Skip to contentSkip to conversation

bpf: verifier: initialize imm in kfunc_tab in add_kfunc_call()

PATCH Superseded Puranjay Mohan 3 participants 3 messages Nov 11, 2025 – Nov 12, 2025

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 bpf: verifier: initialize imm in kfunc_tab in add_kfunc_call()

Metadata about a kfunc call is added to the kfunc_tab in add_kfunc_call() but the call instruction itself could get removed by opt_remove_dead_code() later if it is not reachable.

If the call instruction is removed, specialize_kfunc() is never called for it and the desc->imm in the kfunc_tab is never initialized for this kfunc call. In this case, sort_kfunc_descs_by_imm_off(env->prog); in do_misc_fixups() doesn't sort the table correctly.
This is a problem from s390 as its JIT uses this table to find the addresses for kfuncs, and if this table is not sorted properly, JIT can fail to find addresses for valid kfunc calls.

This was exposed by:

commit d869d56ca848 ("bpf: verifier: refactor kfunc specialization")

as before this commit, desc->imm was initialised in add_kfunc_call().

Initialize desc->imm to func_id, it will be overwritten in specialize_kfunc() if the instruction is not removed.

Fixes d869d56ca848 ("bpf: verifier: refactor kfunc specialization")

---

This bug is not triggered by the CI currently, I am working on another set for non-sleepbale arena allocations and as part of that I am adding a new selftest that triggers this bug.

Selftest: https://github.com/kernel-patches/bpf/pull/10242/commits/1f681f022c6d685fd76695e5eafbe9d9ab4c0002
CI run: https://github.com/kernel-patches/bpf/actions/runs/19238699806/job/54996376908

---
 kernel/bpf/verifier.c | 1 +
 1 file changed, 1 insertion(+)
kernel/bpf/verifier.c +1 -0
index 1268fa075d4c..a667f761173c 100644
@@ -3371,6 +3371,7 @@ static int add_kfunc_call(struct bpf_verifier_env *env, u32 func_id, s16 offset)
 
 	desc = &tab->descs[tab->nr_descs++];
 	desc->func_id = func_id;
+	desc->imm = func_id;
 	desc->offset = offset;
 	desc->addr = addr;
 	desc->func_model = func_model;
Signature · 1 line
2.47.3
#1 lore raw
Quoted 41 lines · Puranjay Mohan

Metadata about a kfunc call is added to the kfunc_tab in add_kfunc_call() but the call instruction itself could get removed by opt_remove_dead_code() later if it is not reachable.

If the call instruction is removed, specialize_kfunc() is never called for it and the desc->imm in the kfunc_tab is never initialized for this kfunc call. In this case, sort_kfunc_descs_by_imm_off(env->prog); in do_misc_fixups() doesn't sort the table correctly.
This is a problem from s390 as its JIT uses this table to find the addresses for kfuncs, and if this table is not sorted properly, JIT can fail to find addresses for valid kfunc calls.

This was exposed by:

commit d869d56ca848 ("bpf: verifier: refactor kfunc specialization")

as before this commit, desc->imm was initialised in add_kfunc_call().

Initialize desc->imm to func_id, it will be overwritten in specialize_kfunc() if the instruction is not removed.

Fixes d869d56ca848 ("bpf: verifier: refactor kfunc specialization")

---

This bug is not triggered by the CI currently, I am working on another set for non-sleepbale arena allocations and as part of that I am adding a new selftest that triggers this bug.

Selftest: https://github.com/kernel-patches/bpf/pull/10242/commits/1f681f022c6d685fd76695e5eafbe9d9ab4c0002
CI run: https://github.com/kernel-patches/bpf/actions/runs/19238699806/job/54996376908

---
  kernel/bpf/verifier.c | 1 +
  1 file changed, 1 insertion(+)
kernel/bpf/verifier.c +0 -0
index 1268fa075d4c..a667f761173c 100644

kernel/bpf/verifier.c static int add_kfunc_call(struct bpf_verifier_env *env, u32 func_id, s16 offset) · line 3377

Quoted 7 lines
kernel/bpf/verifier.c +1 -0
@@ -3371,6 +3371,7 @@ static int add_kfunc_call(struct bpf_verifier_env *env, u32 func_id, s16 offset)
 
  	desc = &tab->descs[tab->nr_descs++];
  	desc->func_id = func_id;
+	desc->imm = func_id;
  	desc->offset = offset;
  	desc->addr = addr;
  	desc->func_model = func_model;

Thanks for sending the fix.
I'm not sure if this is enough, though. Don't you need to run entire calculation for the desc->imm, like in the original, before
d869d56ca848 ("bpf: verifier: refactor kfunc specialization")?

    if (bpf_jit_supports_far_kfunc_call()) {
        call_imm = func_id;
    } else {
        call_imm = BPF_CALL_IMM(addr);
        /* Check whether the relative offset overflows desc->imm */
        if ((unsigned long)(s32)call_imm != call_imm) {
            verbose(env, "address of kernel func_id %u is out of range\n", func_id); return -EINVAL;
        }
    }
    desc->imm = call_imm;

I think it would be right to reuse that hunk in both places:
 add_kfunc_call() and specialize_kfunc().

#2 lore raw
Quoted 69 lines · Mykyta Yatsenko
Quoted 41 lines · Puranjay Mohan

Metadata about a kfunc call is added to the kfunc_tab in add_kfunc_call() but the call instruction itself could get removed by opt_remove_dead_code() later if it is not reachable.

If the call instruction is removed, specialize_kfunc() is never called for it and the desc->imm in the kfunc_tab is never initialized for this kfunc call. In this case, sort_kfunc_descs_by_imm_off(env->prog); in do_misc_fixups() doesn't sort the table correctly.
This is a problem from s390 as its JIT uses this table to find the addresses for kfuncs, and if this table is not sorted properly, JIT can fail to find addresses for valid kfunc calls.

This was exposed by:

commit d869d56ca848 ("bpf: verifier: refactor kfunc specialization")

as before this commit, desc->imm was initialised in add_kfunc_call().

Initialize desc->imm to func_id, it will be overwritten in specialize_kfunc() if the instruction is not removed.

Fixes d869d56ca848 ("bpf: verifier: refactor kfunc specialization")

---

This bug is not triggered by the CI currently, I am working on another set for non-sleepbale arena allocations and as part of that I am adding a new selftest that triggers this bug.

Selftest: https://github.com/kernel-patches/bpf/pull/10242/commits/1f681f022c6d685fd76695e5eafbe9d9ab4c0002
CI run: https://github.com/kernel-patches/bpf/actions/runs/19238699806/job/54996376908

---
  kernel/bpf/verifier.c | 1 +
  1 file changed, 1 insertion(+)
kernel/bpf/verifier.c +0 -0
index 1268fa075d4c..a667f761173c 100644

kernel/bpf/verifier.c static int add_kfunc_call(struct bpf_verifier_env *env, u32 func_id, s16 offset) · line 3377

Quoted 7 lines
kernel/bpf/verifier.c +1 -0
@@ -3371,6 +3371,7 @@ static int add_kfunc_call(struct bpf_verifier_env *env, u32 func_id, s16 offset)
 
      desc = &tab->descs[tab->nr_descs++];
      desc->func_id = func_id;
+     desc->imm = func_id;
      desc->offset = offset;
      desc->addr = addr;
      desc->func_model = func_model;

Thanks for sending the fix.
I'm not sure if this is enough, though. Don't you need to run entire calculation
for the desc->imm, like in the original, before d869d56ca848 ("bpf: verifier: refactor kfunc specialization")?

     if (bpf_jit_supports_far_kfunc_call()) {
         call_imm = func_id;
     } else {
         call_imm = BPF_CALL_IMM(addr);
         /* Check whether the relative offset overflows desc->imm */
         if ((unsigned long)(s32)call_imm != call_imm) {
             verbose(env, "address of kernel func_id %u is out of
range\n", func_id);
             return -EINVAL;
         }
     }
     desc->imm = call_imm;
Quoted 2 lines

I think it would be right to reuse that hunk in both places:
add_kfunc_call() and specialize_kfunc().

Yes, I agree with you. Currently it doesn't make a difference because only s390 uses it
and it needs the func_id to be in desc->imm, but for completeness we should do the proper
calculation to not introduce bugs for later developments.

I will send v2 with this entire calculation. and we can remove:

 if (bpf_jit_supports_far_kfunc_call()) {
                call_imm = func_id;

from specialize_kfunc() as it is redundant.

Thanks,
Puranjay