Re: [PATCH 02/10] af_unix: record the pid of the sending thread

On Mon, Sep 07, 2026 at 11:55:23AM +0200, Alexander Mikhalitsyn wrote:

Am Mo., 31. Aug. 2026 um 13:21 Uhr schrieb Christian Brauner
[off-list ref]:

quoted

Currently only the struct pid of the thread-group leader is recorded.
The identity of the actual thread that sent the message or is connected
to a given socket cannot be retrieved.

Add the plumbing to make it possible to retrieve a pidfd for the sender.
Nothing uses the thread-specific struct pid yet. No functional changes.

Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
---
 include/net/scm.h  |  7 +++----
 net/core/scm.c     | 21 +++++++++++++++++----
 net/unix/af_unix.c | 23 +++++++++++++----------
 net/unix/af_unix.h |  3 ++-
 4 files changed, 35 insertions(+), 19 deletions(-)

diff --git a/include/net/scm.h b/include/net/scm.h
index 86ae6bc109ec..aa7d15c5fc27 100644
--- a/include/net/scm.h
+++ b/include/net/scm.h

@@ -42,7 +42,7 @@ struct scm_fp_list {
 };

 struct scm_cookie {
-       struct pid              *pid;           /* Skb credentials */
+       DECLARE_PIDS(pid, PIDTYPE_TGID);        /* Skb credentials by pid type */
        struct scm_fp_list      *fp;            /* Passed files         */
        struct scm_creds        creds;          /* Skb credentials      */
 #ifdef CONFIG_SECURITY_NETWORK

@@ -69,7 +69,7 @@ static __inline__ void unix_get_peersec_dgram(struct socket *sock, struct scm_co
 static __inline__ void scm_set_cred(struct scm_cookie *scm,
                                    struct pid *pid, kuid_t uid, kgid_t gid)
 {
-       scm->pid = get_pid(pid);
+       scm->pid[PIDTYPE_TGID] = get_pid(pid);
        scm->creds.pid = pid_vnr(pid);
        scm->creds.uid = uid;
        scm->creds.gid = gid;

@@ -77,8 +77,7 @@ static __inline__ void scm_set_cred(struct scm_cookie *scm,

 static __inline__ void scm_destroy_cred(struct scm_cookie *scm)
 {
-       put_pid(scm->pid);
-       scm->pid = NULL;
+       put_pids(scm->pid);
 }

 static __inline__ void scm_destroy(struct scm_cookie *scm)

diff --git a/net/core/scm.c b/net/core/scm.c
index f0d44ecdb11f..9b9e119c353a 100644
--- a/net/core/scm.c
+++ b/net/core/scm.c

@@ -149,6 +149,7 @@ EXPORT_SYMBOL(__scm_destroy);

 static inline int scm_replace_pid(struct scm_cookie *scm, struct pid *pid)
 {
+       struct pid *thread_pid;
        int err;

        /* drop all previous references */

@@ -158,7 +159,18 @@ static inline int scm_replace_pid(struct scm_cookie *scm, struct pid *pid)
        if (unlikely(err))
                return err;

-       scm->pid = pid;
+       /* A sender naming its own thread-group sends from the current thread. */
+       if (pid == task_tgid(current))
+               thread_pid = task_pid(current);
+       else
+               thread_pid = pid;

Hi Christian,

The patch looks good to me, but this spot raised some doubts for me.

If scm_replace_pid() is called with pid == task_tgid(current) then all
good, because we
get scm->pid[PIDTYPE_TGID] == pid and  scm->pid[PIDTYPE_PID] ==
task_pid(current). All good.

If scm_replace_pid() is supplied with pid that has a type PIDTYPE_TGID, then:
- scm->pid[PIDTYPE_TGID] == pid. Good.
- scm->pid[PIDTYPE_PID] = pid. Doesn't looks correct, shoudn't we do

After chatting with you, I think I get the concern. You think that
scm->pid[PIDTYPE_PID] always points to a thread-group leader if pid ==
task_tgid(current). Yes, that's true but that's intended and fine. The
thread-group leader is just a thread in the thread-group just with
magical behavior (that we would really like to get rid of but can't
easily...).

Thanks for looking closely at this though!