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/err.h>
15 #include <linux/slab.h>
17 #include <linux/filter.h>
18 #include <linux/perf_event.h>
20 #include "map_in_map.h"
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 = __alloc_percpu_gfp(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 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
54 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
55 int ret, numa_node = bpf_map_attr_numa_node(attr);
56 u32 elem_size, index_mask, max_entries;
57 bool unpriv = !capable(CAP_SYS_ADMIN);
58 u64 cost, array_size, mask64;
59 struct bpf_array *array;
61 /* check sanity of attributes */
62 if (attr->max_entries == 0 || attr->key_size != 4 ||
63 attr->value_size == 0 || attr->map_flags & ~BPF_F_NUMA_NODE ||
64 (percpu && numa_node != NUMA_NO_NODE))
65 return ERR_PTR(-EINVAL);
67 if (attr->value_size > KMALLOC_MAX_SIZE)
68 /* if value_size is bigger, the user space won't be able to
69 * access the elements.
71 return ERR_PTR(-E2BIG);
73 elem_size = round_up(attr->value_size, 8);
75 max_entries = attr->max_entries;
77 /* On 32 bit archs roundup_pow_of_two() with max_entries that has
78 * upper most bit set in u32 space is undefined behavior due to
79 * resulting 1U << 32, so do it manually here in u64 space.
81 mask64 = fls_long(max_entries - 1);
82 mask64 = 1ULL << mask64;
87 /* round up array size to nearest power of 2,
88 * since cpu will speculate within index_mask limits
90 max_entries = index_mask + 1;
91 /* Check for overflows. */
92 if (max_entries < attr->max_entries)
93 return ERR_PTR(-E2BIG);
96 array_size = sizeof(*array);
98 array_size += (u64) max_entries * sizeof(void *);
100 array_size += (u64) max_entries * elem_size;
102 /* make sure there is no u32 overflow later in round_up() */
104 if (cost >= U32_MAX - PAGE_SIZE)
105 return ERR_PTR(-ENOMEM);
107 cost += (u64)attr->max_entries * elem_size * num_possible_cpus();
108 if (cost >= U32_MAX - PAGE_SIZE)
109 return ERR_PTR(-ENOMEM);
111 cost = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
113 ret = bpf_map_precharge_memlock(cost);
117 /* allocate all map elements and zero-initialize them */
118 array = bpf_map_area_alloc(array_size, numa_node);
120 return ERR_PTR(-ENOMEM);
121 array->index_mask = index_mask;
122 array->map.unpriv_array = unpriv;
124 /* copy mandatory map attributes */
125 array->map.map_type = attr->map_type;
126 array->map.key_size = attr->key_size;
127 array->map.value_size = attr->value_size;
128 array->map.max_entries = attr->max_entries;
129 array->map.map_flags = attr->map_flags;
130 array->map.numa_node = numa_node;
131 array->map.pages = cost;
132 array->elem_size = elem_size;
134 if (percpu && bpf_array_alloc_percpu(array)) {
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 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
155 static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
157 struct bpf_array *array = container_of(map, struct bpf_array, map);
158 struct bpf_insn *insn = insn_buf;
159 u32 elem_size = round_up(map->value_size, 8);
160 const int ret = BPF_REG_0;
161 const int map_ptr = BPF_REG_1;
162 const int index = BPF_REG_2;
164 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
165 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
166 if (map->unpriv_array) {
167 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
168 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
170 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
173 if (is_power_of_2(elem_size)) {
174 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
176 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
178 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
179 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
180 *insn++ = BPF_MOV64_IMM(ret, 0);
181 return insn - insn_buf;
184 /* Called from eBPF program */
185 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
187 struct bpf_array *array = container_of(map, struct bpf_array, map);
188 u32 index = *(u32 *)key;
190 if (unlikely(index >= array->map.max_entries))
193 return this_cpu_ptr(array->pptrs[index & array->index_mask]);
196 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
198 struct bpf_array *array = container_of(map, struct bpf_array, map);
199 u32 index = *(u32 *)key;
204 if (unlikely(index >= array->map.max_entries))
207 /* per_cpu areas are zero-filled and bpf programs can only
208 * access 'value_size' of them, so copying rounded areas
209 * will not leak any kernel data
211 size = round_up(map->value_size, 8);
213 pptr = array->pptrs[index & array->index_mask];
214 for_each_possible_cpu(cpu) {
215 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
222 /* Called from syscall */
223 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
225 struct bpf_array *array = container_of(map, struct bpf_array, map);
226 u32 index = key ? *(u32 *)key : U32_MAX;
227 u32 *next = (u32 *)next_key;
229 if (index >= array->map.max_entries) {
234 if (index == array->map.max_entries - 1)
241 /* Called from syscall or from eBPF program */
242 static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
245 struct bpf_array *array = container_of(map, struct bpf_array, map);
246 u32 index = *(u32 *)key;
248 if (unlikely(map_flags > BPF_EXIST))
252 if (unlikely(index >= array->map.max_entries))
253 /* all elements were pre-allocated, cannot insert a new one */
256 if (unlikely(map_flags == BPF_NOEXIST))
257 /* all elements already exist */
260 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
261 memcpy(this_cpu_ptr(array->pptrs[index & array->index_mask]),
262 value, map->value_size);
264 memcpy(array->value +
265 array->elem_size * (index & array->index_mask),
266 value, map->value_size);
270 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
273 struct bpf_array *array = container_of(map, struct bpf_array, map);
274 u32 index = *(u32 *)key;
279 if (unlikely(map_flags > 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 /* the user space will provide round_up(value_size, 8) bytes that
292 * will be copied into per-cpu area. bpf programs can only access
293 * value_size of it. During lookup the same extra bytes will be
294 * returned or zeros which were zero-filled by percpu_alloc,
295 * so no kernel data leaks possible
297 size = round_up(map->value_size, 8);
299 pptr = array->pptrs[index & array->index_mask];
300 for_each_possible_cpu(cpu) {
301 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
308 /* Called from syscall or from eBPF program */
309 static int array_map_delete_elem(struct bpf_map *map, void *key)
314 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
315 static void array_map_free(struct bpf_map *map)
317 struct bpf_array *array = container_of(map, struct bpf_array, map);
319 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
320 * so the programs (can be more than one that used this map) were
321 * disconnected from events. Wait for outstanding programs to complete
326 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
327 bpf_array_free_percpu(array);
329 bpf_map_area_free(array);
332 const struct bpf_map_ops array_map_ops = {
333 .map_alloc = array_map_alloc,
334 .map_free = array_map_free,
335 .map_get_next_key = array_map_get_next_key,
336 .map_lookup_elem = array_map_lookup_elem,
337 .map_update_elem = array_map_update_elem,
338 .map_delete_elem = array_map_delete_elem,
339 .map_gen_lookup = array_map_gen_lookup,
342 const struct bpf_map_ops percpu_array_map_ops = {
343 .map_alloc = array_map_alloc,
344 .map_free = array_map_free,
345 .map_get_next_key = array_map_get_next_key,
346 .map_lookup_elem = percpu_array_map_lookup_elem,
347 .map_update_elem = array_map_update_elem,
348 .map_delete_elem = array_map_delete_elem,
351 static struct bpf_map *fd_array_map_alloc(union bpf_attr *attr)
353 /* only file descriptors can be stored in this type of map */
354 if (attr->value_size != sizeof(u32))
355 return ERR_PTR(-EINVAL);
356 return array_map_alloc(attr);
359 static void fd_array_map_free(struct bpf_map *map)
361 struct bpf_array *array = container_of(map, struct bpf_array, map);
366 /* make sure it's empty */
367 for (i = 0; i < array->map.max_entries; i++)
368 BUG_ON(array->ptrs[i] != NULL);
370 bpf_map_area_free(array);
373 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
378 /* only called from syscall */
379 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
384 if (!map->ops->map_fd_sys_lookup_elem)
388 elem = array_map_lookup_elem(map, key);
389 if (elem && (ptr = READ_ONCE(*elem)))
390 *value = map->ops->map_fd_sys_lookup_elem(ptr);
398 /* only called from syscall */
399 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
400 void *key, void *value, u64 map_flags)
402 struct bpf_array *array = container_of(map, struct bpf_array, map);
403 void *new_ptr, *old_ptr;
404 u32 index = *(u32 *)key, ufd;
406 if (map_flags != BPF_ANY)
409 if (index >= array->map.max_entries)
413 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
415 return PTR_ERR(new_ptr);
417 old_ptr = xchg(array->ptrs + index, new_ptr);
419 map->ops->map_fd_put_ptr(old_ptr);
424 static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
426 struct bpf_array *array = container_of(map, struct bpf_array, map);
428 u32 index = *(u32 *)key;
430 if (index >= array->map.max_entries)
433 old_ptr = xchg(array->ptrs + index, NULL);
435 map->ops->map_fd_put_ptr(old_ptr);
442 static void *prog_fd_array_get_ptr(struct bpf_map *map,
443 struct file *map_file, int fd)
445 struct bpf_array *array = container_of(map, struct bpf_array, map);
446 struct bpf_prog *prog = bpf_prog_get(fd);
451 if (!bpf_prog_array_compatible(array, prog)) {
453 return ERR_PTR(-EINVAL);
459 static void prog_fd_array_put_ptr(void *ptr)
464 static u32 prog_fd_array_sys_lookup_elem(void *ptr)
466 return ((struct bpf_prog *)ptr)->aux->id;
469 /* decrement refcnt of all bpf_progs that are stored in this map */
470 static void bpf_fd_array_map_clear(struct bpf_map *map)
472 struct bpf_array *array = container_of(map, struct bpf_array, map);
475 for (i = 0; i < array->map.max_entries; i++)
476 fd_array_map_delete_elem(map, &i);
479 const struct bpf_map_ops prog_array_map_ops = {
480 .map_alloc = fd_array_map_alloc,
481 .map_free = fd_array_map_free,
482 .map_get_next_key = array_map_get_next_key,
483 .map_lookup_elem = fd_array_map_lookup_elem,
484 .map_delete_elem = fd_array_map_delete_elem,
485 .map_fd_get_ptr = prog_fd_array_get_ptr,
486 .map_fd_put_ptr = prog_fd_array_put_ptr,
487 .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
488 .map_release_uref = bpf_fd_array_map_clear,
491 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
492 struct file *map_file)
494 struct bpf_event_entry *ee;
496 ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
498 ee->event = perf_file->private_data;
499 ee->perf_file = perf_file;
500 ee->map_file = map_file;
506 static void __bpf_event_entry_free(struct rcu_head *rcu)
508 struct bpf_event_entry *ee;
510 ee = container_of(rcu, struct bpf_event_entry, rcu);
515 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
517 call_rcu(&ee->rcu, __bpf_event_entry_free);
520 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
521 struct file *map_file, int fd)
523 struct bpf_event_entry *ee;
524 struct perf_event *event;
525 struct file *perf_file;
528 perf_file = perf_event_get(fd);
529 if (IS_ERR(perf_file))
532 ee = ERR_PTR(-EOPNOTSUPP);
533 event = perf_file->private_data;
534 if (perf_event_read_local(event, &value) == -EOPNOTSUPP)
537 ee = bpf_event_entry_gen(perf_file, map_file);
540 ee = ERR_PTR(-ENOMEM);
546 static void perf_event_fd_array_put_ptr(void *ptr)
548 bpf_event_entry_free_rcu(ptr);
551 static void perf_event_fd_array_release(struct bpf_map *map,
552 struct file *map_file)
554 struct bpf_array *array = container_of(map, struct bpf_array, map);
555 struct bpf_event_entry *ee;
559 for (i = 0; i < array->map.max_entries; i++) {
560 ee = READ_ONCE(array->ptrs[i]);
561 if (ee && ee->map_file == map_file)
562 fd_array_map_delete_elem(map, &i);
567 const struct bpf_map_ops perf_event_array_map_ops = {
568 .map_alloc = fd_array_map_alloc,
569 .map_free = fd_array_map_free,
570 .map_get_next_key = array_map_get_next_key,
571 .map_lookup_elem = fd_array_map_lookup_elem,
572 .map_delete_elem = fd_array_map_delete_elem,
573 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
574 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
575 .map_release = perf_event_fd_array_release,
578 #ifdef CONFIG_CGROUPS
579 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
580 struct file *map_file /* not used */,
583 return cgroup_get_from_fd(fd);
586 static void cgroup_fd_array_put_ptr(void *ptr)
588 /* cgroup_put free cgrp after a rcu grace period */
592 static void cgroup_fd_array_free(struct bpf_map *map)
594 bpf_fd_array_map_clear(map);
595 fd_array_map_free(map);
598 const struct bpf_map_ops cgroup_array_map_ops = {
599 .map_alloc = fd_array_map_alloc,
600 .map_free = cgroup_fd_array_free,
601 .map_get_next_key = array_map_get_next_key,
602 .map_lookup_elem = fd_array_map_lookup_elem,
603 .map_delete_elem = fd_array_map_delete_elem,
604 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
605 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
609 static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
611 struct bpf_map *map, *inner_map_meta;
613 inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
614 if (IS_ERR(inner_map_meta))
615 return inner_map_meta;
617 map = fd_array_map_alloc(attr);
619 bpf_map_meta_free(inner_map_meta);
623 map->inner_map_meta = inner_map_meta;
628 static void array_of_map_free(struct bpf_map *map)
630 /* map->inner_map_meta is only accessed by syscall which
631 * is protected by fdget/fdput.
633 bpf_map_meta_free(map->inner_map_meta);
634 bpf_fd_array_map_clear(map);
635 fd_array_map_free(map);
638 static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
640 struct bpf_map **inner_map = array_map_lookup_elem(map, key);
645 return READ_ONCE(*inner_map);
648 static u32 array_of_map_gen_lookup(struct bpf_map *map,
649 struct bpf_insn *insn_buf)
651 struct bpf_array *array = container_of(map, struct bpf_array, map);
652 u32 elem_size = round_up(map->value_size, 8);
653 struct bpf_insn *insn = insn_buf;
654 const int ret = BPF_REG_0;
655 const int map_ptr = BPF_REG_1;
656 const int index = BPF_REG_2;
658 *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
659 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
660 if (map->unpriv_array) {
661 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
662 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
664 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
666 if (is_power_of_2(elem_size))
667 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
669 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
670 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
671 *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
672 *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
673 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
674 *insn++ = BPF_MOV64_IMM(ret, 0);
676 return insn - insn_buf;
679 const struct bpf_map_ops array_of_maps_map_ops = {
680 .map_alloc = array_of_map_alloc,
681 .map_free = array_of_map_free,
682 .map_get_next_key = array_map_get_next_key,
683 .map_lookup_elem = array_of_map_lookup_elem,
684 .map_delete_elem = fd_array_map_delete_elem,
685 .map_fd_get_ptr = bpf_map_fd_get_ptr,
686 .map_fd_put_ptr = bpf_map_fd_put_ptr,
687 .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
688 .map_gen_lookup = array_of_map_gen_lookup,