On Wed, Apr 13, 2022 at 11:33 AM Stanislav Fomichev [off-list ref] wrote:
Commit 7d08c2c91171 ("bpf: Refactor BPF_PROG_RUN_ARRAY family of macros
into functions") switched a bunch of BPF_PROG_RUN macros to inline
routines. This changed the semantic a bit. Due to arguments expansion
of macros, it used to be:
rcu_read_lock();
array = rcu_dereference(cgrp->bpf.effective[atype]);
...
Now, with with inline routines, we have:
array_rcu = rcu_dereference(cgrp->bpf.effective[atype]);
/* array_rcu can be kfree'd here */
rcu_read_lock();
array = rcu_dereference(array_rcu);
So subtle difference, wow...
But this open-coding of rcu_read_lock() seems very unfortunate as
well. Would making BPF_PROG_RUN_ARRAY back to a macro which only does
rcu lock/unlock and grabs effective array and then calls static inline
function be a viable solution?
#define BPF_PROG_RUN_ARRAY_CG_FLAGS(array_rcu, ctx, run_prog, ret_flags) \
({
int ret;
rcu_read_lock();
ret = __BPF_PROG_RUN_ARRAY_CG_FLAGS(rcu_dereference(array_rcu), ....);
rcu_read_unlock();
ret;
})
where __BPF_PROG_RUN_ARRAY_CG_FLAGS is what
BPF_PROG_RUN_ARRAY_CG_FLAGS is today but with __rcu annotation dropped
(and no internal rcu stuff)?
I'm assuming in practice rcu subsystem isn't fast enough to trigger
this but let's use rcu API properly: ask callers of BPF_PROG_RUN
to manage rcu themselves.
Also, rename to lower caps to not confuse with macros. Additionally,
drop and expand BPF_PROG_CGROUP_INET_EGRESS_RUN_ARRAY.
See [1] for more context.
[1] https://lore.kernel.org/bpf/CAKH8qBs60fOinFdxiiQikK_q0EcVxGvNTQoWvHLEUGbgcj1UYg@mail.gmail.com/T/#u (local)
Cc: Martin KaFai Lau <redacted>
Fixes: 7d08c2c91171 ("bpf: Refactor BPF_PROG_RUN_ARRAY family of macros into functions")
Signed-off-by: Stanislav Fomichev <redacted>
---
drivers/media/rc/bpf-lirc.c | 8 +++-
include/linux/bpf.h | 70 ++++++-------------------------
kernel/bpf/cgroup.c | 84 +++++++++++++++++++++++++++++--------
kernel/trace/bpf_trace.c | 5 ++-
4 files changed, 90 insertions(+), 77 deletions(-)
[...]