helper-functions: add newer bpf_base_func_proto helpers to base - #301

The base group in data/helpers-functions.yaml mirrors bpf_base_func_proto(), but it is missing helpers that the kernel has handed to every program type for a while. For example, the bpf_get_current_comm page lists only tracing, LSM, syscall and cgroup sock types, while XDP and TC programs can call it since v6.16.

Added to base, with since set to the commit that moved each helper into bpf_base_func_proto():

• bpf_strtol, bpf_strtoul: v6.1, 8a67f2de9b1d ("bpf: expose bpf_strtol and bpf_strtoul to all program types")
• bpf_perf_event_read_value: v6.15, ae0a457f5d33
• bpf_get_current_uid_gid, bpf_current_task_under_cgroup, bpf_get_cgroup_classid, bpf_task_storage_get, bpf_task_storage_delete, bpf_get_current_comm, bpf_perf_event_read, bpf_send_signal, bpf_send_signal_thread, bpf_get_task_stack, bpf_get_branch_snapshot, bpf_find_vma: v6.16, ee971630f20f ("bpf: Allow some trace helpers for all prog types")

cap and kconfig follow the section of bpf_base_func_proto() each helper is in (no capability, CAP_BPF, or CAP_BPF + CAP_PERFMON; CONFIG_CGROUPS / CONFIG_CGROUP_NET_CLASSID). Program types that already list one of these helpers keep their own entry, because the generator lets a parent definition override the base one.

I left out bpf_copy_from_user and bpf_copy_from_user_task. The same v6.16 commit added them to the base set, but they are might_sleep helpers, and the verifier rejects them in non-sleepable programs ("helper call might sleep in a non-sleepable prog"). Of the program types that use the base group, only struct_ops can be sleepable. I can add them for BPF_PROG_TYPE_STRUCT_OPS only if you prefer.

The program type and helper pages were regenerated with tools/helper-ref-gen (built from this repo). No other generated files changed.

How I checked it:

• I ran bpftool feature probe (bpftool v7.7.0 / libbpf 1.7) on Ubuntu 25.04 (6.14.0-37-generic) and Ubuntu 26.04 (7.0.0-38-generic). On 7.0, the v6.15/v6.16 helpers above show up for socket_filter, sched_cls, sched_act, xdp, cgroup_skb, lwt_*, sock_ops, sk_skb, sk_msg, flow_dissector, sk_reuseport, sk_lookup and netfilter (where the type didn't already have them). On 6.14 they don't. bpf_strtol/bpf_strtoul show up on both kernels.
• I resolved the YAML per program type the same way the generator does and compared the result with the 7.0 probe output for these 15 helpers. Before this change, 314 (program type, helper) pairs that the kernel reports are missing from the docs. After the change, none are missing, and no pairs are documented that the kernel doesn't report. bpftool cannot probe tracing, LSM or struct_ops, so I did not check those three types this way.
• bpftool also lists bpf_copy_from_user for XDP etc. on 7.0. That is because its heuristic treats any verifier error other than "unknown func" / "cannot use helper" as supported.

This change was drafted with AI assistance and reviewed by me.

🤖 Generated with Claude Code