Re: [PATCH 2/2] clone.2: added clone3() set_tid information

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

Hi Adrian,

Thank you for the patch. I have some review comments below,
and also further comments in my reply to Christian's notes.

On 11/29/19 2:56 PM, Adrian Reber wrote:
> On Thu, Nov 28, 2019 at 06:24:05PM +0100, Christian Brauner wrote:
>> On Thu, Nov 28, 2019 at 01:46:50PM +0100, Adrian Reber wrote:
>>> Signed-off-by: Adrian Reber <areber@redhat.com>
>>> ---
>>>   man2/clone.2 | 90 ++++++++++++++++++++++++++++++++++++++++++++++++++++
>>>   1 file changed, 90 insertions(+)
>>>
>>> diff --git a/man2/clone.2 b/man2/clone.2
>>> index 076b9258e..59c13ec35 100644
>>> --- a/man2/clone.2
>>> +++ b/man2/clone.2
>>> @@ -195,6 +195,8 @@ struct clone_args {
>>>       u64 stack;        /* Pointer to lowest byte of stack */
>>>       u64 stack_size;   /* Size of stack */
>>>       u64 tls;          /* Location of new TLS */
>>> +    u64 set_tid;      /* Pointer to a \fIpid_t\fP array */
>>> +    u64 set_tid_size; /* Number of elements in \fIset_tid\fP */
>>>   };
>>>   .EE
>>>   .in
>>> @@ -262,6 +264,8 @@ flags & 0xff	exit_signal
>>>   stack	stack
>>>   \fP---\fP	stack_size
>>>   tls	tls	See CLONE_SETTLS
>>> +\fP---\fP	set_tid	See below for details
>>> +\fP---\fP	set_tid_size
>>>   .TE
>>>   .RE
>>>   .\"
>>> @@ -285,6 +289,74 @@ options when waiting for the child with
>>>   If no signal (i.e., zero) is specified, then the parent process is not signaled
>>>   when the child terminates.
>>>   .\"
>>> +.SS The set_tid array
>>> +.PP
>>> +The
>>> +.I set_tid
>>> +array is used to select a certain PID for the process to be created by
>>> +.BR clone3 ().
>>> +If the PID of the newly created process should only be set for the current
>>> +PID namespace or in the newly created PID namespace (if
>>> +.I flags
>>> +contains
>>> +.BR CLONE_NEWPID )
>>> +then the first element in the
>>> +.I set_tid
>>> +array has to be the desired PID and
>>> +.I set_tid_size
>>> +needs to be 1.
>>> +.PP
>>> +If the PID of the newly created process should have a certain value in
>>> +multiple PID namespaces the
>>> +.I set_tid
>>> +array can have multiple entries. The first entry defines the PID in the most
>>> +nested PID namespace and all following entries contain the PID of the
>>> +corresponding parent PID namespace. The number of PID namespaces in which a PID
>>> +should be set is defined by
>>> +.I set_tid_size
>>> +which cannot be larger than the number of currently nested PID namespaces.
>>
>> "It's upper cap is the kernel-enforced general nesting limit."
>> or sm like that
> 
> Is that an addition to my sentence or a replacement. I think at this
> point it is more important to point out that it cannot be larger than
> the number of currently nested PID namespaces. Later (at EPERM) I am
> also mentioning that it cannot be larger than the maximum number of
> nested PID namespaces. The code does indeed check if set_tid_size is
> larger than the maximum number of possible nested PID namespaces for
> the user, I think, when calling clone3(), it is more relevant that
> set_tid_size is not larger than the number of currently nested PID
> namespaces. The maximum number of possible nested PID namespaces is more
> likely enforced during unshare() or CLONE_NEWPID (which could be
> happening at the same point in time as set_tid_size larger than maximum
> number of nested PID namespace).
> 
> This definitely feels like too much discussion for a single sentence ;)
> 
> I can add a sentence about the maximum number of nested PID namespaces
> here in addition to the one at EPERM.
> I do not think it is relevant for the user at this point in time.

I'm inclined to agree with you. The maximum nesting depth is a general
detail of PID NSs and is already covered in pid_namespaces((7). I think
there's no need to repeat or allude to that detial here the detail here.

Thanks,

Michael