1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3  * Copyright (c) 2016,2017 Facebook
4  */
5 #include <linux/bpf.h>
6 #include <linux/btf.h>
7 #include <linux/err.h>
8 #include <linux/slab.h>
9 #include <linux/mm.h>
10 #include <linux/filter.h>
11 #include <linux/perf_event.h>
12 #include <uapi/linux/btf.h>
13 #include <linux/rcupdate_trace.h>
14 #include <linux/btf_ids.h>
15 
16 #include "map_in_map.h"
17 
18 #define ARRAY_CREATE_FLAG_MASK \
19 	(BPF_F_NUMA_NODE | BPF_F_MMAPABLE | BPF_F_ACCESS_MASK | \
20 	 BPF_F_PRESERVE_ELEMS | BPF_F_INNER_MAP)
21 
bpf_array_free_percpu(struct bpf_array * array)22 static void bpf_array_free_percpu(struct bpf_array *array)
23 {
24 	int i;
25 
26 	for (i = 0; i < array->map.max_entries; i++) {
27 		free_percpu(array->pptrs[i]);
28 		cond_resched();
29 	}
30 }
31 
bpf_array_alloc_percpu(struct bpf_array * array)32 static int bpf_array_alloc_percpu(struct bpf_array *array)
33 {
34 	void __percpu *ptr;
35 	int i;
36 
37 	for (i = 0; i < array->map.max_entries; i++) {
38 		ptr = bpf_map_alloc_percpu(&array->map, array->elem_size, 8,
39 					   GFP_USER | __GFP_NOWARN);
40 		if (!ptr) {
41 			bpf_array_free_percpu(array);
42 			return -ENOMEM;
43 		}
44 		array->pptrs[i] = ptr;
45 		cond_resched();
46 	}
47 
48 	return 0;
49 }
50 
51 /* Called from syscall */
array_map_alloc_check(union bpf_attr * attr)52 int array_map_alloc_check(union bpf_attr *attr)
53 {
54 	bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
55 	int numa_node = bpf_map_attr_numa_node(attr);
56 
57 	/* check sanity of attributes */
58 	if (attr->max_entries == 0 || attr->key_size != 4 ||
59 	    attr->value_size == 0 ||
60 	    attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
61 	    !bpf_map_flags_access_ok(attr->map_flags) ||
62 	    (percpu && numa_node != NUMA_NO_NODE))
63 		return -EINVAL;
64 
65 	if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
66 	    attr->map_flags & (BPF_F_MMAPABLE | BPF_F_INNER_MAP))
67 		return -EINVAL;
68 
69 	if (attr->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
70 	    attr->map_flags & BPF_F_PRESERVE_ELEMS)
71 		return -EINVAL;
72 
73 	/* avoid overflow on round_up(map->value_size) */
74 	if (attr->value_size > INT_MAX)
75 		return -E2BIG;
76 
77 	return 0;
78 }
79 
array_map_alloc(union bpf_attr * attr)80 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
81 {
82 	bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
83 	int numa_node = bpf_map_attr_numa_node(attr);
84 	u32 elem_size, index_mask, max_entries;
85 	bool bypass_spec_v1 = bpf_bypass_spec_v1();
86 	u64 array_size, mask64;
87 	struct bpf_array *array;
88 
89 	elem_size = round_up(attr->value_size, 8);
90 
91 	max_entries = attr->max_entries;
92 
93 	/* On 32 bit archs roundup_pow_of_two() with max_entries that has
94 	 * upper most bit set in u32 space is undefined behavior due to
95 	 * resulting 1U << 32, so do it manually here in u64 space.
96 	 */
97 	mask64 = fls_long(max_entries - 1);
98 	mask64 = 1ULL << mask64;
99 	mask64 -= 1;
100 
101 	index_mask = mask64;
102 	if (!bypass_spec_v1) {
103 		/* round up array size to nearest power of 2,
104 		 * since cpu will speculate within index_mask limits
105 		 */
106 		max_entries = index_mask + 1;
107 		/* Check for overflows. */
108 		if (max_entries < attr->max_entries)
109 			return ERR_PTR(-E2BIG);
110 	}
111 
112 	array_size = sizeof(*array);
113 	if (percpu) {
114 		array_size += (u64) max_entries * sizeof(void *);
115 	} else {
116 		/* rely on vmalloc() to return page-aligned memory and
117 		 * ensure array->value is exactly page-aligned
118 		 */
119 		if (attr->map_flags & BPF_F_MMAPABLE) {
120 			array_size = PAGE_ALIGN(array_size);
121 			array_size += PAGE_ALIGN((u64) max_entries * elem_size);
122 		} else {
123 			array_size += (u64) max_entries * elem_size;
124 		}
125 	}
126 
127 	/* allocate all map elements and zero-initialize them */
128 	if (attr->map_flags & BPF_F_MMAPABLE) {
129 		void *data;
130 
131 		/* kmalloc'ed memory can't be mmap'ed, use explicit vmalloc */
132 		data = bpf_map_area_mmapable_alloc(array_size, numa_node);
133 		if (!data)
134 			return ERR_PTR(-ENOMEM);
135 		array = data + PAGE_ALIGN(sizeof(struct bpf_array))
136 			- offsetof(struct bpf_array, value);
137 	} else {
138 		array = bpf_map_area_alloc(array_size, numa_node);
139 	}
140 	if (!array)
141 		return ERR_PTR(-ENOMEM);
142 	array->index_mask = index_mask;
143 	array->map.bypass_spec_v1 = bypass_spec_v1;
144 
145 	/* copy mandatory map attributes */
146 	bpf_map_init_from_attr(&array->map, attr);
147 	array->elem_size = elem_size;
148 
149 	if (percpu && bpf_array_alloc_percpu(array)) {
150 		bpf_map_area_free(array);
151 		return ERR_PTR(-ENOMEM);
152 	}
153 
154 	return &array->map;
155 }
156 
array_map_elem_ptr(struct bpf_array * array,u32 index)157 static void *array_map_elem_ptr(struct bpf_array* array, u32 index)
158 {
159 	return array->value + (u64)array->elem_size * index;
160 }
161 
162 /* Called from syscall or from eBPF program */
array_map_lookup_elem(struct bpf_map * map,void * key)163 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
164 {
165 	struct bpf_array *array = container_of(map, struct bpf_array, map);
166 	u32 index = *(u32 *)key;
167 
168 	if (unlikely(index >= array->map.max_entries))
169 		return NULL;
170 
171 	return array->value + (u64)array->elem_size * (index & array->index_mask);
172 }
173 
array_map_direct_value_addr(const struct bpf_map * map,u64 * imm,u32 off)174 static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm,
175 				       u32 off)
176 {
177 	struct bpf_array *array = container_of(map, struct bpf_array, map);
178 
179 	if (map->max_entries != 1)
180 		return -ENOTSUPP;
181 	if (off >= map->value_size)
182 		return -EINVAL;
183 
184 	*imm = (unsigned long)array->value;
185 	return 0;
186 }
187 
array_map_direct_value_meta(const struct bpf_map * map,u64 imm,u32 * off)188 static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm,
189 				       u32 *off)
190 {
191 	struct bpf_array *array = container_of(map, struct bpf_array, map);
192 	u64 base = (unsigned long)array->value;
193 	u64 range = array->elem_size;
194 
195 	if (map->max_entries != 1)
196 		return -ENOTSUPP;
197 	if (imm < base || imm >= base + range)
198 		return -ENOENT;
199 
200 	*off = imm - base;
201 	return 0;
202 }
203 
204 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
array_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)205 static int array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
206 {
207 	struct bpf_array *array = container_of(map, struct bpf_array, map);
208 	struct bpf_insn *insn = insn_buf;
209 	u32 elem_size = array->elem_size;
210 	const int ret = BPF_REG_0;
211 	const int map_ptr = BPF_REG_1;
212 	const int index = BPF_REG_2;
213 
214 	if (map->map_flags & BPF_F_INNER_MAP)
215 		return -EOPNOTSUPP;
216 
217 	*insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
218 	*insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
219 	if (!map->bypass_spec_v1) {
220 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
221 		*insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
222 	} else {
223 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
224 	}
225 
226 	if (is_power_of_2(elem_size)) {
227 		*insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
228 	} else {
229 		*insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
230 	}
231 	*insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
232 	*insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
233 	*insn++ = BPF_MOV64_IMM(ret, 0);
234 	return insn - insn_buf;
235 }
236 
237 /* Called from eBPF program */
percpu_array_map_lookup_elem(struct bpf_map * map,void * key)238 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
239 {
240 	struct bpf_array *array = container_of(map, struct bpf_array, map);
241 	u32 index = *(u32 *)key;
242 
243 	if (unlikely(index >= array->map.max_entries))
244 		return NULL;
245 
246 	return this_cpu_ptr(array->pptrs[index & array->index_mask]);
247 }
248 
percpu_array_map_lookup_percpu_elem(struct bpf_map * map,void * key,u32 cpu)249 static void *percpu_array_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
250 {
251 	struct bpf_array *array = container_of(map, struct bpf_array, map);
252 	u32 index = *(u32 *)key;
253 
254 	if (cpu >= nr_cpu_ids)
255 		return NULL;
256 
257 	if (unlikely(index >= array->map.max_entries))
258 		return NULL;
259 
260 	return per_cpu_ptr(array->pptrs[index & array->index_mask], cpu);
261 }
262 
bpf_percpu_array_copy(struct bpf_map * map,void * key,void * value)263 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
264 {
265 	struct bpf_array *array = container_of(map, struct bpf_array, map);
266 	u32 index = *(u32 *)key;
267 	void __percpu *pptr;
268 	int cpu, off = 0;
269 	u32 size;
270 
271 	if (unlikely(index >= array->map.max_entries))
272 		return -ENOENT;
273 
274 	/* per_cpu areas are zero-filled and bpf programs can only
275 	 * access 'value_size' of them, so copying rounded areas
276 	 * will not leak any kernel data
277 	 */
278 	size = array->elem_size;
279 	rcu_read_lock();
280 	pptr = array->pptrs[index & array->index_mask];
281 	for_each_possible_cpu(cpu) {
282 		copy_map_value_long(map, value + off, per_cpu_ptr(pptr, cpu));
283 		check_and_init_map_value(map, value + off);
284 		off += size;
285 	}
286 	rcu_read_unlock();
287 	return 0;
288 }
289 
290 /* Called from syscall */
array_map_get_next_key(struct bpf_map * map,void * key,void * next_key)291 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
292 {
293 	struct bpf_array *array = container_of(map, struct bpf_array, map);
294 	u32 index = key ? *(u32 *)key : U32_MAX;
295 	u32 *next = (u32 *)next_key;
296 
297 	if (index >= array->map.max_entries) {
298 		*next = 0;
299 		return 0;
300 	}
301 
302 	if (index == array->map.max_entries - 1)
303 		return -ENOENT;
304 
305 	*next = index + 1;
306 	return 0;
307 }
308 
309 /* Called from syscall or from eBPF program */
array_map_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags)310 static long array_map_update_elem(struct bpf_map *map, void *key, void *value,
311 				  u64 map_flags)
312 {
313 	struct bpf_array *array = container_of(map, struct bpf_array, map);
314 	u32 index = *(u32 *)key;
315 	char *val;
316 
317 	if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
318 		/* unknown flags */
319 		return -EINVAL;
320 
321 	if (unlikely(index >= array->map.max_entries))
322 		/* all elements were pre-allocated, cannot insert a new one */
323 		return -E2BIG;
324 
325 	if (unlikely(map_flags & BPF_NOEXIST))
326 		/* all elements already exist */
327 		return -EEXIST;
328 
329 	if (unlikely((map_flags & BPF_F_LOCK) &&
330 		     !btf_record_has_field(map->record, BPF_SPIN_LOCK)))
331 		return -EINVAL;
332 
333 	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
334 		val = this_cpu_ptr(array->pptrs[index & array->index_mask]);
335 		copy_map_value(map, val, value);
336 		bpf_obj_free_fields(array->map.record, val);
337 	} else {
338 		val = array->value +
339 			(u64)array->elem_size * (index & array->index_mask);
340 		if (map_flags & BPF_F_LOCK)
341 			copy_map_value_locked(map, val, value, false);
342 		else
343 			copy_map_value(map, val, value);
344 		bpf_obj_free_fields(array->map.record, val);
345 	}
346 	return 0;
347 }
348 
bpf_percpu_array_update(struct bpf_map * map,void * key,void * value,u64 map_flags)349 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
350 			    u64 map_flags)
351 {
352 	struct bpf_array *array = container_of(map, struct bpf_array, map);
353 	u32 index = *(u32 *)key;
354 	void __percpu *pptr;
355 	int cpu, off = 0;
356 	u32 size;
357 
358 	if (unlikely(map_flags > BPF_EXIST))
359 		/* unknown flags */
360 		return -EINVAL;
361 
362 	if (unlikely(index >= array->map.max_entries))
363 		/* all elements were pre-allocated, cannot insert a new one */
364 		return -E2BIG;
365 
366 	if (unlikely(map_flags == BPF_NOEXIST))
367 		/* all elements already exist */
368 		return -EEXIST;
369 
370 	/* the user space will provide round_up(value_size, 8) bytes that
371 	 * will be copied into per-cpu area. bpf programs can only access
372 	 * value_size of it. During lookup the same extra bytes will be
373 	 * returned or zeros which were zero-filled by percpu_alloc,
374 	 * so no kernel data leaks possible
375 	 */
376 	size = array->elem_size;
377 	rcu_read_lock();
378 	pptr = array->pptrs[index & array->index_mask];
379 	for_each_possible_cpu(cpu) {
380 		copy_map_value_long(map, per_cpu_ptr(pptr, cpu), value + off);
381 		bpf_obj_free_fields(array->map.record, per_cpu_ptr(pptr, cpu));
382 		off += size;
383 	}
384 	rcu_read_unlock();
385 	return 0;
386 }
387 
388 /* Called from syscall or from eBPF program */
array_map_delete_elem(struct bpf_map * map,void * key)389 static long array_map_delete_elem(struct bpf_map *map, void *key)
390 {
391 	return -EINVAL;
392 }
393 
array_map_vmalloc_addr(struct bpf_array * array)394 static void *array_map_vmalloc_addr(struct bpf_array *array)
395 {
396 	return (void *)round_down((unsigned long)array, PAGE_SIZE);
397 }
398 
array_map_free_timers(struct bpf_map * map)399 static void array_map_free_timers(struct bpf_map *map)
400 {
401 	struct bpf_array *array = container_of(map, struct bpf_array, map);
402 	int i;
403 
404 	/* We don't reset or free fields other than timer on uref dropping to zero. */
405 	if (!btf_record_has_field(map->record, BPF_TIMER))
406 		return;
407 
408 	for (i = 0; i < array->map.max_entries; i++)
409 		bpf_obj_free_timer(map->record, array_map_elem_ptr(array, i));
410 }
411 
412 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
array_map_free(struct bpf_map * map)413 static void array_map_free(struct bpf_map *map)
414 {
415 	struct bpf_array *array = container_of(map, struct bpf_array, map);
416 	int i;
417 
418 	if (!IS_ERR_OR_NULL(map->record)) {
419 		if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
420 			for (i = 0; i < array->map.max_entries; i++) {
421 				void __percpu *pptr = array->pptrs[i & array->index_mask];
422 				int cpu;
423 
424 				for_each_possible_cpu(cpu) {
425 					bpf_obj_free_fields(map->record, per_cpu_ptr(pptr, cpu));
426 					cond_resched();
427 				}
428 			}
429 		} else {
430 			for (i = 0; i < array->map.max_entries; i++)
431 				bpf_obj_free_fields(map->record, array_map_elem_ptr(array, i));
432 		}
433 	}
434 
435 	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
436 		bpf_array_free_percpu(array);
437 
438 	if (array->map.map_flags & BPF_F_MMAPABLE)
439 		bpf_map_area_free(array_map_vmalloc_addr(array));
440 	else
441 		bpf_map_area_free(array);
442 }
443 
array_map_seq_show_elem(struct bpf_map * map,void * key,struct seq_file * m)444 static void array_map_seq_show_elem(struct bpf_map *map, void *key,
445 				    struct seq_file *m)
446 {
447 	void *value;
448 
449 	rcu_read_lock();
450 
451 	value = array_map_lookup_elem(map, key);
452 	if (!value) {
453 		rcu_read_unlock();
454 		return;
455 	}
456 
457 	if (map->btf_key_type_id)
458 		seq_printf(m, "%u: ", *(u32 *)key);
459 	btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
460 	seq_puts(m, "\n");
461 
462 	rcu_read_unlock();
463 }
464 
percpu_array_map_seq_show_elem(struct bpf_map * map,void * key,struct seq_file * m)465 static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
466 					   struct seq_file *m)
467 {
468 	struct bpf_array *array = container_of(map, struct bpf_array, map);
469 	u32 index = *(u32 *)key;
470 	void __percpu *pptr;
471 	int cpu;
472 
473 	rcu_read_lock();
474 
475 	seq_printf(m, "%u: {\n", *(u32 *)key);
476 	pptr = array->pptrs[index & array->index_mask];
477 	for_each_possible_cpu(cpu) {
478 		seq_printf(m, "\tcpu%d: ", cpu);
479 		btf_type_seq_show(map->btf, map->btf_value_type_id,
480 				  per_cpu_ptr(pptr, cpu), m);
481 		seq_puts(m, "\n");
482 	}
483 	seq_puts(m, "}\n");
484 
485 	rcu_read_unlock();
486 }
487 
array_map_check_btf(const struct bpf_map * map,const struct btf * btf,const struct btf_type * key_type,const struct btf_type * value_type)488 static int array_map_check_btf(const struct bpf_map *map,
489 			       const struct btf *btf,
490 			       const struct btf_type *key_type,
491 			       const struct btf_type *value_type)
492 {
493 	u32 int_data;
494 
495 	/* One exception for keyless BTF: .bss/.data/.rodata map */
496 	if (btf_type_is_void(key_type)) {
497 		if (map->map_type != BPF_MAP_TYPE_ARRAY ||
498 		    map->max_entries != 1)
499 			return -EINVAL;
500 
501 		if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC)
502 			return -EINVAL;
503 
504 		return 0;
505 	}
506 
507 	if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT)
508 		return -EINVAL;
509 
510 	int_data = *(u32 *)(key_type + 1);
511 	/* bpf array can only take a u32 key. This check makes sure
512 	 * that the btf matches the attr used during map_create.
513 	 */
514 	if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data))
515 		return -EINVAL;
516 
517 	return 0;
518 }
519 
array_map_mmap(struct bpf_map * map,struct vm_area_struct * vma)520 static int array_map_mmap(struct bpf_map *map, struct vm_area_struct *vma)
521 {
522 	struct bpf_array *array = container_of(map, struct bpf_array, map);
523 	pgoff_t pgoff = PAGE_ALIGN(sizeof(*array)) >> PAGE_SHIFT;
524 
525 	if (!(map->map_flags & BPF_F_MMAPABLE))
526 		return -EINVAL;
527 
528 	if (vma->vm_pgoff * PAGE_SIZE + (vma->vm_end - vma->vm_start) >
529 	    PAGE_ALIGN((u64)array->map.max_entries * array->elem_size))
530 		return -EINVAL;
531 
532 	return remap_vmalloc_range(vma, array_map_vmalloc_addr(array),
533 				   vma->vm_pgoff + pgoff);
534 }
535 
array_map_meta_equal(const struct bpf_map * meta0,const struct bpf_map * meta1)536 static bool array_map_meta_equal(const struct bpf_map *meta0,
537 				 const struct bpf_map *meta1)
538 {
539 	if (!bpf_map_meta_equal(meta0, meta1))
540 		return false;
541 	return meta0->map_flags & BPF_F_INNER_MAP ? true :
542 	       meta0->max_entries == meta1->max_entries;
543 }
544 
545 struct bpf_iter_seq_array_map_info {
546 	struct bpf_map *map;
547 	void *percpu_value_buf;
548 	u32 index;
549 };
550 
bpf_array_map_seq_start(struct seq_file * seq,loff_t * pos)551 static void *bpf_array_map_seq_start(struct seq_file *seq, loff_t *pos)
552 {
553 	struct bpf_iter_seq_array_map_info *info = seq->private;
554 	struct bpf_map *map = info->map;
555 	struct bpf_array *array;
556 	u32 index;
557 
558 	if (info->index >= map->max_entries)
559 		return NULL;
560 
561 	if (*pos == 0)
562 		++*pos;
563 	array = container_of(map, struct bpf_array, map);
564 	index = info->index & array->index_mask;
565 	if (info->percpu_value_buf)
566 	       return array->pptrs[index];
567 	return array_map_elem_ptr(array, index);
568 }
569 
bpf_array_map_seq_next(struct seq_file * seq,void * v,loff_t * pos)570 static void *bpf_array_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
571 {
572 	struct bpf_iter_seq_array_map_info *info = seq->private;
573 	struct bpf_map *map = info->map;
574 	struct bpf_array *array;
575 	u32 index;
576 
577 	++*pos;
578 	++info->index;
579 	if (info->index >= map->max_entries)
580 		return NULL;
581 
582 	array = container_of(map, struct bpf_array, map);
583 	index = info->index & array->index_mask;
584 	if (info->percpu_value_buf)
585 	       return array->pptrs[index];
586 	return array_map_elem_ptr(array, index);
587 }
588 
__bpf_array_map_seq_show(struct seq_file * seq,void * v)589 static int __bpf_array_map_seq_show(struct seq_file *seq, void *v)
590 {
591 	struct bpf_iter_seq_array_map_info *info = seq->private;
592 	struct bpf_iter__bpf_map_elem ctx = {};
593 	struct bpf_map *map = info->map;
594 	struct bpf_array *array = container_of(map, struct bpf_array, map);
595 	struct bpf_iter_meta meta;
596 	struct bpf_prog *prog;
597 	int off = 0, cpu = 0;
598 	void __percpu **pptr;
599 	u32 size;
600 
601 	meta.seq = seq;
602 	prog = bpf_iter_get_info(&meta, v == NULL);
603 	if (!prog)
604 		return 0;
605 
606 	ctx.meta = &meta;
607 	ctx.map = info->map;
608 	if (v) {
609 		ctx.key = &info->index;
610 
611 		if (!info->percpu_value_buf) {
612 			ctx.value = v;
613 		} else {
614 			pptr = v;
615 			size = array->elem_size;
616 			for_each_possible_cpu(cpu) {
617 				copy_map_value_long(map, info->percpu_value_buf + off,
618 						    per_cpu_ptr(pptr, cpu));
619 				check_and_init_map_value(map, info->percpu_value_buf + off);
620 				off += size;
621 			}
622 			ctx.value = info->percpu_value_buf;
623 		}
624 	}
625 
626 	return bpf_iter_run_prog(prog, &ctx);
627 }
628 
bpf_array_map_seq_show(struct seq_file * seq,void * v)629 static int bpf_array_map_seq_show(struct seq_file *seq, void *v)
630 {
631 	return __bpf_array_map_seq_show(seq, v);
632 }
633 
bpf_array_map_seq_stop(struct seq_file * seq,void * v)634 static void bpf_array_map_seq_stop(struct seq_file *seq, void *v)
635 {
636 	if (!v)
637 		(void)__bpf_array_map_seq_show(seq, NULL);
638 }
639 
bpf_iter_init_array_map(void * priv_data,struct bpf_iter_aux_info * aux)640 static int bpf_iter_init_array_map(void *priv_data,
641 				   struct bpf_iter_aux_info *aux)
642 {
643 	struct bpf_iter_seq_array_map_info *seq_info = priv_data;
644 	struct bpf_map *map = aux->map;
645 	struct bpf_array *array = container_of(map, struct bpf_array, map);
646 	void *value_buf;
647 	u32 buf_size;
648 
649 	if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
650 		buf_size = array->elem_size * num_possible_cpus();
651 		value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
652 		if (!value_buf)
653 			return -ENOMEM;
654 
655 		seq_info->percpu_value_buf = value_buf;
656 	}
657 
658 	/* bpf_iter_attach_map() acquires a map uref, and the uref may be
659 	 * released before or in the middle of iterating map elements, so
660 	 * acquire an extra map uref for iterator.
661 	 */
662 	bpf_map_inc_with_uref(map);
663 	seq_info->map = map;
664 	return 0;
665 }
666 
bpf_iter_fini_array_map(void * priv_data)667 static void bpf_iter_fini_array_map(void *priv_data)
668 {
669 	struct bpf_iter_seq_array_map_info *seq_info = priv_data;
670 
671 	bpf_map_put_with_uref(seq_info->map);
672 	kfree(seq_info->percpu_value_buf);
673 }
674 
675 static const struct seq_operations bpf_array_map_seq_ops = {
676 	.start	= bpf_array_map_seq_start,
677 	.next	= bpf_array_map_seq_next,
678 	.stop	= bpf_array_map_seq_stop,
679 	.show	= bpf_array_map_seq_show,
680 };
681 
682 static const struct bpf_iter_seq_info iter_seq_info = {
683 	.seq_ops		= &bpf_array_map_seq_ops,
684 	.init_seq_private	= bpf_iter_init_array_map,
685 	.fini_seq_private	= bpf_iter_fini_array_map,
686 	.seq_priv_size		= sizeof(struct bpf_iter_seq_array_map_info),
687 };
688 
bpf_for_each_array_elem(struct bpf_map * map,bpf_callback_t callback_fn,void * callback_ctx,u64 flags)689 static long bpf_for_each_array_elem(struct bpf_map *map, bpf_callback_t callback_fn,
690 				    void *callback_ctx, u64 flags)
691 {
692 	u32 i, key, num_elems = 0;
693 	struct bpf_array *array;
694 	bool is_percpu;
695 	u64 ret = 0;
696 	void *val;
697 
698 	if (flags != 0)
699 		return -EINVAL;
700 
701 	is_percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
702 	array = container_of(map, struct bpf_array, map);
703 	if (is_percpu)
704 		migrate_disable();
705 	for (i = 0; i < map->max_entries; i++) {
706 		if (is_percpu)
707 			val = this_cpu_ptr(array->pptrs[i]);
708 		else
709 			val = array_map_elem_ptr(array, i);
710 		num_elems++;
711 		key = i;
712 		ret = callback_fn((u64)(long)map, (u64)(long)&key,
713 				  (u64)(long)val, (u64)(long)callback_ctx, 0);
714 		/* return value: 0 - continue, 1 - stop and return */
715 		if (ret)
716 			break;
717 	}
718 
719 	if (is_percpu)
720 		migrate_enable();
721 	return num_elems;
722 }
723 
array_map_mem_usage(const struct bpf_map * map)724 static u64 array_map_mem_usage(const struct bpf_map *map)
725 {
726 	struct bpf_array *array = container_of(map, struct bpf_array, map);
727 	bool percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
728 	u32 elem_size = array->elem_size;
729 	u64 entries = map->max_entries;
730 	u64 usage = sizeof(*array);
731 
732 	if (percpu) {
733 		usage += entries * sizeof(void *);
734 		usage += entries * elem_size * num_possible_cpus();
735 	} else {
736 		if (map->map_flags & BPF_F_MMAPABLE) {
737 			usage = PAGE_ALIGN(usage);
738 			usage += PAGE_ALIGN(entries * elem_size);
739 		} else {
740 			usage += entries * elem_size;
741 		}
742 	}
743 	return usage;
744 }
745 
746 BTF_ID_LIST_SINGLE(array_map_btf_ids, struct, bpf_array)
747 const struct bpf_map_ops array_map_ops = {
748 	.map_meta_equal = array_map_meta_equal,
749 	.map_alloc_check = array_map_alloc_check,
750 	.map_alloc = array_map_alloc,
751 	.map_free = array_map_free,
752 	.map_get_next_key = array_map_get_next_key,
753 	.map_release_uref = array_map_free_timers,
754 	.map_lookup_elem = array_map_lookup_elem,
755 	.map_update_elem = array_map_update_elem,
756 	.map_delete_elem = array_map_delete_elem,
757 	.map_gen_lookup = array_map_gen_lookup,
758 	.map_direct_value_addr = array_map_direct_value_addr,
759 	.map_direct_value_meta = array_map_direct_value_meta,
760 	.map_mmap = array_map_mmap,
761 	.map_seq_show_elem = array_map_seq_show_elem,
762 	.map_check_btf = array_map_check_btf,
763 	.map_lookup_batch = generic_map_lookup_batch,
764 	.map_update_batch = generic_map_update_batch,
765 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
766 	.map_for_each_callback = bpf_for_each_array_elem,
767 	.map_mem_usage = array_map_mem_usage,
768 	.map_btf_id = &array_map_btf_ids[0],
769 	.iter_seq_info = &iter_seq_info,
770 };
771 
772 const struct bpf_map_ops percpu_array_map_ops = {
773 	.map_meta_equal = bpf_map_meta_equal,
774 	.map_alloc_check = array_map_alloc_check,
775 	.map_alloc = array_map_alloc,
776 	.map_free = array_map_free,
777 	.map_get_next_key = array_map_get_next_key,
778 	.map_lookup_elem = percpu_array_map_lookup_elem,
779 	.map_update_elem = array_map_update_elem,
780 	.map_delete_elem = array_map_delete_elem,
781 	.map_lookup_percpu_elem = percpu_array_map_lookup_percpu_elem,
782 	.map_seq_show_elem = percpu_array_map_seq_show_elem,
783 	.map_check_btf = array_map_check_btf,
784 	.map_lookup_batch = generic_map_lookup_batch,
785 	.map_update_batch = generic_map_update_batch,
786 	.map_set_for_each_callback_args = map_set_for_each_callback_args,
787 	.map_for_each_callback = bpf_for_each_array_elem,
788 	.map_mem_usage = array_map_mem_usage,
789 	.map_btf_id = &array_map_btf_ids[0],
790 	.iter_seq_info = &iter_seq_info,
791 };
792 
fd_array_map_alloc_check(union bpf_attr * attr)793 static int fd_array_map_alloc_check(union bpf_attr *attr)
794 {
795 	/* only file descriptors can be stored in this type of map */
796 	if (attr->value_size != sizeof(u32))
797 		return -EINVAL;
798 	/* Program read-only/write-only not supported for special maps yet. */
799 	if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG))
800 		return -EINVAL;
801 	return array_map_alloc_check(attr);
802 }
803 
fd_array_map_free(struct bpf_map * map)804 static void fd_array_map_free(struct bpf_map *map)
805 {
806 	struct bpf_array *array = container_of(map, struct bpf_array, map);
807 	int i;
808 
809 	/* make sure it's empty */
810 	for (i = 0; i < array->map.max_entries; i++)
811 		BUG_ON(array->ptrs[i] != NULL);
812 
813 	bpf_map_area_free(array);
814 }
815 
fd_array_map_lookup_elem(struct bpf_map * map,void * key)816 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
817 {
818 	return ERR_PTR(-EOPNOTSUPP);
819 }
820 
821 /* only called from syscall */
bpf_fd_array_map_lookup_elem(struct bpf_map * map,void * key,u32 * value)822 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
823 {
824 	void **elem, *ptr;
825 	int ret =  0;
826 
827 	if (!map->ops->map_fd_sys_lookup_elem)
828 		return -ENOTSUPP;
829 
830 	rcu_read_lock();
831 	elem = array_map_lookup_elem(map, key);
832 	if (elem && (ptr = READ_ONCE(*elem)))
833 		*value = map->ops->map_fd_sys_lookup_elem(ptr);
834 	else
835 		ret = -ENOENT;
836 	rcu_read_unlock();
837 
838 	return ret;
839 }
840 
841 /* only called from syscall */
bpf_fd_array_map_update_elem(struct bpf_map * map,struct file * map_file,void * key,void * value,u64 map_flags)842 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
843 				 void *key, void *value, u64 map_flags)
844 {
845 	struct bpf_array *array = container_of(map, struct bpf_array, map);
846 	void *new_ptr, *old_ptr;
847 	u32 index = *(u32 *)key, ufd;
848 
849 	if (map_flags != BPF_ANY)
850 		return -EINVAL;
851 
852 	if (index >= array->map.max_entries)
853 		return -E2BIG;
854 
855 	ufd = *(u32 *)value;
856 	new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
857 	if (IS_ERR(new_ptr))
858 		return PTR_ERR(new_ptr);
859 
860 	if (map->ops->map_poke_run) {
861 		mutex_lock(&array->aux->poke_mutex);
862 		old_ptr = xchg(array->ptrs + index, new_ptr);
863 		map->ops->map_poke_run(map, index, old_ptr, new_ptr);
864 		mutex_unlock(&array->aux->poke_mutex);
865 	} else {
866 		old_ptr = xchg(array->ptrs + index, new_ptr);
867 	}
868 
869 	if (old_ptr)
870 		map->ops->map_fd_put_ptr(map, old_ptr, true);
871 	return 0;
872 }
873 
__fd_array_map_delete_elem(struct bpf_map * map,void * key,bool need_defer)874 static long __fd_array_map_delete_elem(struct bpf_map *map, void *key, bool need_defer)
875 {
876 	struct bpf_array *array = container_of(map, struct bpf_array, map);
877 	void *old_ptr;
878 	u32 index = *(u32 *)key;
879 
880 	if (index >= array->map.max_entries)
881 		return -E2BIG;
882 
883 	if (map->ops->map_poke_run) {
884 		mutex_lock(&array->aux->poke_mutex);
885 		old_ptr = xchg(array->ptrs + index, NULL);
886 		map->ops->map_poke_run(map, index, old_ptr, NULL);
887 		mutex_unlock(&array->aux->poke_mutex);
888 	} else {
889 		old_ptr = xchg(array->ptrs + index, NULL);
890 	}
891 
892 	if (old_ptr) {
893 		map->ops->map_fd_put_ptr(map, old_ptr, need_defer);
894 		return 0;
895 	} else {
896 		return -ENOENT;
897 	}
898 }
899 
fd_array_map_delete_elem(struct bpf_map * map,void * key)900 static long fd_array_map_delete_elem(struct bpf_map *map, void *key)
901 {
902 	return __fd_array_map_delete_elem(map, key, true);
903 }
904 
prog_fd_array_get_ptr(struct bpf_map * map,struct file * map_file,int fd)905 static void *prog_fd_array_get_ptr(struct bpf_map *map,
906 				   struct file *map_file, int fd)
907 {
908 	struct bpf_prog *prog = bpf_prog_get(fd);
909 
910 	if (IS_ERR(prog))
911 		return prog;
912 
913 	if (!bpf_prog_map_compatible(map, prog)) {
914 		bpf_prog_put(prog);
915 		return ERR_PTR(-EINVAL);
916 	}
917 
918 	return prog;
919 }
920 
prog_fd_array_put_ptr(struct bpf_map * map,void * ptr,bool need_defer)921 static void prog_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
922 {
923 	/* bpf_prog is freed after one RCU or tasks trace grace period */
924 	bpf_prog_put(ptr);
925 }
926 
prog_fd_array_sys_lookup_elem(void * ptr)927 static u32 prog_fd_array_sys_lookup_elem(void *ptr)
928 {
929 	return ((struct bpf_prog *)ptr)->aux->id;
930 }
931 
932 /* decrement refcnt of all bpf_progs that are stored in this map */
bpf_fd_array_map_clear(struct bpf_map * map,bool need_defer)933 static void bpf_fd_array_map_clear(struct bpf_map *map, bool need_defer)
934 {
935 	struct bpf_array *array = container_of(map, struct bpf_array, map);
936 	int i;
937 
938 	for (i = 0; i < array->map.max_entries; i++)
939 		__fd_array_map_delete_elem(map, &i, need_defer);
940 }
941 
prog_array_map_seq_show_elem(struct bpf_map * map,void * key,struct seq_file * m)942 static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
943 					 struct seq_file *m)
944 {
945 	void **elem, *ptr;
946 	u32 prog_id;
947 
948 	rcu_read_lock();
949 
950 	elem = array_map_lookup_elem(map, key);
951 	if (elem) {
952 		ptr = READ_ONCE(*elem);
953 		if (ptr) {
954 			seq_printf(m, "%u: ", *(u32 *)key);
955 			prog_id = prog_fd_array_sys_lookup_elem(ptr);
956 			btf_type_seq_show(map->btf, map->btf_value_type_id,
957 					  &prog_id, m);
958 			seq_puts(m, "\n");
959 		}
960 	}
961 
962 	rcu_read_unlock();
963 }
964 
965 struct prog_poke_elem {
966 	struct list_head list;
967 	struct bpf_prog_aux *aux;
968 };
969 
prog_array_map_poke_track(struct bpf_map * map,struct bpf_prog_aux * prog_aux)970 static int prog_array_map_poke_track(struct bpf_map *map,
971 				     struct bpf_prog_aux *prog_aux)
972 {
973 	struct prog_poke_elem *elem;
974 	struct bpf_array_aux *aux;
975 	int ret = 0;
976 
977 	aux = container_of(map, struct bpf_array, map)->aux;
978 	mutex_lock(&aux->poke_mutex);
979 	list_for_each_entry(elem, &aux->poke_progs, list) {
980 		if (elem->aux == prog_aux)
981 			goto out;
982 	}
983 
984 	elem = kmalloc(sizeof(*elem), GFP_KERNEL);
985 	if (!elem) {
986 		ret = -ENOMEM;
987 		goto out;
988 	}
989 
990 	INIT_LIST_HEAD(&elem->list);
991 	/* We must track the program's aux info at this point in time
992 	 * since the program pointer itself may not be stable yet, see
993 	 * also comment in prog_array_map_poke_run().
994 	 */
995 	elem->aux = prog_aux;
996 
997 	list_add_tail(&elem->list, &aux->poke_progs);
998 out:
999 	mutex_unlock(&aux->poke_mutex);
1000 	return ret;
1001 }
1002 
prog_array_map_poke_untrack(struct bpf_map * map,struct bpf_prog_aux * prog_aux)1003 static void prog_array_map_poke_untrack(struct bpf_map *map,
1004 					struct bpf_prog_aux *prog_aux)
1005 {
1006 	struct prog_poke_elem *elem, *tmp;
1007 	struct bpf_array_aux *aux;
1008 
1009 	aux = container_of(map, struct bpf_array, map)->aux;
1010 	mutex_lock(&aux->poke_mutex);
1011 	list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1012 		if (elem->aux == prog_aux) {
1013 			list_del_init(&elem->list);
1014 			kfree(elem);
1015 			break;
1016 		}
1017 	}
1018 	mutex_unlock(&aux->poke_mutex);
1019 }
1020 
bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor * poke,struct bpf_prog * new,struct bpf_prog * old)1021 void __weak bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor *poke,
1022 				      struct bpf_prog *new, struct bpf_prog *old)
1023 {
1024 	WARN_ON_ONCE(1);
1025 }
1026 
prog_array_map_poke_run(struct bpf_map * map,u32 key,struct bpf_prog * old,struct bpf_prog * new)1027 static void prog_array_map_poke_run(struct bpf_map *map, u32 key,
1028 				    struct bpf_prog *old,
1029 				    struct bpf_prog *new)
1030 {
1031 	struct prog_poke_elem *elem;
1032 	struct bpf_array_aux *aux;
1033 
1034 	aux = container_of(map, struct bpf_array, map)->aux;
1035 	WARN_ON_ONCE(!mutex_is_locked(&aux->poke_mutex));
1036 
1037 	list_for_each_entry(elem, &aux->poke_progs, list) {
1038 		struct bpf_jit_poke_descriptor *poke;
1039 		int i;
1040 
1041 		for (i = 0; i < elem->aux->size_poke_tab; i++) {
1042 			poke = &elem->aux->poke_tab[i];
1043 
1044 			/* Few things to be aware of:
1045 			 *
1046 			 * 1) We can only ever access aux in this context, but
1047 			 *    not aux->prog since it might not be stable yet and
1048 			 *    there could be danger of use after free otherwise.
1049 			 * 2) Initially when we start tracking aux, the program
1050 			 *    is not JITed yet and also does not have a kallsyms
1051 			 *    entry. We skip these as poke->tailcall_target_stable
1052 			 *    is not active yet. The JIT will do the final fixup
1053 			 *    before setting it stable. The various
1054 			 *    poke->tailcall_target_stable are successively
1055 			 *    activated, so tail call updates can arrive from here
1056 			 *    while JIT is still finishing its final fixup for
1057 			 *    non-activated poke entries.
1058 			 * 3) Also programs reaching refcount of zero while patching
1059 			 *    is in progress is okay since we're protected under
1060 			 *    poke_mutex and untrack the programs before the JIT
1061 			 *    buffer is freed.
1062 			 */
1063 			if (!READ_ONCE(poke->tailcall_target_stable))
1064 				continue;
1065 			if (poke->reason != BPF_POKE_REASON_TAIL_CALL)
1066 				continue;
1067 			if (poke->tail_call.map != map ||
1068 			    poke->tail_call.key != key)
1069 				continue;
1070 
1071 			bpf_arch_poke_desc_update(poke, new, old);
1072 		}
1073 	}
1074 }
1075 
prog_array_map_clear_deferred(struct work_struct * work)1076 static void prog_array_map_clear_deferred(struct work_struct *work)
1077 {
1078 	struct bpf_map *map = container_of(work, struct bpf_array_aux,
1079 					   work)->map;
1080 	bpf_fd_array_map_clear(map, true);
1081 	bpf_map_put(map);
1082 }
1083 
prog_array_map_clear(struct bpf_map * map)1084 static void prog_array_map_clear(struct bpf_map *map)
1085 {
1086 	struct bpf_array_aux *aux = container_of(map, struct bpf_array,
1087 						 map)->aux;
1088 	bpf_map_inc(map);
1089 	schedule_work(&aux->work);
1090 }
1091 
prog_array_map_alloc(union bpf_attr * attr)1092 static struct bpf_map *prog_array_map_alloc(union bpf_attr *attr)
1093 {
1094 	struct bpf_array_aux *aux;
1095 	struct bpf_map *map;
1096 
1097 	aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT);
1098 	if (!aux)
1099 		return ERR_PTR(-ENOMEM);
1100 
1101 	INIT_WORK(&aux->work, prog_array_map_clear_deferred);
1102 	INIT_LIST_HEAD(&aux->poke_progs);
1103 	mutex_init(&aux->poke_mutex);
1104 
1105 	map = array_map_alloc(attr);
1106 	if (IS_ERR(map)) {
1107 		kfree(aux);
1108 		return map;
1109 	}
1110 
1111 	container_of(map, struct bpf_array, map)->aux = aux;
1112 	aux->map = map;
1113 
1114 	return map;
1115 }
1116 
prog_array_map_free(struct bpf_map * map)1117 static void prog_array_map_free(struct bpf_map *map)
1118 {
1119 	struct prog_poke_elem *elem, *tmp;
1120 	struct bpf_array_aux *aux;
1121 
1122 	aux = container_of(map, struct bpf_array, map)->aux;
1123 	list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1124 		list_del_init(&elem->list);
1125 		kfree(elem);
1126 	}
1127 	kfree(aux);
1128 	fd_array_map_free(map);
1129 }
1130 
1131 /* prog_array->aux->{type,jited} is a runtime binding.
1132  * Doing static check alone in the verifier is not enough.
1133  * Thus, prog_array_map cannot be used as an inner_map
1134  * and map_meta_equal is not implemented.
1135  */
1136 const struct bpf_map_ops prog_array_map_ops = {
1137 	.map_alloc_check = fd_array_map_alloc_check,
1138 	.map_alloc = prog_array_map_alloc,
1139 	.map_free = prog_array_map_free,
1140 	.map_poke_track = prog_array_map_poke_track,
1141 	.map_poke_untrack = prog_array_map_poke_untrack,
1142 	.map_poke_run = prog_array_map_poke_run,
1143 	.map_get_next_key = array_map_get_next_key,
1144 	.map_lookup_elem = fd_array_map_lookup_elem,
1145 	.map_delete_elem = fd_array_map_delete_elem,
1146 	.map_fd_get_ptr = prog_fd_array_get_ptr,
1147 	.map_fd_put_ptr = prog_fd_array_put_ptr,
1148 	.map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
1149 	.map_release_uref = prog_array_map_clear,
1150 	.map_seq_show_elem = prog_array_map_seq_show_elem,
1151 	.map_mem_usage = array_map_mem_usage,
1152 	.map_btf_id = &array_map_btf_ids[0],
1153 };
1154 
bpf_event_entry_gen(struct file * perf_file,struct file * map_file)1155 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
1156 						   struct file *map_file)
1157 {
1158 	struct bpf_event_entry *ee;
1159 
1160 	ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
1161 	if (ee) {
1162 		ee->event = perf_file->private_data;
1163 		ee->perf_file = perf_file;
1164 		ee->map_file = map_file;
1165 	}
1166 
1167 	return ee;
1168 }
1169 
__bpf_event_entry_free(struct rcu_head * rcu)1170 static void __bpf_event_entry_free(struct rcu_head *rcu)
1171 {
1172 	struct bpf_event_entry *ee;
1173 
1174 	ee = container_of(rcu, struct bpf_event_entry, rcu);
1175 	fput(ee->perf_file);
1176 	kfree(ee);
1177 }
1178 
bpf_event_entry_free_rcu(struct bpf_event_entry * ee)1179 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
1180 {
1181 	call_rcu(&ee->rcu, __bpf_event_entry_free);
1182 }
1183 
perf_event_fd_array_get_ptr(struct bpf_map * map,struct file * map_file,int fd)1184 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
1185 					 struct file *map_file, int fd)
1186 {
1187 	struct bpf_event_entry *ee;
1188 	struct perf_event *event;
1189 	struct file *perf_file;
1190 	u64 value;
1191 
1192 	perf_file = perf_event_get(fd);
1193 	if (IS_ERR(perf_file))
1194 		return perf_file;
1195 
1196 	ee = ERR_PTR(-EOPNOTSUPP);
1197 	event = perf_file->private_data;
1198 	if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
1199 		goto err_out;
1200 
1201 	ee = bpf_event_entry_gen(perf_file, map_file);
1202 	if (ee)
1203 		return ee;
1204 	ee = ERR_PTR(-ENOMEM);
1205 err_out:
1206 	fput(perf_file);
1207 	return ee;
1208 }
1209 
perf_event_fd_array_put_ptr(struct bpf_map * map,void * ptr,bool need_defer)1210 static void perf_event_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1211 {
1212 	/* bpf_perf_event is freed after one RCU grace period */
1213 	bpf_event_entry_free_rcu(ptr);
1214 }
1215 
perf_event_fd_array_release(struct bpf_map * map,struct file * map_file)1216 static void perf_event_fd_array_release(struct bpf_map *map,
1217 					struct file *map_file)
1218 {
1219 	struct bpf_array *array = container_of(map, struct bpf_array, map);
1220 	struct bpf_event_entry *ee;
1221 	int i;
1222 
1223 	if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1224 		return;
1225 
1226 	rcu_read_lock();
1227 	for (i = 0; i < array->map.max_entries; i++) {
1228 		ee = READ_ONCE(array->ptrs[i]);
1229 		if (ee && ee->map_file == map_file)
1230 			__fd_array_map_delete_elem(map, &i, true);
1231 	}
1232 	rcu_read_unlock();
1233 }
1234 
perf_event_fd_array_map_free(struct bpf_map * map)1235 static void perf_event_fd_array_map_free(struct bpf_map *map)
1236 {
1237 	if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1238 		bpf_fd_array_map_clear(map, false);
1239 	fd_array_map_free(map);
1240 }
1241 
1242 const struct bpf_map_ops perf_event_array_map_ops = {
1243 	.map_meta_equal = bpf_map_meta_equal,
1244 	.map_alloc_check = fd_array_map_alloc_check,
1245 	.map_alloc = array_map_alloc,
1246 	.map_free = perf_event_fd_array_map_free,
1247 	.map_get_next_key = array_map_get_next_key,
1248 	.map_lookup_elem = fd_array_map_lookup_elem,
1249 	.map_delete_elem = fd_array_map_delete_elem,
1250 	.map_fd_get_ptr = perf_event_fd_array_get_ptr,
1251 	.map_fd_put_ptr = perf_event_fd_array_put_ptr,
1252 	.map_release = perf_event_fd_array_release,
1253 	.map_check_btf = map_check_no_btf,
1254 	.map_mem_usage = array_map_mem_usage,
1255 	.map_btf_id = &array_map_btf_ids[0],
1256 };
1257 
1258 #ifdef CONFIG_CGROUPS
cgroup_fd_array_get_ptr(struct bpf_map * map,struct file * map_file,int fd)1259 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
1260 				     struct file *map_file /* not used */,
1261 				     int fd)
1262 {
1263 	return cgroup_get_from_fd(fd);
1264 }
1265 
cgroup_fd_array_put_ptr(struct bpf_map * map,void * ptr,bool need_defer)1266 static void cgroup_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1267 {
1268 	/* cgroup_put free cgrp after a rcu grace period */
1269 	cgroup_put(ptr);
1270 }
1271 
cgroup_fd_array_free(struct bpf_map * map)1272 static void cgroup_fd_array_free(struct bpf_map *map)
1273 {
1274 	bpf_fd_array_map_clear(map, false);
1275 	fd_array_map_free(map);
1276 }
1277 
1278 const struct bpf_map_ops cgroup_array_map_ops = {
1279 	.map_meta_equal = bpf_map_meta_equal,
1280 	.map_alloc_check = fd_array_map_alloc_check,
1281 	.map_alloc = array_map_alloc,
1282 	.map_free = cgroup_fd_array_free,
1283 	.map_get_next_key = array_map_get_next_key,
1284 	.map_lookup_elem = fd_array_map_lookup_elem,
1285 	.map_delete_elem = fd_array_map_delete_elem,
1286 	.map_fd_get_ptr = cgroup_fd_array_get_ptr,
1287 	.map_fd_put_ptr = cgroup_fd_array_put_ptr,
1288 	.map_check_btf = map_check_no_btf,
1289 	.map_mem_usage = array_map_mem_usage,
1290 	.map_btf_id = &array_map_btf_ids[0],
1291 };
1292 #endif
1293 
array_of_map_alloc(union bpf_attr * attr)1294 static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
1295 {
1296 	struct bpf_map *map, *inner_map_meta;
1297 
1298 	inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
1299 	if (IS_ERR(inner_map_meta))
1300 		return inner_map_meta;
1301 
1302 	map = array_map_alloc(attr);
1303 	if (IS_ERR(map)) {
1304 		bpf_map_meta_free(inner_map_meta);
1305 		return map;
1306 	}
1307 
1308 	map->inner_map_meta = inner_map_meta;
1309 
1310 	return map;
1311 }
1312 
array_of_map_free(struct bpf_map * map)1313 static void array_of_map_free(struct bpf_map *map)
1314 {
1315 	/* map->inner_map_meta is only accessed by syscall which
1316 	 * is protected by fdget/fdput.
1317 	 */
1318 	bpf_map_meta_free(map->inner_map_meta);
1319 	bpf_fd_array_map_clear(map, false);
1320 	fd_array_map_free(map);
1321 }
1322 
array_of_map_lookup_elem(struct bpf_map * map,void * key)1323 static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
1324 {
1325 	struct bpf_map **inner_map = array_map_lookup_elem(map, key);
1326 
1327 	if (!inner_map)
1328 		return NULL;
1329 
1330 	return READ_ONCE(*inner_map);
1331 }
1332 
array_of_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)1333 static int array_of_map_gen_lookup(struct bpf_map *map,
1334 				   struct bpf_insn *insn_buf)
1335 {
1336 	struct bpf_array *array = container_of(map, struct bpf_array, map);
1337 	u32 elem_size = array->elem_size;
1338 	struct bpf_insn *insn = insn_buf;
1339 	const int ret = BPF_REG_0;
1340 	const int map_ptr = BPF_REG_1;
1341 	const int index = BPF_REG_2;
1342 
1343 	*insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
1344 	*insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
1345 	if (!map->bypass_spec_v1) {
1346 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
1347 		*insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
1348 	} else {
1349 		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
1350 	}
1351 	if (is_power_of_2(elem_size))
1352 		*insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
1353 	else
1354 		*insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
1355 	*insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
1356 	*insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
1357 	*insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
1358 	*insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
1359 	*insn++ = BPF_MOV64_IMM(ret, 0);
1360 
1361 	return insn - insn_buf;
1362 }
1363 
1364 const struct bpf_map_ops array_of_maps_map_ops = {
1365 	.map_alloc_check = fd_array_map_alloc_check,
1366 	.map_alloc = array_of_map_alloc,
1367 	.map_free = array_of_map_free,
1368 	.map_get_next_key = array_map_get_next_key,
1369 	.map_lookup_elem = array_of_map_lookup_elem,
1370 	.map_delete_elem = fd_array_map_delete_elem,
1371 	.map_fd_get_ptr = bpf_map_fd_get_ptr,
1372 	.map_fd_put_ptr = bpf_map_fd_put_ptr,
1373 	.map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
1374 	.map_gen_lookup = array_of_map_gen_lookup,
1375 	.map_lookup_batch = generic_map_lookup_batch,
1376 	.map_update_batch = generic_map_update_batch,
1377 	.map_check_btf = map_check_no_btf,
1378 	.map_mem_usage = array_map_mem_usage,
1379 	.map_btf_id = &array_map_btf_ids[0],
1380 };
1381