1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Longest prefix match list implementation
4 *
5 * Copyright (c) 2016,2017 Daniel Mack
6 * Copyright (c) 2016 David Herrmann
7 */
8
9 #include <linux/bpf.h>
10 #include <linux/btf.h>
11 #include <linux/err.h>
12 #include <linux/slab.h>
13 #include <linux/spinlock.h>
14 #include <linux/vmalloc.h>
15 #include <net/ipv6.h>
16 #include <uapi/linux/btf.h>
17 #include <linux/btf_ids.h>
18
19 /* Intermediate node */
20 #define LPM_TREE_NODE_FLAG_IM BIT(0)
21
22 struct lpm_trie_node;
23
24 struct lpm_trie_node {
25 struct rcu_head rcu;
26 struct lpm_trie_node __rcu *child[2];
27 u32 prefixlen;
28 u32 flags;
29 u8 data[];
30 };
31
32 struct lpm_trie {
33 struct bpf_map map;
34 struct lpm_trie_node __rcu *root;
35 size_t n_entries;
36 size_t max_prefixlen;
37 size_t data_size;
38 spinlock_t lock;
39 };
40
41 /* This trie implements a longest prefix match algorithm that can be used to
42 * match IP addresses to a stored set of ranges.
43 *
44 * Data stored in @data of struct bpf_lpm_key and struct lpm_trie_node is
45 * interpreted as big endian, so data[0] stores the most significant byte.
46 *
47 * Match ranges are internally stored in instances of struct lpm_trie_node
48 * which each contain their prefix length as well as two pointers that may
49 * lead to more nodes containing more specific matches. Each node also stores
50 * a value that is defined by and returned to userspace via the update_elem
51 * and lookup functions.
52 *
53 * For instance, let's start with a trie that was created with a prefix length
54 * of 32, so it can be used for IPv4 addresses, and one single element that
55 * matches 192.168.0.0/16. The data array would hence contain
56 * [0xc0, 0xa8, 0x00, 0x00] in big-endian notation. This documentation will
57 * stick to IP-address notation for readability though.
58 *
59 * As the trie is empty initially, the new node (1) will be places as root
60 * node, denoted as (R) in the example below. As there are no other node, both
61 * child pointers are %NULL.
62 *
63 * +----------------+
64 * | (1) (R) |
65 * | 192.168.0.0/16 |
66 * | value: 1 |
67 * | [0] [1] |
68 * +----------------+
69 *
70 * Next, let's add a new node (2) matching 192.168.0.0/24. As there is already
71 * a node with the same data and a smaller prefix (ie, a less specific one),
72 * node (2) will become a child of (1). In child index depends on the next bit
73 * that is outside of what (1) matches, and that bit is 0, so (2) will be
74 * child[0] of (1):
75 *
76 * +----------------+
77 * | (1) (R) |
78 * | 192.168.0.0/16 |
79 * | value: 1 |
80 * | [0] [1] |
81 * +----------------+
82 * |
83 * +----------------+
84 * | (2) |
85 * | 192.168.0.0/24 |
86 * | value: 2 |
87 * | [0] [1] |
88 * +----------------+
89 *
90 * The child[1] slot of (1) could be filled with another node which has bit #17
91 * (the next bit after the ones that (1) matches on) set to 1. For instance,
92 * 192.168.128.0/24:
93 *
94 * +----------------+
95 * | (1) (R) |
96 * | 192.168.0.0/16 |
97 * | value: 1 |
98 * | [0] [1] |
99 * +----------------+
100 * | |
101 * +----------------+ +------------------+
102 * | (2) | | (3) |
103 * | 192.168.0.0/24 | | 192.168.128.0/24 |
104 * | value: 2 | | value: 3 |
105 * | [0] [1] | | [0] [1] |
106 * +----------------+ +------------------+
107 *
108 * Let's add another node (4) to the game for 192.168.1.0/24. In order to place
109 * it, node (1) is looked at first, and because (4) of the semantics laid out
110 * above (bit #17 is 0), it would normally be attached to (1) as child[0].
111 * However, that slot is already allocated, so a new node is needed in between.
112 * That node does not have a value attached to it and it will never be
113 * returned to users as result of a lookup. It is only there to differentiate
114 * the traversal further. It will get a prefix as wide as necessary to
115 * distinguish its two children:
116 *
117 * +----------------+
118 * | (1) (R) |
119 * | 192.168.0.0/16 |
120 * | value: 1 |
121 * | [0] [1] |
122 * +----------------+
123 * | |
124 * +----------------+ +------------------+
125 * | (4) (I) | | (3) |
126 * | 192.168.0.0/23 | | 192.168.128.0/24 |
127 * | value: --- | | value: 3 |
128 * | [0] [1] | | [0] [1] |
129 * +----------------+ +------------------+
130 * | |
131 * +----------------+ +----------------+
132 * | (2) | | (5) |
133 * | 192.168.0.0/24 | | 192.168.1.0/24 |
134 * | value: 2 | | value: 5 |
135 * | [0] [1] | | [0] [1] |
136 * +----------------+ +----------------+
137 *
138 * 192.168.1.1/32 would be a child of (5) etc.
139 *
140 * An intermediate node will be turned into a 'real' node on demand. In the
141 * example above, (4) would be re-used if 192.168.0.0/23 is added to the trie.
142 *
143 * A fully populated trie would have a height of 32 nodes, as the trie was
144 * created with a prefix length of 32.
145 *
146 * The lookup starts at the root node. If the current node matches and if there
147 * is a child that can be used to become more specific, the trie is traversed
148 * downwards. The last node in the traversal that is a non-intermediate one is
149 * returned.
150 */
151
extract_bit(const u8 * data,size_t index)152 static inline int extract_bit(const u8 *data, size_t index)
153 {
154 return !!(data[index / 8] & (1 << (7 - (index % 8))));
155 }
156
157 /**
158 * longest_prefix_match() - determine the longest prefix
159 * @trie: The trie to get internal sizes from
160 * @node: The node to operate on
161 * @key: The key to compare to @node
162 *
163 * Determine the longest prefix of @node that matches the bits in @key.
164 */
longest_prefix_match(const struct lpm_trie * trie,const struct lpm_trie_node * node,const struct bpf_lpm_trie_key_u8 * key)165 static size_t longest_prefix_match(const struct lpm_trie *trie,
166 const struct lpm_trie_node *node,
167 const struct bpf_lpm_trie_key_u8 *key)
168 {
169 u32 limit = min(node->prefixlen, key->prefixlen);
170 u32 prefixlen = 0, i = 0;
171
172 BUILD_BUG_ON(offsetof(struct lpm_trie_node, data) % sizeof(u32));
173 BUILD_BUG_ON(offsetof(struct bpf_lpm_trie_key_u8, data) % sizeof(u32));
174
175 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && defined(CONFIG_64BIT)
176
177 /* data_size >= 16 has very small probability.
178 * We do not use a loop for optimal code generation.
179 */
180 if (trie->data_size >= 8) {
181 u64 diff = be64_to_cpu(*(__be64 *)node->data ^
182 *(__be64 *)key->data);
183
184 prefixlen = 64 - fls64(diff);
185 if (prefixlen >= limit)
186 return limit;
187 if (diff)
188 return prefixlen;
189 i = 8;
190 }
191 #endif
192
193 while (trie->data_size >= i + 4) {
194 u32 diff = be32_to_cpu(*(__be32 *)&node->data[i] ^
195 *(__be32 *)&key->data[i]);
196
197 prefixlen += 32 - fls(diff);
198 if (prefixlen >= limit)
199 return limit;
200 if (diff)
201 return prefixlen;
202 i += 4;
203 }
204
205 if (trie->data_size >= i + 2) {
206 u16 diff = be16_to_cpu(*(__be16 *)&node->data[i] ^
207 *(__be16 *)&key->data[i]);
208
209 prefixlen += 16 - fls(diff);
210 if (prefixlen >= limit)
211 return limit;
212 if (diff)
213 return prefixlen;
214 i += 2;
215 }
216
217 if (trie->data_size >= i + 1) {
218 prefixlen += 8 - fls(node->data[i] ^ key->data[i]);
219
220 if (prefixlen >= limit)
221 return limit;
222 }
223
224 return prefixlen;
225 }
226
227 /* Called from syscall or from eBPF program */
trie_lookup_elem(struct bpf_map * map,void * _key)228 static void *trie_lookup_elem(struct bpf_map *map, void *_key)
229 {
230 struct lpm_trie *trie = container_of(map, struct lpm_trie, map);
231 struct lpm_trie_node *node, *found = NULL;
232 struct bpf_lpm_trie_key_u8 *key = _key;
233
234 if (key->prefixlen > trie->max_prefixlen)
235 return NULL;
236
237 /* Start walking the trie from the root node ... */
238
239 for (node = rcu_dereference_check(trie->root, rcu_read_lock_bh_held());
240 node;) {
241 unsigned int next_bit;
242 size_t matchlen;
243
244 /* Determine the longest prefix of @node that matches @key.
245 * If it's the maximum possible prefix for this trie, we have
246 * an exact match and can return it directly.
247 */
248 matchlen = longest_prefix_match(trie, node, key);
249 if (matchlen == trie->max_prefixlen) {
250 found = node;
251 break;
252 }
253
254 /* If the number of bits that match is smaller than the prefix
255 * length of @node, bail out and return the node we have seen
256 * last in the traversal (ie, the parent).
257 */
258 if (matchlen < node->prefixlen)
259 break;
260
261 /* Consider this node as return candidate unless it is an
262 * artificially added intermediate one.
263 */
264 if (!(node->flags & LPM_TREE_NODE_FLAG_IM))
265 found = node;
266
267 /* If the node match is fully satisfied, let's see if we can
268 * become more specific. Determine the next bit in the key and
269 * traverse down.
270 */
271 next_bit = extract_bit(key->data, node->prefixlen);
272 node = rcu_dereference_check(node->child[next_bit],
273 rcu_read_lock_bh_held());
274 }
275
276 if (!found)
277 return NULL;
278
279 return found->data + trie->data_size;
280 }
281
lpm_trie_node_alloc(const struct lpm_trie * trie,const void * value)282 static struct lpm_trie_node *lpm_trie_node_alloc(const struct lpm_trie *trie,
283 const void *value)
284 {
285 struct lpm_trie_node *node;
286 size_t size = sizeof(struct lpm_trie_node) + trie->data_size;
287
288 if (value)
289 size += trie->map.value_size;
290
291 node = bpf_map_kmalloc_node(&trie->map, size, GFP_NOWAIT | __GFP_NOWARN,
292 trie->map.numa_node);
293 if (!node)
294 return NULL;
295
296 node->flags = 0;
297
298 if (value)
299 memcpy(node->data + trie->data_size, value,
300 trie->map.value_size);
301
302 return node;
303 }
304
trie_check_add_elem(struct lpm_trie * trie,u64 flags)305 static int trie_check_add_elem(struct lpm_trie *trie, u64 flags)
306 {
307 if (flags == BPF_EXIST)
308 return -ENOENT;
309 if (trie->n_entries == trie->map.max_entries)
310 return -ENOSPC;
311 trie->n_entries++;
312 return 0;
313 }
314
315 /* Called from syscall or from eBPF program */
trie_update_elem(struct bpf_map * map,void * _key,void * value,u64 flags)316 static long trie_update_elem(struct bpf_map *map,
317 void *_key, void *value, u64 flags)
318 {
319 struct lpm_trie *trie = container_of(map, struct lpm_trie, map);
320 struct lpm_trie_node *node, *im_node, *new_node = NULL;
321 struct lpm_trie_node *free_node = NULL;
322 struct lpm_trie_node __rcu **slot;
323 struct bpf_lpm_trie_key_u8 *key = _key;
324 unsigned long irq_flags;
325 unsigned int next_bit;
326 size_t matchlen = 0;
327 int ret = 0;
328
329 if (unlikely(flags > BPF_EXIST))
330 return -EINVAL;
331
332 if (key->prefixlen > trie->max_prefixlen)
333 return -EINVAL;
334
335 spin_lock_irqsave(&trie->lock, irq_flags);
336
337 /* Allocate and fill a new node */
338 new_node = lpm_trie_node_alloc(trie, value);
339 if (!new_node) {
340 ret = -ENOMEM;
341 goto out;
342 }
343
344 new_node->prefixlen = key->prefixlen;
345 RCU_INIT_POINTER(new_node->child[0], NULL);
346 RCU_INIT_POINTER(new_node->child[1], NULL);
347 memcpy(new_node->data, key->data, trie->data_size);
348
349 /* Now find a slot to attach the new node. To do that, walk the tree
350 * from the root and match as many bits as possible for each node until
351 * we either find an empty slot or a slot that needs to be replaced by
352 * an intermediate node.
353 */
354 slot = &trie->root;
355
356 while ((node = rcu_dereference_protected(*slot,
357 lockdep_is_held(&trie->lock)))) {
358 matchlen = longest_prefix_match(trie, node, key);
359
360 if (node->prefixlen != matchlen ||
361 node->prefixlen == key->prefixlen ||
362 node->prefixlen == trie->max_prefixlen)
363 break;
364
365 next_bit = extract_bit(key->data, node->prefixlen);
366 slot = &node->child[next_bit];
367 }
368
369 /* If the slot is empty (a free child pointer or an empty root),
370 * simply assign the @new_node to that slot and be done.
371 */
372 if (!node) {
373 ret = trie_check_add_elem(trie, flags);
374 if (ret)
375 goto out;
376
377 rcu_assign_pointer(*slot, new_node);
378 goto out;
379 }
380
381 /* If the slot we picked already exists, replace it with @new_node
382 * which already has the correct data array set.
383 */
384 if (node->prefixlen == matchlen) {
385 if (!(node->flags & LPM_TREE_NODE_FLAG_IM)) {
386 if (flags == BPF_NOEXIST) {
387 ret = -EEXIST;
388 goto out;
389 }
390 } else {
391 ret = trie_check_add_elem(trie, flags);
392 if (ret)
393 goto out;
394 }
395
396 new_node->child[0] = node->child[0];
397 new_node->child[1] = node->child[1];
398
399 rcu_assign_pointer(*slot, new_node);
400 free_node = node;
401
402 goto out;
403 }
404
405 ret = trie_check_add_elem(trie, flags);
406 if (ret)
407 goto out;
408
409 /* If the new node matches the prefix completely, it must be inserted
410 * as an ancestor. Simply insert it between @node and *@slot.
411 */
412 if (matchlen == key->prefixlen) {
413 next_bit = extract_bit(node->data, matchlen);
414 rcu_assign_pointer(new_node->child[next_bit], node);
415 rcu_assign_pointer(*slot, new_node);
416 goto out;
417 }
418
419 im_node = lpm_trie_node_alloc(trie, NULL);
420 if (!im_node) {
421 trie->n_entries--;
422 ret = -ENOMEM;
423 goto out;
424 }
425
426 im_node->prefixlen = matchlen;
427 im_node->flags |= LPM_TREE_NODE_FLAG_IM;
428 memcpy(im_node->data, node->data, trie->data_size);
429
430 /* Now determine which child to install in which slot */
431 if (extract_bit(key->data, matchlen)) {
432 rcu_assign_pointer(im_node->child[0], node);
433 rcu_assign_pointer(im_node->child[1], new_node);
434 } else {
435 rcu_assign_pointer(im_node->child[0], new_node);
436 rcu_assign_pointer(im_node->child[1], node);
437 }
438
439 /* Finally, assign the intermediate node to the determined slot */
440 rcu_assign_pointer(*slot, im_node);
441
442 out:
443 if (ret)
444 kfree(new_node);
445 spin_unlock_irqrestore(&trie->lock, irq_flags);
446 kfree_rcu(free_node, rcu);
447
448 return ret;
449 }
450
451 /* Called from syscall or from eBPF program */
trie_delete_elem(struct bpf_map * map,void * _key)452 static long trie_delete_elem(struct bpf_map *map, void *_key)
453 {
454 struct lpm_trie *trie = container_of(map, struct lpm_trie, map);
455 struct lpm_trie_node *free_node = NULL, *free_parent = NULL;
456 struct bpf_lpm_trie_key_u8 *key = _key;
457 struct lpm_trie_node __rcu **trim, **trim2;
458 struct lpm_trie_node *node, *parent;
459 unsigned long irq_flags;
460 unsigned int next_bit;
461 size_t matchlen = 0;
462 int ret = 0;
463
464 if (key->prefixlen > trie->max_prefixlen)
465 return -EINVAL;
466
467 spin_lock_irqsave(&trie->lock, irq_flags);
468
469 /* Walk the tree looking for an exact key/length match and keeping
470 * track of the path we traverse. We will need to know the node
471 * we wish to delete, and the slot that points to the node we want
472 * to delete. We may also need to know the nodes parent and the
473 * slot that contains it.
474 */
475 trim = &trie->root;
476 trim2 = trim;
477 parent = NULL;
478 while ((node = rcu_dereference_protected(
479 *trim, lockdep_is_held(&trie->lock)))) {
480 matchlen = longest_prefix_match(trie, node, key);
481
482 if (node->prefixlen != matchlen ||
483 node->prefixlen == key->prefixlen)
484 break;
485
486 parent = node;
487 trim2 = trim;
488 next_bit = extract_bit(key->data, node->prefixlen);
489 trim = &node->child[next_bit];
490 }
491
492 if (!node || node->prefixlen != key->prefixlen ||
493 node->prefixlen != matchlen ||
494 (node->flags & LPM_TREE_NODE_FLAG_IM)) {
495 ret = -ENOENT;
496 goto out;
497 }
498
499 trie->n_entries--;
500
501 /* If the node we are removing has two children, simply mark it
502 * as intermediate and we are done.
503 */
504 if (rcu_access_pointer(node->child[0]) &&
505 rcu_access_pointer(node->child[1])) {
506 node->flags |= LPM_TREE_NODE_FLAG_IM;
507 goto out;
508 }
509
510 /* If the parent of the node we are about to delete is an intermediate
511 * node, and the deleted node doesn't have any children, we can delete
512 * the intermediate parent as well and promote its other child
513 * up the tree. Doing this maintains the invariant that all
514 * intermediate nodes have exactly 2 children and that there are no
515 * unnecessary intermediate nodes in the tree.
516 */
517 if (parent && (parent->flags & LPM_TREE_NODE_FLAG_IM) &&
518 !node->child[0] && !node->child[1]) {
519 if (node == rcu_access_pointer(parent->child[0]))
520 rcu_assign_pointer(
521 *trim2, rcu_access_pointer(parent->child[1]));
522 else
523 rcu_assign_pointer(
524 *trim2, rcu_access_pointer(parent->child[0]));
525 free_parent = parent;
526 free_node = node;
527 goto out;
528 }
529
530 /* The node we are removing has either zero or one child. If there
531 * is a child, move it into the removed node's slot then delete
532 * the node. Otherwise just clear the slot and delete the node.
533 */
534 if (node->child[0])
535 rcu_assign_pointer(*trim, rcu_access_pointer(node->child[0]));
536 else if (node->child[1])
537 rcu_assign_pointer(*trim, rcu_access_pointer(node->child[1]));
538 else
539 RCU_INIT_POINTER(*trim, NULL);
540 free_node = node;
541
542 out:
543 spin_unlock_irqrestore(&trie->lock, irq_flags);
544 kfree_rcu(free_parent, rcu);
545 kfree_rcu(free_node, rcu);
546
547 return ret;
548 }
549
550 #define LPM_DATA_SIZE_MAX 256
551 #define LPM_DATA_SIZE_MIN 1
552
553 #define LPM_VAL_SIZE_MAX (KMALLOC_MAX_SIZE - LPM_DATA_SIZE_MAX - \
554 sizeof(struct lpm_trie_node))
555 #define LPM_VAL_SIZE_MIN 1
556
557 #define LPM_KEY_SIZE(X) (sizeof(struct bpf_lpm_trie_key_u8) + (X))
558 #define LPM_KEY_SIZE_MAX LPM_KEY_SIZE(LPM_DATA_SIZE_MAX)
559 #define LPM_KEY_SIZE_MIN LPM_KEY_SIZE(LPM_DATA_SIZE_MIN)
560
561 #define LPM_CREATE_FLAG_MASK (BPF_F_NO_PREALLOC | BPF_F_NUMA_NODE | \
562 BPF_F_ACCESS_MASK)
563
trie_alloc(union bpf_attr * attr)564 static struct bpf_map *trie_alloc(union bpf_attr *attr)
565 {
566 struct lpm_trie *trie;
567
568 /* check sanity of attributes */
569 if (attr->max_entries == 0 ||
570 !(attr->map_flags & BPF_F_NO_PREALLOC) ||
571 attr->map_flags & ~LPM_CREATE_FLAG_MASK ||
572 !bpf_map_flags_access_ok(attr->map_flags) ||
573 attr->key_size < LPM_KEY_SIZE_MIN ||
574 attr->key_size > LPM_KEY_SIZE_MAX ||
575 attr->value_size < LPM_VAL_SIZE_MIN ||
576 attr->value_size > LPM_VAL_SIZE_MAX)
577 return ERR_PTR(-EINVAL);
578
579 trie = bpf_map_area_alloc(sizeof(*trie), NUMA_NO_NODE);
580 if (!trie)
581 return ERR_PTR(-ENOMEM);
582
583 /* copy mandatory map attributes */
584 bpf_map_init_from_attr(&trie->map, attr);
585 trie->data_size = attr->key_size -
586 offsetof(struct bpf_lpm_trie_key_u8, data);
587 trie->max_prefixlen = trie->data_size * 8;
588
589 spin_lock_init(&trie->lock);
590
591 return &trie->map;
592 }
593
trie_free(struct bpf_map * map)594 static void trie_free(struct bpf_map *map)
595 {
596 struct lpm_trie *trie = container_of(map, struct lpm_trie, map);
597 struct lpm_trie_node __rcu **slot;
598 struct lpm_trie_node *node;
599
600 /* Always start at the root and walk down to a node that has no
601 * children. Then free that node, nullify its reference in the parent
602 * and start over.
603 */
604
605 for (;;) {
606 slot = &trie->root;
607
608 for (;;) {
609 node = rcu_dereference_protected(*slot, 1);
610 if (!node)
611 goto out;
612
613 if (rcu_access_pointer(node->child[0])) {
614 slot = &node->child[0];
615 continue;
616 }
617
618 if (rcu_access_pointer(node->child[1])) {
619 slot = &node->child[1];
620 continue;
621 }
622
623 kfree(node);
624 RCU_INIT_POINTER(*slot, NULL);
625 break;
626 }
627 }
628
629 out:
630 bpf_map_area_free(trie);
631 }
632
trie_get_next_key(struct bpf_map * map,void * _key,void * _next_key)633 static int trie_get_next_key(struct bpf_map *map, void *_key, void *_next_key)
634 {
635 struct lpm_trie_node *node, *next_node = NULL, *parent, *search_root;
636 struct lpm_trie *trie = container_of(map, struct lpm_trie, map);
637 struct bpf_lpm_trie_key_u8 *key = _key, *next_key = _next_key;
638 struct lpm_trie_node **node_stack = NULL;
639 int err = 0, stack_ptr = -1;
640 unsigned int next_bit;
641 size_t matchlen = 0;
642
643 /* The get_next_key follows postorder. For the 4 node example in
644 * the top of this file, the trie_get_next_key() returns the following
645 * one after another:
646 * 192.168.0.0/24
647 * 192.168.1.0/24
648 * 192.168.128.0/24
649 * 192.168.0.0/16
650 *
651 * The idea is to return more specific keys before less specific ones.
652 */
653
654 /* Empty trie */
655 search_root = rcu_dereference(trie->root);
656 if (!search_root)
657 return -ENOENT;
658
659 /* For invalid key, find the leftmost node in the trie */
660 if (!key || key->prefixlen > trie->max_prefixlen)
661 goto find_leftmost;
662
663 node_stack = kmalloc_array(trie->max_prefixlen + 1,
664 sizeof(struct lpm_trie_node *),
665 GFP_ATOMIC | __GFP_NOWARN);
666 if (!node_stack)
667 return -ENOMEM;
668
669 /* Try to find the exact node for the given key */
670 for (node = search_root; node;) {
671 node_stack[++stack_ptr] = node;
672 matchlen = longest_prefix_match(trie, node, key);
673 if (node->prefixlen != matchlen ||
674 node->prefixlen == key->prefixlen)
675 break;
676
677 next_bit = extract_bit(key->data, node->prefixlen);
678 node = rcu_dereference(node->child[next_bit]);
679 }
680 if (!node || node->prefixlen != matchlen ||
681 (node->flags & LPM_TREE_NODE_FLAG_IM))
682 goto find_leftmost;
683
684 /* The node with the exactly-matching key has been found,
685 * find the first node in postorder after the matched node.
686 */
687 node = node_stack[stack_ptr];
688 while (stack_ptr > 0) {
689 parent = node_stack[stack_ptr - 1];
690 if (rcu_dereference(parent->child[0]) == node) {
691 search_root = rcu_dereference(parent->child[1]);
692 if (search_root)
693 goto find_leftmost;
694 }
695 if (!(parent->flags & LPM_TREE_NODE_FLAG_IM)) {
696 next_node = parent;
697 goto do_copy;
698 }
699
700 node = parent;
701 stack_ptr--;
702 }
703
704 /* did not find anything */
705 err = -ENOENT;
706 goto free_stack;
707
708 find_leftmost:
709 /* Find the leftmost non-intermediate node, all intermediate nodes
710 * have exact two children, so this function will never return NULL.
711 */
712 for (node = search_root; node;) {
713 if (node->flags & LPM_TREE_NODE_FLAG_IM) {
714 node = rcu_dereference(node->child[0]);
715 } else {
716 next_node = node;
717 node = rcu_dereference(node->child[0]);
718 if (!node)
719 node = rcu_dereference(next_node->child[1]);
720 }
721 }
722 do_copy:
723 next_key->prefixlen = next_node->prefixlen;
724 memcpy((void *)next_key + offsetof(struct bpf_lpm_trie_key_u8, data),
725 next_node->data, trie->data_size);
726 free_stack:
727 kfree(node_stack);
728 return err;
729 }
730
trie_check_btf(const struct bpf_map * map,const struct btf * btf,const struct btf_type * key_type,const struct btf_type * value_type)731 static int trie_check_btf(const struct bpf_map *map,
732 const struct btf *btf,
733 const struct btf_type *key_type,
734 const struct btf_type *value_type)
735 {
736 /* Keys must have struct bpf_lpm_trie_key_u8 embedded. */
737 return BTF_INFO_KIND(key_type->info) != BTF_KIND_STRUCT ?
738 -EINVAL : 0;
739 }
740
trie_mem_usage(const struct bpf_map * map)741 static u64 trie_mem_usage(const struct bpf_map *map)
742 {
743 struct lpm_trie *trie = container_of(map, struct lpm_trie, map);
744 u64 elem_size;
745
746 elem_size = sizeof(struct lpm_trie_node) + trie->data_size +
747 trie->map.value_size;
748 return elem_size * READ_ONCE(trie->n_entries);
749 }
750
751 BTF_ID_LIST_SINGLE(trie_map_btf_ids, struct, lpm_trie)
752 const struct bpf_map_ops trie_map_ops = {
753 .map_meta_equal = bpf_map_meta_equal,
754 .map_alloc = trie_alloc,
755 .map_free = trie_free,
756 .map_get_next_key = trie_get_next_key,
757 .map_lookup_elem = trie_lookup_elem,
758 .map_update_elem = trie_update_elem,
759 .map_delete_elem = trie_delete_elem,
760 .map_lookup_batch = generic_map_lookup_batch,
761 .map_update_batch = generic_map_update_batch,
762 .map_delete_batch = generic_map_delete_batch,
763 .map_check_btf = trie_check_btf,
764 .map_mem_usage = trie_mem_usage,
765 .map_btf_id = &trie_map_btf_ids[0],
766 };
767