mm/memcg: Exclude mem_cgroup pointer from kABI signature computation

Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=2036995
MR: https://gitlab.com/redhat/centos-stream/src/kernel/centos-stream-9/-/merge_requests/326
Upstream Status: RHEL only

The mem_cgroup structure is allocated dynamically, self-contained and
is not embedded in other data structures. Third-party kernel modules
are not supposed to access mem_cgroup structure directly.

To support new features, mem_cgroup structure often changes quite a
lot over time. However, a number of other important data structures
like the task_struct contain pointers to mem_cgroup. So any change to
mem_cgroup can affect the kABI signatures of the other data structures
leading to kABI breakage.

Instead of carefully messaging the mem_cgroup structure every time
when change is needed in order not to break kABI signature, it is more
efficient to just decouple changes to mem_cgroup from affecting kABI
signatures of other data structures by using the RH_KABI_EXCLUDE() macro
to exclude mem_cgroup pointers from being included in the computation
of kABI signature.

This macro is applied to mem_cgroup pointers in the task_struct, bpf_map,
fsnotify_group and sock data structures.

Signed-off-by: Waiman Long <longman@redhat.com>

include/linux/bpf.h

@@ -174,7 +174,7 @@ struct bpf_map {
 	u32 btf_value_type_id;
 	struct btf *btf;
 #ifdef CONFIG_MEMCG_KMEM
-	struct mem_cgroup *memcg;
+	RH_KABI_EXCLUDE(struct mem_cgroup *memcg)
 #endif
 	char name[BPF_OBJ_NAME_LEN];
 	u32 btf_vmlinux_value_type_id;

include/linux/fsnotify_backend.h

@@ -216,7 +216,7 @@ struct fsnotify_group {
 						 * notification list is too
 						 * full */
 
-	struct mem_cgroup *memcg;	/* memcg to charge allocations */
+	RH_KABI_EXCLUDE(struct mem_cgroup *memcg)	/* memcg to charge allocations */
 
 	/* groups can define private fields here or use the void *private */
 	union {

include/linux/memcontrol.h

@@ -233,6 +233,30 @@ struct obj_cgroup {
  * page cache and RSS per cgroup. We would eventually like to provide
  * statistics based on the statistics developed by Rik Van Riel for clock-pro,
  * to help the administrator determine what knobs to tune.
+ *
+ * RHEL9 Notes
+ * -----------
+ * The mem_cgroup structure is allocated dynamically and is not embedded in
+ * other data structures. To support new features, mem_cgroup structure can
+ * change quite a lot over time. Therefore, the kABI signature of mem_cgroup
+ * will change over time. The mem_cgroup structure is excluded from the kABI
+ * signature generation to avoid impacting kABI signature of other structures
+ * that include a pointer to mem_cgroup. Third-party kernel modules should not
+ * access mem_cgroup or mem_cgroup_per_node directly or use inlined functions
+ * that access them directly.
+ *
+ * In addition, the following enum types will have new upstream fields added
+ * in a way that may break kABI if they are used or referenced by 3rd party
+ * kernel drivers.
+ *  - memcg_memory_event: affect MEMCG_NR_MEMORY_EVENTS
+ *  - node_stat_item: NR_VM_NODE_STAT_ITEMS does get used in lruvec_stat[] of
+ *    mem_cgroup_per_node (memcontrol.h) and lumped into MEMCG_NR_STAT.
+ *    The change in MEMCG_NR_STAT does increase the size of stat[] of
+ *    memcg_vmstats_percpu and vmstats[] of mem_cgroup structures.
+ *    These arrays are not at the end of the structure and so changing
+ *    the array size will break offsets of existing fields that
+ *    follows the arrays.
+ *  - vm_event_item: NR_VM_EVENT_ITEMS used in vmevents[] of memcg.
  */
 struct mem_cgroup {
 	struct cgroup_subsys_state css;

include/linux/mm_types.h

@@ -19,6 +19,8 @@
 
 #include <asm/mmu.h>
 
+#include <linux/rh_kabi.h>
+
 #ifndef AT_VECTOR_SIZE_ARCH
 #define AT_VECTOR_SIZE_ARCH 0
 #endif

include/linux/sched.h

@@ -36,6 +36,8 @@
 #include <linux/kcsan.h>
 #include <asm/kmap_size.h>
 
+#include <linux/rh_kabi.h>
+
 /* task_struct member predeclarations (sorted alphabetically): */
 struct audit_context;
 struct backing_dev_info;
@@ -1403,7 +1405,7 @@ struct task_struct {
 #endif
 
 #ifdef CONFIG_MEMCG
-	struct mem_cgroup		*memcg_in_oom;
+	RH_KABI_EXCLUDE(struct mem_cgroup *memcg_in_oom)
 	gfp_t				memcg_oom_gfp_mask;
 	int				memcg_oom_order;
 
@@ -1411,7 +1413,7 @@ struct task_struct {
 	unsigned int			memcg_nr_pages_over_high;
 
 	/* Used by memcontrol for targeted memcg charge: */
-	struct mem_cgroup		*active_memcg;
+	RH_KABI_EXCLUDE(struct mem_cgroup *active_memcg)
 #endif
 
 #ifdef CONFIG_BLK_CGROUP

include/net/sock.h

@@ -511,7 +511,7 @@ struct sock {
 	void			*sk_security;
 #endif
 	struct sock_cgroup_data	sk_cgrp_data;
-	struct mem_cgroup	*sk_memcg;
+	RH_KABI_EXCLUDE(struct mem_cgroup *sk_memcg)
 	void			(*sk_state_change)(struct sock *sk);
 	void			(*sk_data_ready)(struct sock *sk);
 	void			(*sk_write_space)(struct sock *sk);