S vmlinux
>> >   [Nr] Name              Type             Address           Offset
>> >        Size              EntSize          Flags  Link  Info  Align
>> >   [ 4] .rodata           PROGBITS         ffffffff81a00000  00c00000
>> >        00000000002663d0  0000000000000000  WA       0     0     4096
>> >
>> > Both have WA on that section in the object file and the final vmlinux
>> > ELF too.
>>
>> Hm, very true, I never noticed that. Oddly, the LOAD flags don't pay
>> any attention on x86:
>>
>> $ readelf -l vmlinux
>>
>> Elf file type is EXEC (Executable file)
>> Entry point 0x1000000
>> There are 5 program headers, starting at offset 64
>>
>> Program Headers:
>>   Type           Offset             VirtAddr           PhysAddr
>>                  FileSiz            MemSiz              Flags  Align
>>   LOAD           0x0000000000200000 0xffffffff81000000 0x0000000001000000
>>                  0x0000000000fdc000 0x0000000000fdc000  R E    200000
>>   LOAD           0x0000000001200000 0xffffffff82000000 0x0000000002000000
>>                  0x0000000000155000 0x0000000000155000  RW     200000
>>   LOAD           0x0000000001400000 0x0000000000000000 0x0000000002155000
>>                  0x0000000000019488 0x0000000000019488  RW     200000
>>   LOAD           0x000000000156f000 0xffffffff8216f000 0x000000000216f000
>>                  0x0000000000122000 0x0000000000eb4000  RWE    200000
>>   NOTE           0x0000000000ca0248 0xffffffff81aa0248 0x0000000001aa0248
>>                  0x0000000000000024 0x0000000000000024         4
>>
>>  Section to Segment mapping:
>>   Segment Sections...
>>    00     .text .notes __ex_table .rodata __bug_table .pci_fixup
>> .builtin_fw .tracedata __ksymtab __ksymtab_gpl __ksymtab_strings
>> __param __modver
>>    01     .data .vvar
>>    02     .data..percpu
>>    03     .init.text .altinstr_aux .init.data .x86_cpu_dev.init
>> .altinstructions .altinstr_replacement .iommu_table .apicdrivers
>> .exit.text .smp_locks .bss .brk
>>    04     .notes
>>
>> The first load (containing .rodata) is "R E".
>
> Aah, right, I think thats because the program headers are specified
> explicitly in arch/x86/kernel/vmlinux.lds.S:
>
> PHDRS {
>         text PT_LOAD FLAGS(5);          /* R_E */
>         data PT_LOAD FLAGS(6);          /* RW_ */
> #ifdef CONFIG_X86_64
> #ifdef CONFIG_SMP
>         percpu PT_LOAD FLAGS(6);        /* RW_ */
> #endif
>         init PT_LOAD FLAGS(7);          /* RWE */
> #endif
>         note PT_NOTE FLAGS(0);          /* ___ */
> }

Ah-ha! That solves that mystery for me. :)

> The bit I was missing is that RO_DATA() is after .text, but before
> .data, so counts as part of the PT_LOAD program header for text.

Right, originally, it was so that there could be a single read-only
mapping covering both, but ultimately it doesn't matter now since they
can't share a mapping anyway: text needs to be read-only and
executable and rodata needs to be read-only and non-executable.

>> But, the point is: the kernel is what sets up the permissions, so the
>> flags are ignored anyway.
>
> Indeed.
>
> Thanks for your patience working through this stuff with me :)

No problem; I learned some stuff too. :)

-Kees

-- 
Kees Cook
Nexus Security