bpf: Fix test_progs -j error with fentry/fexit tests

When multiple threads are attaching/detaching fentry/fexit programs to
the same trampoline, we may call register_fentry on the same trampoline
twice: register_fentry(), unregister_fentry(), then register_fentry again.
This causes ftrace_set_filter_ip() for the same ip on tr->fops twice,
which leaves duplicated ip in tr->fops. The extra ip is not cleaned up
properly on unregister and thus causes failures with further register in
register_ftrace_direct_multi():

register_ftrace_direct_multi()
{
        ...
        for (i = 0; i < size; i++) {
                hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
                        if (ftrace_find_rec_direct(entry->ip))
                                goto out_unlock;
                }
        }
        ...
}

This can be triggered with parallel fentry/fexit tests with test_progs:

  ./test_progs -t fentry,fexit -j

Fix this by resetting tr->fops in ftrace_set_filter_ip(), so that there
will never be duplicated entries in tr->fops.

Fixes: 00963a2e75a8 ("bpf: Support bpf_trampoline on functions with IPMODIFY (e.g. livepatch)")
Reported-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Song Liu <song@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220729194106.1207472-1-song@kernel.org

diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c
index 42e387a..7ec7e23 100644 (file)
--- a/kernel/bpf/trampoline.c
+++ b/kernel/bpf/trampoline.c
@@ -255,7 +255,7 @@ static int register_fentry(struct bpf_trampoline *tr, void *new_addr)
                return -ENOENT;
 
        if (tr->func.ftrace_managed) {
-               ftrace_set_filter_ip(tr->fops, (unsigned long)ip, 0, 0);
+               ftrace_set_filter_ip(tr->fops, (unsigned long)ip, 0, 1);
                ret = register_ftrace_direct_multi(tr->fops, (long)new_addr);
        } else {
                ret = bpf_arch_text_poke(ip, BPF_MOD_CALL, NULL, new_addr);