1 // SPDX-License-Identifier: GPL-2.0
3 * queue_stack_maps.c: BPF queue and stack maps
5 * Copyright (c) 2018 Politecnico di Torino
8 #include <linux/list.h>
9 #include <linux/slab.h>
10 #include <linux/capability.h>
11 #include "percpu_freelist.h"
13 #define QUEUE_STACK_CREATE_FLAG_MASK \
14 (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK)
16 struct bpf_queue_stack {
20 u32 size; /* max_entries + 1 */
22 char elements[0] __aligned(8);
25 static struct bpf_queue_stack *bpf_queue_stack(struct bpf_map *map)
27 return container_of(map, struct bpf_queue_stack, map);
30 static bool queue_stack_map_is_empty(struct bpf_queue_stack *qs)
32 return qs->head == qs->tail;
35 static bool queue_stack_map_is_full(struct bpf_queue_stack *qs)
37 u32 head = qs->head + 1;
39 if (unlikely(head >= qs->size))
42 return head == qs->tail;
45 /* Called from syscall */
46 static int queue_stack_map_alloc_check(union bpf_attr *attr)
48 if (!capable(CAP_SYS_ADMIN))
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))
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.
67 static struct bpf_map *queue_stack_map_alloc(union bpf_attr *attr)
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;
74 size = (u64) attr->max_entries + 1;
75 cost = queue_size = sizeof(*qs) + size * attr->value_size;
77 ret = bpf_map_charge_init(&mem, cost);
81 qs = bpf_map_area_alloc(queue_size, numa_node);
83 bpf_map_charge_finish(&mem);
84 return ERR_PTR(-ENOMEM);
87 memset(qs, 0, sizeof(*qs));
89 bpf_map_init_from_attr(&qs->map, attr);
91 bpf_map_charge_move(&qs->map.memory, &mem);
94 raw_spin_lock_init(&qs->lock);
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)
102 struct bpf_queue_stack *qs = bpf_queue_stack(map);
104 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
105 * so the programs (can be more than one that used this map) were
106 * disconnected from events. Wait for outstanding critical sections in
107 * these programs to complete
111 bpf_map_area_free(qs);
114 static int __queue_map_get(struct bpf_map *map, void *value, bool delete)
116 struct bpf_queue_stack *qs = bpf_queue_stack(map);
121 raw_spin_lock_irqsave(&qs->lock, flags);
123 if (queue_stack_map_is_empty(qs)) {
124 memset(value, 0, qs->map.value_size);
129 ptr = &qs->elements[qs->tail * qs->map.value_size];
130 memcpy(value, ptr, qs->map.value_size);
133 if (unlikely(++qs->tail >= qs->size))
138 raw_spin_unlock_irqrestore(&qs->lock, flags);
143 static int __stack_map_get(struct bpf_map *map, void *value, bool delete)
145 struct bpf_queue_stack *qs = bpf_queue_stack(map);
151 raw_spin_lock_irqsave(&qs->lock, flags);
153 if (queue_stack_map_is_empty(qs)) {
154 memset(value, 0, qs->map.value_size);
159 index = qs->head - 1;
160 if (unlikely(index >= qs->size))
161 index = qs->size - 1;
163 ptr = &qs->elements[index * qs->map.value_size];
164 memcpy(value, ptr, qs->map.value_size);
170 raw_spin_unlock_irqrestore(&qs->lock, flags);
174 /* Called from syscall or from eBPF program */
175 static int queue_map_peek_elem(struct bpf_map *map, void *value)
177 return __queue_map_get(map, value, false);
180 /* Called from syscall or from eBPF program */
181 static int stack_map_peek_elem(struct bpf_map *map, void *value)
183 return __stack_map_get(map, value, false);
186 /* Called from syscall or from eBPF program */
187 static int queue_map_pop_elem(struct bpf_map *map, void *value)
189 return __queue_map_get(map, value, true);
192 /* Called from syscall or from eBPF program */
193 static int stack_map_pop_elem(struct bpf_map *map, void *value)
195 return __stack_map_get(map, value, true);
198 /* Called from syscall or from eBPF program */
199 static int queue_stack_map_push_elem(struct bpf_map *map, void *value,
202 struct bpf_queue_stack *qs = bpf_queue_stack(map);
203 unsigned long irq_flags;
207 /* BPF_EXIST is used to force making room for a new element in case the
210 bool replace = (flags & BPF_EXIST);
212 /* Check supported flags for queue and stack maps */
213 if (flags & BPF_NOEXIST || flags > BPF_EXIST)
216 raw_spin_lock_irqsave(&qs->lock, irq_flags);
218 if (queue_stack_map_is_full(qs)) {
223 /* advance tail pointer to overwrite oldest element */
224 if (unlikely(++qs->tail >= qs->size))
228 dst = &qs->elements[qs->head * qs->map.value_size];
229 memcpy(dst, value, qs->map.value_size);
231 if (unlikely(++qs->head >= qs->size))
235 raw_spin_unlock_irqrestore(&qs->lock, irq_flags);
239 /* Called from syscall or from eBPF program */
240 static void *queue_stack_map_lookup_elem(struct bpf_map *map, void *key)
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)
252 /* Called from syscall or from eBPF program */
253 static int queue_stack_map_delete_elem(struct bpf_map *map, void *key)
258 /* Called from syscall */
259 static int queue_stack_map_get_next_key(struct bpf_map *map, void *key,
265 const struct bpf_map_ops queue_map_ops = {
266 .map_alloc_check = queue_stack_map_alloc_check,
267 .map_alloc = queue_stack_map_alloc,
268 .map_free = queue_stack_map_free,
269 .map_lookup_elem = queue_stack_map_lookup_elem,
270 .map_update_elem = queue_stack_map_update_elem,
271 .map_delete_elem = queue_stack_map_delete_elem,
272 .map_push_elem = queue_stack_map_push_elem,
273 .map_pop_elem = queue_map_pop_elem,
274 .map_peek_elem = queue_map_peek_elem,
275 .map_get_next_key = queue_stack_map_get_next_key,
278 const struct bpf_map_ops stack_map_ops = {
279 .map_alloc_check = queue_stack_map_alloc_check,
280 .map_alloc = queue_stack_map_alloc,
281 .map_free = queue_stack_map_free,
282 .map_lookup_elem = queue_stack_map_lookup_elem,
283 .map_update_elem = queue_stack_map_update_elem,
284 .map_delete_elem = queue_stack_map_delete_elem,
285 .map_push_elem = queue_stack_map_push_elem,
286 .map_pop_elem = stack_map_pop_elem,
287 .map_peek_elem = stack_map_peek_elem,
288 .map_get_next_key = queue_stack_map_get_next_key,