[PATCH 21/23] linux: Add kcmp

Florian Weimer fweimer@redhat.com
Thu Nov 5 12:22:18 GMT 2026


* Adhemerval Zanella Netto:

> On 04/11/26 19:51, Florian Weimer wrote:
>> * Adhemerval Zanella:
>> 
>>> +#ifndef _LINUX_KCMP_H
>>> +/* Types of kernel resources compared by kcmp.  */
>>> +enum kcmp_type
>>> +{
>>> +  KCMP_FILE,
>>> +  KCMP_VM,
>>> +  KCMP_FILES,
>>> +  KCMP_FS,
>>> +  KCMP_SIGHAND,
>>> +  KCMP_IO,
>>> +  KCMP_SYSVSEM,
>>> +  KCMP_EPOLL_TFD,
>>> +
>>> +  KCMP_TYPES
>>> +};
>> 
>> I see that it's exactly like this in the kernel UAPI header, but
>> KCMP_TYPES is not actual constant and will change as more operation
>> types are added.
>
> We can keep as define or remove, I don't have a strong opinion.

I think we have to keep it in sync with the kernel.

>>> +__BEGIN_DECLS
>>> +
>>> +/* Compare the kernel resource of type TYPE (one of the KCMP_* values) used by
>>> +   the process PID1 with the one used by the process PID2.  For KCMP_FILE,
>>> +   IDX1 and IDX2 are file descriptors of the processes; for KCMP_EPOLL_TFD,
>>> +   IDX1 is a file descriptor of PID1 and IDX2 points to a
>>> +   struct kcmp_epoll_slot describing a target file registered in an epoll
>>> +   set of PID2.
>>> +   Return 0 if the resources are the same, 1 or 2 if the first one is
>>> +   respectively less or greater than the second one (in an arbitrary
>>> +   ordering), or -1 and set errno on failure.  */
>>> +extern int kcmp (__pid_t __pid1, __pid_t __pid2, int __type,
>>> +		 unsigned long int __idx1, unsigned long int __idx2) __THROW;
>> 
>> Should we expose the sub-options as individual functions?
>> 
> I don't want to keep adding new wrappers for every kernel feature, and/or
> having do deprecated some comparison that kernel might remove (like futex
> did for some case).

Given that the wrapper does not provide type safety, I think it doesn't
add much value.

Thanks,
Florian



More information about the Libc-alpha mailing list