kernel/bpf/queue_stack_maps.c

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