1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 * Copyright (c) 2016,2017 Facebook
8 #include <linux/slab.h>
10 #include <linux/filter.h>
11 #include <linux/perf_event.h>
12 #include <uapi/linux/btf.h>
14 #include "map_in_map.h"
16 #define ARRAY_CREATE_FLAG_MASK \
17 (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK)
19 static void bpf_array_free_percpu(struct bpf_array *array)
23 for (i = 0; i < array->map.max_entries; i++) {
24 free_percpu(array->pptrs[i]);
29 static int bpf_array_alloc_percpu(struct bpf_array *array)
34 for (i = 0; i < array->map.max_entries; i++) {
35 ptr = __alloc_percpu_gfp(array->elem_size, 8,
36 GFP_USER | __GFP_NOWARN);
38 bpf_array_free_percpu(array);
41 array->pptrs[i] = ptr;
48 /* Called from syscall */
49 int array_map_alloc_check(union bpf_attr *attr)
51 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
52 int numa_node = bpf_map_attr_numa_node(attr);
54 /* check sanity of attributes */
55 if (attr->max_entries == 0 || attr->key_size != 4 ||
56 attr->value_size == 0 ||
57 attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
58 !bpf_map_flags_access_ok(attr->map_flags) ||
59 (percpu && numa_node != NUMA_NO_NODE))
62 if (attr->value_size > KMALLOC_MAX_SIZE)
63 /* if value_size is bigger, the user space won't be able to
64 * access the elements.
71 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
73 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
74 int ret, numa_node = bpf_map_attr_numa_node(attr);
75 u32 elem_size, index_mask, max_entries;
76 bool unpriv = !capable(CAP_SYS_ADMIN);
77 u64 cost, array_size, mask64;
78 struct bpf_map_memory mem;
79 struct bpf_array *array;
81 elem_size = round_up(attr->value_size, 8);
83 max_entries = attr->max_entries;
85 /* On 32 bit archs roundup_pow_of_two() with max_entries that has
86 * upper most bit set in u32 space is undefined behavior due to
87 * resulting 1U << 32, so do it manually here in u64 space.
89 mask64 = fls_long(max_entries - 1);
90 mask64 = 1ULL << mask64;
95 /* round up array size to nearest power of 2,
96 * since cpu will speculate within index_mask limits
98 max_entries = index_mask + 1;
99 /* Check for overflows. */
100 if (max_entries < attr->max_entries)
101 return ERR_PTR(-E2BIG);
104 array_size = sizeof(*array);
106 array_size += (u64) max_entries * sizeof(void *);
108 array_size += (u64) max_entries * elem_size;
110 /* make sure there is no u32 overflow later in round_up() */
113 cost += (u64)attr->max_entries * elem_size * num_possible_cpus();
115 ret = bpf_map_charge_init(&mem, cost);
119 /* allocate all map elements and zero-initialize them */
120 array = bpf_map_area_alloc(array_size, numa_node);
122 bpf_map_charge_finish(&mem);
123 return ERR_PTR(-ENOMEM);
125 array->index_mask = index_mask;
126 array->map.unpriv_array = unpriv;
128 /* copy mandatory map attributes */
129 bpf_map_init_from_attr(&array->map, attr);
130 bpf_map_charge_move(&array->map.memory, &mem);
131 array->elem_size = elem_size;
133 if (percpu && bpf_array_alloc_percpu(array)) {
134 bpf_map_charge_finish(&array->map.memory);
135 bpf_map_area_free(array);
136 return ERR_PTR(-ENOMEM);
142 /* Called from syscall or from eBPF program */
143 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
145 struct bpf_array *array = container_of(map, struct bpf_array, map);
146 u32 index = *(u32 *)key;
148 if (unlikely(index >= array->map.max_entries))
151 return array->value + array->elem_size * (index & array->index_mask);
154 static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm,
157 struct bpf_array *array = container_of(map, struct bpf_array, map);
159 if (map->max_entries != 1)
161 if (off >= map->value_size)
164 *imm = (unsigned long)array->value;
168 static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm,
171 struct bpf_array *array = container_of(map, struct bpf_array, map);
172 u64 base = (unsigned long)array->value;
173 u64 range = array->elem_size;
175 if (map->max_entries != 1)
177 if (imm < base || imm >= base + range)
184 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
185 static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
187 struct bpf_array *array = container_of(map, struct bpf_array, map);
188 struct bpf_insn *insn = insn_buf;
189 u32 elem_size = round_up(map->value_size, 8);
190 const int ret = BPF_REG_0;
191 const int map_ptr = BPF_REG_1;
192 const int index = BPF_REG_2;
194 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
195 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
196 if (map->unpriv_array) {
197 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
198 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
200 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
203 if (is_power_of_2(elem_size)) {
204 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
206 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
208 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
209 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
210 *insn++ = BPF_MOV64_IMM(ret, 0);
211 return insn - insn_buf;
214 /* Called from eBPF program */
215 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
217 struct bpf_array *array = container_of(map, struct bpf_array, map);
218 u32 index = *(u32 *)key;
220 if (unlikely(index >= array->map.max_entries))
223 return this_cpu_ptr(array->pptrs[index & array->index_mask]);
226 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
228 struct bpf_array *array = container_of(map, struct bpf_array, map);
229 u32 index = *(u32 *)key;
234 if (unlikely(index >= array->map.max_entries))
237 /* per_cpu areas are zero-filled and bpf programs can only
238 * access 'value_size' of them, so copying rounded areas
239 * will not leak any kernel data
241 size = round_up(map->value_size, 8);
243 pptr = array->pptrs[index & array->index_mask];
244 for_each_possible_cpu(cpu) {
245 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
252 /* Called from syscall */
253 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
255 struct bpf_array *array = container_of(map, struct bpf_array, map);
256 u32 index = key ? *(u32 *)key : U32_MAX;
257 u32 *next = (u32 *)next_key;
259 if (index >= array->map.max_entries) {
264 if (index == array->map.max_entries - 1)
271 /* Called from syscall or from eBPF program */
272 static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
275 struct bpf_array *array = container_of(map, struct bpf_array, map);
276 u32 index = *(u32 *)key;
279 if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
283 if (unlikely(index >= array->map.max_entries))
284 /* all elements were pre-allocated, cannot insert a new one */
287 if (unlikely(map_flags & BPF_NOEXIST))
288 /* all elements already exist */
291 if (unlikely((map_flags & BPF_F_LOCK) &&
292 !map_value_has_spin_lock(map)))
295 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
296 memcpy(this_cpu_ptr(array->pptrs[index & array->index_mask]),
297 value, map->value_size);
300 array->elem_size * (index & array->index_mask);
301 if (map_flags & BPF_F_LOCK)
302 copy_map_value_locked(map, val, value, false);
304 copy_map_value(map, val, value);
309 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
312 struct bpf_array *array = container_of(map, struct bpf_array, map);
313 u32 index = *(u32 *)key;
318 if (unlikely(map_flags > BPF_EXIST))
322 if (unlikely(index >= array->map.max_entries))
323 /* all elements were pre-allocated, cannot insert a new one */
326 if (unlikely(map_flags == BPF_NOEXIST))
327 /* all elements already exist */
330 /* the user space will provide round_up(value_size, 8) bytes that
331 * will be copied into per-cpu area. bpf programs can only access
332 * value_size of it. During lookup the same extra bytes will be
333 * returned or zeros which were zero-filled by percpu_alloc,
334 * so no kernel data leaks possible
336 size = round_up(map->value_size, 8);
338 pptr = array->pptrs[index & array->index_mask];
339 for_each_possible_cpu(cpu) {
340 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
347 /* Called from syscall or from eBPF program */
348 static int array_map_delete_elem(struct bpf_map *map, void *key)
353 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
354 static void array_map_free(struct bpf_map *map)
356 struct bpf_array *array = container_of(map, struct bpf_array, map);
358 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
359 * so the programs (can be more than one that used this map) were
360 * disconnected from events. Wait for outstanding programs to complete
365 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
366 bpf_array_free_percpu(array);
368 bpf_map_area_free(array);
371 static void array_map_seq_show_elem(struct bpf_map *map, void *key,
378 value = array_map_lookup_elem(map, key);
384 if (map->btf_key_type_id)
385 seq_printf(m, "%u: ", *(u32 *)key);
386 btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
392 static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
395 struct bpf_array *array = container_of(map, struct bpf_array, map);
396 u32 index = *(u32 *)key;
402 seq_printf(m, "%u: {\n", *(u32 *)key);
403 pptr = array->pptrs[index & array->index_mask];
404 for_each_possible_cpu(cpu) {
405 seq_printf(m, "\tcpu%d: ", cpu);
406 btf_type_seq_show(map->btf, map->btf_value_type_id,
407 per_cpu_ptr(pptr, cpu), m);
415 static int array_map_check_btf(const struct bpf_map *map,
416 const struct btf *btf,
417 const struct btf_type *key_type,
418 const struct btf_type *value_type)
422 /* One exception for keyless BTF: .bss/.data/.rodata map */
423 if (btf_type_is_void(key_type)) {
424 if (map->map_type != BPF_MAP_TYPE_ARRAY ||
425 map->max_entries != 1)
428 if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC)
434 if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT)
437 int_data = *(u32 *)(key_type + 1);
438 /* bpf array can only take a u32 key. This check makes sure
439 * that the btf matches the attr used during map_create.
441 if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data))
447 const struct bpf_map_ops array_map_ops = {
448 .map_alloc_check = array_map_alloc_check,
449 .map_alloc = array_map_alloc,
450 .map_free = array_map_free,
451 .map_get_next_key = array_map_get_next_key,
452 .map_lookup_elem = array_map_lookup_elem,
453 .map_update_elem = array_map_update_elem,
454 .map_delete_elem = array_map_delete_elem,
455 .map_gen_lookup = array_map_gen_lookup,
456 .map_direct_value_addr = array_map_direct_value_addr,
457 .map_direct_value_meta = array_map_direct_value_meta,
458 .map_seq_show_elem = array_map_seq_show_elem,
459 .map_check_btf = array_map_check_btf,
462 const struct bpf_map_ops percpu_array_map_ops = {
463 .map_alloc_check = array_map_alloc_check,
464 .map_alloc = array_map_alloc,
465 .map_free = array_map_free,
466 .map_get_next_key = array_map_get_next_key,
467 .map_lookup_elem = percpu_array_map_lookup_elem,
468 .map_update_elem = array_map_update_elem,
469 .map_delete_elem = array_map_delete_elem,
470 .map_seq_show_elem = percpu_array_map_seq_show_elem,
471 .map_check_btf = array_map_check_btf,
474 static int fd_array_map_alloc_check(union bpf_attr *attr)
476 /* only file descriptors can be stored in this type of map */
477 if (attr->value_size != sizeof(u32))
479 /* Program read-only/write-only not supported for special maps yet. */
480 if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG))
482 return array_map_alloc_check(attr);
485 static void fd_array_map_free(struct bpf_map *map)
487 struct bpf_array *array = container_of(map, struct bpf_array, map);
492 /* make sure it's empty */
493 for (i = 0; i < array->map.max_entries; i++)
494 BUG_ON(array->ptrs[i] != NULL);
496 bpf_map_area_free(array);
499 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
501 return ERR_PTR(-EOPNOTSUPP);
504 /* only called from syscall */
505 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
510 if (!map->ops->map_fd_sys_lookup_elem)
514 elem = array_map_lookup_elem(map, key);
515 if (elem && (ptr = READ_ONCE(*elem)))
516 *value = map->ops->map_fd_sys_lookup_elem(ptr);
524 /* only called from syscall */
525 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
526 void *key, void *value, u64 map_flags)
528 struct bpf_array *array = container_of(map, struct bpf_array, map);
529 void *new_ptr, *old_ptr;
530 u32 index = *(u32 *)key, ufd;
532 if (map_flags != BPF_ANY)
535 if (index >= array->map.max_entries)
539 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
541 return PTR_ERR(new_ptr);
543 old_ptr = xchg(array->ptrs + index, new_ptr);
545 map->ops->map_fd_put_ptr(old_ptr);
550 static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
552 struct bpf_array *array = container_of(map, struct bpf_array, map);
554 u32 index = *(u32 *)key;
556 if (index >= array->map.max_entries)
559 old_ptr = xchg(array->ptrs + index, NULL);
561 map->ops->map_fd_put_ptr(old_ptr);
568 static void *prog_fd_array_get_ptr(struct bpf_map *map,
569 struct file *map_file, int fd)
571 struct bpf_array *array = container_of(map, struct bpf_array, map);
572 struct bpf_prog *prog = bpf_prog_get(fd);
577 if (!bpf_prog_array_compatible(array, prog)) {
579 return ERR_PTR(-EINVAL);
585 static void prog_fd_array_put_ptr(void *ptr)
590 static u32 prog_fd_array_sys_lookup_elem(void *ptr)
592 return ((struct bpf_prog *)ptr)->aux->id;
595 /* decrement refcnt of all bpf_progs that are stored in this map */
596 static void bpf_fd_array_map_clear(struct bpf_map *map)
598 struct bpf_array *array = container_of(map, struct bpf_array, map);
601 for (i = 0; i < array->map.max_entries; i++)
602 fd_array_map_delete_elem(map, &i);
605 static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
613 elem = array_map_lookup_elem(map, key);
615 ptr = READ_ONCE(*elem);
617 seq_printf(m, "%u: ", *(u32 *)key);
618 prog_id = prog_fd_array_sys_lookup_elem(ptr);
619 btf_type_seq_show(map->btf, map->btf_value_type_id,
628 const struct bpf_map_ops prog_array_map_ops = {
629 .map_alloc_check = fd_array_map_alloc_check,
630 .map_alloc = array_map_alloc,
631 .map_free = fd_array_map_free,
632 .map_get_next_key = array_map_get_next_key,
633 .map_lookup_elem = fd_array_map_lookup_elem,
634 .map_delete_elem = fd_array_map_delete_elem,
635 .map_fd_get_ptr = prog_fd_array_get_ptr,
636 .map_fd_put_ptr = prog_fd_array_put_ptr,
637 .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
638 .map_release_uref = bpf_fd_array_map_clear,
639 .map_seq_show_elem = prog_array_map_seq_show_elem,
642 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
643 struct file *map_file)
645 struct bpf_event_entry *ee;
647 ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
649 ee->event = perf_file->private_data;
650 ee->perf_file = perf_file;
651 ee->map_file = map_file;
657 static void __bpf_event_entry_free(struct rcu_head *rcu)
659 struct bpf_event_entry *ee;
661 ee = container_of(rcu, struct bpf_event_entry, rcu);
666 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
668 call_rcu(&ee->rcu, __bpf_event_entry_free);
671 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
672 struct file *map_file, int fd)
674 struct bpf_event_entry *ee;
675 struct perf_event *event;
676 struct file *perf_file;
679 perf_file = perf_event_get(fd);
680 if (IS_ERR(perf_file))
683 ee = ERR_PTR(-EOPNOTSUPP);
684 event = perf_file->private_data;
685 if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
688 ee = bpf_event_entry_gen(perf_file, map_file);
691 ee = ERR_PTR(-ENOMEM);
697 static void perf_event_fd_array_put_ptr(void *ptr)
699 bpf_event_entry_free_rcu(ptr);
702 static void perf_event_fd_array_release(struct bpf_map *map,
703 struct file *map_file)
705 struct bpf_array *array = container_of(map, struct bpf_array, map);
706 struct bpf_event_entry *ee;
710 for (i = 0; i < array->map.max_entries; i++) {
711 ee = READ_ONCE(array->ptrs[i]);
712 if (ee && ee->map_file == map_file)
713 fd_array_map_delete_elem(map, &i);
718 const struct bpf_map_ops perf_event_array_map_ops = {
719 .map_alloc_check = fd_array_map_alloc_check,
720 .map_alloc = array_map_alloc,
721 .map_free = fd_array_map_free,
722 .map_get_next_key = array_map_get_next_key,
723 .map_lookup_elem = fd_array_map_lookup_elem,
724 .map_delete_elem = fd_array_map_delete_elem,
725 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
726 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
727 .map_release = perf_event_fd_array_release,
728 .map_check_btf = map_check_no_btf,
731 #ifdef CONFIG_CGROUPS
732 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
733 struct file *map_file /* not used */,
736 return cgroup_get_from_fd(fd);
739 static void cgroup_fd_array_put_ptr(void *ptr)
741 /* cgroup_put free cgrp after a rcu grace period */
745 static void cgroup_fd_array_free(struct bpf_map *map)
747 bpf_fd_array_map_clear(map);
748 fd_array_map_free(map);
751 const struct bpf_map_ops cgroup_array_map_ops = {
752 .map_alloc_check = fd_array_map_alloc_check,
753 .map_alloc = array_map_alloc,
754 .map_free = cgroup_fd_array_free,
755 .map_get_next_key = array_map_get_next_key,
756 .map_lookup_elem = fd_array_map_lookup_elem,
757 .map_delete_elem = fd_array_map_delete_elem,
758 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
759 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
760 .map_check_btf = map_check_no_btf,
764 static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
766 struct bpf_map *map, *inner_map_meta;
768 inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
769 if (IS_ERR(inner_map_meta))
770 return inner_map_meta;
772 map = array_map_alloc(attr);
774 bpf_map_meta_free(inner_map_meta);
778 map->inner_map_meta = inner_map_meta;
783 static void array_of_map_free(struct bpf_map *map)
785 /* map->inner_map_meta is only accessed by syscall which
786 * is protected by fdget/fdput.
788 bpf_map_meta_free(map->inner_map_meta);
789 bpf_fd_array_map_clear(map);
790 fd_array_map_free(map);
793 static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
795 struct bpf_map **inner_map = array_map_lookup_elem(map, key);
800 return READ_ONCE(*inner_map);
803 static u32 array_of_map_gen_lookup(struct bpf_map *map,
804 struct bpf_insn *insn_buf)
806 struct bpf_array *array = container_of(map, struct bpf_array, map);
807 u32 elem_size = round_up(map->value_size, 8);
808 struct bpf_insn *insn = insn_buf;
809 const int ret = BPF_REG_0;
810 const int map_ptr = BPF_REG_1;
811 const int index = BPF_REG_2;
813 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
814 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
815 if (map->unpriv_array) {
816 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
817 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
819 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
821 if (is_power_of_2(elem_size))
822 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
824 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
825 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
826 *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
827 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
828 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
829 *insn++ = BPF_MOV64_IMM(ret, 0);
831 return insn - insn_buf;
834 const struct bpf_map_ops array_of_maps_map_ops = {
835 .map_alloc_check = fd_array_map_alloc_check,
836 .map_alloc = array_of_map_alloc,
837 .map_free = array_of_map_free,
838 .map_get_next_key = array_map_get_next_key,
839 .map_lookup_elem = array_of_map_lookup_elem,
840 .map_delete_elem = fd_array_map_delete_elem,
841 .map_fd_get_ptr = bpf_map_fd_get_ptr,
842 .map_fd_put_ptr = bpf_map_fd_put_ptr,
843 .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
844 .map_gen_lookup = array_of_map_gen_lookup,
845 .map_check_btf = map_check_no_btf,