| Message ID | 20231222113102.4148-4-laoar.shao@gmail.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | BPF |
| Headers | show |
| Series | bpf: Add bpf_iter_cpumask | expand |
Hello, On Fri, Dec 22, 2023 at 11:31:01AM +0000, Yafang Shao wrote: > Introducing a new kfunc: bpf_cpumask_set_from_pid. This function serves the > purpose of retrieving the cpumask associated with a specific PID. Its > utility is particularly evident within container environments. For > instance, it allows for extracting the cpuset of a container using the > init task within it. > > Signed-off-by: Yafang Shao <laoar.shao@gmail.com> ... > +__bpf_kfunc bool bpf_cpumask_set_from_pid(struct cpumask *cpumask, u32 pid) > +{ > + struct task_struct *task; > + > + if (!cpumask) > + return false; > + > + task = get_pid_task(find_vpid(pid), PIDTYPE_PID); > + if (!task) > + return false; > + > + cpumask_copy(cpumask, task->cpus_ptr); > + put_task_struct(task); > + return true; > +} This seems awfully specific. Why is this necessary? Shouldn't the BPF prog get the task and bpf_cpumask_copy() its ->cpus_ptr instead? Thanks.
On Sat, Dec 23, 2023 at 1:51 AM Tejun Heo <tj@kernel.org> wrote: > > Hello, > > On Fri, Dec 22, 2023 at 11:31:01AM +0000, Yafang Shao wrote: > > Introducing a new kfunc: bpf_cpumask_set_from_pid. This function serves the > > purpose of retrieving the cpumask associated with a specific PID. Its > > utility is particularly evident within container environments. For > > instance, it allows for extracting the cpuset of a container using the > > init task within it. > > > > Signed-off-by: Yafang Shao <laoar.shao@gmail.com> > ... > > +__bpf_kfunc bool bpf_cpumask_set_from_pid(struct cpumask *cpumask, u32 pid) > > +{ > > + struct task_struct *task; > > + > > + if (!cpumask) > > + return false; > > + > > + task = get_pid_task(find_vpid(pid), PIDTYPE_PID); > > + if (!task) > > + return false; > > + > > + cpumask_copy(cpumask, task->cpus_ptr); > > + put_task_struct(task); > > + return true; > > +} > > This seems awfully specific. Why is this necessary? Shouldn't the BPF prog > get the task and bpf_cpumask_copy() its ->cpus_ptr instead? Good point. I missed the bpf_cpumask_copy(). Will use it instead.
diff --git a/kernel/bpf/cpumask.c b/kernel/bpf/cpumask.c index 4ae07a4..5755bb6 100644 --- a/kernel/bpf/cpumask.c +++ b/kernel/bpf/cpumask.c @@ -467,6 +467,22 @@ __bpf_kfunc void bpf_iter_cpumask_destroy(struct bpf_iter_cpumask *it) bpf_mem_free(&bpf_global_ma, kit->cpu); } +__bpf_kfunc bool bpf_cpumask_set_from_pid(struct cpumask *cpumask, u32 pid) +{ + struct task_struct *task; + + if (!cpumask) + return false; + + task = get_pid_task(find_vpid(pid), PIDTYPE_PID); + if (!task) + return false; + + cpumask_copy(cpumask, task->cpus_ptr); + put_task_struct(task); + return true; +} + __bpf_kfunc_end_defs(); BTF_SET8_START(cpumask_kfunc_btf_ids) @@ -498,6 +514,7 @@ __bpf_kfunc void bpf_iter_cpumask_destroy(struct bpf_iter_cpumask *it) BTF_ID_FLAGS(func, bpf_iter_cpumask_new, KF_ITER_NEW | KF_RCU) BTF_ID_FLAGS(func, bpf_iter_cpumask_next, KF_ITER_NEXT | KF_RET_NULL | KF_RCU) BTF_ID_FLAGS(func, bpf_iter_cpumask_destroy, KF_ITER_DESTROY) +BTF_ID_FLAGS(func, bpf_cpumask_set_from_pid, KF_RCU) BTF_SET8_END(cpumask_kfunc_btf_ids) static const struct btf_kfunc_id_set cpumask_kfunc_set = {
Introducing a new kfunc: bpf_cpumask_set_from_pid. This function serves the purpose of retrieving the cpumask associated with a specific PID. Its utility is particularly evident within container environments. For instance, it allows for extracting the cpuset of a container using the init task within it. Signed-off-by: Yafang Shao <laoar.shao@gmail.com> --- kernel/bpf/cpumask.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)