Under CONFIG_FORTIFY_SOURCE, memcpy() will check the size of destination and source buffers. Defining kernel_headers_data as "char" would trip this check. Since these addresses are treated as byte arrays, define them as arrays (as done everywhere else).
This was seen with:
$ cat /sys/kernel/kheaders.tar.xz >> /dev/null
detected buffer overflow in memcpy kernel BUG at lib/string_helpers.c:1027! ... RIP: 0010:fortify_panic+0xf/0x20 [...] Call Trace: <TASK> ikheaders_read+0x45/0x50 [kheaders] kernfs_fop_read_iter+0x1a4/0x2f0 ...
Reported-by: Jakub Kicinski kuba@kernel.org Link: https://lore.kernel.org/bpf/20230302112130.6e402a98@kernel.org/ Tested-by: Jakub Kicinski kuba@kernel.org Fixes: 43d8ce9d65a5 ("Provide in-kernel headers to make extending kernel easier") Cc: Joel Fernandes (Google) joel@joelfernandes.org Cc: stable@vger.kernel.org Signed-off-by: Kees Cook keescook@chromium.org --- kernel/kheaders.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/kernel/kheaders.c b/kernel/kheaders.c index 8f69772af77b..42163c9e94e5 100644 --- a/kernel/kheaders.c +++ b/kernel/kheaders.c @@ -26,15 +26,15 @@ asm ( " .popsection \n" );
-extern char kernel_headers_data; -extern char kernel_headers_data_end; +extern char kernel_headers_data[]; +extern char kernel_headers_data_end[];
static ssize_t ikheaders_read(struct file *file, struct kobject *kobj, struct bin_attribute *bin_attr, char *buf, loff_t off, size_t len) { - memcpy(buf, &kernel_headers_data + off, len); + memcpy(buf, &kernel_headers_data[off], len); return len; }
@@ -48,8 +48,8 @@ static struct bin_attribute kheaders_attr __ro_after_init = {
static int __init ikheaders_init(void) { - kheaders_attr.size = (&kernel_headers_data_end - - &kernel_headers_data); + kheaders_attr.size = (kernel_headers_data_end - + kernel_headers_data); return sysfs_create_bin_file(kernel_kobj, &kheaders_attr); }
On Thu, Mar 02, 2023 at 02:49:50PM -0800, Kees Cook wrote:
Under CONFIG_FORTIFY_SOURCE, memcpy() will check the size of destination and source buffers. Defining kernel_headers_data as "char" would trip this check. Since these addresses are treated as byte arrays, define them as arrays (as done everywhere else).
This was seen with:
$ cat /sys/kernel/kheaders.tar.xz >> /dev/null
detected buffer overflow in memcpy kernel BUG at lib/string_helpers.c:1027! ... RIP: 0010:fortify_panic+0xf/0x20 [...] Call Trace:
<TASK> ikheaders_read+0x45/0x50 [kheaders] kernfs_fop_read_iter+0x1a4/0x2f0 ...
Reported-by: Jakub Kicinski kuba@kernel.org Link: https://lore.kernel.org/bpf/20230302112130.6e402a98@kernel.org/ Tested-by: Jakub Kicinski kuba@kernel.org
Acked-by: Joel Fernandes (Google) joel@joelfernandes.org
thanks,
- Joel
Fixes: 43d8ce9d65a5 ("Provide in-kernel headers to make extending kernel easier") Cc: Joel Fernandes (Google) joel@joelfernandes.org Cc: stable@vger.kernel.org Signed-off-by: Kees Cook keescook@chromium.org
kernel/kheaders.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/kernel/kheaders.c b/kernel/kheaders.c index 8f69772af77b..42163c9e94e5 100644 --- a/kernel/kheaders.c +++ b/kernel/kheaders.c @@ -26,15 +26,15 @@ asm ( " .popsection \n" ); -extern char kernel_headers_data; -extern char kernel_headers_data_end; +extern char kernel_headers_data[]; +extern char kernel_headers_data_end[]; static ssize_t ikheaders_read(struct file *file, struct kobject *kobj, struct bin_attribute *bin_attr, char *buf, loff_t off, size_t len) {
- memcpy(buf, &kernel_headers_data + off, len);
- memcpy(buf, &kernel_headers_data[off], len); return len;
} @@ -48,8 +48,8 @@ static struct bin_attribute kheaders_attr __ro_after_init = { static int __init ikheaders_init(void) {
- kheaders_attr.size = (&kernel_headers_data_end -
&kernel_headers_data);
- kheaders_attr.size = (kernel_headers_data_end -
kernel_headers_data);} -- 2.34.1
From: Kees Cook keescook@chromium.org Date: Thu, 2 Mar 2023 14:49:50 -0800
Under CONFIG_FORTIFY_SOURCE, memcpy() will check the size of destination and source buffers. Defining kernel_headers_data as "char" would trip this check. Since these addresses are treated as byte arrays, define them as arrays (as done everywhere else).
Yet another array-as-one-char, I wonder how many are still here...
This was seen with:
$ cat /sys/kernel/kheaders.tar.xz >> /dev/null
detected buffer overflow in memcpy kernel BUG at lib/string_helpers.c:1027! ... RIP: 0010:fortify_panic+0xf/0x20 [...] Call Trace:
<TASK> ikheaders_read+0x45/0x50 [kheaders] kernfs_fop_read_iter+0x1a4/0x2f0 ...
Reported-by: Jakub Kicinski kuba@kernel.org Link: https://lore.kernel.org/bpf/20230302112130.6e402a98@kernel.org/ Tested-by: Jakub Kicinski kuba@kernel.org Fixes: 43d8ce9d65a5 ("Provide in-kernel headers to make extending kernel easier") Cc: Joel Fernandes (Google) joel@joelfernandes.org Cc: stable@vger.kernel.org Signed-off-by: Kees Cook keescook@chromium.org
Reviewed-by: Alexander Lobakin aleksander.lobakin@intel.com
kernel/kheaders.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
[...]
Thanks, Olek
On Fri, Mar 03, 2023 at 04:19:51PM +0100, Alexander Lobakin wrote:
From: Kees Cook keescook@chromium.org Date: Thu, 2 Mar 2023 14:49:50 -0800
Under CONFIG_FORTIFY_SOURCE, memcpy() will check the size of destination and source buffers. Defining kernel_headers_data as "char" would trip this check. Since these addresses are treated as byte arrays, define them as arrays (as done everywhere else).
Yet another array-as-one-char, I wonder how many are still here...
Yeah, good point. They do tend to stand out; we could find them:
$ git grep 'extern char [^*[( ]*;' arch/alpha/boot/bootp.c:extern char _end; arch/alpha/boot/bootpz.c:extern char _end; arch/alpha/boot/main.c: extern char _end; arch/arm/mach-rockchip/core.h:extern char rockchip_secondary_trampoline; arch/arm/mach-rockchip/core.h:extern char rockchip_secondary_trampoline_end; arch/arm/mach-zynq/common.h:extern char zynq_secondary_trampoline; arch/arm/mach-zynq/common.h:extern char zynq_secondary_trampoline_jump; arch/arm/mach-zynq/common.h:extern char zynq_secondary_trampoline_end; arch/hexagon/include/uapi/asm/setup.h:extern char external_cmdline_buffer; arch/ia64/include/asm/smp.h:extern char no_int_routing; arch/ia64/kernel/process.c: extern char ia64_ret_from_clone; arch/mips/dec/prom/memory.c: extern char genexcept_early; arch/mips/kernel/traps.c: extern char except_vec3_generic; arch/mips/kernel/traps.c: extern char except_vec4; arch/mips/kernel/traps.c: extern char except_vec3_r4000; arch/mips/mm/c-octeon.c: extern char except_vec2_octeon; arch/parisc/boot/compressed/misc.c:extern char output_len; arch/parisc/boot/compressed/misc.c:extern char _startcode_end; arch/powerpc/include/asm/smp.h:extern char __secondary_hold; arch/s390/include/asm/kvm_host.h:extern char sie_exit; arch/sh/boards/mach-ap325rxa/setup.c:extern char ap325rxa_sdram_enter_start; arch/sh/boards/mach-ap325rxa/setup.c:extern char ap325rxa_sdram_enter_end; arch/sh/boards/mach-ap325rxa/setup.c:extern char ap325rxa_sdram_leave_start; arch/sh/boards/mach-ap325rxa/setup.c:extern char ap325rxa_sdram_leave_end; arch/sh/boards/mach-ecovec24/setup.c:extern char ecovec24_sdram_enter_start; arch/sh/boards/mach-ecovec24/setup.c:extern char ecovec24_sdram_enter_end; arch/sh/boards/mach-ecovec24/setup.c:extern char ecovec24_sdram_leave_start; arch/sh/boards/mach-ecovec24/setup.c:extern char ecovec24_sdram_leave_end; arch/sh/boards/mach-kfr2r09/setup.c:extern char kfr2r09_sdram_enter_start; arch/sh/boards/mach-kfr2r09/setup.c:extern char kfr2r09_sdram_enter_end; arch/sh/boards/mach-kfr2r09/setup.c:extern char kfr2r09_sdram_leave_start; arch/sh/boards/mach-kfr2r09/setup.c:extern char kfr2r09_sdram_leave_end; arch/sh/boards/mach-migor/setup.c:extern char migor_sdram_enter_start; arch/sh/boards/mach-migor/setup.c:extern char migor_sdram_enter_end; arch/sh/boards/mach-migor/setup.c:extern char migor_sdram_leave_start; arch/sh/boards/mach-migor/setup.c:extern char migor_sdram_leave_end; arch/sh/boards/mach-se/7724/setup.c:extern char ms7724se_sdram_enter_start; arch/sh/boards/mach-se/7724/setup.c:extern char ms7724se_sdram_enter_end; arch/sh/boards/mach-se/7724/setup.c:extern char ms7724se_sdram_leave_start; arch/sh/boards/mach-se/7724/setup.c:extern char ms7724se_sdram_leave_end; arch/sh/kernel/cpu/shmobile/pm.c:extern char sh_mobile_sleep_enter_start; arch/sh/kernel/cpu/shmobile/pm.c:extern char sh_mobile_sleep_enter_end; arch/sh/kernel/cpu/shmobile/pm.c:extern char sh_mobile_sleep_resume_start; arch/sh/kernel/cpu/shmobile/pm.c:extern char sh_mobile_sleep_resume_end; arch/x86/entry/vsyscall/vsyscall_64.c: extern char __vsyscall_page; arch/x86/include/asm/vvar.h:extern char __vvar_page; kernel/configs.c:extern char kernel_config_data; kernel/configs.c:extern char kernel_config_data_end; net/bpfilter/bpfilter_kern.c:extern char bpfilter_umh_start; net/bpfilter/bpfilter_kern.c:extern char bpfilter_umh_end; samples/bpf/task_fd_query_user.c: extern char __executable_start; tools/testing/selftests/kvm/lib/aarch64/processor.c: extern char vectors; tools/testing/selftests/x86/test_syscall_vdso.c:extern char int80;
Of those, it looks like only a handful might trip FORTIFY:
$ for i in $(git grep 'extern char [^*[( ]*;' | grep -v boot/ | awk -F' ' '{print $NF}' | cut -d';' -f1); do git grep -E '(strcpy|memcpy|memset).*'"$i",; done arch/arm/mach-rockchip/platsmp.c: memcpy_toio(sram_base_addr, &rockchip_secondary_trampoline, trampoline_sz); arch/arm/mach-zynq/platsmp.c: memcpy_toio(zero, &zynq_secondary_trampoline, arch/mips/dec/prom/memory.c: memcpy((void *)(CKSEG0 + 0x80), &genexcept_early, 0x80); arch/sh/kernel/cpu/shmobile/pm.c: memcpy(vp, &sh_mobile_sleep_enter_start, n); arch/sh/kernel/cpu/shmobile/pm.c: memcpy(vp, &sh_mobile_sleep_resume_start, n); arch/arm64/mm/trans_pgd.c: memcpy(hyp_stub, &trans_pgd_stub_vectors, ARM64_VECTOR_TABLE_LEN);
From: Kees Cook keescook@chromium.org Date: Fri, 3 Mar 2023 12:36:09 -0800
On Fri, Mar 03, 2023 at 04:19:51PM +0100, Alexander Lobakin wrote:
From: Kees Cook keescook@chromium.org Date: Thu, 2 Mar 2023 14:49:50 -0800
Under CONFIG_FORTIFY_SOURCE, memcpy() will check the size of destination and source buffers. Defining kernel_headers_data as "char" would trip this check. Since these addresses are treated as byte arrays, define them as arrays (as done everywhere else).
Yet another array-as-one-char, I wonder how many are still here...
Yeah, good point. They do tend to stand out; we could find them:
$ git grep 'extern char [^*[( ]*;' arch/alpha/boot/bootp.c:extern char _end; arch/alpha/boot/bootpz.c:extern char _end; arch/alpha/boot/main.c: extern char _end; arch/arm/mach-rockchip/core.h:extern char rockchip_secondary_trampoline; arch/arm/mach-rockchip/core.h:extern char rockchip_secondary_trampoline_end; arch/arm/mach-zynq/common.h:extern char zynq_secondary_trampoline; arch/arm/mach-zynq/common.h:extern char zynq_secondary_trampoline_jump; arch/arm/mach-zynq/common.h:extern char zynq_secondary_trampoline_end; arch/hexagon/include/uapi/asm/setup.h:extern char external_cmdline_buffer; arch/ia64/include/asm/smp.h:extern char no_int_routing; arch/ia64/kernel/process.c: extern char ia64_ret_from_clone; arch/mips/dec/prom/memory.c: extern char genexcept_early; arch/mips/kernel/traps.c: extern char except_vec3_generic; arch/mips/kernel/traps.c: extern char except_vec4; arch/mips/kernel/traps.c: extern char except_vec3_r4000; arch/mips/mm/c-octeon.c: extern char except_vec2_octeon; arch/parisc/boot/compressed/misc.c:extern char output_len; arch/parisc/boot/compressed/misc.c:extern char _startcode_end; arch/powerpc/include/asm/smp.h:extern char __secondary_hold; arch/s390/include/asm/kvm_host.h:extern char sie_exit; arch/sh/boards/mach-ap325rxa/setup.c:extern char ap325rxa_sdram_enter_start; arch/sh/boards/mach-ap325rxa/setup.c:extern char ap325rxa_sdram_enter_end; arch/sh/boards/mach-ap325rxa/setup.c:extern char ap325rxa_sdram_leave_start; arch/sh/boards/mach-ap325rxa/setup.c:extern char ap325rxa_sdram_leave_end; arch/sh/boards/mach-ecovec24/setup.c:extern char ecovec24_sdram_enter_start; arch/sh/boards/mach-ecovec24/setup.c:extern char ecovec24_sdram_enter_end; arch/sh/boards/mach-ecovec24/setup.c:extern char ecovec24_sdram_leave_start; arch/sh/boards/mach-ecovec24/setup.c:extern char ecovec24_sdram_leave_end; arch/sh/boards/mach-kfr2r09/setup.c:extern char kfr2r09_sdram_enter_start; arch/sh/boards/mach-kfr2r09/setup.c:extern char kfr2r09_sdram_enter_end; arch/sh/boards/mach-kfr2r09/setup.c:extern char kfr2r09_sdram_leave_start; arch/sh/boards/mach-kfr2r09/setup.c:extern char kfr2r09_sdram_leave_end; arch/sh/boards/mach-migor/setup.c:extern char migor_sdram_enter_start; arch/sh/boards/mach-migor/setup.c:extern char migor_sdram_enter_end; arch/sh/boards/mach-migor/setup.c:extern char migor_sdram_leave_start; arch/sh/boards/mach-migor/setup.c:extern char migor_sdram_leave_end; arch/sh/boards/mach-se/7724/setup.c:extern char ms7724se_sdram_enter_start; arch/sh/boards/mach-se/7724/setup.c:extern char ms7724se_sdram_enter_end; arch/sh/boards/mach-se/7724/setup.c:extern char ms7724se_sdram_leave_start; arch/sh/boards/mach-se/7724/setup.c:extern char ms7724se_sdram_leave_end; arch/sh/kernel/cpu/shmobile/pm.c:extern char sh_mobile_sleep_enter_start; arch/sh/kernel/cpu/shmobile/pm.c:extern char sh_mobile_sleep_enter_end; arch/sh/kernel/cpu/shmobile/pm.c:extern char sh_mobile_sleep_resume_start; arch/sh/kernel/cpu/shmobile/pm.c:extern char sh_mobile_sleep_resume_end; arch/x86/entry/vsyscall/vsyscall_64.c: extern char __vsyscall_page; arch/x86/include/asm/vvar.h:extern char __vvar_page; kernel/configs.c:extern char kernel_config_data; kernel/configs.c:extern char kernel_config_data_end; net/bpfilter/bpfilter_kern.c:extern char bpfilter_umh_start; net/bpfilter/bpfilter_kern.c:extern char bpfilter_umh_end; samples/bpf/task_fd_query_user.c: extern char __executable_start; tools/testing/selftests/kvm/lib/aarch64/processor.c: extern char vectors; tools/testing/selftests/x86/test_syscall_vdso.c:extern char int80;
Of those, it looks like only a handful might trip FORTIFY:
$ for i in $(git grep 'extern char [^*[( ]*;' | grep -v boot/ | awk -F' ' '{print $NF}' | cut -d';' -f1); do git grep -E '(strcpy|memcpy|memset).*'"$i",; done arch/arm/mach-rockchip/platsmp.c: memcpy_toio(sram_base_addr, &rockchip_secondary_trampoline, trampoline_sz); arch/arm/mach-zynq/platsmp.c: memcpy_toio(zero, &zynq_secondary_trampoline, arch/mips/dec/prom/memory.c: memcpy((void *)(CKSEG0 + 0x80), &genexcept_early, 0x80); arch/sh/kernel/cpu/shmobile/pm.c: memcpy(vp, &sh_mobile_sleep_enter_start, n); arch/sh/kernel/cpu/shmobile/pm.c: memcpy(vp, &sh_mobile_sleep_resume_start, n); arch/arm64/mm/trans_pgd.c: memcpy(hyp_stub, &trans_pgd_stub_vectors, ARM64_VECTOR_TABLE_LEN);
This list looks much better than the first one :D
I remember I caught a couple fortify bugs due to such array declaration on MIPS only when testing with `KCFLAGS=-O3`. The thing was that a pointer to such array-not-array was passed as a function argument and that function wasn't inlined on -O2. So I sometimes use this optlevel for sanity-checking the code :P
Thanks, Olek
linux-stable-mirror@lists.linaro.org
-
Alexander Lobakin -
Joel Fernandes -
Kees Cook