Re: [PATCH bpf-next 2/5] bpf: add main_thread_only customization for task/task_file iterators

Currently, task and task_file by default iterates through
all tasks. For task_file, by default, all files from all tasks
will be traversed.

But for a user process, the file_table is shared by all threads
of that process. So traversing the main thread per process should
be enough to traverse all files and this can save a lot of cpu
time if some process has large number of threads and each thread
has lots of open files.

This patch implemented a customization for task/task_file iterator,
permitting to traverse only the kernel task where its pid equal
to tgid in the kernel. This includes some kernel threads, and
main threads of user processes. This will solve the above potential
performance issue for task_file. This customization may be useful
for task iterator too if only traversing main threads is enough.

Signed-off-by: Yonghong Song <redacted>
---
 include/linux/bpf.h            |  3 ++-
 include/uapi/linux/bpf.h       |  5 ++++
 kernel/bpf/task_iter.c         | 46 +++++++++++++++++++++++-----------
 tools/include/uapi/linux/bpf.h |  5 ++++
 4 files changed, 43 insertions(+), 16 deletions(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index a6131d95e31e..058eb9b0ba78 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h

@@ -1220,7 +1220,8 @@ int bpf_obj_get_user(const char __user *pathname, int flags);
        int __init bpf_iter_ ## target(args) { return 0; }

 struct bpf_iter_aux_info {
-       struct bpf_map *map;
+       struct bpf_map *map;    /* for iterator traversing map elements */
+       bool main_thread_only;  /* for task/task_file iterator */