Mauricio Vasquez B | f1a2e44 | 2018-10-18 15:16:25 +0200 | [diff] [blame^] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * queue_stack_maps.c: BPF queue and stack maps |
| 4 | * |
| 5 | * Copyright (c) 2018 Politecnico di Torino |
| 6 | */ |
| 7 | #include <linux/bpf.h> |
| 8 | #include <linux/list.h> |
| 9 | #include <linux/slab.h> |
| 10 | #include "percpu_freelist.h" |
| 11 | |
| 12 | #define QUEUE_STACK_CREATE_FLAG_MASK \ |
| 13 | (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY) |
| 14 | |
| 15 | |
| 16 | struct bpf_queue_stack { |
| 17 | struct bpf_map map; |
| 18 | raw_spinlock_t lock; |
| 19 | u32 head, tail; |
| 20 | u32 size; /* max_entries + 1 */ |
| 21 | |
| 22 | char elements[0] __aligned(8); |
| 23 | }; |
| 24 | |
| 25 | static struct bpf_queue_stack *bpf_queue_stack(struct bpf_map *map) |
| 26 | { |
| 27 | return container_of(map, struct bpf_queue_stack, map); |
| 28 | } |
| 29 | |
| 30 | static bool queue_stack_map_is_empty(struct bpf_queue_stack *qs) |
| 31 | { |
| 32 | return qs->head == qs->tail; |
| 33 | } |
| 34 | |
| 35 | static bool queue_stack_map_is_full(struct bpf_queue_stack *qs) |
| 36 | { |
| 37 | u32 head = qs->head + 1; |
| 38 | |
| 39 | if (unlikely(head >= qs->size)) |
| 40 | head = 0; |
| 41 | |
| 42 | return head == qs->tail; |
| 43 | } |
| 44 | |
| 45 | /* Called from syscall */ |
| 46 | static int queue_stack_map_alloc_check(union bpf_attr *attr) |
| 47 | { |
| 48 | /* check sanity of attributes */ |
| 49 | if (attr->max_entries == 0 || attr->key_size != 0 || |
| 50 | attr->map_flags & ~QUEUE_STACK_CREATE_FLAG_MASK) |
| 51 | return -EINVAL; |
| 52 | |
| 53 | if (attr->value_size > KMALLOC_MAX_SIZE) |
| 54 | /* if value_size is bigger, the user space won't be able to |
| 55 | * access the elements. |
| 56 | */ |
| 57 | return -E2BIG; |
| 58 | |
| 59 | return 0; |
| 60 | } |
| 61 | |
| 62 | static struct bpf_map *queue_stack_map_alloc(union bpf_attr *attr) |
| 63 | { |
| 64 | int ret, numa_node = bpf_map_attr_numa_node(attr); |
| 65 | struct bpf_queue_stack *qs; |
| 66 | u32 size, value_size; |
| 67 | u64 queue_size, cost; |
| 68 | |
| 69 | size = attr->max_entries + 1; |
| 70 | value_size = attr->value_size; |
| 71 | |
| 72 | queue_size = sizeof(*qs) + (u64) value_size * size; |
| 73 | |
| 74 | cost = queue_size; |
| 75 | if (cost >= U32_MAX - PAGE_SIZE) |
| 76 | return ERR_PTR(-E2BIG); |
| 77 | |
| 78 | cost = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT; |
| 79 | |
| 80 | ret = bpf_map_precharge_memlock(cost); |
| 81 | if (ret < 0) |
| 82 | return ERR_PTR(ret); |
| 83 | |
| 84 | qs = bpf_map_area_alloc(queue_size, numa_node); |
| 85 | if (!qs) |
| 86 | return ERR_PTR(-ENOMEM); |
| 87 | |
| 88 | memset(qs, 0, sizeof(*qs)); |
| 89 | |
| 90 | bpf_map_init_from_attr(&qs->map, attr); |
| 91 | |
| 92 | qs->map.pages = cost; |
| 93 | qs->size = size; |
| 94 | |
| 95 | raw_spin_lock_init(&qs->lock); |
| 96 | |
| 97 | return &qs->map; |
| 98 | } |
| 99 | |
| 100 | /* Called when map->refcnt goes to zero, either from workqueue or from syscall */ |
| 101 | static void queue_stack_map_free(struct bpf_map *map) |
| 102 | { |
| 103 | struct bpf_queue_stack *qs = bpf_queue_stack(map); |
| 104 | |
| 105 | /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0, |
| 106 | * so the programs (can be more than one that used this map) were |
| 107 | * disconnected from events. Wait for outstanding critical sections in |
| 108 | * these programs to complete |
| 109 | */ |
| 110 | synchronize_rcu(); |
| 111 | |
| 112 | bpf_map_area_free(qs); |
| 113 | } |
| 114 | |
| 115 | static int __queue_map_get(struct bpf_map *map, void *value, bool delete) |
| 116 | { |
| 117 | struct bpf_queue_stack *qs = bpf_queue_stack(map); |
| 118 | unsigned long flags; |
| 119 | int err = 0; |
| 120 | void *ptr; |
| 121 | |
| 122 | raw_spin_lock_irqsave(&qs->lock, flags); |
| 123 | |
| 124 | if (queue_stack_map_is_empty(qs)) { |
| 125 | err = -ENOENT; |
| 126 | goto out; |
| 127 | } |
| 128 | |
| 129 | ptr = &qs->elements[qs->tail * qs->map.value_size]; |
| 130 | memcpy(value, ptr, qs->map.value_size); |
| 131 | |
| 132 | if (delete) { |
| 133 | if (unlikely(++qs->tail >= qs->size)) |
| 134 | qs->tail = 0; |
| 135 | } |
| 136 | |
| 137 | out: |
| 138 | raw_spin_unlock_irqrestore(&qs->lock, flags); |
| 139 | return err; |
| 140 | } |
| 141 | |
| 142 | |
| 143 | static int __stack_map_get(struct bpf_map *map, void *value, bool delete) |
| 144 | { |
| 145 | struct bpf_queue_stack *qs = bpf_queue_stack(map); |
| 146 | unsigned long flags; |
| 147 | int err = 0; |
| 148 | void *ptr; |
| 149 | u32 index; |
| 150 | |
| 151 | raw_spin_lock_irqsave(&qs->lock, flags); |
| 152 | |
| 153 | if (queue_stack_map_is_empty(qs)) { |
| 154 | err = -ENOENT; |
| 155 | goto out; |
| 156 | } |
| 157 | |
| 158 | index = qs->head - 1; |
| 159 | if (unlikely(index >= qs->size)) |
| 160 | index = qs->size - 1; |
| 161 | |
| 162 | ptr = &qs->elements[index * qs->map.value_size]; |
| 163 | memcpy(value, ptr, qs->map.value_size); |
| 164 | |
| 165 | if (delete) |
| 166 | qs->head = index; |
| 167 | |
| 168 | out: |
| 169 | raw_spin_unlock_irqrestore(&qs->lock, flags); |
| 170 | return err; |
| 171 | } |
| 172 | |
| 173 | /* Called from syscall or from eBPF program */ |
| 174 | static int queue_map_peek_elem(struct bpf_map *map, void *value) |
| 175 | { |
| 176 | return __queue_map_get(map, value, false); |
| 177 | } |
| 178 | |
| 179 | /* Called from syscall or from eBPF program */ |
| 180 | static int stack_map_peek_elem(struct bpf_map *map, void *value) |
| 181 | { |
| 182 | return __stack_map_get(map, value, false); |
| 183 | } |
| 184 | |
| 185 | /* Called from syscall or from eBPF program */ |
| 186 | static int queue_map_pop_elem(struct bpf_map *map, void *value) |
| 187 | { |
| 188 | return __queue_map_get(map, value, true); |
| 189 | } |
| 190 | |
| 191 | /* Called from syscall or from eBPF program */ |
| 192 | static int stack_map_pop_elem(struct bpf_map *map, void *value) |
| 193 | { |
| 194 | return __stack_map_get(map, value, true); |
| 195 | } |
| 196 | |
| 197 | /* Called from syscall or from eBPF program */ |
| 198 | static int queue_stack_map_push_elem(struct bpf_map *map, void *value, |
| 199 | u64 flags) |
| 200 | { |
| 201 | struct bpf_queue_stack *qs = bpf_queue_stack(map); |
| 202 | unsigned long irq_flags; |
| 203 | int err = 0; |
| 204 | void *dst; |
| 205 | |
| 206 | /* BPF_EXIST is used to force making room for a new element in case the |
| 207 | * map is full |
| 208 | */ |
| 209 | bool replace = (flags & BPF_EXIST); |
| 210 | |
| 211 | /* Check supported flags for queue and stack maps */ |
| 212 | if (flags & BPF_NOEXIST || flags > BPF_EXIST) |
| 213 | return -EINVAL; |
| 214 | |
| 215 | raw_spin_lock_irqsave(&qs->lock, irq_flags); |
| 216 | |
| 217 | if (queue_stack_map_is_full(qs)) { |
| 218 | if (!replace) { |
| 219 | err = -E2BIG; |
| 220 | goto out; |
| 221 | } |
| 222 | /* advance tail pointer to overwrite oldest element */ |
| 223 | if (unlikely(++qs->tail >= qs->size)) |
| 224 | qs->tail = 0; |
| 225 | } |
| 226 | |
| 227 | dst = &qs->elements[qs->head * qs->map.value_size]; |
| 228 | memcpy(dst, value, qs->map.value_size); |
| 229 | |
| 230 | if (unlikely(++qs->head >= qs->size)) |
| 231 | qs->head = 0; |
| 232 | |
| 233 | out: |
| 234 | raw_spin_unlock_irqrestore(&qs->lock, irq_flags); |
| 235 | return err; |
| 236 | } |
| 237 | |
| 238 | /* Called from syscall or from eBPF program */ |
| 239 | static void *queue_stack_map_lookup_elem(struct bpf_map *map, void *key) |
| 240 | { |
| 241 | return NULL; |
| 242 | } |
| 243 | |
| 244 | /* Called from syscall or from eBPF program */ |
| 245 | static int queue_stack_map_update_elem(struct bpf_map *map, void *key, |
| 246 | void *value, u64 flags) |
| 247 | { |
| 248 | return -EINVAL; |
| 249 | } |
| 250 | |
| 251 | /* Called from syscall or from eBPF program */ |
| 252 | static int queue_stack_map_delete_elem(struct bpf_map *map, void *key) |
| 253 | { |
| 254 | return -EINVAL; |
| 255 | } |
| 256 | |
| 257 | /* Called from syscall */ |
| 258 | static int queue_stack_map_get_next_key(struct bpf_map *map, void *key, |
| 259 | void *next_key) |
| 260 | { |
| 261 | return -EINVAL; |
| 262 | } |
| 263 | |
| 264 | const struct bpf_map_ops queue_map_ops = { |
| 265 | .map_alloc_check = queue_stack_map_alloc_check, |
| 266 | .map_alloc = queue_stack_map_alloc, |
| 267 | .map_free = queue_stack_map_free, |
| 268 | .map_lookup_elem = queue_stack_map_lookup_elem, |
| 269 | .map_update_elem = queue_stack_map_update_elem, |
| 270 | .map_delete_elem = queue_stack_map_delete_elem, |
| 271 | .map_push_elem = queue_stack_map_push_elem, |
| 272 | .map_pop_elem = queue_map_pop_elem, |
| 273 | .map_peek_elem = queue_map_peek_elem, |
| 274 | .map_get_next_key = queue_stack_map_get_next_key, |
| 275 | }; |
| 276 | |
| 277 | const struct bpf_map_ops stack_map_ops = { |
| 278 | .map_alloc_check = queue_stack_map_alloc_check, |
| 279 | .map_alloc = queue_stack_map_alloc, |
| 280 | .map_free = queue_stack_map_free, |
| 281 | .map_lookup_elem = queue_stack_map_lookup_elem, |
| 282 | .map_update_elem = queue_stack_map_update_elem, |
| 283 | .map_delete_elem = queue_stack_map_delete_elem, |
| 284 | .map_push_elem = queue_stack_map_push_elem, |
| 285 | .map_pop_elem = stack_map_pop_elem, |
| 286 | .map_peek_elem = stack_map_peek_elem, |
| 287 | .map_get_next_key = queue_stack_map_get_next_key, |
| 288 | }; |