]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - kernel/bpf/local_storage.c
bpf: rework cgroup storage pointer passing
[mirror_ubuntu-eoan-kernel.git] / kernel / bpf / local_storage.c
CommitLineData
de9cbbaa
RG
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
f294b37e
RG
10DEFINE_PER_CPU(struct bpf_cgroup_storage*,
11 bpf_cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]);
aa0ad5b0 12
de9cbbaa
RG
13#ifdef CONFIG_CGROUP_BPF
14
15#define LOCAL_STORAGE_CREATE_FLAG_MASK \
16 (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
17
18struct bpf_cgroup_storage_map {
19 struct bpf_map map;
20
21 spinlock_t lock;
22 struct bpf_prog *prog;
23 struct rb_root root;
24 struct list_head list;
25};
26
27static struct bpf_cgroup_storage_map *map_to_storage(struct bpf_map *map)
28{
29 return container_of(map, struct bpf_cgroup_storage_map, map);
30}
31
32static int bpf_cgroup_storage_key_cmp(
33 const struct bpf_cgroup_storage_key *key1,
34 const struct bpf_cgroup_storage_key *key2)
35{
36 if (key1->cgroup_inode_id < key2->cgroup_inode_id)
37 return -1;
38 else if (key1->cgroup_inode_id > key2->cgroup_inode_id)
39 return 1;
40 else if (key1->attach_type < key2->attach_type)
41 return -1;
42 else if (key1->attach_type > key2->attach_type)
43 return 1;
44 return 0;
45}
46
47static struct bpf_cgroup_storage *cgroup_storage_lookup(
48 struct bpf_cgroup_storage_map *map, struct bpf_cgroup_storage_key *key,
49 bool locked)
50{
51 struct rb_root *root = &map->root;
52 struct rb_node *node;
53
54 if (!locked)
55 spin_lock_bh(&map->lock);
56
57 node = root->rb_node;
58 while (node) {
59 struct bpf_cgroup_storage *storage;
60
61 storage = container_of(node, struct bpf_cgroup_storage, node);
62
63 switch (bpf_cgroup_storage_key_cmp(key, &storage->key)) {
64 case -1:
65 node = node->rb_left;
66 break;
67 case 1:
68 node = node->rb_right;
69 break;
70 default:
71 if (!locked)
72 spin_unlock_bh(&map->lock);
73 return storage;
74 }
75 }
76
77 if (!locked)
78 spin_unlock_bh(&map->lock);
79
80 return NULL;
81}
82
83static int cgroup_storage_insert(struct bpf_cgroup_storage_map *map,
84 struct bpf_cgroup_storage *storage)
85{
86 struct rb_root *root = &map->root;
87 struct rb_node **new = &(root->rb_node), *parent = NULL;
88
89 while (*new) {
90 struct bpf_cgroup_storage *this;
91
92 this = container_of(*new, struct bpf_cgroup_storage, node);
93
94 parent = *new;
95 switch (bpf_cgroup_storage_key_cmp(&storage->key, &this->key)) {
96 case -1:
97 new = &((*new)->rb_left);
98 break;
99 case 1:
100 new = &((*new)->rb_right);
101 break;
102 default:
103 return -EEXIST;
104 }
105 }
106
107 rb_link_node(&storage->node, parent, new);
108 rb_insert_color(&storage->node, root);
109
110 return 0;
111}
112
113static void *cgroup_storage_lookup_elem(struct bpf_map *_map, void *_key)
114{
115 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
116 struct bpf_cgroup_storage_key *key = _key;
117 struct bpf_cgroup_storage *storage;
118
119 storage = cgroup_storage_lookup(map, key, false);
120 if (!storage)
121 return NULL;
122
123 return &READ_ONCE(storage->buf)->data[0];
124}
125
126static int cgroup_storage_update_elem(struct bpf_map *map, void *_key,
127 void *value, u64 flags)
128{
129 struct bpf_cgroup_storage_key *key = _key;
130 struct bpf_cgroup_storage *storage;
131 struct bpf_storage_buffer *new;
132
133 if (flags & BPF_NOEXIST)
134 return -EINVAL;
135
136 storage = cgroup_storage_lookup((struct bpf_cgroup_storage_map *)map,
137 key, false);
138 if (!storage)
139 return -ENOENT;
140
141 new = kmalloc_node(sizeof(struct bpf_storage_buffer) +
142 map->value_size, __GFP_ZERO | GFP_USER,
143 map->numa_node);
144 if (!new)
145 return -ENOMEM;
146
147 memcpy(&new->data[0], value, map->value_size);
148
149 new = xchg(&storage->buf, new);
150 kfree_rcu(new, rcu);
151
152 return 0;
153}
154
155static int cgroup_storage_get_next_key(struct bpf_map *_map, void *_key,
156 void *_next_key)
157{
158 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
159 struct bpf_cgroup_storage_key *key = _key;
160 struct bpf_cgroup_storage_key *next = _next_key;
161 struct bpf_cgroup_storage *storage;
162
163 spin_lock_bh(&map->lock);
164
165 if (list_empty(&map->list))
166 goto enoent;
167
168 if (key) {
169 storage = cgroup_storage_lookup(map, key, true);
170 if (!storage)
171 goto enoent;
172
173 storage = list_next_entry(storage, list);
174 if (!storage)
175 goto enoent;
176 } else {
177 storage = list_first_entry(&map->list,
178 struct bpf_cgroup_storage, list);
179 }
180
181 spin_unlock_bh(&map->lock);
182 next->attach_type = storage->key.attach_type;
183 next->cgroup_inode_id = storage->key.cgroup_inode_id;
184 return 0;
185
186enoent:
187 spin_unlock_bh(&map->lock);
188 return -ENOENT;
189}
190
191static struct bpf_map *cgroup_storage_map_alloc(union bpf_attr *attr)
192{
193 int numa_node = bpf_map_attr_numa_node(attr);
194 struct bpf_cgroup_storage_map *map;
195
196 if (attr->key_size != sizeof(struct bpf_cgroup_storage_key))
197 return ERR_PTR(-EINVAL);
198
199 if (attr->value_size > PAGE_SIZE)
200 return ERR_PTR(-E2BIG);
201
202 if (attr->map_flags & ~LOCAL_STORAGE_CREATE_FLAG_MASK)
203 /* reserved bits should not be used */
204 return ERR_PTR(-EINVAL);
205
206 if (attr->max_entries)
207 /* max_entries is not used and enforced to be 0 */
208 return ERR_PTR(-EINVAL);
209
210 map = kmalloc_node(sizeof(struct bpf_cgroup_storage_map),
211 __GFP_ZERO | GFP_USER, numa_node);
212 if (!map)
213 return ERR_PTR(-ENOMEM);
214
215 map->map.pages = round_up(sizeof(struct bpf_cgroup_storage_map),
216 PAGE_SIZE) >> PAGE_SHIFT;
217
218 /* copy mandatory map attributes */
219 bpf_map_init_from_attr(&map->map, attr);
220
221 spin_lock_init(&map->lock);
222 map->root = RB_ROOT;
223 INIT_LIST_HEAD(&map->list);
224
225 return &map->map;
226}
227
228static void cgroup_storage_map_free(struct bpf_map *_map)
229{
230 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
231
232 WARN_ON(!RB_EMPTY_ROOT(&map->root));
233 WARN_ON(!list_empty(&map->list));
234
235 kfree(map);
236}
237
238static int cgroup_storage_delete_elem(struct bpf_map *map, void *key)
239{
240 return -EINVAL;
241}
242
243const struct bpf_map_ops cgroup_storage_map_ops = {
244 .map_alloc = cgroup_storage_map_alloc,
245 .map_free = cgroup_storage_map_free,
246 .map_get_next_key = cgroup_storage_get_next_key,
247 .map_lookup_elem = cgroup_storage_lookup_elem,
248 .map_update_elem = cgroup_storage_update_elem,
249 .map_delete_elem = cgroup_storage_delete_elem,
e8d2bec0 250 .map_check_btf = map_check_no_btf,
de9cbbaa
RG
251};
252
253int bpf_cgroup_storage_assign(struct bpf_prog *prog, struct bpf_map *_map)
254{
8bad74f9 255 enum bpf_cgroup_storage_type stype = cgroup_storage_type(_map);
de9cbbaa
RG
256 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
257 int ret = -EBUSY;
258
259 spin_lock_bh(&map->lock);
260
261 if (map->prog && map->prog != prog)
262 goto unlock;
8bad74f9
RG
263 if (prog->aux->cgroup_storage[stype] &&
264 prog->aux->cgroup_storage[stype] != _map)
de9cbbaa
RG
265 goto unlock;
266
267 map->prog = prog;
8bad74f9 268 prog->aux->cgroup_storage[stype] = _map;
de9cbbaa
RG
269 ret = 0;
270unlock:
271 spin_unlock_bh(&map->lock);
272
273 return ret;
274}
275
276void bpf_cgroup_storage_release(struct bpf_prog *prog, struct bpf_map *_map)
277{
8bad74f9 278 enum bpf_cgroup_storage_type stype = cgroup_storage_type(_map);
de9cbbaa
RG
279 struct bpf_cgroup_storage_map *map = map_to_storage(_map);
280
281 spin_lock_bh(&map->lock);
282 if (map->prog == prog) {
8bad74f9 283 WARN_ON(prog->aux->cgroup_storage[stype] != _map);
de9cbbaa 284 map->prog = NULL;
8bad74f9 285 prog->aux->cgroup_storage[stype] = NULL;
de9cbbaa
RG
286 }
287 spin_unlock_bh(&map->lock);
288}
289
8bad74f9
RG
290struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(struct bpf_prog *prog,
291 enum bpf_cgroup_storage_type stype)
de9cbbaa
RG
292{
293 struct bpf_cgroup_storage *storage;
294 struct bpf_map *map;
295 u32 pages;
296
8bad74f9 297 map = prog->aux->cgroup_storage[stype];
de9cbbaa
RG
298 if (!map)
299 return NULL;
300
301 pages = round_up(sizeof(struct bpf_cgroup_storage) +
302 sizeof(struct bpf_storage_buffer) +
303 map->value_size, PAGE_SIZE) >> PAGE_SHIFT;
304 if (bpf_map_charge_memlock(map, pages))
305 return ERR_PTR(-EPERM);
306
307 storage = kmalloc_node(sizeof(struct bpf_cgroup_storage),
308 __GFP_ZERO | GFP_USER, map->numa_node);
309 if (!storage) {
310 bpf_map_uncharge_memlock(map, pages);
311 return ERR_PTR(-ENOMEM);
312 }
313
314 storage->buf = kmalloc_node(sizeof(struct bpf_storage_buffer) +
315 map->value_size, __GFP_ZERO | GFP_USER,
316 map->numa_node);
317 if (!storage->buf) {
318 bpf_map_uncharge_memlock(map, pages);
319 kfree(storage);
320 return ERR_PTR(-ENOMEM);
321 }
322
323 storage->map = (struct bpf_cgroup_storage_map *)map;
324
325 return storage;
326}
327
328void bpf_cgroup_storage_free(struct bpf_cgroup_storage *storage)
329{
330 u32 pages;
331 struct bpf_map *map;
332
333 if (!storage)
334 return;
335
336 map = &storage->map->map;
337 pages = round_up(sizeof(struct bpf_cgroup_storage) +
338 sizeof(struct bpf_storage_buffer) +
339 map->value_size, PAGE_SIZE) >> PAGE_SHIFT;
340 bpf_map_uncharge_memlock(map, pages);
341
342 kfree_rcu(storage->buf, rcu);
343 kfree_rcu(storage, rcu);
344}
345
346void bpf_cgroup_storage_link(struct bpf_cgroup_storage *storage,
347 struct cgroup *cgroup,
348 enum bpf_attach_type type)
349{
350 struct bpf_cgroup_storage_map *map;
351
352 if (!storage)
353 return;
354
355 storage->key.attach_type = type;
356 storage->key.cgroup_inode_id = cgroup->kn->id.id;
357
358 map = storage->map;
359
360 spin_lock_bh(&map->lock);
361 WARN_ON(cgroup_storage_insert(map, storage));
362 list_add(&storage->list, &map->list);
363 spin_unlock_bh(&map->lock);
364}
365
366void bpf_cgroup_storage_unlink(struct bpf_cgroup_storage *storage)
367{
368 struct bpf_cgroup_storage_map *map;
369 struct rb_root *root;
370
371 if (!storage)
372 return;
373
374 map = storage->map;
375
376 spin_lock_bh(&map->lock);
377 root = &map->root;
378 rb_erase(&storage->node, root);
379
380 list_del(&storage->list);
381 spin_unlock_bh(&map->lock);
382}
383
384#endif