Skip to contentSkip to conversation

bpf: Fix ld_{abs,ind} failure path analysis in subprogs

PATCH Applied Philo Lu 3 participants 7 messages Jul 22 – Jul 28

Subsystems (3)
lore
Share & tools
lore

Philo Lu

author
#0 lore raw

cover bpf: Fix ld_{abs,ind} failure path analysis in subprogs

Backport of the CVE-2026-53090 fix.

1/2 ee861486e377, the CVE fix.
2/2 e0fcb42bc6f4, selftests for the fix.

Daniel Borkmann (2):
bpf: Fix ld_{abs,ind} failure path analysis in subprogs selftests/bpf: Add tests for ld_{abs,ind} failure path in subprogs

 kernel/bpf/verifier.c                         |  17 +++
 .../selftests/bpf/progs/verifier_ld_ind.c     | 142 ++++++++++++++++++
 2 files changed, 159 insertions(+)

--
2.47.3

author
#1 lore raw

PATCH 1/2 bpf: Fix ld_{abs,ind} failure path analysis in subprogs

From: Daniel Borkmann <[email protected]>

commit ee861486e377edc55361c08dcbceab3f6b6577bd upstream.

Usage of ld_{abs,ind} instructions got extended into subprogs some time ago via commit 09b28d76eac4 ("bpf: Add abnormal return checks."). These are only allowed in subprograms when the latter are BTF annotated and have scalar return types.

The code generator in bpf_gen_ld_abs() has an abnormal exit path (r0=0 + exit) from legacy cBPF times. While the enforcement is on scalar return types, the verifier must also simulate the path of abnormal exit if the packet data load via ld_{abs,ind} failed.

This is currently not the case. Fix it by having the verifier simulate both success and failure paths, and extend it in similar ways as we do for tail calls. The success path (r0=unknown, continue to next insn) is pushed onto stack for later validation and the r0=0 and return to the caller is done on the fall-through side.

[ Dropped visit_abnormal_return_insn changes: depends on 7.0 symbols from
e40f5a6bf88a ("bpf: correct stack liveness for tail calls");
Hunk1: adapted IS_ERR/PTR_ERR to !branch/-EFAULT to match push_stack() NULL-on-failure convention. ]

---
 kernel/bpf/verifier.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
kernel/bpf/verifier.c +17 -0
index f638b2d3a42fb..3f74767e47484 100644
@@ -15119,6 +15119,23 @@ static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
 	mark_reg_unknown(env, regs, BPF_REG_0);
 	/* ld_abs load up to 32-bit skb data. */
 	regs[BPF_REG_0].subreg_def = env->insn_idx + 1;
+	/*
+	 * See bpf_gen_ld_abs() which emits a hidden BPF_EXIT with r0=0
+	 * which must be explored by the verifier when in a subprog.
+	 */
+	if (env->cur_state->curframe) {
+		struct bpf_verifier_state *branch;
+
+		mark_reg_scratched(env, BPF_REG_0);
+		branch = push_stack(env, env->insn_idx + 1, env->insn_idx, false);
+		if (!branch)
+			return -EFAULT;
+		mark_reg_known_zero(env, regs, BPF_REG_0);
+		err = prepare_func_exit(env, &env->insn_idx);
+		if (err)
+			return err;
+		env->insn_idx--;
+	}
 	return 0;
 }
 

--
2.47.3

Reviewed-by on v1 1/2
#2 lore raw
Quoted 25 lines · Philo Lu

From: Daniel Borkmann <[email protected]>

commit ee861486e377edc55361c08dcbceab3f6b6577bd upstream.

Usage of ld_{abs,ind} instructions got extended into subprogs some time ago via commit 09b28d76eac4 ("bpf: Add abnormal return checks."). These are only allowed in subprograms when the latter are BTF annotated and have scalar return types.

The code generator in bpf_gen_ld_abs() has an abnormal exit path (r0=0 + exit) from legacy cBPF times. While the enforcement is on scalar return types, the verifier must also simulate the path of abnormal exit if the packet data load via ld_{abs,ind} failed.

This is currently not the case. Fix it by having the verifier simulate both success and failure paths, and extend it in similar ways as we do for tail calls. The success path (r0=unknown, continue to next insn) is pushed onto stack for later validation and the r0=0 and return to the caller is done on the fall-through side.

Quoted 4 lines

[ Dropped visit_abnormal_return_insn changes: depends on 7.0 symbols from
e40f5a6bf88a ("bpf: correct stack liveness for tail calls");
Hunk1: adapted IS_ERR/PTR_ERR to !branch/-EFAULT to match push_stack() NULL-on-failure convention. ]

These versions does not introduce consumers corresponding to the output of jump tables; therefore, the assignment of jump tables is not involved.

Quoted 37 lines
---
  kernel/bpf/verifier.c | 17 +++++++++++++++++
  1 file changed, 17 insertions(+)
kernel/bpf/verifier.c +17 -1
index f638b2d3a42fb..3f74767e47484 100644
@@ -15119,6 +15119,23 @@ static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
  	mark_reg_unknown(env, regs, BPF_REG_0);
  	/* ld_abs load up to 32-bit skb data. */
  	regs[BPF_REG_0].subreg_def = env->insn_idx + 1;
+	/*
+	 * See bpf_gen_ld_abs() which emits a hidden BPF_EXIT with r0=0
+	 * which must be explored by the verifier when in a subprog.
+	 */
+	if (env->cur_state->curframe) {
+		struct bpf_verifier_state *branch;
+
+		mark_reg_scratched(env, BPF_REG_0);
+		branch = push_stack(env, env->insn_idx + 1, env->insn_idx, false);
+		if (!branch)
+			return -EFAULT;
+		mark_reg_known_zero(env, regs, BPF_REG_0);
+		err = prepare_func_exit(env, &env->insn_idx);
+		if (err)
+			return err;
+		env->insn_idx--;
+	}
  	return 0;
  }
 
--

2.47.3

author
#3 lore raw

PATCH 2/2 selftests/bpf: Add tests for ld_{abs,ind} failure path in subprogs

In reply to Philo Lu

From: Daniel Borkmann <[email protected]>

commit e0fcb42bc6f41bab2895757d6610616b3820eff7 upstream.

Extend the verifier_ld_ind BPF selftests with subprogs containing ld_{abs,ind} and craft the test in a way where the invalid register read is rejected in the fixed case. Also add a success case each, and add additional coverage related to the BTF return type enforcement.

# LDLIBS=-static PKG_CONFIG='pkg-config --static' ./vmtest.sh -- ./test_progs -t verifier_ld_ind [...]

  #611/1   verifier_ld_ind/ld_ind: check calling conv, r1:OK
  #611/2   verifier_ld_ind/ld_ind: check calling conv, r1 @unpriv:OK
  #611/3   verifier_ld_ind/ld_ind: check calling conv, r2:OK
  #611/4   verifier_ld_ind/ld_ind: check calling conv, r2 @unpriv:OK
  #611/5   verifier_ld_ind/ld_ind: check calling conv, r3:OK
  #611/6   verifier_ld_ind/ld_ind: check calling conv, r3 @unpriv:OK
  #611/7   verifier_ld_ind/ld_ind: check calling conv, r4:OK
  #611/8   verifier_ld_ind/ld_ind: check calling conv, r4 @unpriv:OK
  #611/9   verifier_ld_ind/ld_ind: check calling conv, r5:OK
  #611/10  verifier_ld_ind/ld_ind: check calling conv, r5 @unpriv:OK
  #611/11  verifier_ld_ind/ld_ind: check calling conv, r7:OK
  #611/12  verifier_ld_ind/ld_ind: check calling conv, r7 @unpriv:OK
  #611/13  verifier_ld_ind/ld_abs: subprog early exit on ld_abs failure:OK
  #611/14  verifier_ld_ind/ld_ind: subprog early exit on ld_ind failure:OK
  #611/15  verifier_ld_ind/ld_abs: subprog with both paths safe:OK
  #611/16  verifier_ld_ind/ld_ind: subprog with both paths safe:OK
  #611/17  verifier_ld_ind/ld_abs: reject void return subprog:OK
  #611/18  verifier_ld_ind/ld_ind: reject void return subprog:OK
  #611     verifier_ld_ind:OK

Summary: 1/18 PASSED, 0 SKIPPED, 0 FAILED

---
 .../selftests/bpf/progs/verifier_ld_ind.c     | 142 ++++++++++++++++++
 1 file changed, 142 insertions(+)
tools/testing/selftests/bpf/progs/verifier_ld_ind.c +142 -0
index c925ba9a2e74c..09e81b99eecb4 100644
@@ -107,4 +107,146 @@ __naked void ind_check_calling_conv_r7(void)
 	: __clobber_all);
 }
 
+/*
+ * ld_{abs,ind} subprog that always sets r0=1 on the success path.
+ * bpf_gen_ld_abs() emits a hidden exit with r0=0 when the load helper
+ * fails. The verifier must model this failure return so that callers
+ * account for r0=0 as a possible return value.
+ */
+__naked __noinline __used
+static int ldabs_subprog(void)
+{
+	asm volatile (
+	"r6 = r1;"
+	".8byte %[ld_abs];"
+	"r0 = 1;"
+	"exit;"
+	:
+	: __imm_insn(ld_abs, BPF_LD_ABS(BPF_W, 0))
+	: __clobber_all);
+}
+
+__naked __noinline __used
+static int ldind_subprog(void)
+{
+	asm volatile (
+	"r6 = r1;"
+	"r7 = 0;"
+	".8byte %[ld_ind];"
+	"r0 = 1;"
+	"exit;"
+	:
+	: __imm_insn(ld_ind, BPF_LD_IND(BPF_W, BPF_REG_7, 0))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_abs: subprog early exit on ld_abs failure")
+__failure __msg("R9 !read_ok")
+__naked void ld_abs_subprog_early_exit(void)
+{
+	asm volatile (
+	"call ldabs_subprog;"
+	"if r0 != 0 goto l_exit_%=;"
+	"r0 = r9;"
+	"l_exit_%=:"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_ind: subprog early exit on ld_ind failure")
+__failure __msg("R9 !read_ok")
+__naked void ld_ind_subprog_early_exit(void)
+{
+	asm volatile (
+	"call ldind_subprog;"
+	"if r0 != 0 goto l_exit_%=;"
+	"r0 = r9;"
+	"l_exit_%=:"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_abs: subprog with both paths safe")
+__success
+__naked void ld_abs_subprog_both_paths_safe(void)
+{
+	asm volatile (
+	"call ldabs_subprog;"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_ind: subprog with both paths safe")
+__success
+__naked void ld_ind_subprog_both_paths_safe(void)
+{
+	asm volatile (
+	"call ldind_subprog;"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
+/*
+ * ld_{abs,ind} in subprogs require scalar (int) return type in BTF.
+ * A test with void return must be rejected.
+ */
+__naked __noinline __used
+static void ldabs_void_subprog(void)
+{
+	asm volatile (
+	"r6 = r1;"
+	".8byte %[ld_abs];"
+	"r0 = 1;"
+	"exit;"
+	:
+	: __imm_insn(ld_abs, BPF_LD_ABS(BPF_W, 0))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_abs: reject void return subprog")
+__failure __msg("LD_ABS is only allowed in functions that return 'int'")
+__naked void ld_abs_void_subprog_reject(void)
+{
+	asm volatile (
+	"call ldabs_void_subprog;"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
+__naked __noinline __used
+static void ldind_void_subprog(void)
+{
+	asm volatile (
+	"r6 = r1;"
+	"r7 = 0;"
+	".8byte %[ld_ind];"
+	"r0 = 1;"
+	"exit;"
+	:
+	: __imm_insn(ld_ind, BPF_LD_IND(BPF_W, BPF_REG_7, 0))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_ind: reject void return subprog")
+__failure __msg("LD_ABS is only allowed in functions that return 'int'")
+__naked void ld_ind_void_subprog_reject(void)
+{
+	asm volatile (
+	"call ldind_void_subprog;"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
 char _license[] SEC("license") = "GPL";
Signature · 1 line
2.47.3
Reviewed-by on v1 2/2
#4 lore raw

Re: [PATCH 6.6.y/6.12.y/6.18.y 2/2] selftests/bpf: Add tests for ld_{abs,ind} failure path in subprogs

Quoted 30 lines · Philo Lu

From: Daniel Borkmann <[email protected]>

commit e0fcb42bc6f41bab2895757d6610616b3820eff7 upstream.

Extend the verifier_ld_ind BPF selftests with subprogs containing ld_{abs,ind} and craft the test in a way where the invalid register read is rejected in the fixed case. Also add a success case each, and add additional coverage related to the BTF return type enforcement.

# LDLIBS=-static PKG_CONFIG='pkg-config --static' ./vmtest.sh -- ./test_progs -t verifier_ld_ind [...]

   #611/1   verifier_ld_ind/ld_ind: check calling conv, r1:OK
   #611/2   verifier_ld_ind/ld_ind: check calling conv, r1 @unpriv:OK
   #611/3   verifier_ld_ind/ld_ind: check calling conv, r2:OK
   #611/4   verifier_ld_ind/ld_ind: check calling conv, r2 @unpriv:OK
   #611/5   verifier_ld_ind/ld_ind: check calling conv, r3:OK
   #611/6   verifier_ld_ind/ld_ind: check calling conv, r3 @unpriv:OK
   #611/7   verifier_ld_ind/ld_ind: check calling conv, r4:OK
   #611/8   verifier_ld_ind/ld_ind: check calling conv, r4 @unpriv:OK
   #611/9   verifier_ld_ind/ld_ind: check calling conv, r5:OK
   #611/10  verifier_ld_ind/ld_ind: check calling conv, r5 @unpriv:OK
   #611/11  verifier_ld_ind/ld_ind: check calling conv, r7:OK
   #611/12  verifier_ld_ind/ld_ind: check calling conv, r7 @unpriv:OK
   #611/13  verifier_ld_ind/ld_abs: subprog early exit on ld_abs failure:OK
   #611/14  verifier_ld_ind/ld_ind: subprog early exit on ld_ind failure:OK
   #611/15  verifier_ld_ind/ld_abs: subprog with both paths safe:OK
   #611/16  verifier_ld_ind/ld_ind: subprog with both paths safe:OK
   #611/17  verifier_ld_ind/ld_abs: reject void return subprog:OK
   #611/18  verifier_ld_ind/ld_ind: reject void return subprog:OK
   #611     verifier_ld_ind:OK
Quoted 6 lines

Summary: 1/18 PASSED, 0 SKIPPED, 0 FAILED

Quoted 155 lines
---
  .../selftests/bpf/progs/verifier_ld_ind.c     | 142 ++++++++++++++++++
  1 file changed, 142 insertions(+)
tools/testing/selftests/bpf/progs/verifier_ld_ind.c +142 -0
index c925ba9a2e74c..09e81b99eecb4 100644
@@ -107,4 +107,146 @@ __naked void ind_check_calling_conv_r7(void)
  	: __clobber_all);
  }
 
+/*
+ * ld_{abs,ind} subprog that always sets r0=1 on the success path.
+ * bpf_gen_ld_abs() emits a hidden exit with r0=0 when the load helper
+ * fails. The verifier must model this failure return so that callers
+ * account for r0=0 as a possible return value.
+ */
+__naked __noinline __used
+static int ldabs_subprog(void)
+{
+	asm volatile (
+	"r6 = r1;"
+	".8byte %[ld_abs];"
+	"r0 = 1;"
+	"exit;"
+	:
+	: __imm_insn(ld_abs, BPF_LD_ABS(BPF_W, 0))
+	: __clobber_all);
+}
+
+__naked __noinline __used
+static int ldind_subprog(void)
+{
+	asm volatile (
+	"r6 = r1;"
+	"r7 = 0;"
+	".8byte %[ld_ind];"
+	"r0 = 1;"
+	"exit;"
+	:
+	: __imm_insn(ld_ind, BPF_LD_IND(BPF_W, BPF_REG_7, 0))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_abs: subprog early exit on ld_abs failure")
+__failure __msg("R9 !read_ok")
+__naked void ld_abs_subprog_early_exit(void)
+{
+	asm volatile (
+	"call ldabs_subprog;"
+	"if r0 != 0 goto l_exit_%=;"
+	"r0 = r9;"
+	"l_exit_%=:"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_ind: subprog early exit on ld_ind failure")
+__failure __msg("R9 !read_ok")
+__naked void ld_ind_subprog_early_exit(void)
+{
+	asm volatile (
+	"call ldind_subprog;"
+	"if r0 != 0 goto l_exit_%=;"
+	"r0 = r9;"
+	"l_exit_%=:"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_abs: subprog with both paths safe")
+__success
+__naked void ld_abs_subprog_both_paths_safe(void)
+{
+	asm volatile (
+	"call ldabs_subprog;"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_ind: subprog with both paths safe")
+__success
+__naked void ld_ind_subprog_both_paths_safe(void)
+{
+	asm volatile (
+	"call ldind_subprog;"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
+/*
+ * ld_{abs,ind} in subprogs require scalar (int) return type in BTF.
+ * A test with void return must be rejected.
+ */
+__naked __noinline __used
+static void ldabs_void_subprog(void)
+{
+	asm volatile (
+	"r6 = r1;"
+	".8byte %[ld_abs];"
+	"r0 = 1;"
+	"exit;"
+	:
+	: __imm_insn(ld_abs, BPF_LD_ABS(BPF_W, 0))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_abs: reject void return subprog")
+__failure __msg("LD_ABS is only allowed in functions that return 'int'")
+__naked void ld_abs_void_subprog_reject(void)
+{
+	asm volatile (
+	"call ldabs_void_subprog;"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
+__naked __noinline __used
+static void ldind_void_subprog(void)
+{
+	asm volatile (
+	"r6 = r1;"
+	"r7 = 0;"
+	".8byte %[ld_ind];"
+	"r0 = 1;"
+	"exit;"
+	:
+	: __imm_insn(ld_ind, BPF_LD_IND(BPF_W, BPF_REG_7, 0))
+	: __clobber_all);
+}
+
+SEC("socket")
+__description("ld_ind: reject void return subprog")
+__failure __msg("LD_ABS is only allowed in functions that return 'int'")
+__naked void ld_ind_void_subprog_reject(void)
+{
+	asm volatile (
+	"call ldind_void_subprog;"
+	"r0 = 0;"
+	"exit;"
+	::: __clobber_all);
+}
+
  char _license[] SEC("license") = "GPL";
maintainer Applied
#5 lore raw
In reply to Philo Lu
Quoted 4 lines

Backport of the CVE-2026-53090 fix.

1/2 ee861486e377, the CVE fix.
2/2 e0fcb42bc6f4, selftests for the fix.

Queued the series for 6.6, 6.12, and 6.18, thanks.

One follow-up: the Fixes commit 09b28d76eac4 ("bpf: Add abnormal return checks.") is also present in 6.1, 5.15 and 5.10, and the fix is absent there, so those trees are affected as well but were not covered by this submission. They likely need a further adaptation. Could you extend coverage with adapted backports for 6.1/5.15/5.10?

Signature · 2 lines
Thanks,
Sasha
author
#6 lore raw
Quoted 6 lines · Sasha Levin

One follow-up: the Fixes commit 09b28d76eac4 ("bpf: Add abnormal return checks.") is also present in 6.1, 5.15 and 5.10, and the fix is absent there, so those trees are affected as well but were not covered by this submission. They likely need a further adaptation. Could you extend coverage with adapted backports for 6.1/5.15/5.10?

I've posted patches for 6.1/5.15/5.10. The 6.1 one is identical; 5.10/5.15 required minor adaptation.

The selftests patch can't go into any of the three trees due to missing dependencies.

Thanks.

Signature · 1 line
Philo