[PATCH bpf-next 2/3] bpf: Relax the max_entries check for inner map

[Martin KaFai Lau <kafai@fb.com>]

This patch relaxes the max_entries check for most of the inner map types
during an update to the outer map.  The max_entries of those map types
are only used in runtime.  By doing this, an inner map with different
size can be updated to the outer map in runtime.

The max_entries of arraymap and xskmap are used statically
in verification time to generate the inline code, so they
are excluded in this patch.

Cc: Andrey Ignatov <rdna@fb.com>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
---
 include/linux/bpf.h       | 12 ++++++++++++
 include/linux/bpf_types.h |  6 ++++--
 kernel/bpf/map_in_map.c   |  3 ++-
 3 files changed, 18 insertions(+), 3 deletions(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 1e20b9911d48..1488d2aa41f2 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -99,6 +99,18 @@ struct bpf_map_memory {
 
 /* Cannot be used as an inner map */
 #define BPF_MAP_CAP_NO_INNER_MAP (1 << 0)
+/* When a prog has used map-in-map, the verifier requires
+ * an inner-map as a template to verify the access operations
+ * on the outer and inner map.  For some inner map-types,
+ * the verifier uses the inner_map's max_entries statically
+ * (e.g. to generate inline code).  If this verification
+ * time usage on max_entries applies to an inner map-type,
+ * during runtime, only the inner map with the same
+ * max_entries can be updated to this outer map.
+ *
+ * Please see bpf_map_meta_equal() for details.
+ */
+#define BPF_MAP_CAP_NO_DYNAMIC_INNER_MAP_SIZE (1 << 1)
 
 struct bpf_map {
 	/* The first two cachelines with read-mostly members of which some
diff --git a/include/linux/bpf_types.h b/include/linux/bpf_types.h
index 652f17d646dd..4b350f9ad486 100644
--- a/include/linux/bpf_types.h
+++ b/include/linux/bpf_types.h
@@ -76,7 +76,8 @@ BPF_PROG_TYPE(BPF_PROG_TYPE_LSM, lsm,
 #endif /* CONFIG_BPF_LSM */
 #endif
 
-BPF_MAP_TYPE(BPF_MAP_TYPE_ARRAY, array_map_ops, 0)
+BPF_MAP_TYPE(BPF_MAP_TYPE_ARRAY, array_map_ops,
+	     BPF_MAP_CAP_NO_DYNAMIC_INNER_MAP_SIZE)
 BPF_MAP_TYPE(BPF_MAP_TYPE_PERCPU_ARRAY, percpu_array_map_ops, 0)
 /* prog_array->aux->{type,jited} is a runtime binding.
  * Doing static check alone in the verifier is not enough,
@@ -114,7 +115,8 @@ BPF_MAP_TYPE(BPF_MAP_TYPE_SOCKHASH, sock_hash_ops, 0)
 #endif
 BPF_MAP_TYPE(BPF_MAP_TYPE_CPUMAP, cpu_map_ops, 0)
 #if defined(CONFIG_XDP_SOCKETS)
-BPF_MAP_TYPE(BPF_MAP_TYPE_XSKMAP, xsk_map_ops, 0)
+BPF_MAP_TYPE(BPF_MAP_TYPE_XSKMAP, xsk_map_ops,
+	     BPF_MAP_CAP_NO_DYNAMIC_INNER_MAP_SIZE)
 #endif
 #ifdef CONFIG_INET
 BPF_MAP_TYPE(BPF_MAP_TYPE_REUSEPORT_SOCKARRAY, reuseport_array_ops, 0)
diff --git a/kernel/bpf/map_in_map.c b/kernel/bpf/map_in_map.c
index 6e1286ad7b76..bee1fcfd64f2 100644
--- a/kernel/bpf/map_in_map.c
+++ b/kernel/bpf/map_in_map.c
@@ -77,7 +77,8 @@ bool bpf_map_meta_equal(const struct bpf_map *meta0,
 		meta0->key_size == meta1->key_size &&
 		meta0->value_size == meta1->value_size &&
 		meta0->map_flags == meta1->map_flags &&
-		meta0->max_entries == meta1->max_entries;
+		(meta0->max_entries == meta1->max_entries ||
+		 !(meta0->capability & BPF_MAP_CAP_NO_DYNAMIC_INNER_MAP_SIZE));
 }
 
 void *bpf_map_fd_get_ptr(struct bpf_map *map,
-- 
2.24.1