Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Jens Axboe <axboe@kernel.dk>
To:     io-uring@vger.kernel.org
Cc:     torvalds@linux-foundation.org, ebiederm@xmission.com,
        linux-kernel@vger.kernel.org, oleg@redhat.com, metze@samba.org,
        Jens Axboe <axboe@kernel.dk>
Subject: [PATCH 2/2] proc: don't show PF_IO_WORKER threads as threads in /proc/<pid>/task/
Date:   Thu, 25 Mar 2021 10:43:43 -0600
Message-Id: <20210325164343.807498-3-axboe@kernel.dk>
In-Reply-To: <20210325164343.807498-1-axboe@kernel.dk>
References: <20210325164343.807498-1-axboe@kernel.dk>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

We don't allow SIGSTOP and ptrace attach to these threads, and that
confuses applications like gdb that assume they can attach to any thread
listed in /proc/<pid>/task/. gdb then enters an infinite loop of retrying
attach, even though it fails with the same error (-EPERM) every time.

Skip over PF_IO_WORKER threads in the proc task setup. We can't just
terminate the when we find a PF_IO_WORKER thread, as there's no real
ordering here. It's perfectly feasible to have the first thread be an
IO worker, and then a real thread after that. Hence just implement the
skip.

Reported-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/proc/base.c | 22 ++++++++++++++--------
 1 file changed, 14 insertions(+), 8 deletions(-)

diff --git a/fs/proc/base.c b/fs/proc/base.c
index 3851bfcdba56..abff2fe10bfa 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -3723,7 +3723,7 @@ static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
 	 */
 	pos = task = task->group_leader;
 	do {
-		if (!nr--)
+		if (same_thread_group(task, pos) && !nr--)
 			goto found;
 	} while_each_thread(task, pos);
 fail:
@@ -3744,16 +3744,22 @@ static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
  */
 static struct task_struct *next_tid(struct task_struct *start)
 {
-	struct task_struct *pos = NULL;
+	struct task_struct *tmp, *pos = NULL;
+
 	rcu_read_lock();
-	if (pid_alive(start)) {
-		pos = next_thread(start);
-		if (thread_group_leader(pos))
-			pos = NULL;
-		else
-			get_task_struct(pos);
+	if (!pid_alive(start))
+		goto no_thread;
+	list_for_each_entry_rcu(tmp, &start->thread_group, thread_group) {
+		if (!thread_group_leader(tmp) && same_thread_group(start, tmp)) {
+			get_task_struct(tmp);
+			pos = tmp;
+			break;
+		}
 	}
+no_thread:
 	rcu_read_unlock();
+	if (!pos)
+		return NULL;
 	put_task_struct(start);
 	return pos;
 }
-- 
2.31.0