xref: /openbmc/linux/kernel/bpf/arraymap.c (revision 78700c0a)
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 #include <linux/perf_event.h>
19 
20 static void bpf_array_free_percpu(struct bpf_array *array)
21 {
22 	int i;
23 
24 	for (i = 0; i < array->map.max_entries; i++)
25 		free_percpu(array->pptrs[i]);
26 }
27 
28 static int bpf_array_alloc_percpu(struct bpf_array *array)
29 {
30 	void __percpu *ptr;
31 	int i;
32 
33 	for (i = 0; i < array->map.max_entries; i++) {
34 		ptr = __alloc_percpu_gfp(array->elem_size, 8,
35 					 GFP_USER | __GFP_NOWARN);
36 		if (!ptr) {
37 			bpf_array_free_percpu(array);
38 			return -ENOMEM;
39 		}
40 		array->pptrs[i] = ptr;
41 	}
42 
43 	return 0;
44 }
45 
46 /* Called from syscall */
47 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
48 {
49 	bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
50 	struct bpf_array *array;
51 	u64 array_size;
52 	u32 elem_size;
53 
54 	/* check sanity of attributes */
55 	if (attr->max_entries == 0 || attr->key_size != 4 ||
56 	    attr->value_size == 0 || attr->map_flags)
57 		return ERR_PTR(-EINVAL);
58 
59 	if (attr->value_size >= 1 << (KMALLOC_SHIFT_MAX - 1))
60 		/* if value_size is bigger, the user space won't be able to
61 		 * access the elements.
62 		 */
63 		return ERR_PTR(-E2BIG);
64 
65 	elem_size = round_up(attr->value_size, 8);
66 
67 	array_size = sizeof(*array);
68 	if (percpu)
69 		array_size += (u64) attr->max_entries * sizeof(void *);
70 	else
71 		array_size += (u64) attr->max_entries * elem_size;
72 
73 	/* make sure there is no u32 overflow later in round_up() */
74 	if (array_size >= U32_MAX - PAGE_SIZE)
75 		return ERR_PTR(-ENOMEM);
76 
77 
78 	/* allocate all map elements and zero-initialize them */
79 	array = kzalloc(array_size, GFP_USER | __GFP_NOWARN);
80 	if (!array) {
81 		array = vzalloc(array_size);
82 		if (!array)
83 			return ERR_PTR(-ENOMEM);
84 	}
85 
86 	/* copy mandatory map attributes */
87 	array->map.map_type = attr->map_type;
88 	array->map.key_size = attr->key_size;
89 	array->map.value_size = attr->value_size;
90 	array->map.max_entries = attr->max_entries;
91 	array->elem_size = elem_size;
92 
93 	if (!percpu)
94 		goto out;
95 
96 	array_size += (u64) attr->max_entries * elem_size * num_possible_cpus();
97 
98 	if (array_size >= U32_MAX - PAGE_SIZE ||
99 	    elem_size > PCPU_MIN_UNIT_SIZE || bpf_array_alloc_percpu(array)) {
100 		kvfree(array);
101 		return ERR_PTR(-ENOMEM);
102 	}
103 out:
104 	array->map.pages = round_up(array_size, PAGE_SIZE) >> PAGE_SHIFT;
105 
106 	return &array->map;
107 }
108 
109 /* Called from syscall or from eBPF program */
110 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
111 {
112 	struct bpf_array *array = container_of(map, struct bpf_array, map);
113 	u32 index = *(u32 *)key;
114 
115 	if (unlikely(index >= array->map.max_entries))
116 		return NULL;
117 
118 	return array->value + array->elem_size * index;
119 }
120 
121 /* Called from eBPF program */
122 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
123 {
124 	struct bpf_array *array = container_of(map, struct bpf_array, map);
125 	u32 index = *(u32 *)key;
126 
127 	if (unlikely(index >= array->map.max_entries))
128 		return NULL;
129 
130 	return this_cpu_ptr(array->pptrs[index]);
131 }
132 
133 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
134 {
135 	struct bpf_array *array = container_of(map, struct bpf_array, map);
136 	u32 index = *(u32 *)key;
137 	void __percpu *pptr;
138 	int cpu, off = 0;
139 	u32 size;
140 
141 	if (unlikely(index >= array->map.max_entries))
142 		return -ENOENT;
143 
144 	/* per_cpu areas are zero-filled and bpf programs can only
145 	 * access 'value_size' of them, so copying rounded areas
146 	 * will not leak any kernel data
147 	 */
148 	size = round_up(map->value_size, 8);
149 	rcu_read_lock();
150 	pptr = array->pptrs[index];
151 	for_each_possible_cpu(cpu) {
152 		bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
153 		off += size;
154 	}
155 	rcu_read_unlock();
156 	return 0;
157 }
158 
159 /* Called from syscall */
160 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
161 {
162 	struct bpf_array *array = container_of(map, struct bpf_array, map);
163 	u32 index = *(u32 *)key;
164 	u32 *next = (u32 *)next_key;
165 
166 	if (index >= array->map.max_entries) {
167 		*next = 0;
168 		return 0;
169 	}
170 
171 	if (index == array->map.max_entries - 1)
172 		return -ENOENT;
173 
174 	*next = index + 1;
175 	return 0;
176 }
177 
178 /* Called from syscall or from eBPF program */
179 static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
180 				 u64 map_flags)
181 {
182 	struct bpf_array *array = container_of(map, struct bpf_array, map);
183 	u32 index = *(u32 *)key;
184 
185 	if (unlikely(map_flags > BPF_EXIST))
186 		/* unknown flags */
187 		return -EINVAL;
188 
189 	if (unlikely(index >= array->map.max_entries))
190 		/* all elements were pre-allocated, cannot insert a new one */
191 		return -E2BIG;
192 
193 	if (unlikely(map_flags == BPF_NOEXIST))
194 		/* all elements already exist */
195 		return -EEXIST;
196 
197 	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
198 		memcpy(this_cpu_ptr(array->pptrs[index]),
199 		       value, map->value_size);
200 	else
201 		memcpy(array->value + array->elem_size * index,
202 		       value, map->value_size);
203 	return 0;
204 }
205 
206 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
207 			    u64 map_flags)
208 {
209 	struct bpf_array *array = container_of(map, struct bpf_array, map);
210 	u32 index = *(u32 *)key;
211 	void __percpu *pptr;
212 	int cpu, off = 0;
213 	u32 size;
214 
215 	if (unlikely(map_flags > BPF_EXIST))
216 		/* unknown flags */
217 		return -EINVAL;
218 
219 	if (unlikely(index >= array->map.max_entries))
220 		/* all elements were pre-allocated, cannot insert a new one */
221 		return -E2BIG;
222 
223 	if (unlikely(map_flags == BPF_NOEXIST))
224 		/* all elements already exist */
225 		return -EEXIST;
226 
227 	/* the user space will provide round_up(value_size, 8) bytes that
228 	 * will be copied into per-cpu area. bpf programs can only access
229 	 * value_size of it. During lookup the same extra bytes will be
230 	 * returned or zeros which were zero-filled by percpu_alloc,
231 	 * so no kernel data leaks possible
232 	 */
233 	size = round_up(map->value_size, 8);
234 	rcu_read_lock();
235 	pptr = array->pptrs[index];
236 	for_each_possible_cpu(cpu) {
237 		bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
238 		off += size;
239 	}
240 	rcu_read_unlock();
241 	return 0;
242 }
243 
244 /* Called from syscall or from eBPF program */
245 static int array_map_delete_elem(struct bpf_map *map, void *key)
246 {
247 	return -EINVAL;
248 }
249 
250 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
251 static void array_map_free(struct bpf_map *map)
252 {
253 	struct bpf_array *array = container_of(map, struct bpf_array, map);
254 
255 	/* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
256 	 * so the programs (can be more than one that used this map) were
257 	 * disconnected from events. Wait for outstanding programs to complete
258 	 * and free the array
259 	 */
260 	synchronize_rcu();
261 
262 	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
263 		bpf_array_free_percpu(array);
264 
265 	kvfree(array);
266 }
267 
268 static const struct bpf_map_ops array_ops = {
269 	.map_alloc = array_map_alloc,
270 	.map_free = array_map_free,
271 	.map_get_next_key = array_map_get_next_key,
272 	.map_lookup_elem = array_map_lookup_elem,
273 	.map_update_elem = array_map_update_elem,
274 	.map_delete_elem = array_map_delete_elem,
275 };
276 
277 static struct bpf_map_type_list array_type __read_mostly = {
278 	.ops = &array_ops,
279 	.type = BPF_MAP_TYPE_ARRAY,
280 };
281 
282 static const struct bpf_map_ops percpu_array_ops = {
283 	.map_alloc = array_map_alloc,
284 	.map_free = array_map_free,
285 	.map_get_next_key = array_map_get_next_key,
286 	.map_lookup_elem = percpu_array_map_lookup_elem,
287 	.map_update_elem = array_map_update_elem,
288 	.map_delete_elem = array_map_delete_elem,
289 };
290 
291 static struct bpf_map_type_list percpu_array_type __read_mostly = {
292 	.ops = &percpu_array_ops,
293 	.type = BPF_MAP_TYPE_PERCPU_ARRAY,
294 };
295 
296 static int __init register_array_map(void)
297 {
298 	bpf_register_map_type(&array_type);
299 	bpf_register_map_type(&percpu_array_type);
300 	return 0;
301 }
302 late_initcall(register_array_map);
303 
304 static struct bpf_map *fd_array_map_alloc(union bpf_attr *attr)
305 {
306 	/* only file descriptors can be stored in this type of map */
307 	if (attr->value_size != sizeof(u32))
308 		return ERR_PTR(-EINVAL);
309 	return array_map_alloc(attr);
310 }
311 
312 static void fd_array_map_free(struct bpf_map *map)
313 {
314 	struct bpf_array *array = container_of(map, struct bpf_array, map);
315 	int i;
316 
317 	synchronize_rcu();
318 
319 	/* make sure it's empty */
320 	for (i = 0; i < array->map.max_entries; i++)
321 		BUG_ON(array->ptrs[i] != NULL);
322 	kvfree(array);
323 }
324 
325 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
326 {
327 	return NULL;
328 }
329 
330 /* only called from syscall */
331 static int fd_array_map_update_elem(struct bpf_map *map, void *key,
332 				    void *value, u64 map_flags)
333 {
334 	struct bpf_array *array = container_of(map, struct bpf_array, map);
335 	void *new_ptr, *old_ptr;
336 	u32 index = *(u32 *)key, ufd;
337 
338 	if (map_flags != BPF_ANY)
339 		return -EINVAL;
340 
341 	if (index >= array->map.max_entries)
342 		return -E2BIG;
343 
344 	ufd = *(u32 *)value;
345 	new_ptr = map->ops->map_fd_get_ptr(map, ufd);
346 	if (IS_ERR(new_ptr))
347 		return PTR_ERR(new_ptr);
348 
349 	old_ptr = xchg(array->ptrs + index, new_ptr);
350 	if (old_ptr)
351 		map->ops->map_fd_put_ptr(old_ptr);
352 
353 	return 0;
354 }
355 
356 static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
357 {
358 	struct bpf_array *array = container_of(map, struct bpf_array, map);
359 	void *old_ptr;
360 	u32 index = *(u32 *)key;
361 
362 	if (index >= array->map.max_entries)
363 		return -E2BIG;
364 
365 	old_ptr = xchg(array->ptrs + index, NULL);
366 	if (old_ptr) {
367 		map->ops->map_fd_put_ptr(old_ptr);
368 		return 0;
369 	} else {
370 		return -ENOENT;
371 	}
372 }
373 
374 static void *prog_fd_array_get_ptr(struct bpf_map *map, int fd)
375 {
376 	struct bpf_array *array = container_of(map, struct bpf_array, map);
377 	struct bpf_prog *prog = bpf_prog_get(fd);
378 	if (IS_ERR(prog))
379 		return prog;
380 
381 	if (!bpf_prog_array_compatible(array, prog)) {
382 		bpf_prog_put(prog);
383 		return ERR_PTR(-EINVAL);
384 	}
385 	return prog;
386 }
387 
388 static void prog_fd_array_put_ptr(void *ptr)
389 {
390 	struct bpf_prog *prog = ptr;
391 
392 	bpf_prog_put_rcu(prog);
393 }
394 
395 /* decrement refcnt of all bpf_progs that are stored in this map */
396 void bpf_fd_array_map_clear(struct bpf_map *map)
397 {
398 	struct bpf_array *array = container_of(map, struct bpf_array, map);
399 	int i;
400 
401 	for (i = 0; i < array->map.max_entries; i++)
402 		fd_array_map_delete_elem(map, &i);
403 }
404 
405 static const struct bpf_map_ops prog_array_ops = {
406 	.map_alloc = fd_array_map_alloc,
407 	.map_free = fd_array_map_free,
408 	.map_get_next_key = array_map_get_next_key,
409 	.map_lookup_elem = fd_array_map_lookup_elem,
410 	.map_update_elem = fd_array_map_update_elem,
411 	.map_delete_elem = fd_array_map_delete_elem,
412 	.map_fd_get_ptr = prog_fd_array_get_ptr,
413 	.map_fd_put_ptr = prog_fd_array_put_ptr,
414 };
415 
416 static struct bpf_map_type_list prog_array_type __read_mostly = {
417 	.ops = &prog_array_ops,
418 	.type = BPF_MAP_TYPE_PROG_ARRAY,
419 };
420 
421 static int __init register_prog_array_map(void)
422 {
423 	bpf_register_map_type(&prog_array_type);
424 	return 0;
425 }
426 late_initcall(register_prog_array_map);
427 
428 static void perf_event_array_map_free(struct bpf_map *map)
429 {
430 	bpf_fd_array_map_clear(map);
431 	fd_array_map_free(map);
432 }
433 
434 static void *perf_event_fd_array_get_ptr(struct bpf_map *map, int fd)
435 {
436 	struct perf_event *event;
437 	const struct perf_event_attr *attr;
438 	struct file *file;
439 
440 	file = perf_event_get(fd);
441 	if (IS_ERR(file))
442 		return file;
443 
444 	event = file->private_data;
445 
446 	attr = perf_event_attrs(event);
447 	if (IS_ERR(attr))
448 		goto err;
449 
450 	if (attr->inherit)
451 		goto err;
452 
453 	if (attr->type == PERF_TYPE_RAW)
454 		return file;
455 
456 	if (attr->type == PERF_TYPE_HARDWARE)
457 		return file;
458 
459 	if (attr->type == PERF_TYPE_SOFTWARE &&
460 	    attr->config == PERF_COUNT_SW_BPF_OUTPUT)
461 		return file;
462 err:
463 	fput(file);
464 	return ERR_PTR(-EINVAL);
465 }
466 
467 static void perf_event_fd_array_put_ptr(void *ptr)
468 {
469 	fput((struct file *)ptr);
470 }
471 
472 static const struct bpf_map_ops perf_event_array_ops = {
473 	.map_alloc = fd_array_map_alloc,
474 	.map_free = perf_event_array_map_free,
475 	.map_get_next_key = array_map_get_next_key,
476 	.map_lookup_elem = fd_array_map_lookup_elem,
477 	.map_update_elem = fd_array_map_update_elem,
478 	.map_delete_elem = fd_array_map_delete_elem,
479 	.map_fd_get_ptr = perf_event_fd_array_get_ptr,
480 	.map_fd_put_ptr = perf_event_fd_array_put_ptr,
481 };
482 
483 static struct bpf_map_type_list perf_event_array_type __read_mostly = {
484 	.ops = &perf_event_array_ops,
485 	.type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
486 };
487 
488 static int __init register_perf_event_array_map(void)
489 {
490 	bpf_register_map_type(&perf_event_array_type);
491 	return 0;
492 }
493 late_initcall(register_perf_event_array_map);
494