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         if (in_nmi()) {
 102                 if (!raw_spin_trylock_irqsave(&qs->lock, flags))
 103                         return -EBUSY;
 104         } else {
 105                 raw_spin_lock_irqsave(&qs->lock, flags);
 106         }
 107
 108         if (queue_stack_map_is_empty(qs)) {
 109                 memset(value, 0, qs->map.value_size);
 110                 err = -ENOENT;
 111                 goto out;
 112         }
 113
 114         ptr = &qs->elements[qs->tail * qs->map.value_size];
 115         memcpy(value, ptr, qs->map.value_size);
 116
 117         if (delete) {
 118                 if (unlikely(++qs->tail >= qs->size))
 119                         qs->tail = 0;
 120         }
 121
 122 out:
 123         raw_spin_unlock_irqrestore(&qs->lock, flags);
 124         return err;
 125 }
 126
 127
 128 static long __stack_map_get(struct bpf_map *map, void *value, bool delete)
 129 {
 130         struct bpf_queue_stack *qs = bpf_queue_stack(map);
 131         unsigned long flags;
 132         int err = 0;
 133         void *ptr;
 134         u32 index;
 135
 136         if (in_nmi()) {
 137                 if (!raw_spin_trylock_irqsave(&qs->lock, flags))
 138                         return -EBUSY;
 139         } else {
 140                 raw_spin_lock_irqsave(&qs->lock, flags);
 141         }
 142
 143         if (queue_stack_map_is_empty(qs)) {
 144                 memset(value, 0, qs->map.value_size);
 145                 err = -ENOENT;
 146                 goto out;
 147         }
 148
 149         index = qs->head - 1;
 150         if (unlikely(index >= qs->size))
 151                 index = qs->size - 1;
 152
 153         ptr = &qs->elements[index * qs->map.value_size];
 154         memcpy(value, ptr, qs->map.value_size);
 155
 156         if (delete)
 157                 qs->head = index;
 158
 159 out:
 160         raw_spin_unlock_irqrestore(&qs->lock, flags);
 161         return err;
 162 }
 163
 164 /* Called from syscall or from eBPF program */
 165 static long queue_map_peek_elem(struct bpf_map *map, void *value)
 166 {
 167         return __queue_map_get(map, value, false);
 168 }
 169
 170 /* Called from syscall or from eBPF program */
 171 static long stack_map_peek_elem(struct bpf_map *map, void *value)
 172 {
 173         return __stack_map_get(map, value, false);
 174 }
 175
 176 /* Called from syscall or from eBPF program */
 177 static long queue_map_pop_elem(struct bpf_map *map, void *value)
 178 {
 179         return __queue_map_get(map, value, true);
 180 }
 181
 182 /* Called from syscall or from eBPF program */
 183 static long stack_map_pop_elem(struct bpf_map *map, void *value)
 184 {
 185         return __stack_map_get(map, value, true);
 186 }
 187
 188 /* Called from syscall or from eBPF program */
 189 static long queue_stack_map_push_elem(struct bpf_map *map, void *value,
 190                                       u64 flags)
 191 {
 192         struct bpf_queue_stack *qs = bpf_queue_stack(map);
 193         unsigned long irq_flags;
 194         int err = 0;
 195         void *dst;
 196
 197         /* BPF_EXIST is used to force making room for a new element in case the
 198          * map is full
 199          */
 200         bool replace = (flags & BPF_EXIST);
 201
 202         /* Check supported flags for queue and stack maps */
 203         if (flags & BPF_NOEXIST || flags > BPF_EXIST)
 204                 return -EINVAL;
 205
 206         if (in_nmi()) {
 207                 if (!raw_spin_trylock_irqsave(&qs->lock, irq_flags))
 208                         return -EBUSY;
 209         } else {
 210                 raw_spin_lock_irqsave(&qs->lock, irq_flags);
 211         }
 212
 213         if (queue_stack_map_is_full(qs)) {
 214                 if (!replace) {
 215                         err = -E2BIG;
 216                         goto out;
 217                 }
 218                 /* advance tail pointer to overwrite oldest element */
 219                 if (unlikely(++qs->tail >= qs->size))
 220                         qs->tail = 0;
 221         }
 222
 223         dst = &qs->elements[qs->head * qs->map.value_size];
 224         memcpy(dst, value, qs->map.value_size);
 225
 226         if (unlikely(++qs->head >= qs->size))
 227                 qs->head = 0;
 228
 229 out:
 230         raw_spin_unlock_irqrestore(&qs->lock, irq_flags);
 231         return err;
 232 }
 233
 234 /* Called from syscall or from eBPF program */
 235 static void *queue_stack_map_lookup_elem(struct bpf_map *map, void *key)
 236 {
 237         return NULL;
 238 }
 239
 240 /* Called from syscall or from eBPF program */
 241 static long queue_stack_map_update_elem(struct bpf_map *map, void *key,
 242                                         void *value, u64 flags)
 243 {
 244         return -EINVAL;
 245 }
 246
 247 /* Called from syscall or from eBPF program */
 248 static long queue_stack_map_delete_elem(struct bpf_map *map, void *key)
 249 {
 250         return -EINVAL;
 251 }
 252
 253 /* Called from syscall */
 254 static int queue_stack_map_get_next_key(struct bpf_map *map, void *key,
 255                                         void *next_key)
 256 {
 257         return -EINVAL;
 258 }
 259
 260 static u64 queue_stack_map_mem_usage(const struct bpf_map *map)
 261 {
 262         u64 usage = sizeof(struct bpf_queue_stack);
 263
 264         usage += ((u64)map->max_entries + 1) * map->value_size;
 265         return usage;
 266 }
 267
 268 BTF_ID_LIST_SINGLE(queue_map_btf_ids, struct, bpf_queue_stack)
 269 const struct bpf_map_ops queue_map_ops = {
 270         .map_meta_equal = bpf_map_meta_equal,
 271         .map_alloc_check = queue_stack_map_alloc_check,
 272         .map_alloc = queue_stack_map_alloc,
 273         .map_free = queue_stack_map_free,
 274         .map_lookup_elem = queue_stack_map_lookup_elem,
 275         .map_update_elem = queue_stack_map_update_elem,
 276         .map_delete_elem = queue_stack_map_delete_elem,
 277         .map_push_elem = queue_stack_map_push_elem,
 278         .map_pop_elem = queue_map_pop_elem,
 279         .map_peek_elem = queue_map_peek_elem,
 280         .map_get_next_key = queue_stack_map_get_next_key,
 281         .map_mem_usage = queue_stack_map_mem_usage,
 282         .map_btf_id = &queue_map_btf_ids[0],
 283 };
 284
 285 const struct bpf_map_ops stack_map_ops = {
 286         .map_meta_equal = bpf_map_meta_equal,
 287         .map_alloc_check = queue_stack_map_alloc_check,
 288         .map_alloc = queue_stack_map_alloc,
 289         .map_free = queue_stack_map_free,
 290         .map_lookup_elem = queue_stack_map_lookup_elem,
 291         .map_update_elem = queue_stack_map_update_elem,
 292         .map_delete_elem = queue_stack_map_delete_elem,
 293         .map_push_elem = queue_stack_map_push_elem,
 294         .map_pop_elem = stack_map_pop_elem,
 295         .map_peek_elem = stack_map_peek_elem,
 296         .map_get_next_key = queue_stack_map_get_next_key,
 297         .map_mem_usage = queue_stack_map_mem_usage,
 298         .map_btf_id = &queue_map_btf_ids[0],
 299 };