Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 #include <linux/bpf-cgroup.h>
0003 #include <linux/bpf.h>
0004 #include <linux/bpf_local_storage.h>
0005 #include <linux/btf.h>
0006 #include <linux/bug.h>
0007 #include <linux/filter.h>
0008 #include <linux/mm.h>
0009 #include <linux/rbtree.h>
0010 #include <linux/slab.h>
0011 #include <uapi/linux/btf.h>
0012 #include <linux/btf_ids.h>
0013 
0014 #ifdef CONFIG_CGROUP_BPF
0015 
0016 #include "../cgroup/cgroup-internal.h"
0017 
0018 #define LOCAL_STORAGE_CREATE_FLAG_MASK                  \
0019     (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK)
0020 
0021 struct bpf_cgroup_storage_map {
0022     struct bpf_map map;
0023 
0024     spinlock_t lock;
0025     struct rb_root root;
0026     struct list_head list;
0027 };
0028 
0029 static struct bpf_cgroup_storage_map *map_to_storage(struct bpf_map *map)
0030 {
0031     return container_of(map, struct bpf_cgroup_storage_map, map);
0032 }
0033 
0034 static bool attach_type_isolated(const struct bpf_map *map)
0035 {
0036     return map->key_size == sizeof(struct bpf_cgroup_storage_key);
0037 }
0038 
0039 static int bpf_cgroup_storage_key_cmp(const struct bpf_cgroup_storage_map *map,
0040                       const void *_key1, const void *_key2)
0041 {
0042     if (attach_type_isolated(&map->map)) {
0043         const struct bpf_cgroup_storage_key *key1 = _key1;
0044         const struct bpf_cgroup_storage_key *key2 = _key2;
0045 
0046         if (key1->cgroup_inode_id < key2->cgroup_inode_id)
0047             return -1;
0048         else if (key1->cgroup_inode_id > key2->cgroup_inode_id)
0049             return 1;
0050         else if (key1->attach_type < key2->attach_type)
0051             return -1;
0052         else if (key1->attach_type > key2->attach_type)
0053             return 1;
0054     } else {
0055         const __u64 *cgroup_inode_id1 = _key1;
0056         const __u64 *cgroup_inode_id2 = _key2;
0057 
0058         if (*cgroup_inode_id1 < *cgroup_inode_id2)
0059             return -1;
0060         else if (*cgroup_inode_id1 > *cgroup_inode_id2)
0061             return 1;
0062     }
0063     return 0;
0064 }
0065 
0066 struct bpf_cgroup_storage *
0067 cgroup_storage_lookup(struct bpf_cgroup_storage_map *map,
0068               void *key, bool locked)
0069 {
0070     struct rb_root *root = &map->root;
0071     struct rb_node *node;
0072 
0073     if (!locked)
0074         spin_lock_bh(&map->lock);
0075 
0076     node = root->rb_node;
0077     while (node) {
0078         struct bpf_cgroup_storage *storage;
0079 
0080         storage = container_of(node, struct bpf_cgroup_storage, node);
0081 
0082         switch (bpf_cgroup_storage_key_cmp(map, key, &storage->key)) {
0083         case -1:
0084             node = node->rb_left;
0085             break;
0086         case 1:
0087             node = node->rb_right;
0088             break;
0089         default:
0090             if (!locked)
0091                 spin_unlock_bh(&map->lock);
0092             return storage;
0093         }
0094     }
0095 
0096     if (!locked)
0097         spin_unlock_bh(&map->lock);
0098 
0099     return NULL;
0100 }
0101 
0102 static int cgroup_storage_insert(struct bpf_cgroup_storage_map *map,
0103                  struct bpf_cgroup_storage *storage)
0104 {
0105     struct rb_root *root = &map->root;
0106     struct rb_node **new = &(root->rb_node), *parent = NULL;
0107 
0108     while (*new) {
0109         struct bpf_cgroup_storage *this;
0110 
0111         this = container_of(*new, struct bpf_cgroup_storage, node);
0112 
0113         parent = *new;
0114         switch (bpf_cgroup_storage_key_cmp(map, &storage->key, &this->key)) {
0115         case -1:
0116             new = &((*new)->rb_left);
0117             break;
0118         case 1:
0119             new = &((*new)->rb_right);
0120             break;
0121         default:
0122             return -EEXIST;
0123         }
0124     }
0125 
0126     rb_link_node(&storage->node, parent, new);
0127     rb_insert_color(&storage->node, root);
0128 
0129     return 0;
0130 }
0131 
0132 static void *cgroup_storage_lookup_elem(struct bpf_map *_map, void *key)
0133 {
0134     struct bpf_cgroup_storage_map *map = map_to_storage(_map);
0135     struct bpf_cgroup_storage *storage;
0136 
0137     storage = cgroup_storage_lookup(map, key, false);
0138     if (!storage)
0139         return NULL;
0140 
0141     return &READ_ONCE(storage->buf)->data[0];
0142 }
0143 
0144 static int cgroup_storage_update_elem(struct bpf_map *map, void *key,
0145                       void *value, u64 flags)
0146 {
0147     struct bpf_cgroup_storage *storage;
0148     struct bpf_storage_buffer *new;
0149 
0150     if (unlikely(flags & ~(BPF_F_LOCK | BPF_EXIST)))
0151         return -EINVAL;
0152 
0153     if (unlikely((flags & BPF_F_LOCK) &&
0154              !map_value_has_spin_lock(map)))
0155         return -EINVAL;
0156 
0157     storage = cgroup_storage_lookup((struct bpf_cgroup_storage_map *)map,
0158                     key, false);
0159     if (!storage)
0160         return -ENOENT;
0161 
0162     if (flags & BPF_F_LOCK) {
0163         copy_map_value_locked(map, storage->buf->data, value, false);
0164         return 0;
0165     }
0166 
0167     new = bpf_map_kmalloc_node(map, struct_size(new, data, map->value_size),
0168                    __GFP_ZERO | GFP_NOWAIT | __GFP_NOWARN,
0169                    map->numa_node);
0170     if (!new)
0171         return -ENOMEM;
0172 
0173     memcpy(&new->data[0], value, map->value_size);
0174     check_and_init_map_value(map, new->data);
0175 
0176     new = xchg(&storage->buf, new);
0177     kfree_rcu(new, rcu);
0178 
0179     return 0;
0180 }
0181 
0182 int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *key,
0183                    void *value)
0184 {
0185     struct bpf_cgroup_storage_map *map = map_to_storage(_map);
0186     struct bpf_cgroup_storage *storage;
0187     int cpu, off = 0;
0188     u32 size;
0189 
0190     rcu_read_lock();
0191     storage = cgroup_storage_lookup(map, key, false);
0192     if (!storage) {
0193         rcu_read_unlock();
0194         return -ENOENT;
0195     }
0196 
0197     /* per_cpu areas are zero-filled and bpf programs can only
0198      * access 'value_size' of them, so copying rounded areas
0199      * will not leak any kernel data
0200      */
0201     size = round_up(_map->value_size, 8);
0202     for_each_possible_cpu(cpu) {
0203         bpf_long_memcpy(value + off,
0204                 per_cpu_ptr(storage->percpu_buf, cpu), size);
0205         off += size;
0206     }
0207     rcu_read_unlock();
0208     return 0;
0209 }
0210 
0211 int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *key,
0212                      void *value, u64 map_flags)
0213 {
0214     struct bpf_cgroup_storage_map *map = map_to_storage(_map);
0215     struct bpf_cgroup_storage *storage;
0216     int cpu, off = 0;
0217     u32 size;
0218 
0219     if (map_flags != BPF_ANY && map_flags != BPF_EXIST)
0220         return -EINVAL;
0221 
0222     rcu_read_lock();
0223     storage = cgroup_storage_lookup(map, key, false);
0224     if (!storage) {
0225         rcu_read_unlock();
0226         return -ENOENT;
0227     }
0228 
0229     /* the user space will provide round_up(value_size, 8) bytes that
0230      * will be copied into per-cpu area. bpf programs can only access
0231      * value_size of it. During lookup the same extra bytes will be
0232      * returned or zeros which were zero-filled by percpu_alloc,
0233      * so no kernel data leaks possible
0234      */
0235     size = round_up(_map->value_size, 8);
0236     for_each_possible_cpu(cpu) {
0237         bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu),
0238                 value + off, size);
0239         off += size;
0240     }
0241     rcu_read_unlock();
0242     return 0;
0243 }
0244 
0245 static int cgroup_storage_get_next_key(struct bpf_map *_map, void *key,
0246                        void *_next_key)
0247 {
0248     struct bpf_cgroup_storage_map *map = map_to_storage(_map);
0249     struct bpf_cgroup_storage *storage;
0250 
0251     spin_lock_bh(&map->lock);
0252 
0253     if (list_empty(&map->list))
0254         goto enoent;
0255 
0256     if (key) {
0257         storage = cgroup_storage_lookup(map, key, true);
0258         if (!storage)
0259             goto enoent;
0260 
0261         storage = list_next_entry(storage, list_map);
0262         if (!storage)
0263             goto enoent;
0264     } else {
0265         storage = list_first_entry(&map->list,
0266                      struct bpf_cgroup_storage, list_map);
0267     }
0268 
0269     spin_unlock_bh(&map->lock);
0270 
0271     if (attach_type_isolated(&map->map)) {
0272         struct bpf_cgroup_storage_key *next = _next_key;
0273         *next = storage->key;
0274     } else {
0275         __u64 *next = _next_key;
0276         *next = storage->key.cgroup_inode_id;
0277     }
0278     return 0;
0279 
0280 enoent:
0281     spin_unlock_bh(&map->lock);
0282     return -ENOENT;
0283 }
0284 
0285 static struct bpf_map *cgroup_storage_map_alloc(union bpf_attr *attr)
0286 {
0287     __u32 max_value_size = BPF_LOCAL_STORAGE_MAX_VALUE_SIZE;
0288     int numa_node = bpf_map_attr_numa_node(attr);
0289     struct bpf_cgroup_storage_map *map;
0290 
0291     /* percpu is bound by PCPU_MIN_UNIT_SIZE, non-percu
0292      * is the same as other local storages.
0293      */
0294     if (attr->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
0295         max_value_size = min_t(__u32, max_value_size,
0296                        PCPU_MIN_UNIT_SIZE);
0297 
0298     if (attr->key_size != sizeof(struct bpf_cgroup_storage_key) &&
0299         attr->key_size != sizeof(__u64))
0300         return ERR_PTR(-EINVAL);
0301 
0302     if (attr->value_size == 0)
0303         return ERR_PTR(-EINVAL);
0304 
0305     if (attr->value_size > max_value_size)
0306         return ERR_PTR(-E2BIG);
0307 
0308     if (attr->map_flags & ~LOCAL_STORAGE_CREATE_FLAG_MASK ||
0309         !bpf_map_flags_access_ok(attr->map_flags))
0310         return ERR_PTR(-EINVAL);
0311 
0312     if (attr->max_entries)
0313         /* max_entries is not used and enforced to be 0 */
0314         return ERR_PTR(-EINVAL);
0315 
0316     map = kmalloc_node(sizeof(struct bpf_cgroup_storage_map),
0317                __GFP_ZERO | GFP_USER | __GFP_ACCOUNT, numa_node);
0318     if (!map)
0319         return ERR_PTR(-ENOMEM);
0320 
0321     /* copy mandatory map attributes */
0322     bpf_map_init_from_attr(&map->map, attr);
0323 
0324     spin_lock_init(&map->lock);
0325     map->root = RB_ROOT;
0326     INIT_LIST_HEAD(&map->list);
0327 
0328     return &map->map;
0329 }
0330 
0331 static void cgroup_storage_map_free(struct bpf_map *_map)
0332 {
0333     struct bpf_cgroup_storage_map *map = map_to_storage(_map);
0334     struct list_head *storages = &map->list;
0335     struct bpf_cgroup_storage *storage, *stmp;
0336 
0337     mutex_lock(&cgroup_mutex);
0338 
0339     list_for_each_entry_safe(storage, stmp, storages, list_map) {
0340         bpf_cgroup_storage_unlink(storage);
0341         bpf_cgroup_storage_free(storage);
0342     }
0343 
0344     mutex_unlock(&cgroup_mutex);
0345 
0346     WARN_ON(!RB_EMPTY_ROOT(&map->root));
0347     WARN_ON(!list_empty(&map->list));
0348 
0349     kfree(map);
0350 }
0351 
0352 static int cgroup_storage_delete_elem(struct bpf_map *map, void *key)
0353 {
0354     return -EINVAL;
0355 }
0356 
0357 static int cgroup_storage_check_btf(const struct bpf_map *map,
0358                     const struct btf *btf,
0359                     const struct btf_type *key_type,
0360                     const struct btf_type *value_type)
0361 {
0362     if (attach_type_isolated(map)) {
0363         struct btf_member *m;
0364         u32 offset, size;
0365 
0366         /* Key is expected to be of struct bpf_cgroup_storage_key type,
0367          * which is:
0368          * struct bpf_cgroup_storage_key {
0369          *  __u64   cgroup_inode_id;
0370          *  __u32   attach_type;
0371          * };
0372          */
0373 
0374         /*
0375          * Key_type must be a structure with two fields.
0376          */
0377         if (BTF_INFO_KIND(key_type->info) != BTF_KIND_STRUCT ||
0378             BTF_INFO_VLEN(key_type->info) != 2)
0379             return -EINVAL;
0380 
0381         /*
0382          * The first field must be a 64 bit integer at 0 offset.
0383          */
0384         m = (struct btf_member *)(key_type + 1);
0385         size = sizeof_field(struct bpf_cgroup_storage_key, cgroup_inode_id);
0386         if (!btf_member_is_reg_int(btf, key_type, m, 0, size))
0387             return -EINVAL;
0388 
0389         /*
0390          * The second field must be a 32 bit integer at 64 bit offset.
0391          */
0392         m++;
0393         offset = offsetof(struct bpf_cgroup_storage_key, attach_type);
0394         size = sizeof_field(struct bpf_cgroup_storage_key, attach_type);
0395         if (!btf_member_is_reg_int(btf, key_type, m, offset, size))
0396             return -EINVAL;
0397     } else {
0398         u32 int_data;
0399 
0400         /*
0401          * Key is expected to be u64, which stores the cgroup_inode_id
0402          */
0403 
0404         if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT)
0405             return -EINVAL;
0406 
0407         int_data = *(u32 *)(key_type + 1);
0408         if (BTF_INT_BITS(int_data) != 64 || BTF_INT_OFFSET(int_data))
0409             return -EINVAL;
0410     }
0411 
0412     return 0;
0413 }
0414 
0415 static void cgroup_storage_seq_show_elem(struct bpf_map *map, void *key,
0416                      struct seq_file *m)
0417 {
0418     enum bpf_cgroup_storage_type stype;
0419     struct bpf_cgroup_storage *storage;
0420     int cpu;
0421 
0422     rcu_read_lock();
0423     storage = cgroup_storage_lookup(map_to_storage(map), key, false);
0424     if (!storage) {
0425         rcu_read_unlock();
0426         return;
0427     }
0428 
0429     btf_type_seq_show(map->btf, map->btf_key_type_id, key, m);
0430     stype = cgroup_storage_type(map);
0431     if (stype == BPF_CGROUP_STORAGE_SHARED) {
0432         seq_puts(m, ": ");
0433         btf_type_seq_show(map->btf, map->btf_value_type_id,
0434                   &READ_ONCE(storage->buf)->data[0], m);
0435         seq_puts(m, "\n");
0436     } else {
0437         seq_puts(m, ": {\n");
0438         for_each_possible_cpu(cpu) {
0439             seq_printf(m, "\tcpu%d: ", cpu);
0440             btf_type_seq_show(map->btf, map->btf_value_type_id,
0441                       per_cpu_ptr(storage->percpu_buf, cpu),
0442                       m);
0443             seq_puts(m, "\n");
0444         }
0445         seq_puts(m, "}\n");
0446     }
0447     rcu_read_unlock();
0448 }
0449 
0450 BTF_ID_LIST_SINGLE(cgroup_storage_map_btf_ids, struct,
0451            bpf_cgroup_storage_map)
0452 const struct bpf_map_ops cgroup_storage_map_ops = {
0453     .map_alloc = cgroup_storage_map_alloc,
0454     .map_free = cgroup_storage_map_free,
0455     .map_get_next_key = cgroup_storage_get_next_key,
0456     .map_lookup_elem = cgroup_storage_lookup_elem,
0457     .map_update_elem = cgroup_storage_update_elem,
0458     .map_delete_elem = cgroup_storage_delete_elem,
0459     .map_check_btf = cgroup_storage_check_btf,
0460     .map_seq_show_elem = cgroup_storage_seq_show_elem,
0461     .map_btf_id = &cgroup_storage_map_btf_ids[0],
0462 };
0463 
0464 int bpf_cgroup_storage_assign(struct bpf_prog_aux *aux, struct bpf_map *_map)
0465 {
0466     enum bpf_cgroup_storage_type stype = cgroup_storage_type(_map);
0467 
0468     if (aux->cgroup_storage[stype] &&
0469         aux->cgroup_storage[stype] != _map)
0470         return -EBUSY;
0471 
0472     aux->cgroup_storage[stype] = _map;
0473     return 0;
0474 }
0475 
0476 static size_t bpf_cgroup_storage_calculate_size(struct bpf_map *map, u32 *pages)
0477 {
0478     size_t size;
0479 
0480     if (cgroup_storage_type(map) == BPF_CGROUP_STORAGE_SHARED) {
0481         size = sizeof(struct bpf_storage_buffer) + map->value_size;
0482         *pages = round_up(sizeof(struct bpf_cgroup_storage) + size,
0483                   PAGE_SIZE) >> PAGE_SHIFT;
0484     } else {
0485         size = map->value_size;
0486         *pages = round_up(round_up(size, 8) * num_possible_cpus(),
0487                   PAGE_SIZE) >> PAGE_SHIFT;
0488     }
0489 
0490     return size;
0491 }
0492 
0493 struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(struct bpf_prog *prog,
0494                     enum bpf_cgroup_storage_type stype)
0495 {
0496     const gfp_t gfp = __GFP_ZERO | GFP_USER;
0497     struct bpf_cgroup_storage *storage;
0498     struct bpf_map *map;
0499     size_t size;
0500     u32 pages;
0501 
0502     map = prog->aux->cgroup_storage[stype];
0503     if (!map)
0504         return NULL;
0505 
0506     size = bpf_cgroup_storage_calculate_size(map, &pages);
0507 
0508     storage = bpf_map_kmalloc_node(map, sizeof(struct bpf_cgroup_storage),
0509                        gfp, map->numa_node);
0510     if (!storage)
0511         goto enomem;
0512 
0513     if (stype == BPF_CGROUP_STORAGE_SHARED) {
0514         storage->buf = bpf_map_kmalloc_node(map, size, gfp,
0515                             map->numa_node);
0516         if (!storage->buf)
0517             goto enomem;
0518         check_and_init_map_value(map, storage->buf->data);
0519     } else {
0520         storage->percpu_buf = bpf_map_alloc_percpu(map, size, 8, gfp);
0521         if (!storage->percpu_buf)
0522             goto enomem;
0523     }
0524 
0525     storage->map = (struct bpf_cgroup_storage_map *)map;
0526 
0527     return storage;
0528 
0529 enomem:
0530     kfree(storage);
0531     return ERR_PTR(-ENOMEM);
0532 }
0533 
0534 static void free_shared_cgroup_storage_rcu(struct rcu_head *rcu)
0535 {
0536     struct bpf_cgroup_storage *storage =
0537         container_of(rcu, struct bpf_cgroup_storage, rcu);
0538 
0539     kfree(storage->buf);
0540     kfree(storage);
0541 }
0542 
0543 static void free_percpu_cgroup_storage_rcu(struct rcu_head *rcu)
0544 {
0545     struct bpf_cgroup_storage *storage =
0546         container_of(rcu, struct bpf_cgroup_storage, rcu);
0547 
0548     free_percpu(storage->percpu_buf);
0549     kfree(storage);
0550 }
0551 
0552 void bpf_cgroup_storage_free(struct bpf_cgroup_storage *storage)
0553 {
0554     enum bpf_cgroup_storage_type stype;
0555     struct bpf_map *map;
0556 
0557     if (!storage)
0558         return;
0559 
0560     map = &storage->map->map;
0561     stype = cgroup_storage_type(map);
0562     if (stype == BPF_CGROUP_STORAGE_SHARED)
0563         call_rcu(&storage->rcu, free_shared_cgroup_storage_rcu);
0564     else
0565         call_rcu(&storage->rcu, free_percpu_cgroup_storage_rcu);
0566 }
0567 
0568 void bpf_cgroup_storage_link(struct bpf_cgroup_storage *storage,
0569                  struct cgroup *cgroup,
0570                  enum bpf_attach_type type)
0571 {
0572     struct bpf_cgroup_storage_map *map;
0573 
0574     if (!storage)
0575         return;
0576 
0577     storage->key.attach_type = type;
0578     storage->key.cgroup_inode_id = cgroup_id(cgroup);
0579 
0580     map = storage->map;
0581 
0582     spin_lock_bh(&map->lock);
0583     WARN_ON(cgroup_storage_insert(map, storage));
0584     list_add(&storage->list_map, &map->list);
0585     list_add(&storage->list_cg, &cgroup->bpf.storages);
0586     spin_unlock_bh(&map->lock);
0587 }
0588 
0589 void bpf_cgroup_storage_unlink(struct bpf_cgroup_storage *storage)
0590 {
0591     struct bpf_cgroup_storage_map *map;
0592     struct rb_root *root;
0593 
0594     if (!storage)
0595         return;
0596 
0597     map = storage->map;
0598 
0599     spin_lock_bh(&map->lock);
0600     root = &map->root;
0601     rb_erase(&storage->node, root);
0602 
0603     list_del(&storage->list_map);
0604     list_del(&storage->list_cg);
0605     spin_unlock_bh(&map->lock);
0606 }
0607 
0608 #endif
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!