xref: /openbmc/linux/kernel/bpf/arraymap.c (revision b96fc2f3)
1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2  *
3  * This program is free software; you can redistribute it and/or
4  * modify it under the terms of version 2 of the GNU General Public
5  * License as published by the Free Software Foundation.
6  *
7  * This program is distributed in the hope that it will be useful, but
8  * WITHOUT ANY WARRANTY; without even the implied warranty of
9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
10  * General Public License for more details.
11  */
12 #include <linux/bpf.h>
13 #include <linux/err.h>
14 #include <linux/vmalloc.h>
15 #include <linux/slab.h>
16 #include <linux/mm.h>
17 #include <linux/filter.h>
18 
19 /* Called from syscall */
20 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
21 {
22 	struct bpf_array *array;
23 	u32 elem_size, array_size;
24 
25 	/* check sanity of attributes */
26 	if (attr->max_entries == 0 || attr->key_size != 4 ||
27 	    attr->value_size == 0)
28 		return ERR_PTR(-EINVAL);
29 
30 	elem_size = round_up(attr->value_size, 8);
31 
32 	/* check round_up into zero and u32 overflow */
33 	if (elem_size == 0 ||
34 	    attr->max_entries > (U32_MAX - sizeof(*array)) / elem_size)
35 		return ERR_PTR(-ENOMEM);
36 
37 	array_size = sizeof(*array) + attr->max_entries * elem_size;
38 
39 	/* allocate all map elements and zero-initialize them */
40 	array = kzalloc(array_size, GFP_USER | __GFP_NOWARN);
41 	if (!array) {
42 		array = vzalloc(array_size);
43 		if (!array)
44 			return ERR_PTR(-ENOMEM);
45 	}
46 
47 	/* copy mandatory map attributes */
48 	array->map.key_size = attr->key_size;
49 	array->map.value_size = attr->value_size;
50 	array->map.max_entries = attr->max_entries;
51 
52 	array->elem_size = elem_size;
53 
54 	return &array->map;
55 }
56 
57 /* Called from syscall or from eBPF program */
58 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
59 {
60 	struct bpf_array *array = container_of(map, struct bpf_array, map);
61 	u32 index = *(u32 *)key;
62 
63 	if (index >= array->map.max_entries)
64 		return NULL;
65 
66 	return array->value + array->elem_size * index;
67 }
68 
69 /* Called from syscall */
70 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
71 {
72 	struct bpf_array *array = container_of(map, struct bpf_array, map);
73 	u32 index = *(u32 *)key;
74 	u32 *next = (u32 *)next_key;
75 
76 	if (index >= array->map.max_entries) {
77 		*next = 0;
78 		return 0;
79 	}
80 
81 	if (index == array->map.max_entries - 1)
82 		return -ENOENT;
83 
84 	*next = index + 1;
85 	return 0;
86 }
87 
88 /* Called from syscall or from eBPF program */
89 static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
90 				 u64 map_flags)
91 {
92 	struct bpf_array *array = container_of(map, struct bpf_array, map);
93 	u32 index = *(u32 *)key;
94 
95 	if (map_flags > BPF_EXIST)
96 		/* unknown flags */
97 		return -EINVAL;
98 
99 	if (index >= array->map.max_entries)
100 		/* all elements were pre-allocated, cannot insert a new one */
101 		return -E2BIG;
102 
103 	if (map_flags == BPF_NOEXIST)
104 		/* all elements already exist */
105 		return -EEXIST;
106 
107 	memcpy(array->value + array->elem_size * index, value, array->elem_size);
108 	return 0;
109 }
110 
111 /* Called from syscall or from eBPF program */
112 static int array_map_delete_elem(struct bpf_map *map, void *key)
113 {
114 	return -EINVAL;
115 }
116 
117 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
118 static void array_map_free(struct bpf_map *map)
119 {
120 	struct bpf_array *array = container_of(map, struct bpf_array, map);
121 
122 	/* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
123 	 * so the programs (can be more than one that used this map) were
124 	 * disconnected from events. Wait for outstanding programs to complete
125 	 * and free the array
126 	 */
127 	synchronize_rcu();
128 
129 	kvfree(array);
130 }
131 
132 static const struct bpf_map_ops array_ops = {
133 	.map_alloc = array_map_alloc,
134 	.map_free = array_map_free,
135 	.map_get_next_key = array_map_get_next_key,
136 	.map_lookup_elem = array_map_lookup_elem,
137 	.map_update_elem = array_map_update_elem,
138 	.map_delete_elem = array_map_delete_elem,
139 };
140 
141 static struct bpf_map_type_list array_type __read_mostly = {
142 	.ops = &array_ops,
143 	.type = BPF_MAP_TYPE_ARRAY,
144 };
145 
146 static int __init register_array_map(void)
147 {
148 	bpf_register_map_type(&array_type);
149 	return 0;
150 }
151 late_initcall(register_array_map);
152 
153 static struct bpf_map *fd_array_map_alloc(union bpf_attr *attr)
154 {
155 	/* only file descriptors can be stored in this type of map */
156 	if (attr->value_size != sizeof(u32))
157 		return ERR_PTR(-EINVAL);
158 	return array_map_alloc(attr);
159 }
160 
161 static void fd_array_map_free(struct bpf_map *map)
162 {
163 	struct bpf_array *array = container_of(map, struct bpf_array, map);
164 	int i;
165 
166 	synchronize_rcu();
167 
168 	/* make sure it's empty */
169 	for (i = 0; i < array->map.max_entries; i++)
170 		BUG_ON(array->ptrs[i] != NULL);
171 	kvfree(array);
172 }
173 
174 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
175 {
176 	return NULL;
177 }
178 
179 /* only called from syscall */
180 static int fd_array_map_update_elem(struct bpf_map *map, void *key,
181 				    void *value, u64 map_flags)
182 {
183 	struct bpf_array *array = container_of(map, struct bpf_array, map);
184 	void *new_ptr, *old_ptr;
185 	u32 index = *(u32 *)key, ufd;
186 
187 	if (map_flags != BPF_ANY)
188 		return -EINVAL;
189 
190 	if (index >= array->map.max_entries)
191 		return -E2BIG;
192 
193 	ufd = *(u32 *)value;
194 	new_ptr = map->ops->map_fd_get_ptr(map, ufd);
195 	if (IS_ERR(new_ptr))
196 		return PTR_ERR(new_ptr);
197 
198 	old_ptr = xchg(array->ptrs + index, new_ptr);
199 	if (old_ptr)
200 		map->ops->map_fd_put_ptr(old_ptr);
201 
202 	return 0;
203 }
204 
205 static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
206 {
207 	struct bpf_array *array = container_of(map, struct bpf_array, map);
208 	void *old_ptr;
209 	u32 index = *(u32 *)key;
210 
211 	if (index >= array->map.max_entries)
212 		return -E2BIG;
213 
214 	old_ptr = xchg(array->ptrs + index, NULL);
215 	if (old_ptr) {
216 		map->ops->map_fd_put_ptr(old_ptr);
217 		return 0;
218 	} else {
219 		return -ENOENT;
220 	}
221 }
222 
223 static void *prog_fd_array_get_ptr(struct bpf_map *map, int fd)
224 {
225 	struct bpf_array *array = container_of(map, struct bpf_array, map);
226 	struct bpf_prog *prog = bpf_prog_get(fd);
227 	if (IS_ERR(prog))
228 		return prog;
229 
230 	if (!bpf_prog_array_compatible(array, prog)) {
231 		bpf_prog_put(prog);
232 		return ERR_PTR(-EINVAL);
233 	}
234 	return prog;
235 }
236 
237 static void prog_fd_array_put_ptr(void *ptr)
238 {
239 	struct bpf_prog *prog = ptr;
240 
241 	bpf_prog_put_rcu(prog);
242 }
243 
244 /* decrement refcnt of all bpf_progs that are stored in this map */
245 void bpf_fd_array_map_clear(struct bpf_map *map)
246 {
247 	struct bpf_array *array = container_of(map, struct bpf_array, map);
248 	int i;
249 
250 	for (i = 0; i < array->map.max_entries; i++)
251 		fd_array_map_delete_elem(map, &i);
252 }
253 
254 static const struct bpf_map_ops prog_array_ops = {
255 	.map_alloc = fd_array_map_alloc,
256 	.map_free = fd_array_map_free,
257 	.map_get_next_key = array_map_get_next_key,
258 	.map_lookup_elem = fd_array_map_lookup_elem,
259 	.map_update_elem = fd_array_map_update_elem,
260 	.map_delete_elem = fd_array_map_delete_elem,
261 	.map_fd_get_ptr = prog_fd_array_get_ptr,
262 	.map_fd_put_ptr = prog_fd_array_put_ptr,
263 };
264 
265 static struct bpf_map_type_list prog_array_type __read_mostly = {
266 	.ops = &prog_array_ops,
267 	.type = BPF_MAP_TYPE_PROG_ARRAY,
268 };
269 
270 static int __init register_prog_array_map(void)
271 {
272 	bpf_register_map_type(&prog_array_type);
273 	return 0;
274 }
275 late_initcall(register_prog_array_map);
276 
277 static void perf_event_array_map_free(struct bpf_map *map)
278 {
279 	bpf_fd_array_map_clear(map);
280 	fd_array_map_free(map);
281 }
282 
283 static void *perf_event_fd_array_get_ptr(struct bpf_map *map, int fd)
284 {
285 	struct perf_event *event;
286 	const struct perf_event_attr *attr;
287 
288 	event = perf_event_get(fd);
289 	if (IS_ERR(event))
290 		return event;
291 
292 	attr = perf_event_attrs(event);
293 	if (IS_ERR(attr))
294 		return (void *)attr;
295 
296 	if (attr->type != PERF_TYPE_RAW &&
297 	    attr->type != PERF_TYPE_HARDWARE) {
298 		perf_event_release_kernel(event);
299 		return ERR_PTR(-EINVAL);
300 	}
301 	return event;
302 }
303 
304 static void perf_event_fd_array_put_ptr(void *ptr)
305 {
306 	struct perf_event *event = ptr;
307 
308 	perf_event_release_kernel(event);
309 }
310 
311 static const struct bpf_map_ops perf_event_array_ops = {
312 	.map_alloc = fd_array_map_alloc,
313 	.map_free = perf_event_array_map_free,
314 	.map_get_next_key = array_map_get_next_key,
315 	.map_lookup_elem = fd_array_map_lookup_elem,
316 	.map_update_elem = fd_array_map_update_elem,
317 	.map_delete_elem = fd_array_map_delete_elem,
318 	.map_fd_get_ptr = perf_event_fd_array_get_ptr,
319 	.map_fd_put_ptr = perf_event_fd_array_put_ptr,
320 };
321 
322 static struct bpf_map_type_list perf_event_array_type __read_mostly = {
323 	.ops = &perf_event_array_ops,
324 	.type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
325 };
326 
327 static int __init register_perf_event_array_map(void)
328 {
329 	bpf_register_map_type(&perf_event_array_type);
330 	return 0;
331 }
332 late_initcall(register_perf_event_array_map);
333