| 1 |
|
| 2 | |
| 3 | * Copyright (c) 2019 Facebook
|
| 4 | * Copyright 2020 Google LLC.
|
| 5 |
|
| 6 |
|
| 7 | #ifndef _BPF_LOCAL_STORAGE_H
|
| 8 | #define _BPF_LOCAL_STORAGE_H
|
| 9 |
|
| 10 | #include <linux/bpf.h>
|
| 11 | #include <linux/filter.h>
|
| 12 | #include <linux/rculist.h>
|
| 13 | #include <linux/list.h>
|
| 14 | #include <linux/hash.h>
|
| 15 | #include <linux/types.h>
|
| 16 | #include <linux/bpf_mem_alloc.h>
|
| 17 | #include <uapi/linux/btf.h>
|
| 18 | #include <asm/rqspinlock.h>
|
| 19 |
|
| 20 | #define BPF_LOCAL_STORAGE_CACHE_SIZE 16
|
| 21 |
|
| 22 | struct bpf_local_storage_map_bucket {
|
| 23 | struct hlist_head list;
|
| 24 | rqspinlock_t lock;
|
| 25 | };
|
| 26 |
|
| 27 | |
| 28 | * Instead, the container object (eg. sk->sk_bpf_storage) is.
|
| 29 | *
|
| 30 | * The map (bpf_local_storage_map) is for two purposes
|
| 31 | * 1. Define the size of the "local storage". It is
|
| 32 | * the map's value_size.
|
| 33 | *
|
| 34 | * 2. Maintain a list to keep track of all elems such
|
| 35 | * that they can be cleaned up during the map destruction.
|
| 36 | *
|
| 37 | * When a bpf local storage is being looked up for a
|
| 38 | * particular object, the "bpf_map" pointer is actually used
|
| 39 | * as the "key" to search in the list of elem in
|
| 40 | * the respective bpf_local_storage owned by the object.
|
| 41 | *
|
| 42 | * e.g. sk->sk_bpf_storage is the mini-map with the "bpf_map" pointer
|
| 43 | * as the searching key.
|
| 44 |
|
| 45 | struct bpf_local_storage_map {
|
| 46 | struct bpf_map map;
|
| 47 | |
| 48 | *
|
| 49 | * Updating/Deleting requires a bucket lock to
|
| 50 | * link/unlink the elem from the map. Having
|
| 51 | * multiple buckets to improve contention.
|
| 52 |
|
| 53 | struct bpf_local_storage_map_bucket *buckets;
|
| 54 | u32 bucket_log;
|
| 55 | u16 elem_size;
|
| 56 | u16 cache_idx;
|
| 57 | };
|
| 58 |
|
| 59 | struct bpf_local_storage_data {
|
| 60 | |
| 61 | * from the object's bpf_local_storage.
|
| 62 | *
|
| 63 | * Put it in the same cacheline as the data to minimize
|
| 64 | * the number of cachelines accessed during the cache hit case.
|
| 65 |
|
| 66 | struct bpf_local_storage_map __rcu *smap;
|
| 67 | u8 data[] __aligned(8);
|
| 68 | };
|
| 69 |
|
| 70 | #define SELEM_MAP_UNLINKED (1 << 0)
|
| 71 | #define SELEM_STORAGE_UNLINKED (1 << 1)
|
| 72 | #define SELEM_UNLINKED (SELEM_MAP_UNLINKED | SELEM_STORAGE_UNLINKED)
|
| 73 | #define SELEM_TOFREE (1 << 2)
|
| 74 |
|
| 75 |
|
| 76 | struct bpf_local_storage_elem {
|
| 77 | struct hlist_node map_node;
|
| 78 | struct hlist_node snode;
|
| 79 | struct bpf_local_storage __rcu *local_storage;
|
| 80 | union {
|
| 81 | struct rcu_head rcu;
|
| 82 | struct hlist_node free_node; |
| 83 | * bpf_selem_free
|
| 84 | * after raw_spin_unlock
|
| 85 |
|
| 86 | };
|
| 87 | atomic_t state;
|
| 88 |
|
| 89 | |
| 90 | * the number of cachelines access during a cache hit.
|
| 91 |
|
| 92 | struct bpf_local_storage_data sdata ____cacheline_aligned;
|
| 93 | };
|
| 94 |
|
| 95 | struct bpf_local_storage {
|
| 96 | struct bpf_local_storage_data __rcu *cache[BPF_LOCAL_STORAGE_CACHE_SIZE];
|
| 97 | struct hlist_head list;
|
| 98 | void *owner; |
| 99 | * bpf_local_storage_elem.
|
| 100 |
|
| 101 | struct rcu_head rcu;
|
| 102 | rqspinlock_t lock;
|
| 103 | u64 mem_charge;
|
| 104 | refcount_t owner_refcnt;
|
| 105 | };
|
| 106 |
|
| 107 | |
| 108 | * considering a tcp_sock is ~2k.
|
| 109 |
|
| 110 | #define BPF_LOCAL_STORAGE_MAX_VALUE_SIZE \
|
| 111 | min_t(u32, \
|
| 112 | (KMALLOC_MAX_SIZE - MAX_BPF_STACK - \
|
| 113 | sizeof(struct bpf_local_storage_elem)), \
|
| 114 | (U16_MAX - sizeof(struct bpf_local_storage_elem)))
|
| 115 |
|
| 116 | #define SELEM(_SDATA) \
|
| 117 | container_of((_SDATA), struct bpf_local_storage_elem, sdata)
|
| 118 | #define SDATA(_SELEM) (&(_SELEM)->sdata)
|
| 119 |
|
| 120 | #define BPF_LOCAL_STORAGE_CACHE_SIZE 16
|
| 121 |
|
| 122 | struct bpf_local_storage_cache {
|
| 123 | spinlock_t idx_lock;
|
| 124 | u64 idx_usage_counts[BPF_LOCAL_STORAGE_CACHE_SIZE];
|
| 125 | };
|
| 126 |
|
| 127 | #define DEFINE_BPF_STORAGE_CACHE(name) \
|
| 128 | static struct bpf_local_storage_cache name = { \
|
| 129 | .idx_lock = __SPIN_LOCK_UNLOCKED(name.idx_lock), \
|
| 130 | }
|
| 131 |
|
| 132 |
|
| 133 | int bpf_local_storage_map_alloc_check(union bpf_attr *attr);
|
| 134 |
|
| 135 | struct bpf_map *
|
| 136 | bpf_local_storage_map_alloc(union bpf_attr *attr,
|
| 137 | struct bpf_local_storage_cache *cache);
|
| 138 |
|
| 139 | void __bpf_local_storage_insert_cache(struct bpf_local_storage *local_storage,
|
| 140 | struct bpf_local_storage_map *smap,
|
| 141 | struct bpf_local_storage_elem *selem);
|
| 142 |
|
| 143 | static inline struct bpf_local_storage_data *
|
| 144 | bpf_local_storage_lookup(struct bpf_local_storage *local_storage,
|
| 145 | struct bpf_local_storage_map *smap,
|
| 146 | bool cacheit_lockit)
|
| 147 | {
|
| 148 | struct bpf_local_storage_data *sdata;
|
| 149 | struct bpf_local_storage_elem *selem;
|
| 150 |
|
| 151 |
|
| 152 | sdata = rcu_dereference_check(local_storage->cache[smap->cache_idx],
|
| 153 | bpf_rcu_lock_held());
|
| 154 | if (sdata && rcu_access_pointer(sdata->smap) == smap)
|
| 155 | return sdata;
|
| 156 |
|
| 157 |
|
| 158 | hlist_for_each_entry_rcu(selem, &local_storage->list, snode,
|
| 159 | rcu_read_lock_trace_held())
|
| 160 | if (rcu_access_pointer(SDATA(selem)->smap) == smap)
|
| 161 | break;
|
| 162 |
|
| 163 | if (!selem)
|
| 164 | return NULL;
|
| 165 | if (cacheit_lockit)
|
| 166 | __bpf_local_storage_insert_cache(local_storage, smap, selem);
|
| 167 | return SDATA(selem);
|
| 168 | }
|
| 169 |
|
| 170 | u32 bpf_local_storage_destroy(struct bpf_local_storage *local_storage);
|
| 171 |
|
| 172 | void bpf_local_storage_map_free(struct bpf_map *map,
|
| 173 | struct bpf_local_storage_cache *cache);
|
| 174 |
|
| 175 | int bpf_local_storage_map_check_btf(struct bpf_map *map,
|
| 176 | const struct btf *btf,
|
| 177 | const struct btf_type *key_type,
|
| 178 | const struct btf_type *value_type);
|
| 179 |
|
| 180 | void bpf_selem_link_storage_nolock(struct bpf_local_storage *local_storage,
|
| 181 | struct bpf_local_storage_elem *selem);
|
| 182 |
|
| 183 | int bpf_selem_unlink(struct bpf_local_storage_elem *selem);
|
| 184 |
|
| 185 | int bpf_selem_link_map(struct bpf_local_storage_map *smap,
|
| 186 | struct bpf_local_storage *local_storage,
|
| 187 | struct bpf_local_storage_elem *selem);
|
| 188 |
|
| 189 | struct bpf_local_storage_elem *
|
| 190 | bpf_selem_alloc(struct bpf_local_storage_map *smap, void *owner, void *value,
|
| 191 | bool swap_uptrs);
|
| 192 |
|
| 193 | void bpf_selem_free(struct bpf_local_storage_elem *selem,
|
| 194 | bool reuse_now);
|
| 195 |
|
| 196 | int
|
| 197 | bpf_local_storage_alloc(void *owner,
|
| 198 | struct bpf_local_storage_map *smap,
|
| 199 | struct bpf_local_storage_elem *first_selem);
|
| 200 |
|
| 201 | struct bpf_local_storage_data *
|
| 202 | bpf_local_storage_update(void *owner, struct bpf_local_storage_map *smap,
|
| 203 | void *value, u64 map_flags, bool swap_uptrs);
|
| 204 |
|
| 205 | u64 bpf_local_storage_map_mem_usage(const struct bpf_map *map);
|
| 206 |
|
| 207 | #endif
|