[PATCH V2] libxcmd: return error from cvtnum() on overflow

[Eric Sandeen]

On 3/1/11 3:00 PM, Alex Elder wrote:
> On Mon, 2011-02-28 at 15:26 -0600, Eric Sandeen wrote:
>> Test 071 was failing in weird ways, partly because it was trying
>> to pass in offsets larger than strtoll() could accept, which then
>> silently returned LLONG_MAX instead.  For DIO tests, this was
>> unaligned, so we got unexpected (to me, anyay) alignment errors.
>>
>> At least printing out the perror() makes this more obvious,
>> but unfortunately we then get the somewhat odd output:
>>
>> # xfs_io -f -d -c "pwrite 9223373136366403584  4096" /mnt/test/grrr
>> cvtnum: Numerical result out of range
>> non-numeric offset argument -- 9223373136366403584
>>
>> Test 071 still fails, but at least it's a bit more obvious as to why.
> 
> Your change looks good.  But here are a few more general questions
> (for anyone who cares to respond--not just you):
> - Do you plan to get test 071 working?  (Just curious.)

some day maybe, and I'd like to make it a generic test.

> - mkfs/xfs_mkfs.c and extimate/xfs_estimate.c each define their
>   own version of the same function.  Do you know why?  Is there
>   any reason we couldn't just have one?

I don't know ;)

> - The three version of cvtnum() are each a bit different.  Two
>   of them (the other two) return -1 for an empty string, while
>   this one returns 0.

hrm.

> - I'm not sure what you meant by "non-numeric" versus "invalid"
>   in call sites.

I mean perror says:

cvtnum: Numerical result out of range

but then the caller says:

non-numeric offset argument -- 9223373136366403584

"9223373136366403584" is not non-numeric; it is out of range.  :)

> - Call sites seem to be a bit varied on how (or whether) they
>   look for errors.  Kind of a mess...

yeah.

> Regardless, you can consider this one reviewed.  We should
> fix all three instances of the function to fix this problem
> though--either the same as this (and in the same commit)
> or separeately.

ok I may fix up the others, I'd forgotten about that.

-Eric

> Reviewed-by: Alex Elder <aelder at sgi.com>
> 
>> Signed-off-by: Eric Sandeen <sandeen at redhat.com>
>> ---
>>
>> V2: zero errno first so we don't pick up a stale errno.
>>
>> Note:
>> ... should I change all callsites from "non-numeric" to "invalid" perhaps?
> 
>