History

bpf: Don't run arg-tracking analysis twice on main subprog

Because subprog 0, the main subprog, is considered a global function, we end up running the arg-tracking dataflow analysis twice on it. That results in slightly longer verification but mostly in more verbose verifier logs. This patch fixes it by keeping only the iteration over global subprogs. When running over all of Cilium's programs with BPF_LOG_LEVEL2, this reduces verbosity by ~20% on average.

Paul Chaignon committed 6 months ago

Commit 512809b1 parent 18fc6501 file · +7 −18Browse files

Unified · Split

kernel/bpf/liveness.c+7−18
@@ -1914,26 +1914,15 @@ int bpf_compute_subprog_arg_access(struct bpf_verifier_env *env) · 1913 unmodified lines
1914 return -ENOMEM;1914 return -ENOMEM;
1915 }1915 }
1916 1916
1917 instance = call_instance(env, NULL, 0, 0);
1918 if (IS_ERR(instance)) {
1919 err = PTR_ERR(instance);
1920 goto out;
1921 }
1922 err = analyze_subprog(env, NULL, info, instance, callsites);
1923 if (err)
1924 goto out;
1925
1926 /*1917 /*
1927 * Subprogs and callbacks that don't receive FP-derived arguments1918 * Analyze every subprog in reverse topological order (callers
1928 * cannot access ancestor stack frames, so they were skipped during1919 * before callees) so that each subprog is analyzed before its
1929 * the recursive walk above. Async callbacks (timer, workqueue) are1920 * callees, allowing the recursive walk inside analyze_subprog()
1930 * also not reachable from the main program's call graph. Analyze1921 * to naturally reach callees that receive FP-derived args.
1931 * all unvisited subprogs as independent roots at depth 0.
1932 *1922 *
1933 * Use reverse topological order (callers before callees) so that1923 * Subprogs and callbacks that don't receive FP-derived arguments
1934 * each subprog is analyzed before its callees, allowing the1924 * cannot access ancestor stack frames are analyzed independently.
1935 * recursive walk inside analyze_subprog() to naturally1925 * Async callbacks (timer, workqueue) are handled the same way.
1936 * reach nested callees that also lack FP-derived args.
1937 */1926 */
1938 for (k = env->subprog_cnt - 1; k >= 0; k--) {1927 for (k = env->subprog_cnt - 1; k >= 0; k--) {
1939 int sub = env->subprog_topo_order[k];1928 int sub = env->subprog_topo_order[k];