Re: [PATCH bpf-next v5 2/2] bpf: fix stackmap overflow check in __bpf_get_stackid()

[Song Liu <song@kernel.org>]

On Fri, Aug 29, 2025 at 10:29 AM Alexei Starovoitov
<alexei.starovoitov@gmail.com> wrote:
[...]
> >
> >  static long __bpf_get_stackid(struct bpf_map *map,
> > -                             struct perf_callchain_entry *trace, u64 flags)
> > +                             struct perf_callchain_entry *trace, u64 flags, u32 max_depth)
> >  {
> >         struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
> >         struct stack_map_bucket *bucket, *new_bucket, *old_bucket;
> > @@ -263,6 +263,8 @@ static long __bpf_get_stackid(struct bpf_map *map,
> >
> >         trace_nr = trace->nr - skip;
> >         trace_len = trace_nr * sizeof(u64);
> > +       trace_nr = min(trace_nr, max_depth - skip);
> > +
>
> The patch might have fixed this particular syzbot repro
> with OOB in stackmap-with-buildid case,
> but above two line looks wrong.
> trace_len is computed before being capped by max_depth.
> So non-buildid case below is using
> memcpy(new_bucket->data, ips, trace_len);
>
> so OOB is still there?

+1 for this observation.

We are calling __bpf_get_stackid() from two functions: bpf_get_stackid
and bpf_get_stackid_pe. The check against max_depth is only needed
from bpf_get_stackid_pe, so it is better to just check here.

I have got the following on top of patch 1/2. This makes more sense to
me.

PS: The following also includes some clean up in __bpf_get_stack.
I include those because it also uses stack_map_calculate_max_depth.

Does this look better?

Thanks,
Song


diff --git c/kernel/bpf/stackmap.c w/kernel/bpf/stackmap.c
index 796cc105eacb..08554fb146e1 100644
--- c/kernel/bpf/stackmap.c
+++ w/kernel/bpf/stackmap.c
@@ -262,7 +262,7 @@ static long __bpf_get_stackid(struct bpf_map *map,
                return -EFAULT;

        trace_nr = trace->nr - skip;
-       trace_len = trace_nr * sizeof(u64);
+
        ips = trace->ip + skip;
        hash = jhash2((u32 *)ips, trace_len / sizeof(u32), 0);
        id = hash & (smap->n_buckets - 1);
@@ -297,6 +297,7 @@ static long __bpf_get_stackid(struct bpf_map *map,
                        return -EEXIST;
                }
        } else {
+               trace_len = trace_nr * sizeof(u64);
                if (hash_matches && bucket->nr == trace_nr &&
                    memcmp(bucket->data, ips, trace_len) == 0)
                        return id;
@@ -322,19 +323,17 @@ static long __bpf_get_stackid(struct bpf_map *map,
 BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
           u64, flags)
 {
-       u32 max_depth = map->value_size / stack_map_data_size(map);
-       u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
+       u32 elem_size = stack_map_data_size(map);
        bool user = flags & BPF_F_USER_STACK;
        struct perf_callchain_entry *trace;
        bool kernel = !user;
+       u32 max_depth;

        if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
                               BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
                return -EINVAL;

-       max_depth += skip;
-       if (max_depth > sysctl_perf_event_max_stack)
-               max_depth = sysctl_perf_event_max_stack;
+       max_depth = stack_map_calculate_max_depth(map->value_size,
elem_size, flags);

        trace = get_perf_callchain(regs, 0, kernel, user, max_depth,
                                   false, false);
@@ -375,6 +374,7 @@ BPF_CALL_3(bpf_get_stackid_pe, struct
bpf_perf_event_data_kern *, ctx,
        bool kernel, user;
        __u64 nr_kernel;
        int ret;
+       u32 elem_size, max_depth;

        /* perf_sample_data doesn't have callchain, use bpf_get_stackid */
        if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN))
@@ -393,11 +393,12 @@ BPF_CALL_3(bpf_get_stackid_pe, struct
bpf_perf_event_data_kern *, ctx,
                return -EFAULT;

        nr_kernel = count_kernel_ip(trace);
-
+       elem_size = stack_map_data_size(map);
        if (kernel) {
                __u64 nr = trace->nr;

-               trace->nr = nr_kernel;
+               max_depth =
stack_map_calculate_max_depth(map->value_size, elem_size, flags);
+               trace->nr = min_t(u32, nr_kernel, max_depth);
                ret = __bpf_get_stackid(map, trace, flags);

                /* restore nr */
@@ -410,6 +411,8 @@ BPF_CALL_3(bpf_get_stackid_pe, struct
bpf_perf_event_data_kern *, ctx,
                        return -EFAULT;

                flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
+               max_depth =
stack_map_calculate_max_depth(map->value_size, elem_size, flags);
+               trace->nr = min_t(u32, trace->nr, max_depth);
                ret = __bpf_get_stackid(map, trace, flags);
        }
        return ret;
@@ -428,7 +431,7 @@ static long __bpf_get_stack(struct pt_regs *regs,
struct task_struct *task,
                            struct perf_callchain_entry *trace_in,
                            void *buf, u32 size, u64 flags, bool may_fault)
 {
-       u32 trace_nr, copy_len, elem_size, num_elem, max_depth;
+       u32 trace_nr, copy_len, elem_size, max_depth;
        bool user_build_id = flags & BPF_F_USER_BUILD_ID;
        bool crosstask = task && task != current;
        u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
@@ -465,13 +468,15 @@ static long __bpf_get_stack(struct pt_regs
*regs, struct task_struct *task,
        if (may_fault)
                rcu_read_lock(); /* need RCU for perf's callchain below */

-       if (trace_in)
+       if (trace_in) {
                trace = trace_in;
-       else if (kernel && task)
+               trace->nr = min_t(u32, trace->nr, max_depth);
+       } else if (kernel && task) {
                trace = get_callchain_entry_for_task(task, max_depth);
-       else
+       } else {
                trace = get_perf_callchain(regs, 0, kernel, user, max_depth,
                                           crosstask, false);
+       }

        if (unlikely(!trace) || trace->nr < skip) {
                if (may_fault)
@@ -479,9 +484,7 @@ static long __bpf_get_stack(struct pt_regs *regs,
struct task_struct *task,
                goto err_fault;
        }

-       num_elem = size / elem_size;
        trace_nr = trace->nr - skip;
-       trace_nr = (trace_nr <= num_elem) ? trace_nr : num_elem;
        copy_len = trace_nr * elem_size;

        ips = trace->ip + skip;