xref: /openbmc/linux/kernel/bpf/arraymap.c (revision 82e6fdd6)
1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2  * Copyright (c) 2016,2017 Facebook
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  */
13 #include <linux/bpf.h>
14 #include <linux/err.h>
15 #include <linux/slab.h>
16 #include <linux/mm.h>
17 #include <linux/filter.h>
18 #include <linux/perf_event.h>
19 
20 #include "map_in_map.h"
21 
22 #define ARRAY_CREATE_FLAG_MASK \
23 	(BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
24 
25 static void bpf_array_free_percpu(struct bpf_array *array)
26 {
27 	int i;
28 
29 	for (i = 0; i < array->map.max_entries; i++) {
30 		free_percpu(array->pptrs[i]);
31 		cond_resched();
32 	}
33 }
34 
35 static int bpf_array_alloc_percpu(struct bpf_array *array)
36 {
37 	void __percpu *ptr;
38 	int i;
39 
40 	for (i = 0; i < array->map.max_entries; i++) {
41 		ptr = __alloc_percpu_gfp(array->elem_size, 8,
42 					 GFP_USER | __GFP_NOWARN);
43 		if (!ptr) {
44 			bpf_array_free_percpu(array);
45 			return -ENOMEM;
46 		}
47 		array->pptrs[i] = ptr;
48 		cond_resched();
49 	}
50 
51 	return 0;
52 }
53 
54 /* Called from syscall */
55 static int array_map_alloc_check(union bpf_attr *attr)
56 {
57 	bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
58 	int numa_node = bpf_map_attr_numa_node(attr);
59 
60 	/* check sanity of attributes */
61 	if (attr->max_entries == 0 || attr->key_size != 4 ||
62 	    attr->value_size == 0 ||
63 	    attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
64 	    (percpu && numa_node != NUMA_NO_NODE))
65 		return -EINVAL;
66 
67 	if (attr->value_size > KMALLOC_MAX_SIZE)
68 		/* if value_size is bigger, the user space won't be able to
69 		 * access the elements.
70 		 */
71 		return -E2BIG;
72 
73 	return 0;
74 }
75 
76 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
77 {
78 	bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
79 	int ret, numa_node = bpf_map_attr_numa_node(attr);
80 	u32 elem_size, index_mask, max_entries;
81 	bool unpriv = !capable(CAP_SYS_ADMIN);
82 	u64 cost, array_size, mask64;
83 	struct bpf_array *array;
84 
85 	elem_size = round_up(attr->value_size, 8);
86 
87 	max_entries = attr->max_entries;
88 
89 	/* On 32 bit archs roundup_pow_of_two() with max_entries that has
90 	 * upper most bit set in u32 space is undefined behavior due to
91 	 * resulting 1U << 32, so do it manually here in u64 space.
92 	 */
93 	mask64 = fls_long(max_entries - 1);
94 	mask64 = 1ULL << mask64;
95 	mask64 -= 1;
96 
97 	index_mask = mask64;
98 	if (unpriv) {
99 		/* round up array size to nearest power of 2,
100 		 * since cpu will speculate within index_mask limits
101 		 */
102 		max_entries = index_mask + 1;
103 		/* Check for overflows. */
104 		if (max_entries < attr->max_entries)
105 			return ERR_PTR(-E2BIG);
106 	}
107 
108 	array_size = sizeof(*array);
109 	if (percpu)
110 		array_size += (u64) max_entries * sizeof(void *);
111 	else
112 		array_size += (u64) max_entries * elem_size;
113 
114 	/* make sure there is no u32 overflow later in round_up() */
115 	cost = array_size;
116 	if (cost >= U32_MAX - PAGE_SIZE)
117 		return ERR_PTR(-ENOMEM);
118 	if (percpu) {
119 		cost += (u64)attr->max_entries * elem_size * num_possible_cpus();
120 		if (cost >= U32_MAX - PAGE_SIZE)
121 			return ERR_PTR(-ENOMEM);
122 	}
123 	cost = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
124 
125 	ret = bpf_map_precharge_memlock(cost);
126 	if (ret < 0)
127 		return ERR_PTR(ret);
128 
129 	/* allocate all map elements and zero-initialize them */
130 	array = bpf_map_area_alloc(array_size, numa_node);
131 	if (!array)
132 		return ERR_PTR(-ENOMEM);
133 	array->index_mask = index_mask;
134 	array->map.unpriv_array = unpriv;
135 
136 	/* copy mandatory map attributes */
137 	bpf_map_init_from_attr(&array->map, attr);
138 	array->map.pages = cost;
139 	array->elem_size = elem_size;
140 
141 	if (percpu && bpf_array_alloc_percpu(array)) {
142 		bpf_map_area_free(array);
143 		return ERR_PTR(-ENOMEM);
144 	}
145 
146 	return &array->map;
147 }
148 
149 /* Called from syscall or from eBPF program */
150 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
151 {
152 	struct bpf_array *array = container_of(map, struct bpf_array, map);
153 	u32 index = *(u32 *)key;
154 
155 	if (unlikely(index >= array->map.max_entries))
156 		return NULL;
157 
158 	return array->value + array->elem_size * (index & array->index_mask);
159 }
160 
161 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
162 static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
163 {
164 	struct bpf_array *array = container_of(map, struct bpf_array, map);
165 	struct bpf_insn *insn = insn_buf;
166 	u32 elem_size = round_up(map->value_size, 8);
167 	const int ret = BPF_REG_0;
168 	const int map_ptr = BPF_REG_1;
169 	const int index = BPF_REG_2;
170 
171 	*insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
172 	*insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
173 	if (map->unpriv_array) {
174 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
175 		*insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
176 	} else {
177 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
178 	}
179 
180 	if (is_power_of_2(elem_size)) {
181 		*insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
182 	} else {
183 		*insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
184 	}
185 	*insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
186 	*insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
187 	*insn++ = BPF_MOV64_IMM(ret, 0);
188 	return insn - insn_buf;
189 }
190 
191 /* Called from eBPF program */
192 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
193 {
194 	struct bpf_array *array = container_of(map, struct bpf_array, map);
195 	u32 index = *(u32 *)key;
196 
197 	if (unlikely(index >= array->map.max_entries))
198 		return NULL;
199 
200 	return this_cpu_ptr(array->pptrs[index & array->index_mask]);
201 }
202 
203 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
204 {
205 	struct bpf_array *array = container_of(map, struct bpf_array, map);
206 	u32 index = *(u32 *)key;
207 	void __percpu *pptr;
208 	int cpu, off = 0;
209 	u32 size;
210 
211 	if (unlikely(index >= array->map.max_entries))
212 		return -ENOENT;
213 
214 	/* per_cpu areas are zero-filled and bpf programs can only
215 	 * access 'value_size' of them, so copying rounded areas
216 	 * will not leak any kernel data
217 	 */
218 	size = round_up(map->value_size, 8);
219 	rcu_read_lock();
220 	pptr = array->pptrs[index & array->index_mask];
221 	for_each_possible_cpu(cpu) {
222 		bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
223 		off += size;
224 	}
225 	rcu_read_unlock();
226 	return 0;
227 }
228 
229 /* Called from syscall */
230 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
231 {
232 	struct bpf_array *array = container_of(map, struct bpf_array, map);
233 	u32 index = key ? *(u32 *)key : U32_MAX;
234 	u32 *next = (u32 *)next_key;
235 
236 	if (index >= array->map.max_entries) {
237 		*next = 0;
238 		return 0;
239 	}
240 
241 	if (index == array->map.max_entries - 1)
242 		return -ENOENT;
243 
244 	*next = index + 1;
245 	return 0;
246 }
247 
248 /* Called from syscall or from eBPF program */
249 static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
250 				 u64 map_flags)
251 {
252 	struct bpf_array *array = container_of(map, struct bpf_array, map);
253 	u32 index = *(u32 *)key;
254 
255 	if (unlikely(map_flags > BPF_EXIST))
256 		/* unknown flags */
257 		return -EINVAL;
258 
259 	if (unlikely(index >= array->map.max_entries))
260 		/* all elements were pre-allocated, cannot insert a new one */
261 		return -E2BIG;
262 
263 	if (unlikely(map_flags == BPF_NOEXIST))
264 		/* all elements already exist */
265 		return -EEXIST;
266 
267 	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
268 		memcpy(this_cpu_ptr(array->pptrs[index & array->index_mask]),
269 		       value, map->value_size);
270 	else
271 		memcpy(array->value +
272 		       array->elem_size * (index & array->index_mask),
273 		       value, map->value_size);
274 	return 0;
275 }
276 
277 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
278 			    u64 map_flags)
279 {
280 	struct bpf_array *array = container_of(map, struct bpf_array, map);
281 	u32 index = *(u32 *)key;
282 	void __percpu *pptr;
283 	int cpu, off = 0;
284 	u32 size;
285 
286 	if (unlikely(map_flags > BPF_EXIST))
287 		/* unknown flags */
288 		return -EINVAL;
289 
290 	if (unlikely(index >= array->map.max_entries))
291 		/* all elements were pre-allocated, cannot insert a new one */
292 		return -E2BIG;
293 
294 	if (unlikely(map_flags == BPF_NOEXIST))
295 		/* all elements already exist */
296 		return -EEXIST;
297 
298 	/* the user space will provide round_up(value_size, 8) bytes that
299 	 * will be copied into per-cpu area. bpf programs can only access
300 	 * value_size of it. During lookup the same extra bytes will be
301 	 * returned or zeros which were zero-filled by percpu_alloc,
302 	 * so no kernel data leaks possible
303 	 */
304 	size = round_up(map->value_size, 8);
305 	rcu_read_lock();
306 	pptr = array->pptrs[index & array->index_mask];
307 	for_each_possible_cpu(cpu) {
308 		bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
309 		off += size;
310 	}
311 	rcu_read_unlock();
312 	return 0;
313 }
314 
315 /* Called from syscall or from eBPF program */
316 static int array_map_delete_elem(struct bpf_map *map, void *key)
317 {
318 	return -EINVAL;
319 }
320 
321 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
322 static void array_map_free(struct bpf_map *map)
323 {
324 	struct bpf_array *array = container_of(map, struct bpf_array, map);
325 
326 	/* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
327 	 * so the programs (can be more than one that used this map) were
328 	 * disconnected from events. Wait for outstanding programs to complete
329 	 * and free the array
330 	 */
331 	synchronize_rcu();
332 
333 	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
334 		bpf_array_free_percpu(array);
335 
336 	bpf_map_area_free(array);
337 }
338 
339 const struct bpf_map_ops array_map_ops = {
340 	.map_alloc_check = array_map_alloc_check,
341 	.map_alloc = array_map_alloc,
342 	.map_free = array_map_free,
343 	.map_get_next_key = array_map_get_next_key,
344 	.map_lookup_elem = array_map_lookup_elem,
345 	.map_update_elem = array_map_update_elem,
346 	.map_delete_elem = array_map_delete_elem,
347 	.map_gen_lookup = array_map_gen_lookup,
348 };
349 
350 const struct bpf_map_ops percpu_array_map_ops = {
351 	.map_alloc_check = array_map_alloc_check,
352 	.map_alloc = array_map_alloc,
353 	.map_free = array_map_free,
354 	.map_get_next_key = array_map_get_next_key,
355 	.map_lookup_elem = percpu_array_map_lookup_elem,
356 	.map_update_elem = array_map_update_elem,
357 	.map_delete_elem = array_map_delete_elem,
358 };
359 
360 static int fd_array_map_alloc_check(union bpf_attr *attr)
361 {
362 	/* only file descriptors can be stored in this type of map */
363 	if (attr->value_size != sizeof(u32))
364 		return -EINVAL;
365 	return array_map_alloc_check(attr);
366 }
367 
368 static void fd_array_map_free(struct bpf_map *map)
369 {
370 	struct bpf_array *array = container_of(map, struct bpf_array, map);
371 	int i;
372 
373 	synchronize_rcu();
374 
375 	/* make sure it's empty */
376 	for (i = 0; i < array->map.max_entries; i++)
377 		BUG_ON(array->ptrs[i] != NULL);
378 
379 	bpf_map_area_free(array);
380 }
381 
382 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
383 {
384 	return NULL;
385 }
386 
387 /* only called from syscall */
388 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
389 {
390 	void **elem, *ptr;
391 	int ret =  0;
392 
393 	if (!map->ops->map_fd_sys_lookup_elem)
394 		return -ENOTSUPP;
395 
396 	rcu_read_lock();
397 	elem = array_map_lookup_elem(map, key);
398 	if (elem && (ptr = READ_ONCE(*elem)))
399 		*value = map->ops->map_fd_sys_lookup_elem(ptr);
400 	else
401 		ret = -ENOENT;
402 	rcu_read_unlock();
403 
404 	return ret;
405 }
406 
407 /* only called from syscall */
408 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
409 				 void *key, void *value, u64 map_flags)
410 {
411 	struct bpf_array *array = container_of(map, struct bpf_array, map);
412 	void *new_ptr, *old_ptr;
413 	u32 index = *(u32 *)key, ufd;
414 
415 	if (map_flags != BPF_ANY)
416 		return -EINVAL;
417 
418 	if (index >= array->map.max_entries)
419 		return -E2BIG;
420 
421 	ufd = *(u32 *)value;
422 	new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
423 	if (IS_ERR(new_ptr))
424 		return PTR_ERR(new_ptr);
425 
426 	old_ptr = xchg(array->ptrs + index, new_ptr);
427 	if (old_ptr)
428 		map->ops->map_fd_put_ptr(old_ptr);
429 
430 	return 0;
431 }
432 
433 static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
434 {
435 	struct bpf_array *array = container_of(map, struct bpf_array, map);
436 	void *old_ptr;
437 	u32 index = *(u32 *)key;
438 
439 	if (index >= array->map.max_entries)
440 		return -E2BIG;
441 
442 	old_ptr = xchg(array->ptrs + index, NULL);
443 	if (old_ptr) {
444 		map->ops->map_fd_put_ptr(old_ptr);
445 		return 0;
446 	} else {
447 		return -ENOENT;
448 	}
449 }
450 
451 static void *prog_fd_array_get_ptr(struct bpf_map *map,
452 				   struct file *map_file, int fd)
453 {
454 	struct bpf_array *array = container_of(map, struct bpf_array, map);
455 	struct bpf_prog *prog = bpf_prog_get(fd);
456 
457 	if (IS_ERR(prog))
458 		return prog;
459 
460 	if (!bpf_prog_array_compatible(array, prog)) {
461 		bpf_prog_put(prog);
462 		return ERR_PTR(-EINVAL);
463 	}
464 
465 	return prog;
466 }
467 
468 static void prog_fd_array_put_ptr(void *ptr)
469 {
470 	bpf_prog_put(ptr);
471 }
472 
473 static u32 prog_fd_array_sys_lookup_elem(void *ptr)
474 {
475 	return ((struct bpf_prog *)ptr)->aux->id;
476 }
477 
478 /* decrement refcnt of all bpf_progs that are stored in this map */
479 void bpf_fd_array_map_clear(struct bpf_map *map)
480 {
481 	struct bpf_array *array = container_of(map, struct bpf_array, map);
482 	int i;
483 
484 	for (i = 0; i < array->map.max_entries; i++)
485 		fd_array_map_delete_elem(map, &i);
486 }
487 
488 const struct bpf_map_ops prog_array_map_ops = {
489 	.map_alloc_check = fd_array_map_alloc_check,
490 	.map_alloc = array_map_alloc,
491 	.map_free = fd_array_map_free,
492 	.map_get_next_key = array_map_get_next_key,
493 	.map_lookup_elem = fd_array_map_lookup_elem,
494 	.map_delete_elem = fd_array_map_delete_elem,
495 	.map_fd_get_ptr = prog_fd_array_get_ptr,
496 	.map_fd_put_ptr = prog_fd_array_put_ptr,
497 	.map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
498 };
499 
500 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
501 						   struct file *map_file)
502 {
503 	struct bpf_event_entry *ee;
504 
505 	ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
506 	if (ee) {
507 		ee->event = perf_file->private_data;
508 		ee->perf_file = perf_file;
509 		ee->map_file = map_file;
510 	}
511 
512 	return ee;
513 }
514 
515 static void __bpf_event_entry_free(struct rcu_head *rcu)
516 {
517 	struct bpf_event_entry *ee;
518 
519 	ee = container_of(rcu, struct bpf_event_entry, rcu);
520 	fput(ee->perf_file);
521 	kfree(ee);
522 }
523 
524 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
525 {
526 	call_rcu(&ee->rcu, __bpf_event_entry_free);
527 }
528 
529 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
530 					 struct file *map_file, int fd)
531 {
532 	struct bpf_event_entry *ee;
533 	struct perf_event *event;
534 	struct file *perf_file;
535 	u64 value;
536 
537 	perf_file = perf_event_get(fd);
538 	if (IS_ERR(perf_file))
539 		return perf_file;
540 
541 	ee = ERR_PTR(-EOPNOTSUPP);
542 	event = perf_file->private_data;
543 	if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
544 		goto err_out;
545 
546 	ee = bpf_event_entry_gen(perf_file, map_file);
547 	if (ee)
548 		return ee;
549 	ee = ERR_PTR(-ENOMEM);
550 err_out:
551 	fput(perf_file);
552 	return ee;
553 }
554 
555 static void perf_event_fd_array_put_ptr(void *ptr)
556 {
557 	bpf_event_entry_free_rcu(ptr);
558 }
559 
560 static void perf_event_fd_array_release(struct bpf_map *map,
561 					struct file *map_file)
562 {
563 	struct bpf_array *array = container_of(map, struct bpf_array, map);
564 	struct bpf_event_entry *ee;
565 	int i;
566 
567 	rcu_read_lock();
568 	for (i = 0; i < array->map.max_entries; i++) {
569 		ee = READ_ONCE(array->ptrs[i]);
570 		if (ee && ee->map_file == map_file)
571 			fd_array_map_delete_elem(map, &i);
572 	}
573 	rcu_read_unlock();
574 }
575 
576 const struct bpf_map_ops perf_event_array_map_ops = {
577 	.map_alloc_check = fd_array_map_alloc_check,
578 	.map_alloc = array_map_alloc,
579 	.map_free = fd_array_map_free,
580 	.map_get_next_key = array_map_get_next_key,
581 	.map_lookup_elem = fd_array_map_lookup_elem,
582 	.map_delete_elem = fd_array_map_delete_elem,
583 	.map_fd_get_ptr = perf_event_fd_array_get_ptr,
584 	.map_fd_put_ptr = perf_event_fd_array_put_ptr,
585 	.map_release = perf_event_fd_array_release,
586 };
587 
588 #ifdef CONFIG_CGROUPS
589 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
590 				     struct file *map_file /* not used */,
591 				     int fd)
592 {
593 	return cgroup_get_from_fd(fd);
594 }
595 
596 static void cgroup_fd_array_put_ptr(void *ptr)
597 {
598 	/* cgroup_put free cgrp after a rcu grace period */
599 	cgroup_put(ptr);
600 }
601 
602 static void cgroup_fd_array_free(struct bpf_map *map)
603 {
604 	bpf_fd_array_map_clear(map);
605 	fd_array_map_free(map);
606 }
607 
608 const struct bpf_map_ops cgroup_array_map_ops = {
609 	.map_alloc_check = fd_array_map_alloc_check,
610 	.map_alloc = array_map_alloc,
611 	.map_free = cgroup_fd_array_free,
612 	.map_get_next_key = array_map_get_next_key,
613 	.map_lookup_elem = fd_array_map_lookup_elem,
614 	.map_delete_elem = fd_array_map_delete_elem,
615 	.map_fd_get_ptr = cgroup_fd_array_get_ptr,
616 	.map_fd_put_ptr = cgroup_fd_array_put_ptr,
617 };
618 #endif
619 
620 static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
621 {
622 	struct bpf_map *map, *inner_map_meta;
623 
624 	inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
625 	if (IS_ERR(inner_map_meta))
626 		return inner_map_meta;
627 
628 	map = array_map_alloc(attr);
629 	if (IS_ERR(map)) {
630 		bpf_map_meta_free(inner_map_meta);
631 		return map;
632 	}
633 
634 	map->inner_map_meta = inner_map_meta;
635 
636 	return map;
637 }
638 
639 static void array_of_map_free(struct bpf_map *map)
640 {
641 	/* map->inner_map_meta is only accessed by syscall which
642 	 * is protected by fdget/fdput.
643 	 */
644 	bpf_map_meta_free(map->inner_map_meta);
645 	bpf_fd_array_map_clear(map);
646 	fd_array_map_free(map);
647 }
648 
649 static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
650 {
651 	struct bpf_map **inner_map = array_map_lookup_elem(map, key);
652 
653 	if (!inner_map)
654 		return NULL;
655 
656 	return READ_ONCE(*inner_map);
657 }
658 
659 static u32 array_of_map_gen_lookup(struct bpf_map *map,
660 				   struct bpf_insn *insn_buf)
661 {
662 	struct bpf_array *array = container_of(map, struct bpf_array, map);
663 	u32 elem_size = round_up(map->value_size, 8);
664 	struct bpf_insn *insn = insn_buf;
665 	const int ret = BPF_REG_0;
666 	const int map_ptr = BPF_REG_1;
667 	const int index = BPF_REG_2;
668 
669 	*insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
670 	*insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
671 	if (map->unpriv_array) {
672 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
673 		*insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
674 	} else {
675 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
676 	}
677 	if (is_power_of_2(elem_size))
678 		*insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
679 	else
680 		*insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
681 	*insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
682 	*insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
683 	*insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
684 	*insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
685 	*insn++ = BPF_MOV64_IMM(ret, 0);
686 
687 	return insn - insn_buf;
688 }
689 
690 const struct bpf_map_ops array_of_maps_map_ops = {
691 	.map_alloc_check = fd_array_map_alloc_check,
692 	.map_alloc = array_of_map_alloc,
693 	.map_free = array_of_map_free,
694 	.map_get_next_key = array_map_get_next_key,
695 	.map_lookup_elem = array_of_map_lookup_elem,
696 	.map_delete_elem = fd_array_map_delete_elem,
697 	.map_fd_get_ptr = bpf_map_fd_get_ptr,
698 	.map_fd_put_ptr = bpf_map_fd_put_ptr,
699 	.map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
700 	.map_gen_lookup = array_of_map_gen_lookup,
701 };
702