Similar to previous patch that drops signed_add*_overflows() and uses
(compiler) builtin-based check_add_overflow(), do the same for
signed_sub*_overflows() and replace them with the generic
check_sub_overflow() to make future refactoring easier and have the
checks implemented more efficiently.
Unsigned overflow check for subtraction does not use helpers and are
simple enough already, so they're left untouched.
After the change GCC 13.3.0 generates cleaner assembly on x86_64:
if (check_sub_overflow(*dst_smin, src_reg->smax_value, dst_smin) ||
139bf: mov 0x28(%r12),%rax
139c4: mov %edx,0x54(%r12)
139c9: sub %r11,%rax
139cc: mov %rax,0x28(%r12)
139d1: jo 14627 <adjust_reg_min_max_vals+0x1237>
check_sub_overflow(*dst_smax, src_reg->smin_value, dst_smax)) {
139d7: mov 0x30(%r12),%rax
139dc: sub %r9,%rax
139df: mov %rax,0x30(%r12)
if (check_sub_overflow(*dst_smin, src_reg->smax_value, dst_smin) ||
139e4: jo 14627 <adjust_reg_min_max_vals+0x1237>
...
*dst_smin = S64_MIN;
14627: movabs $0x8000000000000000,%rax
14631: mov %rax,0x28(%r12)
*dst_smax = S64_MAX;
14636: sub $0x1,%rax
1463a: mov %rax,0x30(%r12)
Before the change it gives:
if (signed_sub_overflows(dst_reg->smin_value, smax_val) ||
13a50: mov 0x28(%r12),%rdi
13a55: mov %edx,0x54(%r12)
dst_reg->smax_value = S64_MAX;
13a5a: movabs $0x7fffffffffffffff,%rdx
13a64: mov %eax,0x50(%r12)
dst_reg->smin_value = S64_MIN;
13a69: movabs $0x8000000000000000,%rax
s64 res = (s64)((u64)a - (u64)b);
13a73: mov %rdi,%rsi
13a76: sub %rcx,%rsi
if (b < 0)
13a79: test %rcx,%rcx
13a7c: js 145ea <adjust_reg_min_max_vals+0x119a>
if (signed_sub_overflows(dst_reg->smin_value, smax_val) ||
13a82: cmp %rsi,%rdi
13a85: jl 13ac7 <adjust_reg_min_max_vals+0x677>
signed_sub_overflows(dst_reg->smax_value, smin_val)) {
13a87: mov 0x30(%r12),%r8
s64 res = (s64)((u64)a - (u64)b);
13a8c: mov %r8,%rax
13a8f: sub %r9,%rax
return res > a;
13a92: cmp %rax,%r8
13a95: setl %sil
if (b < 0)
13a99: test %r9,%r9
13a9c: js 147d1 <adjust_reg_min_max_vals+0x1381>
dst_reg->smax_value = S64_MAX;
13aa2: movabs $0x7fffffffffffffff,%rdx
dst_reg->smin_value = S64_MIN;
13aac: movabs $0x8000000000000000,%rax
if (signed_sub_overflows(dst_reg->smin_value, smax_val) ||
13ab6: test %sil,%sil
13ab9: jne 13ac7 <adjust_reg_min_max_vals+0x677>
dst_reg->smin_value -= smax_val;
13abb: mov %rdi,%rax
dst_reg->smax_value -= smin_val;
13abe: mov %r8,%rdx
dst_reg->smin_value -= smax_val;
13ac1: sub %rcx,%rax
dst_reg->smax_value -= smin_val;
13ac4: sub %r9,%rdx
13ac7: mov %rax,0x28(%r12)
...
13ad1: mov %rdx,0x30(%r12)
...
if (signed_sub_overflows(dst_reg->smin_value, smax_val) ||
145ea: cmp %rsi,%rdi
145ed: jg 13ac7 <adjust_reg_min_max_vals+0x677>
145f3: jmp 13a87 <adjust_reg_min_max_vals+0x637>
Suggested-by: Jiri Olsa <jolsa@xxxxxxxxxx>
Signed-off-by: Shung-Hsi Yu <shung-hsi.yu@xxxxxxxx>
---
Some bike shedding. For aesthetic symmetry, we can also change unsigned
overflow check for subtraction to
if (check_sub_overflow(*dst_umin, src_reg->umax_value, dst_umin) ||
check_sub_overflow(*dst_umax, src_reg->umin_value, dst_umax)) {
*dst_umin = 0;
*dst_umax = U64_MAX;
}
The 2nd check_sub_overflow(*dst_umax, src_reg->umin_value, dst_umax) is
redundant (though likely don't matter much in terms of performance), but
without that check we technically can't use the value at the dst_umax
pointer. So that, or
*dst_umax -= src_reg->src_reg->umin_value;
/* if dst_umin does not overflow we know that dst_umax won't either */
if (check_sub_overflow(*dst_umin, src_reg->umax_value, dst_umin)) {
*dst_umin = 0;
*dst_umax = U64_MAX;
}
OTOH current unsigned subtraction check already gives pretty clean
assembly right now (see below), so in terms of assembly I don't think
using check_sub_overflow would make much of a difference.
if (dst_reg->umin_value < umax_val) {
139ea: mov 0x38(%r12),%rax
139ef: cmp %r10,%rax
139f2: jb 14247 <adjust_reg_min_max_vals+0xe57>
dst_reg->umax_value -= umin_val;
139f8: mov 0x40(%r12),%rdx
139fd: mov 0x30(%rsp),%rcx
dst_reg->umin_value -= umax_val;
13a02: sub %r10,%rax
dst_reg->umax_value -= umin_val;
13a05: sub %rcx,%rdx
...
13a12: mov %rdx,0x40(%r12)
13a17: mov %rax,0x38(%r12)
...
dst_reg->umax_value = U64_MAX;
14247: mov $0xffffffffffffffff,%rdx
dst_reg->umin_value = 0;
1424e: xor %eax,%eax
---
kernel/bpf/verifier.c | 57 +++++++++++--------------------------------
1 file changed, 14 insertions(+), 43 deletions(-)
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 26c2b7527942..8417f6187961 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -12725,26 +12725,6 @@ static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
return 0;
}
-static bool signed_sub_overflows(s64 a, s64 b)
-{
- /* Do the sub in u64, where overflow is well-defined */
- s64 res = (s64)((u64)a - (u64)b);
-
- if (b < 0)
- return res < a;
- return res > a;
-}
-
-static bool signed_sub32_overflows(s32 a, s32 b)
-{
- /* Do the sub in u32, where overflow is well-defined */
- s32 res = (s32)((u32)a - (u32)b);
-
- if (b < 0)
- return res < a;
- return res > a;
-}
-
static bool check_reg_sane_offset(struct bpf_verifier_env *env,
const struct bpf_reg_state *reg,
enum bpf_reg_type type)
@@ -13277,14 +13257,11 @@ static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env,
/* A new variable offset is created. If the subtrahend is known
* nonnegative, then any reg->range we had before is still good.
*/
- if (signed_sub_overflows(smin_ptr, smax_val) ||
- signed_sub_overflows(smax_ptr, smin_val)) {
+ if (check_sub_overflow(smin_ptr, smax_val, &dst_reg->smin_value) ||
+ check_sub_overflow(smax_ptr, smin_val, &dst_reg->smax_value)) {
/* Overflow possible, we know nothing */
dst_reg->smin_value = S64_MIN;
dst_reg->smax_value = S64_MAX;
- } else {
- dst_reg->smin_value = smin_ptr - smax_val;
- dst_reg->smax_value = smax_ptr - smin_val;
}
if (umin_ptr < umax_val) {
/* Overflow possible, we know nothing */
@@ -13377,19 +13354,16 @@ static void scalar_min_max_add(struct bpf_reg_state *dst_reg,
static void scalar32_min_max_sub(struct bpf_reg_state *dst_reg,
struct bpf_reg_state *src_reg)
{
- s32 smin_val = src_reg->s32_min_value;
- s32 smax_val = src_reg->s32_max_value;
+ s32 *dst_smin = &dst_reg->s32_min_value;
+ s32 *dst_smax = &dst_reg->s32_max_value;
u32 umin_val = src_reg->u32_min_value;
u32 umax_val = src_reg->u32_max_value;
- if (signed_sub32_overflows(dst_reg->s32_min_value, smax_val) ||
- signed_sub32_overflows(dst_reg->s32_max_value, smin_val)) {
+ if (check_sub_overflow(*dst_smin, src_reg->s32_max_value, dst_smin) ||
+ check_sub_overflow(*dst_smax, src_reg->s32_min_value, dst_smax)) {
/* Overflow possible, we know nothing */
- dst_reg->s32_min_value = S32_MIN;
- dst_reg->s32_max_value = S32_MAX;
- } else {
- dst_reg->s32_min_value -= smax_val;
- dst_reg->s32_max_value -= smin_val;
+ *dst_smin = S32_MIN;
+ *dst_smax = S32_MAX;
}
if (dst_reg->u32_min_value < umax_val) {
/* Overflow possible, we know nothing */
@@ -13405,19 +13379,16 @@ static void scalar32_min_max_sub(struct bpf_reg_state *dst_reg,
static void scalar_min_max_sub(struct bpf_reg_state *dst_reg,
struct bpf_reg_state *src_reg)
{
- s64 smin_val = src_reg->smin_value;
- s64 smax_val = src_reg->smax_value;
+ s64 *dst_smin = &dst_reg->smin_value;
+ s64 *dst_smax = &dst_reg->smax_value;
u64 umin_val = src_reg->umin_value;
u64 umax_val = src_reg->umax_value;
- if (signed_sub_overflows(dst_reg->smin_value, smax_val) ||
- signed_sub_overflows(dst_reg->smax_value, smin_val)) {
+ if (check_sub_overflow(*dst_smin, src_reg->smax_value, dst_smin) ||
+ check_sub_overflow(*dst_smax, src_reg->smin_value, dst_smax)) {
/* Overflow possible, we know nothing */
- dst_reg->smin_value = S64_MIN;
- dst_reg->smax_value = S64_MAX;
- } else {
- dst_reg->smin_value -= smax_val;
- dst_reg->smax_value -= smin_val;
+ *dst_smin = S64_MIN;
+ *dst_smax = S64_MAX;
}
if (dst_reg->umin_value < umax_val) {
/* Overflow possible, we know nothing */
--
2.45.2
