• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
bpf_queue_stack(struct bpf_map * map)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 
queue_stack_map_is_empty(struct bpf_queue_stack * qs)30 static bool queue_stack_map_is_empty(struct bpf_queue_stack *qs)
31 {
32 	return qs->head == qs->tail;
33 }
34 
queue_stack_map_is_full(struct bpf_queue_stack * qs)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 */
queue_stack_map_alloc_check(union bpf_attr * attr)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 
queue_stack_map_alloc(union bpf_attr * attr)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 */
queue_stack_map_free(struct bpf_map * map)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 
__queue_map_get(struct bpf_map * map,void * value,bool delete)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 	if (in_nmi()) {
115 		if (!raw_spin_trylock_irqsave(&qs->lock, flags))
116 			return -EBUSY;
117 	} else {
118 		raw_spin_lock_irqsave(&qs->lock, flags);
119 	}
120 
121 	if (queue_stack_map_is_empty(qs)) {
122 		memset(value, 0, qs->map.value_size);
123 		err = -ENOENT;
124 		goto out;
125 	}
126 
127 	ptr = &qs->elements[qs->tail * qs->map.value_size];
128 	memcpy(value, ptr, qs->map.value_size);
129 
130 	if (delete) {
131 		if (unlikely(++qs->tail >= qs->size))
132 			qs->tail = 0;
133 	}
134 
135 out:
136 	raw_spin_unlock_irqrestore(&qs->lock, flags);
137 	return err;
138 }
139 
140 
__stack_map_get(struct bpf_map * map,void * value,bool delete)141 static int __stack_map_get(struct bpf_map *map, void *value, bool delete)
142 {
143 	struct bpf_queue_stack *qs = bpf_queue_stack(map);
144 	unsigned long flags;
145 	int err = 0;
146 	void *ptr;
147 	u32 index;
148 
149 	if (in_nmi()) {
150 		if (!raw_spin_trylock_irqsave(&qs->lock, flags))
151 			return -EBUSY;
152 	} else {
153 		raw_spin_lock_irqsave(&qs->lock, flags);
154 	}
155 
156 	if (queue_stack_map_is_empty(qs)) {
157 		memset(value, 0, qs->map.value_size);
158 		err = -ENOENT;
159 		goto out;
160 	}
161 
162 	index = qs->head - 1;
163 	if (unlikely(index >= qs->size))
164 		index = qs->size - 1;
165 
166 	ptr = &qs->elements[index * qs->map.value_size];
167 	memcpy(value, ptr, qs->map.value_size);
168 
169 	if (delete)
170 		qs->head = index;
171 
172 out:
173 	raw_spin_unlock_irqrestore(&qs->lock, flags);
174 	return err;
175 }
176 
177 /* Called from syscall or from eBPF program */
queue_map_peek_elem(struct bpf_map * map,void * value)178 static int queue_map_peek_elem(struct bpf_map *map, void *value)
179 {
180 	return __queue_map_get(map, value, false);
181 }
182 
183 /* Called from syscall or from eBPF program */
stack_map_peek_elem(struct bpf_map * map,void * value)184 static int stack_map_peek_elem(struct bpf_map *map, void *value)
185 {
186 	return __stack_map_get(map, value, false);
187 }
188 
189 /* Called from syscall or from eBPF program */
queue_map_pop_elem(struct bpf_map * map,void * value)190 static int queue_map_pop_elem(struct bpf_map *map, void *value)
191 {
192 	return __queue_map_get(map, value, true);
193 }
194 
195 /* Called from syscall or from eBPF program */
stack_map_pop_elem(struct bpf_map * map,void * value)196 static int stack_map_pop_elem(struct bpf_map *map, void *value)
197 {
198 	return __stack_map_get(map, value, true);
199 }
200 
201 /* Called from syscall or from eBPF program */
queue_stack_map_push_elem(struct bpf_map * map,void * value,u64 flags)202 static int queue_stack_map_push_elem(struct bpf_map *map, void *value,
203 				     u64 flags)
204 {
205 	struct bpf_queue_stack *qs = bpf_queue_stack(map);
206 	unsigned long irq_flags;
207 	int err = 0;
208 	void *dst;
209 
210 	/* BPF_EXIST is used to force making room for a new element in case the
211 	 * map is full
212 	 */
213 	bool replace = (flags & BPF_EXIST);
214 
215 	/* Check supported flags for queue and stack maps */
216 	if (flags & BPF_NOEXIST || flags > BPF_EXIST)
217 		return -EINVAL;
218 
219 	if (in_nmi()) {
220 		if (!raw_spin_trylock_irqsave(&qs->lock, irq_flags))
221 			return -EBUSY;
222 	} else {
223 		raw_spin_lock_irqsave(&qs->lock, irq_flags);
224 	}
225 
226 	if (queue_stack_map_is_full(qs)) {
227 		if (!replace) {
228 			err = -E2BIG;
229 			goto out;
230 		}
231 		/* advance tail pointer to overwrite oldest element */
232 		if (unlikely(++qs->tail >= qs->size))
233 			qs->tail = 0;
234 	}
235 
236 	dst = &qs->elements[qs->head * qs->map.value_size];
237 	memcpy(dst, value, qs->map.value_size);
238 
239 	if (unlikely(++qs->head >= qs->size))
240 		qs->head = 0;
241 
242 out:
243 	raw_spin_unlock_irqrestore(&qs->lock, irq_flags);
244 	return err;
245 }
246 
247 /* Called from syscall or from eBPF program */
queue_stack_map_lookup_elem(struct bpf_map * map,void * key)248 static void *queue_stack_map_lookup_elem(struct bpf_map *map, void *key)
249 {
250 	return NULL;
251 }
252 
253 /* Called from syscall or from eBPF program */
queue_stack_map_update_elem(struct bpf_map * map,void * key,void * value,u64 flags)254 static int queue_stack_map_update_elem(struct bpf_map *map, void *key,
255 				       void *value, u64 flags)
256 {
257 	return -EINVAL;
258 }
259 
260 /* Called from syscall or from eBPF program */
queue_stack_map_delete_elem(struct bpf_map * map,void * key)261 static int queue_stack_map_delete_elem(struct bpf_map *map, void *key)
262 {
263 	return -EINVAL;
264 }
265 
266 /* Called from syscall */
queue_stack_map_get_next_key(struct bpf_map * map,void * key,void * next_key)267 static int queue_stack_map_get_next_key(struct bpf_map *map, void *key,
268 					void *next_key)
269 {
270 	return -EINVAL;
271 }
272 
273 static int queue_map_btf_id;
274 const struct bpf_map_ops queue_map_ops = {
275 	.map_meta_equal = bpf_map_meta_equal,
276 	.map_alloc_check = queue_stack_map_alloc_check,
277 	.map_alloc = queue_stack_map_alloc,
278 	.map_free = queue_stack_map_free,
279 	.map_lookup_elem = queue_stack_map_lookup_elem,
280 	.map_update_elem = queue_stack_map_update_elem,
281 	.map_delete_elem = queue_stack_map_delete_elem,
282 	.map_push_elem = queue_stack_map_push_elem,
283 	.map_pop_elem = queue_map_pop_elem,
284 	.map_peek_elem = queue_map_peek_elem,
285 	.map_get_next_key = queue_stack_map_get_next_key,
286 	.map_btf_name = "bpf_queue_stack",
287 	.map_btf_id = &queue_map_btf_id,
288 };
289 
290 static int stack_map_btf_id;
291 const struct bpf_map_ops stack_map_ops = {
292 	.map_meta_equal = bpf_map_meta_equal,
293 	.map_alloc_check = queue_stack_map_alloc_check,
294 	.map_alloc = queue_stack_map_alloc,
295 	.map_free = queue_stack_map_free,
296 	.map_lookup_elem = queue_stack_map_lookup_elem,
297 	.map_update_elem = queue_stack_map_update_elem,
298 	.map_delete_elem = queue_stack_map_delete_elem,
299 	.map_push_elem = queue_stack_map_push_elem,
300 	.map_pop_elem = stack_map_pop_elem,
301 	.map_peek_elem = stack_map_peek_elem,
302 	.map_get_next_key = queue_stack_map_get_next_key,
303 	.map_btf_name = "bpf_queue_stack",
304 	.map_btf_id = &stack_map_btf_id,
305 };
306