bpf: Fix fexit trampoline.

The fexit/fmod_ret programs can be attached to kernel functions that can sleep.
The synchronize_rcu_tasks() will not wait for such tasks to complete.
In such case the trampoline image will be freed and when the task
wakes up the return IP will point to freed memory causing the crash.
Solve this by adding percpu_ref_get/put for the duration of trampoline
and separate trampoline vs its image life times.
The "half page" optimization has to be removed, since
first_half->second_half->first_half transition cannot be guaranteed to
complete in deterministic time. Every trampoline update becomes a new image.
The image with fmod_ret or fexit progs will be freed via percpu_ref_kill and
call_rcu_tasks. Together they will wait for the original function and
trampoline asm to complete. The trampoline is patched from nop to jmp to skip
fexit progs. They are freed independently from the trampoline. The image with
fentry progs only will be freed via call_rcu_tasks_trace+call_rcu_tasks which
will wait for both sleepable and non-sleepable progs to complete.

Fixes: fec56f5890d9 ("bpf: Introduce BPF trampoline")
Reported-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Paul E. McKenney <paulmck@kernel.org>  # for RCU
Link: https://lore.kernel.org/bpf/20210316210007.38949-1-alexei.starovoitov@gmail.com

1 files changed, 22 insertions, 4 deletions

diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
index 747bba0a584a..72b5a57e9e31 100644
--- a/arch/x86/net/bpf_jit_comp.c
+++ b/arch/x86/net/bpf_jit_comp.c
@@ -1936,7 +1936,7 @@ static int invoke_bpf_mod_ret(const struct btf_func_model *m, u8 **pprog,
  * add rsp, 8                      // skip eth_type_trans's frame
  * ret                             // return to its caller
  */
-int arch_prepare_bpf_trampoline(void *image, void *image_end,
+int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *image_end,
 				const struct btf_func_model *m, u32 flags,
 				struct bpf_tramp_progs *tprogs,
 				void *orig_call)
@@ -1975,6 +1975,15 @@ int arch_prepare_bpf_trampoline(void *image, void *image_end,
 
 	save_regs(m, &prog, nr_args, stack_size);
 
+	if (flags & BPF_TRAMP_F_CALL_ORIG) {
+		/* arg1: mov rdi, im */
+		emit_mov_imm64(&prog, BPF_REG_1, (long) im >> 32, (u32) (long) im);
+		if (emit_call(&prog, __bpf_tramp_enter, prog)) {
+			ret = -EINVAL;
+			goto cleanup;
+		}
+	}
+
 	if (fentry->nr_progs)
 		if (invoke_bpf(m, &prog, fentry, stack_size))
 			return -EINVAL;
@@ -1993,8 +2002,7 @@ int arch_prepare_bpf_trampoline(void *image, void *image_end,
 	}
 
 	if (flags & BPF_TRAMP_F_CALL_ORIG) {
-		if (fentry->nr_progs || fmod_ret->nr_progs)
-			restore_regs(m, &prog, nr_args, stack_size);
+		restore_regs(m, &prog, nr_args, stack_size);
 
 		/* call original function */
 		if (emit_call(&prog, orig_call, prog)) {
@@ -2003,6 +2011,8 @@ int arch_prepare_bpf_trampoline(void *image, void *image_end,
 		}
 		/* remember return value in a stack for bpf prog to access */
 		emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_0, -8);
+		im->ip_after_call = prog;
+		emit_nops(&prog, 5);
 	}
 
 	if (fmod_ret->nr_progs) {
@@ -2033,9 +2043,17 @@ int arch_prepare_bpf_trampoline(void *image, void *image_end,
 	 * the return value is only updated on the stack and still needs to be
 	 * restored to R0.
 	 */
-	if (flags & BPF_TRAMP_F_CALL_ORIG)
+	if (flags & BPF_TRAMP_F_CALL_ORIG) {
+		im->ip_epilogue = prog;
+		/* arg1: mov rdi, im */
+		emit_mov_imm64(&prog, BPF_REG_1, (long) im >> 32, (u32) (long) im);
+		if (emit_call(&prog, __bpf_tramp_exit, prog)) {
+			ret = -EINVAL;
+			goto cleanup;
+		}
 		/* restore original return value back into RAX */
 		emit_ldx(&prog, BPF_DW, BPF_REG_0, BPF_REG_FP, -8);
+	}
 
 	EMIT1(0x5B); /* pop rbx */
 	EMIT1(0xC9); /* leave */