Re: dd - behavior on read error

[Pádraig Brady]

On 10/04/17 09:39, David Balažic wrote:
> On 10 April 2017 at 18:04, David Balažic wrote:
>> On 10 April 2017 at 17:05, Pádraig Brady  wrote:
>>> On 10/04/17 06:37, David Balažic wrote:
>>>> Hi!
>>>>
>>>> Using dd on Ubuntu 16.04.2 LTS (64 bit) I discovered this behavior
>>>> that was (for me) unexpected.
>>>>
>>>> There is a bad sector on /dev/sda which on read attempt returns an i/o 
>>>> error.
>>>> Suppose it is sector number 50.
>>>>
>>>> So: dd if=/dev/sda bs=512 count=1 skip=50
>>>> result: I/O error printed on stderr, exit status code is 1 (non zero,
>>>> don't remember the exact value)
>>>>
>>>> But this: dd if=/dev/sda bs=2048 count=1 skip=12
>>>> returns exit code 0 and no error. It reads 1024 bytes (up to the
>>>> erroring sector)
>>>>
>>>> Is this by design? I would expect an I/O error to be reported (in exit 
>>>> code).
>>>
>>> Your probably hitting the default short read behavior of dd
>>> which is defined by POSIX.  I.E. the read() system call will
>>> return what's available before returning the read() error,
>>> however with a count=1 dd will only consider the first read operation.
>>>
>>> You can change this operation with the iflag=fullblock flag.
>>> Note also the conv=sync,noerror settings which would be useful
>>> for reading as much as possible from the disk.
>>
>> Thanks.
>> The dd docs (at [1]) say:
>>
>> ‘fullblock’
>>
>> Accumulate full blocks from input. The read system call may return
>> early if a full block is not available. When that happens, continue
>> calling read to fill the remainder of the block. This flag can be used
>> only with iflag. This flag is useful with pipes for example as they
>> may return short reads. In that case, this flag is needed to ensure
>> that a ‘count=’ argument is interpreted as a block count rather than a
>> count of read operations.
>>
>>
>> Does POSIX mandate treating short reads and error reads the same? Can
>> you point me to the relevant text?
>>
>>
>> [1] 
>> https://www.gnu.org/software/coreutils/manual/html_node/dd-invocation.html#dd-invocation
> 
> 
> I checked the reading code in dd (see below) and apparently if would
> fail on read error, if it were reported by the read() function.
> So apparently it is read() not reporting the error. I'll check with kernel 
> guys.

Everything is behaving as expected here.
The read() call returns less than asked for,
but that's fine and best to do as it's better
to provide what's available.
A subsequent read() will fail though, but because
dd is limited to only 1 read() with count=1 (and without iflag=fullblock)
that error is never encountered by dd.

cheers,
Pádraig