selftests/bpf: Add tests for bpf_rbtree_{root,left,right}
authorMartin KaFai Lau <martin.lau@kernel.org>
Tue, 6 May 2025 01:58:52 +0000 (18:58 -0700)
committerAlexei Starovoitov <ast@kernel.org>
Tue, 6 May 2025 17:21:05 +0000 (10:21 -0700)
This patch has a much simplified rbtree usage from the
kernel sch_fq qdisc. It has a "struct node_data" which can be
added to two different rbtrees which are ordered by different keys.

The test first populates both rbtrees. Then search for a lookup_key
from the "groot0" rbtree. Once the lookup_key is found, that node
refcount is taken. The node is then removed from another "groot1"
rbtree.

While searching the lookup_key, the test will also try to remove
all rbnodes in the path leading to the lookup_key.

The test_{root,left,right}_spinlock_true tests ensure that the
return value of the bpf_rbtree functions is a non_own_ref node pointer.
This is done by forcing an verifier error by calling a helper
bpf_jiffies64() while holding the spinlock. The tests then
check for the verifier message
"call bpf_rbtree...R0=rcu_ptr_or_null_node..."

The other test_{root,left,right}_spinlock_false tests ensure that
they must be called with spinlock held.

Suggested-by: Kumar Kartikeya Dwivedi <memxor@gmail.com> # Check non_own_ref marking
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Link: https://lore.kernel.org/r/20250506015857.817950-6-martin.lau@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
tools/testing/selftests/bpf/prog_tests/rbtree.c
tools/testing/selftests/bpf/progs/rbtree_search.c [new file with mode: 0644]

index 9818f06c97c5a11deb0d04cfbf09ba448fda0180..d8f3d7a45fe9f9e7c71c1dbc01f64adcae2c55de 100644 (file)
@@ -8,6 +8,7 @@
 #include "rbtree_fail.skel.h"
 #include "rbtree_btf_fail__wrong_node_type.skel.h"
 #include "rbtree_btf_fail__add_wrong_type.skel.h"
+#include "rbtree_search.skel.h"
 
 static void test_rbtree_add_nodes(void)
 {
@@ -187,3 +188,8 @@ void test_rbtree_fail(void)
 {
        RUN_TESTS(rbtree_fail);
 }
+
+void test_rbtree_search(void)
+{
+       RUN_TESTS(rbtree_search);
+}
diff --git a/tools/testing/selftests/bpf/progs/rbtree_search.c b/tools/testing/selftests/bpf/progs/rbtree_search.c
new file mode 100644 (file)
index 0000000..098ef97
--- /dev/null
@@ -0,0 +1,206 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2025 Meta Platforms, Inc. and affiliates. */
+
+#include <vmlinux.h>
+#include <bpf/bpf_helpers.h>
+#include "bpf_misc.h"
+#include "bpf_experimental.h"
+
+struct node_data {
+       struct bpf_refcount ref;
+       struct bpf_rb_node r0;
+       struct bpf_rb_node r1;
+       int key0;
+       int key1;
+};
+
+#define private(name) SEC(".data." #name) __hidden __attribute__((aligned(8)))
+private(A) struct bpf_spin_lock glock0;
+private(A) struct bpf_rb_root groot0 __contains(node_data, r0);
+
+private(B) struct bpf_spin_lock glock1;
+private(B) struct bpf_rb_root groot1 __contains(node_data, r1);
+
+#define rb_entry(ptr, type, member) container_of(ptr, type, member)
+#define NR_NODES 16
+
+int zero = 0;
+
+static bool less0(struct bpf_rb_node *a, const struct bpf_rb_node *b)
+{
+       struct node_data *node_a;
+       struct node_data *node_b;
+
+       node_a = rb_entry(a, struct node_data, r0);
+       node_b = rb_entry(b, struct node_data, r0);
+
+       return node_a->key0 < node_b->key0;
+}
+
+static bool less1(struct bpf_rb_node *a, const struct bpf_rb_node *b)
+{
+       struct node_data *node_a;
+       struct node_data *node_b;
+
+       node_a = rb_entry(a, struct node_data, r1);
+       node_b = rb_entry(b, struct node_data, r1);
+
+       return node_a->key1 < node_b->key1;
+}
+
+SEC("syscall")
+__retval(0)
+long rbtree_search(void *ctx)
+{
+       struct bpf_rb_node *rb_n, *rb_m, *gc_ns[NR_NODES];
+       long lookup_key = NR_NODES / 2;
+       struct node_data *n, *m;
+       int i, nr_gc = 0;
+
+       for (i = zero; i < NR_NODES && can_loop; i++) {
+               n = bpf_obj_new(typeof(*n));
+               if (!n)
+                       return __LINE__;
+
+               m = bpf_refcount_acquire(n);
+
+               n->key0 = i;
+               m->key1 = i;
+
+               bpf_spin_lock(&glock0);
+               bpf_rbtree_add(&groot0, &n->r0, less0);
+               bpf_spin_unlock(&glock0);
+
+               bpf_spin_lock(&glock1);
+               bpf_rbtree_add(&groot1, &m->r1, less1);
+               bpf_spin_unlock(&glock1);
+       }
+
+       n = NULL;
+       bpf_spin_lock(&glock0);
+       rb_n = bpf_rbtree_root(&groot0);
+       while (can_loop) {
+               if (!rb_n) {
+                       bpf_spin_unlock(&glock0);
+                       return __LINE__;
+               }
+
+               n = rb_entry(rb_n, struct node_data, r0);
+               if (lookup_key == n->key0)
+                       break;
+               if (nr_gc < NR_NODES)
+                       gc_ns[nr_gc++] = rb_n;
+               if (lookup_key < n->key0)
+                       rb_n = bpf_rbtree_left(&groot0, rb_n);
+               else
+                       rb_n = bpf_rbtree_right(&groot0, rb_n);
+       }
+
+       if (!n || lookup_key != n->key0) {
+               bpf_spin_unlock(&glock0);
+               return __LINE__;
+       }
+
+       for (i = 0; i < nr_gc; i++) {
+               rb_n = gc_ns[i];
+               gc_ns[i] = bpf_rbtree_remove(&groot0, rb_n);
+       }
+
+       m = bpf_refcount_acquire(n);
+       bpf_spin_unlock(&glock0);
+
+       for (i = 0; i < nr_gc; i++) {
+               rb_n = gc_ns[i];
+               if (rb_n) {
+                       n = rb_entry(rb_n, struct node_data, r0);
+                       bpf_obj_drop(n);
+               }
+       }
+
+       if (!m)
+               return __LINE__;
+
+       bpf_spin_lock(&glock1);
+       rb_m = bpf_rbtree_remove(&groot1, &m->r1);
+       bpf_spin_unlock(&glock1);
+       bpf_obj_drop(m);
+       if (!rb_m)
+               return __LINE__;
+       bpf_obj_drop(rb_entry(rb_m, struct node_data, r1));
+
+       return 0;
+}
+
+#define TEST_ROOT(dolock)                              \
+SEC("syscall")                                         \
+__failure __msg(MSG)                                   \
+long test_root_spinlock_##dolock(void *ctx)            \
+{                                                      \
+       struct bpf_rb_node *rb_n;                       \
+       __u64 jiffies = 0;                              \
+                                                       \
+       if (dolock)                                     \
+               bpf_spin_lock(&glock0);                 \
+       rb_n = bpf_rbtree_root(&groot0);                \
+       if (rb_n)                                       \
+               jiffies = bpf_jiffies64();              \
+       if (dolock)                                     \
+               bpf_spin_unlock(&glock0);               \
+                                                       \
+       return !!jiffies;                               \
+}
+
+#define TEST_LR(op, dolock)                            \
+SEC("syscall")                                         \
+__failure __msg(MSG)                                   \
+long test_##op##_spinlock_##dolock(void *ctx)          \
+{                                                      \
+       struct bpf_rb_node *rb_n;                       \
+       struct node_data *n;                            \
+       __u64 jiffies = 0;                              \
+                                                       \
+       bpf_spin_lock(&glock0);                         \
+       rb_n = bpf_rbtree_root(&groot0);                \
+       if (!rb_n) {                                    \
+               bpf_spin_unlock(&glock0);               \
+               return 1;                               \
+       }                                               \
+       n = rb_entry(rb_n, struct node_data, r0);       \
+       n = bpf_refcount_acquire(n);                    \
+       bpf_spin_unlock(&glock0);                       \
+       if (!n)                                         \
+               return 1;                               \
+                                                       \
+       if (dolock)                                     \
+               bpf_spin_lock(&glock0);                 \
+       rb_n = bpf_rbtree_##op(&groot0, &n->r0);        \
+       if (rb_n)                                       \
+               jiffies = bpf_jiffies64();              \
+       if (dolock)                                     \
+               bpf_spin_unlock(&glock0);               \
+                                                       \
+       return !!jiffies;                               \
+}
+
+/*
+ * Use a spearate MSG macro instead of passing to TEST_XXX(..., MSG)
+ * to ensure the message itself is not in the bpf prog lineinfo
+ * which the verifier includes in its log.
+ * Otherwise, the test_loader will incorrectly match the prog lineinfo
+ * instead of the log generated by the verifier.
+ */
+#define MSG "call bpf_rbtree_root{{.+}}; R0{{(_w)?}}=rcu_ptr_or_null_node_data(id={{[0-9]+}},non_own_ref"
+TEST_ROOT(true)
+#undef MSG
+#define MSG "call bpf_rbtree_{{(left|right).+}}; R0{{(_w)?}}=rcu_ptr_or_null_node_data(id={{[0-9]+}},non_own_ref"
+TEST_LR(left,  true)
+TEST_LR(right, true)
+#undef MSG
+
+#define MSG "bpf_spin_lock at off=0 must be held for bpf_rb_root"
+TEST_ROOT(false)
+TEST_LR(left, false)
+TEST_LR(right, false)
+#undef MSG
+
+char _license[] SEC("license") = "GPL";