[bpf-next,1/5] bpf: Add __bpf_prog_{enter,exit}_struct_ops for struct_ops trampoline

Commit Message

Martin KaFai Lau Sept. 22, 2022, 10:56 p.m. UTC

From: Martin KaFai Lau <[e-mail]>

The struct_ops prog is to allow using bpf to implement the functions in
a struct (eg. kernel module).  The current usage is to implement the
tcp_congestion.  The kernel does not call the tcp-cc's ops (ie.
the bpf prog) in a recursive way.

The struct_ops is sharing the tracing-trampoline's enter/exit
function which tracks prog->active to avoid recursion.  It is
needed for tracing prog.  However, it turns out the struct_ops
bpf prog will hit this prog->active and unnecessarily skipped
running the struct_ops prog.  eg.  The '.ssthresh' may run in_task()
and then interrupted by softirq that runs the same '.ssthresh'.
Skip running the '.ssthresh' will end up returning random value
to the caller.

The patch adds __bpf_prog_{enter,exit}_struct_ops for the
struct_ops trampoline.  They do not track the prog->active
to detect recursion.

One exception is when the tcp_congestion's '.init' ops is doing
bpf_setsockopt(TCP_CONGESTION) and then recurs to the same
'.init' ops.  This will be addressed in the following patches.

Fixes: ca06f55b9002 ("bpf: Add per-program recursion prevention mechanism")
Signed-off-by: Martin KaFai Lau <[e-mail]>
---
 arch/x86/net/bpf_jit_comp.c |  3 +++
 include/linux/bpf.h         |  4 ++++
 kernel/bpf/trampoline.c     | 23 +++++++++++++++++++++++
 3 files changed, 30 insertions(+)

Patch

diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
index ae89f4143eb4..58a131dec954 100644
--- a/arch/x86/net/bpf_jit_comp.c
+++ b/arch/x86/net/bpf_jit_comp.c
@@ -1836,6 +1836,9 @@  static int invoke_bpf_prog(const struct btf_func_model *m, u8 **pprog,
 	if (p->aux->sleepable) {
 		enter = __bpf_prog_enter_sleepable;
 		exit = __bpf_prog_exit_sleepable;
+	} else if (p->type == BPF_PROG_TYPE_STRUCT_OPS) {
+		enter = __bpf_prog_enter_struct_ops;
+		exit = __bpf_prog_exit_struct_ops;
 	} else if (p->expected_attach_type == BPF_LSM_CGROUP) {
 		enter = __bpf_prog_enter_lsm_cgroup;
 		exit = __bpf_prog_exit_lsm_cgroup;
diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index edd43edb27d6..6fdbc1398b8a 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -864,6 +864,10 @@  u64 notrace __bpf_prog_enter_lsm_cgroup(struct bpf_prog *prog,
 					struct bpf_tramp_run_ctx *run_ctx);
 void notrace __bpf_prog_exit_lsm_cgroup(struct bpf_prog *prog, u64 start,
 					struct bpf_tramp_run_ctx *run_ctx);
+u64 notrace __bpf_prog_enter_struct_ops(struct bpf_prog *prog,
+					struct bpf_tramp_run_ctx *run_ctx);
+void notrace __bpf_prog_exit_struct_ops(struct bpf_prog *prog, u64 start,
+					struct bpf_tramp_run_ctx *run_ctx);
 void notrace __bpf_tramp_enter(struct bpf_tramp_image *tr);
 void notrace __bpf_tramp_exit(struct bpf_tramp_image *tr);
 
diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c
index 41b67eb83ab3..e6551e4a6064 100644
--- a/kernel/bpf/trampoline.c
+++ b/kernel/bpf/trampoline.c
@@ -976,6 +976,29 @@  void notrace __bpf_prog_exit_sleepable(struct bpf_prog *prog, u64 start,
 	rcu_read_unlock_trace();
 }
 
+u64 notrace __bpf_prog_enter_struct_ops(struct bpf_prog *prog,
+					struct bpf_tramp_run_ctx *run_ctx)
+	__acquires(RCU)
+{
+	rcu_read_lock();
+	migrate_disable();
+
+	run_ctx->saved_run_ctx = bpf_set_run_ctx(&run_ctx->run_ctx);
+
+	return bpf_prog_start_time();
+}
+
+void notrace __bpf_prog_exit_struct_ops(struct bpf_prog *prog, u64 start,
+					struct bpf_tramp_run_ctx *run_ctx)
+	__releases(RCU)
+{
+	bpf_reset_run_ctx(run_ctx->saved_run_ctx);
+
+	update_prog_stats(prog, start);
+	migrate_enable();
+	rcu_read_unlock();
+}
+
 void notrace __bpf_tramp_enter(struct bpf_tramp_image *tr)
 {
 	percpu_ref_get(&tr->pcref);