Re: [PATCH bpf-next 1/4] bpf: enable task local storage for tracing programs

On Sat, Jan 9, 2021 at 12:35 AM Song Liu [off-list ref] wrote:

To access per-task data, BPF program typically creates a hash table with
pid as the key. This is not ideal because:
 1. The use need to estimate requires size of the hash table, with may be
    inaccurate;
 2. Big hash tables are slow;
 3. To clean up the data properly during task terminations, the user need
    to write code.

Task local storage overcomes these issues and becomes a better option for
these per-task data. Task local storage is only available to BPF_LSM. Now
enable it for tracing programs.

Also mention here that you change the pointer from being a security blob to a
dedicated member in the task struct. I assume this is because you want to
use it without CONFIG_BPF_LSM?

Can you also mention the reasons for changing the
raw_spin_lock_bh to raw_spin_lock_irqsave in the commit log?

Reported-by: kernel test robot <redacted>
Signed-off-by: Song Liu <redacted>
---
 include/linux/bpf.h            |  7 +++++++
 include/linux/bpf_lsm.h        | 22 ----------------------
 include/linux/bpf_types.h      |  2 +-
 include/linux/sched.h          |  5 +++++
 kernel/bpf/Makefile            |  3 +--
 kernel/bpf/bpf_local_storage.c | 28 +++++++++++++++++-----------
 kernel/bpf/bpf_lsm.c           |  4 ----
 kernel/bpf/bpf_task_storage.c  | 26 ++++++--------------------
 kernel/fork.c                  |  5 +++++
 kernel/trace/bpf_trace.c       |  4 ++++
 10 files changed, 46 insertions(+), 60 deletions(-)

[...]

On Jan 11, 2021, at 2:14 AM, KP Singh [off-list ref] wrote:

On Sat, Jan 9, 2021 at 12:35 AM Song Liu [off-list ref] wrote:

quoted

To access per-task data, BPF program typically creates a hash table with
pid as the key. This is not ideal because:
1. The use need to estimate requires size of the hash table, with may be
   inaccurate;
2. Big hash tables are slow;
3. To clean up the data properly during task terminations, the user need
   to write code.

Task local storage overcomes these issues and becomes a better option for
these per-task data. Task local storage is only available to BPF_LSM. Now
enable it for tracing programs.

Also mention here that you change the pointer from being a security blob to a
dedicated member in the task struct. I assume this is because you want to
use it without CONFIG_BPF_LSM?

Yes, exactly. I will add this to the commit log. 

quoted

Can you also mention the reasons for changing the
raw_spin_lock_bh to raw_spin_lock_irqsave in the commit log?

This is because we will use these in irq context. I will add this to the 
commit log. 

Thanks,
Song