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 <linux/btf_ids.h> 11 #include "percpu_freelist.h" 12 #include <asm/rqspinlock.h> 13 14 #define QUEUE_STACK_CREATE_FLAG_MASK \ 15 (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK) 16 17 struct bpf_queue_stack { 18 struct bpf_map map; 19 rqspinlock_t lock; 20 u32 head, tail; 21 u32 size; /* max_entries + 1 */ 22 23 char elements[] __aligned(8); 24 }; 25 26 static struct bpf_queue_stack *bpf_queue_stack(struct bpf_map *map) 27 { 28 return container_of(map, struct bpf_queue_stack, map); 29 } 30 31 static bool queue_stack_map_is_empty(struct bpf_queue_stack *qs) 32 { 33 return qs->head == qs->tail; 34 } 35 36 static bool queue_stack_map_is_full(struct bpf_queue_stack *qs) 37 { 38 u32 head = qs->head + 1; 39 40 if (unlikely(head >= qs->size)) 41 head = 0; 42 43 return head == qs->tail; 44 } 45 46 /* Called from syscall */ 47 static int queue_stack_map_alloc_check(union bpf_attr *attr) 48 { 49 /* check sanity of attributes */ 50 if (attr->max_entries == 0 || attr->key_size != 0 || 51 attr->value_size == 0 || 52 attr->map_flags & ~QUEUE_STACK_CREATE_FLAG_MASK || 53 !bpf_map_flags_access_ok(attr->map_flags)) 54 return -EINVAL; 55 56 if (attr->value_size > KMALLOC_MAX_SIZE) 57 /* if value_size is bigger, the user space won't be able to 58 * access the elements. 59 */ 60 return -E2BIG; 61 62 return 0; 63 } 64 65 static struct bpf_map *queue_stack_map_alloc(union bpf_attr *attr) 66 { 67 int numa_node = bpf_map_attr_numa_node(attr); 68 struct bpf_queue_stack *qs; 69 u64 size, queue_size; 70 71 size = (u64) attr->max_entries + 1; 72 queue_size = sizeof(*qs) + size * attr->value_size; 73 74 qs = bpf_map_area_alloc(queue_size, numa_node); 75 if (!qs) 76 return ERR_PTR(-ENOMEM); 77 78 bpf_map_init_from_attr(&qs->map, attr); 79 80 qs->size = size; 81 82 raw_res_spin_lock_init(&qs->lock); 83 84 return &qs->map; 85 } 86 87 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */ 88 static void queue_stack_map_free(struct bpf_map *map) 89 { 90 struct bpf_queue_stack *qs = bpf_queue_stack(map); 91 92 bpf_map_area_free(qs); 93 } 94 95 static long __queue_map_get(struct bpf_map *map, void *value, bool delete) 96 { 97 struct bpf_queue_stack *qs = bpf_queue_stack(map); 98 unsigned long flags; 99 int err = 0; 100 void *ptr; 101 102 if (raw_res_spin_lock_irqsave(&qs->lock, flags)) { 103 memset(value, 0, qs->map.value_size); 104 return -EBUSY; 105 } 106 107 if (queue_stack_map_is_empty(qs)) { 108 memset(value, 0, qs->map.value_size); 109 err = -ENOENT; 110 goto out; 111 } 112 113 ptr = &qs->elements[qs->tail * qs->map.value_size]; 114 memcpy(value, ptr, qs->map.value_size); 115 116 if (delete) { 117 if (unlikely(++qs->tail >= qs->size)) 118 qs->tail = 0; 119 } 120 121 out: 122 raw_res_spin_unlock_irqrestore(&qs->lock, flags); 123 return err; 124 } 125 126 127 static long __stack_map_get(struct bpf_map *map, void *value, bool delete) 128 { 129 struct bpf_queue_stack *qs = bpf_queue_stack(map); 130 unsigned long flags; 131 int err = 0; 132 void *ptr; 133 u32 index; 134 135 if (raw_res_spin_lock_irqsave(&qs->lock, flags)) { 136 memset(value, 0, qs->map.value_size); 137 return -EBUSY; 138 } 139 140 if (queue_stack_map_is_empty(qs)) { 141 memset(value, 0, qs->map.value_size); 142 err = -ENOENT; 143 goto out; 144 } 145 146 index = qs->head - 1; 147 if (unlikely(index >= qs->size)) 148 index = qs->size - 1; 149 150 ptr = &qs->elements[index * qs->map.value_size]; 151 memcpy(value, ptr, qs->map.value_size); 152 153 if (delete) 154 qs->head = index; 155 156 out: 157 raw_res_spin_unlock_irqrestore(&qs->lock, flags); 158 return err; 159 } 160 161 /* Called from syscall or from eBPF program */ 162 static long queue_map_peek_elem(struct bpf_map *map, void *value) 163 { 164 return __queue_map_get(map, value, false); 165 } 166 167 /* Called from syscall or from eBPF program */ 168 static long stack_map_peek_elem(struct bpf_map *map, void *value) 169 { 170 return __stack_map_get(map, value, false); 171 } 172 173 /* Called from syscall or from eBPF program */ 174 static long queue_map_pop_elem(struct bpf_map *map, void *value) 175 { 176 return __queue_map_get(map, value, true); 177 } 178 179 /* Called from syscall or from eBPF program */ 180 static long stack_map_pop_elem(struct bpf_map *map, void *value) 181 { 182 return __stack_map_get(map, value, true); 183 } 184 185 /* Called from syscall or from eBPF program */ 186 static long queue_stack_map_push_elem(struct bpf_map *map, void *value, 187 u64 flags) 188 { 189 struct bpf_queue_stack *qs = bpf_queue_stack(map); 190 unsigned long irq_flags; 191 int err = 0; 192 void *dst; 193 194 /* BPF_EXIST is used to force making room for a new element in case the 195 * map is full 196 */ 197 bool replace = (flags & BPF_EXIST); 198 199 /* Check supported flags for queue and stack maps */ 200 if (flags & BPF_NOEXIST || flags > BPF_EXIST) 201 return -EINVAL; 202 203 if (raw_res_spin_lock_irqsave(&qs->lock, irq_flags)) 204 return -EBUSY; 205 206 if (queue_stack_map_is_full(qs)) { 207 if (!replace) { 208 err = -E2BIG; 209 goto out; 210 } 211 /* advance tail pointer to overwrite oldest element */ 212 if (unlikely(++qs->tail >= qs->size)) 213 qs->tail = 0; 214 } 215 216 dst = &qs->elements[qs->head * qs->map.value_size]; 217 memcpy(dst, value, qs->map.value_size); 218 219 if (unlikely(++qs->head >= qs->size)) 220 qs->head = 0; 221 222 out: 223 raw_res_spin_unlock_irqrestore(&qs->lock, irq_flags); 224 return err; 225 } 226 227 /* Called from syscall or from eBPF program */ 228 static void *queue_stack_map_lookup_elem(struct bpf_map *map, void *key) 229 { 230 return NULL; 231 } 232 233 /* Called from syscall or from eBPF program */ 234 static long queue_stack_map_update_elem(struct bpf_map *map, void *key, 235 void *value, u64 flags) 236 { 237 return -EINVAL; 238 } 239 240 /* Called from syscall or from eBPF program */ 241 static long queue_stack_map_delete_elem(struct bpf_map *map, void *key) 242 { 243 return -EINVAL; 244 } 245 246 /* Called from syscall */ 247 static int queue_stack_map_get_next_key(struct bpf_map *map, void *key, 248 void *next_key) 249 { 250 return -EINVAL; 251 } 252 253 static u64 queue_stack_map_mem_usage(const struct bpf_map *map) 254 { 255 u64 usage = sizeof(struct bpf_queue_stack); 256 257 usage += ((u64)map->max_entries + 1) * map->value_size; 258 return usage; 259 } 260 261 BTF_ID_LIST_SINGLE(queue_map_btf_ids, struct, bpf_queue_stack) 262 const struct bpf_map_ops queue_map_ops = { 263 .map_meta_equal = bpf_map_meta_equal, 264 .map_alloc_check = queue_stack_map_alloc_check, 265 .map_alloc = queue_stack_map_alloc, 266 .map_free = queue_stack_map_free, 267 .map_lookup_elem = queue_stack_map_lookup_elem, 268 .map_update_elem = queue_stack_map_update_elem, 269 .map_delete_elem = queue_stack_map_delete_elem, 270 .map_push_elem = queue_stack_map_push_elem, 271 .map_pop_elem = queue_map_pop_elem, 272 .map_peek_elem = queue_map_peek_elem, 273 .map_get_next_key = queue_stack_map_get_next_key, 274 .map_mem_usage = queue_stack_map_mem_usage, 275 .map_btf_id = &queue_map_btf_ids[0], 276 }; 277 278 const struct bpf_map_ops stack_map_ops = { 279 .map_meta_equal = bpf_map_meta_equal, 280 .map_alloc_check = queue_stack_map_alloc_check, 281 .map_alloc = queue_stack_map_alloc, 282 .map_free = queue_stack_map_free, 283 .map_lookup_elem = queue_stack_map_lookup_elem, 284 .map_update_elem = queue_stack_map_update_elem, 285 .map_delete_elem = queue_stack_map_delete_elem, 286 .map_push_elem = queue_stack_map_push_elem, 287 .map_pop_elem = stack_map_pop_elem, 288 .map_peek_elem = stack_map_peek_elem, 289 .map_get_next_key = queue_stack_map_get_next_key, 290 .map_mem_usage = queue_stack_map_mem_usage, 291 .map_btf_id = &queue_map_btf_ids[0], 292 }; 293