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>
13 #include <linux/rcupdate_trace.h>
14 #include <linux/btf_ids.h>
16 #include "map_in_map.h"
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)
22 static void bpf_array_free_percpu(struct bpf_array *array)
26 for (i = 0; i < array->map.max_entries; i++) {
27 free_percpu(array->pptrs[i]);
32 static int bpf_array_alloc_percpu(struct bpf_array *array)
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);
41 bpf_array_free_percpu(array);
44 array->pptrs[i] = ptr;
51 /* Called from syscall */
52 int array_map_alloc_check(union bpf_attr *attr)
54 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
55 int numa_node = bpf_map_attr_numa_node(attr);
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))
65 if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
66 attr->map_flags & (BPF_F_MMAPABLE | BPF_F_INNER_MAP))
69 if (attr->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
70 attr->map_flags & BPF_F_PRESERVE_ELEMS)
73 /* avoid overflow on round_up(map->value_size) */
74 if (attr->value_size > INT_MAX)
80 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
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;
89 elem_size = round_up(attr->value_size, 8);
91 max_entries = attr->max_entries;
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.
97 mask64 = fls_long(max_entries - 1);
98 mask64 = 1ULL << 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
106 max_entries = index_mask + 1;
107 /* Check for overflows. */
108 if (max_entries < attr->max_entries)
109 return ERR_PTR(-E2BIG);
112 array_size = sizeof(*array);
114 array_size += (u64) max_entries * sizeof(void *);
116 /* rely on vmalloc() to return page-aligned memory and
117 * ensure array->value is exactly page-aligned
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);
123 array_size += (u64) max_entries * elem_size;
127 /* allocate all map elements and zero-initialize them */
128 if (attr->map_flags & BPF_F_MMAPABLE) {
131 /* kmalloc'ed memory can't be mmap'ed, use explicit vmalloc */
132 data = bpf_map_area_mmapable_alloc(array_size, numa_node);
134 return ERR_PTR(-ENOMEM);
135 array = data + PAGE_ALIGN(sizeof(struct bpf_array))
136 - offsetof(struct bpf_array, value);
138 array = bpf_map_area_alloc(array_size, numa_node);
141 return ERR_PTR(-ENOMEM);
142 array->index_mask = index_mask;
143 array->map.bypass_spec_v1 = bypass_spec_v1;
145 /* copy mandatory map attributes */
146 bpf_map_init_from_attr(&array->map, attr);
147 array->elem_size = elem_size;
149 if (percpu && bpf_array_alloc_percpu(array)) {
150 bpf_map_area_free(array);
151 return ERR_PTR(-ENOMEM);
157 static void *array_map_elem_ptr(struct bpf_array* array, u32 index)
159 return array->value + (u64)array->elem_size * index;
162 /* Called from syscall or from eBPF program */
163 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
165 struct bpf_array *array = container_of(map, struct bpf_array, map);
166 u32 index = *(u32 *)key;
168 if (unlikely(index >= array->map.max_entries))
171 return array->value + (u64)array->elem_size * (index & array->index_mask);
174 static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm,
177 struct bpf_array *array = container_of(map, struct bpf_array, map);
179 if (map->max_entries != 1)
181 if (off >= map->value_size)
184 *imm = (unsigned long)array->value;
188 static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm,
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;
195 if (map->max_entries != 1)
197 if (imm < base || imm >= base + range)
204 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
205 static int array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
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;
214 if (map->map_flags & BPF_F_INNER_MAP)
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);
223 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
226 if (is_power_of_2(elem_size)) {
227 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
229 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
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;
237 /* Called from eBPF program */
238 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
240 struct bpf_array *array = container_of(map, struct bpf_array, map);
241 u32 index = *(u32 *)key;
243 if (unlikely(index >= array->map.max_entries))
246 return this_cpu_ptr(array->pptrs[index & array->index_mask]);
249 static void *percpu_array_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
251 struct bpf_array *array = container_of(map, struct bpf_array, map);
252 u32 index = *(u32 *)key;
254 if (cpu >= nr_cpu_ids)
257 if (unlikely(index >= array->map.max_entries))
260 return per_cpu_ptr(array->pptrs[index & array->index_mask], cpu);
263 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
265 struct bpf_array *array = container_of(map, struct bpf_array, map);
266 u32 index = *(u32 *)key;
271 if (unlikely(index >= array->map.max_entries))
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
278 size = array->elem_size;
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);
290 /* Called from syscall */
291 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
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;
297 if (index >= array->map.max_entries) {
302 if (index == array->map.max_entries - 1)
309 /* Called from syscall or from eBPF program */
310 static long array_map_update_elem(struct bpf_map *map, void *key, void *value,
313 struct bpf_array *array = container_of(map, struct bpf_array, map);
314 u32 index = *(u32 *)key;
317 if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
321 if (unlikely(index >= array->map.max_entries))
322 /* all elements were pre-allocated, cannot insert a new one */
325 if (unlikely(map_flags & BPF_NOEXIST))
326 /* all elements already exist */
329 if (unlikely((map_flags & BPF_F_LOCK) &&
330 !btf_record_has_field(map->record, BPF_SPIN_LOCK)))
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);
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);
343 copy_map_value(map, val, value);
344 bpf_obj_free_fields(array->map.record, val);
349 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
352 struct bpf_array *array = container_of(map, struct bpf_array, map);
353 u32 index = *(u32 *)key;
358 if (unlikely(map_flags > BPF_EXIST))
362 if (unlikely(index >= array->map.max_entries))
363 /* all elements were pre-allocated, cannot insert a new one */
366 if (unlikely(map_flags == BPF_NOEXIST))
367 /* all elements already exist */
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
376 size = array->elem_size;
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));
388 /* Called from syscall or from eBPF program */
389 static long array_map_delete_elem(struct bpf_map *map, void *key)
394 static void *array_map_vmalloc_addr(struct bpf_array *array)
396 return (void *)round_down((unsigned long)array, PAGE_SIZE);
399 static void array_map_free_timers(struct bpf_map *map)
401 struct bpf_array *array = container_of(map, struct bpf_array, map);
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))
408 for (i = 0; i < array->map.max_entries; i++)
409 bpf_obj_free_timer(map->record, array_map_elem_ptr(array, i));
412 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
413 static void array_map_free(struct bpf_map *map)
415 struct bpf_array *array = container_of(map, struct bpf_array, map);
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];
424 for_each_possible_cpu(cpu) {
425 bpf_obj_free_fields(map->record, per_cpu_ptr(pptr, cpu));
430 for (i = 0; i < array->map.max_entries; i++)
431 bpf_obj_free_fields(map->record, array_map_elem_ptr(array, i));
435 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
436 bpf_array_free_percpu(array);
438 if (array->map.map_flags & BPF_F_MMAPABLE)
439 bpf_map_area_free(array_map_vmalloc_addr(array));
441 bpf_map_area_free(array);
444 static void array_map_seq_show_elem(struct bpf_map *map, void *key,
451 value = array_map_lookup_elem(map, key);
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);
465 static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
468 struct bpf_array *array = container_of(map, struct bpf_array, map);
469 u32 index = *(u32 *)key;
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);
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)
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)
501 if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC)
507 if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT)
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.
514 if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data))
520 static int array_map_mmap(struct bpf_map *map, struct vm_area_struct *vma)
522 struct bpf_array *array = container_of(map, struct bpf_array, map);
523 pgoff_t pgoff = PAGE_ALIGN(sizeof(*array)) >> PAGE_SHIFT;
525 if (!(map->map_flags & BPF_F_MMAPABLE))
528 if (vma->vm_pgoff * PAGE_SIZE + (vma->vm_end - vma->vm_start) >
529 PAGE_ALIGN((u64)array->map.max_entries * array->elem_size))
532 return remap_vmalloc_range(vma, array_map_vmalloc_addr(array),
533 vma->vm_pgoff + pgoff);
536 static bool array_map_meta_equal(const struct bpf_map *meta0,
537 const struct bpf_map *meta1)
539 if (!bpf_map_meta_equal(meta0, meta1))
541 return meta0->map_flags & BPF_F_INNER_MAP ? true :
542 meta0->max_entries == meta1->max_entries;
545 struct bpf_iter_seq_array_map_info {
547 void *percpu_value_buf;
551 static void *bpf_array_map_seq_start(struct seq_file *seq, loff_t *pos)
553 struct bpf_iter_seq_array_map_info *info = seq->private;
554 struct bpf_map *map = info->map;
555 struct bpf_array *array;
558 if (info->index >= map->max_entries)
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);
570 static void *bpf_array_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
572 struct bpf_iter_seq_array_map_info *info = seq->private;
573 struct bpf_map *map = info->map;
574 struct bpf_array *array;
579 if (info->index >= map->max_entries)
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);
589 static int __bpf_array_map_seq_show(struct seq_file *seq, void *v)
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;
602 prog = bpf_iter_get_info(&meta, v == NULL);
609 ctx.key = &info->index;
611 if (!info->percpu_value_buf) {
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);
622 ctx.value = info->percpu_value_buf;
626 return bpf_iter_run_prog(prog, &ctx);
629 static int bpf_array_map_seq_show(struct seq_file *seq, void *v)
631 return __bpf_array_map_seq_show(seq, v);
634 static void bpf_array_map_seq_stop(struct seq_file *seq, void *v)
637 (void)__bpf_array_map_seq_show(seq, NULL);
640 static int bpf_iter_init_array_map(void *priv_data,
641 struct bpf_iter_aux_info *aux)
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);
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);
655 seq_info->percpu_value_buf = value_buf;
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.
662 bpf_map_inc_with_uref(map);
667 static void bpf_iter_fini_array_map(void *priv_data)
669 struct bpf_iter_seq_array_map_info *seq_info = priv_data;
671 bpf_map_put_with_uref(seq_info->map);
672 kfree(seq_info->percpu_value_buf);
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,
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),
689 static long bpf_for_each_array_elem(struct bpf_map *map, bpf_callback_t callback_fn,
690 void *callback_ctx, u64 flags)
692 u32 i, key, num_elems = 0;
693 struct bpf_array *array;
701 is_percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
702 array = container_of(map, struct bpf_array, map);
705 for (i = 0; i < map->max_entries; i++) {
707 val = this_cpu_ptr(array->pptrs[i]);
709 val = array_map_elem_ptr(array, 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 */
724 static u64 array_map_mem_usage(const struct bpf_map *map)
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);
733 usage += entries * sizeof(void *);
734 usage += entries * elem_size * num_possible_cpus();
736 if (map->map_flags & BPF_F_MMAPABLE) {
737 usage = PAGE_ALIGN(usage);
738 usage += PAGE_ALIGN(entries * elem_size);
740 usage += entries * elem_size;
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,
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,
793 static int fd_array_map_alloc_check(union bpf_attr *attr)
795 /* only file descriptors can be stored in this type of map */
796 if (attr->value_size != sizeof(u32))
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))
801 return array_map_alloc_check(attr);
804 static void fd_array_map_free(struct bpf_map *map)
806 struct bpf_array *array = container_of(map, struct bpf_array, map);
809 /* make sure it's empty */
810 for (i = 0; i < array->map.max_entries; i++)
811 BUG_ON(array->ptrs[i] != NULL);
813 bpf_map_area_free(array);
816 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
818 return ERR_PTR(-EOPNOTSUPP);
821 /* only called from syscall */
822 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
827 if (!map->ops->map_fd_sys_lookup_elem)
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);
841 /* only called from syscall */
842 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
843 void *key, void *value, u64 map_flags)
845 struct bpf_array *array = container_of(map, struct bpf_array, map);
846 void *new_ptr, *old_ptr;
847 u32 index = *(u32 *)key, ufd;
849 if (map_flags != BPF_ANY)
852 if (index >= array->map.max_entries)
856 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
858 return PTR_ERR(new_ptr);
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);
866 old_ptr = xchg(array->ptrs + index, new_ptr);
870 map->ops->map_fd_put_ptr(old_ptr);
874 static long fd_array_map_delete_elem(struct bpf_map *map, void *key)
876 struct bpf_array *array = container_of(map, struct bpf_array, map);
878 u32 index = *(u32 *)key;
880 if (index >= array->map.max_entries)
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);
889 old_ptr = xchg(array->ptrs + index, NULL);
893 map->ops->map_fd_put_ptr(old_ptr);
900 static void *prog_fd_array_get_ptr(struct bpf_map *map,
901 struct file *map_file, int fd)
903 struct bpf_prog *prog = bpf_prog_get(fd);
908 if (!bpf_prog_map_compatible(map, prog)) {
910 return ERR_PTR(-EINVAL);
916 static void prog_fd_array_put_ptr(void *ptr)
921 static u32 prog_fd_array_sys_lookup_elem(void *ptr)
923 return ((struct bpf_prog *)ptr)->aux->id;
926 /* decrement refcnt of all bpf_progs that are stored in this map */
927 static void bpf_fd_array_map_clear(struct bpf_map *map)
929 struct bpf_array *array = container_of(map, struct bpf_array, map);
932 for (i = 0; i < array->map.max_entries; i++)
933 fd_array_map_delete_elem(map, &i);
936 static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
944 elem = array_map_lookup_elem(map, key);
946 ptr = READ_ONCE(*elem);
948 seq_printf(m, "%u: ", *(u32 *)key);
949 prog_id = prog_fd_array_sys_lookup_elem(ptr);
950 btf_type_seq_show(map->btf, map->btf_value_type_id,
959 struct prog_poke_elem {
960 struct list_head list;
961 struct bpf_prog_aux *aux;
964 static int prog_array_map_poke_track(struct bpf_map *map,
965 struct bpf_prog_aux *prog_aux)
967 struct prog_poke_elem *elem;
968 struct bpf_array_aux *aux;
971 aux = container_of(map, struct bpf_array, map)->aux;
972 mutex_lock(&aux->poke_mutex);
973 list_for_each_entry(elem, &aux->poke_progs, list) {
974 if (elem->aux == prog_aux)
978 elem = kmalloc(sizeof(*elem), GFP_KERNEL);
984 INIT_LIST_HEAD(&elem->list);
985 /* We must track the program's aux info at this point in time
986 * since the program pointer itself may not be stable yet, see
987 * also comment in prog_array_map_poke_run().
989 elem->aux = prog_aux;
991 list_add_tail(&elem->list, &aux->poke_progs);
993 mutex_unlock(&aux->poke_mutex);
997 static void prog_array_map_poke_untrack(struct bpf_map *map,
998 struct bpf_prog_aux *prog_aux)
1000 struct prog_poke_elem *elem, *tmp;
1001 struct bpf_array_aux *aux;
1003 aux = container_of(map, struct bpf_array, map)->aux;
1004 mutex_lock(&aux->poke_mutex);
1005 list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1006 if (elem->aux == prog_aux) {
1007 list_del_init(&elem->list);
1012 mutex_unlock(&aux->poke_mutex);
1015 static void prog_array_map_poke_run(struct bpf_map *map, u32 key,
1016 struct bpf_prog *old,
1017 struct bpf_prog *new)
1019 u8 *old_addr, *new_addr, *old_bypass_addr;
1020 struct prog_poke_elem *elem;
1021 struct bpf_array_aux *aux;
1023 aux = container_of(map, struct bpf_array, map)->aux;
1024 WARN_ON_ONCE(!mutex_is_locked(&aux->poke_mutex));
1026 list_for_each_entry(elem, &aux->poke_progs, list) {
1027 struct bpf_jit_poke_descriptor *poke;
1030 for (i = 0; i < elem->aux->size_poke_tab; i++) {
1031 poke = &elem->aux->poke_tab[i];
1033 /* Few things to be aware of:
1035 * 1) We can only ever access aux in this context, but
1036 * not aux->prog since it might not be stable yet and
1037 * there could be danger of use after free otherwise.
1038 * 2) Initially when we start tracking aux, the program
1039 * is not JITed yet and also does not have a kallsyms
1040 * entry. We skip these as poke->tailcall_target_stable
1041 * is not active yet. The JIT will do the final fixup
1042 * before setting it stable. The various
1043 * poke->tailcall_target_stable are successively
1044 * activated, so tail call updates can arrive from here
1045 * while JIT is still finishing its final fixup for
1046 * non-activated poke entries.
1047 * 3) On program teardown, the program's kallsym entry gets
1048 * removed out of RCU callback, but we can only untrack
1049 * from sleepable context, therefore bpf_arch_text_poke()
1050 * might not see that this is in BPF text section and
1051 * bails out with -EINVAL. As these are unreachable since
1052 * RCU grace period already passed, we simply skip them.
1053 * 4) Also programs reaching refcount of zero while patching
1054 * is in progress is okay since we're protected under
1055 * poke_mutex and untrack the programs before the JIT
1056 * buffer is freed. When we're still in the middle of
1057 * patching and suddenly kallsyms entry of the program
1058 * gets evicted, we just skip the rest which is fine due
1060 * 5) Any other error happening below from bpf_arch_text_poke()
1061 * is a unexpected bug.
1063 if (!READ_ONCE(poke->tailcall_target_stable))
1065 if (poke->reason != BPF_POKE_REASON_TAIL_CALL)
1067 if (poke->tail_call.map != map ||
1068 poke->tail_call.key != key)
1071 old_bypass_addr = old ? NULL : poke->bypass_addr;
1072 old_addr = old ? (u8 *)old->bpf_func + poke->adj_off : NULL;
1073 new_addr = new ? (u8 *)new->bpf_func + poke->adj_off : NULL;
1076 ret = bpf_arch_text_poke(poke->tailcall_target,
1078 old_addr, new_addr);
1079 BUG_ON(ret < 0 && ret != -EINVAL);
1081 ret = bpf_arch_text_poke(poke->tailcall_bypass,
1085 BUG_ON(ret < 0 && ret != -EINVAL);
1088 ret = bpf_arch_text_poke(poke->tailcall_bypass,
1092 BUG_ON(ret < 0 && ret != -EINVAL);
1093 /* let other CPUs finish the execution of program
1094 * so that it will not possible to expose them
1095 * to invalid nop, stack unwind, nop state
1099 ret = bpf_arch_text_poke(poke->tailcall_target,
1102 BUG_ON(ret < 0 && ret != -EINVAL);
1108 static void prog_array_map_clear_deferred(struct work_struct *work)
1110 struct bpf_map *map = container_of(work, struct bpf_array_aux,
1112 bpf_fd_array_map_clear(map);
1116 static void prog_array_map_clear(struct bpf_map *map)
1118 struct bpf_array_aux *aux = container_of(map, struct bpf_array,
1121 schedule_work(&aux->work);
1124 static struct bpf_map *prog_array_map_alloc(union bpf_attr *attr)
1126 struct bpf_array_aux *aux;
1127 struct bpf_map *map;
1129 aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT);
1131 return ERR_PTR(-ENOMEM);
1133 INIT_WORK(&aux->work, prog_array_map_clear_deferred);
1134 INIT_LIST_HEAD(&aux->poke_progs);
1135 mutex_init(&aux->poke_mutex);
1137 map = array_map_alloc(attr);
1143 container_of(map, struct bpf_array, map)->aux = aux;
1149 static void prog_array_map_free(struct bpf_map *map)
1151 struct prog_poke_elem *elem, *tmp;
1152 struct bpf_array_aux *aux;
1154 aux = container_of(map, struct bpf_array, map)->aux;
1155 list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1156 list_del_init(&elem->list);
1160 fd_array_map_free(map);
1163 /* prog_array->aux->{type,jited} is a runtime binding.
1164 * Doing static check alone in the verifier is not enough.
1165 * Thus, prog_array_map cannot be used as an inner_map
1166 * and map_meta_equal is not implemented.
1168 const struct bpf_map_ops prog_array_map_ops = {
1169 .map_alloc_check = fd_array_map_alloc_check,
1170 .map_alloc = prog_array_map_alloc,
1171 .map_free = prog_array_map_free,
1172 .map_poke_track = prog_array_map_poke_track,
1173 .map_poke_untrack = prog_array_map_poke_untrack,
1174 .map_poke_run = prog_array_map_poke_run,
1175 .map_get_next_key = array_map_get_next_key,
1176 .map_lookup_elem = fd_array_map_lookup_elem,
1177 .map_delete_elem = fd_array_map_delete_elem,
1178 .map_fd_get_ptr = prog_fd_array_get_ptr,
1179 .map_fd_put_ptr = prog_fd_array_put_ptr,
1180 .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
1181 .map_release_uref = prog_array_map_clear,
1182 .map_seq_show_elem = prog_array_map_seq_show_elem,
1183 .map_mem_usage = array_map_mem_usage,
1184 .map_btf_id = &array_map_btf_ids[0],
1187 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
1188 struct file *map_file)
1190 struct bpf_event_entry *ee;
1192 ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
1194 ee->event = perf_file->private_data;
1195 ee->perf_file = perf_file;
1196 ee->map_file = map_file;
1202 static void __bpf_event_entry_free(struct rcu_head *rcu)
1204 struct bpf_event_entry *ee;
1206 ee = container_of(rcu, struct bpf_event_entry, rcu);
1207 fput(ee->perf_file);
1211 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
1213 call_rcu(&ee->rcu, __bpf_event_entry_free);
1216 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
1217 struct file *map_file, int fd)
1219 struct bpf_event_entry *ee;
1220 struct perf_event *event;
1221 struct file *perf_file;
1224 perf_file = perf_event_get(fd);
1225 if (IS_ERR(perf_file))
1228 ee = ERR_PTR(-EOPNOTSUPP);
1229 event = perf_file->private_data;
1230 if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
1233 ee = bpf_event_entry_gen(perf_file, map_file);
1236 ee = ERR_PTR(-ENOMEM);
1242 static void perf_event_fd_array_put_ptr(void *ptr)
1244 bpf_event_entry_free_rcu(ptr);
1247 static void perf_event_fd_array_release(struct bpf_map *map,
1248 struct file *map_file)
1250 struct bpf_array *array = container_of(map, struct bpf_array, map);
1251 struct bpf_event_entry *ee;
1254 if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1258 for (i = 0; i < array->map.max_entries; i++) {
1259 ee = READ_ONCE(array->ptrs[i]);
1260 if (ee && ee->map_file == map_file)
1261 fd_array_map_delete_elem(map, &i);
1266 static void perf_event_fd_array_map_free(struct bpf_map *map)
1268 if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1269 bpf_fd_array_map_clear(map);
1270 fd_array_map_free(map);
1273 const struct bpf_map_ops perf_event_array_map_ops = {
1274 .map_meta_equal = bpf_map_meta_equal,
1275 .map_alloc_check = fd_array_map_alloc_check,
1276 .map_alloc = array_map_alloc,
1277 .map_free = perf_event_fd_array_map_free,
1278 .map_get_next_key = array_map_get_next_key,
1279 .map_lookup_elem = fd_array_map_lookup_elem,
1280 .map_delete_elem = fd_array_map_delete_elem,
1281 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
1282 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
1283 .map_release = perf_event_fd_array_release,
1284 .map_check_btf = map_check_no_btf,
1285 .map_mem_usage = array_map_mem_usage,
1286 .map_btf_id = &array_map_btf_ids[0],
1289 #ifdef CONFIG_CGROUPS
1290 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
1291 struct file *map_file /* not used */,
1294 return cgroup_get_from_fd(fd);
1297 static void cgroup_fd_array_put_ptr(void *ptr)
1299 /* cgroup_put free cgrp after a rcu grace period */
1303 static void cgroup_fd_array_free(struct bpf_map *map)
1305 bpf_fd_array_map_clear(map);
1306 fd_array_map_free(map);
1309 const struct bpf_map_ops cgroup_array_map_ops = {
1310 .map_meta_equal = bpf_map_meta_equal,
1311 .map_alloc_check = fd_array_map_alloc_check,
1312 .map_alloc = array_map_alloc,
1313 .map_free = cgroup_fd_array_free,
1314 .map_get_next_key = array_map_get_next_key,
1315 .map_lookup_elem = fd_array_map_lookup_elem,
1316 .map_delete_elem = fd_array_map_delete_elem,
1317 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
1318 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
1319 .map_check_btf = map_check_no_btf,
1320 .map_mem_usage = array_map_mem_usage,
1321 .map_btf_id = &array_map_btf_ids[0],
1325 static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
1327 struct bpf_map *map, *inner_map_meta;
1329 inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
1330 if (IS_ERR(inner_map_meta))
1331 return inner_map_meta;
1333 map = array_map_alloc(attr);
1335 bpf_map_meta_free(inner_map_meta);
1339 map->inner_map_meta = inner_map_meta;
1344 static void array_of_map_free(struct bpf_map *map)
1346 /* map->inner_map_meta is only accessed by syscall which
1347 * is protected by fdget/fdput.
1349 bpf_map_meta_free(map->inner_map_meta);
1350 bpf_fd_array_map_clear(map);
1351 fd_array_map_free(map);
1354 static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
1356 struct bpf_map **inner_map = array_map_lookup_elem(map, key);
1361 return READ_ONCE(*inner_map);
1364 static int array_of_map_gen_lookup(struct bpf_map *map,
1365 struct bpf_insn *insn_buf)
1367 struct bpf_array *array = container_of(map, struct bpf_array, map);
1368 u32 elem_size = array->elem_size;
1369 struct bpf_insn *insn = insn_buf;
1370 const int ret = BPF_REG_0;
1371 const int map_ptr = BPF_REG_1;
1372 const int index = BPF_REG_2;
1374 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
1375 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
1376 if (!map->bypass_spec_v1) {
1377 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
1378 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
1380 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
1382 if (is_power_of_2(elem_size))
1383 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
1385 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
1386 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
1387 *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
1388 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
1389 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
1390 *insn++ = BPF_MOV64_IMM(ret, 0);
1392 return insn - insn_buf;
1395 const struct bpf_map_ops array_of_maps_map_ops = {
1396 .map_alloc_check = fd_array_map_alloc_check,
1397 .map_alloc = array_of_map_alloc,
1398 .map_free = array_of_map_free,
1399 .map_get_next_key = array_map_get_next_key,
1400 .map_lookup_elem = array_of_map_lookup_elem,
1401 .map_delete_elem = fd_array_map_delete_elem,
1402 .map_fd_get_ptr = bpf_map_fd_get_ptr,
1403 .map_fd_put_ptr = bpf_map_fd_put_ptr,
1404 .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
1405 .map_gen_lookup = array_of_map_gen_lookup,
1406 .map_lookup_batch = generic_map_lookup_batch,
1407 .map_update_batch = generic_map_update_batch,
1408 .map_check_btf = map_check_no_btf,
1409 .map_mem_usage = array_map_mem_usage,
1410 .map_btf_id = &array_map_btf_ids[0],