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 numa_node = bpf_map_attr_numa_node(attr);
  70         struct bpf_queue_stack *qs;
  71         u64 size, queue_size;
  72
  73         size = (u64) attr->max_entries + 1;
  74         queue_size = sizeof(*qs) + size * attr->value_size;
  75
  76         qs = bpf_map_area_alloc(queue_size, numa_node);
  77         if (!qs)
  78                 return ERR_PTR(-ENOMEM);
  79
  80         memset(qs, 0, sizeof(*qs));
  81
  82         bpf_map_init_from_attr(&qs->map, attr);
  83
  84         qs->size = size;
  85
  86         raw_spin_lock_init(&qs->lock);
  87
  88         return &qs->map;
  89 }
  90
  91 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
  92 static void queue_stack_map_free(struct bpf_map *map)
  93 {
  94         struct bpf_queue_stack *qs = bpf_queue_stack(map);
  95
  96         bpf_map_area_free(qs);
  97 }
  98
  99 static int __queue_map_get(struct bpf_map *map, void *value, bool delete)
 100 {
 101         struct bpf_queue_stack *qs = bpf_queue_stack(map);
 102         unsigned long flags;
 103         int err = 0;
 104         void *ptr;
 105
 106         if (in_nmi()) {
 107                 if (!raw_spin_trylock_irqsave(&qs->lock, flags))
 108                         return -EBUSY;
 109         } else {
 110                 raw_spin_lock_irqsave(&qs->lock, flags);
 111         }
 112
 113         if (queue_stack_map_is_empty(qs)) {
 114                 memset(value, 0, qs->map.value_size);
 115                 err = -ENOENT;
 116                 goto out;
 117         }
 118
 119         ptr = &qs->elements[qs->tail * qs->map.value_size];
 120         memcpy(value, ptr, qs->map.value_size);
 121
 122         if (delete) {
 123                 if (unlikely(++qs->tail >= qs->size))
 124                         qs->tail = 0;
 125         }
 126
 127 out:
 128         raw_spin_unlock_irqrestore(&qs->lock, flags);
 129         return err;
 130 }
 131
 132
 133 static int __stack_map_get(struct bpf_map *map, void *value, bool delete)
 134 {
 135         struct bpf_queue_stack *qs = bpf_queue_stack(map);
 136         unsigned long flags;
 137         int err = 0;
 138         void *ptr;
 139         u32 index;
 140
 141         if (in_nmi()) {
 142                 if (!raw_spin_trylock_irqsave(&qs->lock, flags))
 143                         return -EBUSY;
 144         } else {
 145                 raw_spin_lock_irqsave(&qs->lock, flags);
 146         }
 147
 148         if (queue_stack_map_is_empty(qs)) {
 149                 memset(value, 0, qs->map.value_size);
 150                 err = -ENOENT;
 151                 goto out;
 152         }
 153
 154         index = qs->head - 1;
 155         if (unlikely(index >= qs->size))
 156                 index = qs->size - 1;
 157
 158         ptr = &qs->elements[index * qs->map.value_size];
 159         memcpy(value, ptr, qs->map.value_size);
 160
 161         if (delete)
 162                 qs->head = index;
 163
 164 out:
 165         raw_spin_unlock_irqrestore(&qs->lock, flags);
 166         return err;
 167 }
 168
 169 /* Called from syscall or from eBPF program */
 170 static int queue_map_peek_elem(struct bpf_map *map, void *value)
 171 {
 172         return __queue_map_get(map, value, false);
 173 }
 174
 175 /* Called from syscall or from eBPF program */
 176 static int stack_map_peek_elem(struct bpf_map *map, void *value)
 177 {
 178         return __stack_map_get(map, value, false);
 179 }
 180
 181 /* Called from syscall or from eBPF program */
 182 static int queue_map_pop_elem(struct bpf_map *map, void *value)
 183 {
 184         return __queue_map_get(map, value, true);
 185 }
 186
 187 /* Called from syscall or from eBPF program */
 188 static int stack_map_pop_elem(struct bpf_map *map, void *value)
 189 {
 190         return __stack_map_get(map, value, true);
 191 }
 192
 193 /* Called from syscall or from eBPF program */
 194 static int queue_stack_map_push_elem(struct bpf_map *map, void *value,
 195                                      u64 flags)
 196 {
 197         struct bpf_queue_stack *qs = bpf_queue_stack(map);
 198         unsigned long irq_flags;
 199         int err = 0;
 200         void *dst;
 201
 202         /* BPF_EXIST is used to force making room for a new element in case the
 203          * map is full
 204          */
 205         bool replace = (flags & BPF_EXIST);
 206
 207         /* Check supported flags for queue and stack maps */
 208         if (flags & BPF_NOEXIST || flags > BPF_EXIST)
 209                 return -EINVAL;
 210
 211         if (in_nmi()) {
 212                 if (!raw_spin_trylock_irqsave(&qs->lock, irq_flags))
 213                         return -EBUSY;
 214         } else {
 215                 raw_spin_lock_irqsave(&qs->lock, irq_flags);
 216         }
 217
 218         if (queue_stack_map_is_full(qs)) {
 219                 if (!replace) {
 220                         err = -E2BIG;
 221                         goto out;
 222                 }
 223                 /* advance tail pointer to overwrite oldest element */
 224                 if (unlikely(++qs->tail >= qs->size))
 225                         qs->tail = 0;
 226         }
 227
 228         dst = &qs->elements[qs->head * qs->map.value_size];
 229         memcpy(dst, value, qs->map.value_size);
 230
 231         if (unlikely(++qs->head >= qs->size))
 232                 qs->head = 0;
 233
 234 out:
 235         raw_spin_unlock_irqrestore(&qs->lock, irq_flags);
 236         return err;
 237 }
 238
 239 /* Called from syscall or from eBPF program */
 240 static void *queue_stack_map_lookup_elem(struct bpf_map *map, void *key)
 241 {
 242         return NULL;
 243 }
 244
 245 /* Called from syscall or from eBPF program */
 246 static int queue_stack_map_update_elem(struct bpf_map *map, void *key,
 247                                        void *value, u64 flags)
 248 {
 249         return -EINVAL;
 250 }
 251
 252 /* Called from syscall or from eBPF program */
 253 static int queue_stack_map_delete_elem(struct bpf_map *map, void *key)
 254 {
 255         return -EINVAL;
 256 }
 257
 258 /* Called from syscall */
 259 static int queue_stack_map_get_next_key(struct bpf_map *map, void *key,
 260                                         void *next_key)
 261 {
 262         return -EINVAL;
 263 }
 264
 265 static int queue_map_btf_id;
 266 const struct bpf_map_ops queue_map_ops = {
 267         .map_meta_equal = bpf_map_meta_equal,
 268         .map_alloc_check = queue_stack_map_alloc_check,
 269         .map_alloc = queue_stack_map_alloc,
 270         .map_free = queue_stack_map_free,
 271         .map_lookup_elem = queue_stack_map_lookup_elem,
 272         .map_update_elem = queue_stack_map_update_elem,
 273         .map_delete_elem = queue_stack_map_delete_elem,
 274         .map_push_elem = queue_stack_map_push_elem,
 275         .map_pop_elem = queue_map_pop_elem,
 276         .map_peek_elem = queue_map_peek_elem,
 277         .map_get_next_key = queue_stack_map_get_next_key,
 278         .map_btf_name = "bpf_queue_stack",
 279         .map_btf_id = &queue_map_btf_id,
 280 };
 281
 282 static int stack_map_btf_id;
 283 const struct bpf_map_ops stack_map_ops = {
 284         .map_meta_equal = bpf_map_meta_equal,
 285         .map_alloc_check = queue_stack_map_alloc_check,
 286         .map_alloc = queue_stack_map_alloc,
 287         .map_free = queue_stack_map_free,
 288         .map_lookup_elem = queue_stack_map_lookup_elem,
 289         .map_update_elem = queue_stack_map_update_elem,
 290         .map_delete_elem = queue_stack_map_delete_elem,
 291         .map_push_elem = queue_stack_map_push_elem,
 292         .map_pop_elem = stack_map_pop_elem,
 293         .map_peek_elem = stack_map_peek_elem,
 294         .map_get_next_key = queue_stack_map_get_next_key,
 295         .map_btf_name = "bpf_queue_stack",
 296         .map_btf_id = &stack_map_btf_id,
 297 };