Skip to contentSkip to conversation

limit bpf_core_types_are_compat recursion

PATCH Applied Matteo Croce 3 participants 7 messages Feb 4, 2022 – Feb 6, 2022

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

v3 cover limit bpf_core_types_are_compat recursion

From: Matteo Croce <[email protected]>

As formerly discussed on the BPF mailing list:
https://lore.kernel.org/bpf/CAADnVQJDax2j0-7uyqdqFEnpB57om_z+Cqmi1O2QyLpHqkVKwA@mail.gmail.com/

changes from v2:
test the bpf_core_type_exists() return value, and check that the recursion limit is enforced.

Matteo Croce (2):
bpf: limit bpf_core_types_are_compat() recursion selftests/bpf: test maximum recursion depth for
bpf_core_types_are_compat()

 include/linux/btf.h                           |   5 +
 kernel/bpf/btf.c                              | 105 +++++++++++++++++-
 tools/testing/selftests/bpf/Makefile          |   2 +-
 .../selftests/bpf/bpf_testmod/bpf_testmod.c   |   5 +
 .../selftests/bpf/prog_tests/core_kern.c      |  15 ++-
 .../bpf/prog_tests/core_kern_overflow.c       |  13 +++
 tools/testing/selftests/bpf/progs/core_kern.c |  14 +++
 .../selftests/bpf/progs/core_kern_overflow.c  |  21 ++++
 8 files changed, 177 insertions(+), 3 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/core_kern_overflow.c
 create mode 100644 tools/testing/selftests/bpf/progs/core_kern_overflow.c
Signature · 1 line
2.34.1
#1 lore raw

PATCH v3 1/2 bpf: limit bpf_core_types_are_compat() recursion

From: Matteo Croce <[email protected]>

In userspace, bpf_core_types_are_compat() is a recursive function which can't be put in the kernel as is.
Limit the recursion depth to 2, to avoid potential stack overflows in kernel.

---
 include/linux/btf.h |   5 +++
 kernel/bpf/btf.c    | 105 +++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 109 insertions(+), 1 deletion(-)
include/linux/btf.h +5 -0
index f6c43dd513fa..36bc09b8e890 100644
@@ -327,6 +327,11 @@ static inline const struct btf_var_secinfo *btf_type_var_secinfo(
 	return (const struct btf_var_secinfo *)(t + 1);
 }
 
+static inline struct btf_param *btf_params(const struct btf_type *t)
+{
+	return (struct btf_param *)(t + 1);
+}
+
 #ifdef CONFIG_BPF_SYSCALL
 struct bpf_prog;
 
kernel/bpf/btf.c +104 -1

114 lines not on this page · raw message

Signature · 1 line
2.34.1
#3 lore raw

PATCH v3 2/2 selftests/bpf: test maximum recursion depth for bpf_core_types_are_compat()

In reply to Matteo Croce

From: Matteo Croce <[email protected]>

bpf_core_types_are_compat() was limited to 2 recursion levels, which are enough to parse a function prototype.
Add a test which checks the existence of a function prototype, so to test the bpf_core_types_are_compat() code path.

The test for the recursion limit being hit is done in a separate object, because the kernel failure makes the whole load to fail.

Sample run log with extra prints:

	[ 5689.913751] bpf_core_apply_relo_insn:1200 cands->len: 2
	[ 5689.913902] bpf_core_types_are_compat:6896: ret: 1
	[ 5689.913994] bpf_core_types_are_compat:6896: ret: 0
	[ 5689.914025] bpf_core_apply_relo_insn:1200 cands->len: 2
	[ 5689.914141] bpf_core_types_are_compat:6896: ret: 0
	[ 5689.914246] bpf_core_types_are_compat:6896: ret: 0
	test_core_kern_lskel:PASS:open_and_load 0 nsec
	test_core_kern_lskel:PASS:attach(core_relo_proto) 0 nsec
	test_core_kern_lskel:PASS:bpf_core_type_exists 0 nsec
	test_core_kern_lskel:PASS:!bpf_core_type_exists 0 nsec
	#41 core_kern_lskel:OK
	[ 5689.915267] bpf_core_apply_relo_insn:1200 cands->len: 2
	[ 5689.915399] bpf_core_types_are_compat:6896: ret: 0
	[ 5689.915504] bpf_core_types_are_compat:6896: ret: -22
	test_core_kern_overflow_lskel:PASS:open_and_load 0 nsec
	#42 core_kern_overflow_lskel:OK
	Summary: 2/0 PASSED, 0 SKIPPED, 0 FAILED
	Successfully unloaded bpf_testmod.ko.
---
 tools/testing/selftests/bpf/Makefile          |  2 +-
 .../selftests/bpf/bpf_testmod/bpf_testmod.c   |  5 +++++
 .../selftests/bpf/prog_tests/core_kern.c      | 15 ++++++++++++-
 .../bpf/prog_tests/core_kern_overflow.c       | 13 ++++++++++++
 tools/testing/selftests/bpf/progs/core_kern.c | 14 +++++++++++++
 .../selftests/bpf/progs/core_kern_overflow.c  | 21 +++++++++++++++++++
 6 files changed, 68 insertions(+), 2 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/core_kern_overflow.c
 create mode 100644 tools/testing/selftests/bpf/progs/core_kern_overflow.c
tools/testing/selftests/bpf/Makefile +1 -1
index 945f92d71db3..91ea729990da 100644
@@ -330,7 +330,7 @@ LINKED_SKELS := test_static_linked.skel.h linked_funcs.skel.h		\
 
 LSKELS := kfunc_call_test.c fentry_test.c fexit_test.c fexit_sleep.c \
 	test_ringbuf.c atomics.c trace_printk.c trace_vprintk.c \
-	map_ptr_kern.c core_kern.c
+	map_ptr_kern.c core_kern.c core_kern_overflow.c
 # Generate both light skeleton and libbpf skeleton for these
 LSKELS_EXTRA := test_ksyms_module.c test_ksyms_weak.c kfunc_call_test_subprog.c
 SKEL_BLACKLIST += $$(LSKELS)
tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c +5 -0
index 595d32ab285a..e5ba8d8a17da 100644
@@ -13,6 +13,11 @@
 #define CREATE_TRACE_POINTS
 #include "bpf_testmod-events.h"
 
+typedef int (*func_proto_typedef___match)(long);
+typedef int (*func_proto_typedef___overflow)(func_proto_typedef___match);
+func_proto_typedef___match funcp = NULL;
+func_proto_typedef___overflow funcp_of = NULL;
+
 DEFINE_PER_CPU(int, bpf_testmod_ksym_percpu) = 123;
 
 noinline void
tools/testing/selftests/bpf/prog_tests/core_kern.c +14 -1
index 561c5185d886..91493f5836ff 100644
@@ -7,8 +7,21 @@
 void test_core_kern_lskel(void)
 {
 	struct core_kern_lskel *skel;
+	int link_fd;
 
 	skel = core_kern_lskel__open_and_load();
-	ASSERT_OK_PTR(skel, "open_and_load");
+	if (!ASSERT_OK_PTR(skel, "open_and_load"))
+		return;
+
+	link_fd = core_kern_lskel__core_relo_proto__attach(skel);
+	if (!ASSERT_GT(link_fd, 0, "attach(core_relo_proto)"))
+		goto cleanup;
+
+	/* trigger tracepoints */
+	usleep(1);
+	ASSERT_TRUE(skel->bss->proto_out[0], "bpf_core_type_exists");
+	ASSERT_FALSE(skel->bss->proto_out[1], "!bpf_core_type_exists");
+
+cleanup:
 	core_kern_lskel__destroy(skel);
 }
tools/testing/selftests/bpf/prog_tests/core_kern_overflow.c new file +13 -0
new file mode 100644
index 000000000000..04cc145bc26a
@@ -0,0 +1,13 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "test_progs.h"
+#include "core_kern_overflow.lskel.h"
+
+void test_core_kern_overflow_lskel(void)
+{
+	struct core_kern_overflow_lskel *skel;
+
+	skel = core_kern_overflow_lskel__open_and_load();
+	if (!ASSERT_NULL(skel, "open_and_load"))
+		core_kern_overflow_lskel__destroy(skel);
+}
tools/testing/selftests/bpf/progs/core_kern.c +14 -0
index 13499cc15c7d..acabe4cb0480 100644
@@ -101,4 +101,18 @@ int balancer_ingress(struct __sk_buff *ctx)
 	return 0;
 }
 
+typedef int (*func_proto_typedef___match)(long);
+typedef void (*func_proto_typedef___doesnt_match)(char*);
+
+int proto_out[2];
+
+SEC("raw_tracepoint/sys_enter")
+int core_relo_proto(void *ctx)
+{
+	proto_out[0] = bpf_core_type_exists(func_proto_typedef___match);
+	proto_out[1] = bpf_core_type_exists(func_proto_typedef___doesnt_match);
+
+	return 0;
+}
+
 char LICENSE[] SEC("license") = "GPL";
tools/testing/selftests/bpf/progs/core_kern_overflow.c new file +21 -0
new file mode 100644
index 000000000000..70417413af55
@@ -0,0 +1,21 @@
+// SPDX-License-Identifier: GPL-2.0
+#include "vmlinux.h"
+
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+#include <bpf/bpf_core_read.h>
+
+typedef int (*func_proto_typedef___match)(long);
+typedef int (*func_proto_typedef___overflow)(func_proto_typedef___match);
+
+int proto_out;
+
+SEC("raw_tracepoint/sys_enter")
+int core_relo_proto(void *ctx)
+{
+	proto_out = bpf_core_type_exists(func_proto_typedef___overflow);
+
+	return 0;
+}
+
+char LICENSE[] SEC("license") = "GPL";
Signature · 1 line
2.34.1
#2 lore raw

Re: [PATCH bpf-next v3 1/2] bpf: limit bpf_core_types_are_compat() recursion

In reply to Matteo Croce

kernel/bpf/btf.c int register_btf_kfunc_id_set(enum bpf_prog_type prog_type, · line 6849

Quoted 4 lines · Matteo Croce
kernel/bpf/btf.c +4 -0
+
+               for (i = 0; i < local_vlen; i++, local_p++, targ_p++) {
+                       if (level <= 1)
+                               return -EINVAL;

<= 1 is not correct, since it makes MAX_TYPES_ARE_COMPAT_DEPTH 2
misleading.
With <= 1 the recursion depth is just 1.
So I've changed to to <=0 to make the actual depth of 2.

#4 lore raw

Re: [PATCH bpf-next v3 2/2] selftests/bpf: test maximum recursion depth for bpf_core_types_are_compat()

In reply to Matteo Croce

tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c line 17

Quoted 8 lines · Matteo Croce
tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c +2 -0
@@ -13,6 +13,11 @@
 #define CREATE_TRACE_POINTS
 #include "bpf_testmod-events.h"
 
+typedef int (*func_proto_typedef___match)(long);
+typedef int (*func_proto_typedef___overflow)(func_proto_typedef___match);

There is no need for "___flavor" on the kernel side of type definition. It makes the test confusing to read.

tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c line 19

Quoted 2 lines
tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c +2 -0
+func_proto_typedef___match funcp = NULL;
+func_proto_typedef___overflow funcp_of = NULL;

We have BTF_TYPE_EMIT() macro to avoid unnecessary declaration.

tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c line 21

Quoted 2 lines
tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c +2 -0
+typedef int (*func_proto_typedef___match)(long);
+typedef int (*func_proto_typedef___overflow)(func_proto_typedef___match);

With <=1 in the previous patch such single depth of func_proto was reaching the recursion limit.
Hence the fix <=0 was necessary.
I've also changed this test to:

+typedef int (*func_proto_typedef)(long);
+typedef int (*func_proto_typedef_nested1)(func_proto_typedef); +typedef int (*func_proto_typedef_nested2)(func_proto_typedef_nested1);

in bpf_testmod.c and in progs/core_kern_overflow.c and bpf_core_type_exists(func_proto_typedef_nested2); to go above the limit.

Also added bpf_core_type_exists(func_proto_typedef_nested1) to progs/core_kern.c to stay at the limit.

Please see the result in bpf-next.

#6 lore raw
In reply to Matteo Croce
This series was applied to bpf/bpf-next.git (master) by Alexei Starovoitov <[email protected]>:

Hello:

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

Quoted 10 lines · On Fri, 4 Feb 2022 01:55:17 +0100 you wrote:

From: Matteo Croce <[email protected]>

As formerly discussed on the BPF mailing list:
https://lore.kernel.org/bpf/CAADnVQJDax2j0-7uyqdqFEnpB57om_z+Cqmi1O2QyLpHqkVKwA@mail.gmail.com/

changes from v2:
test the bpf_core_type_exists() return value, and check that the recursion limit is enforced.

[...]

Here is the summary with links:
- [bpf-next,v3,1/2] bpf: limit bpf_core_types_are_compat() recursion

    https://git.kernel.org/bpf/bpf-next/c/e70e13e7d4ab
  - [bpf-next,v3,2/2] selftests/bpf: test maximum recursion depth for bpf_core_types_are_compat()
    https://git.kernel.org/bpf/bpf-next/c/976a38e05a49

You are awesome, thank you!

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

Re: [PATCH bpf-next v3 2/2] selftests/bpf: test maximum recursion depth for bpf_core_types_are_compat()

In reply to Alexei Starovoitov
Quoted 36 lines · Alexei Starovoitov

tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c line 17

Quoted 8 lines · Matteo Croce
tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c +2 -0
@@ -13,6 +13,11 @@
 #define CREATE_TRACE_POINTS
 #include "bpf_testmod-events.h"
 
+typedef int (*func_proto_typedef___match)(long);
+typedef int (*func_proto_typedef___overflow)(func_proto_typedef___match);

There is no need for "___flavor" on the kernel side of type definition. It makes the test confusing to read.

tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c line 19

Quoted 2 lines
tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c +2 -0
+func_proto_typedef___match funcp = NULL;
+func_proto_typedef___overflow funcp_of = NULL;

We have BTF_TYPE_EMIT() macro to avoid unnecessary declaration.

tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c line 21

Quoted 2 lines
tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c +2 -0
+typedef int (*func_proto_typedef___match)(long);
+typedef int (*func_proto_typedef___overflow)(func_proto_typedef___match);

With <=1 in the previous patch such single depth of func_proto was reaching the recursion limit.
Hence the fix <=0 was necessary.
I've also changed this test to:

(quoted diff) +3 -0
+typedef int (*func_proto_typedef)(long);
+typedef int (*func_proto_typedef_nested1)(func_proto_typedef);
+typedef int (*func_proto_typedef_nested2)(func_proto_typedef_nested1);

in bpf_testmod.c and in progs/core_kern_overflow.c and bpf_core_type_exists(func_proto_typedef_nested2); to go above the limit.

Quoted 4 lines

Also added bpf_core_type_exists(func_proto_typedef_nested1) to progs/core_kern.c to stay at the limit.

Please see the result in bpf-next.

Awesome.
I've seen both patches in the repo, LGTM.

Signature · 1 line
per aspera ad upstream