xref: /linux/kernel/bpf/queue_stack_maps.c (revision 3c4fc7bf4c9e66fe71abcbf93f62f4ddb89b7f15)
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 <linux/btf_ids.h>
12 #include "percpu_freelist.h"
13 
14 #define QUEUE_STACK_CREATE_FLAG_MASK \
15 	(BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK)
16 
17 struct bpf_queue_stack {
18 	struct bpf_map map;
19 	raw_spinlock_t lock;
20 	u32 head, tail;
21 	u32 size; /* max_entries + 1 */
22 
23 	char elements[] __aligned(8);
24 };
25 
26 static struct bpf_queue_stack *bpf_queue_stack(struct bpf_map *map)
27 {
28 	return container_of(map, struct bpf_queue_stack, map);
29 }
30 
31 static bool queue_stack_map_is_empty(struct bpf_queue_stack *qs)
32 {
33 	return qs->head == qs->tail;
34 }
35 
36 static bool queue_stack_map_is_full(struct bpf_queue_stack *qs)
37 {
38 	u32 head = qs->head + 1;
39 
40 	if (unlikely(head >= qs->size))
41 		head = 0;
42 
43 	return head == qs->tail;
44 }
45 
46 /* Called from syscall */
47 static int queue_stack_map_alloc_check(union bpf_attr *attr)
48 {
49 	if (!bpf_capable())
50 		return -EPERM;
51 
52 	/* check sanity of attributes */
53 	if (attr->max_entries == 0 || attr->key_size != 0 ||
54 	    attr->value_size == 0 ||
55 	    attr->map_flags & ~QUEUE_STACK_CREATE_FLAG_MASK ||
56 	    !bpf_map_flags_access_ok(attr->map_flags))
57 		return -EINVAL;
58 
59 	if (attr->value_size > KMALLOC_MAX_SIZE)
60 		/* if value_size is bigger, the user space won't be able to
61 		 * access the elements.
62 		 */
63 		return -E2BIG;
64 
65 	return 0;
66 }
67 
68 static struct bpf_map *queue_stack_map_alloc(union bpf_attr *attr)
69 {
70 	int numa_node = bpf_map_attr_numa_node(attr);
71 	struct bpf_queue_stack *qs;
72 	u64 size, queue_size;
73 
74 	size = (u64) attr->max_entries + 1;
75 	queue_size = sizeof(*qs) + size * attr->value_size;
76 
77 	qs = bpf_map_area_alloc(queue_size, numa_node);
78 	if (!qs)
79 		return ERR_PTR(-ENOMEM);
80 
81 	memset(qs, 0, sizeof(*qs));
82 
83 	bpf_map_init_from_attr(&qs->map, attr);
84 
85 	qs->size = size;
86 
87 	raw_spin_lock_init(&qs->lock);
88 
89 	return &qs->map;
90 }
91 
92 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
93 static void queue_stack_map_free(struct bpf_map *map)
94 {
95 	struct bpf_queue_stack *qs = bpf_queue_stack(map);
96 
97 	bpf_map_area_free(qs);
98 }
99 
100 static int __queue_map_get(struct bpf_map *map, void *value, bool delete)
101 {
102 	struct bpf_queue_stack *qs = bpf_queue_stack(map);
103 	unsigned long flags;
104 	int err = 0;
105 	void *ptr;
106 
107 	raw_spin_lock_irqsave(&qs->lock, flags);
108 
109 	if (queue_stack_map_is_empty(qs)) {
110 		memset(value, 0, qs->map.value_size);
111 		err = -ENOENT;
112 		goto out;
113 	}
114 
115 	ptr = &qs->elements[qs->tail * qs->map.value_size];
116 	memcpy(value, ptr, qs->map.value_size);
117 
118 	if (delete) {
119 		if (unlikely(++qs->tail >= qs->size))
120 			qs->tail = 0;
121 	}
122 
123 out:
124 	raw_spin_unlock_irqrestore(&qs->lock, flags);
125 	return err;
126 }
127 
128 
129 static int __stack_map_get(struct bpf_map *map, void *value, bool delete)
130 {
131 	struct bpf_queue_stack *qs = bpf_queue_stack(map);
132 	unsigned long flags;
133 	int err = 0;
134 	void *ptr;
135 	u32 index;
136 
137 	raw_spin_lock_irqsave(&qs->lock, flags);
138 
139 	if (queue_stack_map_is_empty(qs)) {
140 		memset(value, 0, qs->map.value_size);
141 		err = -ENOENT;
142 		goto out;
143 	}
144 
145 	index = qs->head - 1;
146 	if (unlikely(index >= qs->size))
147 		index = qs->size - 1;
148 
149 	ptr = &qs->elements[index * qs->map.value_size];
150 	memcpy(value, ptr, qs->map.value_size);
151 
152 	if (delete)
153 		qs->head = index;
154 
155 out:
156 	raw_spin_unlock_irqrestore(&qs->lock, flags);
157 	return err;
158 }
159 
160 /* Called from syscall or from eBPF program */
161 static int queue_map_peek_elem(struct bpf_map *map, void *value)
162 {
163 	return __queue_map_get(map, value, false);
164 }
165 
166 /* Called from syscall or from eBPF program */
167 static int stack_map_peek_elem(struct bpf_map *map, void *value)
168 {
169 	return __stack_map_get(map, value, false);
170 }
171 
172 /* Called from syscall or from eBPF program */
173 static int queue_map_pop_elem(struct bpf_map *map, void *value)
174 {
175 	return __queue_map_get(map, value, true);
176 }
177 
178 /* Called from syscall or from eBPF program */
179 static int stack_map_pop_elem(struct bpf_map *map, void *value)
180 {
181 	return __stack_map_get(map, value, true);
182 }
183 
184 /* Called from syscall or from eBPF program */
185 static int queue_stack_map_push_elem(struct bpf_map *map, void *value,
186 				     u64 flags)
187 {
188 	struct bpf_queue_stack *qs = bpf_queue_stack(map);
189 	unsigned long irq_flags;
190 	int err = 0;
191 	void *dst;
192 
193 	/* BPF_EXIST is used to force making room for a new element in case the
194 	 * map is full
195 	 */
196 	bool replace = (flags & BPF_EXIST);
197 
198 	/* Check supported flags for queue and stack maps */
199 	if (flags & BPF_NOEXIST || flags > BPF_EXIST)
200 		return -EINVAL;
201 
202 	raw_spin_lock_irqsave(&qs->lock, irq_flags);
203 
204 	if (queue_stack_map_is_full(qs)) {
205 		if (!replace) {
206 			err = -E2BIG;
207 			goto out;
208 		}
209 		/* advance tail pointer to overwrite oldest element */
210 		if (unlikely(++qs->tail >= qs->size))
211 			qs->tail = 0;
212 	}
213 
214 	dst = &qs->elements[qs->head * qs->map.value_size];
215 	memcpy(dst, value, qs->map.value_size);
216 
217 	if (unlikely(++qs->head >= qs->size))
218 		qs->head = 0;
219 
220 out:
221 	raw_spin_unlock_irqrestore(&qs->lock, irq_flags);
222 	return err;
223 }
224 
225 /* Called from syscall or from eBPF program */
226 static void *queue_stack_map_lookup_elem(struct bpf_map *map, void *key)
227 {
228 	return NULL;
229 }
230 
231 /* Called from syscall or from eBPF program */
232 static int queue_stack_map_update_elem(struct bpf_map *map, void *key,
233 				       void *value, u64 flags)
234 {
235 	return -EINVAL;
236 }
237 
238 /* Called from syscall or from eBPF program */
239 static int queue_stack_map_delete_elem(struct bpf_map *map, void *key)
240 {
241 	return -EINVAL;
242 }
243 
244 /* Called from syscall */
245 static int queue_stack_map_get_next_key(struct bpf_map *map, void *key,
246 					void *next_key)
247 {
248 	return -EINVAL;
249 }
250 
251 BTF_ID_LIST_SINGLE(queue_map_btf_ids, struct, bpf_queue_stack)
252 const struct bpf_map_ops queue_map_ops = {
253 	.map_meta_equal = bpf_map_meta_equal,
254 	.map_alloc_check = queue_stack_map_alloc_check,
255 	.map_alloc = queue_stack_map_alloc,
256 	.map_free = queue_stack_map_free,
257 	.map_lookup_elem = queue_stack_map_lookup_elem,
258 	.map_update_elem = queue_stack_map_update_elem,
259 	.map_delete_elem = queue_stack_map_delete_elem,
260 	.map_push_elem = queue_stack_map_push_elem,
261 	.map_pop_elem = queue_map_pop_elem,
262 	.map_peek_elem = queue_map_peek_elem,
263 	.map_get_next_key = queue_stack_map_get_next_key,
264 	.map_btf_id = &queue_map_btf_ids[0],
265 };
266 
267 const struct bpf_map_ops stack_map_ops = {
268 	.map_meta_equal = bpf_map_meta_equal,
269 	.map_alloc_check = queue_stack_map_alloc_check,
270 	.map_alloc = queue_stack_map_alloc,
271 	.map_free = queue_stack_map_free,
272 	.map_lookup_elem = queue_stack_map_lookup_elem,
273 	.map_update_elem = queue_stack_map_update_elem,
274 	.map_delete_elem = queue_stack_map_delete_elem,
275 	.map_push_elem = queue_stack_map_push_elem,
276 	.map_pop_elem = stack_map_pop_elem,
277 	.map_peek_elem = stack_map_peek_elem,
278 	.map_get_next_key = queue_stack_map_get_next_key,
279 	.map_btf_id = &queue_map_btf_ids[0],
280 };
281