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