xref: /openbmc/linux/kernel/bpf/local_storage.c (revision 4b0aaacee51eb6592a03fdefd5ce97558518e291)
1 //SPDX-License-Identifier: GPL-2.0
2 #include <linux/bpf-cgroup.h>
3 #include <linux/bpf.h>
4 #include <linux/bug.h>
5 #include <linux/filter.h>
6 #include <linux/mm.h>
7 #include <linux/rbtree.h>
8 #include <linux/slab.h>
9 
10 DEFINE_PER_CPU(void*, bpf_cgroup_storage);
11 
12 #ifdef CONFIG_CGROUP_BPF
13 
14 #define LOCAL_STORAGE_CREATE_FLAG_MASK					\
15 	(BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
16 
17 struct bpf_cgroup_storage_map {
18 	struct bpf_map map;
19 
20 	spinlock_t lock;
21 	struct bpf_prog *prog;
22 	struct rb_root root;
23 	struct list_head list;
24 };
25 
26 static struct bpf_cgroup_storage_map *map_to_storage(struct bpf_map *map)
27 {
28 	return container_of(map, struct bpf_cgroup_storage_map, map);
29 }
30 
31 static int bpf_cgroup_storage_key_cmp(
32 	const struct bpf_cgroup_storage_key *key1,
33 	const struct bpf_cgroup_storage_key *key2)
34 {
35 	if (key1->cgroup_inode_id < key2->cgroup_inode_id)
36 		return -1;
37 	else if (key1->cgroup_inode_id > key2->cgroup_inode_id)
38 		return 1;
39 	else if (key1->attach_type < key2->attach_type)
40 		return -1;
41 	else if (key1->attach_type > key2->attach_type)
42 		return 1;
43 	return 0;
44 }
45 
46 static struct bpf_cgroup_storage *cgroup_storage_lookup(
47 	struct bpf_cgroup_storage_map *map, struct bpf_cgroup_storage_key *key,
48 	bool locked)
49 {
50 	struct rb_root *root = &map->root;
51 	struct rb_node *node;
52 
53 	if (!locked)
54 		spin_lock_bh(&map->lock);
55 
56 	node = root->rb_node;
57 	while (node) {
58 		struct bpf_cgroup_storage *storage;
59 
60 		storage = container_of(node, struct bpf_cgroup_storage, node);
61 
62 		switch (bpf_cgroup_storage_key_cmp(key, &storage->key)) {
63 		case -1:
64 			node = node->rb_left;
65 			break;
66 		case 1:
67 			node = node->rb_right;
68 			break;
69 		default:
70 			if (!locked)
71 				spin_unlock_bh(&map->lock);
72 			return storage;
73 		}
74 	}
75 
76 	if (!locked)
77 		spin_unlock_bh(&map->lock);
78 
79 	return NULL;
80 }
81 
82 static int cgroup_storage_insert(struct bpf_cgroup_storage_map *map,
83 				 struct bpf_cgroup_storage *storage)
84 {
85 	struct rb_root *root = &map->root;
86 	struct rb_node **new = &(root->rb_node), *parent = NULL;
87 
88 	while (*new) {
89 		struct bpf_cgroup_storage *this;
90 
91 		this = container_of(*new, struct bpf_cgroup_storage, node);
92 
93 		parent = *new;
94 		switch (bpf_cgroup_storage_key_cmp(&storage->key, &this->key)) {
95 		case -1:
96 			new = &((*new)->rb_left);
97 			break;
98 		case 1:
99 			new = &((*new)->rb_right);
100 			break;
101 		default:
102 			return -EEXIST;
103 		}
104 	}
105 
106 	rb_link_node(&storage->node, parent, new);
107 	rb_insert_color(&storage->node, root);
108 
109 	return 0;
110 }
111 
112 static void *cgroup_storage_lookup_elem(struct bpf_map *_map, void *_key)
113 {
114 	struct bpf_cgroup_storage_map *map = map_to_storage(_map);
115 	struct bpf_cgroup_storage_key *key = _key;
116 	struct bpf_cgroup_storage *storage;
117 
118 	storage = cgroup_storage_lookup(map, key, false);
119 	if (!storage)
120 		return NULL;
121 
122 	return &READ_ONCE(storage->buf)->data[0];
123 }
124 
125 static int cgroup_storage_update_elem(struct bpf_map *map, void *_key,
126 				      void *value, u64 flags)
127 {
128 	struct bpf_cgroup_storage_key *key = _key;
129 	struct bpf_cgroup_storage *storage;
130 	struct bpf_storage_buffer *new;
131 
132 	if (flags != BPF_ANY && flags != BPF_EXIST)
133 		return -EINVAL;
134 
135 	storage = cgroup_storage_lookup((struct bpf_cgroup_storage_map *)map,
136 					key, false);
137 	if (!storage)
138 		return -ENOENT;
139 
140 	new = kmalloc_node(sizeof(struct bpf_storage_buffer) +
141 			   map->value_size, __GFP_ZERO | GFP_USER,
142 			   map->numa_node);
143 	if (!new)
144 		return -ENOMEM;
145 
146 	memcpy(&new->data[0], value, map->value_size);
147 
148 	new = xchg(&storage->buf, new);
149 	kfree_rcu(new, rcu);
150 
151 	return 0;
152 }
153 
154 static int cgroup_storage_get_next_key(struct bpf_map *_map, void *_key,
155 				       void *_next_key)
156 {
157 	struct bpf_cgroup_storage_map *map = map_to_storage(_map);
158 	struct bpf_cgroup_storage_key *key = _key;
159 	struct bpf_cgroup_storage_key *next = _next_key;
160 	struct bpf_cgroup_storage *storage;
161 
162 	spin_lock_bh(&map->lock);
163 
164 	if (list_empty(&map->list))
165 		goto enoent;
166 
167 	if (key) {
168 		storage = cgroup_storage_lookup(map, key, true);
169 		if (!storage)
170 			goto enoent;
171 
172 		storage = list_next_entry(storage, list);
173 		if (!storage)
174 			goto enoent;
175 	} else {
176 		storage = list_first_entry(&map->list,
177 					 struct bpf_cgroup_storage, list);
178 	}
179 
180 	spin_unlock_bh(&map->lock);
181 	next->attach_type = storage->key.attach_type;
182 	next->cgroup_inode_id = storage->key.cgroup_inode_id;
183 	return 0;
184 
185 enoent:
186 	spin_unlock_bh(&map->lock);
187 	return -ENOENT;
188 }
189 
190 static struct bpf_map *cgroup_storage_map_alloc(union bpf_attr *attr)
191 {
192 	int numa_node = bpf_map_attr_numa_node(attr);
193 	struct bpf_cgroup_storage_map *map;
194 
195 	if (attr->key_size != sizeof(struct bpf_cgroup_storage_key))
196 		return ERR_PTR(-EINVAL);
197 
198 	if (attr->value_size == 0)
199 		return ERR_PTR(-EINVAL);
200 
201 	if (attr->value_size > PAGE_SIZE)
202 		return ERR_PTR(-E2BIG);
203 
204 	if (attr->map_flags & ~LOCAL_STORAGE_CREATE_FLAG_MASK)
205 		/* reserved bits should not be used */
206 		return ERR_PTR(-EINVAL);
207 
208 	if (attr->max_entries)
209 		/* max_entries is not used and enforced to be 0 */
210 		return ERR_PTR(-EINVAL);
211 
212 	map = kmalloc_node(sizeof(struct bpf_cgroup_storage_map),
213 			   __GFP_ZERO | GFP_USER, numa_node);
214 	if (!map)
215 		return ERR_PTR(-ENOMEM);
216 
217 	map->map.pages = round_up(sizeof(struct bpf_cgroup_storage_map),
218 				  PAGE_SIZE) >> PAGE_SHIFT;
219 
220 	/* copy mandatory map attributes */
221 	bpf_map_init_from_attr(&map->map, attr);
222 
223 	spin_lock_init(&map->lock);
224 	map->root = RB_ROOT;
225 	INIT_LIST_HEAD(&map->list);
226 
227 	return &map->map;
228 }
229 
230 static void cgroup_storage_map_free(struct bpf_map *_map)
231 {
232 	struct bpf_cgroup_storage_map *map = map_to_storage(_map);
233 
234 	WARN_ON(!RB_EMPTY_ROOT(&map->root));
235 	WARN_ON(!list_empty(&map->list));
236 
237 	kfree(map);
238 }
239 
240 static int cgroup_storage_delete_elem(struct bpf_map *map, void *key)
241 {
242 	return -EINVAL;
243 }
244 
245 const struct bpf_map_ops cgroup_storage_map_ops = {
246 	.map_alloc = cgroup_storage_map_alloc,
247 	.map_free = cgroup_storage_map_free,
248 	.map_get_next_key = cgroup_storage_get_next_key,
249 	.map_lookup_elem = cgroup_storage_lookup_elem,
250 	.map_update_elem = cgroup_storage_update_elem,
251 	.map_delete_elem = cgroup_storage_delete_elem,
252 	.map_check_btf = map_check_no_btf,
253 };
254 
255 int bpf_cgroup_storage_assign(struct bpf_prog *prog, struct bpf_map *_map)
256 {
257 	struct bpf_cgroup_storage_map *map = map_to_storage(_map);
258 	int ret = -EBUSY;
259 
260 	spin_lock_bh(&map->lock);
261 
262 	if (map->prog && map->prog != prog)
263 		goto unlock;
264 	if (prog->aux->cgroup_storage && prog->aux->cgroup_storage != _map)
265 		goto unlock;
266 
267 	map->prog = prog;
268 	prog->aux->cgroup_storage = _map;
269 	ret = 0;
270 unlock:
271 	spin_unlock_bh(&map->lock);
272 
273 	return ret;
274 }
275 
276 void bpf_cgroup_storage_release(struct bpf_prog *prog, struct bpf_map *_map)
277 {
278 	struct bpf_cgroup_storage_map *map = map_to_storage(_map);
279 
280 	spin_lock_bh(&map->lock);
281 	if (map->prog == prog) {
282 		WARN_ON(prog->aux->cgroup_storage != _map);
283 		map->prog = NULL;
284 		prog->aux->cgroup_storage = NULL;
285 	}
286 	spin_unlock_bh(&map->lock);
287 }
288 
289 struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(struct bpf_prog *prog)
290 {
291 	struct bpf_cgroup_storage *storage;
292 	struct bpf_map *map;
293 	u32 pages;
294 
295 	map = prog->aux->cgroup_storage;
296 	if (!map)
297 		return NULL;
298 
299 	pages = round_up(sizeof(struct bpf_cgroup_storage) +
300 			 sizeof(struct bpf_storage_buffer) +
301 			 map->value_size, PAGE_SIZE) >> PAGE_SHIFT;
302 	if (bpf_map_charge_memlock(map, pages))
303 		return ERR_PTR(-EPERM);
304 
305 	storage = kmalloc_node(sizeof(struct bpf_cgroup_storage),
306 			       __GFP_ZERO | GFP_USER, map->numa_node);
307 	if (!storage) {
308 		bpf_map_uncharge_memlock(map, pages);
309 		return ERR_PTR(-ENOMEM);
310 	}
311 
312 	storage->buf = kmalloc_node(sizeof(struct bpf_storage_buffer) +
313 				    map->value_size, __GFP_ZERO | GFP_USER,
314 				    map->numa_node);
315 	if (!storage->buf) {
316 		bpf_map_uncharge_memlock(map, pages);
317 		kfree(storage);
318 		return ERR_PTR(-ENOMEM);
319 	}
320 
321 	storage->map = (struct bpf_cgroup_storage_map *)map;
322 
323 	return storage;
324 }
325 
326 void bpf_cgroup_storage_free(struct bpf_cgroup_storage *storage)
327 {
328 	u32 pages;
329 	struct bpf_map *map;
330 
331 	if (!storage)
332 		return;
333 
334 	map = &storage->map->map;
335 	pages = round_up(sizeof(struct bpf_cgroup_storage) +
336 			 sizeof(struct bpf_storage_buffer) +
337 			 map->value_size, PAGE_SIZE) >> PAGE_SHIFT;
338 	bpf_map_uncharge_memlock(map, pages);
339 
340 	kfree_rcu(storage->buf, rcu);
341 	kfree_rcu(storage, rcu);
342 }
343 
344 void bpf_cgroup_storage_link(struct bpf_cgroup_storage *storage,
345 			     struct cgroup *cgroup,
346 			     enum bpf_attach_type type)
347 {
348 	struct bpf_cgroup_storage_map *map;
349 
350 	if (!storage)
351 		return;
352 
353 	storage->key.attach_type = type;
354 	storage->key.cgroup_inode_id = cgroup->kn->id.id;
355 
356 	map = storage->map;
357 
358 	spin_lock_bh(&map->lock);
359 	WARN_ON(cgroup_storage_insert(map, storage));
360 	list_add(&storage->list, &map->list);
361 	spin_unlock_bh(&map->lock);
362 }
363 
364 void bpf_cgroup_storage_unlink(struct bpf_cgroup_storage *storage)
365 {
366 	struct bpf_cgroup_storage_map *map;
367 	struct rb_root *root;
368 
369 	if (!storage)
370 		return;
371 
372 	map = storage->map;
373 
374 	spin_lock_bh(&map->lock);
375 	root = &map->root;
376 	rb_erase(&storage->node, root);
377 
378 	list_del(&storage->list);
379 	spin_unlock_bh(&map->lock);
380 }
381 
382 #endif
383