[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20210809080432.l3rf6e5fzjqlz42f@wittgenstein>
Date: Mon, 9 Aug 2021 10:04:32 +0200
From: Christian Brauner <christian.brauner@...ntu.com>
To: Eugene Syromiatnikov <esyr@...hat.com>
Cc: "Peter Zijlstra (Intel)" <peterz@...radead.org>,
"Joel Fernandes (Google)" <joel@...lfernandes.org>,
Chris Hyser <chris.hyser@...cle.com>,
Josh Don <joshdon@...gle.com>, Ingo Molnar <mingo@...nel.org>,
Vincent Guittot <vincent.guittot@...aro.org>,
Valentin Schneider <valentin.schneider@....com>,
Mel Gorman <mgorman@...e.de>, linux-kernel@...r.kernel.org,
Thomas Gleixner <tglx@...utronix.de>,
"Dmitry V. Levin" <ldv@...ace.io>, linux-doc@...r.kernel.org,
linux-api@...r.kernel.org, Jonathan Corbet <corbet@....net>,
libc-alpha@...rceware.org, fweimer@...hat.com
Subject: Re: [PATCH] uapi: expose enum pid_type
On Sat, Aug 07, 2021 at 03:01:23AM +0200, Eugene Syromiatnikov wrote:
> Commit 7ac592aa35a684ff ("sched: prctl() core-scheduling interface")
> made use of enum pid_type in prctl's arg4; this type and the associated
(Hm, I thought we were trying to stop exposing the thread-group leader
concept directly to userspace. But that ship probably has sailed now.)
> + __PIDTYPE_PID,
> + __PIDTYPE_TGID,
> + __PIDTYPE_PGID,
What's rather odd is that the interface seems to not be using
PIDTYPE_PID, PIDTYPE_TGID, or PIDTYPE_PGID to indicate the type of the
pid but rather the scope of the operation which conflicts with the
names in the enum.
So I think exposing these values as PIDTYPE_* is misleading as it
doesn't really apply to the passed in pid. For example, if I pass in a
non-thread group leader and specify PIDTYPE_TGID then this interface
goes and digs out the thread-group leader from signal->pids when
arguably it should've failed because I passed in a non-thread-group
leader pid. So I think we shouldn't expose the enum with __PIDTYPE_*
prefix.
I think there's two ways. First, the prctl() would need to change a bit
and then we could use the P_PID, P_PGID definitions from wait.h that we
already expose to userspace (adding a P_TGID). But that would mean a
uapi change for the prctl() and I've never really liked those defines
anyway as they are equally strangely named.
So I would suggest we expose a custom enum or a set of defines in
prctl.h. Like (sorry, probably terrible names):
#define PR_SCHED_CORE_SCOPE_THREAD
#define PR_SCHED_CORE_SCOPE_THREADGROUP
#define PR_SCHED_CORE_SCOPE_PROCESSGROUP
that have the same values as PIDTYPE_* but express clearly that this
indicates the scope of the operation independent of whether or not the
passed in pid is a thread-group leader, process-group leader or not.
Thoughts?
Christian
> enumeration definitions are not exposed to userspace. Try to fix that
> by providing enum _kernel_pid_type and tying in-kernel enum pid_type
> definitions to it. Note that enum pid_type cannot be exposed as is,
> since "enum pid_type" is already exists in various projects [1] (notably
> gcc and strace), and "enum __pid_type" is defined by glibc and uclibc
> for fcntl(F_SETOWN_EX) owner ID type.
>
> [1] https://codesearch.debian.net/search?q=enum+pid_type
>
> Complements: 7ac592aa35a684ff ("sched: prctl() core-scheduling interface")
> Signed-off-by: Eugene Syromiatnikov <esyr@...hat.com>
> ---
> .../admin-guide/hw-vuln/core-scheduling.rst | 7 ++++---
> include/linux/pid.h | 12 +++++++-----
> include/uapi/linux/pid.h | 20 ++++++++++++++++++++
> include/uapi/linux/prctl.h | 1 +
> 4 files changed, 32 insertions(+), 8 deletions(-)
> create mode 100644 include/uapi/linux/pid.h
>
> diff --git a/Documentation/admin-guide/hw-vuln/core-scheduling.rst b/Documentation/admin-guide/hw-vuln/core-scheduling.rst
> index 7b410ae..3eb2b7c 100644
> --- a/Documentation/admin-guide/hw-vuln/core-scheduling.rst
> +++ b/Documentation/admin-guide/hw-vuln/core-scheduling.rst
> @@ -61,9 +61,10 @@ arg3:
> ``pid`` of the task for which the operation applies.
>
> arg4:
> - ``pid_type`` for which the operation applies. It is of type ``enum pid_type``.
> - For example, if arg4 is ``PIDTYPE_TGID``, then the operation of this command
> - will be performed for all tasks in the task group of ``pid``.
> + ``pid_type`` for which the operation applies. It is of type
> + ``enum __kernel_pid_type``. For example, if arg4 is ``__PIDTYPE_TGID``,
> + then the operation of this command will be performed for all tasks
> + in the task group of ``pid``.
>
> arg5:
> userspace pointer to an unsigned long for storing the cookie returned by
> diff --git a/include/linux/pid.h b/include/linux/pid.h
> index fa10acb..f8ca4c9 100644
> --- a/include/linux/pid.h
> +++ b/include/linux/pid.h
> @@ -5,14 +5,16 @@
> #include <linux/rculist.h>
> #include <linux/wait.h>
> #include <linux/refcount.h>
> +#include <uapi/linux/pid.h>
>
> enum pid_type
> {
> - PIDTYPE_PID,
> - PIDTYPE_TGID,
> - PIDTYPE_PGID,
> - PIDTYPE_SID,
> - PIDTYPE_MAX,
> + PIDTYPE_PID = __PIDTYPE_PID,
> + PIDTYPE_TGID = __PIDTYPE_TGID,
> + PIDTYPE_PGID = __PIDTYPE_PGID,
> + PIDTYPE_SID = __PIDTYPE_SID,
> +
> + PIDTYPE_MAX = __PIDTYPE_MAX
> };
>
> /*
> diff --git a/include/uapi/linux/pid.h b/include/uapi/linux/pid.h
> new file mode 100644
> index 0000000..91d08e4
> --- /dev/null
> +++ b/include/uapi/linux/pid.h
> @@ -0,0 +1,20 @@
> +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
> +#ifndef _UAPI_LINUX_PID_H
> +#define _UAPI_LINUX_PID_H
> +
> +/*
> + * Type of process-related ID. So far, it is used only for prctl(PR_SCHED_CORE);
> + * not to be confused with type field of f_owner_ex structure argument
> + * of fcntl(F_SETOWN_EX).
> + */
> +enum __kernel_pid_type
> +{
> + __PIDTYPE_PID,
> + __PIDTYPE_TGID,
> + __PIDTYPE_PGID,
> + __PIDTYPE_SID,
> +
> + __PIDTYPE_MAX /* Non-UAPI */
> +};
> +
> +#endif /* _UAPI_LINUX_PID_H */
> diff --git a/include/uapi/linux/prctl.h b/include/uapi/linux/prctl.h
> index 967d9c5..4e794aa 100644
> --- a/include/uapi/linux/prctl.h
> +++ b/include/uapi/linux/prctl.h
> @@ -3,6 +3,7 @@
> #define _LINUX_PRCTL_H
>
> #include <linux/types.h>
> +#include <linux/pid.h> /* enum __kernel_pid_type */
>
> /* Values to pass as first argument to prctl() */
>
> --
> 2.1.4
>
Powered by blists - more mailing lists