Re: [PATCH 3/7] contrib/elf2dmp: Ensure segment fits in file

[Akihiko Odaki <akihiko.odaki@daynix.com>]

On 2024/03/05 2:52, Peter Maydell wrote:
> On Sun, 3 Mar 2024 at 10:53, Akihiko Odaki <akihiko.odaki@daynix.com> wrote:
>>
>> This makes elf2dmp more robust against corrupted inputs.
>>
>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
>> ---
>>   contrib/elf2dmp/addrspace.c | 5 +++--
>>   1 file changed, 3 insertions(+), 2 deletions(-)
>>
>> diff --git a/contrib/elf2dmp/addrspace.c b/contrib/elf2dmp/addrspace.c
>> index 980a7aa5f8fb..d546a400dfda 100644
>> --- a/contrib/elf2dmp/addrspace.c
>> +++ b/contrib/elf2dmp/addrspace.c
>> @@ -88,11 +88,12 @@ int pa_space_create(struct pa_space *ps, QEMU_Elf *qemu_elf)
>>       ps->block = g_new(struct pa_block, ps->block_nr);
>>
>>       for (i = 0; i < phdr_nr; i++) {
>> -        if (phdr[i].p_type == PT_LOAD) {
>> +        if (phdr[i].p_type == PT_LOAD && phdr[i].p_offset < qemu_elf->size) {
>>               ps->block[block_i] = (struct pa_block) {
>>                   .addr = (uint8_t *)qemu_elf->map + phdr[i].p_offset,
>>                   .paddr = phdr[i].p_paddr,
>> -                .size = phdr[i].p_filesz,
>> +                .size = MIN(phdr[i].p_filesz,
>> +                            qemu_elf->size - phdr[i].p_offset),
> 
> Shouldn't "p_filesz is smaller than the actual amount of data in the
> file" be a failure condition? In include/hw/elf_ops.h we treat it
> that way:
> 
>              mem_size = ph->p_memsz; /* Size of the ROM */
>              file_size = ph->p_filesz; /* Size of the allocated data */
>              data_offset = ph->p_offset; /* Offset where the data is located */
> 
>              if (file_size > 0) {
>                  if (g_mapped_file_get_length(mapped_file) <
>                      file_size + data_offset) {
>                      goto fail;
>                  }
>                  [etc]
> 
> Like that code, we could then only check if p_offset + p_filesz is off
> the end of the file, rather than checking p_offset separately.
> 
>>               };
>>               pa_block_align(&ps->block[block_i]);
>>               block_i = ps->block[block_i].size ? (block_i + 1) : block_i;
> 
> thanks
> -- PMM

I'm making this permissive for corrupted dumps since they may still 
include valuable information.

It is different from include/hw/elf_ops.h, which is presumably used to 
load executables rather than dumps. Loading a corrupted executable does 
nothing good.

Regards,
Akihiko Odaki