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 #define ARRAY_CREATE_FLAG_MASK \
23 (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
25 static void bpf_array_free_percpu(struct bpf_array
*array
)
29 for (i
= 0; i
< array
->map
.max_entries
; i
++) {
30 free_percpu(array
->pptrs
[i
]);
35 static int bpf_array_alloc_percpu(struct bpf_array
*array
)
40 for (i
= 0; i
< array
->map
.max_entries
; i
++) {
41 ptr
= __alloc_percpu_gfp(array
->elem_size
, 8,
42 GFP_USER
| __GFP_NOWARN
);
44 bpf_array_free_percpu(array
);
47 array
->pptrs
[i
] = ptr
;
54 /* Called from syscall */
55 static struct bpf_map
*array_map_alloc(union bpf_attr
*attr
)
57 bool percpu
= attr
->map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
;
58 int ret
, numa_node
= bpf_map_attr_numa_node(attr
);
59 u32 elem_size
, index_mask
, max_entries
;
60 bool unpriv
= !capable(CAP_SYS_ADMIN
);
61 u64 cost
, array_size
, mask64
;
62 struct bpf_array
*array
;
64 /* check sanity of attributes */
65 if (attr
->max_entries
== 0 || attr
->key_size
!= 4 ||
66 attr
->value_size
== 0 ||
67 attr
->map_flags
& ~ARRAY_CREATE_FLAG_MASK
||
68 (percpu
&& numa_node
!= NUMA_NO_NODE
))
69 return ERR_PTR(-EINVAL
);
71 if (attr
->value_size
> KMALLOC_MAX_SIZE
)
72 /* if value_size is bigger, the user space won't be able to
73 * access the elements.
75 return ERR_PTR(-E2BIG
);
77 elem_size
= round_up(attr
->value_size
, 8);
79 max_entries
= attr
->max_entries
;
81 /* On 32 bit archs roundup_pow_of_two() with max_entries that has
82 * upper most bit set in u32 space is undefined behavior due to
83 * resulting 1U << 32, so do it manually here in u64 space.
85 mask64
= fls_long(max_entries
- 1);
86 mask64
= 1ULL << mask64
;
91 /* round up array size to nearest power of 2,
92 * since cpu will speculate within index_mask limits
94 max_entries
= index_mask
+ 1;
95 /* Check for overflows. */
96 if (max_entries
< attr
->max_entries
)
97 return ERR_PTR(-E2BIG
);
100 array_size
= sizeof(*array
);
102 array_size
+= (u64
) max_entries
* sizeof(void *);
104 array_size
+= (u64
) max_entries
* elem_size
;
106 /* make sure there is no u32 overflow later in round_up() */
108 if (cost
>= U32_MAX
- PAGE_SIZE
)
109 return ERR_PTR(-ENOMEM
);
111 cost
+= (u64
)attr
->max_entries
* elem_size
* num_possible_cpus();
112 if (cost
>= U32_MAX
- PAGE_SIZE
)
113 return ERR_PTR(-ENOMEM
);
115 cost
= round_up(cost
, PAGE_SIZE
) >> PAGE_SHIFT
;
117 ret
= bpf_map_precharge_memlock(cost
);
121 /* allocate all map elements and zero-initialize them */
122 array
= bpf_map_area_alloc(array_size
, numa_node
);
124 return ERR_PTR(-ENOMEM
);
125 array
->index_mask
= index_mask
;
126 array
->map
.unpriv_array
= unpriv
;
128 /* copy mandatory map attributes */
129 array
->map
.map_type
= attr
->map_type
;
130 array
->map
.key_size
= attr
->key_size
;
131 array
->map
.value_size
= attr
->value_size
;
132 array
->map
.max_entries
= attr
->max_entries
;
133 array
->map
.map_flags
= attr
->map_flags
;
134 array
->map
.numa_node
= numa_node
;
135 array
->map
.pages
= cost
;
136 array
->elem_size
= elem_size
;
138 if (percpu
&& bpf_array_alloc_percpu(array
)) {
139 bpf_map_area_free(array
);
140 return ERR_PTR(-ENOMEM
);
146 /* Called from syscall or from eBPF program */
147 static void *array_map_lookup_elem(struct bpf_map
*map
, void *key
)
149 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
150 u32 index
= *(u32
*)key
;
152 if (unlikely(index
>= array
->map
.max_entries
))
155 return array
->value
+ array
->elem_size
* (index
& array
->index_mask
);
158 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
159 static u32
array_map_gen_lookup(struct bpf_map
*map
, struct bpf_insn
*insn_buf
)
161 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
162 struct bpf_insn
*insn
= insn_buf
;
163 u32 elem_size
= round_up(map
->value_size
, 8);
164 const int ret
= BPF_REG_0
;
165 const int map_ptr
= BPF_REG_1
;
166 const int index
= BPF_REG_2
;
168 *insn
++ = BPF_ALU64_IMM(BPF_ADD
, map_ptr
, offsetof(struct bpf_array
, value
));
169 *insn
++ = BPF_LDX_MEM(BPF_W
, ret
, index
, 0);
170 if (map
->unpriv_array
) {
171 *insn
++ = BPF_JMP_IMM(BPF_JGE
, ret
, map
->max_entries
, 4);
172 *insn
++ = BPF_ALU32_IMM(BPF_AND
, ret
, array
->index_mask
);
174 *insn
++ = BPF_JMP_IMM(BPF_JGE
, ret
, map
->max_entries
, 3);
177 if (is_power_of_2(elem_size
)) {
178 *insn
++ = BPF_ALU64_IMM(BPF_LSH
, ret
, ilog2(elem_size
));
180 *insn
++ = BPF_ALU64_IMM(BPF_MUL
, ret
, elem_size
);
182 *insn
++ = BPF_ALU64_REG(BPF_ADD
, ret
, map_ptr
);
183 *insn
++ = BPF_JMP_IMM(BPF_JA
, 0, 0, 1);
184 *insn
++ = BPF_MOV64_IMM(ret
, 0);
185 return insn
- insn_buf
;
188 /* Called from eBPF program */
189 static void *percpu_array_map_lookup_elem(struct bpf_map
*map
, void *key
)
191 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
192 u32 index
= *(u32
*)key
;
194 if (unlikely(index
>= array
->map
.max_entries
))
197 return this_cpu_ptr(array
->pptrs
[index
& array
->index_mask
]);
200 int bpf_percpu_array_copy(struct bpf_map
*map
, void *key
, void *value
)
202 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
203 u32 index
= *(u32
*)key
;
208 if (unlikely(index
>= array
->map
.max_entries
))
211 /* per_cpu areas are zero-filled and bpf programs can only
212 * access 'value_size' of them, so copying rounded areas
213 * will not leak any kernel data
215 size
= round_up(map
->value_size
, 8);
217 pptr
= array
->pptrs
[index
& array
->index_mask
];
218 for_each_possible_cpu(cpu
) {
219 bpf_long_memcpy(value
+ off
, per_cpu_ptr(pptr
, cpu
), size
);
226 /* Called from syscall */
227 static int array_map_get_next_key(struct bpf_map
*map
, void *key
, void *next_key
)
229 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
230 u32 index
= key
? *(u32
*)key
: U32_MAX
;
231 u32
*next
= (u32
*)next_key
;
233 if (index
>= array
->map
.max_entries
) {
238 if (index
== array
->map
.max_entries
- 1)
245 /* Called from syscall or from eBPF program */
246 static int array_map_update_elem(struct bpf_map
*map
, void *key
, void *value
,
249 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
250 u32 index
= *(u32
*)key
;
252 if (unlikely(map_flags
> BPF_EXIST
))
256 if (unlikely(index
>= array
->map
.max_entries
))
257 /* all elements were pre-allocated, cannot insert a new one */
260 if (unlikely(map_flags
== BPF_NOEXIST
))
261 /* all elements already exist */
264 if (array
->map
.map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
)
265 memcpy(this_cpu_ptr(array
->pptrs
[index
& array
->index_mask
]),
266 value
, map
->value_size
);
268 memcpy(array
->value
+
269 array
->elem_size
* (index
& array
->index_mask
),
270 value
, map
->value_size
);
274 int bpf_percpu_array_update(struct bpf_map
*map
, void *key
, void *value
,
277 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
278 u32 index
= *(u32
*)key
;
283 if (unlikely(map_flags
> BPF_EXIST
))
287 if (unlikely(index
>= array
->map
.max_entries
))
288 /* all elements were pre-allocated, cannot insert a new one */
291 if (unlikely(map_flags
== BPF_NOEXIST
))
292 /* all elements already exist */
295 /* the user space will provide round_up(value_size, 8) bytes that
296 * will be copied into per-cpu area. bpf programs can only access
297 * value_size of it. During lookup the same extra bytes will be
298 * returned or zeros which were zero-filled by percpu_alloc,
299 * so no kernel data leaks possible
301 size
= round_up(map
->value_size
, 8);
303 pptr
= array
->pptrs
[index
& array
->index_mask
];
304 for_each_possible_cpu(cpu
) {
305 bpf_long_memcpy(per_cpu_ptr(pptr
, cpu
), value
+ off
, size
);
312 /* Called from syscall or from eBPF program */
313 static int array_map_delete_elem(struct bpf_map
*map
, void *key
)
318 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
319 static void array_map_free(struct bpf_map
*map
)
321 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
323 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
324 * so the programs (can be more than one that used this map) were
325 * disconnected from events. Wait for outstanding programs to complete
330 if (array
->map
.map_type
== BPF_MAP_TYPE_PERCPU_ARRAY
)
331 bpf_array_free_percpu(array
);
333 bpf_map_area_free(array
);
336 const struct bpf_map_ops array_map_ops
= {
337 .map_alloc
= array_map_alloc
,
338 .map_free
= array_map_free
,
339 .map_get_next_key
= array_map_get_next_key
,
340 .map_lookup_elem
= array_map_lookup_elem
,
341 .map_update_elem
= array_map_update_elem
,
342 .map_delete_elem
= array_map_delete_elem
,
343 .map_gen_lookup
= array_map_gen_lookup
,
346 const struct bpf_map_ops percpu_array_map_ops
= {
347 .map_alloc
= array_map_alloc
,
348 .map_free
= array_map_free
,
349 .map_get_next_key
= array_map_get_next_key
,
350 .map_lookup_elem
= percpu_array_map_lookup_elem
,
351 .map_update_elem
= array_map_update_elem
,
352 .map_delete_elem
= array_map_delete_elem
,
355 static struct bpf_map
*fd_array_map_alloc(union bpf_attr
*attr
)
357 /* only file descriptors can be stored in this type of map */
358 if (attr
->value_size
!= sizeof(u32
))
359 return ERR_PTR(-EINVAL
);
360 return array_map_alloc(attr
);
363 static void fd_array_map_free(struct bpf_map
*map
)
365 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
370 /* make sure it's empty */
371 for (i
= 0; i
< array
->map
.max_entries
; i
++)
372 BUG_ON(array
->ptrs
[i
] != NULL
);
374 bpf_map_area_free(array
);
377 static void *fd_array_map_lookup_elem(struct bpf_map
*map
, void *key
)
382 /* only called from syscall */
383 int bpf_fd_array_map_lookup_elem(struct bpf_map
*map
, void *key
, u32
*value
)
388 if (!map
->ops
->map_fd_sys_lookup_elem
)
392 elem
= array_map_lookup_elem(map
, key
);
393 if (elem
&& (ptr
= READ_ONCE(*elem
)))
394 *value
= map
->ops
->map_fd_sys_lookup_elem(ptr
);
402 /* only called from syscall */
403 int bpf_fd_array_map_update_elem(struct bpf_map
*map
, struct file
*map_file
,
404 void *key
, void *value
, u64 map_flags
)
406 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
407 void *new_ptr
, *old_ptr
;
408 u32 index
= *(u32
*)key
, ufd
;
410 if (map_flags
!= BPF_ANY
)
413 if (index
>= array
->map
.max_entries
)
417 new_ptr
= map
->ops
->map_fd_get_ptr(map
, map_file
, ufd
);
419 return PTR_ERR(new_ptr
);
421 old_ptr
= xchg(array
->ptrs
+ index
, new_ptr
);
423 map
->ops
->map_fd_put_ptr(old_ptr
);
428 static int fd_array_map_delete_elem(struct bpf_map
*map
, void *key
)
430 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
432 u32 index
= *(u32
*)key
;
434 if (index
>= array
->map
.max_entries
)
437 old_ptr
= xchg(array
->ptrs
+ index
, NULL
);
439 map
->ops
->map_fd_put_ptr(old_ptr
);
446 static void *prog_fd_array_get_ptr(struct bpf_map
*map
,
447 struct file
*map_file
, int fd
)
449 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
450 struct bpf_prog
*prog
= bpf_prog_get(fd
);
455 if (!bpf_prog_array_compatible(array
, prog
)) {
457 return ERR_PTR(-EINVAL
);
463 static void prog_fd_array_put_ptr(void *ptr
)
468 static u32
prog_fd_array_sys_lookup_elem(void *ptr
)
470 return ((struct bpf_prog
*)ptr
)->aux
->id
;
473 /* decrement refcnt of all bpf_progs that are stored in this map */
474 static void bpf_fd_array_map_clear(struct bpf_map
*map
)
476 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
479 for (i
= 0; i
< array
->map
.max_entries
; i
++)
480 fd_array_map_delete_elem(map
, &i
);
483 const struct bpf_map_ops prog_array_map_ops
= {
484 .map_alloc
= fd_array_map_alloc
,
485 .map_free
= fd_array_map_free
,
486 .map_get_next_key
= array_map_get_next_key
,
487 .map_lookup_elem
= fd_array_map_lookup_elem
,
488 .map_delete_elem
= fd_array_map_delete_elem
,
489 .map_fd_get_ptr
= prog_fd_array_get_ptr
,
490 .map_fd_put_ptr
= prog_fd_array_put_ptr
,
491 .map_fd_sys_lookup_elem
= prog_fd_array_sys_lookup_elem
,
492 .map_release_uref
= bpf_fd_array_map_clear
,
495 static struct bpf_event_entry
*bpf_event_entry_gen(struct file
*perf_file
,
496 struct file
*map_file
)
498 struct bpf_event_entry
*ee
;
500 ee
= kzalloc(sizeof(*ee
), GFP_ATOMIC
);
502 ee
->event
= perf_file
->private_data
;
503 ee
->perf_file
= perf_file
;
504 ee
->map_file
= map_file
;
510 static void __bpf_event_entry_free(struct rcu_head
*rcu
)
512 struct bpf_event_entry
*ee
;
514 ee
= container_of(rcu
, struct bpf_event_entry
, rcu
);
519 static void bpf_event_entry_free_rcu(struct bpf_event_entry
*ee
)
521 call_rcu(&ee
->rcu
, __bpf_event_entry_free
);
524 static void *perf_event_fd_array_get_ptr(struct bpf_map
*map
,
525 struct file
*map_file
, int fd
)
527 struct bpf_event_entry
*ee
;
528 struct perf_event
*event
;
529 struct file
*perf_file
;
532 perf_file
= perf_event_get(fd
);
533 if (IS_ERR(perf_file
))
536 ee
= ERR_PTR(-EOPNOTSUPP
);
537 event
= perf_file
->private_data
;
538 if (perf_event_read_local(event
, &value
, NULL
, NULL
) == -EOPNOTSUPP
)
541 ee
= bpf_event_entry_gen(perf_file
, map_file
);
544 ee
= ERR_PTR(-ENOMEM
);
550 static void perf_event_fd_array_put_ptr(void *ptr
)
552 bpf_event_entry_free_rcu(ptr
);
555 static void perf_event_fd_array_release(struct bpf_map
*map
,
556 struct file
*map_file
)
558 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
559 struct bpf_event_entry
*ee
;
563 for (i
= 0; i
< array
->map
.max_entries
; i
++) {
564 ee
= READ_ONCE(array
->ptrs
[i
]);
565 if (ee
&& ee
->map_file
== map_file
)
566 fd_array_map_delete_elem(map
, &i
);
571 const struct bpf_map_ops perf_event_array_map_ops
= {
572 .map_alloc
= fd_array_map_alloc
,
573 .map_free
= fd_array_map_free
,
574 .map_get_next_key
= array_map_get_next_key
,
575 .map_lookup_elem
= fd_array_map_lookup_elem
,
576 .map_delete_elem
= fd_array_map_delete_elem
,
577 .map_fd_get_ptr
= perf_event_fd_array_get_ptr
,
578 .map_fd_put_ptr
= perf_event_fd_array_put_ptr
,
579 .map_release
= perf_event_fd_array_release
,
582 #ifdef CONFIG_CGROUPS
583 static void *cgroup_fd_array_get_ptr(struct bpf_map
*map
,
584 struct file
*map_file
/* not used */,
587 return cgroup_get_from_fd(fd
);
590 static void cgroup_fd_array_put_ptr(void *ptr
)
592 /* cgroup_put free cgrp after a rcu grace period */
596 static void cgroup_fd_array_free(struct bpf_map
*map
)
598 bpf_fd_array_map_clear(map
);
599 fd_array_map_free(map
);
602 const struct bpf_map_ops cgroup_array_map_ops
= {
603 .map_alloc
= fd_array_map_alloc
,
604 .map_free
= cgroup_fd_array_free
,
605 .map_get_next_key
= array_map_get_next_key
,
606 .map_lookup_elem
= fd_array_map_lookup_elem
,
607 .map_delete_elem
= fd_array_map_delete_elem
,
608 .map_fd_get_ptr
= cgroup_fd_array_get_ptr
,
609 .map_fd_put_ptr
= cgroup_fd_array_put_ptr
,
613 static struct bpf_map
*array_of_map_alloc(union bpf_attr
*attr
)
615 struct bpf_map
*map
, *inner_map_meta
;
617 inner_map_meta
= bpf_map_meta_alloc(attr
->inner_map_fd
);
618 if (IS_ERR(inner_map_meta
))
619 return inner_map_meta
;
621 map
= fd_array_map_alloc(attr
);
623 bpf_map_meta_free(inner_map_meta
);
627 map
->inner_map_meta
= inner_map_meta
;
632 static void array_of_map_free(struct bpf_map
*map
)
634 /* map->inner_map_meta is only accessed by syscall which
635 * is protected by fdget/fdput.
637 bpf_map_meta_free(map
->inner_map_meta
);
638 bpf_fd_array_map_clear(map
);
639 fd_array_map_free(map
);
642 static void *array_of_map_lookup_elem(struct bpf_map
*map
, void *key
)
644 struct bpf_map
**inner_map
= array_map_lookup_elem(map
, key
);
649 return READ_ONCE(*inner_map
);
652 static u32
array_of_map_gen_lookup(struct bpf_map
*map
,
653 struct bpf_insn
*insn_buf
)
655 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
656 u32 elem_size
= round_up(map
->value_size
, 8);
657 struct bpf_insn
*insn
= insn_buf
;
658 const int ret
= BPF_REG_0
;
659 const int map_ptr
= BPF_REG_1
;
660 const int index
= BPF_REG_2
;
662 *insn
++ = BPF_ALU64_IMM(BPF_ADD
, map_ptr
, offsetof(struct bpf_array
, value
));
663 *insn
++ = BPF_LDX_MEM(BPF_W
, ret
, index
, 0);
664 if (map
->unpriv_array
) {
665 *insn
++ = BPF_JMP_IMM(BPF_JGE
, ret
, map
->max_entries
, 6);
666 *insn
++ = BPF_ALU32_IMM(BPF_AND
, ret
, array
->index_mask
);
668 *insn
++ = BPF_JMP_IMM(BPF_JGE
, ret
, map
->max_entries
, 5);
670 if (is_power_of_2(elem_size
))
671 *insn
++ = BPF_ALU64_IMM(BPF_LSH
, ret
, ilog2(elem_size
));
673 *insn
++ = BPF_ALU64_IMM(BPF_MUL
, ret
, elem_size
);
674 *insn
++ = BPF_ALU64_REG(BPF_ADD
, ret
, map_ptr
);
675 *insn
++ = BPF_LDX_MEM(BPF_DW
, ret
, ret
, 0);
676 *insn
++ = BPF_JMP_IMM(BPF_JEQ
, ret
, 0, 1);
677 *insn
++ = BPF_JMP_IMM(BPF_JA
, 0, 0, 1);
678 *insn
++ = BPF_MOV64_IMM(ret
, 0);
680 return insn
- insn_buf
;
683 const struct bpf_map_ops array_of_maps_map_ops
= {
684 .map_alloc
= array_of_map_alloc
,
685 .map_free
= array_of_map_free
,
686 .map_get_next_key
= array_map_get_next_key
,
687 .map_lookup_elem
= array_of_map_lookup_elem
,
688 .map_delete_elem
= fd_array_map_delete_elem
,
689 .map_fd_get_ptr
= bpf_map_fd_get_ptr
,
690 .map_fd_put_ptr
= bpf_map_fd_put_ptr
,
691 .map_fd_sys_lookup_elem
= bpf_map_fd_sys_lookup_elem
,
692 .map_gen_lookup
= array_of_map_gen_lookup
,