Skip to contentSkip to conversation

bpf: Explicitly zero-extend R0 after 32-bit cmpxchg

PATCH Superseded Brendan Jackman 4 participants 6 messages Mar 3, 2021

Subsystems (4)

v5 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

PATCH v5 1/1 bpf: Explicitly zero-extend R0 after 32-bit cmpxchg

As pointed out by Ilya and explained in the new comment, there's a discrepancy between x86 and BPF CMPXCHG semantics: BPF always loads the value from memory into r0, while x86 only does so when r0 and the value in memory are different. The same issue affects s390.

At first this might sound like pure semantics, but it makes a real difference when the comparison is 32-bit, since the load will zero-extend r0/rax.

The fix is to explicitly zero-extend rax after doing such a CMPXCHG. Since this problem affects multiple archs, this is done in the verifier by patching in a BPF_ZEXT_REG instruction after every 32-bit cmpxchg. Any archs that don't need such manual zero-extension can do a look-ahead with insn_is_zext to skip the unnecessary mov.

Fixes 5ffa25502b5a ("bpf: Add instructions for atomic_[cmp]xchg")

---

Differences v4->v5[1]:
- Moved the logic entirely into opt_subreg_zext_lo32_rnd_hi32, thanks to Martin for suggesting this.

Differences v3->v4[1]:
- Moved the optimization against pointless zext into the correct place:
opt_subreg_zext_lo32_rnd_hi32 is called _after_ fixup_bpf_calls.

Differences v2->v3[1]:
- Moved patching into fixup_bpf_calls (patch incoming to rename this function)
- Added extra commentary on bpf_jit_needs_zext
- Added check to avoid adding a pointless zext(r0) if there's already one there.

Difference v1->v2[1]: Now solved centrally in the verifier instead of
specifically for the x86 JIT. Thanks to Ilya and Daniel for the suggestions!

[1] v4: https://lore.kernel.org/bpf/CA+i-1C3ytZz6FjcPmUg5s4L51pMQDxWcZNvM86w4RHZ_o2khwg@mail.gmail.com/T/#t

    v3: https://lore.kernel.org/bpf/[email protected]/T/#t
    v2: https://lore.kernel.org/bpf/[email protected]/T/#t
    v1: https://lore.kernel.org/bpf/[email protected]/T/#t
 kernel/bpf/core.c                             |  4 +++
 kernel/bpf/verifier.c                         | 17 +++++++++++-
 .../selftests/bpf/verifier/atomic_cmpxchg.c   | 25 ++++++++++++++++++
 .../selftests/bpf/verifier/atomic_or.c        | 26 +++++++++++++++++++
 4 files changed, 71 insertions(+), 1 deletion(-)
kernel/bpf/core.c +4 -0
index 0ae015ad1e05..dcf18612841b 100644
@@ -2342,6 +2342,10 @@ bool __weak bpf_helper_changes_pkt_data(void *func)
 /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
  * analysis code and wants explicit zero extension inserted by verifier.
  * Otherwise, return FALSE.
+ *
+ * The verifier inserts an explicit zero extension after BPF_CMPXCHGs even if
+ * you don't override this. JITs that don't want these extra insns can detect
+ * them using insn_is_zext.
  */
 bool __weak bpf_jit_needs_zext(void)
 {
kernel/bpf/verifier.c +16 -1
index 4c373589273b..37076e4c6175 100644
@@ -11237,6 +11237,11 @@ static int opt_remove_nops(struct bpf_verifier_env *env)
 	return 0;
 }
 
+static inline bool is_cmpxchg(struct bpf_insn *insn)
+{
+	return (BPF_MODE(insn->code) == BPF_ATOMIC && insn->imm == BPF_CMPXCHG);
+}
+
 static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env,
 					 const union bpf_attr *attr)
 {
@@ -11296,7 +11301,17 @@ static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env,
 			goto apply_patch_buffer;
 		}
 
-		if (!bpf_jit_needs_zext())
+		/* Add in an zero-extend instruction if a) the JIT has requested
+		 * it or b) it's a CMPXCHG.
+		 *
+		 * The latter is because: BPF_CMPXCHG always loads a value into
+		 * R0, therefore always zero-extends. However some archs'
+		 * equivalent instruction only does this load when the
+		 * comparison is successful. This detail of CMPXCHG is
+		 * orthogonal to the general zero-extension behaviour of the
+		 * CPU, so it's treated independently of bpf_jit_needs_zext.
+		 */
+		if (!bpf_jit_needs_zext() && !is_cmpxchg(&insn))
 			continue;
 
 		if (WARN_ON_ONCE(load_reg == -1)) {
tools/testing/selftests/bpf/verifier/atomic_cmpxchg.c +25 -0
index 2efd8bcf57a1..6e52dfc64415 100644
@@ -94,3 +94,28 @@
 	.result = REJECT,
 	.errstr = "invalid read from stack",
 },
+{
+	"BPF_W cmpxchg should zero top 32 bits",
+	.insns = {
+		/* r0 = U64_MAX; */
+		BPF_MOV64_IMM(BPF_REG_0, 0),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_0, 1),
+		/* u64 val = r0; */
+		BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_0, -8),
+		/* r0 = (u32)atomic_cmpxchg((u32 *)&val, r0, 1); */
+		BPF_MOV32_IMM(BPF_REG_1, 1),
+		BPF_ATOMIC_OP(BPF_W, BPF_CMPXCHG, BPF_REG_10, BPF_REG_1, -8),
+		/* r1 = 0x00000000FFFFFFFFull; */
+		BPF_MOV64_IMM(BPF_REG_1, 1),
+		BPF_ALU64_IMM(BPF_LSH, BPF_REG_1, 32),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_1, 1),
+		/* if (r0 != r1) exit(1); */
+		BPF_JMP_REG(BPF_JEQ, BPF_REG_0, BPF_REG_1, 2),
+		BPF_MOV32_IMM(BPF_REG_0, 1),
+		BPF_EXIT_INSN(),
+		/* exit(0); */
+		BPF_MOV32_IMM(BPF_REG_0, 0),
+		BPF_EXIT_INSN(),
+	},
+	.result = ACCEPT,
+},
tools/testing/selftests/bpf/verifier/atomic_or.c +26 -0
index 70f982e1f9f0..0a08b99e6ddd 100644
@@ -75,3 +75,29 @@
 	},
 	.result = ACCEPT,
 },
+{
+	"BPF_W atomic_fetch_or should zero top 32 bits",
+	.insns = {
+		/* r1 = U64_MAX; */
+		BPF_MOV64_IMM(BPF_REG_1, 0),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_1, 1),
+		/* u64 val = r0; */
+		BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_1, -8),
+		/* r1 = (u32)atomic_sub((u32 *)&val, 1); */
+		BPF_MOV32_IMM(BPF_REG_1, 2),
+		BPF_ATOMIC_OP(BPF_W, BPF_OR | BPF_FETCH, BPF_REG_10, BPF_REG_1, -8),
+		/* r2 = 0x00000000FFFFFFFF; */
+		BPF_MOV64_IMM(BPF_REG_2, 1),
+		BPF_ALU64_IMM(BPF_LSH, BPF_REG_2, 32),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_2, 1),
+		/* if (r2 != r1) exit(1); */
+		BPF_JMP_REG(BPF_JEQ, BPF_REG_2, BPF_REG_1, 2),
+		/* BPF_MOV32_IMM(BPF_REG_0, 1), */
+		BPF_MOV64_REG(BPF_REG_0, BPF_REG_1),
+		BPF_EXIT_INSN(),
+		/* exit(0); */
+		BPF_MOV32_IMM(BPF_REG_0, 0),
+		BPF_EXIT_INSN(),
+	},
+	.result = ACCEPT,
+},

base-commit: f2cfe32e8a965a86e512dcb2e6251371d4a60c63 --
2.30.1.766.gb4fecdf3b7-goog

maintainer
#2 lore raw
Quoted 7 lines · Brendan Jackman
Quoted 1 line · Brendan Jackman

base-commit: f2cfe32e8a965a86e512dcb2e6251371d4a60c63

Oh yeah, this is based on Ilya's patch [1]. Is that OK or should I just resend it once that one is merged?

[1] https://lore.kernel.org/bpf/[email protected]/T/#u

No need to resend, enough to have it stated here.

Thanks,
Daniel

#3 lore raw
In reply to Brendan Jackman
Quoted 69 lines · Brendan Jackman

As pointed out by Ilya and explained in the new comment, there's a discrepancy between x86 and BPF CMPXCHG semantics: BPF always loads the value from memory into r0, while x86 only does so when r0 and the value in memory are different. The same issue affects s390.

At first this might sound like pure semantics, but it makes a real difference when the comparison is 32-bit, since the load will zero-extend r0/rax.

The fix is to explicitly zero-extend rax after doing such a CMPXCHG. Since this problem affects multiple archs, this is done in the verifier by patching in a BPF_ZEXT_REG instruction after every 32-bit cmpxchg. Any archs that don't need such manual zero-extension can do a look-ahead with insn_is_zext to skip the unnecessary mov.

Fixes 5ffa25502b5a ("bpf: Add instructions for atomic_[cmp]xchg")

---

Differences v4->v5[1]:
- Moved the logic entirely into opt_subreg_zext_lo32_rnd_hi32, thanks to Martin for suggesting this.

Differences v3->v4[1]:
- Moved the optimization against pointless zext into the correct place:
opt_subreg_zext_lo32_rnd_hi32 is called _after_ fixup_bpf_calls.

Differences v2->v3[1]:
- Moved patching into fixup_bpf_calls (patch incoming to rename this function)
- Added extra commentary on bpf_jit_needs_zext
- Added check to avoid adding a pointless zext(r0) if there's already one there.

Difference v1->v2[1]: Now solved centrally in the verifier instead of
specifically for the x86 JIT. Thanks to Ilya and Daniel for the suggestions!

[1] v4: https://lore.kernel.org/bpf/CA+i-1C3ytZz6FjcPmUg5s4L51pMQDxWcZNvM86w4RHZ_o2khwg@mail.gmail.com/T/#t

    v3: https://lore.kernel.org/bpf/[email protected]/T/#t
    v2: https://lore.kernel.org/bpf/[email protected]/T/#t
    v1: https://lore.kernel.org/bpf/[email protected]/T/#t
 kernel/bpf/core.c                             |  4 +++
 kernel/bpf/verifier.c                         | 17 +++++++++++-
 .../selftests/bpf/verifier/atomic_cmpxchg.c   | 25 ++++++++++++++++++
 .../selftests/bpf/verifier/atomic_or.c        | 26 +++++++++++++++++++
 4 files changed, 71 insertions(+), 1 deletion(-)
kernel/bpf/core.c +4 -0
index 0ae015ad1e05..dcf18612841b 100644
@@ -2342,6 +2342,10 @@ bool __weak bpf_helper_changes_pkt_data(void *func)
 /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
  * analysis code and wants explicit zero extension inserted by verifier.
  * Otherwise, return FALSE.
+ *
+ * The verifier inserts an explicit zero extension after BPF_CMPXCHGs even if
+ * you don't override this. JITs that don't want these extra insns can detect
+ * them using insn_is_zext.
  */
 bool __weak bpf_jit_needs_zext(void)
 {
kernel/bpf/verifier.c +0 -0
index 4c373589273b..37076e4c6175 100644

kernel/bpf/verifier.c static int opt_remove_nops(struct bpf_verifier_env *env) · line 11240

Quoted 4 lines
kernel/bpf/verifier.c +1 -0
@@ -11237,6 +11237,11 @@ static int opt_remove_nops(struct bpf_verifier_env *env)
 	return 0;
 }
 
+static inline bool is_cmpxchg(struct bpf_insn *insn)

nit. "const" struct bpf_insn *insn.

kernel/bpf/verifier.c static int opt_remove_nops(struct bpf_verifier_env *env) · line 11242

Quoted 2 lines
kernel/bpf/verifier.c +2 -0
+{
+	return (BPF_MODE(insn->code) == BPF_ATOMIC && insn->imm == BPF_CMPXCHG);

I think it is better to check BPF_CLASS(insn->code) == BPF_STX also in case in the future this helper will be reused before do_check() has a chance to verify the instructions.

Quoted 62 lines
kernel/bpf/verifier.c +13 -1
+}
+
 static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env,
 					 const union bpf_attr *attr)
 {
@@ -11296,7 +11301,17 @@ static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env,
 			goto apply_patch_buffer;
 		}
 
-		if (!bpf_jit_needs_zext())
+		/* Add in an zero-extend instruction if a) the JIT has requested
+		 * it or b) it's a CMPXCHG.
+		 *
+		 * The latter is because: BPF_CMPXCHG always loads a value into
+		 * R0, therefore always zero-extends. However some archs'
+		 * equivalent instruction only does this load when the
+		 * comparison is successful. This detail of CMPXCHG is
+		 * orthogonal to the general zero-extension behaviour of the
+		 * CPU, so it's treated independently of bpf_jit_needs_zext.
+		 */
+		if (!bpf_jit_needs_zext() && !is_cmpxchg(&insn))
 			continue;
 
 		if (WARN_ON_ONCE(load_reg == -1)) {
tools/testing/selftests/bpf/verifier/atomic_cmpxchg.c +25 -0
index 2efd8bcf57a1..6e52dfc64415 100644
@@ -94,3 +94,28 @@
 	.result = REJECT,
 	.errstr = "invalid read from stack",
 },
+{
+	"BPF_W cmpxchg should zero top 32 bits",
+	.insns = {
+		/* r0 = U64_MAX; */
+		BPF_MOV64_IMM(BPF_REG_0, 0),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_0, 1),
+		/* u64 val = r0; */
+		BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_0, -8),
+		/* r0 = (u32)atomic_cmpxchg((u32 *)&val, r0, 1); */
+		BPF_MOV32_IMM(BPF_REG_1, 1),
+		BPF_ATOMIC_OP(BPF_W, BPF_CMPXCHG, BPF_REG_10, BPF_REG_1, -8),
+		/* r1 = 0x00000000FFFFFFFFull; */
+		BPF_MOV64_IMM(BPF_REG_1, 1),
+		BPF_ALU64_IMM(BPF_LSH, BPF_REG_1, 32),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_1, 1),
+		/* if (r0 != r1) exit(1); */
+		BPF_JMP_REG(BPF_JEQ, BPF_REG_0, BPF_REG_1, 2),
+		BPF_MOV32_IMM(BPF_REG_0, 1),
+		BPF_EXIT_INSN(),
+		/* exit(0); */
+		BPF_MOV32_IMM(BPF_REG_0, 0),
+		BPF_EXIT_INSN(),
+	},
+	.result = ACCEPT,
+},
tools/testing/selftests/bpf/verifier/atomic_or.c +0 -0
index 70f982e1f9f0..0a08b99e6ddd 100644

tools/testing/selftests/bpf/verifier/atomic_or.c line 84

Quoted 10 lines
tools/testing/selftests/bpf/verifier/atomic_or.c +7 -0
@@ -75,3 +75,29 @@
 	},
 	.result = ACCEPT,
 },
+{
+	"BPF_W atomic_fetch_or should zero top 32 bits",
+	.insns = {
+		/* r1 = U64_MAX; */
+		BPF_MOV64_IMM(BPF_REG_1, 0),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_1, 1),
+		/* u64 val = r0; */

s/r0/r1/

tools/testing/selftests/bpf/verifier/atomic_or.c line 86

Quoted 2 lines
tools/testing/selftests/bpf/verifier/atomic_or.c +2 -0
+		BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_1, -8),
+		/* r1 = (u32)atomic_sub((u32 *)&val, 1); */
		   r1 = (u32)atomic_fetch_or((u32 *)&val, 2)
Quoted 22 lines
tools/testing/selftests/bpf/verifier/atomic_or.c +17 -0
+		BPF_MOV32_IMM(BPF_REG_1, 2),
+		BPF_ATOMIC_OP(BPF_W, BPF_OR | BPF_FETCH, BPF_REG_10, BPF_REG_1, -8),
+		/* r2 = 0x00000000FFFFFFFF; */
+		BPF_MOV64_IMM(BPF_REG_2, 1),
+		BPF_ALU64_IMM(BPF_LSH, BPF_REG_2, 32),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_2, 1),
+		/* if (r2 != r1) exit(1); */
+		BPF_JMP_REG(BPF_JEQ, BPF_REG_2, BPF_REG_1, 2),
+		/* BPF_MOV32_IMM(BPF_REG_0, 1), */
+		BPF_MOV64_REG(BPF_REG_0, BPF_REG_1),
+		BPF_EXIT_INSN(),
+		/* exit(0); */
+		BPF_MOV32_IMM(BPF_REG_0, 0),
+		BPF_EXIT_INSN(),
+	},
+	.result = ACCEPT,
+},

base-commit: f2cfe32e8a965a86e512dcb2e6251371d4a60c63 --
2.30.1.766.gb4fecdf3b7-goog

#4 lore raw
Quoted 73 lines · Martin KaFai Lau
Quoted 72 lines · Brendan Jackman

As pointed out by Ilya and explained in the new comment, there's a discrepancy between x86 and BPF CMPXCHG semantics: BPF always loads the value from memory into r0, while x86 only does so when r0 and the value in memory are different. The same issue affects s390.

At first this might sound like pure semantics, but it makes a real difference when the comparison is 32-bit, since the load will zero-extend r0/rax.

The fix is to explicitly zero-extend rax after doing such a CMPXCHG. Since this problem affects multiple archs, this is done in the verifier by patching in a BPF_ZEXT_REG instruction after every 32-bit cmpxchg. Any archs that don't need such manual zero-extension can do a look-ahead with insn_is_zext to skip the unnecessary mov.

Fixes 5ffa25502b5a ("bpf: Add instructions for atomic_[cmp]xchg")

---

Differences v4->v5[1]:
- Moved the logic entirely into opt_subreg_zext_lo32_rnd_hi32, thanks to Martin for suggesting this.

Differences v3->v4[1]:
- Moved the optimization against pointless zext into the correct place:
opt_subreg_zext_lo32_rnd_hi32 is called _after_ fixup_bpf_calls.

Differences v2->v3[1]:
- Moved patching into fixup_bpf_calls (patch incoming to rename this function)
- Added extra commentary on bpf_jit_needs_zext
- Added check to avoid adding a pointless zext(r0) if there's already one there.

Difference v1->v2[1]: Now solved centrally in the verifier instead of
specifically for the x86 JIT. Thanks to Ilya and Daniel for the suggestions!

[1] v4: https://lore.kernel.org/bpf/CA+i-1C3ytZz6FjcPmUg5s4L51pMQDxWcZNvM86w4RHZ_o2khwg@mail.gmail.com/T/#t

     v3: https://lore.kernel.org/bpf/[email protected]/T/#t
     v2: https://lore.kernel.org/bpf/[email protected]/T/#t
     v1: https://lore.kernel.org/bpf/[email protected]/T/#t
  kernel/bpf/core.c                             |  4 +++
  kernel/bpf/verifier.c                         | 17 +++++++++++-
  .../selftests/bpf/verifier/atomic_cmpxchg.c   | 25 ++++++++++++++++++
  .../selftests/bpf/verifier/atomic_or.c        | 26 +++++++++++++++++++
  4 files changed, 71 insertions(+), 1 deletion(-)
kernel/bpf/core.c +4 -0
index 0ae015ad1e05..dcf18612841b 100644
@@ -2342,6 +2342,10 @@ bool __weak bpf_helper_changes_pkt_data(void *func)
  /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
   * analysis code and wants explicit zero extension inserted by verifier.
   * Otherwise, return FALSE.
+ *
+ * The verifier inserts an explicit zero extension after BPF_CMPXCHGs even if
+ * you don't override this. JITs that don't want these extra insns can detect
+ * them using insn_is_zext.
   */
  bool __weak bpf_jit_needs_zext(void)
  {
kernel/bpf/verifier.c +0 -0
index 4c373589273b..37076e4c6175 100644
@@ -11237,6 +11237,11 @@ static int opt_remove_nops(struct bpf_verifier_env *env)
  	return 0;
  }
 
Quoted 8 lines

kernel/bpf/verifier.c static int opt_remove_nops(struct bpf_verifier_env *env) · line 11240

Quoted 1 line
kernel/bpf/verifier.c +1 -0
+static inline bool is_cmpxchg(struct bpf_insn *insn)

nit. "const" struct bpf_insn *insn.

kernel/bpf/verifier.c static int opt_remove_nops(struct bpf_verifier_env *env) · line 11242

Quoted 2 lines
kernel/bpf/verifier.c +2 -0
+{
+	return (BPF_MODE(insn->code) == BPF_ATOMIC && insn->imm == BPF_CMPXCHG);

I think it is better to check BPF_CLASS(insn->code) == BPF_STX also in case in the future this helper will be reused before do_check() has a chance to verify the instructions.

If this is the case, I would suggest to move is_cmpxchg() earlier
in verifier.c so later on for reuse we do not need to move this function. Also, in the return statement, there is no need for outmost ().

Quoted 101 lines
Quoted 63 lines
kernel/bpf/verifier.c +13 -1
+}
+
  static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env,
  					 const union bpf_attr *attr)
  {
@@ -11296,7 +11301,17 @@ static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env,
  			goto apply_patch_buffer;
  		}
 
-		if (!bpf_jit_needs_zext())
+		/* Add in an zero-extend instruction if a) the JIT has requested
+		 * it or b) it's a CMPXCHG.
+		 *
+		 * The latter is because: BPF_CMPXCHG always loads a value into
+		 * R0, therefore always zero-extends. However some archs'
+		 * equivalent instruction only does this load when the
+		 * comparison is successful. This detail of CMPXCHG is
+		 * orthogonal to the general zero-extension behaviour of the
+		 * CPU, so it's treated independently of bpf_jit_needs_zext.
+		 */
+		if (!bpf_jit_needs_zext() && !is_cmpxchg(&insn))
  			continue;
 
  		if (WARN_ON_ONCE(load_reg == -1)) {
tools/testing/selftests/bpf/verifier/atomic_cmpxchg.c +25 -0
index 2efd8bcf57a1..6e52dfc64415 100644
@@ -94,3 +94,28 @@
  	.result = REJECT,
  	.errstr = "invalid read from stack",
  },
+{
+	"BPF_W cmpxchg should zero top 32 bits",
+	.insns = {
+		/* r0 = U64_MAX; */
+		BPF_MOV64_IMM(BPF_REG_0, 0),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_0, 1),
+		/* u64 val = r0; */
+		BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_0, -8),
+		/* r0 = (u32)atomic_cmpxchg((u32 *)&val, r0, 1); */
+		BPF_MOV32_IMM(BPF_REG_1, 1),
+		BPF_ATOMIC_OP(BPF_W, BPF_CMPXCHG, BPF_REG_10, BPF_REG_1, -8),
+		/* r1 = 0x00000000FFFFFFFFull; */
+		BPF_MOV64_IMM(BPF_REG_1, 1),
+		BPF_ALU64_IMM(BPF_LSH, BPF_REG_1, 32),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_1, 1),
+		/* if (r0 != r1) exit(1); */
+		BPF_JMP_REG(BPF_JEQ, BPF_REG_0, BPF_REG_1, 2),
+		BPF_MOV32_IMM(BPF_REG_0, 1),
+		BPF_EXIT_INSN(),
+		/* exit(0); */
+		BPF_MOV32_IMM(BPF_REG_0, 0),
+		BPF_EXIT_INSN(),
+	},
+	.result = ACCEPT,
+},
tools/testing/selftests/bpf/verifier/atomic_or.c +0 -0
index 70f982e1f9f0..0a08b99e6ddd 100644

tools/testing/selftests/bpf/verifier/atomic_or.c line 84

Quoted 10 lines
tools/testing/selftests/bpf/verifier/atomic_or.c +7 -0
@@ -75,3 +75,29 @@
  	},
  	.result = ACCEPT,
  },
+{
+	"BPF_W atomic_fetch_or should zero top 32 bits",
+	.insns = {
+		/* r1 = U64_MAX; */
+		BPF_MOV64_IMM(BPF_REG_1, 0),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_1, 1),
+		/* u64 val = r0; */

s/r0/r1/

tools/testing/selftests/bpf/verifier/atomic_or.c line 86

Quoted 2 lines
tools/testing/selftests/bpf/verifier/atomic_or.c +2 -0
+		BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_1, -8),
+		/* r1 = (u32)atomic_sub((u32 *)&val, 1); */
		   r1 = (u32)atomic_fetch_or((u32 *)&val, 2)
Quoted 22 lines
tools/testing/selftests/bpf/verifier/atomic_or.c +17 -0
+		BPF_MOV32_IMM(BPF_REG_1, 2),
+		BPF_ATOMIC_OP(BPF_W, BPF_OR | BPF_FETCH, BPF_REG_10, BPF_REG_1, -8),
+		/* r2 = 0x00000000FFFFFFFF; */
+		BPF_MOV64_IMM(BPF_REG_2, 1),
+		BPF_ALU64_IMM(BPF_LSH, BPF_REG_2, 32),
+		BPF_ALU64_IMM(BPF_SUB, BPF_REG_2, 1),
+		/* if (r2 != r1) exit(1); */
+		BPF_JMP_REG(BPF_JEQ, BPF_REG_2, BPF_REG_1, 2),
+		/* BPF_MOV32_IMM(BPF_REG_0, 1), */
+		BPF_MOV64_REG(BPF_REG_0, BPF_REG_1),
+		BPF_EXIT_INSN(),
+		/* exit(0); */
+		BPF_MOV32_IMM(BPF_REG_0, 0),
+		BPF_EXIT_INSN(),
+	},
+	.result = ACCEPT,
+},

base-commit: f2cfe32e8a965a86e512dcb2e6251371d4a60c63 --
2.30.1.766.gb4fecdf3b7-goog

#5 lore raw

Thanks once again for the reviews.

Quoted 82 lines · Yonghong Song
Quoted 78 lines · Martin KaFai Lau
Quoted 69 lines · Brendan Jackman

As pointed out by Ilya and explained in the new comment, there's a discrepancy between x86 and BPF CMPXCHG semantics: BPF always loads the value from memory into r0, while x86 only does so when r0 and the value in memory are different. The same issue affects s390.

At first this might sound like pure semantics, but it makes a real difference when the comparison is 32-bit, since the load will zero-extend r0/rax.

The fix is to explicitly zero-extend rax after doing such a CMPXCHG. Since this problem affects multiple archs, this is done in the verifier by patching in a BPF_ZEXT_REG instruction after every 32-bit cmpxchg. Any archs that don't need such manual zero-extension can do a look-ahead with insn_is_zext to skip the unnecessary mov.

Fixes 5ffa25502b5a ("bpf: Add instructions for atomic_[cmp]xchg")

---

Differences v4->v5[1]:
- Moved the logic entirely into opt_subreg_zext_lo32_rnd_hi32, thanks to Martin for suggesting this.

Differences v3->v4[1]:
- Moved the optimization against pointless zext into the correct place:
opt_subreg_zext_lo32_rnd_hi32 is called _after_ fixup_bpf_calls.

Differences v2->v3[1]:
- Moved patching into fixup_bpf_calls (patch incoming to rename this function)
- Added extra commentary on bpf_jit_needs_zext
- Added check to avoid adding a pointless zext(r0) if there's already one there.

Difference v1->v2[1]: Now solved centrally in the verifier instead of
specifically for the x86 JIT. Thanks to Ilya and Daniel for the suggestions!

[1] v4: https://lore.kernel.org/bpf/CA+i-1C3ytZz6FjcPmUg5s4L51pMQDxWcZNvM86w4RHZ_o2khwg@mail.gmail.com/T/#t

     v3: https://lore.kernel.org/bpf/[email protected]/T/#t
     v2: https://lore.kernel.org/bpf/[email protected]/T/#t
     v1: https://lore.kernel.org/bpf/[email protected]/T/#t
  kernel/bpf/core.c                             |  4 +++
  kernel/bpf/verifier.c                         | 17 +++++++++++-
  .../selftests/bpf/verifier/atomic_cmpxchg.c   | 25 ++++++++++++++++++
  .../selftests/bpf/verifier/atomic_or.c        | 26 +++++++++++++++++++
  4 files changed, 71 insertions(+), 1 deletion(-)
kernel/bpf/core.c +4 -0
index 0ae015ad1e05..dcf18612841b 100644
@@ -2342,6 +2342,10 @@ bool __weak bpf_helper_changes_pkt_data(void *func)
  /* Return TRUE if the JIT backend wants verifier to enable sub-register usage
   * analysis code and wants explicit zero extension inserted by verifier.
   * Otherwise, return FALSE.
+ *
+ * The verifier inserts an explicit zero extension after BPF_CMPXCHGs even if
+ * you don't override this. JITs that don't want these extra insns can detect
+ * them using insn_is_zext.
   */
  bool __weak bpf_jit_needs_zext(void)
  {
kernel/bpf/verifier.c +0 -0
index 4c373589273b..37076e4c6175 100644

kernel/bpf/verifier.c static int opt_remove_nops(struct bpf_verifier_env *env) · line 11240

Quoted 4 lines
kernel/bpf/verifier.c +1 -0
@@ -11237,6 +11237,11 @@ static int opt_remove_nops(struct bpf_verifier_env *env)
     return 0;
  }
 
+static inline bool is_cmpxchg(struct bpf_insn *insn)

nit. "const" struct bpf_insn *insn.

Quoted 2 lines
kernel/bpf/verifier.c +2 -0
+{
+    return (BPF_MODE(insn->code) == BPF_ATOMIC && insn->imm == BPF_CMPXCHG);
Quoted 8 lines
Quoted 3 lines

I think it is better to check BPF_CLASS(insn->code) == BPF_STX also in case in the future this helper will be reused before do_check() has a chance to verify the instructions.

If this is the case, I would suggest to move is_cmpxchg() earlier in verifier.c so later on for reuse we do not need to move this function. Also, in the return statement, there is no need for outmost ().

Yep, all good points, thanks. Spinning v6 now.

Quoted 101 lines
Quoted 101 lines
Quoted 63 lines
kernel/bpf/verifier.c +13 -1
+}
+
  static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env,
                                      const union bpf_attr *attr)
  {
@@ -11296,7 +11301,17 @@ static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env,
                     goto apply_patch_buffer;
             }
 
-            if (!bpf_jit_needs_zext())
+            /* Add in an zero-extend instruction if a) the JIT has requested
+             * it or b) it's a CMPXCHG.
+             *
+             * The latter is because: BPF_CMPXCHG always loads a value into
+             * R0, therefore always zero-extends. However some archs'
+             * equivalent instruction only does this load when the
+             * comparison is successful. This detail of CMPXCHG is
+             * orthogonal to the general zero-extension behaviour of the
+             * CPU, so it's treated independently of bpf_jit_needs_zext.
+             */
+            if (!bpf_jit_needs_zext() && !is_cmpxchg(&insn))
                     continue;
 
             if (WARN_ON_ONCE(load_reg == -1)) {
tools/testing/selftests/bpf/verifier/atomic_cmpxchg.c +25 -0
index 2efd8bcf57a1..6e52dfc64415 100644
@@ -94,3 +94,28 @@
     .result = REJECT,
     .errstr = "invalid read from stack",
  },
+{
+    "BPF_W cmpxchg should zero top 32 bits",
+    .insns = {
+            /* r0 = U64_MAX; */
+            BPF_MOV64_IMM(BPF_REG_0, 0),
+            BPF_ALU64_IMM(BPF_SUB, BPF_REG_0, 1),
+            /* u64 val = r0; */
+            BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_0, -8),
+            /* r0 = (u32)atomic_cmpxchg((u32 *)&val, r0, 1); */
+            BPF_MOV32_IMM(BPF_REG_1, 1),
+            BPF_ATOMIC_OP(BPF_W, BPF_CMPXCHG, BPF_REG_10, BPF_REG_1, -8),
+            /* r1 = 0x00000000FFFFFFFFull; */
+            BPF_MOV64_IMM(BPF_REG_1, 1),
+            BPF_ALU64_IMM(BPF_LSH, BPF_REG_1, 32),
+            BPF_ALU64_IMM(BPF_SUB, BPF_REG_1, 1),
+            /* if (r0 != r1) exit(1); */
+            BPF_JMP_REG(BPF_JEQ, BPF_REG_0, BPF_REG_1, 2),
+            BPF_MOV32_IMM(BPF_REG_0, 1),
+            BPF_EXIT_INSN(),
+            /* exit(0); */
+            BPF_MOV32_IMM(BPF_REG_0, 0),
+            BPF_EXIT_INSN(),
+    },
+    .result = ACCEPT,
+},
tools/testing/selftests/bpf/verifier/atomic_or.c +0 -0
index 70f982e1f9f0..0a08b99e6ddd 100644

tools/testing/selftests/bpf/verifier/atomic_or.c line 84

Quoted 10 lines
tools/testing/selftests/bpf/verifier/atomic_or.c +7 -0
@@ -75,3 +75,29 @@
     },
     .result = ACCEPT,
  },
+{
+    "BPF_W atomic_fetch_or should zero top 32 bits",
+    .insns = {
+            /* r1 = U64_MAX; */
+            BPF_MOV64_IMM(BPF_REG_1, 0),
+            BPF_ALU64_IMM(BPF_SUB, BPF_REG_1, 1),
+            /* u64 val = r0; */

s/r0/r1/

tools/testing/selftests/bpf/verifier/atomic_or.c line 86

Quoted 2 lines
tools/testing/selftests/bpf/verifier/atomic_or.c +2 -0
+            BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_1, -8),
+            /* r1 = (u32)atomic_sub((u32 *)&val, 1); */
                 r1 = (u32)atomic_fetch_or((u32 *)&val, 2)
Quoted 22 lines
tools/testing/selftests/bpf/verifier/atomic_or.c +17 -0
+            BPF_MOV32_IMM(BPF_REG_1, 2),
+            BPF_ATOMIC_OP(BPF_W, BPF_OR | BPF_FETCH, BPF_REG_10, BPF_REG_1, -8),
+            /* r2 = 0x00000000FFFFFFFF; */
+            BPF_MOV64_IMM(BPF_REG_2, 1),
+            BPF_ALU64_IMM(BPF_LSH, BPF_REG_2, 32),
+            BPF_ALU64_IMM(BPF_SUB, BPF_REG_2, 1),
+            /* if (r2 != r1) exit(1); */
+            BPF_JMP_REG(BPF_JEQ, BPF_REG_2, BPF_REG_1, 2),
+            /* BPF_MOV32_IMM(BPF_REG_0, 1), */
+            BPF_MOV64_REG(BPF_REG_0, BPF_REG_1),
+            BPF_EXIT_INSN(),
+            /* exit(0); */
+            BPF_MOV32_IMM(BPF_REG_0, 0),
+            BPF_EXIT_INSN(),
+    },
+    .result = ACCEPT,
+},

base-commit: f2cfe32e8a965a86e512dcb2e6251371d4a60c63 --
2.30.1.766.gb4fecdf3b7-goog