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