Check cfi_stubs before registering a struct_ops type.
Subsystems (4)
v3 cover Check cfi_stubs before registering a struct_ops type.
From: Kui-Feng Lee <[email protected]>
Recently, cfi_stubs were introduced. However, existing struct_ops types that are not in the upstream may not be aware of this, resulting in kernel crashes. By rejecting struct_ops types that do not provide cfi_stubs properly during registration, these crashes can be avoided.
---
Changes from v2:
- Add a stub function for get_info of struct tcp_congestion_ops.
Changes from v1:
- Check *(void **)(cfi_stubs + moff) to make sure stub functions are provided for every operator.
- Add a test case to ensure that struct_ops rejects incomplete cfi_stub.
v2: https://lore.kernel.org/all/[email protected]/
v1: https://lore.kernel.org/all/[email protected]/
Kui-Feng Lee (3):
x86/cfi,bpf: Add a stub function for get_info of struct
tcp_congestion_ops.
bpf: Check cfi_stubs before registering a struct_ops type. selftests/bpf: Test case for lacking CFI stub functions.
kernel/bpf/bpf_struct_ops.c | 14 +++ net/ipv4/bpf_tcp_ca.c | 7 ++ tools/testing/selftests/bpf/Makefile | 10 +- .../selftests/bpf/bpf_test_no_cfi/Makefile | 19 ++++ .../bpf/bpf_test_no_cfi/bpf_test_no_cfi.c | 93 +++++++++++++++++++ .../bpf/prog_tests/test_struct_ops_no_cfi.c | 31 +++++++ tools/testing/selftests/bpf/testing_helpers.c | 4 +- tools/testing/selftests/bpf/testing_helpers.h | 2 + 8 files changed, 177 insertions(+), 3 deletions(-) create mode 100644 tools/testing/selftests/bpf/bpf_test_no_cfi/Makefile create mode 100644 tools/testing/selftests/bpf/bpf_test_no_cfi/bpf_test_no_cfi.c create mode 100644 tools/testing/selftests/bpf/prog_tests/test_struct_ops_no_cfi.c
Signature · 1 line
2.34.1
PATCH v3 1/3 x86/cfi,bpf: Add a stub function for get_info of struct tcp_congestion_ops.
From: Kui-Feng Lee <[email protected]>
struct tcp_congestion_ops is missing a stub function for get_info. This is required for checking the consistency of cfi_stubs of struct_ops.
--- net/ipv4/bpf_tcp_ca.c | 7 +++++++ 1 file changed, 7 insertions(+)
net/ipv4/bpf_tcp_ca.c +7 -0
@@ -321,6 +321,12 @@ static u32 bpf_tcp_ca_sndbuf_expand(struct sock *sk)return 0 ; } +static size_t bpf_tcp_ca_get_info (struct sock *sk,u32 ext,int *attr, +union tcp_cc_info *info) +{ +return 0 ; +} +static void __bpf_tcp_ca_init (struct sock *sk) { }
@@ -340,6 +346,7 @@ static struct tcp_congestion_ops __bpf_ops_tcp_congestion_ops = { .cong_control = bpf_tcp_ca_cong_control, .undo_cwnd = bpf_tcp_ca_undo_cwnd, .sndbuf_expand = bpf_tcp_ca_sndbuf_expand, + .get_info = bpf_tcp_ca_get_info, .init = __bpf_tcp_ca_init, .release = __bpf_tcp_ca_release,
Signature · 1 line
2.34.1
PATCH v3 2/3 bpf: Check cfi_stubs before registering a struct_ops type.
In reply to [email protected]From: Kui-Feng Lee <[email protected]>
Recently, cfi_stubs were introduced. However, existing struct_ops types that are not in the upstream may not be aware of this, resulting in kernel crashes. By rejecting struct_ops types that do not provide cfi_stubs during registration, these crashes can be avoided.
--- kernel/bpf/bpf_struct_ops.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+)
kernel/bpf/bpf_struct_ops.c +14 -0
@@ -302,6 +302,11 @@ int bpf_struct_ops_desc_init(struct bpf_struct_ops_desc *st_ops_desc, }sprintf (value_name,"%s%s" ,VALUE_PREFIX , st_ops->name); +if (!st_ops->cfi_stubs) { +pr_warn ("struct %s has no cfi_stubs\n" , st_ops->name); +return -EINVAL ; + } + type_id =btf_find_by_name_kind (btf, st_ops->name,BTF_KIND_STRUCT );if (type_id <0 ) {
@@ -339,6 +344,7 @@ int bpf_struct_ops_desc_init(struct bpf_struct_ops_desc *st_ops_desc,for_each_member (i, t, member) {const struct btf_type *func_proto; +u32 moff; mname =btf_name_by_offset (btf, member->name_off);if (!*mname) {
@@ -361,6 +367,14 @@ int bpf_struct_ops_desc_init(struct bpf_struct_ops_desc *st_ops_desc,if (!func_proto)continue ; + moff =__btf_member_bit_offset (t, member) /8 ; +if (!*(void **)(st_ops->cfi_stubs + moff)) { +pr_warn ("member %s in struct %s has no cfi stub function\n" , + mname, st_ops->name); + err = -EINVAL ; + goto errout; + } +if (btf_distill_func_proto (log, btf, func_proto, mname, &st_ops->func_models[i])) {
Signature · 1 line
2.34.1
PATCH v3 3/3 selftests/bpf: Test case for lacking CFI stub functions.
In reply to [email protected]From: Kui-Feng Lee <[email protected]>
Ensure struct_ops rejects the registration of struct_ops types without proper CFI stub functions.
bpf_test_no_cfi.ko is a module that attempts to register a struct_ops type called "bpf_test_no_cfi_ops" with varying levels of cfi_stubs. It starts with a NULL cfi_stub and ends with a fully complete cfi_stub. Only the fully complete cfi_stub should be accepted by struct_ops. The module can only be loaded successfully if these registrations yield the expected results.
--- tools/testing/selftests/bpf/Makefile | 10 +- .../selftests/bpf/bpf_test_no_cfi/Makefile | 19 ++++ .../bpf/bpf_test_no_cfi/bpf_test_no_cfi.c | 93 +++++++++++++++++++ .../bpf/prog_tests/test_struct_ops_no_cfi.c | 31 +++++++ tools/testing/selftests/bpf/testing_helpers.c | 4 +- tools/testing/selftests/bpf/testing_helpers.h | 2 + 6 files changed, 156 insertions(+), 3 deletions(-) create mode 100644 tools/testing/selftests/bpf/bpf_test_no_cfi/Makefile create mode 100644 tools/testing/selftests/bpf/bpf_test_no_cfi/bpf_test_no_cfi.c create mode 100644 tools/testing/selftests/bpf/prog_tests/test_struct_ops_no_cfi.c
tools/testing/selftests/bpf/Makefile +9 -1
@@ -132,7 +132,7 @@ TEST_GEN_PROGS_EXTENDED = test_sock_addr test_skb_cgroup_id_user \ flow_dissector_load test_flow_dissector test_tcp_check_syncookie_user \ test_lirc_mode2_user xdping test_cpp runqslower bench bpf_testmod.ko \ xskxceiver xdp_redirect_multi xdp_synproxy veristat xdp_hw_metadata \ - xdp_features + xdp_features bpf_test_no_cfi.ko TEST_GEN_FILES += liburandom_read.so urandom_read sign-file uprobe_multi
@@ -254,6 +254,12 @@ $(OUTPUT)/bpf_testmod.ko: $(VMLINUX_BTF) $(RESOLVE_BTFIDS) $(wildcard bpf_testmo $(Q)$(MAKE) $(submake_extras) RESOLVE_BTFIDS=$(RESOLVE_BTFIDS) -C bpf_testmod $(Q)cp bpf_testmod/bpf_testmod.ko $@ +$(OUTPUT)/bpf_test_no_cfi.ko: $(VMLINUX_BTF) $(RESOLVE_BTFIDS) $(wildcard bpf_test_no_cfi/Makefile bpf_test_no_cfi/*.[ch]) + $(call msg,MOD,,$@) + $(Q)$(RM) bpf_test_no_cfi/bpf_test_no_cfi.ko # force re-compilation + $(Q)$(MAKE) $(submake_extras) RESOLVE_BTFIDS=$(RESOLVE_BTFIDS) -C bpf_test_no_cfi + $(Q)cp bpf_test_no_cfi/bpf_test_no_cfi.ko $@ + DEFAULT_BPFTOOL := $(HOST_SCRATCH_DIR)/sbin/bpftool ifneq ($(CROSS_COMPILE),) CROSS_BPFTOOL := $(SCRATCH_DIR)/sbin/bpftool
@@ -628,6 +634,7 @@ TRUNNER_EXTRA_SOURCES := test_progs.c \ flow_dissector_load.h \ ip_check_defrag_frags.h TRUNNER_EXTRA_FILES := $(OUTPUT)/urandom_read $(OUTPUT)/bpf_testmod.ko \ + $(OUTPUT)/bpf_test_no_cfi.ko \ $(OUTPUT)/liburandom_read.so \ $(OUTPUT)/xdp_synproxy \ $(OUTPUT)/sign-file \
@@ -756,6 +763,7 @@ EXTRA_CLEAN := $(SCRATCH_DIR) $(HOST_SCRATCH_DIR) \ feature bpftool \ $(addprefix $(OUTPUT)/,*.o *.skel.h *.lskel.h *.subskel.h \ no_alu32 cpuv4 bpf_gcc bpf_testmod.ko \ + bpf_test_no_cfi.ko \ liburandom_read.so) .PHONY: docs docs-clean
tools/testing/selftests/bpf/bpf_test_no_cfi/Makefile new file +19 -0
@@ -0,0 +1,19 @@ +BPF_TEST_NO_CFI_DIR := $(realpath $(dir $(abspath $(lastword $(MAKEFILE_LIST))))) +KDIR ?= $(abspath $(BPF_TEST_NO_CFI_DIR)/../../../../..) + +ifeq ($(V),1) +Q = +else +Q = @ +endif + +MODULES = bpf_test_no_cfi.ko + +obj-m += bpf_test_no_cfi.o + +all: + +$(Q)make -C $(KDIR) M=$(BPF_TEST_NO_CFI_DIR) modules + +clean: + +$(Q)make -C $(KDIR) M=$(BPF_TEST_NO_CFI_DIR) clean +
tools/testing/selftests/bpf/bpf_test_no_cfi/bpf_test_no_cfi.c new file +93 -0
93 lines not on this page · raw message
tools/testing/selftests/bpf/prog_tests/test_struct_ops_no_cfi.c new file +31 -0
@@ -0,0 +1,31 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2024 Meta Platforms, Inc. and affiliates. */ +#include <test_progs.h> +#include <testing_helpers.h> + +void test_struct_ops_no_cfi (void ) +{ +int fd; +int err; + + fd =open ("bpf_test_no_cfi.ko" ,O_RDONLY ); +if (!ASSERT_GT (fd,0 ,"open" )) { +close (fd); +return ; + } + +/* The module will try to register a struct_ops type with +* no cfi_stubs, incomplete cfi_stubs, and full cfi_stubs. +* +* Only full cfi_stubs should be allowed. The module will be loaded +* successfully if the result of the registration is as expected, +* or it fails. +*/ + err =finit_module (fd,"" ,0 ); +close (fd); +if (!ASSERT_OK (err,"finit_module" )) +return ; + + err =delete_module ("bpf_test_no_cfi" ,0 ); +ASSERT_OK (err,"delete_module" ); +}
tools/testing/selftests/bpf/testing_helpers.c +2 -2
@@ -356,12 +356,12 @@ __u64 read_perf_max_sample_freq(void)return sample_freq; } -static int finit_module (int fd,const char *param_values,int flags) +int finit_module (int fd,const char *param_values,int flags) {return syscall (__NR_finit_module, fd, param_values, flags); } -static int delete_module (const char *name,int flags) +int delete_module (const char *name,int flags) {return syscall (__NR_delete_module, name, flags); }
tools/testing/selftests/bpf/testing_helpers.h +2 -0
@@ -36,6 +36,8 @@ __u64 read_perf_max_sample_freq(void);int load_bpf_testmod (bool verbose);int unload_bpf_testmod (bool verbose);int kern_sync_rcu (void ); +int finit_module (int fd,const char *param_values,int flags); +int delete_module (const char *name,int flags);static inline __u64 get_time_ns (void ) {
Signature · 1 line
2.34.1
Re: [PATCH bpf-next v3 1/3] x86/cfi,bpf: Add a stub function for get_info of struct tcp_congestion_ops.
In reply to [email protected]Quoted 30 lines · On Fri, Feb 16, 2024 at 11:34 AM <[email protected]> wrote:
From: Kui-Feng Lee <[email protected]>
struct
tcp_congestion_opsis missing a stub function forget_info. This is required for checking the consistency ofcfi_stubsofstruct_ops.Cc Peter Zijlstra <[email protected]>Signed-off-by Kui-Feng Lee <[email protected]>--- net/ipv4/bpf_tcp_ca.c | 7 +++++++ 1 file changed, 7 insertions(+)net/ipv4/bpf_tcp_ca.c +6
-0@@ -321,6 +321,12 @@ static u32 bpf_tcp_ca_sndbuf_expand(struct sock *sk)return 0 ; } +static size_t bpf_tcp_ca_get_info (struct sock *sk,u32 ext,int *attr, +union tcp_cc_info *info) +{ +return 0 ; +} +static void __bpf_tcp_ca_init (struct sock *sk) { }
net/ipv4/bpf_tcp_ca.c static struct tcp_congestion_ops __bpf_ops_tcp_congestion_ops = { · line 349
Quoted 4 lines
net/ipv4/bpf_tcp_ca.c +1
-0@@ -340,6 +346,7 @@ static struct tcp_congestion_ops __bpf_ops_tcp_congestion_ops = { .cong_control = bpf_tcp_ca_cong_control, .undo_cwnd = bpf_tcp_ca_undo_cwnd, .sndbuf_expand = bpf_tcp_ca_sndbuf_expand, + .get_info = bpf_tcp_ca_get_info,
No. It was explicitly skipped.
The plan is to use NULL in cfi_stubs to remove static u32 unsupported_ops[] = {
offsetof(struct tcp_congestion_ops, get_info),
};
and manual check of such fields in is_unsupported(). NULL is cfi_stubs will be such an indication.
pw-bot: cr
Re: [PATCH bpf-next v3 1/3] x86/cfi,bpf: Add a stub function for get_info of struct tcp_congestion_ops.
Quoted 34 lines · Alexei Starovoitov
Quoted 33 lines · On Fri, Feb 16, 2024 at 11:34 AM <[email protected]> wrote:
From: Kui-Feng Lee <[email protected]>
struct
tcp_congestion_opsis missing a stub function forget_info. This is required for checking the consistency ofcfi_stubsofstruct_ops.Cc Peter Zijlstra <[email protected]>Signed-off-by Kui-Feng Lee <[email protected]>--- net/ipv4/bpf_tcp_ca.c | 7 +++++++ 1 file changed, 7 insertions(+)net/ipv4/bpf_tcp_ca.c +6
-0@@ -321,6 +321,12 @@ static u32 bpf_tcp_ca_sndbuf_expand(struct sock *sk)return 0 ; } +static size_t bpf_tcp_ca_get_info (struct sock *sk,u32 ext,int *attr, +union tcp_cc_info *info) +{ +return 0 ; +} +static void __bpf_tcp_ca_init (struct sock *sk) { }@@ -340,6 +346,7 @@ static struct tcp_congestion_ops __bpf_ops_tcp_congestion_ops = { .cong_control = bpf_tcp_ca_cong_control, .undo_cwnd = bpf_tcp_ca_undo_cwnd, .sndbuf_expand = bpf_tcp_ca_sndbuf_expand,
Quoted 8 lines
net/ipv4/bpf_tcp_ca.c static struct tcp_congestion_ops __bpf_ops_tcp_congestion_ops = { · line 349
Quoted 1 line
net/ipv4/bpf_tcp_ca.c +1
-0+ .get_info = bpf_tcp_ca_get_info,No. It was explicitly skipped.
The plan is to use NULL incfi_stubsto remove static u32 unsupported_ops[] = {
offsetof(struct tcp_congestion_ops, get_info),
};
and manual check of such fields inis_unsupported().
Ok! That means I have to check the result of st_ops->check() to skip unsupported ops.
Quoted 3 lines
NULL is
cfi_stubswill be such an indication.pw-bot: cr