Loading...
1/*
2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
6 *
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
9 *
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
12 *
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
14 *
15 * This work is based on the LPC-trie which is originally described in:
16 *
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
20 *
21 *
22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
24 *
25 *
26 * Code from fib_hash has been reused which includes the following header:
27 *
28 *
29 * INET An implementation of the TCP/IP protocol suite for the LINUX
30 * operating system. INET is implemented using the BSD Socket
31 * interface as the means of communication with the user level.
32 *
33 * IPv4 FIB: lookup engine and maintenance routines.
34 *
35 *
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
37 *
38 * This program is free software; you can redistribute it and/or
39 * modify it under the terms of the GNU General Public License
40 * as published by the Free Software Foundation; either version
41 * 2 of the License, or (at your option) any later version.
42 *
43 * Substantial contributions to this work comes from:
44 *
45 * David S. Miller, <davem@davemloft.net>
46 * Stephen Hemminger <shemminger@osdl.org>
47 * Paul E. McKenney <paulmck@us.ibm.com>
48 * Patrick McHardy <kaber@trash.net>
49 */
50
51#define VERSION "0.409"
52
53#include <linux/cache.h>
54#include <linux/uaccess.h>
55#include <linux/bitops.h>
56#include <linux/types.h>
57#include <linux/kernel.h>
58#include <linux/mm.h>
59#include <linux/string.h>
60#include <linux/socket.h>
61#include <linux/sockios.h>
62#include <linux/errno.h>
63#include <linux/in.h>
64#include <linux/inet.h>
65#include <linux/inetdevice.h>
66#include <linux/netdevice.h>
67#include <linux/if_arp.h>
68#include <linux/proc_fs.h>
69#include <linux/rcupdate.h>
70#include <linux/skbuff.h>
71#include <linux/netlink.h>
72#include <linux/init.h>
73#include <linux/list.h>
74#include <linux/slab.h>
75#include <linux/export.h>
76#include <linux/vmalloc.h>
77#include <linux/notifier.h>
78#include <net/net_namespace.h>
79#include <net/ip.h>
80#include <net/protocol.h>
81#include <net/route.h>
82#include <net/tcp.h>
83#include <net/sock.h>
84#include <net/ip_fib.h>
85#include <net/fib_notifier.h>
86#include <trace/events/fib.h>
87#include "fib_lookup.h"
88
89static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
90 enum fib_event_type event_type, u32 dst,
91 int dst_len, struct fib_alias *fa)
92{
93 struct fib_entry_notifier_info info = {
94 .dst = dst,
95 .dst_len = dst_len,
96 .fi = fa->fa_info,
97 .tos = fa->fa_tos,
98 .type = fa->fa_type,
99 .tb_id = fa->tb_id,
100 };
101 return call_fib4_notifier(nb, net, event_type, &info.info);
102}
103
104static int call_fib_entry_notifiers(struct net *net,
105 enum fib_event_type event_type, u32 dst,
106 int dst_len, struct fib_alias *fa,
107 struct netlink_ext_ack *extack)
108{
109 struct fib_entry_notifier_info info = {
110 .info.extack = extack,
111 .dst = dst,
112 .dst_len = dst_len,
113 .fi = fa->fa_info,
114 .tos = fa->fa_tos,
115 .type = fa->fa_type,
116 .tb_id = fa->tb_id,
117 };
118 return call_fib4_notifiers(net, event_type, &info.info);
119}
120
121#define MAX_STAT_DEPTH 32
122
123#define KEYLENGTH (8*sizeof(t_key))
124#define KEY_MAX ((t_key)~0)
125
126typedef unsigned int t_key;
127
128#define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
129#define IS_TNODE(n) ((n)->bits)
130#define IS_LEAF(n) (!(n)->bits)
131
132struct key_vector {
133 t_key key;
134 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
135 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
136 unsigned char slen;
137 union {
138 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
139 struct hlist_head leaf;
140 /* This array is valid if (pos | bits) > 0 (TNODE) */
141 struct key_vector __rcu *tnode[0];
142 };
143};
144
145struct tnode {
146 struct rcu_head rcu;
147 t_key empty_children; /* KEYLENGTH bits needed */
148 t_key full_children; /* KEYLENGTH bits needed */
149 struct key_vector __rcu *parent;
150 struct key_vector kv[1];
151#define tn_bits kv[0].bits
152};
153
154#define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
155#define LEAF_SIZE TNODE_SIZE(1)
156
157#ifdef CONFIG_IP_FIB_TRIE_STATS
158struct trie_use_stats {
159 unsigned int gets;
160 unsigned int backtrack;
161 unsigned int semantic_match_passed;
162 unsigned int semantic_match_miss;
163 unsigned int null_node_hit;
164 unsigned int resize_node_skipped;
165};
166#endif
167
168struct trie_stat {
169 unsigned int totdepth;
170 unsigned int maxdepth;
171 unsigned int tnodes;
172 unsigned int leaves;
173 unsigned int nullpointers;
174 unsigned int prefixes;
175 unsigned int nodesizes[MAX_STAT_DEPTH];
176};
177
178struct trie {
179 struct key_vector kv[1];
180#ifdef CONFIG_IP_FIB_TRIE_STATS
181 struct trie_use_stats __percpu *stats;
182#endif
183};
184
185static struct key_vector *resize(struct trie *t, struct key_vector *tn);
186static size_t tnode_free_size;
187
188/*
189 * synchronize_rcu after call_rcu for that many pages; it should be especially
190 * useful before resizing the root node with PREEMPT_NONE configs; the value was
191 * obtained experimentally, aiming to avoid visible slowdown.
192 */
193static const int sync_pages = 128;
194
195static struct kmem_cache *fn_alias_kmem __ro_after_init;
196static struct kmem_cache *trie_leaf_kmem __ro_after_init;
197
198static inline struct tnode *tn_info(struct key_vector *kv)
199{
200 return container_of(kv, struct tnode, kv[0]);
201}
202
203/* caller must hold RTNL */
204#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
205#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
206
207/* caller must hold RCU read lock or RTNL */
208#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
209#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
210
211/* wrapper for rcu_assign_pointer */
212static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
213{
214 if (n)
215 rcu_assign_pointer(tn_info(n)->parent, tp);
216}
217
218#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
219
220/* This provides us with the number of children in this node, in the case of a
221 * leaf this will return 0 meaning none of the children are accessible.
222 */
223static inline unsigned long child_length(const struct key_vector *tn)
224{
225 return (1ul << tn->bits) & ~(1ul);
226}
227
228#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
229
230static inline unsigned long get_index(t_key key, struct key_vector *kv)
231{
232 unsigned long index = key ^ kv->key;
233
234 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
235 return 0;
236
237 return index >> kv->pos;
238}
239
240/* To understand this stuff, an understanding of keys and all their bits is
241 * necessary. Every node in the trie has a key associated with it, but not
242 * all of the bits in that key are significant.
243 *
244 * Consider a node 'n' and its parent 'tp'.
245 *
246 * If n is a leaf, every bit in its key is significant. Its presence is
247 * necessitated by path compression, since during a tree traversal (when
248 * searching for a leaf - unless we are doing an insertion) we will completely
249 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
250 * a potentially successful search, that we have indeed been walking the
251 * correct key path.
252 *
253 * Note that we can never "miss" the correct key in the tree if present by
254 * following the wrong path. Path compression ensures that segments of the key
255 * that are the same for all keys with a given prefix are skipped, but the
256 * skipped part *is* identical for each node in the subtrie below the skipped
257 * bit! trie_insert() in this implementation takes care of that.
258 *
259 * if n is an internal node - a 'tnode' here, the various parts of its key
260 * have many different meanings.
261 *
262 * Example:
263 * _________________________________________________________________
264 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
265 * -----------------------------------------------------------------
266 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
267 *
268 * _________________________________________________________________
269 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
270 * -----------------------------------------------------------------
271 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
272 *
273 * tp->pos = 22
274 * tp->bits = 3
275 * n->pos = 13
276 * n->bits = 4
277 *
278 * First, let's just ignore the bits that come before the parent tp, that is
279 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
280 * point we do not use them for anything.
281 *
282 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
283 * index into the parent's child array. That is, they will be used to find
284 * 'n' among tp's children.
285 *
286 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
287 * for the node n.
288 *
289 * All the bits we have seen so far are significant to the node n. The rest
290 * of the bits are really not needed or indeed known in n->key.
291 *
292 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
293 * n's child array, and will of course be different for each child.
294 *
295 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
296 * at this point.
297 */
298
299static const int halve_threshold = 25;
300static const int inflate_threshold = 50;
301static const int halve_threshold_root = 15;
302static const int inflate_threshold_root = 30;
303
304static void __alias_free_mem(struct rcu_head *head)
305{
306 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
307 kmem_cache_free(fn_alias_kmem, fa);
308}
309
310static inline void alias_free_mem_rcu(struct fib_alias *fa)
311{
312 call_rcu(&fa->rcu, __alias_free_mem);
313}
314
315#define TNODE_KMALLOC_MAX \
316 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
317#define TNODE_VMALLOC_MAX \
318 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
319
320static void __node_free_rcu(struct rcu_head *head)
321{
322 struct tnode *n = container_of(head, struct tnode, rcu);
323
324 if (!n->tn_bits)
325 kmem_cache_free(trie_leaf_kmem, n);
326 else
327 kvfree(n);
328}
329
330#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
331
332static struct tnode *tnode_alloc(int bits)
333{
334 size_t size;
335
336 /* verify bits is within bounds */
337 if (bits > TNODE_VMALLOC_MAX)
338 return NULL;
339
340 /* determine size and verify it is non-zero and didn't overflow */
341 size = TNODE_SIZE(1ul << bits);
342
343 if (size <= PAGE_SIZE)
344 return kzalloc(size, GFP_KERNEL);
345 else
346 return vzalloc(size);
347}
348
349static inline void empty_child_inc(struct key_vector *n)
350{
351 ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
352}
353
354static inline void empty_child_dec(struct key_vector *n)
355{
356 tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
357}
358
359static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
360{
361 struct key_vector *l;
362 struct tnode *kv;
363
364 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
365 if (!kv)
366 return NULL;
367
368 /* initialize key vector */
369 l = kv->kv;
370 l->key = key;
371 l->pos = 0;
372 l->bits = 0;
373 l->slen = fa->fa_slen;
374
375 /* link leaf to fib alias */
376 INIT_HLIST_HEAD(&l->leaf);
377 hlist_add_head(&fa->fa_list, &l->leaf);
378
379 return l;
380}
381
382static struct key_vector *tnode_new(t_key key, int pos, int bits)
383{
384 unsigned int shift = pos + bits;
385 struct key_vector *tn;
386 struct tnode *tnode;
387
388 /* verify bits and pos their msb bits clear and values are valid */
389 BUG_ON(!bits || (shift > KEYLENGTH));
390
391 tnode = tnode_alloc(bits);
392 if (!tnode)
393 return NULL;
394
395 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
396 sizeof(struct key_vector *) << bits);
397
398 if (bits == KEYLENGTH)
399 tnode->full_children = 1;
400 else
401 tnode->empty_children = 1ul << bits;
402
403 tn = tnode->kv;
404 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
405 tn->pos = pos;
406 tn->bits = bits;
407 tn->slen = pos;
408
409 return tn;
410}
411
412/* Check whether a tnode 'n' is "full", i.e. it is an internal node
413 * and no bits are skipped. See discussion in dyntree paper p. 6
414 */
415static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
416{
417 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
418}
419
420/* Add a child at position i overwriting the old value.
421 * Update the value of full_children and empty_children.
422 */
423static void put_child(struct key_vector *tn, unsigned long i,
424 struct key_vector *n)
425{
426 struct key_vector *chi = get_child(tn, i);
427 int isfull, wasfull;
428
429 BUG_ON(i >= child_length(tn));
430
431 /* update emptyChildren, overflow into fullChildren */
432 if (!n && chi)
433 empty_child_inc(tn);
434 if (n && !chi)
435 empty_child_dec(tn);
436
437 /* update fullChildren */
438 wasfull = tnode_full(tn, chi);
439 isfull = tnode_full(tn, n);
440
441 if (wasfull && !isfull)
442 tn_info(tn)->full_children--;
443 else if (!wasfull && isfull)
444 tn_info(tn)->full_children++;
445
446 if (n && (tn->slen < n->slen))
447 tn->slen = n->slen;
448
449 rcu_assign_pointer(tn->tnode[i], n);
450}
451
452static void update_children(struct key_vector *tn)
453{
454 unsigned long i;
455
456 /* update all of the child parent pointers */
457 for (i = child_length(tn); i;) {
458 struct key_vector *inode = get_child(tn, --i);
459
460 if (!inode)
461 continue;
462
463 /* Either update the children of a tnode that
464 * already belongs to us or update the child
465 * to point to ourselves.
466 */
467 if (node_parent(inode) == tn)
468 update_children(inode);
469 else
470 node_set_parent(inode, tn);
471 }
472}
473
474static inline void put_child_root(struct key_vector *tp, t_key key,
475 struct key_vector *n)
476{
477 if (IS_TRIE(tp))
478 rcu_assign_pointer(tp->tnode[0], n);
479 else
480 put_child(tp, get_index(key, tp), n);
481}
482
483static inline void tnode_free_init(struct key_vector *tn)
484{
485 tn_info(tn)->rcu.next = NULL;
486}
487
488static inline void tnode_free_append(struct key_vector *tn,
489 struct key_vector *n)
490{
491 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
492 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
493}
494
495static void tnode_free(struct key_vector *tn)
496{
497 struct callback_head *head = &tn_info(tn)->rcu;
498
499 while (head) {
500 head = head->next;
501 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
502 node_free(tn);
503
504 tn = container_of(head, struct tnode, rcu)->kv;
505 }
506
507 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
508 tnode_free_size = 0;
509 synchronize_rcu();
510 }
511}
512
513static struct key_vector *replace(struct trie *t,
514 struct key_vector *oldtnode,
515 struct key_vector *tn)
516{
517 struct key_vector *tp = node_parent(oldtnode);
518 unsigned long i;
519
520 /* setup the parent pointer out of and back into this node */
521 NODE_INIT_PARENT(tn, tp);
522 put_child_root(tp, tn->key, tn);
523
524 /* update all of the child parent pointers */
525 update_children(tn);
526
527 /* all pointers should be clean so we are done */
528 tnode_free(oldtnode);
529
530 /* resize children now that oldtnode is freed */
531 for (i = child_length(tn); i;) {
532 struct key_vector *inode = get_child(tn, --i);
533
534 /* resize child node */
535 if (tnode_full(tn, inode))
536 tn = resize(t, inode);
537 }
538
539 return tp;
540}
541
542static struct key_vector *inflate(struct trie *t,
543 struct key_vector *oldtnode)
544{
545 struct key_vector *tn;
546 unsigned long i;
547 t_key m;
548
549 pr_debug("In inflate\n");
550
551 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
552 if (!tn)
553 goto notnode;
554
555 /* prepare oldtnode to be freed */
556 tnode_free_init(oldtnode);
557
558 /* Assemble all of the pointers in our cluster, in this case that
559 * represents all of the pointers out of our allocated nodes that
560 * point to existing tnodes and the links between our allocated
561 * nodes.
562 */
563 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
564 struct key_vector *inode = get_child(oldtnode, --i);
565 struct key_vector *node0, *node1;
566 unsigned long j, k;
567
568 /* An empty child */
569 if (!inode)
570 continue;
571
572 /* A leaf or an internal node with skipped bits */
573 if (!tnode_full(oldtnode, inode)) {
574 put_child(tn, get_index(inode->key, tn), inode);
575 continue;
576 }
577
578 /* drop the node in the old tnode free list */
579 tnode_free_append(oldtnode, inode);
580
581 /* An internal node with two children */
582 if (inode->bits == 1) {
583 put_child(tn, 2 * i + 1, get_child(inode, 1));
584 put_child(tn, 2 * i, get_child(inode, 0));
585 continue;
586 }
587
588 /* We will replace this node 'inode' with two new
589 * ones, 'node0' and 'node1', each with half of the
590 * original children. The two new nodes will have
591 * a position one bit further down the key and this
592 * means that the "significant" part of their keys
593 * (see the discussion near the top of this file)
594 * will differ by one bit, which will be "0" in
595 * node0's key and "1" in node1's key. Since we are
596 * moving the key position by one step, the bit that
597 * we are moving away from - the bit at position
598 * (tn->pos) - is the one that will differ between
599 * node0 and node1. So... we synthesize that bit in the
600 * two new keys.
601 */
602 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
603 if (!node1)
604 goto nomem;
605 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
606
607 tnode_free_append(tn, node1);
608 if (!node0)
609 goto nomem;
610 tnode_free_append(tn, node0);
611
612 /* populate child pointers in new nodes */
613 for (k = child_length(inode), j = k / 2; j;) {
614 put_child(node1, --j, get_child(inode, --k));
615 put_child(node0, j, get_child(inode, j));
616 put_child(node1, --j, get_child(inode, --k));
617 put_child(node0, j, get_child(inode, j));
618 }
619
620 /* link new nodes to parent */
621 NODE_INIT_PARENT(node1, tn);
622 NODE_INIT_PARENT(node0, tn);
623
624 /* link parent to nodes */
625 put_child(tn, 2 * i + 1, node1);
626 put_child(tn, 2 * i, node0);
627 }
628
629 /* setup the parent pointers into and out of this node */
630 return replace(t, oldtnode, tn);
631nomem:
632 /* all pointers should be clean so we are done */
633 tnode_free(tn);
634notnode:
635 return NULL;
636}
637
638static struct key_vector *halve(struct trie *t,
639 struct key_vector *oldtnode)
640{
641 struct key_vector *tn;
642 unsigned long i;
643
644 pr_debug("In halve\n");
645
646 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
647 if (!tn)
648 goto notnode;
649
650 /* prepare oldtnode to be freed */
651 tnode_free_init(oldtnode);
652
653 /* Assemble all of the pointers in our cluster, in this case that
654 * represents all of the pointers out of our allocated nodes that
655 * point to existing tnodes and the links between our allocated
656 * nodes.
657 */
658 for (i = child_length(oldtnode); i;) {
659 struct key_vector *node1 = get_child(oldtnode, --i);
660 struct key_vector *node0 = get_child(oldtnode, --i);
661 struct key_vector *inode;
662
663 /* At least one of the children is empty */
664 if (!node1 || !node0) {
665 put_child(tn, i / 2, node1 ? : node0);
666 continue;
667 }
668
669 /* Two nonempty children */
670 inode = tnode_new(node0->key, oldtnode->pos, 1);
671 if (!inode)
672 goto nomem;
673 tnode_free_append(tn, inode);
674
675 /* initialize pointers out of node */
676 put_child(inode, 1, node1);
677 put_child(inode, 0, node0);
678 NODE_INIT_PARENT(inode, tn);
679
680 /* link parent to node */
681 put_child(tn, i / 2, inode);
682 }
683
684 /* setup the parent pointers into and out of this node */
685 return replace(t, oldtnode, tn);
686nomem:
687 /* all pointers should be clean so we are done */
688 tnode_free(tn);
689notnode:
690 return NULL;
691}
692
693static struct key_vector *collapse(struct trie *t,
694 struct key_vector *oldtnode)
695{
696 struct key_vector *n, *tp;
697 unsigned long i;
698
699 /* scan the tnode looking for that one child that might still exist */
700 for (n = NULL, i = child_length(oldtnode); !n && i;)
701 n = get_child(oldtnode, --i);
702
703 /* compress one level */
704 tp = node_parent(oldtnode);
705 put_child_root(tp, oldtnode->key, n);
706 node_set_parent(n, tp);
707
708 /* drop dead node */
709 node_free(oldtnode);
710
711 return tp;
712}
713
714static unsigned char update_suffix(struct key_vector *tn)
715{
716 unsigned char slen = tn->pos;
717 unsigned long stride, i;
718 unsigned char slen_max;
719
720 /* only vector 0 can have a suffix length greater than or equal to
721 * tn->pos + tn->bits, the second highest node will have a suffix
722 * length at most of tn->pos + tn->bits - 1
723 */
724 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
725
726 /* search though the list of children looking for nodes that might
727 * have a suffix greater than the one we currently have. This is
728 * why we start with a stride of 2 since a stride of 1 would
729 * represent the nodes with suffix length equal to tn->pos
730 */
731 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
732 struct key_vector *n = get_child(tn, i);
733
734 if (!n || (n->slen <= slen))
735 continue;
736
737 /* update stride and slen based on new value */
738 stride <<= (n->slen - slen);
739 slen = n->slen;
740 i &= ~(stride - 1);
741
742 /* stop searching if we have hit the maximum possible value */
743 if (slen >= slen_max)
744 break;
745 }
746
747 tn->slen = slen;
748
749 return slen;
750}
751
752/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
753 * the Helsinki University of Technology and Matti Tikkanen of Nokia
754 * Telecommunications, page 6:
755 * "A node is doubled if the ratio of non-empty children to all
756 * children in the *doubled* node is at least 'high'."
757 *
758 * 'high' in this instance is the variable 'inflate_threshold'. It
759 * is expressed as a percentage, so we multiply it with
760 * child_length() and instead of multiplying by 2 (since the
761 * child array will be doubled by inflate()) and multiplying
762 * the left-hand side by 100 (to handle the percentage thing) we
763 * multiply the left-hand side by 50.
764 *
765 * The left-hand side may look a bit weird: child_length(tn)
766 * - tn->empty_children is of course the number of non-null children
767 * in the current node. tn->full_children is the number of "full"
768 * children, that is non-null tnodes with a skip value of 0.
769 * All of those will be doubled in the resulting inflated tnode, so
770 * we just count them one extra time here.
771 *
772 * A clearer way to write this would be:
773 *
774 * to_be_doubled = tn->full_children;
775 * not_to_be_doubled = child_length(tn) - tn->empty_children -
776 * tn->full_children;
777 *
778 * new_child_length = child_length(tn) * 2;
779 *
780 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
781 * new_child_length;
782 * if (new_fill_factor >= inflate_threshold)
783 *
784 * ...and so on, tho it would mess up the while () loop.
785 *
786 * anyway,
787 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
788 * inflate_threshold
789 *
790 * avoid a division:
791 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
792 * inflate_threshold * new_child_length
793 *
794 * expand not_to_be_doubled and to_be_doubled, and shorten:
795 * 100 * (child_length(tn) - tn->empty_children +
796 * tn->full_children) >= inflate_threshold * new_child_length
797 *
798 * expand new_child_length:
799 * 100 * (child_length(tn) - tn->empty_children +
800 * tn->full_children) >=
801 * inflate_threshold * child_length(tn) * 2
802 *
803 * shorten again:
804 * 50 * (tn->full_children + child_length(tn) -
805 * tn->empty_children) >= inflate_threshold *
806 * child_length(tn)
807 *
808 */
809static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
810{
811 unsigned long used = child_length(tn);
812 unsigned long threshold = used;
813
814 /* Keep root node larger */
815 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
816 used -= tn_info(tn)->empty_children;
817 used += tn_info(tn)->full_children;
818
819 /* if bits == KEYLENGTH then pos = 0, and will fail below */
820
821 return (used > 1) && tn->pos && ((50 * used) >= threshold);
822}
823
824static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
825{
826 unsigned long used = child_length(tn);
827 unsigned long threshold = used;
828
829 /* Keep root node larger */
830 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
831 used -= tn_info(tn)->empty_children;
832
833 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
834
835 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
836}
837
838static inline bool should_collapse(struct key_vector *tn)
839{
840 unsigned long used = child_length(tn);
841
842 used -= tn_info(tn)->empty_children;
843
844 /* account for bits == KEYLENGTH case */
845 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
846 used -= KEY_MAX;
847
848 /* One child or none, time to drop us from the trie */
849 return used < 2;
850}
851
852#define MAX_WORK 10
853static struct key_vector *resize(struct trie *t, struct key_vector *tn)
854{
855#ifdef CONFIG_IP_FIB_TRIE_STATS
856 struct trie_use_stats __percpu *stats = t->stats;
857#endif
858 struct key_vector *tp = node_parent(tn);
859 unsigned long cindex = get_index(tn->key, tp);
860 int max_work = MAX_WORK;
861
862 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
863 tn, inflate_threshold, halve_threshold);
864
865 /* track the tnode via the pointer from the parent instead of
866 * doing it ourselves. This way we can let RCU fully do its
867 * thing without us interfering
868 */
869 BUG_ON(tn != get_child(tp, cindex));
870
871 /* Double as long as the resulting node has a number of
872 * nonempty nodes that are above the threshold.
873 */
874 while (should_inflate(tp, tn) && max_work) {
875 tp = inflate(t, tn);
876 if (!tp) {
877#ifdef CONFIG_IP_FIB_TRIE_STATS
878 this_cpu_inc(stats->resize_node_skipped);
879#endif
880 break;
881 }
882
883 max_work--;
884 tn = get_child(tp, cindex);
885 }
886
887 /* update parent in case inflate failed */
888 tp = node_parent(tn);
889
890 /* Return if at least one inflate is run */
891 if (max_work != MAX_WORK)
892 return tp;
893
894 /* Halve as long as the number of empty children in this
895 * node is above threshold.
896 */
897 while (should_halve(tp, tn) && max_work) {
898 tp = halve(t, tn);
899 if (!tp) {
900#ifdef CONFIG_IP_FIB_TRIE_STATS
901 this_cpu_inc(stats->resize_node_skipped);
902#endif
903 break;
904 }
905
906 max_work--;
907 tn = get_child(tp, cindex);
908 }
909
910 /* Only one child remains */
911 if (should_collapse(tn))
912 return collapse(t, tn);
913
914 /* update parent in case halve failed */
915 return node_parent(tn);
916}
917
918static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
919{
920 unsigned char node_slen = tn->slen;
921
922 while ((node_slen > tn->pos) && (node_slen > slen)) {
923 slen = update_suffix(tn);
924 if (node_slen == slen)
925 break;
926
927 tn = node_parent(tn);
928 node_slen = tn->slen;
929 }
930}
931
932static void node_push_suffix(struct key_vector *tn, unsigned char slen)
933{
934 while (tn->slen < slen) {
935 tn->slen = slen;
936 tn = node_parent(tn);
937 }
938}
939
940/* rcu_read_lock needs to be hold by caller from readside */
941static struct key_vector *fib_find_node(struct trie *t,
942 struct key_vector **tp, u32 key)
943{
944 struct key_vector *pn, *n = t->kv;
945 unsigned long index = 0;
946
947 do {
948 pn = n;
949 n = get_child_rcu(n, index);
950
951 if (!n)
952 break;
953
954 index = get_cindex(key, n);
955
956 /* This bit of code is a bit tricky but it combines multiple
957 * checks into a single check. The prefix consists of the
958 * prefix plus zeros for the bits in the cindex. The index
959 * is the difference between the key and this value. From
960 * this we can actually derive several pieces of data.
961 * if (index >= (1ul << bits))
962 * we have a mismatch in skip bits and failed
963 * else
964 * we know the value is cindex
965 *
966 * This check is safe even if bits == KEYLENGTH due to the
967 * fact that we can only allocate a node with 32 bits if a
968 * long is greater than 32 bits.
969 */
970 if (index >= (1ul << n->bits)) {
971 n = NULL;
972 break;
973 }
974
975 /* keep searching until we find a perfect match leaf or NULL */
976 } while (IS_TNODE(n));
977
978 *tp = pn;
979
980 return n;
981}
982
983/* Return the first fib alias matching TOS with
984 * priority less than or equal to PRIO.
985 */
986static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
987 u8 tos, u32 prio, u32 tb_id)
988{
989 struct fib_alias *fa;
990
991 if (!fah)
992 return NULL;
993
994 hlist_for_each_entry(fa, fah, fa_list) {
995 if (fa->fa_slen < slen)
996 continue;
997 if (fa->fa_slen != slen)
998 break;
999 if (fa->tb_id > tb_id)
1000 continue;
1001 if (fa->tb_id != tb_id)
1002 break;
1003 if (fa->fa_tos > tos)
1004 continue;
1005 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1006 return fa;
1007 }
1008
1009 return NULL;
1010}
1011
1012static void trie_rebalance(struct trie *t, struct key_vector *tn)
1013{
1014 while (!IS_TRIE(tn))
1015 tn = resize(t, tn);
1016}
1017
1018static int fib_insert_node(struct trie *t, struct key_vector *tp,
1019 struct fib_alias *new, t_key key)
1020{
1021 struct key_vector *n, *l;
1022
1023 l = leaf_new(key, new);
1024 if (!l)
1025 goto noleaf;
1026
1027 /* retrieve child from parent node */
1028 n = get_child(tp, get_index(key, tp));
1029
1030 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1031 *
1032 * Add a new tnode here
1033 * first tnode need some special handling
1034 * leaves us in position for handling as case 3
1035 */
1036 if (n) {
1037 struct key_vector *tn;
1038
1039 tn = tnode_new(key, __fls(key ^ n->key), 1);
1040 if (!tn)
1041 goto notnode;
1042
1043 /* initialize routes out of node */
1044 NODE_INIT_PARENT(tn, tp);
1045 put_child(tn, get_index(key, tn) ^ 1, n);
1046
1047 /* start adding routes into the node */
1048 put_child_root(tp, key, tn);
1049 node_set_parent(n, tn);
1050
1051 /* parent now has a NULL spot where the leaf can go */
1052 tp = tn;
1053 }
1054
1055 /* Case 3: n is NULL, and will just insert a new leaf */
1056 node_push_suffix(tp, new->fa_slen);
1057 NODE_INIT_PARENT(l, tp);
1058 put_child_root(tp, key, l);
1059 trie_rebalance(t, tp);
1060
1061 return 0;
1062notnode:
1063 node_free(l);
1064noleaf:
1065 return -ENOMEM;
1066}
1067
1068/* fib notifier for ADD is sent before calling fib_insert_alias with
1069 * the expectation that the only possible failure ENOMEM
1070 */
1071static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1072 struct key_vector *l, struct fib_alias *new,
1073 struct fib_alias *fa, t_key key)
1074{
1075 if (!l)
1076 return fib_insert_node(t, tp, new, key);
1077
1078 if (fa) {
1079 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1080 } else {
1081 struct fib_alias *last;
1082
1083 hlist_for_each_entry(last, &l->leaf, fa_list) {
1084 if (new->fa_slen < last->fa_slen)
1085 break;
1086 if ((new->fa_slen == last->fa_slen) &&
1087 (new->tb_id > last->tb_id))
1088 break;
1089 fa = last;
1090 }
1091
1092 if (fa)
1093 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1094 else
1095 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1096 }
1097
1098 /* if we added to the tail node then we need to update slen */
1099 if (l->slen < new->fa_slen) {
1100 l->slen = new->fa_slen;
1101 node_push_suffix(tp, new->fa_slen);
1102 }
1103
1104 return 0;
1105}
1106
1107static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1108{
1109 if (plen > KEYLENGTH) {
1110 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1111 return false;
1112 }
1113
1114 if ((plen < KEYLENGTH) && (key << plen)) {
1115 NL_SET_ERR_MSG(extack,
1116 "Invalid prefix for given prefix length");
1117 return false;
1118 }
1119
1120 return true;
1121}
1122
1123/* Caller must hold RTNL. */
1124int fib_table_insert(struct net *net, struct fib_table *tb,
1125 struct fib_config *cfg, struct netlink_ext_ack *extack)
1126{
1127 enum fib_event_type event = FIB_EVENT_ENTRY_ADD;
1128 struct trie *t = (struct trie *)tb->tb_data;
1129 struct fib_alias *fa, *new_fa;
1130 struct key_vector *l, *tp;
1131 u16 nlflags = NLM_F_EXCL;
1132 struct fib_info *fi;
1133 u8 plen = cfg->fc_dst_len;
1134 u8 slen = KEYLENGTH - plen;
1135 u8 tos = cfg->fc_tos;
1136 u32 key;
1137 int err;
1138
1139 key = ntohl(cfg->fc_dst);
1140
1141 if (!fib_valid_key_len(key, plen, extack))
1142 return -EINVAL;
1143
1144 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1145
1146 fi = fib_create_info(cfg, extack);
1147 if (IS_ERR(fi)) {
1148 err = PTR_ERR(fi);
1149 goto err;
1150 }
1151
1152 l = fib_find_node(t, &tp, key);
1153 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1154 tb->tb_id) : NULL;
1155
1156 /* Now fa, if non-NULL, points to the first fib alias
1157 * with the same keys [prefix,tos,priority], if such key already
1158 * exists or to the node before which we will insert new one.
1159 *
1160 * If fa is NULL, we will need to allocate a new one and
1161 * insert to the tail of the section matching the suffix length
1162 * of the new alias.
1163 */
1164
1165 if (fa && fa->fa_tos == tos &&
1166 fa->fa_info->fib_priority == fi->fib_priority) {
1167 struct fib_alias *fa_first, *fa_match;
1168
1169 err = -EEXIST;
1170 if (cfg->fc_nlflags & NLM_F_EXCL)
1171 goto out;
1172
1173 nlflags &= ~NLM_F_EXCL;
1174
1175 /* We have 2 goals:
1176 * 1. Find exact match for type, scope, fib_info to avoid
1177 * duplicate routes
1178 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1179 */
1180 fa_match = NULL;
1181 fa_first = fa;
1182 hlist_for_each_entry_from(fa, fa_list) {
1183 if ((fa->fa_slen != slen) ||
1184 (fa->tb_id != tb->tb_id) ||
1185 (fa->fa_tos != tos))
1186 break;
1187 if (fa->fa_info->fib_priority != fi->fib_priority)
1188 break;
1189 if (fa->fa_type == cfg->fc_type &&
1190 fa->fa_info == fi) {
1191 fa_match = fa;
1192 break;
1193 }
1194 }
1195
1196 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1197 struct fib_info *fi_drop;
1198 u8 state;
1199
1200 nlflags |= NLM_F_REPLACE;
1201 fa = fa_first;
1202 if (fa_match) {
1203 if (fa == fa_match)
1204 err = 0;
1205 goto out;
1206 }
1207 err = -ENOBUFS;
1208 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1209 if (!new_fa)
1210 goto out;
1211
1212 fi_drop = fa->fa_info;
1213 new_fa->fa_tos = fa->fa_tos;
1214 new_fa->fa_info = fi;
1215 new_fa->fa_type = cfg->fc_type;
1216 state = fa->fa_state;
1217 new_fa->fa_state = state & ~FA_S_ACCESSED;
1218 new_fa->fa_slen = fa->fa_slen;
1219 new_fa->tb_id = tb->tb_id;
1220 new_fa->fa_default = -1;
1221
1222 err = call_fib_entry_notifiers(net,
1223 FIB_EVENT_ENTRY_REPLACE,
1224 key, plen, new_fa,
1225 extack);
1226 if (err)
1227 goto out_free_new_fa;
1228
1229 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1230 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1231
1232 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1233
1234 alias_free_mem_rcu(fa);
1235
1236 fib_release_info(fi_drop);
1237 if (state & FA_S_ACCESSED)
1238 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1239
1240 goto succeeded;
1241 }
1242 /* Error if we find a perfect match which
1243 * uses the same scope, type, and nexthop
1244 * information.
1245 */
1246 if (fa_match)
1247 goto out;
1248
1249 if (cfg->fc_nlflags & NLM_F_APPEND) {
1250 event = FIB_EVENT_ENTRY_APPEND;
1251 nlflags |= NLM_F_APPEND;
1252 } else {
1253 fa = fa_first;
1254 }
1255 }
1256 err = -ENOENT;
1257 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1258 goto out;
1259
1260 nlflags |= NLM_F_CREATE;
1261 err = -ENOBUFS;
1262 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1263 if (!new_fa)
1264 goto out;
1265
1266 new_fa->fa_info = fi;
1267 new_fa->fa_tos = tos;
1268 new_fa->fa_type = cfg->fc_type;
1269 new_fa->fa_state = 0;
1270 new_fa->fa_slen = slen;
1271 new_fa->tb_id = tb->tb_id;
1272 new_fa->fa_default = -1;
1273
1274 err = call_fib_entry_notifiers(net, event, key, plen, new_fa, extack);
1275 if (err)
1276 goto out_free_new_fa;
1277
1278 /* Insert new entry to the list. */
1279 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1280 if (err)
1281 goto out_fib_notif;
1282
1283 if (!plen)
1284 tb->tb_num_default++;
1285
1286 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1287 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1288 &cfg->fc_nlinfo, nlflags);
1289succeeded:
1290 return 0;
1291
1292out_fib_notif:
1293 /* notifier was sent that entry would be added to trie, but
1294 * the add failed and need to recover. Only failure for
1295 * fib_insert_alias is ENOMEM.
1296 */
1297 NL_SET_ERR_MSG(extack, "Failed to insert route into trie");
1298 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key,
1299 plen, new_fa, NULL);
1300out_free_new_fa:
1301 kmem_cache_free(fn_alias_kmem, new_fa);
1302out:
1303 fib_release_info(fi);
1304err:
1305 return err;
1306}
1307
1308static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1309{
1310 t_key prefix = n->key;
1311
1312 return (key ^ prefix) & (prefix | -prefix);
1313}
1314
1315/* should be called with rcu_read_lock */
1316int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1317 struct fib_result *res, int fib_flags)
1318{
1319 struct trie *t = (struct trie *) tb->tb_data;
1320#ifdef CONFIG_IP_FIB_TRIE_STATS
1321 struct trie_use_stats __percpu *stats = t->stats;
1322#endif
1323 const t_key key = ntohl(flp->daddr);
1324 struct key_vector *n, *pn;
1325 struct fib_alias *fa;
1326 unsigned long index;
1327 t_key cindex;
1328
1329 trace_fib_table_lookup(tb->tb_id, flp);
1330
1331 pn = t->kv;
1332 cindex = 0;
1333
1334 n = get_child_rcu(pn, cindex);
1335 if (!n)
1336 return -EAGAIN;
1337
1338#ifdef CONFIG_IP_FIB_TRIE_STATS
1339 this_cpu_inc(stats->gets);
1340#endif
1341
1342 /* Step 1: Travel to the longest prefix match in the trie */
1343 for (;;) {
1344 index = get_cindex(key, n);
1345
1346 /* This bit of code is a bit tricky but it combines multiple
1347 * checks into a single check. The prefix consists of the
1348 * prefix plus zeros for the "bits" in the prefix. The index
1349 * is the difference between the key and this value. From
1350 * this we can actually derive several pieces of data.
1351 * if (index >= (1ul << bits))
1352 * we have a mismatch in skip bits and failed
1353 * else
1354 * we know the value is cindex
1355 *
1356 * This check is safe even if bits == KEYLENGTH due to the
1357 * fact that we can only allocate a node with 32 bits if a
1358 * long is greater than 32 bits.
1359 */
1360 if (index >= (1ul << n->bits))
1361 break;
1362
1363 /* we have found a leaf. Prefixes have already been compared */
1364 if (IS_LEAF(n))
1365 goto found;
1366
1367 /* only record pn and cindex if we are going to be chopping
1368 * bits later. Otherwise we are just wasting cycles.
1369 */
1370 if (n->slen > n->pos) {
1371 pn = n;
1372 cindex = index;
1373 }
1374
1375 n = get_child_rcu(n, index);
1376 if (unlikely(!n))
1377 goto backtrace;
1378 }
1379
1380 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1381 for (;;) {
1382 /* record the pointer where our next node pointer is stored */
1383 struct key_vector __rcu **cptr = n->tnode;
1384
1385 /* This test verifies that none of the bits that differ
1386 * between the key and the prefix exist in the region of
1387 * the lsb and higher in the prefix.
1388 */
1389 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1390 goto backtrace;
1391
1392 /* exit out and process leaf */
1393 if (unlikely(IS_LEAF(n)))
1394 break;
1395
1396 /* Don't bother recording parent info. Since we are in
1397 * prefix match mode we will have to come back to wherever
1398 * we started this traversal anyway
1399 */
1400
1401 while ((n = rcu_dereference(*cptr)) == NULL) {
1402backtrace:
1403#ifdef CONFIG_IP_FIB_TRIE_STATS
1404 if (!n)
1405 this_cpu_inc(stats->null_node_hit);
1406#endif
1407 /* If we are at cindex 0 there are no more bits for
1408 * us to strip at this level so we must ascend back
1409 * up one level to see if there are any more bits to
1410 * be stripped there.
1411 */
1412 while (!cindex) {
1413 t_key pkey = pn->key;
1414
1415 /* If we don't have a parent then there is
1416 * nothing for us to do as we do not have any
1417 * further nodes to parse.
1418 */
1419 if (IS_TRIE(pn))
1420 return -EAGAIN;
1421#ifdef CONFIG_IP_FIB_TRIE_STATS
1422 this_cpu_inc(stats->backtrack);
1423#endif
1424 /* Get Child's index */
1425 pn = node_parent_rcu(pn);
1426 cindex = get_index(pkey, pn);
1427 }
1428
1429 /* strip the least significant bit from the cindex */
1430 cindex &= cindex - 1;
1431
1432 /* grab pointer for next child node */
1433 cptr = &pn->tnode[cindex];
1434 }
1435 }
1436
1437found:
1438 /* this line carries forward the xor from earlier in the function */
1439 index = key ^ n->key;
1440
1441 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1442 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1443 struct fib_info *fi = fa->fa_info;
1444 int nhsel, err;
1445
1446 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1447 if (index >= (1ul << fa->fa_slen))
1448 continue;
1449 }
1450 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1451 continue;
1452 if (fi->fib_dead)
1453 continue;
1454 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1455 continue;
1456 fib_alias_accessed(fa);
1457 err = fib_props[fa->fa_type].error;
1458 if (unlikely(err < 0)) {
1459#ifdef CONFIG_IP_FIB_TRIE_STATS
1460 this_cpu_inc(stats->semantic_match_passed);
1461#endif
1462 return err;
1463 }
1464 if (fi->fib_flags & RTNH_F_DEAD)
1465 continue;
1466 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1467 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1468 struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1469
1470 if (nh->nh_flags & RTNH_F_DEAD)
1471 continue;
1472 if (in_dev &&
1473 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1474 nh->nh_flags & RTNH_F_LINKDOWN &&
1475 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1476 continue;
1477 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1478 if (flp->flowi4_oif &&
1479 flp->flowi4_oif != nh->nh_oif)
1480 continue;
1481 }
1482
1483 if (!(fib_flags & FIB_LOOKUP_NOREF))
1484 refcount_inc(&fi->fib_clntref);
1485
1486 res->prefix = htonl(n->key);
1487 res->prefixlen = KEYLENGTH - fa->fa_slen;
1488 res->nh_sel = nhsel;
1489 res->type = fa->fa_type;
1490 res->scope = fi->fib_scope;
1491 res->fi = fi;
1492 res->table = tb;
1493 res->fa_head = &n->leaf;
1494#ifdef CONFIG_IP_FIB_TRIE_STATS
1495 this_cpu_inc(stats->semantic_match_passed);
1496#endif
1497 trace_fib_table_lookup_nh(nh);
1498
1499 return err;
1500 }
1501 }
1502#ifdef CONFIG_IP_FIB_TRIE_STATS
1503 this_cpu_inc(stats->semantic_match_miss);
1504#endif
1505 goto backtrace;
1506}
1507EXPORT_SYMBOL_GPL(fib_table_lookup);
1508
1509static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1510 struct key_vector *l, struct fib_alias *old)
1511{
1512 /* record the location of the previous list_info entry */
1513 struct hlist_node **pprev = old->fa_list.pprev;
1514 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1515
1516 /* remove the fib_alias from the list */
1517 hlist_del_rcu(&old->fa_list);
1518
1519 /* if we emptied the list this leaf will be freed and we can sort
1520 * out parent suffix lengths as a part of trie_rebalance
1521 */
1522 if (hlist_empty(&l->leaf)) {
1523 if (tp->slen == l->slen)
1524 node_pull_suffix(tp, tp->pos);
1525 put_child_root(tp, l->key, NULL);
1526 node_free(l);
1527 trie_rebalance(t, tp);
1528 return;
1529 }
1530
1531 /* only access fa if it is pointing at the last valid hlist_node */
1532 if (*pprev)
1533 return;
1534
1535 /* update the trie with the latest suffix length */
1536 l->slen = fa->fa_slen;
1537 node_pull_suffix(tp, fa->fa_slen);
1538}
1539
1540/* Caller must hold RTNL. */
1541int fib_table_delete(struct net *net, struct fib_table *tb,
1542 struct fib_config *cfg, struct netlink_ext_ack *extack)
1543{
1544 struct trie *t = (struct trie *) tb->tb_data;
1545 struct fib_alias *fa, *fa_to_delete;
1546 struct key_vector *l, *tp;
1547 u8 plen = cfg->fc_dst_len;
1548 u8 slen = KEYLENGTH - plen;
1549 u8 tos = cfg->fc_tos;
1550 u32 key;
1551
1552 key = ntohl(cfg->fc_dst);
1553
1554 if (!fib_valid_key_len(key, plen, extack))
1555 return -EINVAL;
1556
1557 l = fib_find_node(t, &tp, key);
1558 if (!l)
1559 return -ESRCH;
1560
1561 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1562 if (!fa)
1563 return -ESRCH;
1564
1565 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1566
1567 fa_to_delete = NULL;
1568 hlist_for_each_entry_from(fa, fa_list) {
1569 struct fib_info *fi = fa->fa_info;
1570
1571 if ((fa->fa_slen != slen) ||
1572 (fa->tb_id != tb->tb_id) ||
1573 (fa->fa_tos != tos))
1574 break;
1575
1576 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1577 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1578 fa->fa_info->fib_scope == cfg->fc_scope) &&
1579 (!cfg->fc_prefsrc ||
1580 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1581 (!cfg->fc_protocol ||
1582 fi->fib_protocol == cfg->fc_protocol) &&
1583 fib_nh_match(cfg, fi, extack) == 0 &&
1584 fib_metrics_match(cfg, fi)) {
1585 fa_to_delete = fa;
1586 break;
1587 }
1588 }
1589
1590 if (!fa_to_delete)
1591 return -ESRCH;
1592
1593 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1594 fa_to_delete, extack);
1595 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1596 &cfg->fc_nlinfo, 0);
1597
1598 if (!plen)
1599 tb->tb_num_default--;
1600
1601 fib_remove_alias(t, tp, l, fa_to_delete);
1602
1603 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1604 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1605
1606 fib_release_info(fa_to_delete->fa_info);
1607 alias_free_mem_rcu(fa_to_delete);
1608 return 0;
1609}
1610
1611/* Scan for the next leaf starting at the provided key value */
1612static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1613{
1614 struct key_vector *pn, *n = *tn;
1615 unsigned long cindex;
1616
1617 /* this loop is meant to try and find the key in the trie */
1618 do {
1619 /* record parent and next child index */
1620 pn = n;
1621 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1622
1623 if (cindex >> pn->bits)
1624 break;
1625
1626 /* descend into the next child */
1627 n = get_child_rcu(pn, cindex++);
1628 if (!n)
1629 break;
1630
1631 /* guarantee forward progress on the keys */
1632 if (IS_LEAF(n) && (n->key >= key))
1633 goto found;
1634 } while (IS_TNODE(n));
1635
1636 /* this loop will search for the next leaf with a greater key */
1637 while (!IS_TRIE(pn)) {
1638 /* if we exhausted the parent node we will need to climb */
1639 if (cindex >= (1ul << pn->bits)) {
1640 t_key pkey = pn->key;
1641
1642 pn = node_parent_rcu(pn);
1643 cindex = get_index(pkey, pn) + 1;
1644 continue;
1645 }
1646
1647 /* grab the next available node */
1648 n = get_child_rcu(pn, cindex++);
1649 if (!n)
1650 continue;
1651
1652 /* no need to compare keys since we bumped the index */
1653 if (IS_LEAF(n))
1654 goto found;
1655
1656 /* Rescan start scanning in new node */
1657 pn = n;
1658 cindex = 0;
1659 }
1660
1661 *tn = pn;
1662 return NULL; /* Root of trie */
1663found:
1664 /* if we are at the limit for keys just return NULL for the tnode */
1665 *tn = pn;
1666 return n;
1667}
1668
1669static void fib_trie_free(struct fib_table *tb)
1670{
1671 struct trie *t = (struct trie *)tb->tb_data;
1672 struct key_vector *pn = t->kv;
1673 unsigned long cindex = 1;
1674 struct hlist_node *tmp;
1675 struct fib_alias *fa;
1676
1677 /* walk trie in reverse order and free everything */
1678 for (;;) {
1679 struct key_vector *n;
1680
1681 if (!(cindex--)) {
1682 t_key pkey = pn->key;
1683
1684 if (IS_TRIE(pn))
1685 break;
1686
1687 n = pn;
1688 pn = node_parent(pn);
1689
1690 /* drop emptied tnode */
1691 put_child_root(pn, n->key, NULL);
1692 node_free(n);
1693
1694 cindex = get_index(pkey, pn);
1695
1696 continue;
1697 }
1698
1699 /* grab the next available node */
1700 n = get_child(pn, cindex);
1701 if (!n)
1702 continue;
1703
1704 if (IS_TNODE(n)) {
1705 /* record pn and cindex for leaf walking */
1706 pn = n;
1707 cindex = 1ul << n->bits;
1708
1709 continue;
1710 }
1711
1712 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1713 hlist_del_rcu(&fa->fa_list);
1714 alias_free_mem_rcu(fa);
1715 }
1716
1717 put_child_root(pn, n->key, NULL);
1718 node_free(n);
1719 }
1720
1721#ifdef CONFIG_IP_FIB_TRIE_STATS
1722 free_percpu(t->stats);
1723#endif
1724 kfree(tb);
1725}
1726
1727struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1728{
1729 struct trie *ot = (struct trie *)oldtb->tb_data;
1730 struct key_vector *l, *tp = ot->kv;
1731 struct fib_table *local_tb;
1732 struct fib_alias *fa;
1733 struct trie *lt;
1734 t_key key = 0;
1735
1736 if (oldtb->tb_data == oldtb->__data)
1737 return oldtb;
1738
1739 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1740 if (!local_tb)
1741 return NULL;
1742
1743 lt = (struct trie *)local_tb->tb_data;
1744
1745 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1746 struct key_vector *local_l = NULL, *local_tp;
1747
1748 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1749 struct fib_alias *new_fa;
1750
1751 if (local_tb->tb_id != fa->tb_id)
1752 continue;
1753
1754 /* clone fa for new local table */
1755 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1756 if (!new_fa)
1757 goto out;
1758
1759 memcpy(new_fa, fa, sizeof(*fa));
1760
1761 /* insert clone into table */
1762 if (!local_l)
1763 local_l = fib_find_node(lt, &local_tp, l->key);
1764
1765 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1766 NULL, l->key)) {
1767 kmem_cache_free(fn_alias_kmem, new_fa);
1768 goto out;
1769 }
1770 }
1771
1772 /* stop loop if key wrapped back to 0 */
1773 key = l->key + 1;
1774 if (key < l->key)
1775 break;
1776 }
1777
1778 return local_tb;
1779out:
1780 fib_trie_free(local_tb);
1781
1782 return NULL;
1783}
1784
1785/* Caller must hold RTNL */
1786void fib_table_flush_external(struct fib_table *tb)
1787{
1788 struct trie *t = (struct trie *)tb->tb_data;
1789 struct key_vector *pn = t->kv;
1790 unsigned long cindex = 1;
1791 struct hlist_node *tmp;
1792 struct fib_alias *fa;
1793
1794 /* walk trie in reverse order */
1795 for (;;) {
1796 unsigned char slen = 0;
1797 struct key_vector *n;
1798
1799 if (!(cindex--)) {
1800 t_key pkey = pn->key;
1801
1802 /* cannot resize the trie vector */
1803 if (IS_TRIE(pn))
1804 break;
1805
1806 /* update the suffix to address pulled leaves */
1807 if (pn->slen > pn->pos)
1808 update_suffix(pn);
1809
1810 /* resize completed node */
1811 pn = resize(t, pn);
1812 cindex = get_index(pkey, pn);
1813
1814 continue;
1815 }
1816
1817 /* grab the next available node */
1818 n = get_child(pn, cindex);
1819 if (!n)
1820 continue;
1821
1822 if (IS_TNODE(n)) {
1823 /* record pn and cindex for leaf walking */
1824 pn = n;
1825 cindex = 1ul << n->bits;
1826
1827 continue;
1828 }
1829
1830 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1831 /* if alias was cloned to local then we just
1832 * need to remove the local copy from main
1833 */
1834 if (tb->tb_id != fa->tb_id) {
1835 hlist_del_rcu(&fa->fa_list);
1836 alias_free_mem_rcu(fa);
1837 continue;
1838 }
1839
1840 /* record local slen */
1841 slen = fa->fa_slen;
1842 }
1843
1844 /* update leaf slen */
1845 n->slen = slen;
1846
1847 if (hlist_empty(&n->leaf)) {
1848 put_child_root(pn, n->key, NULL);
1849 node_free(n);
1850 }
1851 }
1852}
1853
1854/* Caller must hold RTNL. */
1855int fib_table_flush(struct net *net, struct fib_table *tb)
1856{
1857 struct trie *t = (struct trie *)tb->tb_data;
1858 struct key_vector *pn = t->kv;
1859 unsigned long cindex = 1;
1860 struct hlist_node *tmp;
1861 struct fib_alias *fa;
1862 int found = 0;
1863
1864 /* walk trie in reverse order */
1865 for (;;) {
1866 unsigned char slen = 0;
1867 struct key_vector *n;
1868
1869 if (!(cindex--)) {
1870 t_key pkey = pn->key;
1871
1872 /* cannot resize the trie vector */
1873 if (IS_TRIE(pn))
1874 break;
1875
1876 /* update the suffix to address pulled leaves */
1877 if (pn->slen > pn->pos)
1878 update_suffix(pn);
1879
1880 /* resize completed node */
1881 pn = resize(t, pn);
1882 cindex = get_index(pkey, pn);
1883
1884 continue;
1885 }
1886
1887 /* grab the next available node */
1888 n = get_child(pn, cindex);
1889 if (!n)
1890 continue;
1891
1892 if (IS_TNODE(n)) {
1893 /* record pn and cindex for leaf walking */
1894 pn = n;
1895 cindex = 1ul << n->bits;
1896
1897 continue;
1898 }
1899
1900 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1901 struct fib_info *fi = fa->fa_info;
1902
1903 if (!fi || !(fi->fib_flags & RTNH_F_DEAD) ||
1904 tb->tb_id != fa->tb_id) {
1905 slen = fa->fa_slen;
1906 continue;
1907 }
1908
1909 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1910 n->key,
1911 KEYLENGTH - fa->fa_slen, fa,
1912 NULL);
1913 hlist_del_rcu(&fa->fa_list);
1914 fib_release_info(fa->fa_info);
1915 alias_free_mem_rcu(fa);
1916 found++;
1917 }
1918
1919 /* update leaf slen */
1920 n->slen = slen;
1921
1922 if (hlist_empty(&n->leaf)) {
1923 put_child_root(pn, n->key, NULL);
1924 node_free(n);
1925 }
1926 }
1927
1928 pr_debug("trie_flush found=%d\n", found);
1929 return found;
1930}
1931
1932static void fib_leaf_notify(struct net *net, struct key_vector *l,
1933 struct fib_table *tb, struct notifier_block *nb)
1934{
1935 struct fib_alias *fa;
1936
1937 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1938 struct fib_info *fi = fa->fa_info;
1939
1940 if (!fi)
1941 continue;
1942
1943 /* local and main table can share the same trie,
1944 * so don't notify twice for the same entry.
1945 */
1946 if (tb->tb_id != fa->tb_id)
1947 continue;
1948
1949 call_fib_entry_notifier(nb, net, FIB_EVENT_ENTRY_ADD, l->key,
1950 KEYLENGTH - fa->fa_slen, fa);
1951 }
1952}
1953
1954static void fib_table_notify(struct net *net, struct fib_table *tb,
1955 struct notifier_block *nb)
1956{
1957 struct trie *t = (struct trie *)tb->tb_data;
1958 struct key_vector *l, *tp = t->kv;
1959 t_key key = 0;
1960
1961 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1962 fib_leaf_notify(net, l, tb, nb);
1963
1964 key = l->key + 1;
1965 /* stop in case of wrap around */
1966 if (key < l->key)
1967 break;
1968 }
1969}
1970
1971void fib_notify(struct net *net, struct notifier_block *nb)
1972{
1973 unsigned int h;
1974
1975 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1976 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
1977 struct fib_table *tb;
1978
1979 hlist_for_each_entry_rcu(tb, head, tb_hlist)
1980 fib_table_notify(net, tb, nb);
1981 }
1982}
1983
1984static void __trie_free_rcu(struct rcu_head *head)
1985{
1986 struct fib_table *tb = container_of(head, struct fib_table, rcu);
1987#ifdef CONFIG_IP_FIB_TRIE_STATS
1988 struct trie *t = (struct trie *)tb->tb_data;
1989
1990 if (tb->tb_data == tb->__data)
1991 free_percpu(t->stats);
1992#endif /* CONFIG_IP_FIB_TRIE_STATS */
1993 kfree(tb);
1994}
1995
1996void fib_free_table(struct fib_table *tb)
1997{
1998 call_rcu(&tb->rcu, __trie_free_rcu);
1999}
2000
2001static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2002 struct sk_buff *skb, struct netlink_callback *cb)
2003{
2004 __be32 xkey = htonl(l->key);
2005 struct fib_alias *fa;
2006 int i, s_i;
2007
2008 s_i = cb->args[4];
2009 i = 0;
2010
2011 /* rcu_read_lock is hold by caller */
2012 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2013 int err;
2014
2015 if (i < s_i) {
2016 i++;
2017 continue;
2018 }
2019
2020 if (tb->tb_id != fa->tb_id) {
2021 i++;
2022 continue;
2023 }
2024
2025 err = fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
2026 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2027 tb->tb_id, fa->fa_type,
2028 xkey, KEYLENGTH - fa->fa_slen,
2029 fa->fa_tos, fa->fa_info, NLM_F_MULTI);
2030 if (err < 0) {
2031 cb->args[4] = i;
2032 return err;
2033 }
2034 i++;
2035 }
2036
2037 cb->args[4] = i;
2038 return skb->len;
2039}
2040
2041/* rcu_read_lock needs to be hold by caller from readside */
2042int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2043 struct netlink_callback *cb)
2044{
2045 struct trie *t = (struct trie *)tb->tb_data;
2046 struct key_vector *l, *tp = t->kv;
2047 /* Dump starting at last key.
2048 * Note: 0.0.0.0/0 (ie default) is first key.
2049 */
2050 int count = cb->args[2];
2051 t_key key = cb->args[3];
2052
2053 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2054 int err;
2055
2056 err = fn_trie_dump_leaf(l, tb, skb, cb);
2057 if (err < 0) {
2058 cb->args[3] = key;
2059 cb->args[2] = count;
2060 return err;
2061 }
2062
2063 ++count;
2064 key = l->key + 1;
2065
2066 memset(&cb->args[4], 0,
2067 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2068
2069 /* stop loop if key wrapped back to 0 */
2070 if (key < l->key)
2071 break;
2072 }
2073
2074 cb->args[3] = key;
2075 cb->args[2] = count;
2076
2077 return skb->len;
2078}
2079
2080void __init fib_trie_init(void)
2081{
2082 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2083 sizeof(struct fib_alias),
2084 0, SLAB_PANIC, NULL);
2085
2086 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2087 LEAF_SIZE,
2088 0, SLAB_PANIC, NULL);
2089}
2090
2091struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2092{
2093 struct fib_table *tb;
2094 struct trie *t;
2095 size_t sz = sizeof(*tb);
2096
2097 if (!alias)
2098 sz += sizeof(struct trie);
2099
2100 tb = kzalloc(sz, GFP_KERNEL);
2101 if (!tb)
2102 return NULL;
2103
2104 tb->tb_id = id;
2105 tb->tb_num_default = 0;
2106 tb->tb_data = (alias ? alias->__data : tb->__data);
2107
2108 if (alias)
2109 return tb;
2110
2111 t = (struct trie *) tb->tb_data;
2112 t->kv[0].pos = KEYLENGTH;
2113 t->kv[0].slen = KEYLENGTH;
2114#ifdef CONFIG_IP_FIB_TRIE_STATS
2115 t->stats = alloc_percpu(struct trie_use_stats);
2116 if (!t->stats) {
2117 kfree(tb);
2118 tb = NULL;
2119 }
2120#endif
2121
2122 return tb;
2123}
2124
2125#ifdef CONFIG_PROC_FS
2126/* Depth first Trie walk iterator */
2127struct fib_trie_iter {
2128 struct seq_net_private p;
2129 struct fib_table *tb;
2130 struct key_vector *tnode;
2131 unsigned int index;
2132 unsigned int depth;
2133};
2134
2135static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2136{
2137 unsigned long cindex = iter->index;
2138 struct key_vector *pn = iter->tnode;
2139 t_key pkey;
2140
2141 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2142 iter->tnode, iter->index, iter->depth);
2143
2144 while (!IS_TRIE(pn)) {
2145 while (cindex < child_length(pn)) {
2146 struct key_vector *n = get_child_rcu(pn, cindex++);
2147
2148 if (!n)
2149 continue;
2150
2151 if (IS_LEAF(n)) {
2152 iter->tnode = pn;
2153 iter->index = cindex;
2154 } else {
2155 /* push down one level */
2156 iter->tnode = n;
2157 iter->index = 0;
2158 ++iter->depth;
2159 }
2160
2161 return n;
2162 }
2163
2164 /* Current node exhausted, pop back up */
2165 pkey = pn->key;
2166 pn = node_parent_rcu(pn);
2167 cindex = get_index(pkey, pn) + 1;
2168 --iter->depth;
2169 }
2170
2171 /* record root node so further searches know we are done */
2172 iter->tnode = pn;
2173 iter->index = 0;
2174
2175 return NULL;
2176}
2177
2178static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2179 struct trie *t)
2180{
2181 struct key_vector *n, *pn;
2182
2183 if (!t)
2184 return NULL;
2185
2186 pn = t->kv;
2187 n = rcu_dereference(pn->tnode[0]);
2188 if (!n)
2189 return NULL;
2190
2191 if (IS_TNODE(n)) {
2192 iter->tnode = n;
2193 iter->index = 0;
2194 iter->depth = 1;
2195 } else {
2196 iter->tnode = pn;
2197 iter->index = 0;
2198 iter->depth = 0;
2199 }
2200
2201 return n;
2202}
2203
2204static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2205{
2206 struct key_vector *n;
2207 struct fib_trie_iter iter;
2208
2209 memset(s, 0, sizeof(*s));
2210
2211 rcu_read_lock();
2212 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2213 if (IS_LEAF(n)) {
2214 struct fib_alias *fa;
2215
2216 s->leaves++;
2217 s->totdepth += iter.depth;
2218 if (iter.depth > s->maxdepth)
2219 s->maxdepth = iter.depth;
2220
2221 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2222 ++s->prefixes;
2223 } else {
2224 s->tnodes++;
2225 if (n->bits < MAX_STAT_DEPTH)
2226 s->nodesizes[n->bits]++;
2227 s->nullpointers += tn_info(n)->empty_children;
2228 }
2229 }
2230 rcu_read_unlock();
2231}
2232
2233/*
2234 * This outputs /proc/net/fib_triestats
2235 */
2236static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2237{
2238 unsigned int i, max, pointers, bytes, avdepth;
2239
2240 if (stat->leaves)
2241 avdepth = stat->totdepth*100 / stat->leaves;
2242 else
2243 avdepth = 0;
2244
2245 seq_printf(seq, "\tAver depth: %u.%02d\n",
2246 avdepth / 100, avdepth % 100);
2247 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2248
2249 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2250 bytes = LEAF_SIZE * stat->leaves;
2251
2252 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2253 bytes += sizeof(struct fib_alias) * stat->prefixes;
2254
2255 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2256 bytes += TNODE_SIZE(0) * stat->tnodes;
2257
2258 max = MAX_STAT_DEPTH;
2259 while (max > 0 && stat->nodesizes[max-1] == 0)
2260 max--;
2261
2262 pointers = 0;
2263 for (i = 1; i < max; i++)
2264 if (stat->nodesizes[i] != 0) {
2265 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2266 pointers += (1<<i) * stat->nodesizes[i];
2267 }
2268 seq_putc(seq, '\n');
2269 seq_printf(seq, "\tPointers: %u\n", pointers);
2270
2271 bytes += sizeof(struct key_vector *) * pointers;
2272 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2273 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2274}
2275
2276#ifdef CONFIG_IP_FIB_TRIE_STATS
2277static void trie_show_usage(struct seq_file *seq,
2278 const struct trie_use_stats __percpu *stats)
2279{
2280 struct trie_use_stats s = { 0 };
2281 int cpu;
2282
2283 /* loop through all of the CPUs and gather up the stats */
2284 for_each_possible_cpu(cpu) {
2285 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2286
2287 s.gets += pcpu->gets;
2288 s.backtrack += pcpu->backtrack;
2289 s.semantic_match_passed += pcpu->semantic_match_passed;
2290 s.semantic_match_miss += pcpu->semantic_match_miss;
2291 s.null_node_hit += pcpu->null_node_hit;
2292 s.resize_node_skipped += pcpu->resize_node_skipped;
2293 }
2294
2295 seq_printf(seq, "\nCounters:\n---------\n");
2296 seq_printf(seq, "gets = %u\n", s.gets);
2297 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2298 seq_printf(seq, "semantic match passed = %u\n",
2299 s.semantic_match_passed);
2300 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2301 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2302 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2303}
2304#endif /* CONFIG_IP_FIB_TRIE_STATS */
2305
2306static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2307{
2308 if (tb->tb_id == RT_TABLE_LOCAL)
2309 seq_puts(seq, "Local:\n");
2310 else if (tb->tb_id == RT_TABLE_MAIN)
2311 seq_puts(seq, "Main:\n");
2312 else
2313 seq_printf(seq, "Id %d:\n", tb->tb_id);
2314}
2315
2316
2317static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2318{
2319 struct net *net = (struct net *)seq->private;
2320 unsigned int h;
2321
2322 seq_printf(seq,
2323 "Basic info: size of leaf:"
2324 " %zd bytes, size of tnode: %zd bytes.\n",
2325 LEAF_SIZE, TNODE_SIZE(0));
2326
2327 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2328 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2329 struct fib_table *tb;
2330
2331 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2332 struct trie *t = (struct trie *) tb->tb_data;
2333 struct trie_stat stat;
2334
2335 if (!t)
2336 continue;
2337
2338 fib_table_print(seq, tb);
2339
2340 trie_collect_stats(t, &stat);
2341 trie_show_stats(seq, &stat);
2342#ifdef CONFIG_IP_FIB_TRIE_STATS
2343 trie_show_usage(seq, t->stats);
2344#endif
2345 }
2346 }
2347
2348 return 0;
2349}
2350
2351static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2352{
2353 return single_open_net(inode, file, fib_triestat_seq_show);
2354}
2355
2356static const struct file_operations fib_triestat_fops = {
2357 .open = fib_triestat_seq_open,
2358 .read = seq_read,
2359 .llseek = seq_lseek,
2360 .release = single_release_net,
2361};
2362
2363static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2364{
2365 struct fib_trie_iter *iter = seq->private;
2366 struct net *net = seq_file_net(seq);
2367 loff_t idx = 0;
2368 unsigned int h;
2369
2370 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2371 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2372 struct fib_table *tb;
2373
2374 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2375 struct key_vector *n;
2376
2377 for (n = fib_trie_get_first(iter,
2378 (struct trie *) tb->tb_data);
2379 n; n = fib_trie_get_next(iter))
2380 if (pos == idx++) {
2381 iter->tb = tb;
2382 return n;
2383 }
2384 }
2385 }
2386
2387 return NULL;
2388}
2389
2390static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2391 __acquires(RCU)
2392{
2393 rcu_read_lock();
2394 return fib_trie_get_idx(seq, *pos);
2395}
2396
2397static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2398{
2399 struct fib_trie_iter *iter = seq->private;
2400 struct net *net = seq_file_net(seq);
2401 struct fib_table *tb = iter->tb;
2402 struct hlist_node *tb_node;
2403 unsigned int h;
2404 struct key_vector *n;
2405
2406 ++*pos;
2407 /* next node in same table */
2408 n = fib_trie_get_next(iter);
2409 if (n)
2410 return n;
2411
2412 /* walk rest of this hash chain */
2413 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2414 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2415 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2416 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2417 if (n)
2418 goto found;
2419 }
2420
2421 /* new hash chain */
2422 while (++h < FIB_TABLE_HASHSZ) {
2423 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2424 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2425 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2426 if (n)
2427 goto found;
2428 }
2429 }
2430 return NULL;
2431
2432found:
2433 iter->tb = tb;
2434 return n;
2435}
2436
2437static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2438 __releases(RCU)
2439{
2440 rcu_read_unlock();
2441}
2442
2443static void seq_indent(struct seq_file *seq, int n)
2444{
2445 while (n-- > 0)
2446 seq_puts(seq, " ");
2447}
2448
2449static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2450{
2451 switch (s) {
2452 case RT_SCOPE_UNIVERSE: return "universe";
2453 case RT_SCOPE_SITE: return "site";
2454 case RT_SCOPE_LINK: return "link";
2455 case RT_SCOPE_HOST: return "host";
2456 case RT_SCOPE_NOWHERE: return "nowhere";
2457 default:
2458 snprintf(buf, len, "scope=%d", s);
2459 return buf;
2460 }
2461}
2462
2463static const char *const rtn_type_names[__RTN_MAX] = {
2464 [RTN_UNSPEC] = "UNSPEC",
2465 [RTN_UNICAST] = "UNICAST",
2466 [RTN_LOCAL] = "LOCAL",
2467 [RTN_BROADCAST] = "BROADCAST",
2468 [RTN_ANYCAST] = "ANYCAST",
2469 [RTN_MULTICAST] = "MULTICAST",
2470 [RTN_BLACKHOLE] = "BLACKHOLE",
2471 [RTN_UNREACHABLE] = "UNREACHABLE",
2472 [RTN_PROHIBIT] = "PROHIBIT",
2473 [RTN_THROW] = "THROW",
2474 [RTN_NAT] = "NAT",
2475 [RTN_XRESOLVE] = "XRESOLVE",
2476};
2477
2478static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2479{
2480 if (t < __RTN_MAX && rtn_type_names[t])
2481 return rtn_type_names[t];
2482 snprintf(buf, len, "type %u", t);
2483 return buf;
2484}
2485
2486/* Pretty print the trie */
2487static int fib_trie_seq_show(struct seq_file *seq, void *v)
2488{
2489 const struct fib_trie_iter *iter = seq->private;
2490 struct key_vector *n = v;
2491
2492 if (IS_TRIE(node_parent_rcu(n)))
2493 fib_table_print(seq, iter->tb);
2494
2495 if (IS_TNODE(n)) {
2496 __be32 prf = htonl(n->key);
2497
2498 seq_indent(seq, iter->depth-1);
2499 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2500 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2501 tn_info(n)->full_children,
2502 tn_info(n)->empty_children);
2503 } else {
2504 __be32 val = htonl(n->key);
2505 struct fib_alias *fa;
2506
2507 seq_indent(seq, iter->depth);
2508 seq_printf(seq, " |-- %pI4\n", &val);
2509
2510 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2511 char buf1[32], buf2[32];
2512
2513 seq_indent(seq, iter->depth + 1);
2514 seq_printf(seq, " /%zu %s %s",
2515 KEYLENGTH - fa->fa_slen,
2516 rtn_scope(buf1, sizeof(buf1),
2517 fa->fa_info->fib_scope),
2518 rtn_type(buf2, sizeof(buf2),
2519 fa->fa_type));
2520 if (fa->fa_tos)
2521 seq_printf(seq, " tos=%d", fa->fa_tos);
2522 seq_putc(seq, '\n');
2523 }
2524 }
2525
2526 return 0;
2527}
2528
2529static const struct seq_operations fib_trie_seq_ops = {
2530 .start = fib_trie_seq_start,
2531 .next = fib_trie_seq_next,
2532 .stop = fib_trie_seq_stop,
2533 .show = fib_trie_seq_show,
2534};
2535
2536static int fib_trie_seq_open(struct inode *inode, struct file *file)
2537{
2538 return seq_open_net(inode, file, &fib_trie_seq_ops,
2539 sizeof(struct fib_trie_iter));
2540}
2541
2542static const struct file_operations fib_trie_fops = {
2543 .open = fib_trie_seq_open,
2544 .read = seq_read,
2545 .llseek = seq_lseek,
2546 .release = seq_release_net,
2547};
2548
2549struct fib_route_iter {
2550 struct seq_net_private p;
2551 struct fib_table *main_tb;
2552 struct key_vector *tnode;
2553 loff_t pos;
2554 t_key key;
2555};
2556
2557static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2558 loff_t pos)
2559{
2560 struct key_vector *l, **tp = &iter->tnode;
2561 t_key key;
2562
2563 /* use cached location of previously found key */
2564 if (iter->pos > 0 && pos >= iter->pos) {
2565 key = iter->key;
2566 } else {
2567 iter->pos = 1;
2568 key = 0;
2569 }
2570
2571 pos -= iter->pos;
2572
2573 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2574 key = l->key + 1;
2575 iter->pos++;
2576 l = NULL;
2577
2578 /* handle unlikely case of a key wrap */
2579 if (!key)
2580 break;
2581 }
2582
2583 if (l)
2584 iter->key = l->key; /* remember it */
2585 else
2586 iter->pos = 0; /* forget it */
2587
2588 return l;
2589}
2590
2591static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2592 __acquires(RCU)
2593{
2594 struct fib_route_iter *iter = seq->private;
2595 struct fib_table *tb;
2596 struct trie *t;
2597
2598 rcu_read_lock();
2599
2600 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2601 if (!tb)
2602 return NULL;
2603
2604 iter->main_tb = tb;
2605 t = (struct trie *)tb->tb_data;
2606 iter->tnode = t->kv;
2607
2608 if (*pos != 0)
2609 return fib_route_get_idx(iter, *pos);
2610
2611 iter->pos = 0;
2612 iter->key = KEY_MAX;
2613
2614 return SEQ_START_TOKEN;
2615}
2616
2617static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2618{
2619 struct fib_route_iter *iter = seq->private;
2620 struct key_vector *l = NULL;
2621 t_key key = iter->key + 1;
2622
2623 ++*pos;
2624
2625 /* only allow key of 0 for start of sequence */
2626 if ((v == SEQ_START_TOKEN) || key)
2627 l = leaf_walk_rcu(&iter->tnode, key);
2628
2629 if (l) {
2630 iter->key = l->key;
2631 iter->pos++;
2632 } else {
2633 iter->pos = 0;
2634 }
2635
2636 return l;
2637}
2638
2639static void fib_route_seq_stop(struct seq_file *seq, void *v)
2640 __releases(RCU)
2641{
2642 rcu_read_unlock();
2643}
2644
2645static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2646{
2647 unsigned int flags = 0;
2648
2649 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2650 flags = RTF_REJECT;
2651 if (fi && fi->fib_nh->nh_gw)
2652 flags |= RTF_GATEWAY;
2653 if (mask == htonl(0xFFFFFFFF))
2654 flags |= RTF_HOST;
2655 flags |= RTF_UP;
2656 return flags;
2657}
2658
2659/*
2660 * This outputs /proc/net/route.
2661 * The format of the file is not supposed to be changed
2662 * and needs to be same as fib_hash output to avoid breaking
2663 * legacy utilities
2664 */
2665static int fib_route_seq_show(struct seq_file *seq, void *v)
2666{
2667 struct fib_route_iter *iter = seq->private;
2668 struct fib_table *tb = iter->main_tb;
2669 struct fib_alias *fa;
2670 struct key_vector *l = v;
2671 __be32 prefix;
2672
2673 if (v == SEQ_START_TOKEN) {
2674 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2675 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2676 "\tWindow\tIRTT");
2677 return 0;
2678 }
2679
2680 prefix = htonl(l->key);
2681
2682 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2683 const struct fib_info *fi = fa->fa_info;
2684 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2685 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2686
2687 if ((fa->fa_type == RTN_BROADCAST) ||
2688 (fa->fa_type == RTN_MULTICAST))
2689 continue;
2690
2691 if (fa->tb_id != tb->tb_id)
2692 continue;
2693
2694 seq_setwidth(seq, 127);
2695
2696 if (fi)
2697 seq_printf(seq,
2698 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2699 "%d\t%08X\t%d\t%u\t%u",
2700 fi->fib_dev ? fi->fib_dev->name : "*",
2701 prefix,
2702 fi->fib_nh->nh_gw, flags, 0, 0,
2703 fi->fib_priority,
2704 mask,
2705 (fi->fib_advmss ?
2706 fi->fib_advmss + 40 : 0),
2707 fi->fib_window,
2708 fi->fib_rtt >> 3);
2709 else
2710 seq_printf(seq,
2711 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2712 "%d\t%08X\t%d\t%u\t%u",
2713 prefix, 0, flags, 0, 0, 0,
2714 mask, 0, 0, 0);
2715
2716 seq_pad(seq, '\n');
2717 }
2718
2719 return 0;
2720}
2721
2722static const struct seq_operations fib_route_seq_ops = {
2723 .start = fib_route_seq_start,
2724 .next = fib_route_seq_next,
2725 .stop = fib_route_seq_stop,
2726 .show = fib_route_seq_show,
2727};
2728
2729static int fib_route_seq_open(struct inode *inode, struct file *file)
2730{
2731 return seq_open_net(inode, file, &fib_route_seq_ops,
2732 sizeof(struct fib_route_iter));
2733}
2734
2735static const struct file_operations fib_route_fops = {
2736 .open = fib_route_seq_open,
2737 .read = seq_read,
2738 .llseek = seq_lseek,
2739 .release = seq_release_net,
2740};
2741
2742int __net_init fib_proc_init(struct net *net)
2743{
2744 if (!proc_create("fib_trie", 0444, net->proc_net, &fib_trie_fops))
2745 goto out1;
2746
2747 if (!proc_create("fib_triestat", 0444, net->proc_net,
2748 &fib_triestat_fops))
2749 goto out2;
2750
2751 if (!proc_create("route", 0444, net->proc_net, &fib_route_fops))
2752 goto out3;
2753
2754 return 0;
2755
2756out3:
2757 remove_proc_entry("fib_triestat", net->proc_net);
2758out2:
2759 remove_proc_entry("fib_trie", net->proc_net);
2760out1:
2761 return -ENOMEM;
2762}
2763
2764void __net_exit fib_proc_exit(struct net *net)
2765{
2766 remove_proc_entry("fib_trie", net->proc_net);
2767 remove_proc_entry("fib_triestat", net->proc_net);
2768 remove_proc_entry("route", net->proc_net);
2769}
2770
2771#endif /* CONFIG_PROC_FS */
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 *
4 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
5 * & Swedish University of Agricultural Sciences.
6 *
7 * Jens Laas <jens.laas@data.slu.se> Swedish University of
8 * Agricultural Sciences.
9 *
10 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
11 *
12 * This work is based on the LPC-trie which is originally described in:
13 *
14 * An experimental study of compression methods for dynamic tries
15 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
16 * https://www.csc.kth.se/~snilsson/software/dyntrie2/
17 *
18 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
19 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
20 *
21 * Code from fib_hash has been reused which includes the following header:
22 *
23 * INET An implementation of the TCP/IP protocol suite for the LINUX
24 * operating system. INET is implemented using the BSD Socket
25 * interface as the means of communication with the user level.
26 *
27 * IPv4 FIB: lookup engine and maintenance routines.
28 *
29 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
30 *
31 * Substantial contributions to this work comes from:
32 *
33 * David S. Miller, <davem@davemloft.net>
34 * Stephen Hemminger <shemminger@osdl.org>
35 * Paul E. McKenney <paulmck@us.ibm.com>
36 * Patrick McHardy <kaber@trash.net>
37 */
38#include <linux/cache.h>
39#include <linux/uaccess.h>
40#include <linux/bitops.h>
41#include <linux/types.h>
42#include <linux/kernel.h>
43#include <linux/mm.h>
44#include <linux/string.h>
45#include <linux/socket.h>
46#include <linux/sockios.h>
47#include <linux/errno.h>
48#include <linux/in.h>
49#include <linux/inet.h>
50#include <linux/inetdevice.h>
51#include <linux/netdevice.h>
52#include <linux/if_arp.h>
53#include <linux/proc_fs.h>
54#include <linux/rcupdate.h>
55#include <linux/skbuff.h>
56#include <linux/netlink.h>
57#include <linux/init.h>
58#include <linux/list.h>
59#include <linux/slab.h>
60#include <linux/export.h>
61#include <linux/vmalloc.h>
62#include <linux/notifier.h>
63#include <net/net_namespace.h>
64#include <net/ip.h>
65#include <net/protocol.h>
66#include <net/route.h>
67#include <net/tcp.h>
68#include <net/sock.h>
69#include <net/ip_fib.h>
70#include <net/fib_notifier.h>
71#include <trace/events/fib.h>
72#include "fib_lookup.h"
73
74static int call_fib_entry_notifier(struct notifier_block *nb,
75 enum fib_event_type event_type, u32 dst,
76 int dst_len, struct fib_alias *fa,
77 struct netlink_ext_ack *extack)
78{
79 struct fib_entry_notifier_info info = {
80 .info.extack = extack,
81 .dst = dst,
82 .dst_len = dst_len,
83 .fi = fa->fa_info,
84 .tos = fa->fa_tos,
85 .type = fa->fa_type,
86 .tb_id = fa->tb_id,
87 };
88 return call_fib4_notifier(nb, event_type, &info.info);
89}
90
91static int call_fib_entry_notifiers(struct net *net,
92 enum fib_event_type event_type, u32 dst,
93 int dst_len, struct fib_alias *fa,
94 struct netlink_ext_ack *extack)
95{
96 struct fib_entry_notifier_info info = {
97 .info.extack = extack,
98 .dst = dst,
99 .dst_len = dst_len,
100 .fi = fa->fa_info,
101 .tos = fa->fa_tos,
102 .type = fa->fa_type,
103 .tb_id = fa->tb_id,
104 };
105 return call_fib4_notifiers(net, event_type, &info.info);
106}
107
108#define MAX_STAT_DEPTH 32
109
110#define KEYLENGTH (8*sizeof(t_key))
111#define KEY_MAX ((t_key)~0)
112
113typedef unsigned int t_key;
114
115#define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
116#define IS_TNODE(n) ((n)->bits)
117#define IS_LEAF(n) (!(n)->bits)
118
119struct key_vector {
120 t_key key;
121 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
122 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
123 unsigned char slen;
124 union {
125 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
126 struct hlist_head leaf;
127 /* This array is valid if (pos | bits) > 0 (TNODE) */
128 struct key_vector __rcu *tnode[0];
129 };
130};
131
132struct tnode {
133 struct rcu_head rcu;
134 t_key empty_children; /* KEYLENGTH bits needed */
135 t_key full_children; /* KEYLENGTH bits needed */
136 struct key_vector __rcu *parent;
137 struct key_vector kv[1];
138#define tn_bits kv[0].bits
139};
140
141#define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
142#define LEAF_SIZE TNODE_SIZE(1)
143
144#ifdef CONFIG_IP_FIB_TRIE_STATS
145struct trie_use_stats {
146 unsigned int gets;
147 unsigned int backtrack;
148 unsigned int semantic_match_passed;
149 unsigned int semantic_match_miss;
150 unsigned int null_node_hit;
151 unsigned int resize_node_skipped;
152};
153#endif
154
155struct trie_stat {
156 unsigned int totdepth;
157 unsigned int maxdepth;
158 unsigned int tnodes;
159 unsigned int leaves;
160 unsigned int nullpointers;
161 unsigned int prefixes;
162 unsigned int nodesizes[MAX_STAT_DEPTH];
163};
164
165struct trie {
166 struct key_vector kv[1];
167#ifdef CONFIG_IP_FIB_TRIE_STATS
168 struct trie_use_stats __percpu *stats;
169#endif
170};
171
172static struct key_vector *resize(struct trie *t, struct key_vector *tn);
173static unsigned int tnode_free_size;
174
175/*
176 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
177 * especially useful before resizing the root node with PREEMPT_NONE configs;
178 * the value was obtained experimentally, aiming to avoid visible slowdown.
179 */
180unsigned int sysctl_fib_sync_mem = 512 * 1024;
181unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
182unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
183
184static struct kmem_cache *fn_alias_kmem __ro_after_init;
185static struct kmem_cache *trie_leaf_kmem __ro_after_init;
186
187static inline struct tnode *tn_info(struct key_vector *kv)
188{
189 return container_of(kv, struct tnode, kv[0]);
190}
191
192/* caller must hold RTNL */
193#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
194#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
195
196/* caller must hold RCU read lock or RTNL */
197#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
198#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
199
200/* wrapper for rcu_assign_pointer */
201static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
202{
203 if (n)
204 rcu_assign_pointer(tn_info(n)->parent, tp);
205}
206
207#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
208
209/* This provides us with the number of children in this node, in the case of a
210 * leaf this will return 0 meaning none of the children are accessible.
211 */
212static inline unsigned long child_length(const struct key_vector *tn)
213{
214 return (1ul << tn->bits) & ~(1ul);
215}
216
217#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
218
219static inline unsigned long get_index(t_key key, struct key_vector *kv)
220{
221 unsigned long index = key ^ kv->key;
222
223 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
224 return 0;
225
226 return index >> kv->pos;
227}
228
229/* To understand this stuff, an understanding of keys and all their bits is
230 * necessary. Every node in the trie has a key associated with it, but not
231 * all of the bits in that key are significant.
232 *
233 * Consider a node 'n' and its parent 'tp'.
234 *
235 * If n is a leaf, every bit in its key is significant. Its presence is
236 * necessitated by path compression, since during a tree traversal (when
237 * searching for a leaf - unless we are doing an insertion) we will completely
238 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
239 * a potentially successful search, that we have indeed been walking the
240 * correct key path.
241 *
242 * Note that we can never "miss" the correct key in the tree if present by
243 * following the wrong path. Path compression ensures that segments of the key
244 * that are the same for all keys with a given prefix are skipped, but the
245 * skipped part *is* identical for each node in the subtrie below the skipped
246 * bit! trie_insert() in this implementation takes care of that.
247 *
248 * if n is an internal node - a 'tnode' here, the various parts of its key
249 * have many different meanings.
250 *
251 * Example:
252 * _________________________________________________________________
253 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
254 * -----------------------------------------------------------------
255 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
256 *
257 * _________________________________________________________________
258 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
259 * -----------------------------------------------------------------
260 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
261 *
262 * tp->pos = 22
263 * tp->bits = 3
264 * n->pos = 13
265 * n->bits = 4
266 *
267 * First, let's just ignore the bits that come before the parent tp, that is
268 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
269 * point we do not use them for anything.
270 *
271 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
272 * index into the parent's child array. That is, they will be used to find
273 * 'n' among tp's children.
274 *
275 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
276 * for the node n.
277 *
278 * All the bits we have seen so far are significant to the node n. The rest
279 * of the bits are really not needed or indeed known in n->key.
280 *
281 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
282 * n's child array, and will of course be different for each child.
283 *
284 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
285 * at this point.
286 */
287
288static const int halve_threshold = 25;
289static const int inflate_threshold = 50;
290static const int halve_threshold_root = 15;
291static const int inflate_threshold_root = 30;
292
293static void __alias_free_mem(struct rcu_head *head)
294{
295 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
296 kmem_cache_free(fn_alias_kmem, fa);
297}
298
299static inline void alias_free_mem_rcu(struct fib_alias *fa)
300{
301 call_rcu(&fa->rcu, __alias_free_mem);
302}
303
304#define TNODE_VMALLOC_MAX \
305 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
306
307static void __node_free_rcu(struct rcu_head *head)
308{
309 struct tnode *n = container_of(head, struct tnode, rcu);
310
311 if (!n->tn_bits)
312 kmem_cache_free(trie_leaf_kmem, n);
313 else
314 kvfree(n);
315}
316
317#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
318
319static struct tnode *tnode_alloc(int bits)
320{
321 size_t size;
322
323 /* verify bits is within bounds */
324 if (bits > TNODE_VMALLOC_MAX)
325 return NULL;
326
327 /* determine size and verify it is non-zero and didn't overflow */
328 size = TNODE_SIZE(1ul << bits);
329
330 if (size <= PAGE_SIZE)
331 return kzalloc(size, GFP_KERNEL);
332 else
333 return vzalloc(size);
334}
335
336static inline void empty_child_inc(struct key_vector *n)
337{
338 tn_info(n)->empty_children++;
339
340 if (!tn_info(n)->empty_children)
341 tn_info(n)->full_children++;
342}
343
344static inline void empty_child_dec(struct key_vector *n)
345{
346 if (!tn_info(n)->empty_children)
347 tn_info(n)->full_children--;
348
349 tn_info(n)->empty_children--;
350}
351
352static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
353{
354 struct key_vector *l;
355 struct tnode *kv;
356
357 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
358 if (!kv)
359 return NULL;
360
361 /* initialize key vector */
362 l = kv->kv;
363 l->key = key;
364 l->pos = 0;
365 l->bits = 0;
366 l->slen = fa->fa_slen;
367
368 /* link leaf to fib alias */
369 INIT_HLIST_HEAD(&l->leaf);
370 hlist_add_head(&fa->fa_list, &l->leaf);
371
372 return l;
373}
374
375static struct key_vector *tnode_new(t_key key, int pos, int bits)
376{
377 unsigned int shift = pos + bits;
378 struct key_vector *tn;
379 struct tnode *tnode;
380
381 /* verify bits and pos their msb bits clear and values are valid */
382 BUG_ON(!bits || (shift > KEYLENGTH));
383
384 tnode = tnode_alloc(bits);
385 if (!tnode)
386 return NULL;
387
388 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
389 sizeof(struct key_vector *) << bits);
390
391 if (bits == KEYLENGTH)
392 tnode->full_children = 1;
393 else
394 tnode->empty_children = 1ul << bits;
395
396 tn = tnode->kv;
397 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
398 tn->pos = pos;
399 tn->bits = bits;
400 tn->slen = pos;
401
402 return tn;
403}
404
405/* Check whether a tnode 'n' is "full", i.e. it is an internal node
406 * and no bits are skipped. See discussion in dyntree paper p. 6
407 */
408static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
409{
410 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
411}
412
413/* Add a child at position i overwriting the old value.
414 * Update the value of full_children and empty_children.
415 */
416static void put_child(struct key_vector *tn, unsigned long i,
417 struct key_vector *n)
418{
419 struct key_vector *chi = get_child(tn, i);
420 int isfull, wasfull;
421
422 BUG_ON(i >= child_length(tn));
423
424 /* update emptyChildren, overflow into fullChildren */
425 if (!n && chi)
426 empty_child_inc(tn);
427 if (n && !chi)
428 empty_child_dec(tn);
429
430 /* update fullChildren */
431 wasfull = tnode_full(tn, chi);
432 isfull = tnode_full(tn, n);
433
434 if (wasfull && !isfull)
435 tn_info(tn)->full_children--;
436 else if (!wasfull && isfull)
437 tn_info(tn)->full_children++;
438
439 if (n && (tn->slen < n->slen))
440 tn->slen = n->slen;
441
442 rcu_assign_pointer(tn->tnode[i], n);
443}
444
445static void update_children(struct key_vector *tn)
446{
447 unsigned long i;
448
449 /* update all of the child parent pointers */
450 for (i = child_length(tn); i;) {
451 struct key_vector *inode = get_child(tn, --i);
452
453 if (!inode)
454 continue;
455
456 /* Either update the children of a tnode that
457 * already belongs to us or update the child
458 * to point to ourselves.
459 */
460 if (node_parent(inode) == tn)
461 update_children(inode);
462 else
463 node_set_parent(inode, tn);
464 }
465}
466
467static inline void put_child_root(struct key_vector *tp, t_key key,
468 struct key_vector *n)
469{
470 if (IS_TRIE(tp))
471 rcu_assign_pointer(tp->tnode[0], n);
472 else
473 put_child(tp, get_index(key, tp), n);
474}
475
476static inline void tnode_free_init(struct key_vector *tn)
477{
478 tn_info(tn)->rcu.next = NULL;
479}
480
481static inline void tnode_free_append(struct key_vector *tn,
482 struct key_vector *n)
483{
484 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
485 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
486}
487
488static void tnode_free(struct key_vector *tn)
489{
490 struct callback_head *head = &tn_info(tn)->rcu;
491
492 while (head) {
493 head = head->next;
494 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
495 node_free(tn);
496
497 tn = container_of(head, struct tnode, rcu)->kv;
498 }
499
500 if (tnode_free_size >= sysctl_fib_sync_mem) {
501 tnode_free_size = 0;
502 synchronize_rcu();
503 }
504}
505
506static struct key_vector *replace(struct trie *t,
507 struct key_vector *oldtnode,
508 struct key_vector *tn)
509{
510 struct key_vector *tp = node_parent(oldtnode);
511 unsigned long i;
512
513 /* setup the parent pointer out of and back into this node */
514 NODE_INIT_PARENT(tn, tp);
515 put_child_root(tp, tn->key, tn);
516
517 /* update all of the child parent pointers */
518 update_children(tn);
519
520 /* all pointers should be clean so we are done */
521 tnode_free(oldtnode);
522
523 /* resize children now that oldtnode is freed */
524 for (i = child_length(tn); i;) {
525 struct key_vector *inode = get_child(tn, --i);
526
527 /* resize child node */
528 if (tnode_full(tn, inode))
529 tn = resize(t, inode);
530 }
531
532 return tp;
533}
534
535static struct key_vector *inflate(struct trie *t,
536 struct key_vector *oldtnode)
537{
538 struct key_vector *tn;
539 unsigned long i;
540 t_key m;
541
542 pr_debug("In inflate\n");
543
544 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
545 if (!tn)
546 goto notnode;
547
548 /* prepare oldtnode to be freed */
549 tnode_free_init(oldtnode);
550
551 /* Assemble all of the pointers in our cluster, in this case that
552 * represents all of the pointers out of our allocated nodes that
553 * point to existing tnodes and the links between our allocated
554 * nodes.
555 */
556 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
557 struct key_vector *inode = get_child(oldtnode, --i);
558 struct key_vector *node0, *node1;
559 unsigned long j, k;
560
561 /* An empty child */
562 if (!inode)
563 continue;
564
565 /* A leaf or an internal node with skipped bits */
566 if (!tnode_full(oldtnode, inode)) {
567 put_child(tn, get_index(inode->key, tn), inode);
568 continue;
569 }
570
571 /* drop the node in the old tnode free list */
572 tnode_free_append(oldtnode, inode);
573
574 /* An internal node with two children */
575 if (inode->bits == 1) {
576 put_child(tn, 2 * i + 1, get_child(inode, 1));
577 put_child(tn, 2 * i, get_child(inode, 0));
578 continue;
579 }
580
581 /* We will replace this node 'inode' with two new
582 * ones, 'node0' and 'node1', each with half of the
583 * original children. The two new nodes will have
584 * a position one bit further down the key and this
585 * means that the "significant" part of their keys
586 * (see the discussion near the top of this file)
587 * will differ by one bit, which will be "0" in
588 * node0's key and "1" in node1's key. Since we are
589 * moving the key position by one step, the bit that
590 * we are moving away from - the bit at position
591 * (tn->pos) - is the one that will differ between
592 * node0 and node1. So... we synthesize that bit in the
593 * two new keys.
594 */
595 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
596 if (!node1)
597 goto nomem;
598 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
599
600 tnode_free_append(tn, node1);
601 if (!node0)
602 goto nomem;
603 tnode_free_append(tn, node0);
604
605 /* populate child pointers in new nodes */
606 for (k = child_length(inode), j = k / 2; j;) {
607 put_child(node1, --j, get_child(inode, --k));
608 put_child(node0, j, get_child(inode, j));
609 put_child(node1, --j, get_child(inode, --k));
610 put_child(node0, j, get_child(inode, j));
611 }
612
613 /* link new nodes to parent */
614 NODE_INIT_PARENT(node1, tn);
615 NODE_INIT_PARENT(node0, tn);
616
617 /* link parent to nodes */
618 put_child(tn, 2 * i + 1, node1);
619 put_child(tn, 2 * i, node0);
620 }
621
622 /* setup the parent pointers into and out of this node */
623 return replace(t, oldtnode, tn);
624nomem:
625 /* all pointers should be clean so we are done */
626 tnode_free(tn);
627notnode:
628 return NULL;
629}
630
631static struct key_vector *halve(struct trie *t,
632 struct key_vector *oldtnode)
633{
634 struct key_vector *tn;
635 unsigned long i;
636
637 pr_debug("In halve\n");
638
639 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
640 if (!tn)
641 goto notnode;
642
643 /* prepare oldtnode to be freed */
644 tnode_free_init(oldtnode);
645
646 /* Assemble all of the pointers in our cluster, in this case that
647 * represents all of the pointers out of our allocated nodes that
648 * point to existing tnodes and the links between our allocated
649 * nodes.
650 */
651 for (i = child_length(oldtnode); i;) {
652 struct key_vector *node1 = get_child(oldtnode, --i);
653 struct key_vector *node0 = get_child(oldtnode, --i);
654 struct key_vector *inode;
655
656 /* At least one of the children is empty */
657 if (!node1 || !node0) {
658 put_child(tn, i / 2, node1 ? : node0);
659 continue;
660 }
661
662 /* Two nonempty children */
663 inode = tnode_new(node0->key, oldtnode->pos, 1);
664 if (!inode)
665 goto nomem;
666 tnode_free_append(tn, inode);
667
668 /* initialize pointers out of node */
669 put_child(inode, 1, node1);
670 put_child(inode, 0, node0);
671 NODE_INIT_PARENT(inode, tn);
672
673 /* link parent to node */
674 put_child(tn, i / 2, inode);
675 }
676
677 /* setup the parent pointers into and out of this node */
678 return replace(t, oldtnode, tn);
679nomem:
680 /* all pointers should be clean so we are done */
681 tnode_free(tn);
682notnode:
683 return NULL;
684}
685
686static struct key_vector *collapse(struct trie *t,
687 struct key_vector *oldtnode)
688{
689 struct key_vector *n, *tp;
690 unsigned long i;
691
692 /* scan the tnode looking for that one child that might still exist */
693 for (n = NULL, i = child_length(oldtnode); !n && i;)
694 n = get_child(oldtnode, --i);
695
696 /* compress one level */
697 tp = node_parent(oldtnode);
698 put_child_root(tp, oldtnode->key, n);
699 node_set_parent(n, tp);
700
701 /* drop dead node */
702 node_free(oldtnode);
703
704 return tp;
705}
706
707static unsigned char update_suffix(struct key_vector *tn)
708{
709 unsigned char slen = tn->pos;
710 unsigned long stride, i;
711 unsigned char slen_max;
712
713 /* only vector 0 can have a suffix length greater than or equal to
714 * tn->pos + tn->bits, the second highest node will have a suffix
715 * length at most of tn->pos + tn->bits - 1
716 */
717 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
718
719 /* search though the list of children looking for nodes that might
720 * have a suffix greater than the one we currently have. This is
721 * why we start with a stride of 2 since a stride of 1 would
722 * represent the nodes with suffix length equal to tn->pos
723 */
724 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
725 struct key_vector *n = get_child(tn, i);
726
727 if (!n || (n->slen <= slen))
728 continue;
729
730 /* update stride and slen based on new value */
731 stride <<= (n->slen - slen);
732 slen = n->slen;
733 i &= ~(stride - 1);
734
735 /* stop searching if we have hit the maximum possible value */
736 if (slen >= slen_max)
737 break;
738 }
739
740 tn->slen = slen;
741
742 return slen;
743}
744
745/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
746 * the Helsinki University of Technology and Matti Tikkanen of Nokia
747 * Telecommunications, page 6:
748 * "A node is doubled if the ratio of non-empty children to all
749 * children in the *doubled* node is at least 'high'."
750 *
751 * 'high' in this instance is the variable 'inflate_threshold'. It
752 * is expressed as a percentage, so we multiply it with
753 * child_length() and instead of multiplying by 2 (since the
754 * child array will be doubled by inflate()) and multiplying
755 * the left-hand side by 100 (to handle the percentage thing) we
756 * multiply the left-hand side by 50.
757 *
758 * The left-hand side may look a bit weird: child_length(tn)
759 * - tn->empty_children is of course the number of non-null children
760 * in the current node. tn->full_children is the number of "full"
761 * children, that is non-null tnodes with a skip value of 0.
762 * All of those will be doubled in the resulting inflated tnode, so
763 * we just count them one extra time here.
764 *
765 * A clearer way to write this would be:
766 *
767 * to_be_doubled = tn->full_children;
768 * not_to_be_doubled = child_length(tn) - tn->empty_children -
769 * tn->full_children;
770 *
771 * new_child_length = child_length(tn) * 2;
772 *
773 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
774 * new_child_length;
775 * if (new_fill_factor >= inflate_threshold)
776 *
777 * ...and so on, tho it would mess up the while () loop.
778 *
779 * anyway,
780 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
781 * inflate_threshold
782 *
783 * avoid a division:
784 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
785 * inflate_threshold * new_child_length
786 *
787 * expand not_to_be_doubled and to_be_doubled, and shorten:
788 * 100 * (child_length(tn) - tn->empty_children +
789 * tn->full_children) >= inflate_threshold * new_child_length
790 *
791 * expand new_child_length:
792 * 100 * (child_length(tn) - tn->empty_children +
793 * tn->full_children) >=
794 * inflate_threshold * child_length(tn) * 2
795 *
796 * shorten again:
797 * 50 * (tn->full_children + child_length(tn) -
798 * tn->empty_children) >= inflate_threshold *
799 * child_length(tn)
800 *
801 */
802static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
803{
804 unsigned long used = child_length(tn);
805 unsigned long threshold = used;
806
807 /* Keep root node larger */
808 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
809 used -= tn_info(tn)->empty_children;
810 used += tn_info(tn)->full_children;
811
812 /* if bits == KEYLENGTH then pos = 0, and will fail below */
813
814 return (used > 1) && tn->pos && ((50 * used) >= threshold);
815}
816
817static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
818{
819 unsigned long used = child_length(tn);
820 unsigned long threshold = used;
821
822 /* Keep root node larger */
823 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
824 used -= tn_info(tn)->empty_children;
825
826 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
827
828 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
829}
830
831static inline bool should_collapse(struct key_vector *tn)
832{
833 unsigned long used = child_length(tn);
834
835 used -= tn_info(tn)->empty_children;
836
837 /* account for bits == KEYLENGTH case */
838 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
839 used -= KEY_MAX;
840
841 /* One child or none, time to drop us from the trie */
842 return used < 2;
843}
844
845#define MAX_WORK 10
846static struct key_vector *resize(struct trie *t, struct key_vector *tn)
847{
848#ifdef CONFIG_IP_FIB_TRIE_STATS
849 struct trie_use_stats __percpu *stats = t->stats;
850#endif
851 struct key_vector *tp = node_parent(tn);
852 unsigned long cindex = get_index(tn->key, tp);
853 int max_work = MAX_WORK;
854
855 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
856 tn, inflate_threshold, halve_threshold);
857
858 /* track the tnode via the pointer from the parent instead of
859 * doing it ourselves. This way we can let RCU fully do its
860 * thing without us interfering
861 */
862 BUG_ON(tn != get_child(tp, cindex));
863
864 /* Double as long as the resulting node has a number of
865 * nonempty nodes that are above the threshold.
866 */
867 while (should_inflate(tp, tn) && max_work) {
868 tp = inflate(t, tn);
869 if (!tp) {
870#ifdef CONFIG_IP_FIB_TRIE_STATS
871 this_cpu_inc(stats->resize_node_skipped);
872#endif
873 break;
874 }
875
876 max_work--;
877 tn = get_child(tp, cindex);
878 }
879
880 /* update parent in case inflate failed */
881 tp = node_parent(tn);
882
883 /* Return if at least one inflate is run */
884 if (max_work != MAX_WORK)
885 return tp;
886
887 /* Halve as long as the number of empty children in this
888 * node is above threshold.
889 */
890 while (should_halve(tp, tn) && max_work) {
891 tp = halve(t, tn);
892 if (!tp) {
893#ifdef CONFIG_IP_FIB_TRIE_STATS
894 this_cpu_inc(stats->resize_node_skipped);
895#endif
896 break;
897 }
898
899 max_work--;
900 tn = get_child(tp, cindex);
901 }
902
903 /* Only one child remains */
904 if (should_collapse(tn))
905 return collapse(t, tn);
906
907 /* update parent in case halve failed */
908 return node_parent(tn);
909}
910
911static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
912{
913 unsigned char node_slen = tn->slen;
914
915 while ((node_slen > tn->pos) && (node_slen > slen)) {
916 slen = update_suffix(tn);
917 if (node_slen == slen)
918 break;
919
920 tn = node_parent(tn);
921 node_slen = tn->slen;
922 }
923}
924
925static void node_push_suffix(struct key_vector *tn, unsigned char slen)
926{
927 while (tn->slen < slen) {
928 tn->slen = slen;
929 tn = node_parent(tn);
930 }
931}
932
933/* rcu_read_lock needs to be hold by caller from readside */
934static struct key_vector *fib_find_node(struct trie *t,
935 struct key_vector **tp, u32 key)
936{
937 struct key_vector *pn, *n = t->kv;
938 unsigned long index = 0;
939
940 do {
941 pn = n;
942 n = get_child_rcu(n, index);
943
944 if (!n)
945 break;
946
947 index = get_cindex(key, n);
948
949 /* This bit of code is a bit tricky but it combines multiple
950 * checks into a single check. The prefix consists of the
951 * prefix plus zeros for the bits in the cindex. The index
952 * is the difference between the key and this value. From
953 * this we can actually derive several pieces of data.
954 * if (index >= (1ul << bits))
955 * we have a mismatch in skip bits and failed
956 * else
957 * we know the value is cindex
958 *
959 * This check is safe even if bits == KEYLENGTH due to the
960 * fact that we can only allocate a node with 32 bits if a
961 * long is greater than 32 bits.
962 */
963 if (index >= (1ul << n->bits)) {
964 n = NULL;
965 break;
966 }
967
968 /* keep searching until we find a perfect match leaf or NULL */
969 } while (IS_TNODE(n));
970
971 *tp = pn;
972
973 return n;
974}
975
976/* Return the first fib alias matching TOS with
977 * priority less than or equal to PRIO.
978 * If 'find_first' is set, return the first matching
979 * fib alias, regardless of TOS and priority.
980 */
981static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
982 u8 tos, u32 prio, u32 tb_id,
983 bool find_first)
984{
985 struct fib_alias *fa;
986
987 if (!fah)
988 return NULL;
989
990 hlist_for_each_entry(fa, fah, fa_list) {
991 if (fa->fa_slen < slen)
992 continue;
993 if (fa->fa_slen != slen)
994 break;
995 if (fa->tb_id > tb_id)
996 continue;
997 if (fa->tb_id != tb_id)
998 break;
999 if (find_first)
1000 return fa;
1001 if (fa->fa_tos > tos)
1002 continue;
1003 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1004 return fa;
1005 }
1006
1007 return NULL;
1008}
1009
1010static struct fib_alias *
1011fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1012{
1013 u8 slen = KEYLENGTH - fri->dst_len;
1014 struct key_vector *l, *tp;
1015 struct fib_table *tb;
1016 struct fib_alias *fa;
1017 struct trie *t;
1018
1019 tb = fib_get_table(net, fri->tb_id);
1020 if (!tb)
1021 return NULL;
1022
1023 t = (struct trie *)tb->tb_data;
1024 l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1025 if (!l)
1026 return NULL;
1027
1028 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1029 if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1030 fa->fa_tos == fri->tos && fa->fa_info == fri->fi &&
1031 fa->fa_type == fri->type)
1032 return fa;
1033 }
1034
1035 return NULL;
1036}
1037
1038void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1039{
1040 struct fib_alias *fa_match;
1041
1042 rcu_read_lock();
1043
1044 fa_match = fib_find_matching_alias(net, fri);
1045 if (!fa_match)
1046 goto out;
1047
1048 fa_match->offload = fri->offload;
1049 fa_match->trap = fri->trap;
1050
1051out:
1052 rcu_read_unlock();
1053}
1054EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1055
1056static void trie_rebalance(struct trie *t, struct key_vector *tn)
1057{
1058 while (!IS_TRIE(tn))
1059 tn = resize(t, tn);
1060}
1061
1062static int fib_insert_node(struct trie *t, struct key_vector *tp,
1063 struct fib_alias *new, t_key key)
1064{
1065 struct key_vector *n, *l;
1066
1067 l = leaf_new(key, new);
1068 if (!l)
1069 goto noleaf;
1070
1071 /* retrieve child from parent node */
1072 n = get_child(tp, get_index(key, tp));
1073
1074 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1075 *
1076 * Add a new tnode here
1077 * first tnode need some special handling
1078 * leaves us in position for handling as case 3
1079 */
1080 if (n) {
1081 struct key_vector *tn;
1082
1083 tn = tnode_new(key, __fls(key ^ n->key), 1);
1084 if (!tn)
1085 goto notnode;
1086
1087 /* initialize routes out of node */
1088 NODE_INIT_PARENT(tn, tp);
1089 put_child(tn, get_index(key, tn) ^ 1, n);
1090
1091 /* start adding routes into the node */
1092 put_child_root(tp, key, tn);
1093 node_set_parent(n, tn);
1094
1095 /* parent now has a NULL spot where the leaf can go */
1096 tp = tn;
1097 }
1098
1099 /* Case 3: n is NULL, and will just insert a new leaf */
1100 node_push_suffix(tp, new->fa_slen);
1101 NODE_INIT_PARENT(l, tp);
1102 put_child_root(tp, key, l);
1103 trie_rebalance(t, tp);
1104
1105 return 0;
1106notnode:
1107 node_free(l);
1108noleaf:
1109 return -ENOMEM;
1110}
1111
1112static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1113 struct key_vector *l, struct fib_alias *new,
1114 struct fib_alias *fa, t_key key)
1115{
1116 if (!l)
1117 return fib_insert_node(t, tp, new, key);
1118
1119 if (fa) {
1120 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1121 } else {
1122 struct fib_alias *last;
1123
1124 hlist_for_each_entry(last, &l->leaf, fa_list) {
1125 if (new->fa_slen < last->fa_slen)
1126 break;
1127 if ((new->fa_slen == last->fa_slen) &&
1128 (new->tb_id > last->tb_id))
1129 break;
1130 fa = last;
1131 }
1132
1133 if (fa)
1134 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1135 else
1136 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1137 }
1138
1139 /* if we added to the tail node then we need to update slen */
1140 if (l->slen < new->fa_slen) {
1141 l->slen = new->fa_slen;
1142 node_push_suffix(tp, new->fa_slen);
1143 }
1144
1145 return 0;
1146}
1147
1148static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1149{
1150 if (plen > KEYLENGTH) {
1151 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1152 return false;
1153 }
1154
1155 if ((plen < KEYLENGTH) && (key << plen)) {
1156 NL_SET_ERR_MSG(extack,
1157 "Invalid prefix for given prefix length");
1158 return false;
1159 }
1160
1161 return true;
1162}
1163
1164static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1165 struct key_vector *l, struct fib_alias *old);
1166
1167/* Caller must hold RTNL. */
1168int fib_table_insert(struct net *net, struct fib_table *tb,
1169 struct fib_config *cfg, struct netlink_ext_ack *extack)
1170{
1171 struct trie *t = (struct trie *)tb->tb_data;
1172 struct fib_alias *fa, *new_fa;
1173 struct key_vector *l, *tp;
1174 u16 nlflags = NLM_F_EXCL;
1175 struct fib_info *fi;
1176 u8 plen = cfg->fc_dst_len;
1177 u8 slen = KEYLENGTH - plen;
1178 u8 tos = cfg->fc_tos;
1179 u32 key;
1180 int err;
1181
1182 key = ntohl(cfg->fc_dst);
1183
1184 if (!fib_valid_key_len(key, plen, extack))
1185 return -EINVAL;
1186
1187 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1188
1189 fi = fib_create_info(cfg, extack);
1190 if (IS_ERR(fi)) {
1191 err = PTR_ERR(fi);
1192 goto err;
1193 }
1194
1195 l = fib_find_node(t, &tp, key);
1196 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1197 tb->tb_id, false) : NULL;
1198
1199 /* Now fa, if non-NULL, points to the first fib alias
1200 * with the same keys [prefix,tos,priority], if such key already
1201 * exists or to the node before which we will insert new one.
1202 *
1203 * If fa is NULL, we will need to allocate a new one and
1204 * insert to the tail of the section matching the suffix length
1205 * of the new alias.
1206 */
1207
1208 if (fa && fa->fa_tos == tos &&
1209 fa->fa_info->fib_priority == fi->fib_priority) {
1210 struct fib_alias *fa_first, *fa_match;
1211
1212 err = -EEXIST;
1213 if (cfg->fc_nlflags & NLM_F_EXCL)
1214 goto out;
1215
1216 nlflags &= ~NLM_F_EXCL;
1217
1218 /* We have 2 goals:
1219 * 1. Find exact match for type, scope, fib_info to avoid
1220 * duplicate routes
1221 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1222 */
1223 fa_match = NULL;
1224 fa_first = fa;
1225 hlist_for_each_entry_from(fa, fa_list) {
1226 if ((fa->fa_slen != slen) ||
1227 (fa->tb_id != tb->tb_id) ||
1228 (fa->fa_tos != tos))
1229 break;
1230 if (fa->fa_info->fib_priority != fi->fib_priority)
1231 break;
1232 if (fa->fa_type == cfg->fc_type &&
1233 fa->fa_info == fi) {
1234 fa_match = fa;
1235 break;
1236 }
1237 }
1238
1239 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1240 struct fib_info *fi_drop;
1241 u8 state;
1242
1243 nlflags |= NLM_F_REPLACE;
1244 fa = fa_first;
1245 if (fa_match) {
1246 if (fa == fa_match)
1247 err = 0;
1248 goto out;
1249 }
1250 err = -ENOBUFS;
1251 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1252 if (!new_fa)
1253 goto out;
1254
1255 fi_drop = fa->fa_info;
1256 new_fa->fa_tos = fa->fa_tos;
1257 new_fa->fa_info = fi;
1258 new_fa->fa_type = cfg->fc_type;
1259 state = fa->fa_state;
1260 new_fa->fa_state = state & ~FA_S_ACCESSED;
1261 new_fa->fa_slen = fa->fa_slen;
1262 new_fa->tb_id = tb->tb_id;
1263 new_fa->fa_default = -1;
1264 new_fa->offload = 0;
1265 new_fa->trap = 0;
1266
1267 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1268
1269 if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1270 tb->tb_id, true) == new_fa) {
1271 enum fib_event_type fib_event;
1272
1273 fib_event = FIB_EVENT_ENTRY_REPLACE;
1274 err = call_fib_entry_notifiers(net, fib_event,
1275 key, plen,
1276 new_fa, extack);
1277 if (err) {
1278 hlist_replace_rcu(&new_fa->fa_list,
1279 &fa->fa_list);
1280 goto out_free_new_fa;
1281 }
1282 }
1283
1284 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1285 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1286
1287 alias_free_mem_rcu(fa);
1288
1289 fib_release_info(fi_drop);
1290 if (state & FA_S_ACCESSED)
1291 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1292
1293 goto succeeded;
1294 }
1295 /* Error if we find a perfect match which
1296 * uses the same scope, type, and nexthop
1297 * information.
1298 */
1299 if (fa_match)
1300 goto out;
1301
1302 if (cfg->fc_nlflags & NLM_F_APPEND)
1303 nlflags |= NLM_F_APPEND;
1304 else
1305 fa = fa_first;
1306 }
1307 err = -ENOENT;
1308 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1309 goto out;
1310
1311 nlflags |= NLM_F_CREATE;
1312 err = -ENOBUFS;
1313 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1314 if (!new_fa)
1315 goto out;
1316
1317 new_fa->fa_info = fi;
1318 new_fa->fa_tos = tos;
1319 new_fa->fa_type = cfg->fc_type;
1320 new_fa->fa_state = 0;
1321 new_fa->fa_slen = slen;
1322 new_fa->tb_id = tb->tb_id;
1323 new_fa->fa_default = -1;
1324 new_fa->offload = 0;
1325 new_fa->trap = 0;
1326
1327 /* Insert new entry to the list. */
1328 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1329 if (err)
1330 goto out_free_new_fa;
1331
1332 /* The alias was already inserted, so the node must exist. */
1333 l = l ? l : fib_find_node(t, &tp, key);
1334 if (WARN_ON_ONCE(!l))
1335 goto out_free_new_fa;
1336
1337 if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1338 new_fa) {
1339 enum fib_event_type fib_event;
1340
1341 fib_event = FIB_EVENT_ENTRY_REPLACE;
1342 err = call_fib_entry_notifiers(net, fib_event, key, plen,
1343 new_fa, extack);
1344 if (err)
1345 goto out_remove_new_fa;
1346 }
1347
1348 if (!plen)
1349 tb->tb_num_default++;
1350
1351 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1352 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1353 &cfg->fc_nlinfo, nlflags);
1354succeeded:
1355 return 0;
1356
1357out_remove_new_fa:
1358 fib_remove_alias(t, tp, l, new_fa);
1359out_free_new_fa:
1360 kmem_cache_free(fn_alias_kmem, new_fa);
1361out:
1362 fib_release_info(fi);
1363err:
1364 return err;
1365}
1366
1367static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1368{
1369 t_key prefix = n->key;
1370
1371 return (key ^ prefix) & (prefix | -prefix);
1372}
1373
1374bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1375 const struct flowi4 *flp)
1376{
1377 if (nhc->nhc_flags & RTNH_F_DEAD)
1378 return false;
1379
1380 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1381 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1382 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1383 return false;
1384
1385 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1386 if (flp->flowi4_oif &&
1387 flp->flowi4_oif != nhc->nhc_oif)
1388 return false;
1389 }
1390
1391 return true;
1392}
1393
1394/* should be called with rcu_read_lock */
1395int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1396 struct fib_result *res, int fib_flags)
1397{
1398 struct trie *t = (struct trie *) tb->tb_data;
1399#ifdef CONFIG_IP_FIB_TRIE_STATS
1400 struct trie_use_stats __percpu *stats = t->stats;
1401#endif
1402 const t_key key = ntohl(flp->daddr);
1403 struct key_vector *n, *pn;
1404 struct fib_alias *fa;
1405 unsigned long index;
1406 t_key cindex;
1407
1408 pn = t->kv;
1409 cindex = 0;
1410
1411 n = get_child_rcu(pn, cindex);
1412 if (!n) {
1413 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1414 return -EAGAIN;
1415 }
1416
1417#ifdef CONFIG_IP_FIB_TRIE_STATS
1418 this_cpu_inc(stats->gets);
1419#endif
1420
1421 /* Step 1: Travel to the longest prefix match in the trie */
1422 for (;;) {
1423 index = get_cindex(key, n);
1424
1425 /* This bit of code is a bit tricky but it combines multiple
1426 * checks into a single check. The prefix consists of the
1427 * prefix plus zeros for the "bits" in the prefix. The index
1428 * is the difference between the key and this value. From
1429 * this we can actually derive several pieces of data.
1430 * if (index >= (1ul << bits))
1431 * we have a mismatch in skip bits and failed
1432 * else
1433 * we know the value is cindex
1434 *
1435 * This check is safe even if bits == KEYLENGTH due to the
1436 * fact that we can only allocate a node with 32 bits if a
1437 * long is greater than 32 bits.
1438 */
1439 if (index >= (1ul << n->bits))
1440 break;
1441
1442 /* we have found a leaf. Prefixes have already been compared */
1443 if (IS_LEAF(n))
1444 goto found;
1445
1446 /* only record pn and cindex if we are going to be chopping
1447 * bits later. Otherwise we are just wasting cycles.
1448 */
1449 if (n->slen > n->pos) {
1450 pn = n;
1451 cindex = index;
1452 }
1453
1454 n = get_child_rcu(n, index);
1455 if (unlikely(!n))
1456 goto backtrace;
1457 }
1458
1459 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1460 for (;;) {
1461 /* record the pointer where our next node pointer is stored */
1462 struct key_vector __rcu **cptr = n->tnode;
1463
1464 /* This test verifies that none of the bits that differ
1465 * between the key and the prefix exist in the region of
1466 * the lsb and higher in the prefix.
1467 */
1468 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1469 goto backtrace;
1470
1471 /* exit out and process leaf */
1472 if (unlikely(IS_LEAF(n)))
1473 break;
1474
1475 /* Don't bother recording parent info. Since we are in
1476 * prefix match mode we will have to come back to wherever
1477 * we started this traversal anyway
1478 */
1479
1480 while ((n = rcu_dereference(*cptr)) == NULL) {
1481backtrace:
1482#ifdef CONFIG_IP_FIB_TRIE_STATS
1483 if (!n)
1484 this_cpu_inc(stats->null_node_hit);
1485#endif
1486 /* If we are at cindex 0 there are no more bits for
1487 * us to strip at this level so we must ascend back
1488 * up one level to see if there are any more bits to
1489 * be stripped there.
1490 */
1491 while (!cindex) {
1492 t_key pkey = pn->key;
1493
1494 /* If we don't have a parent then there is
1495 * nothing for us to do as we do not have any
1496 * further nodes to parse.
1497 */
1498 if (IS_TRIE(pn)) {
1499 trace_fib_table_lookup(tb->tb_id, flp,
1500 NULL, -EAGAIN);
1501 return -EAGAIN;
1502 }
1503#ifdef CONFIG_IP_FIB_TRIE_STATS
1504 this_cpu_inc(stats->backtrack);
1505#endif
1506 /* Get Child's index */
1507 pn = node_parent_rcu(pn);
1508 cindex = get_index(pkey, pn);
1509 }
1510
1511 /* strip the least significant bit from the cindex */
1512 cindex &= cindex - 1;
1513
1514 /* grab pointer for next child node */
1515 cptr = &pn->tnode[cindex];
1516 }
1517 }
1518
1519found:
1520 /* this line carries forward the xor from earlier in the function */
1521 index = key ^ n->key;
1522
1523 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1524 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1525 struct fib_info *fi = fa->fa_info;
1526 struct fib_nh_common *nhc;
1527 int nhsel, err;
1528
1529 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1530 if (index >= (1ul << fa->fa_slen))
1531 continue;
1532 }
1533 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1534 continue;
1535 if (fi->fib_dead)
1536 continue;
1537 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1538 continue;
1539 fib_alias_accessed(fa);
1540 err = fib_props[fa->fa_type].error;
1541 if (unlikely(err < 0)) {
1542out_reject:
1543#ifdef CONFIG_IP_FIB_TRIE_STATS
1544 this_cpu_inc(stats->semantic_match_passed);
1545#endif
1546 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1547 return err;
1548 }
1549 if (fi->fib_flags & RTNH_F_DEAD)
1550 continue;
1551
1552 if (unlikely(fi->nh)) {
1553 if (nexthop_is_blackhole(fi->nh)) {
1554 err = fib_props[RTN_BLACKHOLE].error;
1555 goto out_reject;
1556 }
1557
1558 nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1559 &nhsel);
1560 if (nhc)
1561 goto set_result;
1562 goto miss;
1563 }
1564
1565 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1566 nhc = fib_info_nhc(fi, nhsel);
1567
1568 if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1569 continue;
1570set_result:
1571 if (!(fib_flags & FIB_LOOKUP_NOREF))
1572 refcount_inc(&fi->fib_clntref);
1573
1574 res->prefix = htonl(n->key);
1575 res->prefixlen = KEYLENGTH - fa->fa_slen;
1576 res->nh_sel = nhsel;
1577 res->nhc = nhc;
1578 res->type = fa->fa_type;
1579 res->scope = fi->fib_scope;
1580 res->fi = fi;
1581 res->table = tb;
1582 res->fa_head = &n->leaf;
1583#ifdef CONFIG_IP_FIB_TRIE_STATS
1584 this_cpu_inc(stats->semantic_match_passed);
1585#endif
1586 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1587
1588 return err;
1589 }
1590 }
1591miss:
1592#ifdef CONFIG_IP_FIB_TRIE_STATS
1593 this_cpu_inc(stats->semantic_match_miss);
1594#endif
1595 goto backtrace;
1596}
1597EXPORT_SYMBOL_GPL(fib_table_lookup);
1598
1599static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1600 struct key_vector *l, struct fib_alias *old)
1601{
1602 /* record the location of the previous list_info entry */
1603 struct hlist_node **pprev = old->fa_list.pprev;
1604 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1605
1606 /* remove the fib_alias from the list */
1607 hlist_del_rcu(&old->fa_list);
1608
1609 /* if we emptied the list this leaf will be freed and we can sort
1610 * out parent suffix lengths as a part of trie_rebalance
1611 */
1612 if (hlist_empty(&l->leaf)) {
1613 if (tp->slen == l->slen)
1614 node_pull_suffix(tp, tp->pos);
1615 put_child_root(tp, l->key, NULL);
1616 node_free(l);
1617 trie_rebalance(t, tp);
1618 return;
1619 }
1620
1621 /* only access fa if it is pointing at the last valid hlist_node */
1622 if (*pprev)
1623 return;
1624
1625 /* update the trie with the latest suffix length */
1626 l->slen = fa->fa_slen;
1627 node_pull_suffix(tp, fa->fa_slen);
1628}
1629
1630static void fib_notify_alias_delete(struct net *net, u32 key,
1631 struct hlist_head *fah,
1632 struct fib_alias *fa_to_delete,
1633 struct netlink_ext_ack *extack)
1634{
1635 struct fib_alias *fa_next, *fa_to_notify;
1636 u32 tb_id = fa_to_delete->tb_id;
1637 u8 slen = fa_to_delete->fa_slen;
1638 enum fib_event_type fib_event;
1639
1640 /* Do not notify if we do not care about the route. */
1641 if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1642 return;
1643
1644 /* Determine if the route should be replaced by the next route in the
1645 * list.
1646 */
1647 fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1648 struct fib_alias, fa_list);
1649 if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1650 fib_event = FIB_EVENT_ENTRY_REPLACE;
1651 fa_to_notify = fa_next;
1652 } else {
1653 fib_event = FIB_EVENT_ENTRY_DEL;
1654 fa_to_notify = fa_to_delete;
1655 }
1656 call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1657 fa_to_notify, extack);
1658}
1659
1660/* Caller must hold RTNL. */
1661int fib_table_delete(struct net *net, struct fib_table *tb,
1662 struct fib_config *cfg, struct netlink_ext_ack *extack)
1663{
1664 struct trie *t = (struct trie *) tb->tb_data;
1665 struct fib_alias *fa, *fa_to_delete;
1666 struct key_vector *l, *tp;
1667 u8 plen = cfg->fc_dst_len;
1668 u8 slen = KEYLENGTH - plen;
1669 u8 tos = cfg->fc_tos;
1670 u32 key;
1671
1672 key = ntohl(cfg->fc_dst);
1673
1674 if (!fib_valid_key_len(key, plen, extack))
1675 return -EINVAL;
1676
1677 l = fib_find_node(t, &tp, key);
1678 if (!l)
1679 return -ESRCH;
1680
1681 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id, false);
1682 if (!fa)
1683 return -ESRCH;
1684
1685 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1686
1687 fa_to_delete = NULL;
1688 hlist_for_each_entry_from(fa, fa_list) {
1689 struct fib_info *fi = fa->fa_info;
1690
1691 if ((fa->fa_slen != slen) ||
1692 (fa->tb_id != tb->tb_id) ||
1693 (fa->fa_tos != tos))
1694 break;
1695
1696 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1697 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1698 fa->fa_info->fib_scope == cfg->fc_scope) &&
1699 (!cfg->fc_prefsrc ||
1700 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1701 (!cfg->fc_protocol ||
1702 fi->fib_protocol == cfg->fc_protocol) &&
1703 fib_nh_match(net, cfg, fi, extack) == 0 &&
1704 fib_metrics_match(cfg, fi)) {
1705 fa_to_delete = fa;
1706 break;
1707 }
1708 }
1709
1710 if (!fa_to_delete)
1711 return -ESRCH;
1712
1713 fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1714 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1715 &cfg->fc_nlinfo, 0);
1716
1717 if (!plen)
1718 tb->tb_num_default--;
1719
1720 fib_remove_alias(t, tp, l, fa_to_delete);
1721
1722 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1723 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1724
1725 fib_release_info(fa_to_delete->fa_info);
1726 alias_free_mem_rcu(fa_to_delete);
1727 return 0;
1728}
1729
1730/* Scan for the next leaf starting at the provided key value */
1731static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1732{
1733 struct key_vector *pn, *n = *tn;
1734 unsigned long cindex;
1735
1736 /* this loop is meant to try and find the key in the trie */
1737 do {
1738 /* record parent and next child index */
1739 pn = n;
1740 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1741
1742 if (cindex >> pn->bits)
1743 break;
1744
1745 /* descend into the next child */
1746 n = get_child_rcu(pn, cindex++);
1747 if (!n)
1748 break;
1749
1750 /* guarantee forward progress on the keys */
1751 if (IS_LEAF(n) && (n->key >= key))
1752 goto found;
1753 } while (IS_TNODE(n));
1754
1755 /* this loop will search for the next leaf with a greater key */
1756 while (!IS_TRIE(pn)) {
1757 /* if we exhausted the parent node we will need to climb */
1758 if (cindex >= (1ul << pn->bits)) {
1759 t_key pkey = pn->key;
1760
1761 pn = node_parent_rcu(pn);
1762 cindex = get_index(pkey, pn) + 1;
1763 continue;
1764 }
1765
1766 /* grab the next available node */
1767 n = get_child_rcu(pn, cindex++);
1768 if (!n)
1769 continue;
1770
1771 /* no need to compare keys since we bumped the index */
1772 if (IS_LEAF(n))
1773 goto found;
1774
1775 /* Rescan start scanning in new node */
1776 pn = n;
1777 cindex = 0;
1778 }
1779
1780 *tn = pn;
1781 return NULL; /* Root of trie */
1782found:
1783 /* if we are at the limit for keys just return NULL for the tnode */
1784 *tn = pn;
1785 return n;
1786}
1787
1788static void fib_trie_free(struct fib_table *tb)
1789{
1790 struct trie *t = (struct trie *)tb->tb_data;
1791 struct key_vector *pn = t->kv;
1792 unsigned long cindex = 1;
1793 struct hlist_node *tmp;
1794 struct fib_alias *fa;
1795
1796 /* walk trie in reverse order and free everything */
1797 for (;;) {
1798 struct key_vector *n;
1799
1800 if (!(cindex--)) {
1801 t_key pkey = pn->key;
1802
1803 if (IS_TRIE(pn))
1804 break;
1805
1806 n = pn;
1807 pn = node_parent(pn);
1808
1809 /* drop emptied tnode */
1810 put_child_root(pn, n->key, NULL);
1811 node_free(n);
1812
1813 cindex = get_index(pkey, pn);
1814
1815 continue;
1816 }
1817
1818 /* grab the next available node */
1819 n = get_child(pn, cindex);
1820 if (!n)
1821 continue;
1822
1823 if (IS_TNODE(n)) {
1824 /* record pn and cindex for leaf walking */
1825 pn = n;
1826 cindex = 1ul << n->bits;
1827
1828 continue;
1829 }
1830
1831 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1832 hlist_del_rcu(&fa->fa_list);
1833 alias_free_mem_rcu(fa);
1834 }
1835
1836 put_child_root(pn, n->key, NULL);
1837 node_free(n);
1838 }
1839
1840#ifdef CONFIG_IP_FIB_TRIE_STATS
1841 free_percpu(t->stats);
1842#endif
1843 kfree(tb);
1844}
1845
1846struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1847{
1848 struct trie *ot = (struct trie *)oldtb->tb_data;
1849 struct key_vector *l, *tp = ot->kv;
1850 struct fib_table *local_tb;
1851 struct fib_alias *fa;
1852 struct trie *lt;
1853 t_key key = 0;
1854
1855 if (oldtb->tb_data == oldtb->__data)
1856 return oldtb;
1857
1858 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1859 if (!local_tb)
1860 return NULL;
1861
1862 lt = (struct trie *)local_tb->tb_data;
1863
1864 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1865 struct key_vector *local_l = NULL, *local_tp;
1866
1867 hlist_for_each_entry(fa, &l->leaf, fa_list) {
1868 struct fib_alias *new_fa;
1869
1870 if (local_tb->tb_id != fa->tb_id)
1871 continue;
1872
1873 /* clone fa for new local table */
1874 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1875 if (!new_fa)
1876 goto out;
1877
1878 memcpy(new_fa, fa, sizeof(*fa));
1879
1880 /* insert clone into table */
1881 if (!local_l)
1882 local_l = fib_find_node(lt, &local_tp, l->key);
1883
1884 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1885 NULL, l->key)) {
1886 kmem_cache_free(fn_alias_kmem, new_fa);
1887 goto out;
1888 }
1889 }
1890
1891 /* stop loop if key wrapped back to 0 */
1892 key = l->key + 1;
1893 if (key < l->key)
1894 break;
1895 }
1896
1897 return local_tb;
1898out:
1899 fib_trie_free(local_tb);
1900
1901 return NULL;
1902}
1903
1904/* Caller must hold RTNL */
1905void fib_table_flush_external(struct fib_table *tb)
1906{
1907 struct trie *t = (struct trie *)tb->tb_data;
1908 struct key_vector *pn = t->kv;
1909 unsigned long cindex = 1;
1910 struct hlist_node *tmp;
1911 struct fib_alias *fa;
1912
1913 /* walk trie in reverse order */
1914 for (;;) {
1915 unsigned char slen = 0;
1916 struct key_vector *n;
1917
1918 if (!(cindex--)) {
1919 t_key pkey = pn->key;
1920
1921 /* cannot resize the trie vector */
1922 if (IS_TRIE(pn))
1923 break;
1924
1925 /* update the suffix to address pulled leaves */
1926 if (pn->slen > pn->pos)
1927 update_suffix(pn);
1928
1929 /* resize completed node */
1930 pn = resize(t, pn);
1931 cindex = get_index(pkey, pn);
1932
1933 continue;
1934 }
1935
1936 /* grab the next available node */
1937 n = get_child(pn, cindex);
1938 if (!n)
1939 continue;
1940
1941 if (IS_TNODE(n)) {
1942 /* record pn and cindex for leaf walking */
1943 pn = n;
1944 cindex = 1ul << n->bits;
1945
1946 continue;
1947 }
1948
1949 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1950 /* if alias was cloned to local then we just
1951 * need to remove the local copy from main
1952 */
1953 if (tb->tb_id != fa->tb_id) {
1954 hlist_del_rcu(&fa->fa_list);
1955 alias_free_mem_rcu(fa);
1956 continue;
1957 }
1958
1959 /* record local slen */
1960 slen = fa->fa_slen;
1961 }
1962
1963 /* update leaf slen */
1964 n->slen = slen;
1965
1966 if (hlist_empty(&n->leaf)) {
1967 put_child_root(pn, n->key, NULL);
1968 node_free(n);
1969 }
1970 }
1971}
1972
1973/* Caller must hold RTNL. */
1974int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1975{
1976 struct trie *t = (struct trie *)tb->tb_data;
1977 struct key_vector *pn = t->kv;
1978 unsigned long cindex = 1;
1979 struct hlist_node *tmp;
1980 struct fib_alias *fa;
1981 int found = 0;
1982
1983 /* walk trie in reverse order */
1984 for (;;) {
1985 unsigned char slen = 0;
1986 struct key_vector *n;
1987
1988 if (!(cindex--)) {
1989 t_key pkey = pn->key;
1990
1991 /* cannot resize the trie vector */
1992 if (IS_TRIE(pn))
1993 break;
1994
1995 /* update the suffix to address pulled leaves */
1996 if (pn->slen > pn->pos)
1997 update_suffix(pn);
1998
1999 /* resize completed node */
2000 pn = resize(t, pn);
2001 cindex = get_index(pkey, pn);
2002
2003 continue;
2004 }
2005
2006 /* grab the next available node */
2007 n = get_child(pn, cindex);
2008 if (!n)
2009 continue;
2010
2011 if (IS_TNODE(n)) {
2012 /* record pn and cindex for leaf walking */
2013 pn = n;
2014 cindex = 1ul << n->bits;
2015
2016 continue;
2017 }
2018
2019 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2020 struct fib_info *fi = fa->fa_info;
2021
2022 if (!fi || tb->tb_id != fa->tb_id ||
2023 (!(fi->fib_flags & RTNH_F_DEAD) &&
2024 !fib_props[fa->fa_type].error)) {
2025 slen = fa->fa_slen;
2026 continue;
2027 }
2028
2029 /* Do not flush error routes if network namespace is
2030 * not being dismantled
2031 */
2032 if (!flush_all && fib_props[fa->fa_type].error) {
2033 slen = fa->fa_slen;
2034 continue;
2035 }
2036
2037 fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2038 NULL);
2039 hlist_del_rcu(&fa->fa_list);
2040 fib_release_info(fa->fa_info);
2041 alias_free_mem_rcu(fa);
2042 found++;
2043 }
2044
2045 /* update leaf slen */
2046 n->slen = slen;
2047
2048 if (hlist_empty(&n->leaf)) {
2049 put_child_root(pn, n->key, NULL);
2050 node_free(n);
2051 }
2052 }
2053
2054 pr_debug("trie_flush found=%d\n", found);
2055 return found;
2056}
2057
2058/* derived from fib_trie_free */
2059static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2060 struct nl_info *info)
2061{
2062 struct trie *t = (struct trie *)tb->tb_data;
2063 struct key_vector *pn = t->kv;
2064 unsigned long cindex = 1;
2065 struct fib_alias *fa;
2066
2067 for (;;) {
2068 struct key_vector *n;
2069
2070 if (!(cindex--)) {
2071 t_key pkey = pn->key;
2072
2073 if (IS_TRIE(pn))
2074 break;
2075
2076 pn = node_parent(pn);
2077 cindex = get_index(pkey, pn);
2078 continue;
2079 }
2080
2081 /* grab the next available node */
2082 n = get_child(pn, cindex);
2083 if (!n)
2084 continue;
2085
2086 if (IS_TNODE(n)) {
2087 /* record pn and cindex for leaf walking */
2088 pn = n;
2089 cindex = 1ul << n->bits;
2090
2091 continue;
2092 }
2093
2094 hlist_for_each_entry(fa, &n->leaf, fa_list) {
2095 struct fib_info *fi = fa->fa_info;
2096
2097 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2098 continue;
2099
2100 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2101 KEYLENGTH - fa->fa_slen, tb->tb_id,
2102 info, NLM_F_REPLACE);
2103
2104 /* call_fib_entry_notifiers will be removed when
2105 * in-kernel notifier is implemented and supported
2106 * for nexthop objects
2107 */
2108 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
2109 n->key,
2110 KEYLENGTH - fa->fa_slen, fa,
2111 NULL);
2112 }
2113 }
2114}
2115
2116void fib_info_notify_update(struct net *net, struct nl_info *info)
2117{
2118 unsigned int h;
2119
2120 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2121 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2122 struct fib_table *tb;
2123
2124 hlist_for_each_entry_rcu(tb, head, tb_hlist,
2125 lockdep_rtnl_is_held())
2126 __fib_info_notify_update(net, tb, info);
2127 }
2128}
2129
2130static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2131 struct notifier_block *nb,
2132 struct netlink_ext_ack *extack)
2133{
2134 struct fib_alias *fa;
2135 int last_slen = -1;
2136 int err;
2137
2138 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2139 struct fib_info *fi = fa->fa_info;
2140
2141 if (!fi)
2142 continue;
2143
2144 /* local and main table can share the same trie,
2145 * so don't notify twice for the same entry.
2146 */
2147 if (tb->tb_id != fa->tb_id)
2148 continue;
2149
2150 if (fa->fa_slen == last_slen)
2151 continue;
2152
2153 last_slen = fa->fa_slen;
2154 err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2155 l->key, KEYLENGTH - fa->fa_slen,
2156 fa, extack);
2157 if (err)
2158 return err;
2159 }
2160 return 0;
2161}
2162
2163static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2164 struct netlink_ext_ack *extack)
2165{
2166 struct trie *t = (struct trie *)tb->tb_data;
2167 struct key_vector *l, *tp = t->kv;
2168 t_key key = 0;
2169 int err;
2170
2171 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2172 err = fib_leaf_notify(l, tb, nb, extack);
2173 if (err)
2174 return err;
2175
2176 key = l->key + 1;
2177 /* stop in case of wrap around */
2178 if (key < l->key)
2179 break;
2180 }
2181 return 0;
2182}
2183
2184int fib_notify(struct net *net, struct notifier_block *nb,
2185 struct netlink_ext_ack *extack)
2186{
2187 unsigned int h;
2188 int err;
2189
2190 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2191 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2192 struct fib_table *tb;
2193
2194 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2195 err = fib_table_notify(tb, nb, extack);
2196 if (err)
2197 return err;
2198 }
2199 }
2200 return 0;
2201}
2202
2203static void __trie_free_rcu(struct rcu_head *head)
2204{
2205 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2206#ifdef CONFIG_IP_FIB_TRIE_STATS
2207 struct trie *t = (struct trie *)tb->tb_data;
2208
2209 if (tb->tb_data == tb->__data)
2210 free_percpu(t->stats);
2211#endif /* CONFIG_IP_FIB_TRIE_STATS */
2212 kfree(tb);
2213}
2214
2215void fib_free_table(struct fib_table *tb)
2216{
2217 call_rcu(&tb->rcu, __trie_free_rcu);
2218}
2219
2220static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2221 struct sk_buff *skb, struct netlink_callback *cb,
2222 struct fib_dump_filter *filter)
2223{
2224 unsigned int flags = NLM_F_MULTI;
2225 __be32 xkey = htonl(l->key);
2226 int i, s_i, i_fa, s_fa, err;
2227 struct fib_alias *fa;
2228
2229 if (filter->filter_set ||
2230 !filter->dump_exceptions || !filter->dump_routes)
2231 flags |= NLM_F_DUMP_FILTERED;
2232
2233 s_i = cb->args[4];
2234 s_fa = cb->args[5];
2235 i = 0;
2236
2237 /* rcu_read_lock is hold by caller */
2238 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2239 struct fib_info *fi = fa->fa_info;
2240
2241 if (i < s_i)
2242 goto next;
2243
2244 i_fa = 0;
2245
2246 if (tb->tb_id != fa->tb_id)
2247 goto next;
2248
2249 if (filter->filter_set) {
2250 if (filter->rt_type && fa->fa_type != filter->rt_type)
2251 goto next;
2252
2253 if ((filter->protocol &&
2254 fi->fib_protocol != filter->protocol))
2255 goto next;
2256
2257 if (filter->dev &&
2258 !fib_info_nh_uses_dev(fi, filter->dev))
2259 goto next;
2260 }
2261
2262 if (filter->dump_routes) {
2263 if (!s_fa) {
2264 struct fib_rt_info fri;
2265
2266 fri.fi = fi;
2267 fri.tb_id = tb->tb_id;
2268 fri.dst = xkey;
2269 fri.dst_len = KEYLENGTH - fa->fa_slen;
2270 fri.tos = fa->fa_tos;
2271 fri.type = fa->fa_type;
2272 fri.offload = fa->offload;
2273 fri.trap = fa->trap;
2274 err = fib_dump_info(skb,
2275 NETLINK_CB(cb->skb).portid,
2276 cb->nlh->nlmsg_seq,
2277 RTM_NEWROUTE, &fri, flags);
2278 if (err < 0)
2279 goto stop;
2280 }
2281
2282 i_fa++;
2283 }
2284
2285 if (filter->dump_exceptions) {
2286 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2287 &i_fa, s_fa, flags);
2288 if (err < 0)
2289 goto stop;
2290 }
2291
2292next:
2293 i++;
2294 }
2295
2296 cb->args[4] = i;
2297 return skb->len;
2298
2299stop:
2300 cb->args[4] = i;
2301 cb->args[5] = i_fa;
2302 return err;
2303}
2304
2305/* rcu_read_lock needs to be hold by caller from readside */
2306int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2307 struct netlink_callback *cb, struct fib_dump_filter *filter)
2308{
2309 struct trie *t = (struct trie *)tb->tb_data;
2310 struct key_vector *l, *tp = t->kv;
2311 /* Dump starting at last key.
2312 * Note: 0.0.0.0/0 (ie default) is first key.
2313 */
2314 int count = cb->args[2];
2315 t_key key = cb->args[3];
2316
2317 /* First time here, count and key are both always 0. Count > 0
2318 * and key == 0 means the dump has wrapped around and we are done.
2319 */
2320 if (count && !key)
2321 return skb->len;
2322
2323 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2324 int err;
2325
2326 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2327 if (err < 0) {
2328 cb->args[3] = key;
2329 cb->args[2] = count;
2330 return err;
2331 }
2332
2333 ++count;
2334 key = l->key + 1;
2335
2336 memset(&cb->args[4], 0,
2337 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2338
2339 /* stop loop if key wrapped back to 0 */
2340 if (key < l->key)
2341 break;
2342 }
2343
2344 cb->args[3] = key;
2345 cb->args[2] = count;
2346
2347 return skb->len;
2348}
2349
2350void __init fib_trie_init(void)
2351{
2352 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2353 sizeof(struct fib_alias),
2354 0, SLAB_PANIC, NULL);
2355
2356 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2357 LEAF_SIZE,
2358 0, SLAB_PANIC, NULL);
2359}
2360
2361struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2362{
2363 struct fib_table *tb;
2364 struct trie *t;
2365 size_t sz = sizeof(*tb);
2366
2367 if (!alias)
2368 sz += sizeof(struct trie);
2369
2370 tb = kzalloc(sz, GFP_KERNEL);
2371 if (!tb)
2372 return NULL;
2373
2374 tb->tb_id = id;
2375 tb->tb_num_default = 0;
2376 tb->tb_data = (alias ? alias->__data : tb->__data);
2377
2378 if (alias)
2379 return tb;
2380
2381 t = (struct trie *) tb->tb_data;
2382 t->kv[0].pos = KEYLENGTH;
2383 t->kv[0].slen = KEYLENGTH;
2384#ifdef CONFIG_IP_FIB_TRIE_STATS
2385 t->stats = alloc_percpu(struct trie_use_stats);
2386 if (!t->stats) {
2387 kfree(tb);
2388 tb = NULL;
2389 }
2390#endif
2391
2392 return tb;
2393}
2394
2395#ifdef CONFIG_PROC_FS
2396/* Depth first Trie walk iterator */
2397struct fib_trie_iter {
2398 struct seq_net_private p;
2399 struct fib_table *tb;
2400 struct key_vector *tnode;
2401 unsigned int index;
2402 unsigned int depth;
2403};
2404
2405static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2406{
2407 unsigned long cindex = iter->index;
2408 struct key_vector *pn = iter->tnode;
2409 t_key pkey;
2410
2411 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2412 iter->tnode, iter->index, iter->depth);
2413
2414 while (!IS_TRIE(pn)) {
2415 while (cindex < child_length(pn)) {
2416 struct key_vector *n = get_child_rcu(pn, cindex++);
2417
2418 if (!n)
2419 continue;
2420
2421 if (IS_LEAF(n)) {
2422 iter->tnode = pn;
2423 iter->index = cindex;
2424 } else {
2425 /* push down one level */
2426 iter->tnode = n;
2427 iter->index = 0;
2428 ++iter->depth;
2429 }
2430
2431 return n;
2432 }
2433
2434 /* Current node exhausted, pop back up */
2435 pkey = pn->key;
2436 pn = node_parent_rcu(pn);
2437 cindex = get_index(pkey, pn) + 1;
2438 --iter->depth;
2439 }
2440
2441 /* record root node so further searches know we are done */
2442 iter->tnode = pn;
2443 iter->index = 0;
2444
2445 return NULL;
2446}
2447
2448static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2449 struct trie *t)
2450{
2451 struct key_vector *n, *pn;
2452
2453 if (!t)
2454 return NULL;
2455
2456 pn = t->kv;
2457 n = rcu_dereference(pn->tnode[0]);
2458 if (!n)
2459 return NULL;
2460
2461 if (IS_TNODE(n)) {
2462 iter->tnode = n;
2463 iter->index = 0;
2464 iter->depth = 1;
2465 } else {
2466 iter->tnode = pn;
2467 iter->index = 0;
2468 iter->depth = 0;
2469 }
2470
2471 return n;
2472}
2473
2474static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2475{
2476 struct key_vector *n;
2477 struct fib_trie_iter iter;
2478
2479 memset(s, 0, sizeof(*s));
2480
2481 rcu_read_lock();
2482 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2483 if (IS_LEAF(n)) {
2484 struct fib_alias *fa;
2485
2486 s->leaves++;
2487 s->totdepth += iter.depth;
2488 if (iter.depth > s->maxdepth)
2489 s->maxdepth = iter.depth;
2490
2491 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2492 ++s->prefixes;
2493 } else {
2494 s->tnodes++;
2495 if (n->bits < MAX_STAT_DEPTH)
2496 s->nodesizes[n->bits]++;
2497 s->nullpointers += tn_info(n)->empty_children;
2498 }
2499 }
2500 rcu_read_unlock();
2501}
2502
2503/*
2504 * This outputs /proc/net/fib_triestats
2505 */
2506static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2507{
2508 unsigned int i, max, pointers, bytes, avdepth;
2509
2510 if (stat->leaves)
2511 avdepth = stat->totdepth*100 / stat->leaves;
2512 else
2513 avdepth = 0;
2514
2515 seq_printf(seq, "\tAver depth: %u.%02d\n",
2516 avdepth / 100, avdepth % 100);
2517 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2518
2519 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2520 bytes = LEAF_SIZE * stat->leaves;
2521
2522 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2523 bytes += sizeof(struct fib_alias) * stat->prefixes;
2524
2525 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2526 bytes += TNODE_SIZE(0) * stat->tnodes;
2527
2528 max = MAX_STAT_DEPTH;
2529 while (max > 0 && stat->nodesizes[max-1] == 0)
2530 max--;
2531
2532 pointers = 0;
2533 for (i = 1; i < max; i++)
2534 if (stat->nodesizes[i] != 0) {
2535 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2536 pointers += (1<<i) * stat->nodesizes[i];
2537 }
2538 seq_putc(seq, '\n');
2539 seq_printf(seq, "\tPointers: %u\n", pointers);
2540
2541 bytes += sizeof(struct key_vector *) * pointers;
2542 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2543 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2544}
2545
2546#ifdef CONFIG_IP_FIB_TRIE_STATS
2547static void trie_show_usage(struct seq_file *seq,
2548 const struct trie_use_stats __percpu *stats)
2549{
2550 struct trie_use_stats s = { 0 };
2551 int cpu;
2552
2553 /* loop through all of the CPUs and gather up the stats */
2554 for_each_possible_cpu(cpu) {
2555 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2556
2557 s.gets += pcpu->gets;
2558 s.backtrack += pcpu->backtrack;
2559 s.semantic_match_passed += pcpu->semantic_match_passed;
2560 s.semantic_match_miss += pcpu->semantic_match_miss;
2561 s.null_node_hit += pcpu->null_node_hit;
2562 s.resize_node_skipped += pcpu->resize_node_skipped;
2563 }
2564
2565 seq_printf(seq, "\nCounters:\n---------\n");
2566 seq_printf(seq, "gets = %u\n", s.gets);
2567 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2568 seq_printf(seq, "semantic match passed = %u\n",
2569 s.semantic_match_passed);
2570 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2571 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2572 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2573}
2574#endif /* CONFIG_IP_FIB_TRIE_STATS */
2575
2576static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2577{
2578 if (tb->tb_id == RT_TABLE_LOCAL)
2579 seq_puts(seq, "Local:\n");
2580 else if (tb->tb_id == RT_TABLE_MAIN)
2581 seq_puts(seq, "Main:\n");
2582 else
2583 seq_printf(seq, "Id %d:\n", tb->tb_id);
2584}
2585
2586
2587static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2588{
2589 struct net *net = (struct net *)seq->private;
2590 unsigned int h;
2591
2592 seq_printf(seq,
2593 "Basic info: size of leaf:"
2594 " %zd bytes, size of tnode: %zd bytes.\n",
2595 LEAF_SIZE, TNODE_SIZE(0));
2596
2597 rcu_read_lock();
2598 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2599 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2600 struct fib_table *tb;
2601
2602 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2603 struct trie *t = (struct trie *) tb->tb_data;
2604 struct trie_stat stat;
2605
2606 if (!t)
2607 continue;
2608
2609 fib_table_print(seq, tb);
2610
2611 trie_collect_stats(t, &stat);
2612 trie_show_stats(seq, &stat);
2613#ifdef CONFIG_IP_FIB_TRIE_STATS
2614 trie_show_usage(seq, t->stats);
2615#endif
2616 }
2617 cond_resched_rcu();
2618 }
2619 rcu_read_unlock();
2620
2621 return 0;
2622}
2623
2624static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2625{
2626 struct fib_trie_iter *iter = seq->private;
2627 struct net *net = seq_file_net(seq);
2628 loff_t idx = 0;
2629 unsigned int h;
2630
2631 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2632 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2633 struct fib_table *tb;
2634
2635 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2636 struct key_vector *n;
2637
2638 for (n = fib_trie_get_first(iter,
2639 (struct trie *) tb->tb_data);
2640 n; n = fib_trie_get_next(iter))
2641 if (pos == idx++) {
2642 iter->tb = tb;
2643 return n;
2644 }
2645 }
2646 }
2647
2648 return NULL;
2649}
2650
2651static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2652 __acquires(RCU)
2653{
2654 rcu_read_lock();
2655 return fib_trie_get_idx(seq, *pos);
2656}
2657
2658static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2659{
2660 struct fib_trie_iter *iter = seq->private;
2661 struct net *net = seq_file_net(seq);
2662 struct fib_table *tb = iter->tb;
2663 struct hlist_node *tb_node;
2664 unsigned int h;
2665 struct key_vector *n;
2666
2667 ++*pos;
2668 /* next node in same table */
2669 n = fib_trie_get_next(iter);
2670 if (n)
2671 return n;
2672
2673 /* walk rest of this hash chain */
2674 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2675 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2676 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2677 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2678 if (n)
2679 goto found;
2680 }
2681
2682 /* new hash chain */
2683 while (++h < FIB_TABLE_HASHSZ) {
2684 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2685 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2686 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2687 if (n)
2688 goto found;
2689 }
2690 }
2691 return NULL;
2692
2693found:
2694 iter->tb = tb;
2695 return n;
2696}
2697
2698static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2699 __releases(RCU)
2700{
2701 rcu_read_unlock();
2702}
2703
2704static void seq_indent(struct seq_file *seq, int n)
2705{
2706 while (n-- > 0)
2707 seq_puts(seq, " ");
2708}
2709
2710static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2711{
2712 switch (s) {
2713 case RT_SCOPE_UNIVERSE: return "universe";
2714 case RT_SCOPE_SITE: return "site";
2715 case RT_SCOPE_LINK: return "link";
2716 case RT_SCOPE_HOST: return "host";
2717 case RT_SCOPE_NOWHERE: return "nowhere";
2718 default:
2719 snprintf(buf, len, "scope=%d", s);
2720 return buf;
2721 }
2722}
2723
2724static const char *const rtn_type_names[__RTN_MAX] = {
2725 [RTN_UNSPEC] = "UNSPEC",
2726 [RTN_UNICAST] = "UNICAST",
2727 [RTN_LOCAL] = "LOCAL",
2728 [RTN_BROADCAST] = "BROADCAST",
2729 [RTN_ANYCAST] = "ANYCAST",
2730 [RTN_MULTICAST] = "MULTICAST",
2731 [RTN_BLACKHOLE] = "BLACKHOLE",
2732 [RTN_UNREACHABLE] = "UNREACHABLE",
2733 [RTN_PROHIBIT] = "PROHIBIT",
2734 [RTN_THROW] = "THROW",
2735 [RTN_NAT] = "NAT",
2736 [RTN_XRESOLVE] = "XRESOLVE",
2737};
2738
2739static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2740{
2741 if (t < __RTN_MAX && rtn_type_names[t])
2742 return rtn_type_names[t];
2743 snprintf(buf, len, "type %u", t);
2744 return buf;
2745}
2746
2747/* Pretty print the trie */
2748static int fib_trie_seq_show(struct seq_file *seq, void *v)
2749{
2750 const struct fib_trie_iter *iter = seq->private;
2751 struct key_vector *n = v;
2752
2753 if (IS_TRIE(node_parent_rcu(n)))
2754 fib_table_print(seq, iter->tb);
2755
2756 if (IS_TNODE(n)) {
2757 __be32 prf = htonl(n->key);
2758
2759 seq_indent(seq, iter->depth-1);
2760 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2761 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2762 tn_info(n)->full_children,
2763 tn_info(n)->empty_children);
2764 } else {
2765 __be32 val = htonl(n->key);
2766 struct fib_alias *fa;
2767
2768 seq_indent(seq, iter->depth);
2769 seq_printf(seq, " |-- %pI4\n", &val);
2770
2771 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2772 char buf1[32], buf2[32];
2773
2774 seq_indent(seq, iter->depth + 1);
2775 seq_printf(seq, " /%zu %s %s",
2776 KEYLENGTH - fa->fa_slen,
2777 rtn_scope(buf1, sizeof(buf1),
2778 fa->fa_info->fib_scope),
2779 rtn_type(buf2, sizeof(buf2),
2780 fa->fa_type));
2781 if (fa->fa_tos)
2782 seq_printf(seq, " tos=%d", fa->fa_tos);
2783 seq_putc(seq, '\n');
2784 }
2785 }
2786
2787 return 0;
2788}
2789
2790static const struct seq_operations fib_trie_seq_ops = {
2791 .start = fib_trie_seq_start,
2792 .next = fib_trie_seq_next,
2793 .stop = fib_trie_seq_stop,
2794 .show = fib_trie_seq_show,
2795};
2796
2797struct fib_route_iter {
2798 struct seq_net_private p;
2799 struct fib_table *main_tb;
2800 struct key_vector *tnode;
2801 loff_t pos;
2802 t_key key;
2803};
2804
2805static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2806 loff_t pos)
2807{
2808 struct key_vector *l, **tp = &iter->tnode;
2809 t_key key;
2810
2811 /* use cached location of previously found key */
2812 if (iter->pos > 0 && pos >= iter->pos) {
2813 key = iter->key;
2814 } else {
2815 iter->pos = 1;
2816 key = 0;
2817 }
2818
2819 pos -= iter->pos;
2820
2821 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2822 key = l->key + 1;
2823 iter->pos++;
2824 l = NULL;
2825
2826 /* handle unlikely case of a key wrap */
2827 if (!key)
2828 break;
2829 }
2830
2831 if (l)
2832 iter->key = l->key; /* remember it */
2833 else
2834 iter->pos = 0; /* forget it */
2835
2836 return l;
2837}
2838
2839static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2840 __acquires(RCU)
2841{
2842 struct fib_route_iter *iter = seq->private;
2843 struct fib_table *tb;
2844 struct trie *t;
2845
2846 rcu_read_lock();
2847
2848 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2849 if (!tb)
2850 return NULL;
2851
2852 iter->main_tb = tb;
2853 t = (struct trie *)tb->tb_data;
2854 iter->tnode = t->kv;
2855
2856 if (*pos != 0)
2857 return fib_route_get_idx(iter, *pos);
2858
2859 iter->pos = 0;
2860 iter->key = KEY_MAX;
2861
2862 return SEQ_START_TOKEN;
2863}
2864
2865static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2866{
2867 struct fib_route_iter *iter = seq->private;
2868 struct key_vector *l = NULL;
2869 t_key key = iter->key + 1;
2870
2871 ++*pos;
2872
2873 /* only allow key of 0 for start of sequence */
2874 if ((v == SEQ_START_TOKEN) || key)
2875 l = leaf_walk_rcu(&iter->tnode, key);
2876
2877 if (l) {
2878 iter->key = l->key;
2879 iter->pos++;
2880 } else {
2881 iter->pos = 0;
2882 }
2883
2884 return l;
2885}
2886
2887static void fib_route_seq_stop(struct seq_file *seq, void *v)
2888 __releases(RCU)
2889{
2890 rcu_read_unlock();
2891}
2892
2893static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2894{
2895 unsigned int flags = 0;
2896
2897 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2898 flags = RTF_REJECT;
2899 if (fi) {
2900 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2901
2902 if (nhc->nhc_gw.ipv4)
2903 flags |= RTF_GATEWAY;
2904 }
2905 if (mask == htonl(0xFFFFFFFF))
2906 flags |= RTF_HOST;
2907 flags |= RTF_UP;
2908 return flags;
2909}
2910
2911/*
2912 * This outputs /proc/net/route.
2913 * The format of the file is not supposed to be changed
2914 * and needs to be same as fib_hash output to avoid breaking
2915 * legacy utilities
2916 */
2917static int fib_route_seq_show(struct seq_file *seq, void *v)
2918{
2919 struct fib_route_iter *iter = seq->private;
2920 struct fib_table *tb = iter->main_tb;
2921 struct fib_alias *fa;
2922 struct key_vector *l = v;
2923 __be32 prefix;
2924
2925 if (v == SEQ_START_TOKEN) {
2926 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2927 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2928 "\tWindow\tIRTT");
2929 return 0;
2930 }
2931
2932 prefix = htonl(l->key);
2933
2934 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2935 struct fib_info *fi = fa->fa_info;
2936 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2937 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2938
2939 if ((fa->fa_type == RTN_BROADCAST) ||
2940 (fa->fa_type == RTN_MULTICAST))
2941 continue;
2942
2943 if (fa->tb_id != tb->tb_id)
2944 continue;
2945
2946 seq_setwidth(seq, 127);
2947
2948 if (fi) {
2949 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2950 __be32 gw = 0;
2951
2952 if (nhc->nhc_gw_family == AF_INET)
2953 gw = nhc->nhc_gw.ipv4;
2954
2955 seq_printf(seq,
2956 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2957 "%d\t%08X\t%d\t%u\t%u",
2958 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2959 prefix, gw, flags, 0, 0,
2960 fi->fib_priority,
2961 mask,
2962 (fi->fib_advmss ?
2963 fi->fib_advmss + 40 : 0),
2964 fi->fib_window,
2965 fi->fib_rtt >> 3);
2966 } else {
2967 seq_printf(seq,
2968 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2969 "%d\t%08X\t%d\t%u\t%u",
2970 prefix, 0, flags, 0, 0, 0,
2971 mask, 0, 0, 0);
2972 }
2973 seq_pad(seq, '\n');
2974 }
2975
2976 return 0;
2977}
2978
2979static const struct seq_operations fib_route_seq_ops = {
2980 .start = fib_route_seq_start,
2981 .next = fib_route_seq_next,
2982 .stop = fib_route_seq_stop,
2983 .show = fib_route_seq_show,
2984};
2985
2986int __net_init fib_proc_init(struct net *net)
2987{
2988 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2989 sizeof(struct fib_trie_iter)))
2990 goto out1;
2991
2992 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
2993 fib_triestat_seq_show, NULL))
2994 goto out2;
2995
2996 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
2997 sizeof(struct fib_route_iter)))
2998 goto out3;
2999
3000 return 0;
3001
3002out3:
3003 remove_proc_entry("fib_triestat", net->proc_net);
3004out2:
3005 remove_proc_entry("fib_trie", net->proc_net);
3006out1:
3007 return -ENOMEM;
3008}
3009
3010void __net_exit fib_proc_exit(struct net *net)
3011{
3012 remove_proc_entry("fib_trie", net->proc_net);
3013 remove_proc_entry("fib_triestat", net->proc_net);
3014 remove_proc_entry("route", net->proc_net);
3015}
3016
3017#endif /* CONFIG_PROC_FS */