~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/bpf/local_storage.c

Version: ~ [ linux-5.17-rc1 ] ~ [ linux-5.16.2 ] ~ [ linux-5.15.16 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.93 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.173 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.225 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.262 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.297 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.299 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 //SPDX-License-Identifier: GPL-2.0
  2 #include <linux/bpf-cgroup.h>
  3 #include <linux/bpf.h>
  4 #include <linux/btf.h>
  5 #include <linux/bug.h>
  6 #include <linux/filter.h>
  7 #include <linux/mm.h>
  8 #include <linux/rbtree.h>
  9 #include <linux/slab.h>
 10 #include <uapi/linux/btf.h>
 11 
 12 DEFINE_PER_CPU(struct bpf_cgroup_storage*, bpf_cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]);
 13 
 14 #ifdef CONFIG_CGROUP_BPF
 15 
 16 #define LOCAL_STORAGE_CREATE_FLAG_MASK                                  \
 17         (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK)
 18 
 19 struct bpf_cgroup_storage_map {
 20         struct bpf_map map;
 21 
 22         spinlock_t lock;
 23         struct bpf_prog *prog;
 24         struct rb_root root;
 25         struct list_head list;
 26 };
 27 
 28 static struct bpf_cgroup_storage_map *map_to_storage(struct bpf_map *map)
 29 {
 30         return container_of(map, struct bpf_cgroup_storage_map, map);
 31 }
 32 
 33 static int bpf_cgroup_storage_key_cmp(
 34         const struct bpf_cgroup_storage_key *key1,
 35         const struct bpf_cgroup_storage_key *key2)
 36 {
 37         if (key1->cgroup_inode_id < key2->cgroup_inode_id)
 38                 return -1;
 39         else if (key1->cgroup_inode_id > key2->cgroup_inode_id)
 40                 return 1;
 41         else if (key1->attach_type < key2->attach_type)
 42                 return -1;
 43         else if (key1->attach_type > key2->attach_type)
 44                 return 1;
 45         return 0;
 46 }
 47 
 48 static struct bpf_cgroup_storage *cgroup_storage_lookup(
 49         struct bpf_cgroup_storage_map *map, struct bpf_cgroup_storage_key *key,
 50         bool locked)
 51 {
 52         struct rb_root *root = &map->root;
 53         struct rb_node *node;
 54 
 55         if (!locked)
 56                 spin_lock_bh(&map->lock);
 57 
 58         node = root->rb_node;
 59         while (node) {
 60                 struct bpf_cgroup_storage *storage;
 61 
 62                 storage = container_of(node, struct bpf_cgroup_storage, node);
 63 
 64                 switch (bpf_cgroup_storage_key_cmp(key, &storage->key)) {
 65                 case -1:
 66                         node = node->rb_left;
 67                         break;
 68                 case 1:
 69                         node = node->rb_right;
 70                         break;
 71                 default:
 72                         if (!locked)
 73                                 spin_unlock_bh(&map->lock);
 74                         return storage;
 75                 }
 76         }
 77 
 78         if (!locked)
 79                 spin_unlock_bh(&map->lock);
 80 
 81         return NULL;
 82 }
 83 
 84 static int cgroup_storage_insert(struct bpf_cgroup_storage_map *map,
 85                                  struct bpf_cgroup_storage *storage)
 86 {
 87         struct rb_root *root = &map->root;
 88         struct rb_node **new = &(root->rb_node), *parent = NULL;
 89 
 90         while (*new) {
 91                 struct bpf_cgroup_storage *this;
 92 
 93                 this = container_of(*new, struct bpf_cgroup_storage, node);
 94 
 95                 parent = *new;
 96                 switch (bpf_cgroup_storage_key_cmp(&storage->key, &this->key)) {
 97                 case -1:
 98                         new = &((*new)->rb_left);
 99                         break;
100                 case 1:
101                         new = &((*new)->rb_right);
102                         break;
103                 default:
104                         return -EEXIST;
105                 }
106         }
107 
108         rb_link_node(&storage->node, parent, new);
109         rb_insert_color(&storage->node, root);
110 
111         return 0;
112 }
113 
114 static void *cgroup_storage_lookup_elem(struct bpf_map *_map, void *_key)
115 {
116         struct bpf_cgroup_storage_map *map = map_to_storage(_map);
117         struct bpf_cgroup_storage_key *key = _key;
118         struct bpf_cgroup_storage *storage;
119 
120         storage = cgroup_storage_lookup(map, key, false);
121         if (!storage)
122                 return NULL;
123 
124         return &READ_ONCE(storage->buf)->data[0];
125 }
126 
127 static int cgroup_storage_update_elem(struct bpf_map *map, void *_key,
128                                       void *value, u64 flags)
129 {
130         struct bpf_cgroup_storage_key *key = _key;
131         struct bpf_cgroup_storage *storage;
132         struct bpf_storage_buffer *new;
133 
134         if (unlikely(flags & ~(BPF_F_LOCK | BPF_EXIST | BPF_NOEXIST)))
135                 return -EINVAL;
136 
137         if (unlikely(flags & BPF_NOEXIST))
138                 return -EINVAL;
139 
140         if (unlikely((flags & BPF_F_LOCK) &&
141                      !map_value_has_spin_lock(map)))
142                 return -EINVAL;
143 
144         storage = cgroup_storage_lookup((struct bpf_cgroup_storage_map *)map,
145                                         key, false);
146         if (!storage)
147                 return -ENOENT;
148 
149         if (flags & BPF_F_LOCK) {
150                 copy_map_value_locked(map, storage->buf->data, value, false);
151                 return 0;
152         }
153 
154         new = kmalloc_node(sizeof(struct bpf_storage_buffer) +
155                            map->value_size,
156                            __GFP_ZERO | GFP_ATOMIC | __GFP_NOWARN,
157                            map->numa_node);
158         if (!new)
159                 return -ENOMEM;
160 
161         memcpy(&new->data[0], value, map->value_size);
162         check_and_init_map_lock(map, new->data);
163 
164         new = xchg(&storage->buf, new);
165         kfree_rcu(new, rcu);
166 
167         return 0;
168 }
169 
170 int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *_key,
171                                    void *value)
172 {
173         struct bpf_cgroup_storage_map *map = map_to_storage(_map);
174         struct bpf_cgroup_storage_key *key = _key;
175         struct bpf_cgroup_storage *storage;
176         int cpu, off = 0;
177         u32 size;
178 
179         rcu_read_lock();
180         storage = cgroup_storage_lookup(map, key, false);
181         if (!storage) {
182                 rcu_read_unlock();
183                 return -ENOENT;
184         }
185 
186         /* per_cpu areas are zero-filled and bpf programs can only
187          * access 'value_size' of them, so copying rounded areas
188          * will not leak any kernel data
189          */
190         size = round_up(_map->value_size, 8);
191         for_each_possible_cpu(cpu) {
192                 bpf_long_memcpy(value + off,
193                                 per_cpu_ptr(storage->percpu_buf, cpu), size);
194                 off += size;
195         }
196         rcu_read_unlock();
197         return 0;
198 }
199 
200 int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *_key,
201                                      void *value, u64 map_flags)
202 {
203         struct bpf_cgroup_storage_map *map = map_to_storage(_map);
204         struct bpf_cgroup_storage_key *key = _key;
205         struct bpf_cgroup_storage *storage;
206         int cpu, off = 0;
207         u32 size;
208 
209         if (map_flags != BPF_ANY && map_flags != BPF_EXIST)
210                 return -EINVAL;
211 
212         rcu_read_lock();
213         storage = cgroup_storage_lookup(map, key, false);
214         if (!storage) {
215                 rcu_read_unlock();
216                 return -ENOENT;
217         }
218 
219         /* the user space will provide round_up(value_size, 8) bytes that
220          * will be copied into per-cpu area. bpf programs can only access
221          * value_size of it. During lookup the same extra bytes will be
222          * returned or zeros which were zero-filled by percpu_alloc,
223          * so no kernel data leaks possible
224          */
225         size = round_up(_map->value_size, 8);
226         for_each_possible_cpu(cpu) {
227                 bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu),
228                                 value + off, size);
229                 off += size;
230         }
231         rcu_read_unlock();
232         return 0;
233 }
234 
235 static int cgroup_storage_get_next_key(struct bpf_map *_map, void *_key,
236                                        void *_next_key)
237 {
238         struct bpf_cgroup_storage_map *map = map_to_storage(_map);
239         struct bpf_cgroup_storage_key *key = _key;
240         struct bpf_cgroup_storage_key *next = _next_key;
241         struct bpf_cgroup_storage *storage;
242 
243         spin_lock_bh(&map->lock);
244 
245         if (list_empty(&map->list))
246                 goto enoent;
247 
248         if (key) {
249                 storage = cgroup_storage_lookup(map, key, true);
250                 if (!storage)
251                         goto enoent;
252 
253                 storage = list_next_entry(storage, list);
254                 if (!storage)
255                         goto enoent;
256         } else {
257                 storage = list_first_entry(&map->list,
258                                          struct bpf_cgroup_storage, list);
259         }
260 
261         spin_unlock_bh(&map->lock);
262         next->attach_type = storage->key.attach_type;
263         next->cgroup_inode_id = storage->key.cgroup_inode_id;
264         return 0;
265 
266 enoent:
267         spin_unlock_bh(&map->lock);
268         return -ENOENT;
269 }
270 
271 static struct bpf_map *cgroup_storage_map_alloc(union bpf_attr *attr)
272 {
273         int numa_node = bpf_map_attr_numa_node(attr);
274         struct bpf_cgroup_storage_map *map;
275 
276         if (attr->key_size != sizeof(struct bpf_cgroup_storage_key))
277                 return ERR_PTR(-EINVAL);
278 
279         if (attr->value_size == 0)
280                 return ERR_PTR(-EINVAL);
281 
282         if (attr->value_size > PAGE_SIZE)
283                 return ERR_PTR(-E2BIG);
284 
285         if (attr->map_flags & ~LOCAL_STORAGE_CREATE_FLAG_MASK ||
286             !bpf_map_flags_access_ok(attr->map_flags))
287                 return ERR_PTR(-EINVAL);
288 
289         if (attr->max_entries)
290                 /* max_entries is not used and enforced to be 0 */
291                 return ERR_PTR(-EINVAL);
292 
293         map = kmalloc_node(sizeof(struct bpf_cgroup_storage_map),
294                            __GFP_ZERO | GFP_USER, numa_node);
295         if (!map)
296                 return ERR_PTR(-ENOMEM);
297 
298         map->map.pages = round_up(sizeof(struct bpf_cgroup_storage_map),
299                                   PAGE_SIZE) >> PAGE_SHIFT;
300 
301         /* copy mandatory map attributes */
302         bpf_map_init_from_attr(&map->map, attr);
303 
304         spin_lock_init(&map->lock);
305         map->root = RB_ROOT;
306         INIT_LIST_HEAD(&map->list);
307 
308         return &map->map;
309 }
310 
311 static void cgroup_storage_map_free(struct bpf_map *_map)
312 {
313         struct bpf_cgroup_storage_map *map = map_to_storage(_map);
314 
315         WARN_ON(!RB_EMPTY_ROOT(&map->root));
316         WARN_ON(!list_empty(&map->list));
317 
318         kfree(map);
319 }
320 
321 static int cgroup_storage_delete_elem(struct bpf_map *map, void *key)
322 {
323         return -EINVAL;
324 }
325 
326 static int cgroup_storage_check_btf(const struct bpf_map *map,
327                                     const struct btf *btf,
328                                     const struct btf_type *key_type,
329                                     const struct btf_type *value_type)
330 {
331         struct btf_member *m;
332         u32 offset, size;
333 
334         /* Key is expected to be of struct bpf_cgroup_storage_key type,
335          * which is:
336          * struct bpf_cgroup_storage_key {
337          *      __u64   cgroup_inode_id;
338          *      __u32   attach_type;
339          * };
340          */
341 
342         /*
343          * Key_type must be a structure with two fields.
344          */
345         if (BTF_INFO_KIND(key_type->info) != BTF_KIND_STRUCT ||
346             BTF_INFO_VLEN(key_type->info) != 2)
347                 return -EINVAL;
348 
349         /*
350          * The first field must be a 64 bit integer at 0 offset.
351          */
352         m = (struct btf_member *)(key_type + 1);
353         size = FIELD_SIZEOF(struct bpf_cgroup_storage_key, cgroup_inode_id);
354         if (!btf_member_is_reg_int(btf, key_type, m, 0, size))
355                 return -EINVAL;
356 
357         /*
358          * The second field must be a 32 bit integer at 64 bit offset.
359          */
360         m++;
361         offset = offsetof(struct bpf_cgroup_storage_key, attach_type);
362         size = FIELD_SIZEOF(struct bpf_cgroup_storage_key, attach_type);
363         if (!btf_member_is_reg_int(btf, key_type, m, offset, size))
364                 return -EINVAL;
365 
366         return 0;
367 }
368 
369 static void cgroup_storage_seq_show_elem(struct bpf_map *map, void *_key,
370                                          struct seq_file *m)
371 {
372         enum bpf_cgroup_storage_type stype = cgroup_storage_type(map);
373         struct bpf_cgroup_storage_key *key = _key;
374         struct bpf_cgroup_storage *storage;
375         int cpu;
376 
377         rcu_read_lock();
378         storage = cgroup_storage_lookup(map_to_storage(map), key, false);
379         if (!storage) {
380                 rcu_read_unlock();
381                 return;
382         }
383 
384         btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
385         stype = cgroup_storage_type(map);
386         if (stype == BPF_CGROUP_STORAGE_SHARED) {
387                 seq_puts(m, ": ");
388                 btf_type_seq_show(map->btf, map->btf_value_type_id,
389                                   &READ_ONCE(storage->buf)->data[0], m);
390                 seq_puts(m, "\n");
391         } else {
392                 seq_puts(m, ": {\n");
393                 for_each_possible_cpu(cpu) {
394                         seq_printf(m, "\tcpu%d: ", cpu);
395                         btf_type_seq_show(map->btf, map->btf_value_type_id,
396                                           per_cpu_ptr(storage->percpu_buf, cpu),
397                                           m);
398                         seq_puts(m, "\n");
399                 }
400                 seq_puts(m, "}\n");
401         }
402         rcu_read_unlock();
403 }
404 
405 const struct bpf_map_ops cgroup_storage_map_ops = {
406         .map_alloc = cgroup_storage_map_alloc,
407         .map_free = cgroup_storage_map_free,
408         .map_get_next_key = cgroup_storage_get_next_key,
409         .map_lookup_elem = cgroup_storage_lookup_elem,
410         .map_update_elem = cgroup_storage_update_elem,
411         .map_delete_elem = cgroup_storage_delete_elem,
412         .map_check_btf = cgroup_storage_check_btf,
413         .map_seq_show_elem = cgroup_storage_seq_show_elem,
414 };
415 
416 int bpf_cgroup_storage_assign(struct bpf_prog *prog, struct bpf_map *_map)
417 {
418         enum bpf_cgroup_storage_type stype = cgroup_storage_type(_map);
419         struct bpf_cgroup_storage_map *map = map_to_storage(_map);
420         int ret = -EBUSY;
421 
422         spin_lock_bh(&map->lock);
423 
424         if (map->prog && map->prog != prog)
425                 goto unlock;
426         if (prog->aux->cgroup_storage[stype] &&
427             prog->aux->cgroup_storage[stype] != _map)
428                 goto unlock;
429 
430         map->prog = prog;
431         prog->aux->cgroup_storage[stype] = _map;
432         ret = 0;
433 unlock:
434         spin_unlock_bh(&map->lock);
435 
436         return ret;
437 }
438 
439 void bpf_cgroup_storage_release(struct bpf_prog *prog, struct bpf_map *_map)
440 {
441         enum bpf_cgroup_storage_type stype = cgroup_storage_type(_map);
442         struct bpf_cgroup_storage_map *map = map_to_storage(_map);
443 
444         spin_lock_bh(&map->lock);
445         if (map->prog == prog) {
446                 WARN_ON(prog->aux->cgroup_storage[stype] != _map);
447                 map->prog = NULL;
448                 prog->aux->cgroup_storage[stype] = NULL;
449         }
450         spin_unlock_bh(&map->lock);
451 }
452 
453 static size_t bpf_cgroup_storage_calculate_size(struct bpf_map *map, u32 *pages)
454 {
455         size_t size;
456 
457         if (cgroup_storage_type(map) == BPF_CGROUP_STORAGE_SHARED) {
458                 size = sizeof(struct bpf_storage_buffer) + map->value_size;
459                 *pages = round_up(sizeof(struct bpf_cgroup_storage) + size,
460                                   PAGE_SIZE) >> PAGE_SHIFT;
461         } else {
462                 size = map->value_size;
463                 *pages = round_up(round_up(size, 8) * num_possible_cpus(),
464                                   PAGE_SIZE) >> PAGE_SHIFT;
465         }
466 
467         return size;
468 }
469 
470 struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(struct bpf_prog *prog,
471                                         enum bpf_cgroup_storage_type stype)
472 {
473         struct bpf_cgroup_storage *storage;
474         struct bpf_map *map;
475         gfp_t flags;
476         size_t size;
477         u32 pages;
478 
479         map = prog->aux->cgroup_storage[stype];
480         if (!map)
481                 return NULL;
482 
483         size = bpf_cgroup_storage_calculate_size(map, &pages);
484 
485         if (bpf_map_charge_memlock(map, pages))
486                 return ERR_PTR(-EPERM);
487 
488         storage = kmalloc_node(sizeof(struct bpf_cgroup_storage),
489                                __GFP_ZERO | GFP_USER, map->numa_node);
490         if (!storage)
491                 goto enomem;
492 
493         flags = __GFP_ZERO | GFP_USER;
494 
495         if (stype == BPF_CGROUP_STORAGE_SHARED) {
496                 storage->buf = kmalloc_node(size, flags, map->numa_node);
497                 if (!storage->buf)
498                         goto enomem;
499                 check_and_init_map_lock(map, storage->buf->data);
500         } else {
501                 storage->percpu_buf = __alloc_percpu_gfp(size, 8, flags);
502                 if (!storage->percpu_buf)
503                         goto enomem;
504         }
505 
506         storage->map = (struct bpf_cgroup_storage_map *)map;
507 
508         return storage;
509 
510 enomem:
511         bpf_map_uncharge_memlock(map, pages);
512         kfree(storage);
513         return ERR_PTR(-ENOMEM);
514 }
515 
516 static void free_shared_cgroup_storage_rcu(struct rcu_head *rcu)
517 {
518         struct bpf_cgroup_storage *storage =
519                 container_of(rcu, struct bpf_cgroup_storage, rcu);
520 
521         kfree(storage->buf);
522         kfree(storage);
523 }
524 
525 static void free_percpu_cgroup_storage_rcu(struct rcu_head *rcu)
526 {
527         struct bpf_cgroup_storage *storage =
528                 container_of(rcu, struct bpf_cgroup_storage, rcu);
529 
530         free_percpu(storage->percpu_buf);
531         kfree(storage);
532 }
533 
534 void bpf_cgroup_storage_free(struct bpf_cgroup_storage *storage)
535 {
536         enum bpf_cgroup_storage_type stype;
537         struct bpf_map *map;
538         u32 pages;
539 
540         if (!storage)
541                 return;
542 
543         map = &storage->map->map;
544 
545         bpf_cgroup_storage_calculate_size(map, &pages);
546         bpf_map_uncharge_memlock(map, pages);
547 
548         stype = cgroup_storage_type(map);
549         if (stype == BPF_CGROUP_STORAGE_SHARED)
550                 call_rcu(&storage->rcu, free_shared_cgroup_storage_rcu);
551         else
552                 call_rcu(&storage->rcu, free_percpu_cgroup_storage_rcu);
553 }
554 
555 void bpf_cgroup_storage_link(struct bpf_cgroup_storage *storage,
556                              struct cgroup *cgroup,
557                              enum bpf_attach_type type)
558 {
559         struct bpf_cgroup_storage_map *map;
560 
561         if (!storage)
562                 return;
563 
564         storage->key.attach_type = type;
565         storage->key.cgroup_inode_id = cgroup->kn->id.id;
566 
567         map = storage->map;
568 
569         spin_lock_bh(&map->lock);
570         WARN_ON(cgroup_storage_insert(map, storage));
571         list_add(&storage->list, &map->list);
572         spin_unlock_bh(&map->lock);
573 }
574 
575 void bpf_cgroup_storage_unlink(struct bpf_cgroup_storage *storage)
576 {
577         struct bpf_cgroup_storage_map *map;
578         struct rb_root *root;
579 
580         if (!storage)
581                 return;
582 
583         map = storage->map;
584 
585         spin_lock_bh(&map->lock);
586         root = &map->root;
587         rb_erase(&storage->node, root);
588 
589         list_del(&storage->list);
590         spin_unlock_bh(&map->lock);
591 }
592 
593 #endif
594 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp