Skip to contentSkip to conversation

bpf: copy BPF token from main program to subprograms

PATCH Applied Eduard Zingerman 2 participants 4 messages Apr 15 – Apr 16

Subsystems (3)
lore
Share & tools
lore

Eduard Zingerman

author maintainer
#0 lore raw

v4 cover bpf: copy BPF token from main program to subprograms

bpf_jit_subprogs() omits aux->token when it creates a struct bpf_prog_aux instances for a subprograms.
This means that for programs loaded via BPF token (i.e., from a non-init user namespace), subprograms fail the bpf_token_capable() check in bpf_prog_kallsyms_add() and don't appear in /proc/kallsyms. Which in-turn makes it impossible to freplace such subprograms.

Changelog:
v3 -> v4:
- check sysctl_set calls for errors (sashiko).
v2 -> v3:
- mark selftest as serial (sashiko).
v1 -> v2:
- target bpf-next tree (fixups.c) instead of bpf tree (verifier.c).

v1: https://lore.kernel.org/bpf/[email protected]/T/
v2: https://lore.kernel.org/bpf/[email protected]/T/
v3: https://lore.kernel.org/bpf/[email protected]/T/
---
Eduard Zingerman (2):
bpf: copy BPF token from main program to subprograms
selftests/bpf: verify kallsyms entries for token-loaded subprograms

 kernel/bpf/fixups.c                                |  1 +
 tools/testing/selftests/bpf/Makefile               |  1 +
 tools/testing/selftests/bpf/prog_tests/token.c     | 86 +++++++++++++++++++++-
 .../selftests/bpf/prog_tests/unpriv_bpf_disabled.c | 21 +-----
 tools/testing/selftests/bpf/progs/token_kallsyms.c | 19 +++++
 tools/testing/selftests/bpf/sysctl_helpers.c       | 37 ++++++++++
 tools/testing/selftests/bpf/sysctl_helpers.h       |  8 ++
 7 files changed, 150 insertions(+), 23 deletions(-)

---
base-commit: 71b500afd2f7336f5b6c6026f2af546fc079be26
change-id: 20260414-subprog-token-fix-96c29ffa7e12

author maintainer
#1 lore raw

PATCH v4 1/2 bpf: copy BPF token from main program to subprograms

bpf_jit_subprogs() copies various fields from the main program's aux to each subprogram's aux, but omits the BPF token. This causes bpf_prog_kallsyms_add() to fail for subprograms loaded via BPF token, as bpf_token_capable() falls back to capable() in init_user_ns when token is NULL.

Copy prog->aux->token to func[i]->aux->token so that subprograms inherit the same capability delegation as the main program.

Fixes d79a35497547 ("bpf: Consistently use BPF token throughout BPF verifier logic")
---
 kernel/bpf/fixups.c | 1 +
 1 file changed, 1 insertion(+)
kernel/bpf/fixups.c +1 -0
index 67c9b28767e1..dd00a680e4ea 100644
@@ -1110,6 +1110,7 @@ int bpf_jit_subprogs(struct bpf_verifier_env *env)
 		func[i]->aux->exception_cb = env->subprog_info[i].is_exception_cb;
 		func[i]->aux->changes_pkt_data = env->subprog_info[i].changes_pkt_data;
 		func[i]->aux->might_sleep = env->subprog_info[i].might_sleep;
+		func[i]->aux->token = prog->aux->token;
 		if (!i)
 			func[i]->aux->exception_boundary = env->seen_exception;
 
Signature · 1 line
2.53.0
author maintainer
#2 lore raw

PATCH v4 2/2 selftests/bpf: verify kallsyms entries for token-loaded subprograms

In reply to Eduard Zingerman

Add a test that loads an XDP program with a global subprogram using a BPF token from a user namespace, then verifies that both the main program and the subprogram appear in /proc/kallsyms.

This exercises the bpf_prog_kallsyms_add() path for subprograms and would have caught the missing aux->token copy in bpf_jit_subprogs().

load_kallsyms_local() filters out kallsyms with zero addresses. For a process with limited capabilities to read kallsym addresses the following sysctl variables have to be set to zero:
- /proc/sys/kernel/perf_event_paranoid
- /proc/sys/kernel/kptr_restrict
Set these variables using sysctl_set() utility function extracted from unpriv_bpf_disabled.c to a separate c/header.
Since the test modifies global system state, mark it as serial.

---
 tools/testing/selftests/bpf/Makefile               |  1 +
 tools/testing/selftests/bpf/prog_tests/token.c     | 86 +++++++++++++++++++++-
 .../selftests/bpf/prog_tests/unpriv_bpf_disabled.c | 21 +-----
 tools/testing/selftests/bpf/progs/token_kallsyms.c | 19 +++++
 tools/testing/selftests/bpf/sysctl_helpers.c       | 37 ++++++++++
 tools/testing/selftests/bpf/sysctl_helpers.h       |  8 ++
 6 files changed, 149 insertions(+), 23 deletions(-)
tools/testing/selftests/bpf/Makefile +1 -0
index f75c4f52c028..da81a1853cfd 100644
@@ -750,6 +750,7 @@ TRUNNER_EXTRA_SOURCES := test_progs.c		\
 			 btf_helpers.c		\
 			 cap_helpers.c		\
 			 unpriv_helpers.c 	\
+			 sysctl_helpers.c	\
 			 netlink_helpers.c	\
 			 jit_disasm_helpers.c	\
 			 io_helpers.c		\
tools/testing/selftests/bpf/prog_tests/token.c +83 -3

118 lines not on this page · raw message

tools/testing/selftests/bpf/prog_tests/unpriv_bpf_disabled.c +1 -20
index 472f4f9fa95f..64404602b9ab 100644
@@ -8,6 +8,7 @@
 
 #include "cap_helpers.h"
 #include "bpf_util.h"
+#include "sysctl_helpers.h"
 
 /* Using CAP_LAST_CAP is risky here, since it can get pulled in from
  * an old /usr/include/linux/capability.h and be < CAP_BPF; as a result
@@ -36,26 +37,6 @@ static void process_perfbuf(void *ctx, int cpu, void *data, __u32 len)
 		got_perfbuf_val = *(__u32 *)data;
 }
 
-static int sysctl_set(const char *sysctl_path, char *old_val, const char *new_val)
-{
-	int ret = 0;
-	FILE *fp;
-
-	fp = fopen(sysctl_path, "r+");
-	if (!fp)
-		return -errno;
-	if (old_val && fscanf(fp, "%s", old_val) <= 0) {
-		ret = -ENOENT;
-	} else if (!old_val || strcmp(old_val, new_val) != 0) {
-		fseek(fp, 0, SEEK_SET);
-		if (fprintf(fp, "%s", new_val) < 0)
-			ret = -errno;
-	}
-	fclose(fp);
-
-	return ret;
-}
-
 static void test_unpriv_bpf_disabled_positive(struct test_unpriv_bpf_disabled *skel,
 					      __u32 prog_id, int prog_fd, int perf_fd,
 					      char **map_paths, int *map_fds)
tools/testing/selftests/bpf/progs/token_kallsyms.c new file +19 -0
new file mode 100644
index 000000000000..c9f9344f3eb2
@@ -0,0 +1,19 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+
+#include "vmlinux.h"
+#include <bpf/bpf_helpers.h>
+
+char _license[] SEC("license") = "GPL";
+
+__weak
+int token_ksym_subprog(void)
+{
+	return 0;
+}
+
+SEC("xdp")
+int xdp_main(struct xdp_md *xdp)
+{
+	return token_ksym_subprog();
+}
tools/testing/selftests/bpf/sysctl_helpers.c new file +37 -0
new file mode 100644
index 000000000000..e2bd824f12d5
@@ -0,0 +1,37 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <stdio.h>
+#include <errno.h>
+#include <string.h>
+
+#include "sysctl_helpers.h"
+#include "test_progs.h"
+
+int sysctl_set(const char *sysctl_path, char *old_val, const char *new_val)
+{
+	int ret = 0;
+	FILE *fp;
+
+	fp = fopen(sysctl_path, "r+");
+	if (!fp)
+		return -errno;
+	if (old_val && fscanf(fp, "%s", old_val) <= 0) {
+		ret = -ENOENT;
+	} else if (!old_val || strcmp(old_val, new_val) != 0) {
+		fseek(fp, 0, SEEK_SET);
+		if (fprintf(fp, "%s", new_val) < 0)
+			ret = -errno;
+	}
+	fclose(fp);
+
+	return ret;
+}
+
+int sysctl_set_or_fail(const char *sysctl_path, char *old_val, const char *new_val)
+{
+	int err;
+
+	err = sysctl_set(sysctl_path, old_val, new_val);
+	if (err)
+		PRINT_FAIL("failed to set %s to %s: %s\n", sysctl_path, new_val, strerror(-err));
+	return err;
+}
tools/testing/selftests/bpf/sysctl_helpers.h new file +8 -0
new file mode 100644
index 000000000000..35e37bfe1b3b
@@ -0,0 +1,8 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __SYSCTL_HELPERS_H
+#define __SYSCTL_HELPERS_H
+
+int sysctl_set(const char *sysctl_path, char *old_val, const char *new_val);
+int sysctl_set_or_fail(const char *sysctl_path, char *old_val, const char *new_val);
+
+#endif
Signature · 1 line
2.53.0
#3 lore raw
In reply to Eduard Zingerman
This series was applied to bpf/bpf.git (master) by Alexei Starovoitov <[email protected]>:

Hello:

This series was applied to bpf/bpf.git (master)
by Alexei Starovoitov <[email protected]>:

Quoted 8 lines · On Wed, 15 Apr 2026 13:03:54 -0700 you wrote:

bpf_jit_subprogs() omits aux->token when it creates a struct bpf_prog_aux instances for a subprograms.
This means that for programs loaded via BPF token (i.e., from a non-init user namespace), subprograms fail the bpf_token_capable() check in bpf_prog_kallsyms_add() and don't appear in /proc/kallsyms. Which in-turn makes it impossible to freplace such subprograms.

[...]

Here is the summary with links:
- [bpf-next,v4,1/2] bpf: copy BPF token from main program to subprograms

    https://git.kernel.org/bpf/bpf/c/0251e40c4829
  - [bpf-next,v4,2/2] selftests/bpf: verify kallsyms entries for token-loaded subprograms
    https://git.kernel.org/bpf/bpf/c/969fb456ffb4

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html