xref: /linux/Documentation/bpf/map_cgrp_storage.rst (revision 0526b56cbc3c489642bd6a5fe4b718dea7ef0ee8)
1.. SPDX-License-Identifier: GPL-2.0-only
2.. Copyright (C) 2022 Meta Platforms, Inc. and affiliates.
3
4=========================
5BPF_MAP_TYPE_CGRP_STORAGE
6=========================
7
8The ``BPF_MAP_TYPE_CGRP_STORAGE`` map type represents a local fix-sized
9storage for cgroups. It is only available with ``CONFIG_CGROUPS``.
10The programs are made available by the same Kconfig. The
11data for a particular cgroup can be retrieved by looking up the map
12with that cgroup.
13
14This document describes the usage and semantics of the
15``BPF_MAP_TYPE_CGRP_STORAGE`` map type.
16
17Usage
18=====
19
20The map key must be ``sizeof(int)`` representing a cgroup fd.
21To access the storage in a program, use ``bpf_cgrp_storage_get``::
22
23    void *bpf_cgrp_storage_get(struct bpf_map *map, struct cgroup *cgroup, void *value, u64 flags)
24
25``flags`` could be 0 or ``BPF_LOCAL_STORAGE_GET_F_CREATE`` which indicates that
26a new local storage will be created if one does not exist.
27
28The local storage can be removed with ``bpf_cgrp_storage_delete``::
29
30    long bpf_cgrp_storage_delete(struct bpf_map *map, struct cgroup *cgroup)
31
32The map is available to all program types.
33
34Examples
35========
36
37A BPF program example with BPF_MAP_TYPE_CGRP_STORAGE::
38
39    #include <vmlinux.h>
40    #include <bpf/bpf_helpers.h>
41    #include <bpf/bpf_tracing.h>
42
43    struct {
44            __uint(type, BPF_MAP_TYPE_CGRP_STORAGE);
45            __uint(map_flags, BPF_F_NO_PREALLOC);
46            __type(key, int);
47            __type(value, long);
48    } cgrp_storage SEC(".maps");
49
50    SEC("tp_btf/sys_enter")
51    int BPF_PROG(on_enter, struct pt_regs *regs, long id)
52    {
53            struct task_struct *task = bpf_get_current_task_btf();
54            long *ptr;
55
56            ptr = bpf_cgrp_storage_get(&cgrp_storage, task->cgroups->dfl_cgrp, 0,
57                                       BPF_LOCAL_STORAGE_GET_F_CREATE);
58            if (ptr)
59                __sync_fetch_and_add(ptr, 1);
60
61            return 0;
62    }
63
64Userspace accessing map declared above::
65
66    #include <linux/bpf.h>
67    #include <linux/libbpf.h>
68
69    __u32 map_lookup(struct bpf_map *map, int cgrp_fd)
70    {
71            __u32 *value;
72            value = bpf_map_lookup_elem(bpf_map__fd(map), &cgrp_fd);
73            if (value)
74                return *value;
75            return 0;
76    }
77
78Difference Between BPF_MAP_TYPE_CGRP_STORAGE and BPF_MAP_TYPE_CGROUP_STORAGE
79============================================================================
80
81The old cgroup storage map ``BPF_MAP_TYPE_CGROUP_STORAGE`` has been marked as
82deprecated (renamed to ``BPF_MAP_TYPE_CGROUP_STORAGE_DEPRECATED``). The new
83``BPF_MAP_TYPE_CGRP_STORAGE`` map should be used instead. The following
84illusates the main difference between ``BPF_MAP_TYPE_CGRP_STORAGE`` and
85``BPF_MAP_TYPE_CGROUP_STORAGE_DEPRECATED``.
86
87(1). ``BPF_MAP_TYPE_CGRP_STORAGE`` can be used by all program types while
88     ``BPF_MAP_TYPE_CGROUP_STORAGE_DEPRECATED`` is available only to cgroup program types
89     like BPF_CGROUP_INET_INGRESS or BPF_CGROUP_SOCK_OPS, etc.
90
91(2). ``BPF_MAP_TYPE_CGRP_STORAGE`` supports local storage for more than one
92     cgroup while ``BPF_MAP_TYPE_CGROUP_STORAGE_DEPRECATED`` only supports one cgroup
93     which is attached by a BPF program.
94
95(3). ``BPF_MAP_TYPE_CGROUP_STORAGE_DEPRECATED`` allocates local storage at attach time so
96     ``bpf_get_local_storage()`` always returns non-NULL local storage.
97     ``BPF_MAP_TYPE_CGRP_STORAGE`` allocates local storage at runtime so
98     it is possible that ``bpf_cgrp_storage_get()`` may return null local storage.
99     To avoid such null local storage issue, user space can do
100     ``bpf_map_update_elem()`` to pre-allocate local storage before a BPF program
101     is attached.
102
103(4). ``BPF_MAP_TYPE_CGRP_STORAGE`` supports deleting local storage by a BPF program
104     while ``BPF_MAP_TYPE_CGROUP_STORAGE_DEPRECATED`` only deletes storage during
105     prog detach time.
106
107So overall, ``BPF_MAP_TYPE_CGRP_STORAGE`` supports all ``BPF_MAP_TYPE_CGROUP_STORAGE_DEPRECATED``
108functionality and beyond. It is recommended to use ``BPF_MAP_TYPE_CGRP_STORAGE``
109instead of ``BPF_MAP_TYPE_CGROUP_STORAGE_DEPRECATED``.
110