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 pn = t->kv;
1330 cindex = 0;
1331
1332 n = get_child_rcu(pn, cindex);
1333 if (!n) {
1334 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1335 return -EAGAIN;
1336 }
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 trace_fib_table_lookup(tb->tb_id, flp,
1421 NULL, -EAGAIN);
1422 return -EAGAIN;
1423 }
1424#ifdef CONFIG_IP_FIB_TRIE_STATS
1425 this_cpu_inc(stats->backtrack);
1426#endif
1427 /* Get Child's index */
1428 pn = node_parent_rcu(pn);
1429 cindex = get_index(pkey, pn);
1430 }
1431
1432 /* strip the least significant bit from the cindex */
1433 cindex &= cindex - 1;
1434
1435 /* grab pointer for next child node */
1436 cptr = &pn->tnode[cindex];
1437 }
1438 }
1439
1440found:
1441 /* this line carries forward the xor from earlier in the function */
1442 index = key ^ n->key;
1443
1444 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1445 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1446 struct fib_info *fi = fa->fa_info;
1447 int nhsel, err;
1448
1449 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1450 if (index >= (1ul << fa->fa_slen))
1451 continue;
1452 }
1453 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1454 continue;
1455 if (fi->fib_dead)
1456 continue;
1457 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1458 continue;
1459 fib_alias_accessed(fa);
1460 err = fib_props[fa->fa_type].error;
1461 if (unlikely(err < 0)) {
1462#ifdef CONFIG_IP_FIB_TRIE_STATS
1463 this_cpu_inc(stats->semantic_match_passed);
1464#endif
1465 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1466 return err;
1467 }
1468 if (fi->fib_flags & RTNH_F_DEAD)
1469 continue;
1470 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1471 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1472 struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1473
1474 if (nh->nh_flags & RTNH_F_DEAD)
1475 continue;
1476 if (in_dev &&
1477 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1478 nh->nh_flags & RTNH_F_LINKDOWN &&
1479 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1480 continue;
1481 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1482 if (flp->flowi4_oif &&
1483 flp->flowi4_oif != nh->nh_oif)
1484 continue;
1485 }
1486
1487 if (!(fib_flags & FIB_LOOKUP_NOREF))
1488 refcount_inc(&fi->fib_clntref);
1489
1490 res->prefix = htonl(n->key);
1491 res->prefixlen = KEYLENGTH - fa->fa_slen;
1492 res->nh_sel = nhsel;
1493 res->type = fa->fa_type;
1494 res->scope = fi->fib_scope;
1495 res->fi = fi;
1496 res->table = tb;
1497 res->fa_head = &n->leaf;
1498#ifdef CONFIG_IP_FIB_TRIE_STATS
1499 this_cpu_inc(stats->semantic_match_passed);
1500#endif
1501 trace_fib_table_lookup(tb->tb_id, flp, nh, err);
1502
1503 return err;
1504 }
1505 }
1506#ifdef CONFIG_IP_FIB_TRIE_STATS
1507 this_cpu_inc(stats->semantic_match_miss);
1508#endif
1509 goto backtrace;
1510}
1511EXPORT_SYMBOL_GPL(fib_table_lookup);
1512
1513static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1514 struct key_vector *l, struct fib_alias *old)
1515{
1516 /* record the location of the previous list_info entry */
1517 struct hlist_node **pprev = old->fa_list.pprev;
1518 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1519
1520 /* remove the fib_alias from the list */
1521 hlist_del_rcu(&old->fa_list);
1522
1523 /* if we emptied the list this leaf will be freed and we can sort
1524 * out parent suffix lengths as a part of trie_rebalance
1525 */
1526 if (hlist_empty(&l->leaf)) {
1527 if (tp->slen == l->slen)
1528 node_pull_suffix(tp, tp->pos);
1529 put_child_root(tp, l->key, NULL);
1530 node_free(l);
1531 trie_rebalance(t, tp);
1532 return;
1533 }
1534
1535 /* only access fa if it is pointing at the last valid hlist_node */
1536 if (*pprev)
1537 return;
1538
1539 /* update the trie with the latest suffix length */
1540 l->slen = fa->fa_slen;
1541 node_pull_suffix(tp, fa->fa_slen);
1542}
1543
1544/* Caller must hold RTNL. */
1545int fib_table_delete(struct net *net, struct fib_table *tb,
1546 struct fib_config *cfg, struct netlink_ext_ack *extack)
1547{
1548 struct trie *t = (struct trie *) tb->tb_data;
1549 struct fib_alias *fa, *fa_to_delete;
1550 struct key_vector *l, *tp;
1551 u8 plen = cfg->fc_dst_len;
1552 u8 slen = KEYLENGTH - plen;
1553 u8 tos = cfg->fc_tos;
1554 u32 key;
1555
1556 key = ntohl(cfg->fc_dst);
1557
1558 if (!fib_valid_key_len(key, plen, extack))
1559 return -EINVAL;
1560
1561 l = fib_find_node(t, &tp, key);
1562 if (!l)
1563 return -ESRCH;
1564
1565 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1566 if (!fa)
1567 return -ESRCH;
1568
1569 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1570
1571 fa_to_delete = NULL;
1572 hlist_for_each_entry_from(fa, fa_list) {
1573 struct fib_info *fi = fa->fa_info;
1574
1575 if ((fa->fa_slen != slen) ||
1576 (fa->tb_id != tb->tb_id) ||
1577 (fa->fa_tos != tos))
1578 break;
1579
1580 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1581 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1582 fa->fa_info->fib_scope == cfg->fc_scope) &&
1583 (!cfg->fc_prefsrc ||
1584 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1585 (!cfg->fc_protocol ||
1586 fi->fib_protocol == cfg->fc_protocol) &&
1587 fib_nh_match(cfg, fi, extack) == 0 &&
1588 fib_metrics_match(cfg, fi)) {
1589 fa_to_delete = fa;
1590 break;
1591 }
1592 }
1593
1594 if (!fa_to_delete)
1595 return -ESRCH;
1596
1597 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1598 fa_to_delete, extack);
1599 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1600 &cfg->fc_nlinfo, 0);
1601
1602 if (!plen)
1603 tb->tb_num_default--;
1604
1605 fib_remove_alias(t, tp, l, fa_to_delete);
1606
1607 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1608 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1609
1610 fib_release_info(fa_to_delete->fa_info);
1611 alias_free_mem_rcu(fa_to_delete);
1612 return 0;
1613}
1614
1615/* Scan for the next leaf starting at the provided key value */
1616static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1617{
1618 struct key_vector *pn, *n = *tn;
1619 unsigned long cindex;
1620
1621 /* this loop is meant to try and find the key in the trie */
1622 do {
1623 /* record parent and next child index */
1624 pn = n;
1625 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1626
1627 if (cindex >> pn->bits)
1628 break;
1629
1630 /* descend into the next child */
1631 n = get_child_rcu(pn, cindex++);
1632 if (!n)
1633 break;
1634
1635 /* guarantee forward progress on the keys */
1636 if (IS_LEAF(n) && (n->key >= key))
1637 goto found;
1638 } while (IS_TNODE(n));
1639
1640 /* this loop will search for the next leaf with a greater key */
1641 while (!IS_TRIE(pn)) {
1642 /* if we exhausted the parent node we will need to climb */
1643 if (cindex >= (1ul << pn->bits)) {
1644 t_key pkey = pn->key;
1645
1646 pn = node_parent_rcu(pn);
1647 cindex = get_index(pkey, pn) + 1;
1648 continue;
1649 }
1650
1651 /* grab the next available node */
1652 n = get_child_rcu(pn, cindex++);
1653 if (!n)
1654 continue;
1655
1656 /* no need to compare keys since we bumped the index */
1657 if (IS_LEAF(n))
1658 goto found;
1659
1660 /* Rescan start scanning in new node */
1661 pn = n;
1662 cindex = 0;
1663 }
1664
1665 *tn = pn;
1666 return NULL; /* Root of trie */
1667found:
1668 /* if we are at the limit for keys just return NULL for the tnode */
1669 *tn = pn;
1670 return n;
1671}
1672
1673static void fib_trie_free(struct fib_table *tb)
1674{
1675 struct trie *t = (struct trie *)tb->tb_data;
1676 struct key_vector *pn = t->kv;
1677 unsigned long cindex = 1;
1678 struct hlist_node *tmp;
1679 struct fib_alias *fa;
1680
1681 /* walk trie in reverse order and free everything */
1682 for (;;) {
1683 struct key_vector *n;
1684
1685 if (!(cindex--)) {
1686 t_key pkey = pn->key;
1687
1688 if (IS_TRIE(pn))
1689 break;
1690
1691 n = pn;
1692 pn = node_parent(pn);
1693
1694 /* drop emptied tnode */
1695 put_child_root(pn, n->key, NULL);
1696 node_free(n);
1697
1698 cindex = get_index(pkey, pn);
1699
1700 continue;
1701 }
1702
1703 /* grab the next available node */
1704 n = get_child(pn, cindex);
1705 if (!n)
1706 continue;
1707
1708 if (IS_TNODE(n)) {
1709 /* record pn and cindex for leaf walking */
1710 pn = n;
1711 cindex = 1ul << n->bits;
1712
1713 continue;
1714 }
1715
1716 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1717 hlist_del_rcu(&fa->fa_list);
1718 alias_free_mem_rcu(fa);
1719 }
1720
1721 put_child_root(pn, n->key, NULL);
1722 node_free(n);
1723 }
1724
1725#ifdef CONFIG_IP_FIB_TRIE_STATS
1726 free_percpu(t->stats);
1727#endif
1728 kfree(tb);
1729}
1730
1731struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1732{
1733 struct trie *ot = (struct trie *)oldtb->tb_data;
1734 struct key_vector *l, *tp = ot->kv;
1735 struct fib_table *local_tb;
1736 struct fib_alias *fa;
1737 struct trie *lt;
1738 t_key key = 0;
1739
1740 if (oldtb->tb_data == oldtb->__data)
1741 return oldtb;
1742
1743 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1744 if (!local_tb)
1745 return NULL;
1746
1747 lt = (struct trie *)local_tb->tb_data;
1748
1749 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1750 struct key_vector *local_l = NULL, *local_tp;
1751
1752 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1753 struct fib_alias *new_fa;
1754
1755 if (local_tb->tb_id != fa->tb_id)
1756 continue;
1757
1758 /* clone fa for new local table */
1759 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1760 if (!new_fa)
1761 goto out;
1762
1763 memcpy(new_fa, fa, sizeof(*fa));
1764
1765 /* insert clone into table */
1766 if (!local_l)
1767 local_l = fib_find_node(lt, &local_tp, l->key);
1768
1769 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1770 NULL, l->key)) {
1771 kmem_cache_free(fn_alias_kmem, new_fa);
1772 goto out;
1773 }
1774 }
1775
1776 /* stop loop if key wrapped back to 0 */
1777 key = l->key + 1;
1778 if (key < l->key)
1779 break;
1780 }
1781
1782 return local_tb;
1783out:
1784 fib_trie_free(local_tb);
1785
1786 return NULL;
1787}
1788
1789/* Caller must hold RTNL */
1790void fib_table_flush_external(struct fib_table *tb)
1791{
1792 struct trie *t = (struct trie *)tb->tb_data;
1793 struct key_vector *pn = t->kv;
1794 unsigned long cindex = 1;
1795 struct hlist_node *tmp;
1796 struct fib_alias *fa;
1797
1798 /* walk trie in reverse order */
1799 for (;;) {
1800 unsigned char slen = 0;
1801 struct key_vector *n;
1802
1803 if (!(cindex--)) {
1804 t_key pkey = pn->key;
1805
1806 /* cannot resize the trie vector */
1807 if (IS_TRIE(pn))
1808 break;
1809
1810 /* update the suffix to address pulled leaves */
1811 if (pn->slen > pn->pos)
1812 update_suffix(pn);
1813
1814 /* resize completed node */
1815 pn = resize(t, pn);
1816 cindex = get_index(pkey, pn);
1817
1818 continue;
1819 }
1820
1821 /* grab the next available node */
1822 n = get_child(pn, cindex);
1823 if (!n)
1824 continue;
1825
1826 if (IS_TNODE(n)) {
1827 /* record pn and cindex for leaf walking */
1828 pn = n;
1829 cindex = 1ul << n->bits;
1830
1831 continue;
1832 }
1833
1834 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1835 /* if alias was cloned to local then we just
1836 * need to remove the local copy from main
1837 */
1838 if (tb->tb_id != fa->tb_id) {
1839 hlist_del_rcu(&fa->fa_list);
1840 alias_free_mem_rcu(fa);
1841 continue;
1842 }
1843
1844 /* record local slen */
1845 slen = fa->fa_slen;
1846 }
1847
1848 /* update leaf slen */
1849 n->slen = slen;
1850
1851 if (hlist_empty(&n->leaf)) {
1852 put_child_root(pn, n->key, NULL);
1853 node_free(n);
1854 }
1855 }
1856}
1857
1858/* Caller must hold RTNL. */
1859int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1860{
1861 struct trie *t = (struct trie *)tb->tb_data;
1862 struct key_vector *pn = t->kv;
1863 unsigned long cindex = 1;
1864 struct hlist_node *tmp;
1865 struct fib_alias *fa;
1866 int found = 0;
1867
1868 /* walk trie in reverse order */
1869 for (;;) {
1870 unsigned char slen = 0;
1871 struct key_vector *n;
1872
1873 if (!(cindex--)) {
1874 t_key pkey = pn->key;
1875
1876 /* cannot resize the trie vector */
1877 if (IS_TRIE(pn))
1878 break;
1879
1880 /* update the suffix to address pulled leaves */
1881 if (pn->slen > pn->pos)
1882 update_suffix(pn);
1883
1884 /* resize completed node */
1885 pn = resize(t, pn);
1886 cindex = get_index(pkey, pn);
1887
1888 continue;
1889 }
1890
1891 /* grab the next available node */
1892 n = get_child(pn, cindex);
1893 if (!n)
1894 continue;
1895
1896 if (IS_TNODE(n)) {
1897 /* record pn and cindex for leaf walking */
1898 pn = n;
1899 cindex = 1ul << n->bits;
1900
1901 continue;
1902 }
1903
1904 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1905 struct fib_info *fi = fa->fa_info;
1906
1907 if (!fi || tb->tb_id != fa->tb_id ||
1908 (!(fi->fib_flags & RTNH_F_DEAD) &&
1909 !fib_props[fa->fa_type].error)) {
1910 slen = fa->fa_slen;
1911 continue;
1912 }
1913
1914 /* Do not flush error routes if network namespace is
1915 * not being dismantled
1916 */
1917 if (!flush_all && fib_props[fa->fa_type].error) {
1918 slen = fa->fa_slen;
1919 continue;
1920 }
1921
1922 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1923 n->key,
1924 KEYLENGTH - fa->fa_slen, fa,
1925 NULL);
1926 hlist_del_rcu(&fa->fa_list);
1927 fib_release_info(fa->fa_info);
1928 alias_free_mem_rcu(fa);
1929 found++;
1930 }
1931
1932 /* update leaf slen */
1933 n->slen = slen;
1934
1935 if (hlist_empty(&n->leaf)) {
1936 put_child_root(pn, n->key, NULL);
1937 node_free(n);
1938 }
1939 }
1940
1941 pr_debug("trie_flush found=%d\n", found);
1942 return found;
1943}
1944
1945static void fib_leaf_notify(struct net *net, struct key_vector *l,
1946 struct fib_table *tb, struct notifier_block *nb)
1947{
1948 struct fib_alias *fa;
1949
1950 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1951 struct fib_info *fi = fa->fa_info;
1952
1953 if (!fi)
1954 continue;
1955
1956 /* local and main table can share the same trie,
1957 * so don't notify twice for the same entry.
1958 */
1959 if (tb->tb_id != fa->tb_id)
1960 continue;
1961
1962 call_fib_entry_notifier(nb, net, FIB_EVENT_ENTRY_ADD, l->key,
1963 KEYLENGTH - fa->fa_slen, fa);
1964 }
1965}
1966
1967static void fib_table_notify(struct net *net, struct fib_table *tb,
1968 struct notifier_block *nb)
1969{
1970 struct trie *t = (struct trie *)tb->tb_data;
1971 struct key_vector *l, *tp = t->kv;
1972 t_key key = 0;
1973
1974 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1975 fib_leaf_notify(net, l, tb, nb);
1976
1977 key = l->key + 1;
1978 /* stop in case of wrap around */
1979 if (key < l->key)
1980 break;
1981 }
1982}
1983
1984void fib_notify(struct net *net, struct notifier_block *nb)
1985{
1986 unsigned int h;
1987
1988 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1989 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
1990 struct fib_table *tb;
1991
1992 hlist_for_each_entry_rcu(tb, head, tb_hlist)
1993 fib_table_notify(net, tb, nb);
1994 }
1995}
1996
1997static void __trie_free_rcu(struct rcu_head *head)
1998{
1999 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2000#ifdef CONFIG_IP_FIB_TRIE_STATS
2001 struct trie *t = (struct trie *)tb->tb_data;
2002
2003 if (tb->tb_data == tb->__data)
2004 free_percpu(t->stats);
2005#endif /* CONFIG_IP_FIB_TRIE_STATS */
2006 kfree(tb);
2007}
2008
2009void fib_free_table(struct fib_table *tb)
2010{
2011 call_rcu(&tb->rcu, __trie_free_rcu);
2012}
2013
2014static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2015 struct sk_buff *skb, struct netlink_callback *cb,
2016 struct fib_dump_filter *filter)
2017{
2018 unsigned int flags = NLM_F_MULTI;
2019 __be32 xkey = htonl(l->key);
2020 struct fib_alias *fa;
2021 int i, s_i;
2022
2023 if (filter->filter_set)
2024 flags |= NLM_F_DUMP_FILTERED;
2025
2026 s_i = cb->args[4];
2027 i = 0;
2028
2029 /* rcu_read_lock is hold by caller */
2030 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2031 int err;
2032
2033 if (i < s_i)
2034 goto next;
2035
2036 if (tb->tb_id != fa->tb_id)
2037 goto next;
2038
2039 if (filter->filter_set) {
2040 if (filter->rt_type && fa->fa_type != filter->rt_type)
2041 goto next;
2042
2043 if ((filter->protocol &&
2044 fa->fa_info->fib_protocol != filter->protocol))
2045 goto next;
2046
2047 if (filter->dev &&
2048 !fib_info_nh_uses_dev(fa->fa_info, filter->dev))
2049 goto next;
2050 }
2051
2052 err = fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
2053 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2054 tb->tb_id, fa->fa_type,
2055 xkey, KEYLENGTH - fa->fa_slen,
2056 fa->fa_tos, fa->fa_info, flags);
2057 if (err < 0) {
2058 cb->args[4] = i;
2059 return err;
2060 }
2061next:
2062 i++;
2063 }
2064
2065 cb->args[4] = i;
2066 return skb->len;
2067}
2068
2069/* rcu_read_lock needs to be hold by caller from readside */
2070int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2071 struct netlink_callback *cb, struct fib_dump_filter *filter)
2072{
2073 struct trie *t = (struct trie *)tb->tb_data;
2074 struct key_vector *l, *tp = t->kv;
2075 /* Dump starting at last key.
2076 * Note: 0.0.0.0/0 (ie default) is first key.
2077 */
2078 int count = cb->args[2];
2079 t_key key = cb->args[3];
2080
2081 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2082 int err;
2083
2084 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2085 if (err < 0) {
2086 cb->args[3] = key;
2087 cb->args[2] = count;
2088 return err;
2089 }
2090
2091 ++count;
2092 key = l->key + 1;
2093
2094 memset(&cb->args[4], 0,
2095 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2096
2097 /* stop loop if key wrapped back to 0 */
2098 if (key < l->key)
2099 break;
2100 }
2101
2102 cb->args[3] = key;
2103 cb->args[2] = count;
2104
2105 return skb->len;
2106}
2107
2108void __init fib_trie_init(void)
2109{
2110 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2111 sizeof(struct fib_alias),
2112 0, SLAB_PANIC, NULL);
2113
2114 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2115 LEAF_SIZE,
2116 0, SLAB_PANIC, NULL);
2117}
2118
2119struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2120{
2121 struct fib_table *tb;
2122 struct trie *t;
2123 size_t sz = sizeof(*tb);
2124
2125 if (!alias)
2126 sz += sizeof(struct trie);
2127
2128 tb = kzalloc(sz, GFP_KERNEL);
2129 if (!tb)
2130 return NULL;
2131
2132 tb->tb_id = id;
2133 tb->tb_num_default = 0;
2134 tb->tb_data = (alias ? alias->__data : tb->__data);
2135
2136 if (alias)
2137 return tb;
2138
2139 t = (struct trie *) tb->tb_data;
2140 t->kv[0].pos = KEYLENGTH;
2141 t->kv[0].slen = KEYLENGTH;
2142#ifdef CONFIG_IP_FIB_TRIE_STATS
2143 t->stats = alloc_percpu(struct trie_use_stats);
2144 if (!t->stats) {
2145 kfree(tb);
2146 tb = NULL;
2147 }
2148#endif
2149
2150 return tb;
2151}
2152
2153#ifdef CONFIG_PROC_FS
2154/* Depth first Trie walk iterator */
2155struct fib_trie_iter {
2156 struct seq_net_private p;
2157 struct fib_table *tb;
2158 struct key_vector *tnode;
2159 unsigned int index;
2160 unsigned int depth;
2161};
2162
2163static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2164{
2165 unsigned long cindex = iter->index;
2166 struct key_vector *pn = iter->tnode;
2167 t_key pkey;
2168
2169 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2170 iter->tnode, iter->index, iter->depth);
2171
2172 while (!IS_TRIE(pn)) {
2173 while (cindex < child_length(pn)) {
2174 struct key_vector *n = get_child_rcu(pn, cindex++);
2175
2176 if (!n)
2177 continue;
2178
2179 if (IS_LEAF(n)) {
2180 iter->tnode = pn;
2181 iter->index = cindex;
2182 } else {
2183 /* push down one level */
2184 iter->tnode = n;
2185 iter->index = 0;
2186 ++iter->depth;
2187 }
2188
2189 return n;
2190 }
2191
2192 /* Current node exhausted, pop back up */
2193 pkey = pn->key;
2194 pn = node_parent_rcu(pn);
2195 cindex = get_index(pkey, pn) + 1;
2196 --iter->depth;
2197 }
2198
2199 /* record root node so further searches know we are done */
2200 iter->tnode = pn;
2201 iter->index = 0;
2202
2203 return NULL;
2204}
2205
2206static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2207 struct trie *t)
2208{
2209 struct key_vector *n, *pn;
2210
2211 if (!t)
2212 return NULL;
2213
2214 pn = t->kv;
2215 n = rcu_dereference(pn->tnode[0]);
2216 if (!n)
2217 return NULL;
2218
2219 if (IS_TNODE(n)) {
2220 iter->tnode = n;
2221 iter->index = 0;
2222 iter->depth = 1;
2223 } else {
2224 iter->tnode = pn;
2225 iter->index = 0;
2226 iter->depth = 0;
2227 }
2228
2229 return n;
2230}
2231
2232static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2233{
2234 struct key_vector *n;
2235 struct fib_trie_iter iter;
2236
2237 memset(s, 0, sizeof(*s));
2238
2239 rcu_read_lock();
2240 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2241 if (IS_LEAF(n)) {
2242 struct fib_alias *fa;
2243
2244 s->leaves++;
2245 s->totdepth += iter.depth;
2246 if (iter.depth > s->maxdepth)
2247 s->maxdepth = iter.depth;
2248
2249 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2250 ++s->prefixes;
2251 } else {
2252 s->tnodes++;
2253 if (n->bits < MAX_STAT_DEPTH)
2254 s->nodesizes[n->bits]++;
2255 s->nullpointers += tn_info(n)->empty_children;
2256 }
2257 }
2258 rcu_read_unlock();
2259}
2260
2261/*
2262 * This outputs /proc/net/fib_triestats
2263 */
2264static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2265{
2266 unsigned int i, max, pointers, bytes, avdepth;
2267
2268 if (stat->leaves)
2269 avdepth = stat->totdepth*100 / stat->leaves;
2270 else
2271 avdepth = 0;
2272
2273 seq_printf(seq, "\tAver depth: %u.%02d\n",
2274 avdepth / 100, avdepth % 100);
2275 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2276
2277 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2278 bytes = LEAF_SIZE * stat->leaves;
2279
2280 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2281 bytes += sizeof(struct fib_alias) * stat->prefixes;
2282
2283 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2284 bytes += TNODE_SIZE(0) * stat->tnodes;
2285
2286 max = MAX_STAT_DEPTH;
2287 while (max > 0 && stat->nodesizes[max-1] == 0)
2288 max--;
2289
2290 pointers = 0;
2291 for (i = 1; i < max; i++)
2292 if (stat->nodesizes[i] != 0) {
2293 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2294 pointers += (1<<i) * stat->nodesizes[i];
2295 }
2296 seq_putc(seq, '\n');
2297 seq_printf(seq, "\tPointers: %u\n", pointers);
2298
2299 bytes += sizeof(struct key_vector *) * pointers;
2300 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2301 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2302}
2303
2304#ifdef CONFIG_IP_FIB_TRIE_STATS
2305static void trie_show_usage(struct seq_file *seq,
2306 const struct trie_use_stats __percpu *stats)
2307{
2308 struct trie_use_stats s = { 0 };
2309 int cpu;
2310
2311 /* loop through all of the CPUs and gather up the stats */
2312 for_each_possible_cpu(cpu) {
2313 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2314
2315 s.gets += pcpu->gets;
2316 s.backtrack += pcpu->backtrack;
2317 s.semantic_match_passed += pcpu->semantic_match_passed;
2318 s.semantic_match_miss += pcpu->semantic_match_miss;
2319 s.null_node_hit += pcpu->null_node_hit;
2320 s.resize_node_skipped += pcpu->resize_node_skipped;
2321 }
2322
2323 seq_printf(seq, "\nCounters:\n---------\n");
2324 seq_printf(seq, "gets = %u\n", s.gets);
2325 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2326 seq_printf(seq, "semantic match passed = %u\n",
2327 s.semantic_match_passed);
2328 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2329 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2330 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2331}
2332#endif /* CONFIG_IP_FIB_TRIE_STATS */
2333
2334static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2335{
2336 if (tb->tb_id == RT_TABLE_LOCAL)
2337 seq_puts(seq, "Local:\n");
2338 else if (tb->tb_id == RT_TABLE_MAIN)
2339 seq_puts(seq, "Main:\n");
2340 else
2341 seq_printf(seq, "Id %d:\n", tb->tb_id);
2342}
2343
2344
2345static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2346{
2347 struct net *net = (struct net *)seq->private;
2348 unsigned int h;
2349
2350 seq_printf(seq,
2351 "Basic info: size of leaf:"
2352 " %zd bytes, size of tnode: %zd bytes.\n",
2353 LEAF_SIZE, TNODE_SIZE(0));
2354
2355 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2356 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2357 struct fib_table *tb;
2358
2359 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2360 struct trie *t = (struct trie *) tb->tb_data;
2361 struct trie_stat stat;
2362
2363 if (!t)
2364 continue;
2365
2366 fib_table_print(seq, tb);
2367
2368 trie_collect_stats(t, &stat);
2369 trie_show_stats(seq, &stat);
2370#ifdef CONFIG_IP_FIB_TRIE_STATS
2371 trie_show_usage(seq, t->stats);
2372#endif
2373 }
2374 }
2375
2376 return 0;
2377}
2378
2379static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2380{
2381 struct fib_trie_iter *iter = seq->private;
2382 struct net *net = seq_file_net(seq);
2383 loff_t idx = 0;
2384 unsigned int h;
2385
2386 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2387 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2388 struct fib_table *tb;
2389
2390 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2391 struct key_vector *n;
2392
2393 for (n = fib_trie_get_first(iter,
2394 (struct trie *) tb->tb_data);
2395 n; n = fib_trie_get_next(iter))
2396 if (pos == idx++) {
2397 iter->tb = tb;
2398 return n;
2399 }
2400 }
2401 }
2402
2403 return NULL;
2404}
2405
2406static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2407 __acquires(RCU)
2408{
2409 rcu_read_lock();
2410 return fib_trie_get_idx(seq, *pos);
2411}
2412
2413static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2414{
2415 struct fib_trie_iter *iter = seq->private;
2416 struct net *net = seq_file_net(seq);
2417 struct fib_table *tb = iter->tb;
2418 struct hlist_node *tb_node;
2419 unsigned int h;
2420 struct key_vector *n;
2421
2422 ++*pos;
2423 /* next node in same table */
2424 n = fib_trie_get_next(iter);
2425 if (n)
2426 return n;
2427
2428 /* walk rest of this hash chain */
2429 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2430 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2431 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2432 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2433 if (n)
2434 goto found;
2435 }
2436
2437 /* new hash chain */
2438 while (++h < FIB_TABLE_HASHSZ) {
2439 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2440 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2441 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2442 if (n)
2443 goto found;
2444 }
2445 }
2446 return NULL;
2447
2448found:
2449 iter->tb = tb;
2450 return n;
2451}
2452
2453static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2454 __releases(RCU)
2455{
2456 rcu_read_unlock();
2457}
2458
2459static void seq_indent(struct seq_file *seq, int n)
2460{
2461 while (n-- > 0)
2462 seq_puts(seq, " ");
2463}
2464
2465static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2466{
2467 switch (s) {
2468 case RT_SCOPE_UNIVERSE: return "universe";
2469 case RT_SCOPE_SITE: return "site";
2470 case RT_SCOPE_LINK: return "link";
2471 case RT_SCOPE_HOST: return "host";
2472 case RT_SCOPE_NOWHERE: return "nowhere";
2473 default:
2474 snprintf(buf, len, "scope=%d", s);
2475 return buf;
2476 }
2477}
2478
2479static const char *const rtn_type_names[__RTN_MAX] = {
2480 [RTN_UNSPEC] = "UNSPEC",
2481 [RTN_UNICAST] = "UNICAST",
2482 [RTN_LOCAL] = "LOCAL",
2483 [RTN_BROADCAST] = "BROADCAST",
2484 [RTN_ANYCAST] = "ANYCAST",
2485 [RTN_MULTICAST] = "MULTICAST",
2486 [RTN_BLACKHOLE] = "BLACKHOLE",
2487 [RTN_UNREACHABLE] = "UNREACHABLE",
2488 [RTN_PROHIBIT] = "PROHIBIT",
2489 [RTN_THROW] = "THROW",
2490 [RTN_NAT] = "NAT",
2491 [RTN_XRESOLVE] = "XRESOLVE",
2492};
2493
2494static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2495{
2496 if (t < __RTN_MAX && rtn_type_names[t])
2497 return rtn_type_names[t];
2498 snprintf(buf, len, "type %u", t);
2499 return buf;
2500}
2501
2502/* Pretty print the trie */
2503static int fib_trie_seq_show(struct seq_file *seq, void *v)
2504{
2505 const struct fib_trie_iter *iter = seq->private;
2506 struct key_vector *n = v;
2507
2508 if (IS_TRIE(node_parent_rcu(n)))
2509 fib_table_print(seq, iter->tb);
2510
2511 if (IS_TNODE(n)) {
2512 __be32 prf = htonl(n->key);
2513
2514 seq_indent(seq, iter->depth-1);
2515 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2516 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2517 tn_info(n)->full_children,
2518 tn_info(n)->empty_children);
2519 } else {
2520 __be32 val = htonl(n->key);
2521 struct fib_alias *fa;
2522
2523 seq_indent(seq, iter->depth);
2524 seq_printf(seq, " |-- %pI4\n", &val);
2525
2526 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2527 char buf1[32], buf2[32];
2528
2529 seq_indent(seq, iter->depth + 1);
2530 seq_printf(seq, " /%zu %s %s",
2531 KEYLENGTH - fa->fa_slen,
2532 rtn_scope(buf1, sizeof(buf1),
2533 fa->fa_info->fib_scope),
2534 rtn_type(buf2, sizeof(buf2),
2535 fa->fa_type));
2536 if (fa->fa_tos)
2537 seq_printf(seq, " tos=%d", fa->fa_tos);
2538 seq_putc(seq, '\n');
2539 }
2540 }
2541
2542 return 0;
2543}
2544
2545static const struct seq_operations fib_trie_seq_ops = {
2546 .start = fib_trie_seq_start,
2547 .next = fib_trie_seq_next,
2548 .stop = fib_trie_seq_stop,
2549 .show = fib_trie_seq_show,
2550};
2551
2552struct fib_route_iter {
2553 struct seq_net_private p;
2554 struct fib_table *main_tb;
2555 struct key_vector *tnode;
2556 loff_t pos;
2557 t_key key;
2558};
2559
2560static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2561 loff_t pos)
2562{
2563 struct key_vector *l, **tp = &iter->tnode;
2564 t_key key;
2565
2566 /* use cached location of previously found key */
2567 if (iter->pos > 0 && pos >= iter->pos) {
2568 key = iter->key;
2569 } else {
2570 iter->pos = 1;
2571 key = 0;
2572 }
2573
2574 pos -= iter->pos;
2575
2576 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2577 key = l->key + 1;
2578 iter->pos++;
2579 l = NULL;
2580
2581 /* handle unlikely case of a key wrap */
2582 if (!key)
2583 break;
2584 }
2585
2586 if (l)
2587 iter->key = l->key; /* remember it */
2588 else
2589 iter->pos = 0; /* forget it */
2590
2591 return l;
2592}
2593
2594static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2595 __acquires(RCU)
2596{
2597 struct fib_route_iter *iter = seq->private;
2598 struct fib_table *tb;
2599 struct trie *t;
2600
2601 rcu_read_lock();
2602
2603 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2604 if (!tb)
2605 return NULL;
2606
2607 iter->main_tb = tb;
2608 t = (struct trie *)tb->tb_data;
2609 iter->tnode = t->kv;
2610
2611 if (*pos != 0)
2612 return fib_route_get_idx(iter, *pos);
2613
2614 iter->pos = 0;
2615 iter->key = KEY_MAX;
2616
2617 return SEQ_START_TOKEN;
2618}
2619
2620static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2621{
2622 struct fib_route_iter *iter = seq->private;
2623 struct key_vector *l = NULL;
2624 t_key key = iter->key + 1;
2625
2626 ++*pos;
2627
2628 /* only allow key of 0 for start of sequence */
2629 if ((v == SEQ_START_TOKEN) || key)
2630 l = leaf_walk_rcu(&iter->tnode, key);
2631
2632 if (l) {
2633 iter->key = l->key;
2634 iter->pos++;
2635 } else {
2636 iter->pos = 0;
2637 }
2638
2639 return l;
2640}
2641
2642static void fib_route_seq_stop(struct seq_file *seq, void *v)
2643 __releases(RCU)
2644{
2645 rcu_read_unlock();
2646}
2647
2648static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2649{
2650 unsigned int flags = 0;
2651
2652 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2653 flags = RTF_REJECT;
2654 if (fi && fi->fib_nh->nh_gw)
2655 flags |= RTF_GATEWAY;
2656 if (mask == htonl(0xFFFFFFFF))
2657 flags |= RTF_HOST;
2658 flags |= RTF_UP;
2659 return flags;
2660}
2661
2662/*
2663 * This outputs /proc/net/route.
2664 * The format of the file is not supposed to be changed
2665 * and needs to be same as fib_hash output to avoid breaking
2666 * legacy utilities
2667 */
2668static int fib_route_seq_show(struct seq_file *seq, void *v)
2669{
2670 struct fib_route_iter *iter = seq->private;
2671 struct fib_table *tb = iter->main_tb;
2672 struct fib_alias *fa;
2673 struct key_vector *l = v;
2674 __be32 prefix;
2675
2676 if (v == SEQ_START_TOKEN) {
2677 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2678 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2679 "\tWindow\tIRTT");
2680 return 0;
2681 }
2682
2683 prefix = htonl(l->key);
2684
2685 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2686 const struct fib_info *fi = fa->fa_info;
2687 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2688 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2689
2690 if ((fa->fa_type == RTN_BROADCAST) ||
2691 (fa->fa_type == RTN_MULTICAST))
2692 continue;
2693
2694 if (fa->tb_id != tb->tb_id)
2695 continue;
2696
2697 seq_setwidth(seq, 127);
2698
2699 if (fi)
2700 seq_printf(seq,
2701 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2702 "%d\t%08X\t%d\t%u\t%u",
2703 fi->fib_dev ? fi->fib_dev->name : "*",
2704 prefix,
2705 fi->fib_nh->nh_gw, flags, 0, 0,
2706 fi->fib_priority,
2707 mask,
2708 (fi->fib_advmss ?
2709 fi->fib_advmss + 40 : 0),
2710 fi->fib_window,
2711 fi->fib_rtt >> 3);
2712 else
2713 seq_printf(seq,
2714 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2715 "%d\t%08X\t%d\t%u\t%u",
2716 prefix, 0, flags, 0, 0, 0,
2717 mask, 0, 0, 0);
2718
2719 seq_pad(seq, '\n');
2720 }
2721
2722 return 0;
2723}
2724
2725static const struct seq_operations fib_route_seq_ops = {
2726 .start = fib_route_seq_start,
2727 .next = fib_route_seq_next,
2728 .stop = fib_route_seq_stop,
2729 .show = fib_route_seq_show,
2730};
2731
2732int __net_init fib_proc_init(struct net *net)
2733{
2734 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2735 sizeof(struct fib_trie_iter)))
2736 goto out1;
2737
2738 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
2739 fib_triestat_seq_show, NULL))
2740 goto out2;
2741
2742 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
2743 sizeof(struct fib_route_iter)))
2744 goto out3;
2745
2746 return 0;
2747
2748out3:
2749 remove_proc_entry("fib_triestat", net->proc_net);
2750out2:
2751 remove_proc_entry("fib_trie", net->proc_net);
2752out1:
2753 return -ENOMEM;
2754}
2755
2756void __net_exit fib_proc_exit(struct net *net)
2757{
2758 remove_proc_entry("fib_trie", net->proc_net);
2759 remove_proc_entry("fib_triestat", net->proc_net);
2760 remove_proc_entry("route", net->proc_net);
2761}
2762
2763#endif /* CONFIG_PROC_FS */
2764