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]);
27 static int bpf_array_alloc_percpu(struct bpf_array *array)
32 for (i = 0; i < array->map.max_entries; i++) {
33 ptr = __alloc_percpu_gfp(array->elem_size, 8,
34 GFP_USER | __GFP_NOWARN);
36 bpf_array_free_percpu(array);
39 array->pptrs[i] = ptr;
45 /* Called from syscall */
46 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
48 bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
49 struct bpf_array *array;
53 /* check sanity of attributes */
54 if (attr->max_entries == 0 || attr->key_size != 4 ||
55 attr->value_size == 0 || attr->map_flags)
56 return ERR_PTR(-EINVAL);
58 if (attr->value_size > KMALLOC_MAX_SIZE)
59 /* if value_size is bigger, the user space won't be able to
60 * access the elements.
62 return ERR_PTR(-E2BIG);
64 elem_size = round_up(attr->value_size, 8);
66 array_size = sizeof(*array);
68 array_size += (u64) attr->max_entries * sizeof(void *);
70 array_size += (u64) attr->max_entries * elem_size;
72 /* make sure there is no u32 overflow later in round_up() */
73 if (array_size >= U32_MAX - PAGE_SIZE)
74 return ERR_PTR(-ENOMEM);
76 /* allocate all map elements and zero-initialize them */
77 array = bpf_map_area_alloc(array_size);
79 return ERR_PTR(-ENOMEM);
81 /* copy mandatory map attributes */
82 array->map.map_type = attr->map_type;
83 array->map.key_size = attr->key_size;
84 array->map.value_size = attr->value_size;
85 array->map.max_entries = attr->max_entries;
86 array->elem_size = elem_size;
91 array_size += (u64) attr->max_entries * elem_size * num_possible_cpus();
93 if (array_size >= U32_MAX - PAGE_SIZE ||
94 elem_size > PCPU_MIN_UNIT_SIZE || bpf_array_alloc_percpu(array)) {
95 bpf_map_area_free(array);
96 return ERR_PTR(-ENOMEM);
99 array->map.pages = round_up(array_size, PAGE_SIZE) >> PAGE_SHIFT;
104 /* Called from syscall or from eBPF program */
105 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
107 struct bpf_array *array = container_of(map, struct bpf_array, map);
108 u32 index = *(u32 *)key;
110 if (unlikely(index >= array->map.max_entries))
113 return array->value + array->elem_size * index;
116 /* Called from eBPF program */
117 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
119 struct bpf_array *array = container_of(map, struct bpf_array, map);
120 u32 index = *(u32 *)key;
122 if (unlikely(index >= array->map.max_entries))
125 return this_cpu_ptr(array->pptrs[index]);
128 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
130 struct bpf_array *array = container_of(map, struct bpf_array, map);
131 u32 index = *(u32 *)key;
136 if (unlikely(index >= array->map.max_entries))
139 /* per_cpu areas are zero-filled and bpf programs can only
140 * access 'value_size' of them, so copying rounded areas
141 * will not leak any kernel data
143 size = round_up(map->value_size, 8);
145 pptr = array->pptrs[index];
146 for_each_possible_cpu(cpu) {
147 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
154 /* Called from syscall */
155 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
157 struct bpf_array *array = container_of(map, struct bpf_array, map);
158 u32 index = *(u32 *)key;
159 u32 *next = (u32 *)next_key;
161 if (index >= array->map.max_entries) {
166 if (index == array->map.max_entries - 1)
173 /* Called from syscall or from eBPF program */
174 static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
177 struct bpf_array *array = container_of(map, struct bpf_array, map);
178 u32 index = *(u32 *)key;
180 if (unlikely(map_flags > BPF_EXIST))
184 if (unlikely(index >= array->map.max_entries))
185 /* all elements were pre-allocated, cannot insert a new one */
188 if (unlikely(map_flags == BPF_NOEXIST))
189 /* all elements already exist */
192 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
193 memcpy(this_cpu_ptr(array->pptrs[index]),
194 value, map->value_size);
196 memcpy(array->value + array->elem_size * index,
197 value, map->value_size);
201 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
204 struct bpf_array *array = container_of(map, struct bpf_array, map);
205 u32 index = *(u32 *)key;
210 if (unlikely(map_flags > BPF_EXIST))
214 if (unlikely(index >= array->map.max_entries))
215 /* all elements were pre-allocated, cannot insert a new one */
218 if (unlikely(map_flags == BPF_NOEXIST))
219 /* all elements already exist */
222 /* the user space will provide round_up(value_size, 8) bytes that
223 * will be copied into per-cpu area. bpf programs can only access
224 * value_size of it. During lookup the same extra bytes will be
225 * returned or zeros which were zero-filled by percpu_alloc,
226 * so no kernel data leaks possible
228 size = round_up(map->value_size, 8);
230 pptr = array->pptrs[index];
231 for_each_possible_cpu(cpu) {
232 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
239 /* Called from syscall or from eBPF program */
240 static int array_map_delete_elem(struct bpf_map *map, void *key)
245 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
246 static void array_map_free(struct bpf_map *map)
248 struct bpf_array *array = container_of(map, struct bpf_array, map);
250 /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
251 * so the programs (can be more than one that used this map) were
252 * disconnected from events. Wait for outstanding programs to complete
257 if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
258 bpf_array_free_percpu(array);
260 bpf_map_area_free(array);
263 static const struct bpf_map_ops array_ops = {
264 .map_alloc = array_map_alloc,
265 .map_free = array_map_free,
266 .map_get_next_key = array_map_get_next_key,
267 .map_lookup_elem = array_map_lookup_elem,
268 .map_update_elem = array_map_update_elem,
269 .map_delete_elem = array_map_delete_elem,
272 static struct bpf_map_type_list array_type __ro_after_init = {
274 .type = BPF_MAP_TYPE_ARRAY,
277 static const struct bpf_map_ops percpu_array_ops = {
278 .map_alloc = array_map_alloc,
279 .map_free = array_map_free,
280 .map_get_next_key = array_map_get_next_key,
281 .map_lookup_elem = percpu_array_map_lookup_elem,
282 .map_update_elem = array_map_update_elem,
283 .map_delete_elem = array_map_delete_elem,
286 static struct bpf_map_type_list percpu_array_type __ro_after_init = {
287 .ops = &percpu_array_ops,
288 .type = BPF_MAP_TYPE_PERCPU_ARRAY,
291 static int __init register_array_map(void)
293 bpf_register_map_type(&array_type);
294 bpf_register_map_type(&percpu_array_type);
297 late_initcall(register_array_map);
299 static struct bpf_map *fd_array_map_alloc(union bpf_attr *attr)
301 /* only file descriptors can be stored in this type of map */
302 if (attr->value_size != sizeof(u32))
303 return ERR_PTR(-EINVAL);
304 return array_map_alloc(attr);
307 static void fd_array_map_free(struct bpf_map *map)
309 struct bpf_array *array = container_of(map, struct bpf_array, map);
314 /* make sure it's empty */
315 for (i = 0; i < array->map.max_entries; i++)
316 BUG_ON(array->ptrs[i] != NULL);
318 bpf_map_area_free(array);
321 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
326 /* only called from syscall */
327 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
328 void *key, void *value, u64 map_flags)
330 struct bpf_array *array = container_of(map, struct bpf_array, map);
331 void *new_ptr, *old_ptr;
332 u32 index = *(u32 *)key, ufd;
334 if (map_flags != BPF_ANY)
337 if (index >= array->map.max_entries)
341 new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
343 return PTR_ERR(new_ptr);
345 old_ptr = xchg(array->ptrs + index, new_ptr);
347 map->ops->map_fd_put_ptr(old_ptr);
352 static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
354 struct bpf_array *array = container_of(map, struct bpf_array, map);
356 u32 index = *(u32 *)key;
358 if (index >= array->map.max_entries)
361 old_ptr = xchg(array->ptrs + index, NULL);
363 map->ops->map_fd_put_ptr(old_ptr);
370 static void *prog_fd_array_get_ptr(struct bpf_map *map,
371 struct file *map_file, int fd)
373 struct bpf_array *array = container_of(map, struct bpf_array, map);
374 struct bpf_prog *prog = bpf_prog_get(fd);
379 if (!bpf_prog_array_compatible(array, prog)) {
381 return ERR_PTR(-EINVAL);
387 static void prog_fd_array_put_ptr(void *ptr)
392 /* decrement refcnt of all bpf_progs that are stored in this map */
393 void bpf_fd_array_map_clear(struct bpf_map *map)
395 struct bpf_array *array = container_of(map, struct bpf_array, map);
398 for (i = 0; i < array->map.max_entries; i++)
399 fd_array_map_delete_elem(map, &i);
402 static const struct bpf_map_ops prog_array_ops = {
403 .map_alloc = fd_array_map_alloc,
404 .map_free = fd_array_map_free,
405 .map_get_next_key = array_map_get_next_key,
406 .map_lookup_elem = fd_array_map_lookup_elem,
407 .map_delete_elem = fd_array_map_delete_elem,
408 .map_fd_get_ptr = prog_fd_array_get_ptr,
409 .map_fd_put_ptr = prog_fd_array_put_ptr,
412 static struct bpf_map_type_list prog_array_type __ro_after_init = {
413 .ops = &prog_array_ops,
414 .type = BPF_MAP_TYPE_PROG_ARRAY,
417 static int __init register_prog_array_map(void)
419 bpf_register_map_type(&prog_array_type);
422 late_initcall(register_prog_array_map);
424 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
425 struct file *map_file)
427 struct bpf_event_entry *ee;
429 ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
431 ee->event = perf_file->private_data;
432 ee->perf_file = perf_file;
433 ee->map_file = map_file;
439 static void __bpf_event_entry_free(struct rcu_head *rcu)
441 struct bpf_event_entry *ee;
443 ee = container_of(rcu, struct bpf_event_entry, rcu);
448 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
450 call_rcu(&ee->rcu, __bpf_event_entry_free);
453 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
454 struct file *map_file, int fd)
456 const struct perf_event_attr *attr;
457 struct bpf_event_entry *ee;
458 struct perf_event *event;
459 struct file *perf_file;
461 perf_file = perf_event_get(fd);
462 if (IS_ERR(perf_file))
465 event = perf_file->private_data;
466 ee = ERR_PTR(-EINVAL);
468 attr = perf_event_attrs(event);
469 if (IS_ERR(attr) || attr->inherit)
472 switch (attr->type) {
473 case PERF_TYPE_SOFTWARE:
474 if (attr->config != PERF_COUNT_SW_BPF_OUTPUT)
478 case PERF_TYPE_HARDWARE:
479 ee = bpf_event_entry_gen(perf_file, map_file);
482 ee = ERR_PTR(-ENOMEM);
493 static void perf_event_fd_array_put_ptr(void *ptr)
495 bpf_event_entry_free_rcu(ptr);
498 static void perf_event_fd_array_release(struct bpf_map *map,
499 struct file *map_file)
501 struct bpf_array *array = container_of(map, struct bpf_array, map);
502 struct bpf_event_entry *ee;
506 for (i = 0; i < array->map.max_entries; i++) {
507 ee = READ_ONCE(array->ptrs[i]);
508 if (ee && ee->map_file == map_file)
509 fd_array_map_delete_elem(map, &i);
514 static const struct bpf_map_ops perf_event_array_ops = {
515 .map_alloc = fd_array_map_alloc,
516 .map_free = fd_array_map_free,
517 .map_get_next_key = array_map_get_next_key,
518 .map_lookup_elem = fd_array_map_lookup_elem,
519 .map_delete_elem = fd_array_map_delete_elem,
520 .map_fd_get_ptr = perf_event_fd_array_get_ptr,
521 .map_fd_put_ptr = perf_event_fd_array_put_ptr,
522 .map_release = perf_event_fd_array_release,
525 static struct bpf_map_type_list perf_event_array_type __ro_after_init = {
526 .ops = &perf_event_array_ops,
527 .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
530 static int __init register_perf_event_array_map(void)
532 bpf_register_map_type(&perf_event_array_type);
535 late_initcall(register_perf_event_array_map);
537 #ifdef CONFIG_CGROUPS
538 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
539 struct file *map_file /* not used */,
542 return cgroup_get_from_fd(fd);
545 static void cgroup_fd_array_put_ptr(void *ptr)
547 /* cgroup_put free cgrp after a rcu grace period */
551 static void cgroup_fd_array_free(struct bpf_map *map)
553 bpf_fd_array_map_clear(map);
554 fd_array_map_free(map);
557 static const struct bpf_map_ops cgroup_array_ops = {
558 .map_alloc = fd_array_map_alloc,
559 .map_free = cgroup_fd_array_free,
560 .map_get_next_key = array_map_get_next_key,
561 .map_lookup_elem = fd_array_map_lookup_elem,
562 .map_delete_elem = fd_array_map_delete_elem,
563 .map_fd_get_ptr = cgroup_fd_array_get_ptr,
564 .map_fd_put_ptr = cgroup_fd_array_put_ptr,
567 static struct bpf_map_type_list cgroup_array_type __ro_after_init = {
568 .ops = &cgroup_array_ops,
569 .type = BPF_MAP_TYPE_CGROUP_ARRAY,
572 static int __init register_cgroup_array_map(void)
574 bpf_register_map_type(&cgroup_array_type);
577 late_initcall(register_cgroup_array_map);