bpf: Make BPF trampoline use register_ftrace_direct() API

Make BPF trampoline attach its generated assembly code to kernel functions via
register_ftrace_direct() API. It helps ftrace-based tracers co-exist with BPF
trampoline on the same kernel function. It also switches attaching logic from
arch specific text_poke to generic ftrace that is available on many
architectures. text_poke is still necessary for bpf-to-bpf attach and for
bpf_tail_call optimization.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20191209000114.1876138-3-ast@kernel.org

Author: Alexei Starovoitov

include/linux/bpf.h

@@ -461,6 +461,7 @@ struct bpf_trampoline {
 	struct {
 		struct btf_func_model model;
 		void *addr;
+		bool ftrace_managed;
 	} func;
 	/* list of BPF programs using this trampoline */
 	struct hlist_head progs_hlist[BPF_TRAMP_MAX];

kernel/bpf/trampoline.c

@@ -3,6 +3,7 @@
 #include <linux/hash.h>
 #include <linux/bpf.h>
 #include <linux/filter.h>
+#include <linux/ftrace.h>
 
 /* btf_vmlinux has ~22k attachable functions. 1k htab is enough. */
 #define TRAMPOLINE_HASH_BITS 10
@@ -59,6 +60,60 @@ struct bpf_trampoline *bpf_trampoline_lookup(u64 key)
 	return tr;
 }
 
+static int is_ftrace_location(void *ip)
+{
+	long addr;
+
+	addr = ftrace_location((long)ip);
+	if (!addr)
+		return 0;
+	if (WARN_ON_ONCE(addr != (long)ip))
+		return -EFAULT;
+	return 1;
+}
+
+static int unregister_fentry(struct bpf_trampoline *tr, void *old_addr)
+{
+	void *ip = tr->func.addr;
+	int ret;
+
+	if (tr->func.ftrace_managed)
+		ret = unregister_ftrace_direct((long)ip, (long)old_addr);
+	else
+		ret = bpf_arch_text_poke(ip, BPF_MOD_CALL, old_addr, NULL);
+	return ret;
+}
+
+static int modify_fentry(struct bpf_trampoline *tr, void *old_addr, void *new_addr)
+{
+	void *ip = tr->func.addr;
+	int ret;
+
+	if (tr->func.ftrace_managed)
+		ret = modify_ftrace_direct((long)ip, (long)old_addr, (long)new_addr);
+	else
+		ret = bpf_arch_text_poke(ip, BPF_MOD_CALL, old_addr, new_addr);
+	return ret;
+}
+
+/* first time registering */
+static int register_fentry(struct bpf_trampoline *tr, void *new_addr)
+{
+	void *ip = tr->func.addr;
+	int ret;
+
+	ret = is_ftrace_location(ip);
+	if (ret < 0)
+		return ret;
+	tr->func.ftrace_managed = ret;
+
+	if (tr->func.ftrace_managed)
+		ret = register_ftrace_direct((long)ip, (long)new_addr);
+	else
+		ret = bpf_arch_text_poke(ip, BPF_MOD_CALL, NULL, new_addr);
+	return ret;
+}
+
 /* Each call __bpf_prog_enter + call bpf_func + call __bpf_prog_exit is ~50
  * bytes on x86.  Pick a number to fit into PAGE_SIZE / 2
  */
@@ -77,8 +132,7 @@ static int bpf_trampoline_update(struct bpf_trampoline *tr)
 	int err;
 
 	if (fentry_cnt + fexit_cnt == 0) {
-		err = bpf_arch_text_poke(tr->func.addr, BPF_MOD_CALL,
-					 old_image, NULL);
+		err = unregister_fentry(tr, old_image);
 		tr->selector = 0;
 		goto out;
 	}
@@ -105,12 +159,10 @@ static int bpf_trampoline_update(struct bpf_trampoline *tr)
 
 	if (tr->selector)
 		/* progs already running at this address */
-		err = bpf_arch_text_poke(tr->func.addr, BPF_MOD_CALL,
-					 old_image, new_image);
+		err = modify_fentry(tr, old_image, new_image);
 	else
 		/* first time registering */
-		err = bpf_arch_text_poke(tr->func.addr, BPF_MOD_CALL, NULL,
-					 new_image);
+		err = register_fentry(tr, new_image);
 	if (err)
 		goto out;
 	tr->selector++;