bpf: Fix a sdiv overflow issue
Subsystems (4)
Yonghong Song
reviewerPATCH v3 1/2 bpf: Fix a sdiv overflow issue
Zac Ecob reported a problem where a bpf program may cause kernel crash due to the following error:
Oops: divide error: 0000 [#1] PREEMPT SMP KASAN PTI
The failure is due to the below signed divide:
LLONG_MIN/-1 where LLONG_MIN equals to -9,223,372,036,854,775,808.
LLONG_MIN/-1 is supposed to give a positive number 9,223,372,036,854,775,808, but it is impossible since for 64-bit system, the maximum positive number is 9,223,372,036,854,775,807. On x86_64, LLONG_MIN/-1 will cause a kernel exception. On arm64, the result for LLONG_MIN/-1 is LLONG_MIN.
Further investigation found all the following sdiv/smod cases may trigger an exception when bpf program is running on x86_64 platform:
- LLONG_MIN/-1 for 64bit operation
- INT_MIN/-1 for 32bit operation
- LLONG_MIN%-1 for 64bit operation
- INT_MIN%-1 for 32bit operation
where -1 can be an immediate or in a register.
On arm64, there are no exceptions:
- LLONG_MIN/-1 = LLONG_MIN
- INT_MIN/-1 = INT_MIN
- LLONG_MIN%-1 = 0
- INT_MIN%-1 = 0
where -1 can be an immediate or in a register.
Insn patching is needed to handle the above cases and the patched codes produced results aligned with above arm64 result. The below are pseudo codes to handle sdiv/smod exceptions including both divisor -1 and divisor 0 and the divisor is stored in a register.
sdiv:
tmp = rX
tmp += 1 /* [-1, 0] -> [0, 1]
if tmp >(unsigned) 1 goto L2
if tmp == 0 goto L1
rY = 0
L1:
rY = -rY;
goto L3 L2:
rY /= rX
L3:
smod:
tmp = rX
tmp += 1 /* [-1, 0] -> [0, 1]
if tmp >(unsigned) 1 goto L1
if tmp == 1 (is64 ? goto L2 : goto L3)
rY = 0;
goto L2 L1:
rY %= rX
L2:
goto L4 // only when !is64
L3:
wY = wY // only when !is64
L4:
--- kernel/bpf/verifier.c | 93 +++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 89 insertions(+), 4 deletions(-)
Changelogs:
v2 -> v3:
- Change sdiv/smod (r/r, r%r) patched insn to be more efficient for default case.
v1 -> v2:
- Handle more crash cases like 32bit operation and modules.
- Add more tests to test new cases.
kernel/bpf/verifier.c +89 -4
@@ -20499,13 +20499,46 @@ static int do_misc_fixups(struct bpf_verifier_env *env)/* Convert BPF_CLASS(insn->code) == BPF_ALU64 to 32-bit ALU */ insn->code =BPF_ALU |BPF_OP (insn->code) |BPF_SRC (insn->code); -/* Make divide-by-zero exceptions impossible. */ +/* Make sdiv/smod divide-by-minus-one exceptions impossible. */ +if ((insn->code == (BPF_ALU64 |BPF_MOD |BPF_K ) || + insn->code == (BPF_ALU64 |BPF_DIV |BPF_K ) || + insn->code == (BPF_ALU |BPF_MOD |BPF_K ) || + insn->code == (BPF_ALU |BPF_DIV |BPF_K )) && + insn->off ==1 && insn->imm ==-1 ) { +bool is64 =BPF_CLASS (insn->code) ==BPF_ALU64 ; +bool isdiv =BPF_OP (insn->code) ==BPF_DIV ; +struct bpf_insn *patchlet; +struct bpf_insn chk_and_sdiv[] = { +BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_NEG ) |BPF_K , insn->dst_reg, +0 ,0 ,0 ), + }; +struct bpf_insn chk_and_smod[] = { +BPF_MOV32_IMM (insn->dst_reg,0 ), + }; + + patchlet = isdiv ? chk_and_sdiv : chk_and_smod; + cnt = isdiv ?ARRAY_SIZE (chk_and_sdiv) :ARRAY_SIZE (chk_and_smod); + + new_prog =bpf_patch_insn_data (env, i + delta, patchlet, cnt); +if (!new_prog) +return -ENOMEM ; + + delta += cnt -1 ; + env->prog = prog = new_prog; + insn = new_prog->insnsi + i + delta; + goto next_insn; + } + +/* Make divide-by-zero and divide-by-minus-one exceptions impossible. */ if (insn->code == (BPF_ALU64 |BPF_MOD |BPF_X ) || insn->code == (BPF_ALU64 |BPF_DIV |BPF_X ) || insn->code == (BPF_ALU |BPF_MOD |BPF_X ) || insn->code == (BPF_ALU |BPF_DIV |BPF_X )) {bool is64 =BPF_CLASS (insn->code) ==BPF_ALU64 ;bool isdiv =BPF_OP (insn->code) ==BPF_DIV ; +bool is_sdiv = isdiv && insn->off ==1 ; +bool is_smod = !isdiv && insn->off ==1 ;struct bpf_insn *patchlet;struct bpf_insn chk_and_div[] = {/* [R,W]x div 0 -> 0 */
@@ -20525,10 +20558,62 @@ static int do_misc_fixups(struct bpf_verifier_env *env)BPF_JMP_IMM (BPF_JA ,0 ,0 ,1 ), BPF_MOV32_REG(insn->dst_reg, insn->dst_reg), }; +struct bpf_insn chk_and_sdiv[] = { +/* [R,W]x sdiv 0 -> 0 +* LLONG_MIN sdiv -1 -> LLONG_MIN +* INT_MIN sdiv -1 -> INT_MIN +*/ +BPF_MOV64_REG (BPF_REG_AX , insn->src_reg), +BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_ADD ) |BPF_K ,BPF_REG_AX , +0 ,0 ,1 ), +BPF_RAW_INSN ((is64 ?BPF_JMP :BPF_JMP32 ) | +BPF_JGT |BPF_K ,BPF_REG_AX , +0 ,4 ,1 ), +BPF_RAW_INSN ((is64 ?BPF_JMP :BPF_JMP32 ) | +BPF_JEQ |BPF_K ,BPF_REG_AX , +0 ,1 ,0 ), +BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_MOV ) |BPF_K , insn->dst_reg, +0 ,0 ,0 ), +/* BPF_NEG(LLONG_MIN) == -LLONG_MIN == LLONG_MIN */ +BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_NEG ) |BPF_K , insn->dst_reg, +0 ,0 ,0 ), +BPF_JMP_IMM (BPF_JA ,0 ,0 ,1 ), + *insn, + }; +struct bpf_insn chk_and_smod[] = { +/* [R,W]x mod 0 -> [R,W]x */ +/* [R,W]x mod -1 -> 0 */ +BPF_MOV64_REG (BPF_REG_AX , insn->src_reg), +BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_ADD ) |BPF_K ,BPF_REG_AX , +0 ,0 ,1 ), +BPF_RAW_INSN ((is64 ?BPF_JMP :BPF_JMP32 ) | +BPF_JGT |BPF_K ,BPF_REG_AX , +0 ,3 ,1 ), +BPF_RAW_INSN ((is64 ?BPF_JMP :BPF_JMP32 ) | +BPF_JEQ |BPF_K ,BPF_REG_AX , +0 ,3 + (is64 ?0 :1 ),1 ), +BPF_MOV32_IMM (insn->dst_reg,0 ), +BPF_JMP_IMM (BPF_JA ,0 ,0 ,1 ), + *insn, +BPF_JMP_IMM (BPF_JA ,0 ,0 ,1 ), +BPF_MOV32_REG (insn->dst_reg, insn->dst_reg), + }; - patchlet = isdiv ? chk_and_div : chk_and_mod; - cnt = isdiv ?ARRAY_SIZE (chk_and_div) : -ARRAY_SIZE (chk_and_mod) - (is64 ?2 :0 ); +if (is_sdiv) { + patchlet = chk_and_sdiv; + cnt =ARRAY_SIZE (chk_and_sdiv); + }else if (is_smod) { + patchlet = chk_and_smod; + cnt =ARRAY_SIZE (chk_and_smod) - (is64 ?2 :0 ); + }else { + patchlet = isdiv ? chk_and_div : chk_and_mod; + cnt = isdiv ?ARRAY_SIZE (chk_and_div) : +ARRAY_SIZE (chk_and_mod) - (is64 ?2 :0 ); + } new_prog =bpf_patch_insn_data (env, i + delta, patchlet, cnt);if (!new_prog)
Signature · 1 line
2.43.5
Yonghong Song
reviewerPATCH v3 2/2 selftests/bpf: Add tests for sdiv/smod overflow cases
Subtests are added to exercise the patched code which handles
- LLONG_MIN/-1
- INT_MIN/-1
- LLONG_MIN%-1
- INT_MIN%-1
where -1 could be an immediate or in a register.
Without the previous patch, all these cases will crash the kernel on x86_64 platform.
Additional tests are added to use small values (e.g. -5/-1, 5%-1, etc.) in order to exercise the additional logic with patched insns.
--- .../selftests/bpf/progs/verifier_sdiv.c | 439 ++++++++++++++++++ 1 file changed, 439 insertions(+)
tools/testing/selftests/bpf/progs/verifier_sdiv.c +439 -0
@@ -1,6 +1,7 @@// SPDX-License-Identifier: GPL-2.0 #include <linux/bpf.h> +#include <limits.h> #include <bpf/bpf_helpers.h> #include "bpf_misc.h"
@@ -770,6 +771,444 @@ __naked void smod64_zero_divisor(void) " ::: __clobber_all); } +SEC("socket") +__description("SDIV64, overflow r/r, LLONG_MIN/-1") +__success __retval(1) +__arch_x86_64 +__xlated("0: r2 = 0x8000000000000000") +__xlated("2: r3 = -1") +__xlated("3: r4 = r2") +__xlated("4: r11 = r3") +__xlated("5: r11 += 1") +__xlated("6: if r11 > 0x1 goto pc+4") +__xlated("7: if r11 == 0x0 goto pc+1") +__xlated("8: r2 = 0") +__xlated("9: r2 = -r2") +__xlated("10: goto pc+1") +__xlated("11: r2 s/= r3") +__xlated("12: r0 = 0") +__xlated("13: if r2 != r4 goto pc+1") +__xlated("14: r0 = 1") +__xlated("15: exit") +__naked void sdiv64_overflow_rr (void ) +{ + asmvolatile (" \ +r2 = %[llong_min] ll; \ +r3 = -1; \ +r4 = r2; \ +r2 s/= r3; \ +r0 = 0; \ +if r2 != r4 goto +1; \ +r0 = 1; \ +exit; \ +" : + : __imm_const(llong_min, LLONG_MIN) + : __clobber_all); +} + +SEC("socket") +__description("SDIV64, r/r, small_val/-1") +__success __retval(-5) +__arch_x86_64 +__xlated("0: r2 = 5") +__xlated("1: r3 = -1") +__xlated("2: r11 = r3") +__xlated("3: r11 += 1") +__xlated("4: if r11 > 0x1 goto pc+4") +__xlated("5: if r11 == 0x0 goto pc+1") +__xlated("6: r2 = 0") +__xlated("7: r2 = -r2") +__xlated("8: goto pc+1") +__xlated("9: r2 s/= r3") +__xlated("10: r0 = r2") +__xlated("11: exit") +__naked void sdiv64_rr_divisor_neg_1 (void ) +{ + asmvolatile (" \ +r2 = 5; \ +r3 = -1; \ +r2 s/= r3; \ +r0 = r2; \ +exit; \ +" : + : + : __clobber_all); +} + +SEC("socket") +__description("SDIV64, overflow r/i, LLONG_MIN/-1") +__success __retval(1) +__arch_x86_64 +__xlated("0: r2 = 0x8000000000000000") +__xlated("2: r4 = r2") +__xlated("3: r2 = -r2") +__xlated("4: r0 = 0") +__xlated("5: if r2 != r4 goto pc+1") +__xlated("6: r0 = 1") +__xlated("7: exit") +__naked void sdiv64_overflow_ri (void ) +{ + asmvolatile (" \ +r2 = %[llong_min] ll; \ +r4 = r2; \ +r2 s/= -1; \ +r0 = 0; \ +if r2 != r4 goto +1; \ +r0 = 1; \ +exit; \ +" : + : __imm_const(llong_min, LLONG_MIN) + : __clobber_all); +} + +SEC("socket") +__description("SDIV64, r/i, small_val/-1") +__success __retval(-5) +__arch_x86_64 +__xlated("0: r2 = 5") +__xlated("1: r4 = r2") +__xlated("2: r2 = -r2") +__xlated("3: r0 = r2") +__xlated("4: exit") +__naked void sdiv64_ri_divisor_neg_1 (void ) +{ + asmvolatile (" \ +r2 = 5; \ +r4 = r2; \ +r2 s/= -1; \ +r0 = r2; \ +exit; \ +" : + : + : __clobber_all); +} + +SEC("socket") +__description("SDIV32, overflow r/r, INT_MIN/-1") +__success __retval(1) +__arch_x86_64 +__xlated("0: w2 = -2147483648") +__xlated("1: w3 = -1") +__xlated("2: w4 = w2") +__xlated("3: r11 = r3") +__xlated("4: w11 += 1") +__xlated("5: if w11 > 0x1 goto pc+4") +__xlated("6: if w11 == 0x0 goto pc+1") +__xlated("7: w2 = 0") +__xlated("8: w2 = -w2") +__xlated("9: goto pc+1") +__xlated("10: w2 s/= w3") +__xlated("11: r0 = 0") +__xlated("12: if w2 != w4 goto pc+1") +__xlated("13: r0 = 1") +__xlated("14: exit") +__naked void sdiv32_overflow_rr (void ) +{ + asmvolatile (" \ +w2 = %[int_min]; \ +w3 = -1; \ +w4 = w2; \ +w2 s/= w3; \ +r0 = 0; \ +if w2 != w4 goto +1; \ +r0 = 1; \ +exit; \ +" : + : __imm_const(int_min, INT_MIN) + : __clobber_all); +} + +SEC("socket") +__description("SDIV32, r/r, small_val/-1") +__success __retval(5) +__arch_x86_64 +__xlated("0: w2 = -5") +__xlated("1: w3 = -1") +__xlated("2: w4 = w2") +__xlated("3: r11 = r3") +__xlated("4: w11 += 1") +__xlated("5: if w11 > 0x1 goto pc+4") +__xlated("6: if w11 == 0x0 goto pc+1") +__xlated("7: w2 = 0") +__xlated("8: w2 = -w2") +__xlated("9: goto pc+1") +__xlated("10: w2 s/= w3") +__xlated("11: w0 = w2") +__xlated("12: exit") +__naked void sdiv32_rr_divisor_neg_1 (void ) +{ + asmvolatile (" \ +w2 = -5; \ +w3 = -1; \ +w4 = w2; \ +w2 s/= w3; \ +w0 = w2; \ +exit; \ +" : + : + : __clobber_all); +} + +SEC("socket") +__description("SDIV32, overflow r/i, INT_MIN/-1") +__success __retval(1) +__arch_x86_64 +__xlated("0: w2 = -2147483648") +__xlated("1: w4 = w2") +__xlated("2: w2 = -w2") +__xlated("3: r0 = 0") +__xlated("4: if w2 != w4 goto pc+1") +__xlated("5: r0 = 1") +__xlated("6: exit") +__naked void sdiv32_overflow_ri (void ) +{ + asmvolatile (" \ +w2 = %[int_min]; \ +w4 = w2; \ +w2 s/= -1; \ +r0 = 0; \ +if w2 != w4 goto +1; \ +r0 = 1; \ +exit; \ +" : + : __imm_const(int_min, INT_MIN) + : __clobber_all); +} + +SEC("socket") +__description("SDIV32, r/i, small_val/-1") +__success __retval(-5) +__arch_x86_64 +__xlated("0: w2 = 5") +__xlated("1: w4 = w2") +__xlated("2: w2 = -w2") +__xlated("3: w0 = w2") +__xlated("4: exit") +__naked void sdiv32_ri_divisor_neg_1 (void ) +{ + asmvolatile (" \ +w2 = 5; \ +w4 = w2; \ +w2 s/= -1; \ +w0 = w2; \ +exit; \ +" : + : + : __clobber_all); +} + +SEC("socket") +__description("SMOD64, overflow r/r, LLONG_MIN/-1") +__success __retval(0) +__arch_x86_64 +__xlated("0: r2 = 0x8000000000000000") +__xlated("2: r3 = -1") +__xlated("3: r4 = r2") +__xlated("4: r11 = r3") +__xlated("5: r11 += 1") +__xlated("6: if r11 > 0x1 goto pc+3") +__xlated("7: if r11 == 0x1 goto pc+3") +__xlated("8: w2 = 0") +__xlated("9: goto pc+1") +__xlated("10: r2 s%= r3") +__xlated("11: r0 = r2") +__xlated("12: exit") +__naked void smod64_overflow_rr (void ) +{ + asmvolatile (" \ +r2 = %[llong_min] ll; \ +r3 = -1; \ +r4 = r2; \ +r2 s%%= r3; \ +r0 = r2; \ +exit; \ +" : + : __imm_const(llong_min, LLONG_MIN) + : __clobber_all); +} + +SEC("socket") +__description("SMOD64, r/r, small_val/-1") +__success __retval(0) +__arch_x86_64 +__xlated("0: r2 = 5") +__xlated("1: r3 = -1") +__xlated("2: r4 = r2") +__xlated("3: r11 = r3") +__xlated("4: r11 += 1") +__xlated("5: if r11 > 0x1 goto pc+3") +__xlated("6: if r11 == 0x1 goto pc+3") +__xlated("7: w2 = 0") +__xlated("8: goto pc+1") +__xlated("9: r2 s%= r3") +__xlated("10: r0 = r2") +__xlated("11: exit") +__naked void smod64_rr_divisor_neg_1 (void ) +{ + asmvolatile (" \ +r2 = 5; \ +r3 = -1; \ +r4 = r2; \ +r2 s%%= r3; \ +r0 = r2; \ +exit; \ +" : + : + : __clobber_all); +} + +SEC("socket") +__description("SMOD64, overflow r/i, LLONG_MIN/-1") +__success __retval(0) +__arch_x86_64 +__xlated("0: r2 = 0x8000000000000000") +__xlated("2: r4 = r2") +__xlated("3: w2 = 0") +__xlated("4: r0 = r2") +__xlated("5: exit") +__naked void smod64_overflow_ri (void ) +{ + asmvolatile (" \ +r2 = %[llong_min] ll; \ +r4 = r2; \ +r2 s%%= -1; \ +r0 = r2; \ +exit; \ +" : + : __imm_const(llong_min, LLONG_MIN) + : __clobber_all); +} + +SEC("socket") +__description("SMOD64, r/i, small_val/-1") +__success __retval(0) +__arch_x86_64 +__xlated("0: r2 = 5") +__xlated("1: r4 = r2") +__xlated("2: w2 = 0") +__xlated("3: r0 = r2") +__xlated("4: exit") +__naked void smod64_ri_divisor_neg_1 (void ) +{ + asmvolatile (" \ +r2 = 5; \ +r4 = r2; \ +r2 s%%= -1; \ +r0 = r2; \ +exit; \ +" : + : + : __clobber_all); +} + +SEC("socket") +__description("SMOD32, overflow r/r, INT_MIN/-1") +__success __retval(0) +__arch_x86_64 +__xlated("0: w2 = -2147483648") +__xlated("1: w3 = -1") +__xlated("2: w4 = w2") +__xlated("3: r11 = r3") +__xlated("4: w11 += 1") +__xlated("5: if w11 > 0x1 goto pc+3") +__xlated("6: if w11 == 0x1 goto pc+4") +__xlated("7: w2 = 0") +__xlated("8: goto pc+1") +__xlated("9: w2 s%= w3") +__xlated("10: goto pc+1") +__xlated("11: w2 = w2") +__xlated("12: r0 = r2") +__xlated("13: exit") +__naked void smod32_overflow_rr (void ) +{ + asmvolatile (" \ +w2 = %[int_min]; \ +w3 = -1; \ +w4 = w2; \ +w2 s%%= w3; \ +r0 = r2; \ +exit; \ +" : + : __imm_const(int_min, INT_MIN) + : __clobber_all); +} + +SEC("socket") +__description("SMOD32, r/r, small_val/-1") +__success __retval(0) +__arch_x86_64 +__xlated("0: w2 = -5") +__xlated("1: w3 = -1") +__xlated("2: w4 = w2") +__xlated("3: r11 = r3") +__xlated("4: w11 += 1") +__xlated("5: if w11 > 0x1 goto pc+3") +__xlated("6: if w11 == 0x1 goto pc+4") +__xlated("7: w2 = 0") +__xlated("8: goto pc+1") +__xlated("9: w2 s%= w3") +__xlated("10: goto pc+1") +__xlated("11: w2 = w2") +__xlated("12: r0 = r2") +__xlated("13: exit") +__naked void smod32_rr_divisor_neg_1 (void ) +{ + asmvolatile (" \ +w2 = -5; \ +w3 = -1; \ +w4 = w2; \ +w2 s%%= w3; \ +r0 = r2; \ +exit; \ +" : + : + : __clobber_all); +} + +SEC("socket") +__description("SMOD32, overflow r/i, INT_MIN/-1") +__success __retval(0) +__arch_x86_64 +__xlated("0: w2 = -2147483648") +__xlated("1: w4 = w2") +__xlated("2: w2 = 0") +__xlated("3: r0 = r2") +__xlated("4: exit") +__naked void smod32_overflow_ri (void ) +{ + asmvolatile (" \ +w2 = %[int_min]; \ +w4 = w2; \ +w2 s%%= -1; \ +r0 = r2; \ +exit; \ +" : + : __imm_const(int_min, INT_MIN) + : __clobber_all); +} + +SEC("socket") +__description("SMOD32, r/i, small_val/-1") +__success __retval(0) +__arch_x86_64 +__xlated("0: w2 = 5") +__xlated("1: w4 = w2") +__xlated("2: w2 = 0") +__xlated("3: w0 = w2") +__xlated("4: exit") +__naked void smod32_ri_divisor_neg_1 (void ) +{ + asmvolatile (" \ +w2 = 5; \ +w4 = w2; \ +w2 s%%= -1; \ +w0 = w2; \ +exit; \ +" : + : + : __clobber_all); +} +#else SEC ("socket" )
Signature · 1 line
2.43.5
Quoted 182 lines · Yonghong Song
Zac Ecob reported a problem where a bpf program may cause kernel crash due to the following error:
Oops: divide error: 0000 [#1] PREEMPT SMP KASAN PTIThe failure is due to the below signed divide:
LLONG_MIN/-1 whereLLONG_MINequals to -9,223,372,036,854,775,808.
LLONG_MIN/-1 is supposed to give a positive number 9,223,372,036,854,775,808, but it is impossible since for 64-bit system, the maximum positive number is 9,223,372,036,854,775,807. Onx86_64, LLONG_MIN/-1 will cause a kernel exception. On arm64, the result for LLONG_MIN/-1 isLLONG_MIN.Further investigation found all the following sdiv/smod cases may trigger an exception when bpf program is running on
x86_64platform:
- LLONG_MIN/-1 for 64bit operation
- INT_MIN/-1 for 32bit operation
- LLONG_MIN%-1 for 64bit operation
- INT_MIN%-1 for 32bit operation
where -1 can be an immediate or in a register.On arm64, there are no exceptions:
- LLONG_MIN/-1 =LLONG_MIN
- INT_MIN/-1 =INT_MIN
- LLONG_MIN%-1 = 0
- INT_MIN%-1 = 0
where -1 can be an immediate or in a register.Insn patching is needed to handle the above cases and the patched codes produced results aligned with above arm64 result. The below are pseudo codes to handle sdiv/smod exceptions including both divisor -1 and divisor 0 and the divisor is stored in a register.
sdiv:
tmp = rX tmp += 1 /* [-1, 0] -> [0, 1] if tmp >(unsigned) 1 goto L2 if tmp == 0 goto L1 rY = 0 L1: rY = -rY; goto L3L2:
rY /= rX
L3:smod:
tmp = rX tmp += 1 /* [-1, 0] -> [0, 1] if tmp >(unsigned) 1 goto L1 if tmp == 1 (is64 ? goto L2 : goto L3) rY = 0; goto L2L1:
rY %= rX
L2:
goto L4 // only when !is64
L3:
wY = wY // only when !is64
L4:Reported-by Zac Ecob <[email protected]>Signed-off-by Yonghong Song <[email protected]>--- kernel/bpf/verifier.c | 93 +++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 89 insertions(+), 4 deletions(-)Changelogs:
v2 -> v3:
- Change sdiv/smod (r/r, r%r) patched insn to be more efficient for default case.
v1 -> v2:
- Handle more crash cases like 32bit operation and modules.
- Add more tests to test new cases.kernel/bpf/verifier.c +79
-4@@ -20499,13 +20499,46 @@ static int do_misc_fixups(struct bpf_verifier_env *env)/* Convert BPF_CLASS(insn->code) == BPF_ALU64 to 32-bit ALU */ insn->code =BPF_ALU |BPF_OP (insn->code) |BPF_SRC (insn->code); -/* Make divide-by-zero exceptions impossible. */ +/* Make sdiv/smod divide-by-minus-one exceptions impossible. */ +if ((insn->code == (BPF_ALU64 |BPF_MOD |BPF_K ) || + insn->code == (BPF_ALU64 |BPF_DIV |BPF_K ) || + insn->code == (BPF_ALU |BPF_MOD |BPF_K ) || + insn->code == (BPF_ALU |BPF_DIV |BPF_K )) && + insn->off ==1 && insn->imm ==-1 ) { +bool is64 =BPF_CLASS (insn->code) ==BPF_ALU64 ; +bool isdiv =BPF_OP (insn->code) ==BPF_DIV ; +struct bpf_insn *patchlet; +struct bpf_insn chk_and_sdiv[] = { +BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_NEG ) |BPF_K , insn->dst_reg, +0 ,0 ,0 ), + }; +struct bpf_insn chk_and_smod[] = { +BPF_MOV32_IMM (insn->dst_reg,0 ), + }; + + patchlet = isdiv ? chk_and_sdiv : chk_and_smod; + cnt = isdiv ?ARRAY_SIZE (chk_and_sdiv) :ARRAY_SIZE (chk_and_smod); + + new_prog =bpf_patch_insn_data (env, i + delta, patchlet, cnt); +if (!new_prog) +return -ENOMEM ; + + delta += cnt -1 ; + env->prog = prog = new_prog; + insn = new_prog->insnsi + i + delta; + goto next_insn; + } + +/* Make divide-by-zero and divide-by-minus-one exceptions impossible. */ if (insn->code == (BPF_ALU64 |BPF_MOD |BPF_X ) || insn->code == (BPF_ALU64 |BPF_DIV |BPF_X ) || insn->code == (BPF_ALU |BPF_MOD |BPF_X ) || insn->code == (BPF_ALU |BPF_DIV |BPF_X )) {bool is64 =BPF_CLASS (insn->code) ==BPF_ALU64 ;bool isdiv =BPF_OP (insn->code) ==BPF_DIV ; +bool is_sdiv = isdiv && insn->off ==1 ; +bool is_smod = !isdiv && insn->off ==1 ;struct bpf_insn *patchlet;struct bpf_insn chk_and_div[] = {/* [R,W]x div 0 -> 0 */ @@ -20525,10 +20558,62 @@ static int do_misc_fixups(struct bpf_verifier_env *env)BPF_JMP_IMM (BPF_JA ,0 ,0 ,1 ), BPF_MOV32_REG(insn->dst_reg, insn->dst_reg), }; +struct bpf_insn chk_and_sdiv[] = { +/* [R,W]x sdiv 0 -> 0 +* LLONG_MIN sdiv -1 -> LLONG_MIN +* INT_MIN sdiv -1 -> INT_MIN +*/ +BPF_MOV64_REG (BPF_REG_AX , insn->src_reg), +BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_ADD ) |BPF_K ,BPF_REG_AX , +0 ,0 ,1 ), +BPF_RAW_INSN ((is64 ?BPF_JMP :BPF_JMP32 ) | +BPF_JGT |BPF_K ,BPF_REG_AX , +0 ,4 ,1 ), +BPF_RAW_INSN ((is64 ?BPF_JMP :BPF_JMP32 ) | +BPF_JEQ |BPF_K ,BPF_REG_AX , +0 ,1 ,0 ), +BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_MOV ) |BPF_K , insn->dst_reg, +0 ,0 ,0 ), +/* BPF_NEG(LLONG_MIN) == -LLONG_MIN == LLONG_MIN */ +BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_NEG ) |BPF_K , insn->dst_reg, +0 ,0 ,0 ), +BPF_JMP_IMM (BPF_JA ,0 ,0 ,1 ), + *insn, + }; +struct bpf_insn chk_and_smod[] = { +/* [R,W]x mod 0 -> [R,W]x */ +/* [R,W]x mod -1 -> 0 */ +BPF_MOV64_REG (BPF_REG_AX , insn->src_reg), +BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_ADD ) |BPF_K ,BPF_REG_AX , +0 ,0 ,1 ), +BPF_RAW_INSN ((is64 ?BPF_JMP :BPF_JMP32 ) | +BPF_JGT |BPF_K ,BPF_REG_AX , +0 ,3 ,1 ), +BPF_RAW_INSN ((is64 ?BPF_JMP :BPF_JMP32 ) | +BPF_JEQ |BPF_K ,BPF_REG_AX , +0 ,3 + (is64 ?0 :1 ),1 ), +BPF_MOV32_IMM (insn->dst_reg,0 ), +BPF_JMP_IMM (BPF_JA ,0 ,0 ,1 ), + *insn, +BPF_JMP_IMM (BPF_JA ,0 ,0 ,1 ), +BPF_MOV32_REG (insn->dst_reg, insn->dst_reg), + }; - patchlet = isdiv ? chk_and_div : chk_and_mod; - cnt = isdiv ?ARRAY_SIZE (chk_and_div) : -ARRAY_SIZE (chk_and_mod) - (is64 ?2 :0 ); +if (is_sdiv) {
kernel/bpf/verifier.c static int do_misc_fixups(struct bpf_verifier_env *env) · line 20616
Quoted 10 lines
kernel/bpf/verifier.c +10
-0+ patchlet = chk_and_sdiv; + cnt =ARRAY_SIZE (chk_and_sdiv); + }else if (is_smod) { + patchlet = chk_and_smod; + cnt =ARRAY_SIZE (chk_and_smod) - (is64 ?2 :0 ); + }else { + patchlet = isdiv ? chk_and_div : chk_and_mod; + cnt = isdiv ?ARRAY_SIZE (chk_and_div) : +ARRAY_SIZE (chk_and_mod) - (is64 ?2 :0 ); + }
looking at how much isdiv vs ismod specific logic we have, it seems like it would be cleaner (and would use less stack as well) to have sdiv vs smod cases handled separately.
But the logic looks good to me, so
Quoted 6 lines
kernel/bpf/verifier.c +0
-1new_prog =bpf_patch_insn_data (env, i + delta, patchlet, cnt);if (!new_prog) --2.43.5
Quoted 6 lines · Yonghong Song
kernel/bpf/verifier.c +5
-0+BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_ADD ) |BPF_K ,BPF_REG_AX , +0 ,0 ,1 ), + BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | + BPF_JGT | BPF_K, BPF_REG_AX,
kernel/bpf/verifier.c static int do_misc_fixups(struct bpf_verifier_env *env) · line 20581
Quoted 10 lines
kernel/bpf/verifier.c +10
-0+ 0, 4, 1), +BPF_RAW_INSN ((is64 ?BPF_JMP :BPF_JMP32 ) | +BPF_JEQ |BPF_K ,BPF_REG_AX , +0 ,1 ,0 ), +BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | +BPF_OP (BPF_MOV ) |BPF_K , insn->dst_reg, +0 ,0 ,0 ), +/* BPF_NEG(LLONG_MIN) == -LLONG_MIN == LLONG_MIN */ + BPF_RAW_INSN((is64 ? BPF_ALU64 : BPF_ALU) | + BPF_OP(BPF_NEG) | BPF_K, insn->dst_reg,
lgtm, but all of BPF_OP(..) are confusing.
What's the point?
We use BPF_OP(insn->code) to reuse the code when we create a new opcode, but BPF_OP(BPF_NEG) == BPF_NEG and BPF_OP(BPF_MOV) == BPF_MOV, so why?
If I'm not missing anything I can remove these BPF_OP wrapping when applying. wdyt?
Yonghong Song
reviewerQuoted 16 lines · Alexei Starovoitov
Quoted 5 lines · Yonghong Song
(quoted diff) +5
-0+ BPF_RAW_INSN((is64 ? BPF_ALU64 : BPF_ALU) | + BPF_OP(BPF_ADD) | BPF_K, BPF_REG_AX, + 0, 0, 1), + BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | + BPF_JGT | BPF_K, BPF_REG_AX,Review
Quoted 10 lines
(quoted diff) +10
-0+ 0, 4, 1), + BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | + BPF_JEQ | BPF_K, BPF_REG_AX, + 0, 1, 0), + BPF_RAW_INSN((is64 ? BPF_ALU64 : BPF_ALU) | + BPF_OP(BPF_MOV) | BPF_K, insn->dst_reg, + 0, 0, 0), + /* BPF_NEG(LLONG_MIN) == -LLONG_MIN == LLONG_MIN */ + BPF_RAW_INSN((is64 ? BPF_ALU64 : BPF_ALU) | + BPF_OP(BPF_NEG) | BPF_K, insn->dst_reg,
Quoted 4 lines
lgtm, but all of
BPF_OP(..)are confusing.
What's the point?
We useBPF_OP(insn->code)to reuse the code when we create a new opcode, butBPF_OP(BPF_NEG)==BPF_NEGandBPF_OP(BPF_MOV)==BPF_MOV, so why?
Sorry, I focused on the algorithm and missed this one. Yes, changing BPF_OP(BPF_NEG) to BPF_NEG and other similar cases are correct.
kernel/bpf/verifier.c +5 -5
@@ -20510,7 +20510,7 @@ static int do_misc_fixups(struct bpf_verifier_env *env)struct bpf_insn *patchlet;struct bpf_insn chk_and_sdiv[] = {BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | -BPF_OP (BPF_NEG ) |BPF_K , insn->dst_reg, +BPF_NEG |BPF_K , insn->dst_reg,0 ,0 ,0 ), };struct bpf_insn chk_and_smod[] = {
@@ -20565,7 +20565,7 @@ static int do_misc_fixups(struct bpf_verifier_env *env)*/ BPF_MOV64_REG (BPF_REG_AX , insn->src_reg),BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | -BPF_OP (BPF_ADD ) |BPF_K ,BPF_REG_AX , +BPF_ADD |BPF_K ,BPF_REG_AX ,0 ,0 ,1 ), BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | BPF_JGT | BPF_K, BPF_REG_AX,
@@ -20574,11 +20574,11 @@ static int do_misc_fixups(struct bpf_verifier_env *env) BPF_JEQ | BPF_K, BPF_REG_AX, 0, 1, 0),BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | -BPF_OP (BPF_MOV ) |BPF_K , insn->dst_reg, +BPF_MOV |BPF_K , insn->dst_reg,0 ,0 ,0 ),/* BPF_NEG(LLONG_MIN) == -LLONG_MIN == LLONG_MIN */ BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | -BPF_OP (BPF_NEG ) |BPF_K , insn->dst_reg, +BPF_NEG |BPF_K , insn->dst_reg,0 ,0 ,0 ),BPF_JMP_IMM (BPF_JA ,0 ,0 ,1 ), *insn,
@@ -20588,7 +20588,7 @@ static int do_misc_fixups(struct bpf_verifier_env *env)/* [R,W]x mod -1 -> 0 */ BPF_MOV64_REG (BPF_REG_AX , insn->src_reg),BPF_RAW_INSN ((is64 ?BPF_ALU64 :BPF_ALU ) | -BPF_OP (BPF_ADD ) |BPF_K ,BPF_REG_AX , +BPF_ADD |BPF_K ,BPF_REG_AX ,0 ,0 ,1 ), BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) |
Quoted 3 lines
If I'm not missing anything I can remove these
BPF_OPwrapping when applying. wdyt?
Yes, pelase do. Thanks!
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 12 lines · On Fri, 13 Sep 2024 08:03:26 -0700 you wrote:
Zac Ecob reported a problem where a bpf program may cause kernel crash due to the following error:
Oops: divide error: 0000 [#1] PREEMPT SMP KASAN PTIThe failure is due to the below signed divide:
LLONG_MIN/-1 whereLLONG_MINequals to -9,223,372,036,854,775,808.
LLONG_MIN/-1 is supposed to give a positive number 9,223,372,036,854,775,808, but it is impossible since for 64-bit system, the maximum positive number is 9,223,372,036,854,775,807. Onx86_64, LLONG_MIN/-1 will cause a kernel exception. On arm64, the result for LLONG_MIN/-1 isLLONG_MIN.
Quoted 1 line
[...]
Here is the summary with links:
- [bpf-next,v3,1/2] bpf: Fix a sdiv overflow issue
https://git.kernel.org/bpf/bpf-next/c/7dd34d7b7dcf - [bpf-next,v3,2/2] selftests/bpf: Add tests for sdiv/smod overflow cases https://git.kernel.org/bpf/bpf-next/c/a18062d54a0b
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html