Re: [Qemu-devel] [Bug 1739304] Re: Passing a directory to (eg.) -cdrom results in misleading error message

[John Snow]

On 01/31/2018 01:36 PM, Michal Suchánek wrote:
> On Wed, 20 Dec 2017 21:57:00 -0500
> John Snow <address@hidden> wrote:
> 
>> On 12/20/2017 05:31 AM, Dr. David Alan Gilbert wrote:
>>> Yep, can repeat it here, it seems pretty random which error it
>>> gives:
>>>
>>> address@hidden try]$ ./x86_64-softmmu/qemu-system-x86_64
>>> -cdrom /tmp qemu-system-x86_64: -cdrom /tmp: Could not refresh
>>> total sector count: Invalid argument address@hidden
>>> try]$ ./x86_64-softmmu/qemu-system-x86_64 -cdrom /var
>>> qemu-system-x86_64: -cdrom /var: Could not read image for
>>> determining its format: File too large
>>>
>>>
>>> ** Changed in: qemu
>>>        Status: New => Confirmed
>>>   
>>
>> Looks like directories play funny games.
>>
> 
> ...
> 
>> specifically:
>>
>> (gdb) s
>> 1963     size = lseek(s->fd, 0, SEEK_END);
>> (gdb) s
>> 1964     if (size < 0) {
>> (gdb) print size
>> $37 = 9223372036854775807
>>
>> cool, cool, cool. This value is 0x7fffffffffffffff and errno isn't
>> set. cool and good function.
> 
> Indeed: The behavior of lseek() on devices which are incapable of
> seeking is implementation-defined.
> 
>>
>> so, lseek on a folder returns crazy nonsense. Perhaps we ought to
>> actually specifically disallow folders, we don't appear to.
> 
> It probably returns -1 which it is supposed to do on error. It should
> also set errno in that case, though.
> 
> So this is probably bug in the error handling code in lseek.
> 
> Thanks
> 
> Michal
> 

Thanks. It looks like we can't make stronger guarantees about the
behavior of lseek, so I submitted a patch to ratchet down QEMU's
acceptable file types:

https://lists.gnu.org/archive/html/qemu-devel/2018-01/msg05055.html

Thanks,
--John