1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016,2017 Facebook
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 #include <linux/bpf.h>
14 #include <linux/btf.h>
15 #include <linux/err.h>
16 #include <linux/slab.h>
18 #include <linux/filter.h>
19 #include <linux/perf_event.h>
20 #include <uapi/linux/btf.h>
22 #include "map_in_map.h"
24 #define ARRAY_CREATE_FLAG_MASK \
25 (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK)
27 static void bpf_array_free_percpu(struct bpf_array *array)
31 for (i = 0; i < array->map.max_entries; i++) {
32 free_percpu(array->pptrs[i]);
37 static int bpf_array_alloc_percpu(struct bpf_array *array)
42 for (i = 0; i < array->map.max_entries; i++) {
43 ptr = __alloc_percpu_gfp(array->elem_size, 8,
44 GFP_USER | __GFP_NOWARN);
46 bpf_array_free_percpu(array);
49 array->pptrs[i] = ptr;
56 /* Called from syscall */
57 int array_map_alloc_check(union bpf_attr *attr)
59 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
60 int numa_node = bpf_map_attr_numa_node(attr);
62 /* check sanity of attributes */
63 if (attr->max_entries == 0 || attr->key_size != 4 ||
64 attr->value_size == 0 ||
65 attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
66 !bpf_map_flags_access_ok(attr->map_flags) ||
67 (percpu && numa_node != NUMA_NO_NODE))
70 if (attr->value_size > KMALLOC_MAX_SIZE)
71 /* if value_size is bigger, the user space won't be able to
72 * access the elements.
79 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
81 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
82 int ret, numa_node = bpf_map_attr_numa_node(attr);
83 u32 elem_size, index_mask, max_entries;
84 bool unpriv = !capable(CAP_SYS_ADMIN);
85 u64 cost, array_size, mask64;
86 struct bpf_array *array;
88 elem_size = round_up(attr->value_size, 8);
90 max_entries = attr->max_entries;
92 /* On 32 bit archs roundup_pow_of_two() with max_entries that has
93 * upper most bit set in u32 space is undefined behavior due to
94 * resulting 1U << 32, so do it manually here in u64 space.
96 mask64 = fls_long(max_entries - 1);
97 mask64 = 1ULL << mask64;
102 /* round up array size to nearest power of 2,
103 * since cpu will speculate within index_mask limits
105 max_entries = index_mask + 1;
106 /* Check for overflows. */
107 if (max_entries < attr->max_entries)
108 return ERR_PTR(-E2BIG);
111 array_size = sizeof(*array);
113 array_size += (u64) max_entries * sizeof(void *);
115 array_size += (u64) max_entries * elem_size;
117 /* make sure there is no u32 overflow later in round_up() */
119 if (cost >= U32_MAX - PAGE_SIZE)
120 return ERR_PTR(-ENOMEM);
122 cost += (u64)attr->max_entries * elem_size * num_possible_cpus();
123 if (cost >= U32_MAX - PAGE_SIZE)
124 return ERR_PTR(-ENOMEM);
126 cost = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
128 ret = bpf_map_precharge_memlock(cost);
132 /* allocate all map elements and zero-initialize them */
133 array = bpf_map_area_alloc(array_size, numa_node);
135 return ERR_PTR(-ENOMEM);
136 array->index_mask = index_mask;
137 array->map.unpriv_array = unpriv;
139 /* copy mandatory map attributes */
140 bpf_map_init_from_attr(&array->map, attr);
141 array->map.pages = cost;
142 array->elem_size = elem_size;
144 if (percpu && bpf_array_alloc_percpu(array)) {
145 bpf_map_area_free(array);
146 return ERR_PTR(-ENOMEM);
152 /* Called from syscall or from eBPF program */
153 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
155 struct bpf_array *array = container_of(map, struct bpf_array, map);
156 u32 index = *(u32 *)key;
158 if (unlikely(index >= array->map.max_entries))
161 return array->value + array->elem_size * (index & array->index_mask);
164 static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm,
167 struct bpf_array *array = container_of(map, struct bpf_array, map);
169 if (map->max_entries != 1)
171 if (off >= map->value_size)
174 *imm = (unsigned long)array->value;
178 static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm,
181 struct bpf_array *array = container_of(map, struct bpf_array, map);
182 u64 base = (unsigned long)array->value;
183 u64 range = array->elem_size;
185 if (map->max_entries != 1)
187 if (imm < base || imm >= base + range)
194 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
195 static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
197 struct bpf_array *array = container_of(map, struct bpf_array, map);
198 struct bpf_insn *insn = insn_buf;
199 u32 elem_size = round_up(map->value_size, 8);
200 const int ret = BPF_REG_0;
201 const int map_ptr = BPF_REG_1;
202 const int index = BPF_REG_2;
204 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
205 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
206 if (map->unpriv_array) {
207 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
208 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
210 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
213 if (is_power_of_2(elem_size)) {
214 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
216 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
218 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
219 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
220 *insn++ = BPF_MOV64_IMM(ret, 0);
221 return insn - insn_buf;
224 /* Called from eBPF program */
225 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
227 struct bpf_array *array = container_of(map, struct bpf_array, map);
228 u32 index = *(u32 *)key;
230 if (unlikely(index >= array->map.max_entries))
233 return this_cpu_ptr(array->pptrs[index & array->index_mask]);
236 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
238 struct bpf_array *array = container_of(map, struct bpf_array, map);
239 u32 index = *(u32 *)key;
244 if (unlikely(index >= array->map.max_entries))
247 /* per_cpu areas are zero-filled and bpf programs can only
248 * access 'value_size' of them, so copying rounded areas
249 * will not leak any kernel data
251 size = round_up(map->value_size, 8);
253 pptr = array->pptrs[index & array->index_mask];
254 for_each_possible_cpu(cpu) {
255 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
262 /* Called from syscall */
263 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
265 struct bpf_array *array = container_of(map, struct bpf_array, map);
266 u32 index = key ? *(u32 *)key : U32_MAX;
267 u32 *next = (u32 *)next_key;
269 if (index >= array->map.max_entries) {
274 if (index == array->map.max_entries - 1)
281 /* Called from syscall or from eBPF program */
282 static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
285 struct bpf_array *array = container_of(map, struct bpf_array, map);
286 u32 index = *(u32 *)key;
289 if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
293 if (unlikely(index >= array->map.max_entries))
294 /* all elements were pre-allocated, cannot insert a new one */
297 if (unlikely(map_flags & BPF_NOEXIST))
298 /* all elements already exist */
301 if (unlikely((map_flags & BPF_F_LOCK) &&
302 !map_value_has_spin_lock(map)))
305 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
306 memcpy(this_cpu_ptr(array->pptrs[index & array->index_mask]),
307 value, map->value_size);
310 array->elem_size * (index & array->index_mask);
311 if (map_flags & BPF_F_LOCK)
312 copy_map_value_locked(map, val, value, false);
314 copy_map_value(map, val, value);
319 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
322 struct bpf_array *array = container_of(map, struct bpf_array, map);
323 u32 index = *(u32 *)key;
328 if (unlikely(map_flags > BPF_EXIST))
332 if (unlikely(index >= array->map.max_entries))
333 /* all elements were pre-allocated, cannot insert a new one */
336 if (unlikely(map_flags == BPF_NOEXIST))
337 /* all elements already exist */
340 /* the user space will provide round_up(value_size, 8) bytes that
341 * will be copied into per-cpu area. bpf programs can only access
342 * value_size of it. During lookup the same extra bytes will be
343 * returned or zeros which were zero-filled by percpu_alloc,
344 * so no kernel data leaks possible
346 size = round_up(map->value_size, 8);
348 pptr = array->pptrs[index & array->index_mask];
349 for_each_possible_cpu(cpu) {
350 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
357 /* Called from syscall or from eBPF program */
358 static int array_map_delete_elem(struct bpf_map *map, void *key)
363 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
364 static void array_map_free(struct bpf_map *map)
366 struct bpf_array *array = container_of(map, struct bpf_array, map);
368 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
369 * so the programs (can be more than one that used this map) were
370 * disconnected from events. Wait for outstanding programs to complete
375 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
376 bpf_array_free_percpu(array);
378 bpf_map_area_free(array);
381 static void array_map_seq_show_elem(struct bpf_map *map, void *key,
388 value = array_map_lookup_elem(map, key);
394 seq_printf(m, "%u: ", *(u32 *)key);
395 btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
401 static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
404 struct bpf_array *array = container_of(map, struct bpf_array, map);
405 u32 index = *(u32 *)key;
411 seq_printf(m, "%u: {\n", *(u32 *)key);
412 pptr = array->pptrs[index & array->index_mask];
413 for_each_possible_cpu(cpu) {
414 seq_printf(m, "\tcpu%d: ", cpu);
415 btf_type_seq_show(map->btf, map->btf_value_type_id,
416 per_cpu_ptr(pptr, cpu), m);
424 static int array_map_check_btf(const struct bpf_map *map,
425 const struct btf *btf,
426 const struct btf_type *key_type,
427 const struct btf_type *value_type)
431 if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT)
434 int_data = *(u32 *)(key_type + 1);
435 /* bpf array can only take a u32 key. This check makes sure
436 * that the btf matches the attr used during map_create.
438 if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data))
444 const struct bpf_map_ops array_map_ops = {
445 .map_alloc_check = array_map_alloc_check,
446 .map_alloc = array_map_alloc,
447 .map_free = array_map_free,
448 .map_get_next_key = array_map_get_next_key,
449 .map_lookup_elem = array_map_lookup_elem,
450 .map_update_elem = array_map_update_elem,
451 .map_delete_elem = array_map_delete_elem,
452 .map_gen_lookup = array_map_gen_lookup,
453 .map_direct_value_addr = array_map_direct_value_addr,
454 .map_direct_value_meta = array_map_direct_value_meta,
455 .map_seq_show_elem = array_map_seq_show_elem,
456 .map_check_btf = array_map_check_btf,
459 const struct bpf_map_ops percpu_array_map_ops = {
460 .map_alloc_check = array_map_alloc_check,
461 .map_alloc = array_map_alloc,
462 .map_free = array_map_free,
463 .map_get_next_key = array_map_get_next_key,
464 .map_lookup_elem = percpu_array_map_lookup_elem,
465 .map_update_elem = array_map_update_elem,
466 .map_delete_elem = array_map_delete_elem,
467 .map_seq_show_elem = percpu_array_map_seq_show_elem,
468 .map_check_btf = array_map_check_btf,
471 static int fd_array_map_alloc_check(union bpf_attr *attr)
473 /* only file descriptors can be stored in this type of map */
474 if (attr->value_size != sizeof(u32))
476 /* Program read-only/write-only not supported for special maps yet. */
477 if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG))
479 return array_map_alloc_check(attr);
482 static void fd_array_map_free(struct bpf_map *map)
484 struct bpf_array *array = container_of(map, struct bpf_array, map);
489 /* make sure it's empty */
490 for (i = 0; i < array->map.max_entries; i++)
491 BUG_ON(array->ptrs[i] != NULL);
493 bpf_map_area_free(array);
496 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
498 return ERR_PTR(-EOPNOTSUPP);
501 /* only called from syscall */
502 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
507 if (!map->ops->map_fd_sys_lookup_elem)
511 elem = array_map_lookup_elem(map, key);
512 if (elem && (ptr = READ_ONCE(*elem)))
513 *value = map->ops->map_fd_sys_lookup_elem(ptr);
521 /* only called from syscall */
522 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
523 void *key, void *value, u64 map_flags)
525 struct bpf_array *array = container_of(map, struct bpf_array, map);
526 void *new_ptr, *old_ptr;
527 u32 index = *(u32 *)key, ufd;
529 if (map_flags != BPF_ANY)
532 if (index >= array->map.max_entries)
536 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
538 return PTR_ERR(new_ptr);
540 old_ptr = xchg(array->ptrs + index, new_ptr);
542 map->ops->map_fd_put_ptr(old_ptr);
547 static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
549 struct bpf_array *array = container_of(map, struct bpf_array, map);
551 u32 index = *(u32 *)key;
553 if (index >= array->map.max_entries)
556 old_ptr = xchg(array->ptrs + index, NULL);
558 map->ops->map_fd_put_ptr(old_ptr);
565 static void *prog_fd_array_get_ptr(struct bpf_map *map,
566 struct file *map_file, int fd)
568 struct bpf_array *array = container_of(map, struct bpf_array, map);
569 struct bpf_prog *prog = bpf_prog_get(fd);
574 if (!bpf_prog_array_compatible(array, prog)) {
576 return ERR_PTR(-EINVAL);
582 static void prog_fd_array_put_ptr(void *ptr)
587 static u32 prog_fd_array_sys_lookup_elem(void *ptr)
589 return ((struct bpf_prog *)ptr)->aux->id;
592 /* decrement refcnt of all bpf_progs that are stored in this map */
593 static void bpf_fd_array_map_clear(struct bpf_map *map)
595 struct bpf_array *array = container_of(map, struct bpf_array, map);
598 for (i = 0; i < array->map.max_entries; i++)
599 fd_array_map_delete_elem(map, &i);
602 static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
610 elem = array_map_lookup_elem(map, key);
612 ptr = READ_ONCE(*elem);
614 seq_printf(m, "%u: ", *(u32 *)key);
615 prog_id = prog_fd_array_sys_lookup_elem(ptr);
616 btf_type_seq_show(map->btf, map->btf_value_type_id,
625 const struct bpf_map_ops prog_array_map_ops = {
626 .map_alloc_check = fd_array_map_alloc_check,
627 .map_alloc = array_map_alloc,
628 .map_free = fd_array_map_free,
629 .map_get_next_key = array_map_get_next_key,
630 .map_lookup_elem = fd_array_map_lookup_elem,
631 .map_delete_elem = fd_array_map_delete_elem,
632 .map_fd_get_ptr = prog_fd_array_get_ptr,
633 .map_fd_put_ptr = prog_fd_array_put_ptr,
634 .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
635 .map_release_uref = bpf_fd_array_map_clear,
636 .map_seq_show_elem = prog_array_map_seq_show_elem,
639 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
640 struct file *map_file)
642 struct bpf_event_entry *ee;
644 ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
646 ee->event = perf_file->private_data;
647 ee->perf_file = perf_file;
648 ee->map_file = map_file;
654 static void __bpf_event_entry_free(struct rcu_head *rcu)
656 struct bpf_event_entry *ee;
658 ee = container_of(rcu, struct bpf_event_entry, rcu);
663 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
665 call_rcu(&ee->rcu, __bpf_event_entry_free);
668 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
669 struct file *map_file, int fd)
671 struct bpf_event_entry *ee;
672 struct perf_event *event;
673 struct file *perf_file;
676 perf_file = perf_event_get(fd);
677 if (IS_ERR(perf_file))
680 ee = ERR_PTR(-EOPNOTSUPP);
681 event = perf_file->private_data;
682 if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
685 ee = bpf_event_entry_gen(perf_file, map_file);
688 ee = ERR_PTR(-ENOMEM);
694 static void perf_event_fd_array_put_ptr(void *ptr)
696 bpf_event_entry_free_rcu(ptr);
699 static void perf_event_fd_array_release(struct bpf_map *map,
700 struct file *map_file)
702 struct bpf_array *array = container_of(map, struct bpf_array, map);
703 struct bpf_event_entry *ee;
707 for (i = 0; i < array->map.max_entries; i++) {
708 ee = READ_ONCE(array->ptrs[i]);
709 if (ee && ee->map_file == map_file)
710 fd_array_map_delete_elem(map, &i);
715 const struct bpf_map_ops perf_event_array_map_ops = {
716 .map_alloc_check = fd_array_map_alloc_check,
717 .map_alloc = array_map_alloc,
718 .map_free = fd_array_map_free,
719 .map_get_next_key = array_map_get_next_key,
720 .map_lookup_elem = fd_array_map_lookup_elem,
721 .map_delete_elem = fd_array_map_delete_elem,
722 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
723 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
724 .map_release = perf_event_fd_array_release,
725 .map_check_btf = map_check_no_btf,
728 #ifdef CONFIG_CGROUPS
729 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
730 struct file *map_file /* not used */,
733 return cgroup_get_from_fd(fd);
736 static void cgroup_fd_array_put_ptr(void *ptr)
738 /* cgroup_put free cgrp after a rcu grace period */
742 static void cgroup_fd_array_free(struct bpf_map *map)
744 bpf_fd_array_map_clear(map);
745 fd_array_map_free(map);
748 const struct bpf_map_ops cgroup_array_map_ops = {
749 .map_alloc_check = fd_array_map_alloc_check,
750 .map_alloc = array_map_alloc,
751 .map_free = cgroup_fd_array_free,
752 .map_get_next_key = array_map_get_next_key,
753 .map_lookup_elem = fd_array_map_lookup_elem,
754 .map_delete_elem = fd_array_map_delete_elem,
755 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
756 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
757 .map_check_btf = map_check_no_btf,
761 static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
763 struct bpf_map *map, *inner_map_meta;
765 inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
766 if (IS_ERR(inner_map_meta))
767 return inner_map_meta;
769 map = array_map_alloc(attr);
771 bpf_map_meta_free(inner_map_meta);
775 map->inner_map_meta = inner_map_meta;
780 static void array_of_map_free(struct bpf_map *map)
782 /* map->inner_map_meta is only accessed by syscall which
783 * is protected by fdget/fdput.
785 bpf_map_meta_free(map->inner_map_meta);
786 bpf_fd_array_map_clear(map);
787 fd_array_map_free(map);
790 static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
792 struct bpf_map **inner_map = array_map_lookup_elem(map, key);
797 return READ_ONCE(*inner_map);
800 static u32 array_of_map_gen_lookup(struct bpf_map *map,
801 struct bpf_insn *insn_buf)
803 struct bpf_array *array = container_of(map, struct bpf_array, map);
804 u32 elem_size = round_up(map->value_size, 8);
805 struct bpf_insn *insn = insn_buf;
806 const int ret = BPF_REG_0;
807 const int map_ptr = BPF_REG_1;
808 const int index = BPF_REG_2;
810 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
811 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
812 if (map->unpriv_array) {
813 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
814 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
816 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
818 if (is_power_of_2(elem_size))
819 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
821 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
822 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
823 *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
824 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
825 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
826 *insn++ = BPF_MOV64_IMM(ret, 0);
828 return insn - insn_buf;
831 const struct bpf_map_ops array_of_maps_map_ops = {
832 .map_alloc_check = fd_array_map_alloc_check,
833 .map_alloc = array_of_map_alloc,
834 .map_free = array_of_map_free,
835 .map_get_next_key = array_map_get_next_key,
836 .map_lookup_elem = array_of_map_lookup_elem,
837 .map_delete_elem = fd_array_map_delete_elem,
838 .map_fd_get_ptr = bpf_map_fd_get_ptr,
839 .map_fd_put_ptr = bpf_map_fd_put_ptr,
840 .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
841 .map_gen_lookup = array_of_map_gen_lookup,
842 .map_check_btf = map_check_no_btf,