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(NULL);
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 /* emit BPF instructions equivalent to C code of percpu_array_map_lookup_elem() */
250 static int percpu_array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
252 struct bpf_array *array = container_of(map, struct bpf_array, map);
253 struct bpf_insn *insn = insn_buf;
255 if (!bpf_jit_supports_percpu_insn())
258 if (map->map_flags & BPF_F_INNER_MAP)
261 BUILD_BUG_ON(offsetof(struct bpf_array, map) != 0);
262 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, offsetof(struct bpf_array, pptrs));
264 *insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_0, BPF_REG_2, 0);
265 if (!map->bypass_spec_v1) {
266 *insn++ = BPF_JMP_IMM(BPF_JGE, BPF_REG_0, map->max_entries, 6);
267 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_0, array->index_mask);
269 *insn++ = BPF_JMP_IMM(BPF_JGE, BPF_REG_0, map->max_entries, 5);
272 *insn++ = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3);
273 *insn++ = BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1);
274 *insn++ = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_0, 0);
275 *insn++ = BPF_MOV64_PERCPU_REG(BPF_REG_0, BPF_REG_0);
276 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
277 *insn++ = BPF_MOV64_IMM(BPF_REG_0, 0);
278 return insn - insn_buf;
281 static void *percpu_array_map_lookup_percpu_elem(struct bpf_map *map, void *key, u32 cpu)
283 struct bpf_array *array = container_of(map, struct bpf_array, map);
284 u32 index = *(u32 *)key;
286 if (cpu >= nr_cpu_ids)
289 if (unlikely(index >= array->map.max_entries))
292 return per_cpu_ptr(array->pptrs[index & array->index_mask], cpu);
295 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
297 struct bpf_array *array = container_of(map, struct bpf_array, map);
298 u32 index = *(u32 *)key;
303 if (unlikely(index >= array->map.max_entries))
306 /* per_cpu areas are zero-filled and bpf programs can only
307 * access 'value_size' of them, so copying rounded areas
308 * will not leak any kernel data
310 size = array->elem_size;
312 pptr = array->pptrs[index & array->index_mask];
313 for_each_possible_cpu(cpu) {
314 copy_map_value_long(map, value + off, per_cpu_ptr(pptr, cpu));
315 check_and_init_map_value(map, value + off);
322 /* Called from syscall */
323 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
325 struct bpf_array *array = container_of(map, struct bpf_array, map);
326 u32 index = key ? *(u32 *)key : U32_MAX;
327 u32 *next = (u32 *)next_key;
329 if (index >= array->map.max_entries) {
334 if (index == array->map.max_entries - 1)
341 /* Called from syscall or from eBPF program */
342 static long array_map_update_elem(struct bpf_map *map, void *key, void *value,
345 struct bpf_array *array = container_of(map, struct bpf_array, map);
346 u32 index = *(u32 *)key;
349 if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
353 if (unlikely(index >= array->map.max_entries))
354 /* all elements were pre-allocated, cannot insert a new one */
357 if (unlikely(map_flags & BPF_NOEXIST))
358 /* all elements already exist */
361 if (unlikely((map_flags & BPF_F_LOCK) &&
362 !btf_record_has_field(map->record, BPF_SPIN_LOCK)))
365 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
366 val = this_cpu_ptr(array->pptrs[index & array->index_mask]);
367 copy_map_value(map, val, value);
368 bpf_obj_free_fields(array->map.record, val);
371 (u64)array->elem_size * (index & array->index_mask);
372 if (map_flags & BPF_F_LOCK)
373 copy_map_value_locked(map, val, value, false);
375 copy_map_value(map, val, value);
376 bpf_obj_free_fields(array->map.record, val);
381 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
384 struct bpf_array *array = container_of(map, struct bpf_array, map);
385 u32 index = *(u32 *)key;
390 if (unlikely(map_flags > BPF_EXIST))
394 if (unlikely(index >= array->map.max_entries))
395 /* all elements were pre-allocated, cannot insert a new one */
398 if (unlikely(map_flags == BPF_NOEXIST))
399 /* all elements already exist */
402 /* the user space will provide round_up(value_size, 8) bytes that
403 * will be copied into per-cpu area. bpf programs can only access
404 * value_size of it. During lookup the same extra bytes will be
405 * returned or zeros which were zero-filled by percpu_alloc,
406 * so no kernel data leaks possible
408 size = array->elem_size;
410 pptr = array->pptrs[index & array->index_mask];
411 for_each_possible_cpu(cpu) {
412 copy_map_value_long(map, per_cpu_ptr(pptr, cpu), value + off);
413 bpf_obj_free_fields(array->map.record, per_cpu_ptr(pptr, cpu));
420 /* Called from syscall or from eBPF program */
421 static long array_map_delete_elem(struct bpf_map *map, void *key)
426 static void *array_map_vmalloc_addr(struct bpf_array *array)
428 return (void *)round_down((unsigned long)array, PAGE_SIZE);
431 static void array_map_free_timers_wq(struct bpf_map *map)
433 struct bpf_array *array = container_of(map, struct bpf_array, map);
436 /* We don't reset or free fields other than timer and workqueue
437 * on uref dropping to zero.
439 if (btf_record_has_field(map->record, BPF_TIMER | BPF_WORKQUEUE)) {
440 for (i = 0; i < array->map.max_entries; i++) {
441 if (btf_record_has_field(map->record, BPF_TIMER))
442 bpf_obj_free_timer(map->record, array_map_elem_ptr(array, i));
443 if (btf_record_has_field(map->record, BPF_WORKQUEUE))
444 bpf_obj_free_workqueue(map->record, array_map_elem_ptr(array, i));
449 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
450 static void array_map_free(struct bpf_map *map)
452 struct bpf_array *array = container_of(map, struct bpf_array, map);
455 if (!IS_ERR_OR_NULL(map->record)) {
456 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
457 for (i = 0; i < array->map.max_entries; i++) {
458 void __percpu *pptr = array->pptrs[i & array->index_mask];
461 for_each_possible_cpu(cpu) {
462 bpf_obj_free_fields(map->record, per_cpu_ptr(pptr, cpu));
467 for (i = 0; i < array->map.max_entries; i++)
468 bpf_obj_free_fields(map->record, array_map_elem_ptr(array, i));
472 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
473 bpf_array_free_percpu(array);
475 if (array->map.map_flags & BPF_F_MMAPABLE)
476 bpf_map_area_free(array_map_vmalloc_addr(array));
478 bpf_map_area_free(array);
481 static void array_map_seq_show_elem(struct bpf_map *map, void *key,
488 value = array_map_lookup_elem(map, key);
494 if (map->btf_key_type_id)
495 seq_printf(m, "%u: ", *(u32 *)key);
496 btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
502 static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
505 struct bpf_array *array = container_of(map, struct bpf_array, map);
506 u32 index = *(u32 *)key;
512 seq_printf(m, "%u: {\n", *(u32 *)key);
513 pptr = array->pptrs[index & array->index_mask];
514 for_each_possible_cpu(cpu) {
515 seq_printf(m, "\tcpu%d: ", cpu);
516 btf_type_seq_show(map->btf, map->btf_value_type_id,
517 per_cpu_ptr(pptr, cpu), m);
525 static int array_map_check_btf(const struct bpf_map *map,
526 const struct btf *btf,
527 const struct btf_type *key_type,
528 const struct btf_type *value_type)
532 /* One exception for keyless BTF: .bss/.data/.rodata map */
533 if (btf_type_is_void(key_type)) {
534 if (map->map_type != BPF_MAP_TYPE_ARRAY ||
535 map->max_entries != 1)
538 if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC)
544 if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT)
547 int_data = *(u32 *)(key_type + 1);
548 /* bpf array can only take a u32 key. This check makes sure
549 * that the btf matches the attr used during map_create.
551 if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data))
557 static int array_map_mmap(struct bpf_map *map, struct vm_area_struct *vma)
559 struct bpf_array *array = container_of(map, struct bpf_array, map);
560 pgoff_t pgoff = PAGE_ALIGN(sizeof(*array)) >> PAGE_SHIFT;
562 if (!(map->map_flags & BPF_F_MMAPABLE))
565 if (vma->vm_pgoff * PAGE_SIZE + (vma->vm_end - vma->vm_start) >
566 PAGE_ALIGN((u64)array->map.max_entries * array->elem_size))
569 return remap_vmalloc_range(vma, array_map_vmalloc_addr(array),
570 vma->vm_pgoff + pgoff);
573 static bool array_map_meta_equal(const struct bpf_map *meta0,
574 const struct bpf_map *meta1)
576 if (!bpf_map_meta_equal(meta0, meta1))
578 return meta0->map_flags & BPF_F_INNER_MAP ? true :
579 meta0->max_entries == meta1->max_entries;
582 struct bpf_iter_seq_array_map_info {
584 void *percpu_value_buf;
588 static void *bpf_array_map_seq_start(struct seq_file *seq, loff_t *pos)
590 struct bpf_iter_seq_array_map_info *info = seq->private;
591 struct bpf_map *map = info->map;
592 struct bpf_array *array;
595 if (info->index >= map->max_entries)
600 array = container_of(map, struct bpf_array, map);
601 index = info->index & array->index_mask;
602 if (info->percpu_value_buf)
603 return array->pptrs[index];
604 return array_map_elem_ptr(array, index);
607 static void *bpf_array_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
609 struct bpf_iter_seq_array_map_info *info = seq->private;
610 struct bpf_map *map = info->map;
611 struct bpf_array *array;
616 if (info->index >= map->max_entries)
619 array = container_of(map, struct bpf_array, map);
620 index = info->index & array->index_mask;
621 if (info->percpu_value_buf)
622 return array->pptrs[index];
623 return array_map_elem_ptr(array, index);
626 static int __bpf_array_map_seq_show(struct seq_file *seq, void *v)
628 struct bpf_iter_seq_array_map_info *info = seq->private;
629 struct bpf_iter__bpf_map_elem ctx = {};
630 struct bpf_map *map = info->map;
631 struct bpf_array *array = container_of(map, struct bpf_array, map);
632 struct bpf_iter_meta meta;
633 struct bpf_prog *prog;
634 int off = 0, cpu = 0;
635 void __percpu **pptr;
639 prog = bpf_iter_get_info(&meta, v == NULL);
646 ctx.key = &info->index;
648 if (!info->percpu_value_buf) {
652 size = array->elem_size;
653 for_each_possible_cpu(cpu) {
654 copy_map_value_long(map, info->percpu_value_buf + off,
655 per_cpu_ptr(pptr, cpu));
656 check_and_init_map_value(map, info->percpu_value_buf + off);
659 ctx.value = info->percpu_value_buf;
663 return bpf_iter_run_prog(prog, &ctx);
666 static int bpf_array_map_seq_show(struct seq_file *seq, void *v)
668 return __bpf_array_map_seq_show(seq, v);
671 static void bpf_array_map_seq_stop(struct seq_file *seq, void *v)
674 (void)__bpf_array_map_seq_show(seq, NULL);
677 static int bpf_iter_init_array_map(void *priv_data,
678 struct bpf_iter_aux_info *aux)
680 struct bpf_iter_seq_array_map_info *seq_info = priv_data;
681 struct bpf_map *map = aux->map;
682 struct bpf_array *array = container_of(map, struct bpf_array, map);
686 if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
687 buf_size = array->elem_size * num_possible_cpus();
688 value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
692 seq_info->percpu_value_buf = value_buf;
695 /* bpf_iter_attach_map() acquires a map uref, and the uref may be
696 * released before or in the middle of iterating map elements, so
697 * acquire an extra map uref for iterator.
699 bpf_map_inc_with_uref(map);
704 static void bpf_iter_fini_array_map(void *priv_data)
706 struct bpf_iter_seq_array_map_info *seq_info = priv_data;
708 bpf_map_put_with_uref(seq_info->map);
709 kfree(seq_info->percpu_value_buf);
712 static const struct seq_operations bpf_array_map_seq_ops = {
713 .start = bpf_array_map_seq_start,
714 .next = bpf_array_map_seq_next,
715 .stop = bpf_array_map_seq_stop,
716 .show = bpf_array_map_seq_show,
719 static const struct bpf_iter_seq_info iter_seq_info = {
720 .seq_ops = &bpf_array_map_seq_ops,
721 .init_seq_private = bpf_iter_init_array_map,
722 .fini_seq_private = bpf_iter_fini_array_map,
723 .seq_priv_size = sizeof(struct bpf_iter_seq_array_map_info),
726 static long bpf_for_each_array_elem(struct bpf_map *map, bpf_callback_t callback_fn,
727 void *callback_ctx, u64 flags)
729 u32 i, key, num_elems = 0;
730 struct bpf_array *array;
738 is_percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
739 array = container_of(map, struct bpf_array, map);
742 for (i = 0; i < map->max_entries; i++) {
744 val = this_cpu_ptr(array->pptrs[i]);
746 val = array_map_elem_ptr(array, i);
749 ret = callback_fn((u64)(long)map, (u64)(long)&key,
750 (u64)(long)val, (u64)(long)callback_ctx, 0);
751 /* return value: 0 - continue, 1 - stop and return */
761 static u64 array_map_mem_usage(const struct bpf_map *map)
763 struct bpf_array *array = container_of(map, struct bpf_array, map);
764 bool percpu = map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
765 u32 elem_size = array->elem_size;
766 u64 entries = map->max_entries;
767 u64 usage = sizeof(*array);
770 usage += entries * sizeof(void *);
771 usage += entries * elem_size * num_possible_cpus();
773 if (map->map_flags & BPF_F_MMAPABLE) {
774 usage = PAGE_ALIGN(usage);
775 usage += PAGE_ALIGN(entries * elem_size);
777 usage += entries * elem_size;
783 BTF_ID_LIST_SINGLE(array_map_btf_ids, struct, bpf_array)
784 const struct bpf_map_ops array_map_ops = {
785 .map_meta_equal = array_map_meta_equal,
786 .map_alloc_check = array_map_alloc_check,
787 .map_alloc = array_map_alloc,
788 .map_free = array_map_free,
789 .map_get_next_key = array_map_get_next_key,
790 .map_release_uref = array_map_free_timers_wq,
791 .map_lookup_elem = array_map_lookup_elem,
792 .map_update_elem = array_map_update_elem,
793 .map_delete_elem = array_map_delete_elem,
794 .map_gen_lookup = array_map_gen_lookup,
795 .map_direct_value_addr = array_map_direct_value_addr,
796 .map_direct_value_meta = array_map_direct_value_meta,
797 .map_mmap = array_map_mmap,
798 .map_seq_show_elem = array_map_seq_show_elem,
799 .map_check_btf = array_map_check_btf,
800 .map_lookup_batch = generic_map_lookup_batch,
801 .map_update_batch = generic_map_update_batch,
802 .map_set_for_each_callback_args = map_set_for_each_callback_args,
803 .map_for_each_callback = bpf_for_each_array_elem,
804 .map_mem_usage = array_map_mem_usage,
805 .map_btf_id = &array_map_btf_ids[0],
806 .iter_seq_info = &iter_seq_info,
809 const struct bpf_map_ops percpu_array_map_ops = {
810 .map_meta_equal = bpf_map_meta_equal,
811 .map_alloc_check = array_map_alloc_check,
812 .map_alloc = array_map_alloc,
813 .map_free = array_map_free,
814 .map_get_next_key = array_map_get_next_key,
815 .map_lookup_elem = percpu_array_map_lookup_elem,
816 .map_gen_lookup = percpu_array_map_gen_lookup,
817 .map_update_elem = array_map_update_elem,
818 .map_delete_elem = array_map_delete_elem,
819 .map_lookup_percpu_elem = percpu_array_map_lookup_percpu_elem,
820 .map_seq_show_elem = percpu_array_map_seq_show_elem,
821 .map_check_btf = array_map_check_btf,
822 .map_lookup_batch = generic_map_lookup_batch,
823 .map_update_batch = generic_map_update_batch,
824 .map_set_for_each_callback_args = map_set_for_each_callback_args,
825 .map_for_each_callback = bpf_for_each_array_elem,
826 .map_mem_usage = array_map_mem_usage,
827 .map_btf_id = &array_map_btf_ids[0],
828 .iter_seq_info = &iter_seq_info,
831 static int fd_array_map_alloc_check(union bpf_attr *attr)
833 /* only file descriptors can be stored in this type of map */
834 if (attr->value_size != sizeof(u32))
836 /* Program read-only/write-only not supported for special maps yet. */
837 if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG))
839 return array_map_alloc_check(attr);
842 static void fd_array_map_free(struct bpf_map *map)
844 struct bpf_array *array = container_of(map, struct bpf_array, map);
847 /* make sure it's empty */
848 for (i = 0; i < array->map.max_entries; i++)
849 BUG_ON(array->ptrs[i] != NULL);
851 bpf_map_area_free(array);
854 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
856 return ERR_PTR(-EOPNOTSUPP);
859 /* only called from syscall */
860 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
865 if (!map->ops->map_fd_sys_lookup_elem)
869 elem = array_map_lookup_elem(map, key);
870 if (elem && (ptr = READ_ONCE(*elem)))
871 *value = map->ops->map_fd_sys_lookup_elem(ptr);
879 /* only called from syscall */
880 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
881 void *key, void *value, u64 map_flags)
883 struct bpf_array *array = container_of(map, struct bpf_array, map);
884 void *new_ptr, *old_ptr;
885 u32 index = *(u32 *)key, ufd;
887 if (map_flags != BPF_ANY)
890 if (index >= array->map.max_entries)
894 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
896 return PTR_ERR(new_ptr);
898 if (map->ops->map_poke_run) {
899 mutex_lock(&array->aux->poke_mutex);
900 old_ptr = xchg(array->ptrs + index, new_ptr);
901 map->ops->map_poke_run(map, index, old_ptr, new_ptr);
902 mutex_unlock(&array->aux->poke_mutex);
904 old_ptr = xchg(array->ptrs + index, new_ptr);
908 map->ops->map_fd_put_ptr(map, old_ptr, true);
912 static long __fd_array_map_delete_elem(struct bpf_map *map, void *key, bool need_defer)
914 struct bpf_array *array = container_of(map, struct bpf_array, map);
916 u32 index = *(u32 *)key;
918 if (index >= array->map.max_entries)
921 if (map->ops->map_poke_run) {
922 mutex_lock(&array->aux->poke_mutex);
923 old_ptr = xchg(array->ptrs + index, NULL);
924 map->ops->map_poke_run(map, index, old_ptr, NULL);
925 mutex_unlock(&array->aux->poke_mutex);
927 old_ptr = xchg(array->ptrs + index, NULL);
931 map->ops->map_fd_put_ptr(map, old_ptr, need_defer);
938 static long fd_array_map_delete_elem(struct bpf_map *map, void *key)
940 return __fd_array_map_delete_elem(map, key, true);
943 static void *prog_fd_array_get_ptr(struct bpf_map *map,
944 struct file *map_file, int fd)
946 struct bpf_prog *prog = bpf_prog_get(fd);
951 if (!bpf_prog_map_compatible(map, prog)) {
953 return ERR_PTR(-EINVAL);
959 static void prog_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
961 /* bpf_prog is freed after one RCU or tasks trace grace period */
965 static u32 prog_fd_array_sys_lookup_elem(void *ptr)
967 return ((struct bpf_prog *)ptr)->aux->id;
970 /* decrement refcnt of all bpf_progs that are stored in this map */
971 static void bpf_fd_array_map_clear(struct bpf_map *map, bool need_defer)
973 struct bpf_array *array = container_of(map, struct bpf_array, map);
976 for (i = 0; i < array->map.max_entries; i++)
977 __fd_array_map_delete_elem(map, &i, need_defer);
980 static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
988 elem = array_map_lookup_elem(map, key);
990 ptr = READ_ONCE(*elem);
992 seq_printf(m, "%u: ", *(u32 *)key);
993 prog_id = prog_fd_array_sys_lookup_elem(ptr);
994 btf_type_seq_show(map->btf, map->btf_value_type_id,
1003 struct prog_poke_elem {
1004 struct list_head list;
1005 struct bpf_prog_aux *aux;
1008 static int prog_array_map_poke_track(struct bpf_map *map,
1009 struct bpf_prog_aux *prog_aux)
1011 struct prog_poke_elem *elem;
1012 struct bpf_array_aux *aux;
1015 aux = container_of(map, struct bpf_array, map)->aux;
1016 mutex_lock(&aux->poke_mutex);
1017 list_for_each_entry(elem, &aux->poke_progs, list) {
1018 if (elem->aux == prog_aux)
1022 elem = kmalloc(sizeof(*elem), GFP_KERNEL);
1028 INIT_LIST_HEAD(&elem->list);
1029 /* We must track the program's aux info at this point in time
1030 * since the program pointer itself may not be stable yet, see
1031 * also comment in prog_array_map_poke_run().
1033 elem->aux = prog_aux;
1035 list_add_tail(&elem->list, &aux->poke_progs);
1037 mutex_unlock(&aux->poke_mutex);
1041 static void prog_array_map_poke_untrack(struct bpf_map *map,
1042 struct bpf_prog_aux *prog_aux)
1044 struct prog_poke_elem *elem, *tmp;
1045 struct bpf_array_aux *aux;
1047 aux = container_of(map, struct bpf_array, map)->aux;
1048 mutex_lock(&aux->poke_mutex);
1049 list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1050 if (elem->aux == prog_aux) {
1051 list_del_init(&elem->list);
1056 mutex_unlock(&aux->poke_mutex);
1059 void __weak bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor *poke,
1060 struct bpf_prog *new, struct bpf_prog *old)
1065 static void prog_array_map_poke_run(struct bpf_map *map, u32 key,
1066 struct bpf_prog *old,
1067 struct bpf_prog *new)
1069 struct prog_poke_elem *elem;
1070 struct bpf_array_aux *aux;
1072 aux = container_of(map, struct bpf_array, map)->aux;
1073 WARN_ON_ONCE(!mutex_is_locked(&aux->poke_mutex));
1075 list_for_each_entry(elem, &aux->poke_progs, list) {
1076 struct bpf_jit_poke_descriptor *poke;
1079 for (i = 0; i < elem->aux->size_poke_tab; i++) {
1080 poke = &elem->aux->poke_tab[i];
1082 /* Few things to be aware of:
1084 * 1) We can only ever access aux in this context, but
1085 * not aux->prog since it might not be stable yet and
1086 * there could be danger of use after free otherwise.
1087 * 2) Initially when we start tracking aux, the program
1088 * is not JITed yet and also does not have a kallsyms
1089 * entry. We skip these as poke->tailcall_target_stable
1090 * is not active yet. The JIT will do the final fixup
1091 * before setting it stable. The various
1092 * poke->tailcall_target_stable are successively
1093 * activated, so tail call updates can arrive from here
1094 * while JIT is still finishing its final fixup for
1095 * non-activated poke entries.
1096 * 3) Also programs reaching refcount of zero while patching
1097 * is in progress is okay since we're protected under
1098 * poke_mutex and untrack the programs before the JIT
1101 if (!READ_ONCE(poke->tailcall_target_stable))
1103 if (poke->reason != BPF_POKE_REASON_TAIL_CALL)
1105 if (poke->tail_call.map != map ||
1106 poke->tail_call.key != key)
1109 bpf_arch_poke_desc_update(poke, new, old);
1114 static void prog_array_map_clear_deferred(struct work_struct *work)
1116 struct bpf_map *map = container_of(work, struct bpf_array_aux,
1118 bpf_fd_array_map_clear(map, true);
1122 static void prog_array_map_clear(struct bpf_map *map)
1124 struct bpf_array_aux *aux = container_of(map, struct bpf_array,
1127 schedule_work(&aux->work);
1130 static struct bpf_map *prog_array_map_alloc(union bpf_attr *attr)
1132 struct bpf_array_aux *aux;
1133 struct bpf_map *map;
1135 aux = kzalloc(sizeof(*aux), GFP_KERNEL_ACCOUNT);
1137 return ERR_PTR(-ENOMEM);
1139 INIT_WORK(&aux->work, prog_array_map_clear_deferred);
1140 INIT_LIST_HEAD(&aux->poke_progs);
1141 mutex_init(&aux->poke_mutex);
1143 map = array_map_alloc(attr);
1149 container_of(map, struct bpf_array, map)->aux = aux;
1155 static void prog_array_map_free(struct bpf_map *map)
1157 struct prog_poke_elem *elem, *tmp;
1158 struct bpf_array_aux *aux;
1160 aux = container_of(map, struct bpf_array, map)->aux;
1161 list_for_each_entry_safe(elem, tmp, &aux->poke_progs, list) {
1162 list_del_init(&elem->list);
1166 fd_array_map_free(map);
1169 /* prog_array->aux->{type,jited} is a runtime binding.
1170 * Doing static check alone in the verifier is not enough.
1171 * Thus, prog_array_map cannot be used as an inner_map
1172 * and map_meta_equal is not implemented.
1174 const struct bpf_map_ops prog_array_map_ops = {
1175 .map_alloc_check = fd_array_map_alloc_check,
1176 .map_alloc = prog_array_map_alloc,
1177 .map_free = prog_array_map_free,
1178 .map_poke_track = prog_array_map_poke_track,
1179 .map_poke_untrack = prog_array_map_poke_untrack,
1180 .map_poke_run = prog_array_map_poke_run,
1181 .map_get_next_key = array_map_get_next_key,
1182 .map_lookup_elem = fd_array_map_lookup_elem,
1183 .map_delete_elem = fd_array_map_delete_elem,
1184 .map_fd_get_ptr = prog_fd_array_get_ptr,
1185 .map_fd_put_ptr = prog_fd_array_put_ptr,
1186 .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
1187 .map_release_uref = prog_array_map_clear,
1188 .map_seq_show_elem = prog_array_map_seq_show_elem,
1189 .map_mem_usage = array_map_mem_usage,
1190 .map_btf_id = &array_map_btf_ids[0],
1193 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
1194 struct file *map_file)
1196 struct bpf_event_entry *ee;
1198 ee = kzalloc(sizeof(*ee), GFP_KERNEL);
1200 ee->event = perf_file->private_data;
1201 ee->perf_file = perf_file;
1202 ee->map_file = map_file;
1208 static void __bpf_event_entry_free(struct rcu_head *rcu)
1210 struct bpf_event_entry *ee;
1212 ee = container_of(rcu, struct bpf_event_entry, rcu);
1213 fput(ee->perf_file);
1217 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
1219 call_rcu(&ee->rcu, __bpf_event_entry_free);
1222 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
1223 struct file *map_file, int fd)
1225 struct bpf_event_entry *ee;
1226 struct perf_event *event;
1227 struct file *perf_file;
1230 perf_file = perf_event_get(fd);
1231 if (IS_ERR(perf_file))
1234 ee = ERR_PTR(-EOPNOTSUPP);
1235 event = perf_file->private_data;
1236 if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
1239 ee = bpf_event_entry_gen(perf_file, map_file);
1242 ee = ERR_PTR(-ENOMEM);
1248 static void perf_event_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1250 /* bpf_perf_event is freed after one RCU grace period */
1251 bpf_event_entry_free_rcu(ptr);
1254 static void perf_event_fd_array_release(struct bpf_map *map,
1255 struct file *map_file)
1257 struct bpf_array *array = container_of(map, struct bpf_array, map);
1258 struct bpf_event_entry *ee;
1261 if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1265 for (i = 0; i < array->map.max_entries; i++) {
1266 ee = READ_ONCE(array->ptrs[i]);
1267 if (ee && ee->map_file == map_file)
1268 __fd_array_map_delete_elem(map, &i, true);
1273 static void perf_event_fd_array_map_free(struct bpf_map *map)
1275 if (map->map_flags & BPF_F_PRESERVE_ELEMS)
1276 bpf_fd_array_map_clear(map, false);
1277 fd_array_map_free(map);
1280 const struct bpf_map_ops perf_event_array_map_ops = {
1281 .map_meta_equal = bpf_map_meta_equal,
1282 .map_alloc_check = fd_array_map_alloc_check,
1283 .map_alloc = array_map_alloc,
1284 .map_free = perf_event_fd_array_map_free,
1285 .map_get_next_key = array_map_get_next_key,
1286 .map_lookup_elem = fd_array_map_lookup_elem,
1287 .map_delete_elem = fd_array_map_delete_elem,
1288 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
1289 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
1290 .map_release = perf_event_fd_array_release,
1291 .map_check_btf = map_check_no_btf,
1292 .map_mem_usage = array_map_mem_usage,
1293 .map_btf_id = &array_map_btf_ids[0],
1296 #ifdef CONFIG_CGROUPS
1297 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
1298 struct file *map_file /* not used */,
1301 return cgroup_get_from_fd(fd);
1304 static void cgroup_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
1306 /* cgroup_put free cgrp after a rcu grace period */
1310 static void cgroup_fd_array_free(struct bpf_map *map)
1312 bpf_fd_array_map_clear(map, false);
1313 fd_array_map_free(map);
1316 const struct bpf_map_ops cgroup_array_map_ops = {
1317 .map_meta_equal = bpf_map_meta_equal,
1318 .map_alloc_check = fd_array_map_alloc_check,
1319 .map_alloc = array_map_alloc,
1320 .map_free = cgroup_fd_array_free,
1321 .map_get_next_key = array_map_get_next_key,
1322 .map_lookup_elem = fd_array_map_lookup_elem,
1323 .map_delete_elem = fd_array_map_delete_elem,
1324 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
1325 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
1326 .map_check_btf = map_check_no_btf,
1327 .map_mem_usage = array_map_mem_usage,
1328 .map_btf_id = &array_map_btf_ids[0],
1332 static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
1334 struct bpf_map *map, *inner_map_meta;
1336 inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
1337 if (IS_ERR(inner_map_meta))
1338 return inner_map_meta;
1340 map = array_map_alloc(attr);
1342 bpf_map_meta_free(inner_map_meta);
1346 map->inner_map_meta = inner_map_meta;
1351 static void array_of_map_free(struct bpf_map *map)
1353 /* map->inner_map_meta is only accessed by syscall which
1354 * is protected by fdget/fdput.
1356 bpf_map_meta_free(map->inner_map_meta);
1357 bpf_fd_array_map_clear(map, false);
1358 fd_array_map_free(map);
1361 static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
1363 struct bpf_map **inner_map = array_map_lookup_elem(map, key);
1368 return READ_ONCE(*inner_map);
1371 static int array_of_map_gen_lookup(struct bpf_map *map,
1372 struct bpf_insn *insn_buf)
1374 struct bpf_array *array = container_of(map, struct bpf_array, map);
1375 u32 elem_size = array->elem_size;
1376 struct bpf_insn *insn = insn_buf;
1377 const int ret = BPF_REG_0;
1378 const int map_ptr = BPF_REG_1;
1379 const int index = BPF_REG_2;
1381 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
1382 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
1383 if (!map->bypass_spec_v1) {
1384 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
1385 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
1387 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
1389 if (is_power_of_2(elem_size))
1390 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
1392 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
1393 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
1394 *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
1395 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
1396 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
1397 *insn++ = BPF_MOV64_IMM(ret, 0);
1399 return insn - insn_buf;
1402 const struct bpf_map_ops array_of_maps_map_ops = {
1403 .map_alloc_check = fd_array_map_alloc_check,
1404 .map_alloc = array_of_map_alloc,
1405 .map_free = array_of_map_free,
1406 .map_get_next_key = array_map_get_next_key,
1407 .map_lookup_elem = array_of_map_lookup_elem,
1408 .map_delete_elem = fd_array_map_delete_elem,
1409 .map_fd_get_ptr = bpf_map_fd_get_ptr,
1410 .map_fd_put_ptr = bpf_map_fd_put_ptr,
1411 .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
1412 .map_gen_lookup = array_of_map_gen_lookup,
1413 .map_lookup_batch = generic_map_lookup_batch,
1414 .map_update_batch = generic_map_update_batch,
1415 .map_check_btf = map_check_no_btf,
1416 .map_mem_usage = array_map_mem_usage,
1417 .map_btf_id = &array_map_btf_ids[0],