Re: [PATCH 1/2] kernel/sys.c: return the current gid when error occurs

["Michael Kerrisk (man-pages)" <mtk.manpages@gmail.com>]

On 08/09/13 02:59, Chen Gang wrote:
> On 08/08/2013 09:37 PM, Michael Kerrisk (man-pages) wrote:
>> On 08/07/13 18:21, Oleg Nesterov wrote:
>>> On 08/06, Andy Lutomirski wrote:
>>>>
>>>> I assume that what the man page means is that the return value is
>>>> whatever fsgid was prior to the call.  On error, fsgid isn't changed, so
>>>> the return value is still "current".
>>>
>>> Probably... Still
>>>
>>> 	On success, the previous value of fsuid is returned.
>>> 	On error, the current value of fsuid is returned.
>>>
>>> looks confusing. sys_setfsuid() always returns the old value.
>>>
>>>> (FWIW, this behavior is awful and is probably the cause of a security
>>>> bug or three, since success and failure are indistinguishable.
>>>
>>> At least this all looks strange.
>>>
>>> I dunno if we can change this old behaviour. I won't be surprized
>>> if someone already uses setfsuid(-1) as getfsuid().
>>>
>>> And perhaps the man page should be changed. Add Michael.
>>
>> Thanks, Oleg. I've applied the following patch to setfsuid.2
>> (and a similar patch to setfsgid.2).
>>
>> Cheers,
>>
>> Michael
>>
>> --- a/man2/setfsuid.2
>> +++ b/man2/setfsuid.2
>> @@ -67,12 +67,8 @@ matches either the real user ID, effective user ID, saved set-user-ID, or
>>  the current value of
>>  .IR fsuid .
>>  .SH RETURN VALUE
>> -On success, the previous value of
>> -.I fsuid
>> -is returned.
>> -On error, the current value of
>> -.I fsuid
>> -is returned.
>> +On both success and failure,
>> +this call returns the previous filesystem user ID of the caller.
>>  .SH VERSIONS
>>  This system call is present in Linux since version 1.2.
>>  .\" This system call is present since Linux 1.1.44
>> @@ -102,7 +98,16 @@ The glibc
>>  .BR setfsuid ()
>>  wrapper function transparently deals with the variation across kernel versions.
>>  .SH BUGS
>> -No error messages of any kind are returned to the caller.
>> +No error indications of any kind are returned to the caller,
>> +and the fact that both successful and unsuccessful calls return
>> +the same value makes it impossible to directly determine
>> +whether the call succeeded or failed.
>> +Instead, the caller must resort to looking at the return value
>> +from a further call such as
>> +.IR setfsuid(\-1)
>> +(which will always fail), in order to determine if a preceding call to
>> +.BR setfsuid ()
>> +changed the filesystem user ID.
>>  At the very
>>  least,
>>  .B EPERM
>>
> 
> Is it suitable to mention this API is obsoleted and unneeded in man page
> ?  ;-)

Yes, that seems reasonable. Done.

Cheers,

Michael