arraymap.c

   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#include <crypto/sha2.h>
  16
  17#include "map_in_map.h"
  18
  19#define ARRAY_CREATE_FLAG_MASK \
  20	(BPF_F_NUMA_NODE | BPF_F_MMAPABLE | BPF_F_ACCESS_MASK | \
  21	 BPF_F_PRESERVE_ELEMS | BPF_F_INNER_MAP)
  22
  23static void bpf_array_free_percpu(struct bpf_array *array)
  24{
  25	int i;
  26
  27	for (i = 0; i < array->map.max_entries; i++) {
  28		free_percpu(array->pptrs[i]);
  29		cond_resched();
  30	}
  31}
  32
  33static int bpf_array_alloc_percpu(struct bpf_array *array)
  34{
  35	void __percpu *ptr;
  36	int i;
  37
  38	for (i = 0; i < array->map.max_entries; i++) {
  39		ptr = bpf_map_alloc_percpu(&array->map, array->elem_size, 8,
  40					   GFP_USER | __GFP_NOWARN);
  41		if (!ptr) {
  42			bpf_array_free_percpu(array);
  43			return -ENOMEM;
  44		}
  45		array->pptrs[i] = ptr;
  46		cond_resched();
  47	}
  48
  49	return 0;
  50}
  51
  52/* Called from syscall */
  53int array_map_alloc_check(union bpf_attr *attr)
  54{
  55	bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
  56	int numa_node = bpf_map_attr_numa_node(attr);
  57
  58	/* check sanity of attributes */
  59	if (attr->max_entries == 0 || attr->key_size != 4 ||
  60	    attr->value_size == 0 ||
  61	    attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
  62	    !bpf_map_flags_access_ok(attr->map_flags) ||
  63	    (percpu && numa_node != NUMA_NO_NODE))
  64		return -EINVAL;
  65
  66	if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
  67	    attr->map_flags & (BPF_F_MMAPABLE | BPF_F_INNER_MAP))
  68		return -EINVAL;
  69
  70	if (attr->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
  71	    attr->map_flags & BPF_F_PRESERVE_ELEMS)
  72		return -EINVAL;
  73
  74	/* avoid overflow on round_up(map->value_size) */
  75	if (attr->value_size > INT_MAX)
  76		return -E2BIG;
  77	/* percpu map value size is bound by PCPU_MIN_UNIT_SIZE */
  78	if (percpu && round_up(attr->value_size, 8) > PCPU_MIN_UNIT_SIZE)
  79		return -E2BIG;
  80
  81	return 0;
  82}
  83
  84static struct bpf_map *array_map_alloc(union bpf_attr *attr)
  85{
  86	bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
  87	int numa_node = bpf_map_attr_numa_node(attr);
  88	u32 elem_size, index_mask, max_entries;
  89	bool bypass_spec_v1 = bpf_bypass_spec_v1(NULL);
  90	u64 array_size, mask64;
  91	struct bpf_array *array;
  92
  93	elem_size = round_up(attr->value_size, 8);
  94
  95	max_entries = attr->max_entries;
  96
  97	/* On 32 bit archs roundup_pow_of_two() with max_entries that has
  98	 * upper most bit set in u32 space is undefined behavior due to
  99	 * resulting 1U << 32, so do it manually here in u64 space.
 100	 */
 101	mask64 = fls_long(max_entries - 1);
 102	mask64 = 1ULL << mask64;
 103	mask64 -= 1;
 104
 105	index_mask = mask64;
 106	if (!bypass_spec_v1) {
 107		/* round up array size to nearest power of 2,
 108		 * since cpu will speculate within index_mask limits
 109		 */
 110		max_entries = index_mask + 1;
 111		/* Check for overflows. */
 112		if (max_entries < attr->max_entries)
 113			return ERR_PTR(-E2BIG);
 114	}
 115
 116	array_size = sizeof(*array);
 117	if (percpu) {
 118		array_size += (u64) max_entries * sizeof(void *);
 119	} else {
 120		/* rely on vmalloc() to return page-aligned memory and
 121		 * ensure array->value is exactly page-aligned
 122		 */
 123		if (attr->map_flags & BPF_F_MMAPABLE) {
 124			array_size = PAGE_ALIGN(array_size);
 125			array_size += PAGE_ALIGN((u64) max_entries * elem_size);
 126		} else {
 127			array_size += (u64) max_entries * elem_size;
 128		}
 129	}
 130
 131	/* allocate all map elements and zero-initialize them */
 132	if (attr->map_flags & BPF_F_MMAPABLE) {
 133		void *data;
 134
 135		/* kmalloc'ed memory can't be mmap'ed, use explicit vmalloc */
 136		data = bpf_map_area_mmapable_alloc(array_size, numa_node);
 137		if (!data)
 138			return ERR_PTR(-ENOMEM);
 139		array = data + PAGE_ALIGN(sizeof(struct bpf_array))
 140			- offsetof(struct bpf_array, value);
 141	} else {
 142		array = bpf_map_area_alloc(array_size, numa_node);
 143	}
 144	if (!array)
 145		return ERR_PTR(-ENOMEM);
 146	array->index_mask = index_mask;
 147	array->map.bypass_spec_v1 = bypass_spec_v1;
 148
 149	/* copy mandatory map attributes */
 150	bpf_map_init_from_attr(&array->map, attr);
 151	array->elem_size = elem_size;
 152
 153	if (percpu && bpf_array_alloc_percpu(array)) {
 154		bpf_map_area_free(array);
 155		return ERR_PTR(-ENOMEM);
 156	}
 157
 158	return &array->map;
 159}
 160
 161static void *array_map_elem_ptr(struct bpf_array* array, u32 index)
 162{
 163	return array->value + (u64)array->elem_size * index;
 164}
 165
 166/* Called from syscall or from eBPF program */
 167static void *array_map_lookup_elem(struct bpf_map *map, void *key)
 168{
 169	struct bpf_array *array = container_of(map, struct bpf_array, map);
 170	u32 index = *(u32 *)key;
 171
 172	if (unlikely(index >= array->map.max_entries))
 173		return NULL;
 174
 175	return array->value + (u64)array->elem_size * (index & array->index_mask);
 176}
 177
 178static int array_map_get_hash(struct bpf_map *map, u32 hash_buf_size,
 179			       void *hash_buf)
 180{
 181	struct bpf_array *array = container_of(map, struct bpf_array, map);
 182
 183	sha256(array->value, (u64)array->elem_size * array->map.max_entries,
 184	       hash_buf);
 185	memcpy(array->map.sha, hash_buf, sizeof(array->map.sha));
 186	return 0;
 187}
 188
 189static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm,
 190				       u32 off)
 191{
 192	struct bpf_array *array = container_of(map, struct bpf_array, map);
 193
 194	if (map->max_entries != 1)
 195		return -ENOTSUPP;
 196	if (off >= map->value_size)
 197		return -EINVAL;
 198
 199	*imm = (unsigned long)array->value;
 200	return 0;
 201}
 202
 203static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm,
 204				       u32 *off)
 205{
 206	struct bpf_array *array = container_of(map, struct bpf_array, map);
 207	u64 base = (unsigned long)array->value;
 208	u64 range = array->elem_size;
 209
 210	if (map->max_entries != 1)
 211		return -ENOTSUPP;
 212	if (imm < base || imm >= base + range)
 213		return -ENOENT;
 214
 215	*off = imm - base;
 216	return 0;
 217}
 218
 219/* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
 220static int array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
 221{
 222	struct bpf_array *array = container_of(map, struct bpf_array, map);
 223	struct bpf_insn *insn = insn_buf;
 224	u32 elem_size = array->elem_size;
 225	const int ret = BPF_REG_0;
 226	const int map_ptr = BPF_REG_1;
 227	const int index = BPF_REG_2;
 228
 229	if (map->map_flags & BPF_F_INNER_MAP)
 230		return -EOPNOTSUPP;
 231
 232	*insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
 233	*insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
 234	if (!map->bypass_spec_v1) {
 235		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
 236		*insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
 237	} else {
 238		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
 239	}
 240
 241	if (is_power_of_2(elem_size)) {
 242		*insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
 243	} else {
 244		*insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
 245	}
 246	*insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
 247	*insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
 248	*insn++ = BPF_MOV64_IMM(ret, 0);
 249	return insn - insn_buf;
 250}
 251
 252/* Called from eBPF program */
 253static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
 254{
 255	struct bpf_array *array = container_of(map, struct bpf_array, map);
 256	u32 index = *(u32 *)key;
 257
 258	if (unlikely(index >= array->map.max_entries))
 259		return NULL;
 260
 261	return this_cpu_ptr(array->pptrs[index & array->index_mask]);
 262}
 263
 264/* emit BPF instructions equivalent to C code of percpu_array_map_lookup_elem() */
 265static int percpu_array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
 266{
 267	struct bpf_array *array = container_of(map, struct bpf_array, map);
 268	struct bpf_insn *insn = insn_buf;
 269
 270	if (!bpf_jit_supports_percpu_insn())
 271		return -EOPNOTSUPP;
 272
 273	if (map->map_flags & BPF_F_INNER_MAP)
 274		return -EOPNOTSUPP;
 275
 276	BUILD_BUG_ON(offsetof(struct bpf_array, map) != 0);
 277	*insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, offsetof(struct bpf_array, pptrs));
 278
 279	*insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_2, 0);
 280	if (!map->bypass_spec_v1) {
 281		*insn++ = BPF_JMP_IMM(BPF_JGE, BPF_REG_0, map->max_entries, 6);
 282		*insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_0, array->index_mask);
 283	} else {
 284		*insn++ = BPF_JMP_IMM(BPF_JGE, BPF_REG_0, map->max_entries, 5);
 285	}
 286
 287	*insn++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3);
 288	*insn++ = BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1);
 289	*insn++ = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_0, 0);
 290	*insn++ = BPF_MOV64_PERCPU_REG(BPF_REG_0, BPF_REG_0);
 291	*insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
 292	*insn++ = BPF_MOV64_IMM(BPF_REG_0, 0);
 293	return insn - insn_buf;
 294}
 295
 296static void *percpu_array_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
 297{
 298	struct bpf_array *array = container_of(map, struct bpf_array, map);
 299	u32 index = *(u32 *)key;
 300
 301	if (cpu >= nr_cpu_ids)
 302		return NULL;
 303
 304	if (unlikely(index >= array->map.max_entries))
 305		return NULL;
 306
 307	return per_cpu_ptr(array->pptrs[index & array->index_mask], cpu);
 308}
 309
 310int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
 311{
 312	struct bpf_array *array = container_of(map, struct bpf_array, map);
 313	u32 index = *(u32 *)key;
 314	void __percpu *pptr;
 315	int cpu, off = 0;
 316	u32 size;
 317
 318	if (unlikely(index >= array->map.max_entries))
 319		return -ENOENT;
 320
 321	/* per_cpu areas are zero-filled and bpf programs can only
 322	 * access 'value_size' of them, so copying rounded areas
 323	 * will not leak any kernel data
 324	 */
 325	size = array->elem_size;
 326	rcu_read_lock();
 327	pptr = array->pptrs[index & array->index_mask];
 328	for_each_possible_cpu(cpu) {
 329		copy_map_value_long(map, value + off, per_cpu_ptr(pptr, cpu));
 330		check_and_init_map_value(map, value + off);
 331		off += size;
 332	}
 333	rcu_read_unlock();
 334	return 0;
 335}
 336
 337/* Called from syscall */
 338int bpf_array_get_next_key(struct bpf_map *map, void *key, void *next_key)
 339{
 340	u32 index = key ? *(u32 *)key : U32_MAX;
 341	u32 *next = (u32 *)next_key;
 342
 343	if (index >= map->max_entries) {
 344		*next = 0;
 345		return 0;
 346	}
 347
 348	if (index == map->max_entries - 1)
 349		return -ENOENT;
 350
 351	*next = index + 1;
 352	return 0;
 353}
 354
 355/* Called from syscall or from eBPF program */
 356static long array_map_update_elem(struct bpf_map *map, void *key, void *value,
 357				  u64 map_flags)
 358{
 359	struct bpf_array *array = container_of(map, struct bpf_array, map);
 360	u32 index = *(u32 *)key;
 361	char *val;
 362
 363	if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
 364		/* unknown flags */
 365		return -EINVAL;
 366
 367	if (unlikely(index >= array->map.max_entries))
 368		/* all elements were pre-allocated, cannot insert a new one */
 369		return -E2BIG;
 370
 371	if (unlikely(map_flags & BPF_NOEXIST))
 372		/* all elements already exist */
 373		return -EEXIST;
 374
 375	if (unlikely((map_flags & BPF_F_LOCK) &&
 376		     !btf_record_has_field(map->record, BPF_SPIN_LOCK)))
 377		return -EINVAL;
 378
 379	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 380		val = this_cpu_ptr(array->pptrs[index & array->index_mask]);
 381		copy_map_value(map, val, value);
 382		bpf_obj_free_fields(array->map.record, val);
 383	} else {
 384		val = array->value +
 385			(u64)array->elem_size * (index & array->index_mask);
 386		if (map_flags & BPF_F_LOCK)
 387			copy_map_value_locked(map, val, value, false);
 388		else
 389			copy_map_value(map, val, value);
 390		bpf_obj_free_fields(array->map.record, val);
 391	}
 392	return 0;
 393}
 394
 395int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
 396			    u64 map_flags)
 397{
 398	struct bpf_array *array = container_of(map, struct bpf_array, map);
 399	u32 index = *(u32 *)key;
 400	void __percpu *pptr;
 401	int cpu, off = 0;
 402	u32 size;
 403
 404	if (unlikely(map_flags > BPF_EXIST))
 405		/* unknown flags */
 406		return -EINVAL;
 407
 408	if (unlikely(index >= array->map.max_entries))
 409		/* all elements were pre-allocated, cannot insert a new one */
 410		return -E2BIG;
 411
 412	if (unlikely(map_flags == BPF_NOEXIST))
 413		/* all elements already exist */
 414		return -EEXIST;
 415
 416	/* the user space will provide round_up(value_size, 8) bytes that
 417	 * will be copied into per-cpu area. bpf programs can only access
 418	 * value_size of it. During lookup the same extra bytes will be
 419	 * returned or zeros which were zero-filled by percpu_alloc,
 420	 * so no kernel data leaks possible
 421	 */
 422	size = array->elem_size;
 423	rcu_read_lock();
 424	pptr = array->pptrs[index & array->index_mask];
 425	for_each_possible_cpu(cpu) {
 426		copy_map_value_long(map, per_cpu_ptr(pptr, cpu), value + off);
 427		bpf_obj_free_fields(array->map.record, per_cpu_ptr(pptr, cpu));
 428		off += size;
 429	}
 430	rcu_read_unlock();
 431	return 0;
 432}
 433
 434/* Called from syscall or from eBPF program */
 435static long array_map_delete_elem(struct bpf_map *map, void *key)
 436{
 437	return -EINVAL;
 438}
 439
 440static void *array_map_vmalloc_addr(struct bpf_array *array)
 441{
 442	return (void *)round_down((unsigned long)array, PAGE_SIZE);
 443}
 444
 445static void array_map_free_internal_structs(struct bpf_map *map)
 446{
 447	struct bpf_array *array = container_of(map, struct bpf_array, map);
 448	int i;
 449
 450	/* We only free internal structs on uref dropping to zero */
 451	if (!bpf_map_has_internal_structs(map))
 452		return;
 453
 454	for (i = 0; i < array->map.max_entries; i++)
 455		bpf_map_free_internal_structs(map, array_map_elem_ptr(array, i));
 456}
 457
 458/* Called when map->refcnt goes to zero, either from workqueue or from syscall */
 459static void array_map_free(struct bpf_map *map)
 460{
 461	struct bpf_array *array = container_of(map, struct bpf_array, map);
 462	int i;
 463
 464	if (!IS_ERR_OR_NULL(map->record)) {
 465		if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 466			for (i = 0; i < array->map.max_entries; i++) {
 467				void __percpu *pptr = array->pptrs[i & array->index_mask];
 468				int cpu;
 469
 470				for_each_possible_cpu(cpu) {
 471					bpf_obj_free_fields(map->record, per_cpu_ptr(pptr, cpu));
 472					cond_resched();
 473				}
 474			}
 475		} else {
 476			for (i = 0; i < array->map.max_entries; i++)
 477				bpf_obj_free_fields(map->record, array_map_elem_ptr(array, i));
 478		}
 479	}
 480
 481	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 482		bpf_array_free_percpu(array);
 483
 484	if (array->map.map_flags & BPF_F_MMAPABLE)
 485		bpf_map_area_free(array_map_vmalloc_addr(array));
 486	else
 487		bpf_map_area_free(array);
 488}
 489
 490static void array_map_seq_show_elem(struct bpf_map *map, void *key,
 491				    struct seq_file *m)
 492{
 493	void *value;
 494
 495	rcu_read_lock();
 496
 497	value = array_map_lookup_elem(map, key);
 498	if (!value) {
 499		rcu_read_unlock();
 500		return;
 501	}
 502
 503	if (map->btf_key_type_id)
 504		seq_printf(m, "%u: ", *(u32 *)key);
 505	btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
 506	seq_putc(m, '\n');
 507
 508	rcu_read_unlock();
 509}
 510
 511static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
 512					   struct seq_file *m)
 513{
 514	struct bpf_array *array = container_of(map, struct bpf_array, map);
 515	u32 index = *(u32 *)key;
 516	void __percpu *pptr;
 517	int cpu;
 518
 519	rcu_read_lock();
 520
 521	seq_printf(m, "%u: {\n", *(u32 *)key);
 522	pptr = array->pptrs[index & array->index_mask];
 523	for_each_possible_cpu(cpu) {
 524		seq_printf(m, "\tcpu%d: ", cpu);
 525		btf_type_seq_show(map->btf, map->btf_value_type_id,
 526				  per_cpu_ptr(pptr, cpu), m);
 527		seq_putc(m, '\n');
 528	}
 529	seq_puts(m, "}\n");
 530
 531	rcu_read_unlock();
 532}
 533
 534static int array_map_check_btf(const struct bpf_map *map,
 535			       const struct btf *btf,
 536			       const struct btf_type *key_type,
 537			       const struct btf_type *value_type)
 538{
 539	/* One exception for keyless BTF: .bss/.data/.rodata map */
 540	if (btf_type_is_void(key_type)) {
 541		if (map->map_type != BPF_MAP_TYPE_ARRAY ||
 542		    map->max_entries != 1)
 543			return -EINVAL;
 544
 545		if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC)
 546			return -EINVAL;
 547
 548		return 0;
 549	}
 550
 551	/*
 552	 * Bpf array can only take a u32 key. This check makes sure
 553	 * that the btf matches the attr used during map_create.
 554	 */
 555	if (!btf_type_is_i32(key_type))
 556		return -EINVAL;
 557
 558	return 0;
 559}
 560
 561static int array_map_mmap(struct bpf_map *map, struct vm_area_struct *vma)
 562{
 563	struct bpf_array *array = container_of(map, struct bpf_array, map);
 564	pgoff_t pgoff = PAGE_ALIGN(sizeof(*array)) >> PAGE_SHIFT;
 565
 566	if (!(map->map_flags & BPF_F_MMAPABLE))
 567		return -EINVAL;
 568
 569	if (vma->vm_pgoff * PAGE_SIZE + (vma->vm_end - vma->vm_start) >
 570	    PAGE_ALIGN((u64)array->map.max_entries * array->elem_size))
 571		return -EINVAL;
 572
 573	return remap_vmalloc_range(vma, array_map_vmalloc_addr(array),
 574				   vma->vm_pgoff + pgoff);
 575}
 576
 577static bool array_map_meta_equal(const struct bpf_map *meta0,
 578				 const struct bpf_map *meta1)
 579{
 580	if (!bpf_map_meta_equal(meta0, meta1))
 581		return false;
 582	return meta0->map_flags & BPF_F_INNER_MAP ? true :
 583	       meta0->max_entries == meta1->max_entries;
 584}
 585
 586struct bpf_iter_seq_array_map_info {
 587	struct bpf_map *map;
 588	void *percpu_value_buf;
 589	u32 index;
 590};
 591
 592static void *bpf_array_map_seq_start(struct seq_file *seq, loff_t *pos)
 593{
 594	struct bpf_iter_seq_array_map_info *info = seq->private;
 595	struct bpf_map *map = info->map;
 596	struct bpf_array *array;
 597	u32 index;
 598
 599	if (info->index >= map->max_entries)
 600		return NULL;
 601
 602	if (*pos == 0)
 603		++*pos;
 604	array = container_of(map, struct bpf_array, map);
 605	index = info->index & array->index_mask;
 606	if (info->percpu_value_buf)
 607		return (void *)(uintptr_t)array->pptrs[index];
 608	return array_map_elem_ptr(array, index);
 609}
 610
 611static void *bpf_array_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 612{
 613	struct bpf_iter_seq_array_map_info *info = seq->private;
 614	struct bpf_map *map = info->map;
 615	struct bpf_array *array;
 616	u32 index;
 617
 618	++*pos;
 619	++info->index;
 620	if (info->index >= map->max_entries)
 621		return NULL;
 622
 623	array = container_of(map, struct bpf_array, map);
 624	index = info->index & array->index_mask;
 625	if (info->percpu_value_buf)
 626		return (void *)(uintptr_t)array->pptrs[index];
 627	return array_map_elem_ptr(array, index);
 628}
 629
 630static int __bpf_array_map_seq_show(struct seq_file *seq, void *v)
 631{
 632	struct bpf_iter_seq_array_map_info *info = seq->private;
 633	struct bpf_iter__bpf_map_elem ctx = {};
 634	struct bpf_map *map = info->map;
 635	struct bpf_array *array = container_of(map, struct bpf_array, map);
 636	struct bpf_iter_meta meta;
 637	struct bpf_prog *prog;
 638	int off = 0, cpu = 0;
 639	void __percpu *pptr;
 640	u32 size;
 641
 642	meta.seq = seq;
 643	prog = bpf_iter_get_info(&meta, v == NULL);
 644	if (!prog)
 645		return 0;
 646
 647	ctx.meta = &meta;
 648	ctx.map = info->map;
 649	if (v) {
 650		ctx.key = &info->index;
 651
 652		if (!info->percpu_value_buf) {
 653			ctx.value = v;
 654		} else {
 655			pptr = (void __percpu *)(uintptr_t)v;
 656			size = array->elem_size;
 657			for_each_possible_cpu(cpu) {
 658				copy_map_value_long(map, info->percpu_value_buf + off,
 659						    per_cpu_ptr(pptr, cpu));
 660				check_and_init_map_value(map, info->percpu_value_buf + off);
 661				off += size;
 662			}
 663			ctx.value = info->percpu_value_buf;
 664		}
 665	}
 666
 667	return bpf_iter_run_prog(prog, &ctx);
 668}
 669
 670static int bpf_array_map_seq_show(struct seq_file *seq, void *v)
 671{
 672	return __bpf_array_map_seq_show(seq, v);
 673}
 674
 675static void bpf_array_map_seq_stop(struct seq_file *seq, void *v)
 676{
 677	if (!v)
 678		(void)__bpf_array_map_seq_show(seq, NULL);
 679}
 680
 681static int bpf_iter_init_array_map(void *priv_data,
 682				   struct bpf_iter_aux_info *aux)
 683{
 684	struct bpf_iter_seq_array_map_info *seq_info = priv_data;
 685	struct bpf_map *map = aux->map;
 686	struct bpf_array *array = container_of(map, struct bpf_array, map);
 687	void *value_buf;
 688	u32 buf_size;
 689
 690	if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 691		buf_size = array->elem_size * num_possible_cpus();
 692		value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
 693		if (!value_buf)
 694			return -ENOMEM;
 695
 696		seq_info->percpu_value_buf = value_buf;
 697	}
 698
 699	/* bpf_iter_attach_map() acquires a map uref, and the uref may be
 700	 * released before or in the middle of iterating map elements, so
 701	 * acquire an extra map uref for iterator.
 702	 */
 703	bpf_map_inc_with_uref(map);
 704	seq_info->map = map;
 705	return 0;
 706}
 707
 708static void bpf_iter_fini_array_map(void *priv_data)
 709{
 710	struct bpf_iter_seq_array_map_info *seq_info = priv_data;
 711
 712	bpf_map_put_with_uref(seq_info->map);
 713	kfree(seq_info->percpu_value_buf);
 714}
 715
 716static const struct seq_operations bpf_array_map_seq_ops = {
 717	.start	= bpf_array_map_seq_start,
 718	.next	= bpf_array_map_seq_next,
 719	.stop	= bpf_array_map_seq_stop,
 720	.show	= bpf_array_map_seq_show,
 721};
 722
 723static const struct bpf_iter_seq_info iter_seq_info = {
 724	.seq_ops		= &bpf_array_map_seq_ops,
 725	.init_seq_private	= bpf_iter_init_array_map,
 726	.fini_seq_private	= bpf_iter_fini_array_map,
 727	.seq_priv_size		= sizeof(struct bpf_iter_seq_array_map_info),
 728};
 729
 730static long bpf_for_each_array_elem(struct bpf_map *map, bpf_callback_t callback_fn,
 731				    void *callback_ctx, u64 flags)
 732{
 733	u32 i, key, num_elems = 0;
 734	struct bpf_array *array;
 735	bool is_percpu;
 736	u64 ret = 0;
 737	void *val;
 738
 739	cant_migrate();
 740
 741	if (flags != 0)
 742		return -EINVAL;
 743
 744	is_percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
 745	array = container_of(map, struct bpf_array, map);
 746	for (i = 0; i < map->max_entries; i++) {
 747		if (is_percpu)
 748			val = this_cpu_ptr(array->pptrs[i]);
 749		else
 750			val = array_map_elem_ptr(array, i);
 751		num_elems++;
 752		key = i;
 753		ret = callback_fn((u64)(long)map, (u64)(long)&key,
 754				  (u64)(long)val, (u64)(long)callback_ctx, 0);
 755		/* return value: 0 - continue, 1 - stop and return */
 756		if (ret)
 757			break;
 758	}
 759
 760	return num_elems;
 761}
 762
 763static u64 array_map_mem_usage(const struct bpf_map *map)
 764{
 765	struct bpf_array *array = container_of(map, struct bpf_array, map);
 766	bool percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
 767	u32 elem_size = array->elem_size;
 768	u64 entries = map->max_entries;
 769	u64 usage = sizeof(*array);
 770
 771	if (percpu) {
 772		usage += entries * sizeof(void *);
 773		usage += entries * elem_size * num_possible_cpus();
 774	} else {
 775		if (map->map_flags & BPF_F_MMAPABLE) {
 776			usage = PAGE_ALIGN(usage);
 777			usage += PAGE_ALIGN(entries * elem_size);
 778		} else {
 779			usage += entries * elem_size;
 780		}
 781	}
 782	return usage;
 783}
 784
 785BTF_ID_LIST_SINGLE(array_map_btf_ids, struct, bpf_array)
 786const struct bpf_map_ops array_map_ops = {
 787	.map_meta_equal = array_map_meta_equal,
 788	.map_alloc_check = array_map_alloc_check,
 789	.map_alloc = array_map_alloc,
 790	.map_free = array_map_free,
 791	.map_get_next_key = bpf_array_get_next_key,
 792	.map_release_uref = array_map_free_internal_structs,
 793	.map_lookup_elem = array_map_lookup_elem,
 794	.map_update_elem = array_map_update_elem,
 795	.map_delete_elem = array_map_delete_elem,
 796	.map_gen_lookup = array_map_gen_lookup,
 797	.map_direct_value_addr = array_map_direct_value_addr,
 798	.map_direct_value_meta = array_map_direct_value_meta,
 799	.map_mmap = array_map_mmap,
 800	.map_seq_show_elem = array_map_seq_show_elem,
 801	.map_check_btf = array_map_check_btf,
 802	.map_lookup_batch = generic_map_lookup_batch,
 803	.map_update_batch = generic_map_update_batch,
 804	.map_set_for_each_callback_args = map_set_for_each_callback_args,
 805	.map_for_each_callback = bpf_for_each_array_elem,
 806	.map_mem_usage = array_map_mem_usage,
 807	.map_btf_id = &array_map_btf_ids[0],
 808	.iter_seq_info = &iter_seq_info,
 809	.map_get_hash = &array_map_get_hash,
 810};
 811
 812const struct bpf_map_ops percpu_array_map_ops = {
 813	.map_meta_equal = bpf_map_meta_equal,
 814	.map_alloc_check = array_map_alloc_check,
 815	.map_alloc = array_map_alloc,
 816	.map_free = array_map_free,
 817	.map_get_next_key = bpf_array_get_next_key,
 818	.map_lookup_elem = percpu_array_map_lookup_elem,
 819	.map_gen_lookup = percpu_array_map_gen_lookup,
 820	.map_update_elem = array_map_update_elem,
 821	.map_delete_elem = array_map_delete_elem,
 822	.map_lookup_percpu_elem = percpu_array_map_lookup_percpu_elem,
 823	.map_seq_show_elem = percpu_array_map_seq_show_elem,
 824	.map_check_btf = array_map_check_btf,
 825	.map_lookup_batch = generic_map_lookup_batch,
 826	.map_update_batch = generic_map_update_batch,
 827	.map_set_for_each_callback_args = map_set_for_each_callback_args,
 828	.map_for_each_callback = bpf_for_each_array_elem,
 829	.map_mem_usage = array_map_mem_usage,
 830	.map_btf_id = &array_map_btf_ids[0],
 831	.iter_seq_info = &iter_seq_info,
 832};
 833
 834static int fd_array_map_alloc_check(union bpf_attr *attr)
 835{
 836	/* only file descriptors can be stored in this type of map */
 837	if (attr->value_size != sizeof(u32))
 838		return -EINVAL;
 839	/* Program read-only/write-only not supported for special maps yet. */
 840	if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG))
 841		return -EINVAL;
 842	return array_map_alloc_check(attr);
 843}
 844
 845static void fd_array_map_free(struct bpf_map *map)
 846{
 847	struct bpf_array *array = container_of(map, struct bpf_array, map);
 848	int i;
 849
 850	/* make sure it's empty */
 851	for (i = 0; i < array->map.max_entries; i++)
 852		BUG_ON(array->ptrs[i] != NULL);
 853
 854	bpf_map_area_free(array);
 855}
 856
 857static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
 858{
 859	return ERR_PTR(-EOPNOTSUPP);
 860}
 861
 862/* only called from syscall */
 863int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
 864{
 865	void **elem, *ptr;
 866	int ret =  0;
 867
 868	if (!map->ops->map_fd_sys_lookup_elem)
 869		return -ENOTSUPP;
 870
 871	rcu_read_lock();
 872	elem = array_map_lookup_elem(map, key);
 873	if (elem && (ptr = READ_ONCE(*elem)))
 874		*value = map->ops->map_fd_sys_lookup_elem(ptr);
 875	else
 876		ret = -ENOENT;
 877	rcu_read_unlock();
 878
 879	return ret;
 880}
 881
 882/* only called from syscall */
 883int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
 884				 void *key, void *value, u64 map_flags)
 885{
 886	struct bpf_array *array = container_of(map, struct bpf_array, map);
 887	void *new_ptr, *old_ptr;
 888	u32 index = *(u32 *)key, ufd;
 889
 890	if (map_flags != BPF_ANY)
 891		return -EINVAL;
 892
 893	if (index >= array->map.max_entries)
 894		return -E2BIG;
 895
 896	ufd = *(u32 *)value;
 897	new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
 898	if (IS_ERR(new_ptr))
 899		return PTR_ERR(new_ptr);
 900
 901	if (map->ops->map_poke_run) {
 902		mutex_lock(&array->aux->poke_mutex);
 903		old_ptr = xchg(array->ptrs + index, new_ptr);
 904		map->ops->map_poke_run(map, index, old_ptr, new_ptr);
 905		mutex_unlock(&array->aux->poke_mutex);
 906	} else {
 907		old_ptr = xchg(array->ptrs + index, new_ptr);
 908	}
 909
 910	if (old_ptr)
 911		map->ops->map_fd_put_ptr(map, old_ptr, true);
 912	return 0;
 913}
 914
 915static long __fd_array_map_delete_elem(struct bpf_map *map, void *key, bool need_defer)
 916{
 917	struct bpf_array *array = container_of(map, struct bpf_array, map);
 918	void *old_ptr;
 919	u32 index = *(u32 *)key;
 920
 921	if (index >= array->map.max_entries)
 922		return -E2BIG;
 923
 924	if (map->ops->map_poke_run) {
 925		mutex_lock(&array->aux->poke_mutex);
 926		old_ptr = xchg(array->ptrs + index, NULL);
 927		map->ops->map_poke_run(map, index, old_ptr, NULL);
 928		mutex_unlock(&array->aux->poke_mutex);
 929	} else {
 930		old_ptr = xchg(array->ptrs + index, NULL);
 931	}
 932
 933	if (old_ptr) {
 934		map->ops->map_fd_put_ptr(map, old_ptr, need_defer);
 935		return 0;
 936	} else {
 937		return -ENOENT;
 938	}
 939}
 940
 941static long fd_array_map_delete_elem(struct bpf_map *map, void *key)
 942{
 943	return __fd_array_map_delete_elem(map, key, true);
 944}
 945
 946static void *prog_fd_array_get_ptr(struct bpf_map *map,
 947				   struct file *map_file, int fd)
 948{
 949	struct bpf_prog *prog = bpf_prog_get(fd);
 950	bool is_extended;
 951
 952	if (IS_ERR(prog))
 953		return prog;
 954
 955	if (prog->type == BPF_PROG_TYPE_EXT ||
 956	    !bpf_prog_map_compatible(map, prog)) {
 957		bpf_prog_put(prog);
 958		return ERR_PTR(-EINVAL);
 959	}
 960
 961	mutex_lock(&prog->aux->ext_mutex);
 962	is_extended = prog->aux->is_extended;
 963	if (!is_extended)
 964		prog->aux->prog_array_member_cnt++;
 965	mutex_unlock(&prog->aux->ext_mutex);
 966	if (is_extended) {
 967		/* Extended prog can not be tail callee. It's to prevent a
 968		 * potential infinite loop like:
 969		 * tail callee prog entry -> tail callee prog subprog ->
 970		 * freplace prog entry --tailcall-> tail callee prog entry.
 971		 */
 972		bpf_prog_put(prog);
 973		return ERR_PTR(-EBUSY);
 974	}
 975
 976	return prog;
 977}
 978
 979static void prog_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
 980{
 981	struct bpf_prog *prog = ptr;
 982
 983	mutex_lock(&prog->aux->ext_mutex);
 984	prog->aux->prog_array_member_cnt--;
 985	mutex_unlock(&prog->aux->ext_mutex);
 986	/* bpf_prog is freed after one RCU or tasks trace grace period */
 987	bpf_prog_put(prog);
 988}
 989
 990static u32 prog_fd_array_sys_lookup_elem(void *ptr)
 991{
 992	return ((struct bpf_prog *)ptr)->aux->id;
 993}
 994
 995/* decrement refcnt of all bpf_progs that are stored in this map */
 996static void bpf_fd_array_map_clear(struct bpf_map *map, bool need_defer)
 997{
 998	struct bpf_array *array = container_of(map, struct bpf_array, map);
 999	int i;
1000
1001	for (i = 0; i < array->map.max_entries; i++)
1002		__fd_array_map_delete_elem(map, &i, need_defer);
1003}
1004
1005static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
1006					 struct seq_file *m)
1007{
1008	void **elem, *ptr;
1009	u32 prog_id;
1010
1011	rcu_read_lock();
1012
1013	elem = array_map_lookup_elem(map, key);
1014	if (elem) {
1015		ptr = READ_ONCE(*elem);
1016		if (ptr) {
1017			seq_printf(m, "%u: ", *(u32 *)key);
1018			prog_id = prog_fd_array_sys_lookup_elem(ptr);
1019			btf_type_seq_show(map->btf, map->btf_value_type_id,
1020					  &prog_id, m);
1021			seq_putc(m, '\n');
1022		}
1023	}
1024
1025	rcu_read_unlock();
1026}
1027
1028struct prog_poke_elem {
1029	struct list_head list;
1030	struct bpf_prog_aux *aux;
1031};
1032
1033static int prog_array_map_poke_track(struct bpf_map *map,
1034				     struct bpf_prog_aux *prog_aux)
1035{
1036	struct prog_poke_elem *elem;
1037	struct bpf_array_aux *aux;
1038	int ret = 0;
1039
1040	aux = container_of(map, struct bpf_array, map)->aux;
1041	mutex_lock(&aux->poke_mutex);
1042	list_for_each_entry(elem, &aux->poke_progs, list) {
1043		if (elem->aux == prog_aux)
1044			goto out;
1045	}
1046
1047	elem = kmalloc(sizeof(*elem), GFP_KERNEL);
1048	if (!elem) {
1049		ret = -ENOMEM;
1050		goto out;
1051	}
1052
1053	INIT_LIST_HEAD(&elem->list);
1054	/* We must track the program's aux info at this point in time
1055	 * since the program pointer itself may not be stable yet, see
1056	 * also comment in prog_array_map_poke_run().
1057	 */
1058	elem->aux = prog_aux;
1059
1060	list_add_tail(&elem->list, &aux->poke_progs);
1061out:
1062	mutex_unlock(&aux->poke_mutex);
1063	return ret;
1064}
1065
1066static void prog_array_map_poke_untrack(struct bpf_map *map,
1067					struct bpf_prog_aux *prog_aux)
1068{
1069	struct prog_poke_elem *elem, *tmp;
1070	struct bpf_array_aux *aux;
1071
1072	aux = container_of(map, struct bpf_array, map)->aux;
1073	mutex_lock(&aux->poke_mutex);
1074	list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1075		if (elem->aux == prog_aux) {
1076			list_del_init(&elem->list);
1077			kfree(elem);
1078			break;
1079		}
1080	}
1081	mutex_unlock(&aux->poke_mutex);
1082}
1083
1084void __weak bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor *poke,
1085				      struct bpf_prog *new, struct bpf_prog *old)
1086{
1087	WARN_ON_ONCE(1);
1088}
1089
1090static void prog_array_map_poke_run(struct bpf_map *map, u32 key,
1091				    struct bpf_prog *old,
1092				    struct bpf_prog *new)
1093{
1094	struct prog_poke_elem *elem;
1095	struct bpf_array_aux *aux;
1096
1097	aux = container_of(map, struct bpf_array, map)->aux;
1098	WARN_ON_ONCE(!mutex_is_locked(&aux->poke_mutex));
1099
1100	list_for_each_entry(elem, &aux->poke_progs, list) {
1101		struct bpf_jit_poke_descriptor *poke;
1102		int i;
1103
1104		for (i = 0; i < elem->aux->size_poke_tab; i++) {
1105			poke = &elem->aux->poke_tab[i];
1106
1107			/* Few things to be aware of:
1108			 *
1109			 * 1) We can only ever access aux in this context, but
1110			 *    not aux->prog since it might not be stable yet and
1111			 *    there could be danger of use after free otherwise.
1112			 * 2) Initially when we start tracking aux, the program
1113			 *    is not JITed yet and also does not have a kallsyms
1114			 *    entry. We skip these as poke->tailcall_target_stable
1115			 *    is not active yet. The JIT will do the final fixup
1116			 *    before setting it stable. The various
1117			 *    poke->tailcall_target_stable are successively
1118			 *    activated, so tail call updates can arrive from here
1119			 *    while JIT is still finishing its final fixup for
1120			 *    non-activated poke entries.
1121			 * 3) Also programs reaching refcount of zero while patching
1122			 *    is in progress is okay since we're protected under
1123			 *    poke_mutex and untrack the programs before the JIT
1124			 *    buffer is freed.
1125			 */
1126			if (!READ_ONCE(poke->tailcall_target_stable))
1127				continue;
1128			if (poke->reason != BPF_POKE_REASON_TAIL_CALL)
1129				continue;
1130			if (poke->tail_call.map != map ||
1131			    poke->tail_call.key != key)
1132				continue;
1133
1134			bpf_arch_poke_desc_update(poke, new, old);
1135		}
1136	}
1137}
1138
1139static void prog_array_map_clear_deferred(struct work_struct *work)
1140{
1141	struct bpf_map *map = container_of(work, struct bpf_array_aux,
1142					   work)->map;
1143	bpf_fd_array_map_clear(map, true);
1144	bpf_map_put(map);
1145}
1146
1147static void prog_array_map_clear(struct bpf_map *map)
1148{
1149	struct bpf_array_aux *aux = container_of(map, struct bpf_array,
1150						 map)->aux;
1151	bpf_map_inc(map);
1152	schedule_work(&aux->work);
1153}
1154
1155static struct bpf_map *prog_array_map_alloc(union bpf_attr *attr)
1156{
1157	struct bpf_array_aux *aux;
1158	struct bpf_map *map;
1159
1160	aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT);
1161	if (!aux)
1162		return ERR_PTR(-ENOMEM);
1163
1164	INIT_WORK(&aux->work, prog_array_map_clear_deferred);
1165	INIT_LIST_HEAD(&aux->poke_progs);
1166	mutex_init(&aux->poke_mutex);
1167
1168	map = array_map_alloc(attr);
1169	if (IS_ERR(map)) {
1170		kfree(aux);
1171		return map;
1172	}
1173
1174	container_of(map, struct bpf_array, map)->aux = aux;
1175	aux->map = map;
1176
1177	return map;
1178}
1179
1180static void prog_array_map_free(struct bpf_map *map)
1181{
1182	struct prog_poke_elem *elem, *tmp;
1183	struct bpf_array_aux *aux;
1184
1185	aux = container_of(map, struct bpf_array, map)->aux;
1186	list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1187		list_del_init(&elem->list);
1188		kfree(elem);
1189	}
1190	kfree(aux);
1191	fd_array_map_free(map);
1192}
1193
1194/* prog_array->aux->{type,jited} is a runtime binding.
1195 * Doing static check alone in the verifier is not enough.
1196 * Thus, prog_array_map cannot be used as an inner_map
1197 * and map_meta_equal is not implemented.
1198 */
1199const struct bpf_map_ops prog_array_map_ops = {
1200	.map_alloc_check = fd_array_map_alloc_check,
1201	.map_alloc = prog_array_map_alloc,
1202	.map_free = prog_array_map_free,
1203	.map_poke_track = prog_array_map_poke_track,
1204	.map_poke_untrack = prog_array_map_poke_untrack,
1205	.map_poke_run = prog_array_map_poke_run,
1206	.map_get_next_key = bpf_array_get_next_key,
1207	.map_lookup_elem = fd_array_map_lookup_elem,
1208	.map_delete_elem = fd_array_map_delete_elem,
1209	.map_fd_get_ptr = prog_fd_array_get_ptr,
1210	.map_fd_put_ptr = prog_fd_array_put_ptr,
1211	.map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
1212	.map_release_uref = prog_array_map_clear,
1213	.map_seq_show_elem = prog_array_map_seq_show_elem,
1214	.map_mem_usage = array_map_mem_usage,
1215	.map_btf_id = &array_map_btf_ids[0],
1216};
1217
1218static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
1219						   struct file *map_file)
1220{
1221	struct bpf_event_entry *ee;
1222
1223	ee = kzalloc(sizeof(*ee), GFP_KERNEL);
1224	if (ee) {
1225		ee->event = perf_file->private_data;
1226		ee->perf_file = perf_file;
1227		ee->map_file = map_file;
1228	}
1229
1230	return ee;
1231}
1232
1233static void __bpf_event_entry_free(struct rcu_head *rcu)
1234{
1235	struct bpf_event_entry *ee;
1236
1237	ee = container_of(rcu, struct bpf_event_entry, rcu);
1238	fput(ee->perf_file);
1239	kfree(ee);
1240}
1241
1242static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
1243{
1244	call_rcu(&ee->rcu, __bpf_event_entry_free);
1245}
1246
1247static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
1248					 struct file *map_file, int fd)
1249{
1250	struct bpf_event_entry *ee;
1251	struct perf_event *event;
1252	struct file *perf_file;
1253	u64 value;
1254
1255	perf_file = perf_event_get(fd);
1256	if (IS_ERR(perf_file))
1257		return perf_file;
1258
1259	ee = ERR_PTR(-EOPNOTSUPP);
1260	event = perf_file->private_data;
1261	if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
1262		goto err_out;
1263
1264	ee = bpf_event_entry_gen(perf_file, map_file);
1265	if (ee)
1266		return ee;
1267	ee = ERR_PTR(-ENOMEM);
1268err_out:
1269	fput(perf_file);
1270	return ee;
1271}
1272
1273static void perf_event_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1274{
1275	/* bpf_perf_event is freed after one RCU grace period */
1276	bpf_event_entry_free_rcu(ptr);
1277}
1278
1279static void perf_event_fd_array_release(struct bpf_map *map,
1280					struct file *map_file)
1281{
1282	struct bpf_array *array = container_of(map, struct bpf_array, map);
1283	struct bpf_event_entry *ee;
1284	int i;
1285
1286	if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1287		return;
1288
1289	rcu_read_lock();
1290	for (i = 0; i < array->map.max_entries; i++) {
1291		ee = READ_ONCE(array->ptrs[i]);
1292		if (ee && ee->map_file == map_file)
1293			__fd_array_map_delete_elem(map, &i, true);
1294	}
1295	rcu_read_unlock();
1296}
1297
1298static void perf_event_fd_array_map_free(struct bpf_map *map)
1299{
1300	if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1301		bpf_fd_array_map_clear(map, false);
1302	fd_array_map_free(map);
1303}
1304
1305const struct bpf_map_ops perf_event_array_map_ops = {
1306	.map_meta_equal = bpf_map_meta_equal,
1307	.map_alloc_check = fd_array_map_alloc_check,
1308	.map_alloc = array_map_alloc,
1309	.map_free = perf_event_fd_array_map_free,
1310	.map_get_next_key = bpf_array_get_next_key,
1311	.map_lookup_elem = fd_array_map_lookup_elem,
1312	.map_delete_elem = fd_array_map_delete_elem,
1313	.map_fd_get_ptr = perf_event_fd_array_get_ptr,
1314	.map_fd_put_ptr = perf_event_fd_array_put_ptr,
1315	.map_release = perf_event_fd_array_release,
1316	.map_check_btf = map_check_no_btf,
1317	.map_mem_usage = array_map_mem_usage,
1318	.map_btf_id = &array_map_btf_ids[0],
1319};
1320
1321#ifdef CONFIG_CGROUPS
1322static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
1323				     struct file *map_file /* not used */,
1324				     int fd)
1325{
1326	return cgroup_get_from_fd(fd);
1327}
1328
1329static void cgroup_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1330{
1331	/* cgroup_put free cgrp after a rcu grace period */
1332	cgroup_put(ptr);
1333}
1334
1335static void cgroup_fd_array_free(struct bpf_map *map)
1336{
1337	bpf_fd_array_map_clear(map, false);
1338	fd_array_map_free(map);
1339}
1340
1341const struct bpf_map_ops cgroup_array_map_ops = {
1342	.map_meta_equal = bpf_map_meta_equal,
1343	.map_alloc_check = fd_array_map_alloc_check,
1344	.map_alloc = array_map_alloc,
1345	.map_free = cgroup_fd_array_free,
1346	.map_get_next_key = bpf_array_get_next_key,
1347	.map_lookup_elem = fd_array_map_lookup_elem,
1348	.map_delete_elem = fd_array_map_delete_elem,
1349	.map_fd_get_ptr = cgroup_fd_array_get_ptr,
1350	.map_fd_put_ptr = cgroup_fd_array_put_ptr,
1351	.map_check_btf = map_check_no_btf,
1352	.map_mem_usage = array_map_mem_usage,
1353	.map_btf_id = &array_map_btf_ids[0],
1354};
1355#endif
1356
1357static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
1358{
1359	struct bpf_map *map, *inner_map_meta;
1360
1361	inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
1362	if (IS_ERR(inner_map_meta))
1363		return inner_map_meta;
1364
1365	map = array_map_alloc(attr);
1366	if (IS_ERR(map)) {
1367		bpf_map_meta_free(inner_map_meta);
1368		return map;
1369	}
1370
1371	map->inner_map_meta = inner_map_meta;
1372
1373	return map;
1374}
1375
1376static void array_of_map_free(struct bpf_map *map)
1377{
1378	/* map->inner_map_meta is only accessed by syscall which
1379	 * is protected by fdget/fdput.
1380	 */
1381	bpf_map_meta_free(map->inner_map_meta);
1382	bpf_fd_array_map_clear(map, false);
1383	fd_array_map_free(map);
1384}
1385
1386static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
1387{
1388	struct bpf_map **inner_map = array_map_lookup_elem(map, key);
1389
1390	if (!inner_map)
1391		return NULL;
1392
1393	return READ_ONCE(*inner_map);
1394}
1395
1396static int array_of_map_gen_lookup(struct bpf_map *map,
1397				   struct bpf_insn *insn_buf)
1398{
1399	struct bpf_array *array = container_of(map, struct bpf_array, map);
1400	u32 elem_size = array->elem_size;
1401	struct bpf_insn *insn = insn_buf;
1402	const int ret = BPF_REG_0;
1403	const int map_ptr = BPF_REG_1;
1404	const int index = BPF_REG_2;
1405
1406	*insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
1407	*insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
1408	if (!map->bypass_spec_v1) {
1409		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
1410		*insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
1411	} else {
1412		*insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
1413	}
1414	if (is_power_of_2(elem_size))
1415		*insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
1416	else
1417		*insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
1418	*insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
1419	*insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
1420	*insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
1421	*insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
1422	*insn++ = BPF_MOV64_IMM(ret, 0);
1423
1424	return insn - insn_buf;
1425}
1426
1427const struct bpf_map_ops array_of_maps_map_ops = {
1428	.map_alloc_check = fd_array_map_alloc_check,
1429	.map_alloc = array_of_map_alloc,
1430	.map_free = array_of_map_free,
1431	.map_get_next_key = bpf_array_get_next_key,
1432	.map_lookup_elem = array_of_map_lookup_elem,
1433	.map_delete_elem = fd_array_map_delete_elem,
1434	.map_fd_get_ptr = bpf_map_fd_get_ptr,
1435	.map_fd_put_ptr = bpf_map_fd_put_ptr,
1436	.map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
1437	.map_gen_lookup = array_of_map_gen_lookup,
1438	.map_lookup_batch = generic_map_lookup_batch,
1439	.map_update_batch = generic_map_update_batch,
1440	.map_check_btf = map_check_no_btf,
1441	.map_mem_usage = array_map_mem_usage,
1442	.map_btf_id = &array_map_btf_ids[0],
1443};