1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of version 2 of the GNU General Public
5 * License as published by the Free Software Foundation.
7 * This program is distributed in the hope that it will be useful, but
8 * WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
10 * General Public License for more details.
12 #include <linux/bpf.h>
13 #include <linux/err.h>
14 #include <linux/slab.h>
16 #include <linux/filter.h>
17 #include <linux/perf_event.h>
19 static void bpf_array_free_percpu(struct bpf_array *array)
23 for (i = 0; i < array->map.max_entries; i++) {
24 free_percpu(array->pptrs[i]);
29 static int bpf_array_alloc_percpu(struct bpf_array *array)
34 for (i = 0; i < array->map.max_entries; i++) {
35 ptr = __alloc_percpu_gfp(array->elem_size, 8,
36 GFP_USER | __GFP_NOWARN);
38 bpf_array_free_percpu(array);
41 array->pptrs[i] = ptr;
48 /* Called from syscall */
49 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
51 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
52 u32 elem_size, index_mask, max_entries;
53 bool unpriv = !capable(CAP_SYS_ADMIN);
54 u64 cost, array_size, mask64;
55 struct bpf_array *array;
58 /* check sanity of attributes */
59 if (attr->max_entries == 0 || attr->key_size != 4 ||
60 attr->value_size == 0 || attr->map_flags)
61 return ERR_PTR(-EINVAL);
63 if (attr->value_size >= 1 << (KMALLOC_SHIFT_MAX - 1))
64 /* if value_size is bigger, the user space won't be able to
65 * access the elements.
67 return ERR_PTR(-E2BIG);
69 elem_size = round_up(attr->value_size, 8);
71 max_entries = attr->max_entries;
73 /* On 32 bit archs roundup_pow_of_two() with max_entries that has
74 * upper most bit set in u32 space is undefined behavior due to
75 * resulting 1U << 32, so do it manually here in u64 space.
77 mask64 = fls_long(max_entries - 1);
78 mask64 = 1ULL << mask64;
83 /* round up array size to nearest power of 2,
84 * since cpu will speculate within index_mask limits
86 max_entries = index_mask + 1;
87 /* Check for overflows. */
88 if (max_entries < attr->max_entries)
89 return ERR_PTR(-E2BIG);
92 array_size = sizeof(*array);
94 array_size += (u64) max_entries * sizeof(void *);
96 array_size += (u64) max_entries * elem_size;
98 /* make sure there is no u32 overflow later in round_up() */
100 if (cost >= U32_MAX - PAGE_SIZE)
101 return ERR_PTR(-ENOMEM);
103 cost += (u64)attr->max_entries * elem_size * num_possible_cpus();
104 if (cost >= U32_MAX - PAGE_SIZE)
105 return ERR_PTR(-ENOMEM);
107 cost = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
109 ret = bpf_map_precharge_memlock(cost);
113 /* allocate all map elements and zero-initialize them */
114 array = bpf_map_area_alloc(array_size);
116 return ERR_PTR(-ENOMEM);
117 array->index_mask = index_mask;
118 array->map.unpriv_array = unpriv;
120 /* copy mandatory map attributes */
121 array->map.map_type = attr->map_type;
122 array->map.key_size = attr->key_size;
123 array->map.value_size = attr->value_size;
124 array->map.max_entries = attr->max_entries;
125 array->map.map_flags = attr->map_flags;
126 array->map.pages = cost;
127 array->elem_size = elem_size;
130 (elem_size > PCPU_MIN_UNIT_SIZE ||
131 bpf_array_alloc_percpu(array))) {
132 bpf_map_area_free(array);
133 return ERR_PTR(-ENOMEM);
139 /* Called from syscall or from eBPF program */
140 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
142 struct bpf_array *array = container_of(map, struct bpf_array, map);
143 u32 index = *(u32 *)key;
145 if (unlikely(index >= array->map.max_entries))
148 return array->value + array->elem_size * (index & array->index_mask);
151 /* Called from eBPF program */
152 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
154 struct bpf_array *array = container_of(map, struct bpf_array, map);
155 u32 index = *(u32 *)key;
157 if (unlikely(index >= array->map.max_entries))
160 return this_cpu_ptr(array->pptrs[index & array->index_mask]);
163 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
165 struct bpf_array *array = container_of(map, struct bpf_array, map);
166 u32 index = *(u32 *)key;
171 if (unlikely(index >= array->map.max_entries))
174 /* per_cpu areas are zero-filled and bpf programs can only
175 * access 'value_size' of them, so copying rounded areas
176 * will not leak any kernel data
178 size = round_up(map->value_size, 8);
180 pptr = array->pptrs[index & array->index_mask];
181 for_each_possible_cpu(cpu) {
182 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
189 /* Called from syscall */
190 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
192 struct bpf_array *array = container_of(map, struct bpf_array, map);
193 u32 index = key ? *(u32 *)key : U32_MAX;
194 u32 *next = (u32 *)next_key;
196 if (index >= array->map.max_entries) {
201 if (index == array->map.max_entries - 1)
208 /* Called from syscall or from eBPF program */
209 static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
212 struct bpf_array *array = container_of(map, struct bpf_array, map);
213 u32 index = *(u32 *)key;
215 if (unlikely(map_flags > BPF_EXIST))
219 if (unlikely(index >= array->map.max_entries))
220 /* all elements were pre-allocated, cannot insert a new one */
223 if (unlikely(map_flags == BPF_NOEXIST))
224 /* all elements already exist */
227 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
228 memcpy(this_cpu_ptr(array->pptrs[index & array->index_mask]),
229 value, map->value_size);
231 memcpy(array->value +
232 array->elem_size * (index & array->index_mask),
233 value, map->value_size);
237 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
240 struct bpf_array *array = container_of(map, struct bpf_array, map);
241 u32 index = *(u32 *)key;
246 if (unlikely(map_flags > BPF_EXIST))
250 if (unlikely(index >= array->map.max_entries))
251 /* all elements were pre-allocated, cannot insert a new one */
254 if (unlikely(map_flags == BPF_NOEXIST))
255 /* all elements already exist */
258 /* the user space will provide round_up(value_size, 8) bytes that
259 * will be copied into per-cpu area. bpf programs can only access
260 * value_size of it. During lookup the same extra bytes will be
261 * returned or zeros which were zero-filled by percpu_alloc,
262 * so no kernel data leaks possible
264 size = round_up(map->value_size, 8);
266 pptr = array->pptrs[index & array->index_mask];
267 for_each_possible_cpu(cpu) {
268 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
275 /* Called from syscall or from eBPF program */
276 static int array_map_delete_elem(struct bpf_map *map, void *key)
281 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
282 static void array_map_free(struct bpf_map *map)
284 struct bpf_array *array = container_of(map, struct bpf_array, map);
286 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
287 * so the programs (can be more than one that used this map) were
288 * disconnected from events. Wait for outstanding programs to complete
293 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
294 bpf_array_free_percpu(array);
296 bpf_map_area_free(array);
299 static const struct bpf_map_ops array_ops = {
300 .map_alloc = array_map_alloc,
301 .map_free = array_map_free,
302 .map_get_next_key = array_map_get_next_key,
303 .map_lookup_elem = array_map_lookup_elem,
304 .map_update_elem = array_map_update_elem,
305 .map_delete_elem = array_map_delete_elem,
308 static struct bpf_map_type_list array_type __read_mostly = {
310 .type = BPF_MAP_TYPE_ARRAY,
313 static const struct bpf_map_ops percpu_array_ops = {
314 .map_alloc = array_map_alloc,
315 .map_free = array_map_free,
316 .map_get_next_key = array_map_get_next_key,
317 .map_lookup_elem = percpu_array_map_lookup_elem,
318 .map_update_elem = array_map_update_elem,
319 .map_delete_elem = array_map_delete_elem,
322 static struct bpf_map_type_list percpu_array_type __read_mostly = {
323 .ops = &percpu_array_ops,
324 .type = BPF_MAP_TYPE_PERCPU_ARRAY,
327 static int __init register_array_map(void)
329 bpf_register_map_type(&array_type);
330 bpf_register_map_type(&percpu_array_type);
333 late_initcall(register_array_map);
335 static struct bpf_map *fd_array_map_alloc(union bpf_attr *attr)
337 /* only file descriptors can be stored in this type of map */
338 if (attr->value_size != sizeof(u32))
339 return ERR_PTR(-EINVAL);
340 return array_map_alloc(attr);
343 static void fd_array_map_free(struct bpf_map *map)
345 struct bpf_array *array = container_of(map, struct bpf_array, map);
350 /* make sure it's empty */
351 for (i = 0; i < array->map.max_entries; i++)
352 BUG_ON(array->ptrs[i] != NULL);
354 bpf_map_area_free(array);
357 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
362 /* only called from syscall */
363 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
364 void *key, void *value, u64 map_flags)
366 struct bpf_array *array = container_of(map, struct bpf_array, map);
367 void *new_ptr, *old_ptr;
368 u32 index = *(u32 *)key, ufd;
370 if (map_flags != BPF_ANY)
373 if (index >= array->map.max_entries)
377 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
379 return PTR_ERR(new_ptr);
381 old_ptr = xchg(array->ptrs + index, new_ptr);
383 map->ops->map_fd_put_ptr(old_ptr);
388 static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
390 struct bpf_array *array = container_of(map, struct bpf_array, map);
392 u32 index = *(u32 *)key;
394 if (index >= array->map.max_entries)
397 old_ptr = xchg(array->ptrs + index, NULL);
399 map->ops->map_fd_put_ptr(old_ptr);
406 static void *prog_fd_array_get_ptr(struct bpf_map *map,
407 struct file *map_file, int fd)
409 struct bpf_array *array = container_of(map, struct bpf_array, map);
410 struct bpf_prog *prog = bpf_prog_get(fd);
415 if (!bpf_prog_array_compatible(array, prog)) {
417 return ERR_PTR(-EINVAL);
423 static void prog_fd_array_put_ptr(void *ptr)
428 /* decrement refcnt of all bpf_progs that are stored in this map */
429 void bpf_fd_array_map_clear(struct bpf_map *map)
431 struct bpf_array *array = container_of(map, struct bpf_array, map);
434 for (i = 0; i < array->map.max_entries; i++)
435 fd_array_map_delete_elem(map, &i);
438 static const struct bpf_map_ops prog_array_ops = {
439 .map_alloc = fd_array_map_alloc,
440 .map_free = fd_array_map_free,
441 .map_get_next_key = array_map_get_next_key,
442 .map_lookup_elem = fd_array_map_lookup_elem,
443 .map_delete_elem = fd_array_map_delete_elem,
444 .map_fd_get_ptr = prog_fd_array_get_ptr,
445 .map_fd_put_ptr = prog_fd_array_put_ptr,
448 static struct bpf_map_type_list prog_array_type __read_mostly = {
449 .ops = &prog_array_ops,
450 .type = BPF_MAP_TYPE_PROG_ARRAY,
453 static int __init register_prog_array_map(void)
455 bpf_register_map_type(&prog_array_type);
458 late_initcall(register_prog_array_map);
460 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
461 struct file *map_file)
463 struct bpf_event_entry *ee;
465 ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
467 ee->event = perf_file->private_data;
468 ee->perf_file = perf_file;
469 ee->map_file = map_file;
475 static void __bpf_event_entry_free(struct rcu_head *rcu)
477 struct bpf_event_entry *ee;
479 ee = container_of(rcu, struct bpf_event_entry, rcu);
484 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
486 call_rcu(&ee->rcu, __bpf_event_entry_free);
489 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
490 struct file *map_file, int fd)
492 const struct perf_event_attr *attr;
493 struct bpf_event_entry *ee;
494 struct perf_event *event;
495 struct file *perf_file;
497 perf_file = perf_event_get(fd);
498 if (IS_ERR(perf_file))
501 event = perf_file->private_data;
502 ee = ERR_PTR(-EINVAL);
504 attr = perf_event_attrs(event);
505 if (IS_ERR(attr) || attr->inherit)
508 switch (attr->type) {
509 case PERF_TYPE_SOFTWARE:
510 if (attr->config != PERF_COUNT_SW_BPF_OUTPUT)
514 case PERF_TYPE_HARDWARE:
515 ee = bpf_event_entry_gen(perf_file, map_file);
518 ee = ERR_PTR(-ENOMEM);
529 static void perf_event_fd_array_put_ptr(void *ptr)
531 bpf_event_entry_free_rcu(ptr);
534 static void perf_event_fd_array_release(struct bpf_map *map,
535 struct file *map_file)
537 struct bpf_array *array = container_of(map, struct bpf_array, map);
538 struct bpf_event_entry *ee;
542 for (i = 0; i < array->map.max_entries; i++) {
543 ee = READ_ONCE(array->ptrs[i]);
544 if (ee && ee->map_file == map_file)
545 fd_array_map_delete_elem(map, &i);
550 static const struct bpf_map_ops perf_event_array_ops = {
551 .map_alloc = fd_array_map_alloc,
552 .map_free = fd_array_map_free,
553 .map_get_next_key = array_map_get_next_key,
554 .map_lookup_elem = fd_array_map_lookup_elem,
555 .map_delete_elem = fd_array_map_delete_elem,
556 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
557 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
558 .map_release = perf_event_fd_array_release,
561 static struct bpf_map_type_list perf_event_array_type __read_mostly = {
562 .ops = &perf_event_array_ops,
563 .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
566 static int __init register_perf_event_array_map(void)
568 bpf_register_map_type(&perf_event_array_type);
571 late_initcall(register_perf_event_array_map);
573 #ifdef CONFIG_CGROUPS
574 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
575 struct file *map_file /* not used */,
578 return cgroup_get_from_fd(fd);
581 static void cgroup_fd_array_put_ptr(void *ptr)
583 /* cgroup_put free cgrp after a rcu grace period */
587 static void cgroup_fd_array_free(struct bpf_map *map)
589 bpf_fd_array_map_clear(map);
590 fd_array_map_free(map);
593 static const struct bpf_map_ops cgroup_array_ops = {
594 .map_alloc = fd_array_map_alloc,
595 .map_free = cgroup_fd_array_free,
596 .map_get_next_key = array_map_get_next_key,
597 .map_lookup_elem = fd_array_map_lookup_elem,
598 .map_delete_elem = fd_array_map_delete_elem,
599 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
600 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
603 static struct bpf_map_type_list cgroup_array_type __read_mostly = {
604 .ops = &cgroup_array_ops,
605 .type = BPF_MAP_TYPE_CGROUP_ARRAY,
608 static int __init register_cgroup_array_map(void)
610 bpf_register_map_type(&cgroup_array_type);
613 late_initcall(register_cgroup_array_map);