Re: [PATCH v1] sd: sdhci: assert data_count is within fifo_buffer

[Philippe Mathieu-Daudé]

On 9/3/20 4:01 PM, Philippe Mathieu-Daudé wrote:
> On 9/3/20 9:08 AM, P J P wrote:
>> From: Prasad J Pandit <pjp@fedoraproject.org>
>>
>> While doing multi block SDMA, transfer block size may exceed
>> the 's->fifo_buffer[s->buf_maxsz]' size. It may leave the
>> current element pointer 's->data_count' pointing out of bounds.
>> Leading the subsequent DMA r/w operation to OOB access issue.
>> Assert that 's->data_count' is within fifo_buffer.
>>
>>  -> 
>> https://ruhr-uni-bochum.sciebo.de/s/NNWP2GfwzYKeKwE?path=%2Fsdhci_oob_write1
>>  ==1459837==ERROR: AddressSanitizer: heap-buffer-overflow
>>  WRITE of size 54722048 at 0x61500001e280 thread T3
>>  #0  __interceptor_memcpy (/lib64/libasan.so.6+0x3a71d)
>>  #1  flatview_read_continue ../exec.c:3245
>>  #2  flatview_read ../exec.c:3278
>>  #3  address_space_read_full ../exec.c:3291
>>  #4  address_space_rw ../exec.c:3319
>>  #5  dma_memory_rw_relaxed ../include/sysemu/dma.h:87
>>  #6  dma_memory_rw ../include/sysemu/dma.h:110
>>  #7  dma_memory_read ../include/sysemu/dma.h:116
>>  #8  sdhci_sdma_transfer_multi_blocks ../hw/sd/sdhci.c:629
>>  #9  sdhci_write ../hw/sd/sdhci.c:1097
>>  #10 memory_region_write_accessor ../softmmu/memory.c:483
>>  ...
>>
>> Reported-by: Ruhr-University <bugs-syssec@rub.de>
>> Suggested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
>> Signed-off-by: Prasad J Pandit <pjp@fedoraproject.org>
>> ---
>>  hw/sd/sdhci.c | 2 ++
>>  1 file changed, 2 insertions(+)
>>
>> Update v1: use assert(3) calls
>>   -> https://lists.nongnu.org/archive/html/qemu-devel/2020-09/msg00966.html
>>
>> diff --git a/hw/sd/sdhci.c b/hw/sd/sdhci.c
>> index 1785d7e1f7..023acbed41 100644
>> --- a/hw/sd/sdhci.c
>> +++ b/hw/sd/sdhci.c
>> @@ -604,6 +604,7 @@ static void sdhci_sdma_transfer_multi_blocks(SDHCIState 
>> *s)
>>                      s->blkcnt--;
>>                  }
>>              }
>> +            assert(s->data_count <= s->buf_maxsz && s->data_count > begin);
>>              dma_memory_write(s->dma_as, s->sdmasysad,
>>                               &s->fifo_buffer[begin], s->data_count - begin);
>>              s->sdmasysad += s->data_count - begin;
>> @@ -626,6 +627,7 @@ static void sdhci_sdma_transfer_multi_blocks(SDHCIState 
>> *s)
>>                  s->data_count = block_size;
>>                  boundary_count -= block_size - begin;
>>              }
>> +            assert(s->data_count <= s->buf_maxsz && s->data_count > begin);
>>              dma_memory_read(s->dma_as, s->sdmasysad,
>>                              &s->fifo_buffer[begin], s->data_count - begin);
>>              s->sdmasysad += s->data_count - begin;
>>
> 
> qemu-system-i386: hw/sd/sdhci.c:632: void
> sdhci_sdma_transfer_multi_blocks(SDHCIState *): Assertion `s->data_count
> <= s->buf_maxsz && s->data_count > begin' failed.
> 
> Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
> 
> If you don't mind I might split the assert in 2 when applying:
> 
> -            assert(s->data_count <= s->buf_maxsz && s->data_count > begin);
> +            assert(s->data_count <= s->buf_maxsz);
> +            assert(s->data_count > begin);

Doesn't seem enough, guest crash here, having:

(gdb) p begin
$1 = 0
(gdb) p s->data_count
$2 = 0
(gdb) p s->blksize
$3 = 0

Beh, something is wrong in this model, because when using ADMA2
length 0 means 65536 bytes (see '1.13.4. Descriptor Table' in
"SD Host Controller Simplified Specification Version 2.00").