Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 *
4 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
5 * & Swedish University of Agricultural Sciences.
6 *
7 * Jens Laas <jens.laas@data.slu.se> Swedish University of
8 * Agricultural Sciences.
9 *
10 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
11 *
12 * This work is based on the LPC-trie which is originally described in:
13 *
14 * An experimental study of compression methods for dynamic tries
15 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
16 * https://www.csc.kth.se/~snilsson/software/dyntrie2/
17 *
18 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
19 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
20 *
21 * Code from fib_hash has been reused which includes the following header:
22 *
23 * INET An implementation of the TCP/IP protocol suite for the LINUX
24 * operating system. INET is implemented using the BSD Socket
25 * interface as the means of communication with the user level.
26 *
27 * IPv4 FIB: lookup engine and maintenance routines.
28 *
29 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
30 *
31 * Substantial contributions to this work comes from:
32 *
33 * David S. Miller, <davem@davemloft.net>
34 * Stephen Hemminger <shemminger@osdl.org>
35 * Paul E. McKenney <paulmck@us.ibm.com>
36 * Patrick McHardy <kaber@trash.net>
37 */
38#include <linux/cache.h>
39#include <linux/uaccess.h>
40#include <linux/bitops.h>
41#include <linux/types.h>
42#include <linux/kernel.h>
43#include <linux/mm.h>
44#include <linux/string.h>
45#include <linux/socket.h>
46#include <linux/sockios.h>
47#include <linux/errno.h>
48#include <linux/in.h>
49#include <linux/inet.h>
50#include <linux/inetdevice.h>
51#include <linux/netdevice.h>
52#include <linux/if_arp.h>
53#include <linux/proc_fs.h>
54#include <linux/rcupdate.h>
55#include <linux/rcupdate_wait.h>
56#include <linux/skbuff.h>
57#include <linux/netlink.h>
58#include <linux/init.h>
59#include <linux/list.h>
60#include <linux/slab.h>
61#include <linux/export.h>
62#include <linux/vmalloc.h>
63#include <linux/notifier.h>
64#include <net/net_namespace.h>
65#include <net/inet_dscp.h>
66#include <net/ip.h>
67#include <net/protocol.h>
68#include <net/route.h>
69#include <net/tcp.h>
70#include <net/sock.h>
71#include <net/ip_fib.h>
72#include <net/fib_notifier.h>
73#include <trace/events/fib.h>
74#include "fib_lookup.h"
75
76static int call_fib_entry_notifier(struct notifier_block *nb,
77 enum fib_event_type event_type, u32 dst,
78 int dst_len, struct fib_alias *fa,
79 struct netlink_ext_ack *extack)
80{
81 struct fib_entry_notifier_info info = {
82 .info.extack = extack,
83 .dst = dst,
84 .dst_len = dst_len,
85 .fi = fa->fa_info,
86 .dscp = fa->fa_dscp,
87 .type = fa->fa_type,
88 .tb_id = fa->tb_id,
89 };
90 return call_fib4_notifier(nb, event_type, &info.info);
91}
92
93static int call_fib_entry_notifiers(struct net *net,
94 enum fib_event_type event_type, u32 dst,
95 int dst_len, struct fib_alias *fa,
96 struct netlink_ext_ack *extack)
97{
98 struct fib_entry_notifier_info info = {
99 .info.extack = extack,
100 .dst = dst,
101 .dst_len = dst_len,
102 .fi = fa->fa_info,
103 .dscp = fa->fa_dscp,
104 .type = fa->fa_type,
105 .tb_id = fa->tb_id,
106 };
107 return call_fib4_notifiers(net, event_type, &info.info);
108}
109
110#define MAX_STAT_DEPTH 32
111
112#define KEYLENGTH (8*sizeof(t_key))
113#define KEY_MAX ((t_key)~0)
114
115typedef unsigned int t_key;
116
117#define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
118#define IS_TNODE(n) ((n)->bits)
119#define IS_LEAF(n) (!(n)->bits)
120
121struct key_vector {
122 t_key key;
123 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
124 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
125 unsigned char slen;
126 union {
127 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
128 struct hlist_head leaf;
129 /* This array is valid if (pos | bits) > 0 (TNODE) */
130 DECLARE_FLEX_ARRAY(struct key_vector __rcu *, tnode);
131 };
132};
133
134struct tnode {
135 struct rcu_head rcu;
136 t_key empty_children; /* KEYLENGTH bits needed */
137 t_key full_children; /* KEYLENGTH bits needed */
138 struct key_vector __rcu *parent;
139 struct key_vector kv[1];
140#define tn_bits kv[0].bits
141};
142
143#define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
144#define LEAF_SIZE TNODE_SIZE(1)
145
146#ifdef CONFIG_IP_FIB_TRIE_STATS
147struct trie_use_stats {
148 unsigned int gets;
149 unsigned int backtrack;
150 unsigned int semantic_match_passed;
151 unsigned int semantic_match_miss;
152 unsigned int null_node_hit;
153 unsigned int resize_node_skipped;
154};
155#endif
156
157struct trie_stat {
158 unsigned int totdepth;
159 unsigned int maxdepth;
160 unsigned int tnodes;
161 unsigned int leaves;
162 unsigned int nullpointers;
163 unsigned int prefixes;
164 unsigned int nodesizes[MAX_STAT_DEPTH];
165};
166
167struct trie {
168 struct key_vector kv[1];
169#ifdef CONFIG_IP_FIB_TRIE_STATS
170 struct trie_use_stats __percpu *stats;
171#endif
172};
173
174static struct key_vector *resize(struct trie *t, struct key_vector *tn);
175static unsigned int tnode_free_size;
176
177/*
178 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
179 * especially useful before resizing the root node with PREEMPT_NONE configs;
180 * the value was obtained experimentally, aiming to avoid visible slowdown.
181 */
182unsigned int sysctl_fib_sync_mem = 512 * 1024;
183unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
184unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
185
186static struct kmem_cache *fn_alias_kmem __ro_after_init;
187static struct kmem_cache *trie_leaf_kmem __ro_after_init;
188
189static inline struct tnode *tn_info(struct key_vector *kv)
190{
191 return container_of(kv, struct tnode, kv[0]);
192}
193
194/* caller must hold RTNL */
195#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
196#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
197
198/* caller must hold RCU read lock or RTNL */
199#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
200#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
201
202/* wrapper for rcu_assign_pointer */
203static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
204{
205 if (n)
206 rcu_assign_pointer(tn_info(n)->parent, tp);
207}
208
209#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
210
211/* This provides us with the number of children in this node, in the case of a
212 * leaf this will return 0 meaning none of the children are accessible.
213 */
214static inline unsigned long child_length(const struct key_vector *tn)
215{
216 return (1ul << tn->bits) & ~(1ul);
217}
218
219#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
220
221static inline unsigned long get_index(t_key key, struct key_vector *kv)
222{
223 unsigned long index = key ^ kv->key;
224
225 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
226 return 0;
227
228 return index >> kv->pos;
229}
230
231/* To understand this stuff, an understanding of keys and all their bits is
232 * necessary. Every node in the trie has a key associated with it, but not
233 * all of the bits in that key are significant.
234 *
235 * Consider a node 'n' and its parent 'tp'.
236 *
237 * If n is a leaf, every bit in its key is significant. Its presence is
238 * necessitated by path compression, since during a tree traversal (when
239 * searching for a leaf - unless we are doing an insertion) we will completely
240 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
241 * a potentially successful search, that we have indeed been walking the
242 * correct key path.
243 *
244 * Note that we can never "miss" the correct key in the tree if present by
245 * following the wrong path. Path compression ensures that segments of the key
246 * that are the same for all keys with a given prefix are skipped, but the
247 * skipped part *is* identical for each node in the subtrie below the skipped
248 * bit! trie_insert() in this implementation takes care of that.
249 *
250 * if n is an internal node - a 'tnode' here, the various parts of its key
251 * have many different meanings.
252 *
253 * Example:
254 * _________________________________________________________________
255 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
256 * -----------------------------------------------------------------
257 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
258 *
259 * _________________________________________________________________
260 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
261 * -----------------------------------------------------------------
262 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
263 *
264 * tp->pos = 22
265 * tp->bits = 3
266 * n->pos = 13
267 * n->bits = 4
268 *
269 * First, let's just ignore the bits that come before the parent tp, that is
270 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
271 * point we do not use them for anything.
272 *
273 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
274 * index into the parent's child array. That is, they will be used to find
275 * 'n' among tp's children.
276 *
277 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
278 * for the node n.
279 *
280 * All the bits we have seen so far are significant to the node n. The rest
281 * of the bits are really not needed or indeed known in n->key.
282 *
283 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
284 * n's child array, and will of course be different for each child.
285 *
286 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
287 * at this point.
288 */
289
290static const int halve_threshold = 25;
291static const int inflate_threshold = 50;
292static const int halve_threshold_root = 15;
293static const int inflate_threshold_root = 30;
294
295static inline void alias_free_mem_rcu(struct fib_alias *fa)
296{
297 kfree_rcu(fa, rcu);
298}
299
300#define TNODE_VMALLOC_MAX \
301 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
302
303static void __node_free_rcu(struct rcu_head *head)
304{
305 struct tnode *n = container_of(head, struct tnode, rcu);
306
307 if (!n->tn_bits)
308 kmem_cache_free(trie_leaf_kmem, n);
309 else
310 kvfree(n);
311}
312
313#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
314
315static struct tnode *tnode_alloc(int bits)
316{
317 size_t size;
318
319 /* verify bits is within bounds */
320 if (bits > TNODE_VMALLOC_MAX)
321 return NULL;
322
323 /* determine size and verify it is non-zero and didn't overflow */
324 size = TNODE_SIZE(1ul << bits);
325
326 if (size <= PAGE_SIZE)
327 return kzalloc(size, GFP_KERNEL);
328 else
329 return vzalloc(size);
330}
331
332static inline void empty_child_inc(struct key_vector *n)
333{
334 tn_info(n)->empty_children++;
335
336 if (!tn_info(n)->empty_children)
337 tn_info(n)->full_children++;
338}
339
340static inline void empty_child_dec(struct key_vector *n)
341{
342 if (!tn_info(n)->empty_children)
343 tn_info(n)->full_children--;
344
345 tn_info(n)->empty_children--;
346}
347
348static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
349{
350 struct key_vector *l;
351 struct tnode *kv;
352
353 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
354 if (!kv)
355 return NULL;
356
357 /* initialize key vector */
358 l = kv->kv;
359 l->key = key;
360 l->pos = 0;
361 l->bits = 0;
362 l->slen = fa->fa_slen;
363
364 /* link leaf to fib alias */
365 INIT_HLIST_HEAD(&l->leaf);
366 hlist_add_head(&fa->fa_list, &l->leaf);
367
368 return l;
369}
370
371static struct key_vector *tnode_new(t_key key, int pos, int bits)
372{
373 unsigned int shift = pos + bits;
374 struct key_vector *tn;
375 struct tnode *tnode;
376
377 /* verify bits and pos their msb bits clear and values are valid */
378 BUG_ON(!bits || (shift > KEYLENGTH));
379
380 tnode = tnode_alloc(bits);
381 if (!tnode)
382 return NULL;
383
384 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
385 sizeof(struct key_vector *) << bits);
386
387 if (bits == KEYLENGTH)
388 tnode->full_children = 1;
389 else
390 tnode->empty_children = 1ul << bits;
391
392 tn = tnode->kv;
393 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
394 tn->pos = pos;
395 tn->bits = bits;
396 tn->slen = pos;
397
398 return tn;
399}
400
401/* Check whether a tnode 'n' is "full", i.e. it is an internal node
402 * and no bits are skipped. See discussion in dyntree paper p. 6
403 */
404static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
405{
406 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
407}
408
409/* Add a child at position i overwriting the old value.
410 * Update the value of full_children and empty_children.
411 */
412static void put_child(struct key_vector *tn, unsigned long i,
413 struct key_vector *n)
414{
415 struct key_vector *chi = get_child(tn, i);
416 int isfull, wasfull;
417
418 BUG_ON(i >= child_length(tn));
419
420 /* update emptyChildren, overflow into fullChildren */
421 if (!n && chi)
422 empty_child_inc(tn);
423 if (n && !chi)
424 empty_child_dec(tn);
425
426 /* update fullChildren */
427 wasfull = tnode_full(tn, chi);
428 isfull = tnode_full(tn, n);
429
430 if (wasfull && !isfull)
431 tn_info(tn)->full_children--;
432 else if (!wasfull && isfull)
433 tn_info(tn)->full_children++;
434
435 if (n && (tn->slen < n->slen))
436 tn->slen = n->slen;
437
438 rcu_assign_pointer(tn->tnode[i], n);
439}
440
441static void update_children(struct key_vector *tn)
442{
443 unsigned long i;
444
445 /* update all of the child parent pointers */
446 for (i = child_length(tn); i;) {
447 struct key_vector *inode = get_child(tn, --i);
448
449 if (!inode)
450 continue;
451
452 /* Either update the children of a tnode that
453 * already belongs to us or update the child
454 * to point to ourselves.
455 */
456 if (node_parent(inode) == tn)
457 update_children(inode);
458 else
459 node_set_parent(inode, tn);
460 }
461}
462
463static inline void put_child_root(struct key_vector *tp, t_key key,
464 struct key_vector *n)
465{
466 if (IS_TRIE(tp))
467 rcu_assign_pointer(tp->tnode[0], n);
468 else
469 put_child(tp, get_index(key, tp), n);
470}
471
472static inline void tnode_free_init(struct key_vector *tn)
473{
474 tn_info(tn)->rcu.next = NULL;
475}
476
477static inline void tnode_free_append(struct key_vector *tn,
478 struct key_vector *n)
479{
480 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
481 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
482}
483
484static void tnode_free(struct key_vector *tn)
485{
486 struct callback_head *head = &tn_info(tn)->rcu;
487
488 while (head) {
489 head = head->next;
490 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
491 node_free(tn);
492
493 tn = container_of(head, struct tnode, rcu)->kv;
494 }
495
496 if (tnode_free_size >= READ_ONCE(sysctl_fib_sync_mem)) {
497 tnode_free_size = 0;
498 synchronize_net();
499 }
500}
501
502static struct key_vector *replace(struct trie *t,
503 struct key_vector *oldtnode,
504 struct key_vector *tn)
505{
506 struct key_vector *tp = node_parent(oldtnode);
507 unsigned long i;
508
509 /* setup the parent pointer out of and back into this node */
510 NODE_INIT_PARENT(tn, tp);
511 put_child_root(tp, tn->key, tn);
512
513 /* update all of the child parent pointers */
514 update_children(tn);
515
516 /* all pointers should be clean so we are done */
517 tnode_free(oldtnode);
518
519 /* resize children now that oldtnode is freed */
520 for (i = child_length(tn); i;) {
521 struct key_vector *inode = get_child(tn, --i);
522
523 /* resize child node */
524 if (tnode_full(tn, inode))
525 tn = resize(t, inode);
526 }
527
528 return tp;
529}
530
531static struct key_vector *inflate(struct trie *t,
532 struct key_vector *oldtnode)
533{
534 struct key_vector *tn;
535 unsigned long i;
536 t_key m;
537
538 pr_debug("In inflate\n");
539
540 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
541 if (!tn)
542 goto notnode;
543
544 /* prepare oldtnode to be freed */
545 tnode_free_init(oldtnode);
546
547 /* Assemble all of the pointers in our cluster, in this case that
548 * represents all of the pointers out of our allocated nodes that
549 * point to existing tnodes and the links between our allocated
550 * nodes.
551 */
552 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
553 struct key_vector *inode = get_child(oldtnode, --i);
554 struct key_vector *node0, *node1;
555 unsigned long j, k;
556
557 /* An empty child */
558 if (!inode)
559 continue;
560
561 /* A leaf or an internal node with skipped bits */
562 if (!tnode_full(oldtnode, inode)) {
563 put_child(tn, get_index(inode->key, tn), inode);
564 continue;
565 }
566
567 /* drop the node in the old tnode free list */
568 tnode_free_append(oldtnode, inode);
569
570 /* An internal node with two children */
571 if (inode->bits == 1) {
572 put_child(tn, 2 * i + 1, get_child(inode, 1));
573 put_child(tn, 2 * i, get_child(inode, 0));
574 continue;
575 }
576
577 /* We will replace this node 'inode' with two new
578 * ones, 'node0' and 'node1', each with half of the
579 * original children. The two new nodes will have
580 * a position one bit further down the key and this
581 * means that the "significant" part of their keys
582 * (see the discussion near the top of this file)
583 * will differ by one bit, which will be "0" in
584 * node0's key and "1" in node1's key. Since we are
585 * moving the key position by one step, the bit that
586 * we are moving away from - the bit at position
587 * (tn->pos) - is the one that will differ between
588 * node0 and node1. So... we synthesize that bit in the
589 * two new keys.
590 */
591 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
592 if (!node1)
593 goto nomem;
594 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
595
596 tnode_free_append(tn, node1);
597 if (!node0)
598 goto nomem;
599 tnode_free_append(tn, node0);
600
601 /* populate child pointers in new nodes */
602 for (k = child_length(inode), j = k / 2; j;) {
603 put_child(node1, --j, get_child(inode, --k));
604 put_child(node0, j, get_child(inode, j));
605 put_child(node1, --j, get_child(inode, --k));
606 put_child(node0, j, get_child(inode, j));
607 }
608
609 /* link new nodes to parent */
610 NODE_INIT_PARENT(node1, tn);
611 NODE_INIT_PARENT(node0, tn);
612
613 /* link parent to nodes */
614 put_child(tn, 2 * i + 1, node1);
615 put_child(tn, 2 * i, node0);
616 }
617
618 /* setup the parent pointers into and out of this node */
619 return replace(t, oldtnode, tn);
620nomem:
621 /* all pointers should be clean so we are done */
622 tnode_free(tn);
623notnode:
624 return NULL;
625}
626
627static struct key_vector *halve(struct trie *t,
628 struct key_vector *oldtnode)
629{
630 struct key_vector *tn;
631 unsigned long i;
632
633 pr_debug("In halve\n");
634
635 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
636 if (!tn)
637 goto notnode;
638
639 /* prepare oldtnode to be freed */
640 tnode_free_init(oldtnode);
641
642 /* Assemble all of the pointers in our cluster, in this case that
643 * represents all of the pointers out of our allocated nodes that
644 * point to existing tnodes and the links between our allocated
645 * nodes.
646 */
647 for (i = child_length(oldtnode); i;) {
648 struct key_vector *node1 = get_child(oldtnode, --i);
649 struct key_vector *node0 = get_child(oldtnode, --i);
650 struct key_vector *inode;
651
652 /* At least one of the children is empty */
653 if (!node1 || !node0) {
654 put_child(tn, i / 2, node1 ? : node0);
655 continue;
656 }
657
658 /* Two nonempty children */
659 inode = tnode_new(node0->key, oldtnode->pos, 1);
660 if (!inode)
661 goto nomem;
662 tnode_free_append(tn, inode);
663
664 /* initialize pointers out of node */
665 put_child(inode, 1, node1);
666 put_child(inode, 0, node0);
667 NODE_INIT_PARENT(inode, tn);
668
669 /* link parent to node */
670 put_child(tn, i / 2, inode);
671 }
672
673 /* setup the parent pointers into and out of this node */
674 return replace(t, oldtnode, tn);
675nomem:
676 /* all pointers should be clean so we are done */
677 tnode_free(tn);
678notnode:
679 return NULL;
680}
681
682static struct key_vector *collapse(struct trie *t,
683 struct key_vector *oldtnode)
684{
685 struct key_vector *n, *tp;
686 unsigned long i;
687
688 /* scan the tnode looking for that one child that might still exist */
689 for (n = NULL, i = child_length(oldtnode); !n && i;)
690 n = get_child(oldtnode, --i);
691
692 /* compress one level */
693 tp = node_parent(oldtnode);
694 put_child_root(tp, oldtnode->key, n);
695 node_set_parent(n, tp);
696
697 /* drop dead node */
698 node_free(oldtnode);
699
700 return tp;
701}
702
703static unsigned char update_suffix(struct key_vector *tn)
704{
705 unsigned char slen = tn->pos;
706 unsigned long stride, i;
707 unsigned char slen_max;
708
709 /* only vector 0 can have a suffix length greater than or equal to
710 * tn->pos + tn->bits, the second highest node will have a suffix
711 * length at most of tn->pos + tn->bits - 1
712 */
713 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
714
715 /* search though the list of children looking for nodes that might
716 * have a suffix greater than the one we currently have. This is
717 * why we start with a stride of 2 since a stride of 1 would
718 * represent the nodes with suffix length equal to tn->pos
719 */
720 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
721 struct key_vector *n = get_child(tn, i);
722
723 if (!n || (n->slen <= slen))
724 continue;
725
726 /* update stride and slen based on new value */
727 stride <<= (n->slen - slen);
728 slen = n->slen;
729 i &= ~(stride - 1);
730
731 /* stop searching if we have hit the maximum possible value */
732 if (slen >= slen_max)
733 break;
734 }
735
736 tn->slen = slen;
737
738 return slen;
739}
740
741/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
742 * the Helsinki University of Technology and Matti Tikkanen of Nokia
743 * Telecommunications, page 6:
744 * "A node is doubled if the ratio of non-empty children to all
745 * children in the *doubled* node is at least 'high'."
746 *
747 * 'high' in this instance is the variable 'inflate_threshold'. It
748 * is expressed as a percentage, so we multiply it with
749 * child_length() and instead of multiplying by 2 (since the
750 * child array will be doubled by inflate()) and multiplying
751 * the left-hand side by 100 (to handle the percentage thing) we
752 * multiply the left-hand side by 50.
753 *
754 * The left-hand side may look a bit weird: child_length(tn)
755 * - tn->empty_children is of course the number of non-null children
756 * in the current node. tn->full_children is the number of "full"
757 * children, that is non-null tnodes with a skip value of 0.
758 * All of those will be doubled in the resulting inflated tnode, so
759 * we just count them one extra time here.
760 *
761 * A clearer way to write this would be:
762 *
763 * to_be_doubled = tn->full_children;
764 * not_to_be_doubled = child_length(tn) - tn->empty_children -
765 * tn->full_children;
766 *
767 * new_child_length = child_length(tn) * 2;
768 *
769 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
770 * new_child_length;
771 * if (new_fill_factor >= inflate_threshold)
772 *
773 * ...and so on, tho it would mess up the while () loop.
774 *
775 * anyway,
776 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
777 * inflate_threshold
778 *
779 * avoid a division:
780 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
781 * inflate_threshold * new_child_length
782 *
783 * expand not_to_be_doubled and to_be_doubled, and shorten:
784 * 100 * (child_length(tn) - tn->empty_children +
785 * tn->full_children) >= inflate_threshold * new_child_length
786 *
787 * expand new_child_length:
788 * 100 * (child_length(tn) - tn->empty_children +
789 * tn->full_children) >=
790 * inflate_threshold * child_length(tn) * 2
791 *
792 * shorten again:
793 * 50 * (tn->full_children + child_length(tn) -
794 * tn->empty_children) >= inflate_threshold *
795 * child_length(tn)
796 *
797 */
798static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
799{
800 unsigned long used = child_length(tn);
801 unsigned long threshold = used;
802
803 /* Keep root node larger */
804 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
805 used -= tn_info(tn)->empty_children;
806 used += tn_info(tn)->full_children;
807
808 /* if bits == KEYLENGTH then pos = 0, and will fail below */
809
810 return (used > 1) && tn->pos && ((50 * used) >= threshold);
811}
812
813static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
814{
815 unsigned long used = child_length(tn);
816 unsigned long threshold = used;
817
818 /* Keep root node larger */
819 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
820 used -= tn_info(tn)->empty_children;
821
822 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
823
824 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
825}
826
827static inline bool should_collapse(struct key_vector *tn)
828{
829 unsigned long used = child_length(tn);
830
831 used -= tn_info(tn)->empty_children;
832
833 /* account for bits == KEYLENGTH case */
834 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
835 used -= KEY_MAX;
836
837 /* One child or none, time to drop us from the trie */
838 return used < 2;
839}
840
841#define MAX_WORK 10
842static struct key_vector *resize(struct trie *t, struct key_vector *tn)
843{
844#ifdef CONFIG_IP_FIB_TRIE_STATS
845 struct trie_use_stats __percpu *stats = t->stats;
846#endif
847 struct key_vector *tp = node_parent(tn);
848 unsigned long cindex = get_index(tn->key, tp);
849 int max_work = MAX_WORK;
850
851 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
852 tn, inflate_threshold, halve_threshold);
853
854 /* track the tnode via the pointer from the parent instead of
855 * doing it ourselves. This way we can let RCU fully do its
856 * thing without us interfering
857 */
858 BUG_ON(tn != get_child(tp, cindex));
859
860 /* Double as long as the resulting node has a number of
861 * nonempty nodes that are above the threshold.
862 */
863 while (should_inflate(tp, tn) && max_work) {
864 tp = inflate(t, tn);
865 if (!tp) {
866#ifdef CONFIG_IP_FIB_TRIE_STATS
867 this_cpu_inc(stats->resize_node_skipped);
868#endif
869 break;
870 }
871
872 max_work--;
873 tn = get_child(tp, cindex);
874 }
875
876 /* update parent in case inflate failed */
877 tp = node_parent(tn);
878
879 /* Return if at least one inflate is run */
880 if (max_work != MAX_WORK)
881 return tp;
882
883 /* Halve as long as the number of empty children in this
884 * node is above threshold.
885 */
886 while (should_halve(tp, tn) && max_work) {
887 tp = halve(t, tn);
888 if (!tp) {
889#ifdef CONFIG_IP_FIB_TRIE_STATS
890 this_cpu_inc(stats->resize_node_skipped);
891#endif
892 break;
893 }
894
895 max_work--;
896 tn = get_child(tp, cindex);
897 }
898
899 /* Only one child remains */
900 if (should_collapse(tn))
901 return collapse(t, tn);
902
903 /* update parent in case halve failed */
904 return node_parent(tn);
905}
906
907static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
908{
909 unsigned char node_slen = tn->slen;
910
911 while ((node_slen > tn->pos) && (node_slen > slen)) {
912 slen = update_suffix(tn);
913 if (node_slen == slen)
914 break;
915
916 tn = node_parent(tn);
917 node_slen = tn->slen;
918 }
919}
920
921static void node_push_suffix(struct key_vector *tn, unsigned char slen)
922{
923 while (tn->slen < slen) {
924 tn->slen = slen;
925 tn = node_parent(tn);
926 }
927}
928
929/* rcu_read_lock needs to be hold by caller from readside */
930static struct key_vector *fib_find_node(struct trie *t,
931 struct key_vector **tp, u32 key)
932{
933 struct key_vector *pn, *n = t->kv;
934 unsigned long index = 0;
935
936 do {
937 pn = n;
938 n = get_child_rcu(n, index);
939
940 if (!n)
941 break;
942
943 index = get_cindex(key, n);
944
945 /* This bit of code is a bit tricky but it combines multiple
946 * checks into a single check. The prefix consists of the
947 * prefix plus zeros for the bits in the cindex. The index
948 * is the difference between the key and this value. From
949 * this we can actually derive several pieces of data.
950 * if (index >= (1ul << bits))
951 * we have a mismatch in skip bits and failed
952 * else
953 * we know the value is cindex
954 *
955 * This check is safe even if bits == KEYLENGTH due to the
956 * fact that we can only allocate a node with 32 bits if a
957 * long is greater than 32 bits.
958 */
959 if (index >= (1ul << n->bits)) {
960 n = NULL;
961 break;
962 }
963
964 /* keep searching until we find a perfect match leaf or NULL */
965 } while (IS_TNODE(n));
966
967 *tp = pn;
968
969 return n;
970}
971
972/* Return the first fib alias matching DSCP with
973 * priority less than or equal to PRIO.
974 * If 'find_first' is set, return the first matching
975 * fib alias, regardless of DSCP and priority.
976 */
977static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
978 dscp_t dscp, u32 prio, u32 tb_id,
979 bool find_first)
980{
981 struct fib_alias *fa;
982
983 if (!fah)
984 return NULL;
985
986 hlist_for_each_entry(fa, fah, fa_list) {
987 /* Avoid Sparse warning when using dscp_t in inequalities */
988 u8 __fa_dscp = inet_dscp_to_dsfield(fa->fa_dscp);
989 u8 __dscp = inet_dscp_to_dsfield(dscp);
990
991 if (fa->fa_slen < slen)
992 continue;
993 if (fa->fa_slen != slen)
994 break;
995 if (fa->tb_id > tb_id)
996 continue;
997 if (fa->tb_id != tb_id)
998 break;
999 if (find_first)
1000 return fa;
1001 if (__fa_dscp > __dscp)
1002 continue;
1003 if (fa->fa_info->fib_priority >= prio || __fa_dscp < __dscp)
1004 return fa;
1005 }
1006
1007 return NULL;
1008}
1009
1010static struct fib_alias *
1011fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1012{
1013 u8 slen = KEYLENGTH - fri->dst_len;
1014 struct key_vector *l, *tp;
1015 struct fib_table *tb;
1016 struct fib_alias *fa;
1017 struct trie *t;
1018
1019 tb = fib_get_table(net, fri->tb_id);
1020 if (!tb)
1021 return NULL;
1022
1023 t = (struct trie *)tb->tb_data;
1024 l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1025 if (!l)
1026 return NULL;
1027
1028 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1029 if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1030 fa->fa_dscp == fri->dscp && fa->fa_info == fri->fi &&
1031 fa->fa_type == fri->type)
1032 return fa;
1033 }
1034
1035 return NULL;
1036}
1037
1038void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1039{
1040 u8 fib_notify_on_flag_change;
1041 struct fib_alias *fa_match;
1042 struct sk_buff *skb;
1043 int err;
1044
1045 rcu_read_lock();
1046
1047 fa_match = fib_find_matching_alias(net, fri);
1048 if (!fa_match)
1049 goto out;
1050
1051 /* These are paired with the WRITE_ONCE() happening in this function.
1052 * The reason is that we are only protected by RCU at this point.
1053 */
1054 if (READ_ONCE(fa_match->offload) == fri->offload &&
1055 READ_ONCE(fa_match->trap) == fri->trap &&
1056 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1057 goto out;
1058
1059 WRITE_ONCE(fa_match->offload, fri->offload);
1060 WRITE_ONCE(fa_match->trap, fri->trap);
1061
1062 fib_notify_on_flag_change = READ_ONCE(net->ipv4.sysctl_fib_notify_on_flag_change);
1063
1064 /* 2 means send notifications only if offload_failed was changed. */
1065 if (fib_notify_on_flag_change == 2 &&
1066 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1067 goto out;
1068
1069 WRITE_ONCE(fa_match->offload_failed, fri->offload_failed);
1070
1071 if (!fib_notify_on_flag_change)
1072 goto out;
1073
1074 skb = nlmsg_new(fib_nlmsg_size(fa_match->fa_info), GFP_ATOMIC);
1075 if (!skb) {
1076 err = -ENOBUFS;
1077 goto errout;
1078 }
1079
1080 err = fib_dump_info(skb, 0, 0, RTM_NEWROUTE, fri, 0);
1081 if (err < 0) {
1082 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
1083 WARN_ON(err == -EMSGSIZE);
1084 kfree_skb(skb);
1085 goto errout;
1086 }
1087
1088 rtnl_notify(skb, net, 0, RTNLGRP_IPV4_ROUTE, NULL, GFP_ATOMIC);
1089 goto out;
1090
1091errout:
1092 rtnl_set_sk_err(net, RTNLGRP_IPV4_ROUTE, err);
1093out:
1094 rcu_read_unlock();
1095}
1096EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1097
1098static void trie_rebalance(struct trie *t, struct key_vector *tn)
1099{
1100 while (!IS_TRIE(tn))
1101 tn = resize(t, tn);
1102}
1103
1104static int fib_insert_node(struct trie *t, struct key_vector *tp,
1105 struct fib_alias *new, t_key key)
1106{
1107 struct key_vector *n, *l;
1108
1109 l = leaf_new(key, new);
1110 if (!l)
1111 goto noleaf;
1112
1113 /* retrieve child from parent node */
1114 n = get_child(tp, get_index(key, tp));
1115
1116 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1117 *
1118 * Add a new tnode here
1119 * first tnode need some special handling
1120 * leaves us in position for handling as case 3
1121 */
1122 if (n) {
1123 struct key_vector *tn;
1124
1125 tn = tnode_new(key, __fls(key ^ n->key), 1);
1126 if (!tn)
1127 goto notnode;
1128
1129 /* initialize routes out of node */
1130 NODE_INIT_PARENT(tn, tp);
1131 put_child(tn, get_index(key, tn) ^ 1, n);
1132
1133 /* start adding routes into the node */
1134 put_child_root(tp, key, tn);
1135 node_set_parent(n, tn);
1136
1137 /* parent now has a NULL spot where the leaf can go */
1138 tp = tn;
1139 }
1140
1141 /* Case 3: n is NULL, and will just insert a new leaf */
1142 node_push_suffix(tp, new->fa_slen);
1143 NODE_INIT_PARENT(l, tp);
1144 put_child_root(tp, key, l);
1145 trie_rebalance(t, tp);
1146
1147 return 0;
1148notnode:
1149 node_free(l);
1150noleaf:
1151 return -ENOMEM;
1152}
1153
1154static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1155 struct key_vector *l, struct fib_alias *new,
1156 struct fib_alias *fa, t_key key)
1157{
1158 if (!l)
1159 return fib_insert_node(t, tp, new, key);
1160
1161 if (fa) {
1162 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1163 } else {
1164 struct fib_alias *last;
1165
1166 hlist_for_each_entry(last, &l->leaf, fa_list) {
1167 if (new->fa_slen < last->fa_slen)
1168 break;
1169 if ((new->fa_slen == last->fa_slen) &&
1170 (new->tb_id > last->tb_id))
1171 break;
1172 fa = last;
1173 }
1174
1175 if (fa)
1176 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1177 else
1178 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1179 }
1180
1181 /* if we added to the tail node then we need to update slen */
1182 if (l->slen < new->fa_slen) {
1183 l->slen = new->fa_slen;
1184 node_push_suffix(tp, new->fa_slen);
1185 }
1186
1187 return 0;
1188}
1189
1190static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1191{
1192 if (plen > KEYLENGTH) {
1193 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1194 return false;
1195 }
1196
1197 if ((plen < KEYLENGTH) && (key << plen)) {
1198 NL_SET_ERR_MSG(extack,
1199 "Invalid prefix for given prefix length");
1200 return false;
1201 }
1202
1203 return true;
1204}
1205
1206static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1207 struct key_vector *l, struct fib_alias *old);
1208
1209/* Caller must hold RTNL. */
1210int fib_table_insert(struct net *net, struct fib_table *tb,
1211 struct fib_config *cfg, struct netlink_ext_ack *extack)
1212{
1213 struct trie *t = (struct trie *)tb->tb_data;
1214 struct fib_alias *fa, *new_fa;
1215 struct key_vector *l, *tp;
1216 u16 nlflags = NLM_F_EXCL;
1217 struct fib_info *fi;
1218 u8 plen = cfg->fc_dst_len;
1219 u8 slen = KEYLENGTH - plen;
1220 dscp_t dscp;
1221 u32 key;
1222 int err;
1223
1224 key = ntohl(cfg->fc_dst);
1225
1226 if (!fib_valid_key_len(key, plen, extack))
1227 return -EINVAL;
1228
1229 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1230
1231 fi = fib_create_info(cfg, extack);
1232 if (IS_ERR(fi)) {
1233 err = PTR_ERR(fi);
1234 goto err;
1235 }
1236
1237 dscp = cfg->fc_dscp;
1238 l = fib_find_node(t, &tp, key);
1239 fa = l ? fib_find_alias(&l->leaf, slen, dscp, fi->fib_priority,
1240 tb->tb_id, false) : NULL;
1241
1242 /* Now fa, if non-NULL, points to the first fib alias
1243 * with the same keys [prefix,dscp,priority], if such key already
1244 * exists or to the node before which we will insert new one.
1245 *
1246 * If fa is NULL, we will need to allocate a new one and
1247 * insert to the tail of the section matching the suffix length
1248 * of the new alias.
1249 */
1250
1251 if (fa && fa->fa_dscp == dscp &&
1252 fa->fa_info->fib_priority == fi->fib_priority) {
1253 struct fib_alias *fa_first, *fa_match;
1254
1255 err = -EEXIST;
1256 if (cfg->fc_nlflags & NLM_F_EXCL)
1257 goto out;
1258
1259 nlflags &= ~NLM_F_EXCL;
1260
1261 /* We have 2 goals:
1262 * 1. Find exact match for type, scope, fib_info to avoid
1263 * duplicate routes
1264 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1265 */
1266 fa_match = NULL;
1267 fa_first = fa;
1268 hlist_for_each_entry_from(fa, fa_list) {
1269 if ((fa->fa_slen != slen) ||
1270 (fa->tb_id != tb->tb_id) ||
1271 (fa->fa_dscp != dscp))
1272 break;
1273 if (fa->fa_info->fib_priority != fi->fib_priority)
1274 break;
1275 if (fa->fa_type == cfg->fc_type &&
1276 fa->fa_info == fi) {
1277 fa_match = fa;
1278 break;
1279 }
1280 }
1281
1282 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1283 struct fib_info *fi_drop;
1284 u8 state;
1285
1286 nlflags |= NLM_F_REPLACE;
1287 fa = fa_first;
1288 if (fa_match) {
1289 if (fa == fa_match)
1290 err = 0;
1291 goto out;
1292 }
1293 err = -ENOBUFS;
1294 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1295 if (!new_fa)
1296 goto out;
1297
1298 fi_drop = fa->fa_info;
1299 new_fa->fa_dscp = fa->fa_dscp;
1300 new_fa->fa_info = fi;
1301 new_fa->fa_type = cfg->fc_type;
1302 state = fa->fa_state;
1303 new_fa->fa_state = state & ~FA_S_ACCESSED;
1304 new_fa->fa_slen = fa->fa_slen;
1305 new_fa->tb_id = tb->tb_id;
1306 new_fa->fa_default = -1;
1307 new_fa->offload = 0;
1308 new_fa->trap = 0;
1309 new_fa->offload_failed = 0;
1310
1311 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1312
1313 if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1314 tb->tb_id, true) == new_fa) {
1315 enum fib_event_type fib_event;
1316
1317 fib_event = FIB_EVENT_ENTRY_REPLACE;
1318 err = call_fib_entry_notifiers(net, fib_event,
1319 key, plen,
1320 new_fa, extack);
1321 if (err) {
1322 hlist_replace_rcu(&new_fa->fa_list,
1323 &fa->fa_list);
1324 goto out_free_new_fa;
1325 }
1326 }
1327
1328 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1329 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1330
1331 alias_free_mem_rcu(fa);
1332
1333 fib_release_info(fi_drop);
1334 if (state & FA_S_ACCESSED)
1335 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1336
1337 goto succeeded;
1338 }
1339 /* Error if we find a perfect match which
1340 * uses the same scope, type, and nexthop
1341 * information.
1342 */
1343 if (fa_match)
1344 goto out;
1345
1346 if (cfg->fc_nlflags & NLM_F_APPEND)
1347 nlflags |= NLM_F_APPEND;
1348 else
1349 fa = fa_first;
1350 }
1351 err = -ENOENT;
1352 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1353 goto out;
1354
1355 nlflags |= NLM_F_CREATE;
1356 err = -ENOBUFS;
1357 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1358 if (!new_fa)
1359 goto out;
1360
1361 new_fa->fa_info = fi;
1362 new_fa->fa_dscp = dscp;
1363 new_fa->fa_type = cfg->fc_type;
1364 new_fa->fa_state = 0;
1365 new_fa->fa_slen = slen;
1366 new_fa->tb_id = tb->tb_id;
1367 new_fa->fa_default = -1;
1368 new_fa->offload = 0;
1369 new_fa->trap = 0;
1370 new_fa->offload_failed = 0;
1371
1372 /* Insert new entry to the list. */
1373 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1374 if (err)
1375 goto out_free_new_fa;
1376
1377 /* The alias was already inserted, so the node must exist. */
1378 l = l ? l : fib_find_node(t, &tp, key);
1379 if (WARN_ON_ONCE(!l)) {
1380 err = -ENOENT;
1381 goto out_free_new_fa;
1382 }
1383
1384 if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1385 new_fa) {
1386 enum fib_event_type fib_event;
1387
1388 fib_event = FIB_EVENT_ENTRY_REPLACE;
1389 err = call_fib_entry_notifiers(net, fib_event, key, plen,
1390 new_fa, extack);
1391 if (err)
1392 goto out_remove_new_fa;
1393 }
1394
1395 if (!plen)
1396 tb->tb_num_default++;
1397
1398 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1399 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1400 &cfg->fc_nlinfo, nlflags);
1401succeeded:
1402 return 0;
1403
1404out_remove_new_fa:
1405 fib_remove_alias(t, tp, l, new_fa);
1406out_free_new_fa:
1407 kmem_cache_free(fn_alias_kmem, new_fa);
1408out:
1409 fib_release_info(fi);
1410err:
1411 return err;
1412}
1413
1414static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1415{
1416 t_key prefix = n->key;
1417
1418 return (key ^ prefix) & (prefix | -prefix);
1419}
1420
1421bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1422 const struct flowi4 *flp)
1423{
1424 if (nhc->nhc_flags & RTNH_F_DEAD)
1425 return false;
1426
1427 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1428 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1429 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1430 return false;
1431
1432 if (flp->flowi4_oif && flp->flowi4_oif != nhc->nhc_oif)
1433 return false;
1434
1435 return true;
1436}
1437
1438/* should be called with rcu_read_lock */
1439int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1440 struct fib_result *res, int fib_flags)
1441{
1442 struct trie *t = (struct trie *) tb->tb_data;
1443#ifdef CONFIG_IP_FIB_TRIE_STATS
1444 struct trie_use_stats __percpu *stats = t->stats;
1445#endif
1446 const t_key key = ntohl(flp->daddr);
1447 struct key_vector *n, *pn;
1448 struct fib_alias *fa;
1449 unsigned long index;
1450 t_key cindex;
1451
1452 pn = t->kv;
1453 cindex = 0;
1454
1455 n = get_child_rcu(pn, cindex);
1456 if (!n) {
1457 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1458 return -EAGAIN;
1459 }
1460
1461#ifdef CONFIG_IP_FIB_TRIE_STATS
1462 this_cpu_inc(stats->gets);
1463#endif
1464
1465 /* Step 1: Travel to the longest prefix match in the trie */
1466 for (;;) {
1467 index = get_cindex(key, n);
1468
1469 /* This bit of code is a bit tricky but it combines multiple
1470 * checks into a single check. The prefix consists of the
1471 * prefix plus zeros for the "bits" in the prefix. The index
1472 * is the difference between the key and this value. From
1473 * this we can actually derive several pieces of data.
1474 * if (index >= (1ul << bits))
1475 * we have a mismatch in skip bits and failed
1476 * else
1477 * we know the value is cindex
1478 *
1479 * This check is safe even if bits == KEYLENGTH due to the
1480 * fact that we can only allocate a node with 32 bits if a
1481 * long is greater than 32 bits.
1482 */
1483 if (index >= (1ul << n->bits))
1484 break;
1485
1486 /* we have found a leaf. Prefixes have already been compared */
1487 if (IS_LEAF(n))
1488 goto found;
1489
1490 /* only record pn and cindex if we are going to be chopping
1491 * bits later. Otherwise we are just wasting cycles.
1492 */
1493 if (n->slen > n->pos) {
1494 pn = n;
1495 cindex = index;
1496 }
1497
1498 n = get_child_rcu(n, index);
1499 if (unlikely(!n))
1500 goto backtrace;
1501 }
1502
1503 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1504 for (;;) {
1505 /* record the pointer where our next node pointer is stored */
1506 struct key_vector __rcu **cptr = n->tnode;
1507
1508 /* This test verifies that none of the bits that differ
1509 * between the key and the prefix exist in the region of
1510 * the lsb and higher in the prefix.
1511 */
1512 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1513 goto backtrace;
1514
1515 /* exit out and process leaf */
1516 if (unlikely(IS_LEAF(n)))
1517 break;
1518
1519 /* Don't bother recording parent info. Since we are in
1520 * prefix match mode we will have to come back to wherever
1521 * we started this traversal anyway
1522 */
1523
1524 while ((n = rcu_dereference(*cptr)) == NULL) {
1525backtrace:
1526#ifdef CONFIG_IP_FIB_TRIE_STATS
1527 if (!n)
1528 this_cpu_inc(stats->null_node_hit);
1529#endif
1530 /* If we are at cindex 0 there are no more bits for
1531 * us to strip at this level so we must ascend back
1532 * up one level to see if there are any more bits to
1533 * be stripped there.
1534 */
1535 while (!cindex) {
1536 t_key pkey = pn->key;
1537
1538 /* If we don't have a parent then there is
1539 * nothing for us to do as we do not have any
1540 * further nodes to parse.
1541 */
1542 if (IS_TRIE(pn)) {
1543 trace_fib_table_lookup(tb->tb_id, flp,
1544 NULL, -EAGAIN);
1545 return -EAGAIN;
1546 }
1547#ifdef CONFIG_IP_FIB_TRIE_STATS
1548 this_cpu_inc(stats->backtrack);
1549#endif
1550 /* Get Child's index */
1551 pn = node_parent_rcu(pn);
1552 cindex = get_index(pkey, pn);
1553 }
1554
1555 /* strip the least significant bit from the cindex */
1556 cindex &= cindex - 1;
1557
1558 /* grab pointer for next child node */
1559 cptr = &pn->tnode[cindex];
1560 }
1561 }
1562
1563found:
1564 /* this line carries forward the xor from earlier in the function */
1565 index = key ^ n->key;
1566
1567 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1568 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1569 struct fib_info *fi = fa->fa_info;
1570 struct fib_nh_common *nhc;
1571 int nhsel, err;
1572
1573 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1574 if (index >= (1ul << fa->fa_slen))
1575 continue;
1576 }
1577 if (fa->fa_dscp && !fib_dscp_masked_match(fa->fa_dscp, flp))
1578 continue;
1579 /* Paired with WRITE_ONCE() in fib_release_info() */
1580 if (READ_ONCE(fi->fib_dead))
1581 continue;
1582 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1583 continue;
1584 fib_alias_accessed(fa);
1585 err = fib_props[fa->fa_type].error;
1586 if (unlikely(err < 0)) {
1587out_reject:
1588#ifdef CONFIG_IP_FIB_TRIE_STATS
1589 this_cpu_inc(stats->semantic_match_passed);
1590#endif
1591 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1592 return err;
1593 }
1594 if (fi->fib_flags & RTNH_F_DEAD)
1595 continue;
1596
1597 if (unlikely(fi->nh)) {
1598 if (nexthop_is_blackhole(fi->nh)) {
1599 err = fib_props[RTN_BLACKHOLE].error;
1600 goto out_reject;
1601 }
1602
1603 nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1604 &nhsel);
1605 if (nhc)
1606 goto set_result;
1607 goto miss;
1608 }
1609
1610 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1611 nhc = fib_info_nhc(fi, nhsel);
1612
1613 if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1614 continue;
1615set_result:
1616 if (!(fib_flags & FIB_LOOKUP_NOREF))
1617 refcount_inc(&fi->fib_clntref);
1618
1619 res->prefix = htonl(n->key);
1620 res->prefixlen = KEYLENGTH - fa->fa_slen;
1621 res->nh_sel = nhsel;
1622 res->nhc = nhc;
1623 res->type = fa->fa_type;
1624 res->scope = fi->fib_scope;
1625 res->dscp = fa->fa_dscp;
1626 res->fi = fi;
1627 res->table = tb;
1628 res->fa_head = &n->leaf;
1629#ifdef CONFIG_IP_FIB_TRIE_STATS
1630 this_cpu_inc(stats->semantic_match_passed);
1631#endif
1632 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1633
1634 return err;
1635 }
1636 }
1637miss:
1638#ifdef CONFIG_IP_FIB_TRIE_STATS
1639 this_cpu_inc(stats->semantic_match_miss);
1640#endif
1641 goto backtrace;
1642}
1643EXPORT_SYMBOL_GPL(fib_table_lookup);
1644
1645static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1646 struct key_vector *l, struct fib_alias *old)
1647{
1648 /* record the location of the previous list_info entry */
1649 struct hlist_node **pprev = old->fa_list.pprev;
1650 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1651
1652 /* remove the fib_alias from the list */
1653 hlist_del_rcu(&old->fa_list);
1654
1655 /* if we emptied the list this leaf will be freed and we can sort
1656 * out parent suffix lengths as a part of trie_rebalance
1657 */
1658 if (hlist_empty(&l->leaf)) {
1659 if (tp->slen == l->slen)
1660 node_pull_suffix(tp, tp->pos);
1661 put_child_root(tp, l->key, NULL);
1662 node_free(l);
1663 trie_rebalance(t, tp);
1664 return;
1665 }
1666
1667 /* only access fa if it is pointing at the last valid hlist_node */
1668 if (*pprev)
1669 return;
1670
1671 /* update the trie with the latest suffix length */
1672 l->slen = fa->fa_slen;
1673 node_pull_suffix(tp, fa->fa_slen);
1674}
1675
1676static void fib_notify_alias_delete(struct net *net, u32 key,
1677 struct hlist_head *fah,
1678 struct fib_alias *fa_to_delete,
1679 struct netlink_ext_ack *extack)
1680{
1681 struct fib_alias *fa_next, *fa_to_notify;
1682 u32 tb_id = fa_to_delete->tb_id;
1683 u8 slen = fa_to_delete->fa_slen;
1684 enum fib_event_type fib_event;
1685
1686 /* Do not notify if we do not care about the route. */
1687 if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1688 return;
1689
1690 /* Determine if the route should be replaced by the next route in the
1691 * list.
1692 */
1693 fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1694 struct fib_alias, fa_list);
1695 if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1696 fib_event = FIB_EVENT_ENTRY_REPLACE;
1697 fa_to_notify = fa_next;
1698 } else {
1699 fib_event = FIB_EVENT_ENTRY_DEL;
1700 fa_to_notify = fa_to_delete;
1701 }
1702 call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1703 fa_to_notify, extack);
1704}
1705
1706/* Caller must hold RTNL. */
1707int fib_table_delete(struct net *net, struct fib_table *tb,
1708 struct fib_config *cfg, struct netlink_ext_ack *extack)
1709{
1710 struct trie *t = (struct trie *) tb->tb_data;
1711 struct fib_alias *fa, *fa_to_delete;
1712 struct key_vector *l, *tp;
1713 u8 plen = cfg->fc_dst_len;
1714 u8 slen = KEYLENGTH - plen;
1715 dscp_t dscp;
1716 u32 key;
1717
1718 key = ntohl(cfg->fc_dst);
1719
1720 if (!fib_valid_key_len(key, plen, extack))
1721 return -EINVAL;
1722
1723 l = fib_find_node(t, &tp, key);
1724 if (!l)
1725 return -ESRCH;
1726
1727 dscp = cfg->fc_dscp;
1728 fa = fib_find_alias(&l->leaf, slen, dscp, 0, tb->tb_id, false);
1729 if (!fa)
1730 return -ESRCH;
1731
1732 pr_debug("Deleting %08x/%d dsfield=0x%02x t=%p\n", key, plen,
1733 inet_dscp_to_dsfield(dscp), t);
1734
1735 fa_to_delete = NULL;
1736 hlist_for_each_entry_from(fa, fa_list) {
1737 struct fib_info *fi = fa->fa_info;
1738
1739 if ((fa->fa_slen != slen) ||
1740 (fa->tb_id != tb->tb_id) ||
1741 (fa->fa_dscp != dscp))
1742 break;
1743
1744 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1745 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1746 fa->fa_info->fib_scope == cfg->fc_scope) &&
1747 (!cfg->fc_prefsrc ||
1748 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1749 (!cfg->fc_protocol ||
1750 fi->fib_protocol == cfg->fc_protocol) &&
1751 fib_nh_match(net, cfg, fi, extack) == 0 &&
1752 fib_metrics_match(cfg, fi)) {
1753 fa_to_delete = fa;
1754 break;
1755 }
1756 }
1757
1758 if (!fa_to_delete)
1759 return -ESRCH;
1760
1761 fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1762 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1763 &cfg->fc_nlinfo, 0);
1764
1765 if (!plen)
1766 tb->tb_num_default--;
1767
1768 fib_remove_alias(t, tp, l, fa_to_delete);
1769
1770 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1771 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1772
1773 fib_release_info(fa_to_delete->fa_info);
1774 alias_free_mem_rcu(fa_to_delete);
1775 return 0;
1776}
1777
1778/* Scan for the next leaf starting at the provided key value */
1779static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1780{
1781 struct key_vector *pn, *n = *tn;
1782 unsigned long cindex;
1783
1784 /* this loop is meant to try and find the key in the trie */
1785 do {
1786 /* record parent and next child index */
1787 pn = n;
1788 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1789
1790 if (cindex >> pn->bits)
1791 break;
1792
1793 /* descend into the next child */
1794 n = get_child_rcu(pn, cindex++);
1795 if (!n)
1796 break;
1797
1798 /* guarantee forward progress on the keys */
1799 if (IS_LEAF(n) && (n->key >= key))
1800 goto found;
1801 } while (IS_TNODE(n));
1802
1803 /* this loop will search for the next leaf with a greater key */
1804 while (!IS_TRIE(pn)) {
1805 /* if we exhausted the parent node we will need to climb */
1806 if (cindex >= (1ul << pn->bits)) {
1807 t_key pkey = pn->key;
1808
1809 pn = node_parent_rcu(pn);
1810 cindex = get_index(pkey, pn) + 1;
1811 continue;
1812 }
1813
1814 /* grab the next available node */
1815 n = get_child_rcu(pn, cindex++);
1816 if (!n)
1817 continue;
1818
1819 /* no need to compare keys since we bumped the index */
1820 if (IS_LEAF(n))
1821 goto found;
1822
1823 /* Rescan start scanning in new node */
1824 pn = n;
1825 cindex = 0;
1826 }
1827
1828 *tn = pn;
1829 return NULL; /* Root of trie */
1830found:
1831 /* if we are at the limit for keys just return NULL for the tnode */
1832 *tn = pn;
1833 return n;
1834}
1835
1836static void fib_trie_free(struct fib_table *tb)
1837{
1838 struct trie *t = (struct trie *)tb->tb_data;
1839 struct key_vector *pn = t->kv;
1840 unsigned long cindex = 1;
1841 struct hlist_node *tmp;
1842 struct fib_alias *fa;
1843
1844 /* walk trie in reverse order and free everything */
1845 for (;;) {
1846 struct key_vector *n;
1847
1848 if (!(cindex--)) {
1849 t_key pkey = pn->key;
1850
1851 if (IS_TRIE(pn))
1852 break;
1853
1854 n = pn;
1855 pn = node_parent(pn);
1856
1857 /* drop emptied tnode */
1858 put_child_root(pn, n->key, NULL);
1859 node_free(n);
1860
1861 cindex = get_index(pkey, pn);
1862
1863 continue;
1864 }
1865
1866 /* grab the next available node */
1867 n = get_child(pn, cindex);
1868 if (!n)
1869 continue;
1870
1871 if (IS_TNODE(n)) {
1872 /* record pn and cindex for leaf walking */
1873 pn = n;
1874 cindex = 1ul << n->bits;
1875
1876 continue;
1877 }
1878
1879 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1880 hlist_del_rcu(&fa->fa_list);
1881 alias_free_mem_rcu(fa);
1882 }
1883
1884 put_child_root(pn, n->key, NULL);
1885 node_free(n);
1886 }
1887
1888#ifdef CONFIG_IP_FIB_TRIE_STATS
1889 free_percpu(t->stats);
1890#endif
1891 kfree(tb);
1892}
1893
1894struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1895{
1896 struct trie *ot = (struct trie *)oldtb->tb_data;
1897 struct key_vector *l, *tp = ot->kv;
1898 struct fib_table *local_tb;
1899 struct fib_alias *fa;
1900 struct trie *lt;
1901 t_key key = 0;
1902
1903 if (oldtb->tb_data == oldtb->__data)
1904 return oldtb;
1905
1906 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1907 if (!local_tb)
1908 return NULL;
1909
1910 lt = (struct trie *)local_tb->tb_data;
1911
1912 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1913 struct key_vector *local_l = NULL, *local_tp;
1914
1915 hlist_for_each_entry(fa, &l->leaf, fa_list) {
1916 struct fib_alias *new_fa;
1917
1918 if (local_tb->tb_id != fa->tb_id)
1919 continue;
1920
1921 /* clone fa for new local table */
1922 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1923 if (!new_fa)
1924 goto out;
1925
1926 memcpy(new_fa, fa, sizeof(*fa));
1927
1928 /* insert clone into table */
1929 if (!local_l)
1930 local_l = fib_find_node(lt, &local_tp, l->key);
1931
1932 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1933 NULL, l->key)) {
1934 kmem_cache_free(fn_alias_kmem, new_fa);
1935 goto out;
1936 }
1937 }
1938
1939 /* stop loop if key wrapped back to 0 */
1940 key = l->key + 1;
1941 if (key < l->key)
1942 break;
1943 }
1944
1945 return local_tb;
1946out:
1947 fib_trie_free(local_tb);
1948
1949 return NULL;
1950}
1951
1952/* Caller must hold RTNL */
1953void fib_table_flush_external(struct fib_table *tb)
1954{
1955 struct trie *t = (struct trie *)tb->tb_data;
1956 struct key_vector *pn = t->kv;
1957 unsigned long cindex = 1;
1958 struct hlist_node *tmp;
1959 struct fib_alias *fa;
1960
1961 /* walk trie in reverse order */
1962 for (;;) {
1963 unsigned char slen = 0;
1964 struct key_vector *n;
1965
1966 if (!(cindex--)) {
1967 t_key pkey = pn->key;
1968
1969 /* cannot resize the trie vector */
1970 if (IS_TRIE(pn))
1971 break;
1972
1973 /* update the suffix to address pulled leaves */
1974 if (pn->slen > pn->pos)
1975 update_suffix(pn);
1976
1977 /* resize completed node */
1978 pn = resize(t, pn);
1979 cindex = get_index(pkey, pn);
1980
1981 continue;
1982 }
1983
1984 /* grab the next available node */
1985 n = get_child(pn, cindex);
1986 if (!n)
1987 continue;
1988
1989 if (IS_TNODE(n)) {
1990 /* record pn and cindex for leaf walking */
1991 pn = n;
1992 cindex = 1ul << n->bits;
1993
1994 continue;
1995 }
1996
1997 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1998 /* if alias was cloned to local then we just
1999 * need to remove the local copy from main
2000 */
2001 if (tb->tb_id != fa->tb_id) {
2002 hlist_del_rcu(&fa->fa_list);
2003 alias_free_mem_rcu(fa);
2004 continue;
2005 }
2006
2007 /* record local slen */
2008 slen = fa->fa_slen;
2009 }
2010
2011 /* update leaf slen */
2012 n->slen = slen;
2013
2014 if (hlist_empty(&n->leaf)) {
2015 put_child_root(pn, n->key, NULL);
2016 node_free(n);
2017 }
2018 }
2019}
2020
2021/* Caller must hold RTNL. */
2022int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
2023{
2024 struct trie *t = (struct trie *)tb->tb_data;
2025 struct nl_info info = { .nl_net = net };
2026 struct key_vector *pn = t->kv;
2027 unsigned long cindex = 1;
2028 struct hlist_node *tmp;
2029 struct fib_alias *fa;
2030 int found = 0;
2031
2032 /* walk trie in reverse order */
2033 for (;;) {
2034 unsigned char slen = 0;
2035 struct key_vector *n;
2036
2037 if (!(cindex--)) {
2038 t_key pkey = pn->key;
2039
2040 /* cannot resize the trie vector */
2041 if (IS_TRIE(pn))
2042 break;
2043
2044 /* update the suffix to address pulled leaves */
2045 if (pn->slen > pn->pos)
2046 update_suffix(pn);
2047
2048 /* resize completed node */
2049 pn = resize(t, pn);
2050 cindex = get_index(pkey, pn);
2051
2052 continue;
2053 }
2054
2055 /* grab the next available node */
2056 n = get_child(pn, cindex);
2057 if (!n)
2058 continue;
2059
2060 if (IS_TNODE(n)) {
2061 /* record pn and cindex for leaf walking */
2062 pn = n;
2063 cindex = 1ul << n->bits;
2064
2065 continue;
2066 }
2067
2068 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2069 struct fib_info *fi = fa->fa_info;
2070
2071 if (!fi || tb->tb_id != fa->tb_id ||
2072 (!(fi->fib_flags & RTNH_F_DEAD) &&
2073 !fib_props[fa->fa_type].error)) {
2074 slen = fa->fa_slen;
2075 continue;
2076 }
2077
2078 /* Do not flush error routes if network namespace is
2079 * not being dismantled
2080 */
2081 if (!flush_all && fib_props[fa->fa_type].error) {
2082 slen = fa->fa_slen;
2083 continue;
2084 }
2085
2086 fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2087 NULL);
2088 if (fi->pfsrc_removed)
2089 rtmsg_fib(RTM_DELROUTE, htonl(n->key), fa,
2090 KEYLENGTH - fa->fa_slen, tb->tb_id, &info, 0);
2091 hlist_del_rcu(&fa->fa_list);
2092 fib_release_info(fa->fa_info);
2093 alias_free_mem_rcu(fa);
2094 found++;
2095 }
2096
2097 /* update leaf slen */
2098 n->slen = slen;
2099
2100 if (hlist_empty(&n->leaf)) {
2101 put_child_root(pn, n->key, NULL);
2102 node_free(n);
2103 }
2104 }
2105
2106 pr_debug("trie_flush found=%d\n", found);
2107 return found;
2108}
2109
2110/* derived from fib_trie_free */
2111static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2112 struct nl_info *info)
2113{
2114 struct trie *t = (struct trie *)tb->tb_data;
2115 struct key_vector *pn = t->kv;
2116 unsigned long cindex = 1;
2117 struct fib_alias *fa;
2118
2119 for (;;) {
2120 struct key_vector *n;
2121
2122 if (!(cindex--)) {
2123 t_key pkey = pn->key;
2124
2125 if (IS_TRIE(pn))
2126 break;
2127
2128 pn = node_parent(pn);
2129 cindex = get_index(pkey, pn);
2130 continue;
2131 }
2132
2133 /* grab the next available node */
2134 n = get_child(pn, cindex);
2135 if (!n)
2136 continue;
2137
2138 if (IS_TNODE(n)) {
2139 /* record pn and cindex for leaf walking */
2140 pn = n;
2141 cindex = 1ul << n->bits;
2142
2143 continue;
2144 }
2145
2146 hlist_for_each_entry(fa, &n->leaf, fa_list) {
2147 struct fib_info *fi = fa->fa_info;
2148
2149 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2150 continue;
2151
2152 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2153 KEYLENGTH - fa->fa_slen, tb->tb_id,
2154 info, NLM_F_REPLACE);
2155 }
2156 }
2157}
2158
2159void fib_info_notify_update(struct net *net, struct nl_info *info)
2160{
2161 unsigned int h;
2162
2163 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2164 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2165 struct fib_table *tb;
2166
2167 hlist_for_each_entry_rcu(tb, head, tb_hlist,
2168 lockdep_rtnl_is_held())
2169 __fib_info_notify_update(net, tb, info);
2170 }
2171}
2172
2173static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2174 struct notifier_block *nb,
2175 struct netlink_ext_ack *extack)
2176{
2177 struct fib_alias *fa;
2178 int last_slen = -1;
2179 int err;
2180
2181 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2182 struct fib_info *fi = fa->fa_info;
2183
2184 if (!fi)
2185 continue;
2186
2187 /* local and main table can share the same trie,
2188 * so don't notify twice for the same entry.
2189 */
2190 if (tb->tb_id != fa->tb_id)
2191 continue;
2192
2193 if (fa->fa_slen == last_slen)
2194 continue;
2195
2196 last_slen = fa->fa_slen;
2197 err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2198 l->key, KEYLENGTH - fa->fa_slen,
2199 fa, extack);
2200 if (err)
2201 return err;
2202 }
2203 return 0;
2204}
2205
2206static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2207 struct netlink_ext_ack *extack)
2208{
2209 struct trie *t = (struct trie *)tb->tb_data;
2210 struct key_vector *l, *tp = t->kv;
2211 t_key key = 0;
2212 int err;
2213
2214 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2215 err = fib_leaf_notify(l, tb, nb, extack);
2216 if (err)
2217 return err;
2218
2219 key = l->key + 1;
2220 /* stop in case of wrap around */
2221 if (key < l->key)
2222 break;
2223 }
2224 return 0;
2225}
2226
2227int fib_notify(struct net *net, struct notifier_block *nb,
2228 struct netlink_ext_ack *extack)
2229{
2230 unsigned int h;
2231 int err;
2232
2233 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2234 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2235 struct fib_table *tb;
2236
2237 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2238 err = fib_table_notify(tb, nb, extack);
2239 if (err)
2240 return err;
2241 }
2242 }
2243 return 0;
2244}
2245
2246static void __trie_free_rcu(struct rcu_head *head)
2247{
2248 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2249#ifdef CONFIG_IP_FIB_TRIE_STATS
2250 struct trie *t = (struct trie *)tb->tb_data;
2251
2252 if (tb->tb_data == tb->__data)
2253 free_percpu(t->stats);
2254#endif /* CONFIG_IP_FIB_TRIE_STATS */
2255 kfree(tb);
2256}
2257
2258void fib_free_table(struct fib_table *tb)
2259{
2260 call_rcu(&tb->rcu, __trie_free_rcu);
2261}
2262
2263static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2264 struct sk_buff *skb, struct netlink_callback *cb,
2265 struct fib_dump_filter *filter)
2266{
2267 unsigned int flags = NLM_F_MULTI;
2268 __be32 xkey = htonl(l->key);
2269 int i, s_i, i_fa, s_fa, err;
2270 struct fib_alias *fa;
2271
2272 if (filter->filter_set ||
2273 !filter->dump_exceptions || !filter->dump_routes)
2274 flags |= NLM_F_DUMP_FILTERED;
2275
2276 s_i = cb->args[4];
2277 s_fa = cb->args[5];
2278 i = 0;
2279
2280 /* rcu_read_lock is hold by caller */
2281 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2282 struct fib_info *fi = fa->fa_info;
2283
2284 if (i < s_i)
2285 goto next;
2286
2287 i_fa = 0;
2288
2289 if (tb->tb_id != fa->tb_id)
2290 goto next;
2291
2292 if (filter->filter_set) {
2293 if (filter->rt_type && fa->fa_type != filter->rt_type)
2294 goto next;
2295
2296 if ((filter->protocol &&
2297 fi->fib_protocol != filter->protocol))
2298 goto next;
2299
2300 if (filter->dev &&
2301 !fib_info_nh_uses_dev(fi, filter->dev))
2302 goto next;
2303 }
2304
2305 if (filter->dump_routes) {
2306 if (!s_fa) {
2307 struct fib_rt_info fri;
2308
2309 fri.fi = fi;
2310 fri.tb_id = tb->tb_id;
2311 fri.dst = xkey;
2312 fri.dst_len = KEYLENGTH - fa->fa_slen;
2313 fri.dscp = fa->fa_dscp;
2314 fri.type = fa->fa_type;
2315 fri.offload = READ_ONCE(fa->offload);
2316 fri.trap = READ_ONCE(fa->trap);
2317 fri.offload_failed = READ_ONCE(fa->offload_failed);
2318 err = fib_dump_info(skb,
2319 NETLINK_CB(cb->skb).portid,
2320 cb->nlh->nlmsg_seq,
2321 RTM_NEWROUTE, &fri, flags);
2322 if (err < 0)
2323 goto stop;
2324 }
2325
2326 i_fa++;
2327 }
2328
2329 if (filter->dump_exceptions) {
2330 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2331 &i_fa, s_fa, flags);
2332 if (err < 0)
2333 goto stop;
2334 }
2335
2336next:
2337 i++;
2338 }
2339
2340 cb->args[4] = i;
2341 return skb->len;
2342
2343stop:
2344 cb->args[4] = i;
2345 cb->args[5] = i_fa;
2346 return err;
2347}
2348
2349/* rcu_read_lock needs to be hold by caller from readside */
2350int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2351 struct netlink_callback *cb, struct fib_dump_filter *filter)
2352{
2353 struct trie *t = (struct trie *)tb->tb_data;
2354 struct key_vector *l, *tp = t->kv;
2355 /* Dump starting at last key.
2356 * Note: 0.0.0.0/0 (ie default) is first key.
2357 */
2358 int count = cb->args[2];
2359 t_key key = cb->args[3];
2360
2361 /* First time here, count and key are both always 0. Count > 0
2362 * and key == 0 means the dump has wrapped around and we are done.
2363 */
2364 if (count && !key)
2365 return 0;
2366
2367 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2368 int err;
2369
2370 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2371 if (err < 0) {
2372 cb->args[3] = key;
2373 cb->args[2] = count;
2374 return err;
2375 }
2376
2377 ++count;
2378 key = l->key + 1;
2379
2380 memset(&cb->args[4], 0,
2381 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2382
2383 /* stop loop if key wrapped back to 0 */
2384 if (key < l->key)
2385 break;
2386 }
2387
2388 cb->args[3] = key;
2389 cb->args[2] = count;
2390
2391 return 0;
2392}
2393
2394void __init fib_trie_init(void)
2395{
2396 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2397 sizeof(struct fib_alias),
2398 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2399
2400 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2401 LEAF_SIZE,
2402 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2403}
2404
2405struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2406{
2407 struct fib_table *tb;
2408 struct trie *t;
2409 size_t sz = sizeof(*tb);
2410
2411 if (!alias)
2412 sz += sizeof(struct trie);
2413
2414 tb = kzalloc(sz, GFP_KERNEL);
2415 if (!tb)
2416 return NULL;
2417
2418 tb->tb_id = id;
2419 tb->tb_num_default = 0;
2420 tb->tb_data = (alias ? alias->__data : tb->__data);
2421
2422 if (alias)
2423 return tb;
2424
2425 t = (struct trie *) tb->tb_data;
2426 t->kv[0].pos = KEYLENGTH;
2427 t->kv[0].slen = KEYLENGTH;
2428#ifdef CONFIG_IP_FIB_TRIE_STATS
2429 t->stats = alloc_percpu(struct trie_use_stats);
2430 if (!t->stats) {
2431 kfree(tb);
2432 tb = NULL;
2433 }
2434#endif
2435
2436 return tb;
2437}
2438
2439#ifdef CONFIG_PROC_FS
2440/* Depth first Trie walk iterator */
2441struct fib_trie_iter {
2442 struct seq_net_private p;
2443 struct fib_table *tb;
2444 struct key_vector *tnode;
2445 unsigned int index;
2446 unsigned int depth;
2447};
2448
2449static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2450{
2451 unsigned long cindex = iter->index;
2452 struct key_vector *pn = iter->tnode;
2453 t_key pkey;
2454
2455 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2456 iter->tnode, iter->index, iter->depth);
2457
2458 while (!IS_TRIE(pn)) {
2459 while (cindex < child_length(pn)) {
2460 struct key_vector *n = get_child_rcu(pn, cindex++);
2461
2462 if (!n)
2463 continue;
2464
2465 if (IS_LEAF(n)) {
2466 iter->tnode = pn;
2467 iter->index = cindex;
2468 } else {
2469 /* push down one level */
2470 iter->tnode = n;
2471 iter->index = 0;
2472 ++iter->depth;
2473 }
2474
2475 return n;
2476 }
2477
2478 /* Current node exhausted, pop back up */
2479 pkey = pn->key;
2480 pn = node_parent_rcu(pn);
2481 cindex = get_index(pkey, pn) + 1;
2482 --iter->depth;
2483 }
2484
2485 /* record root node so further searches know we are done */
2486 iter->tnode = pn;
2487 iter->index = 0;
2488
2489 return NULL;
2490}
2491
2492static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2493 struct trie *t)
2494{
2495 struct key_vector *n, *pn;
2496
2497 if (!t)
2498 return NULL;
2499
2500 pn = t->kv;
2501 n = rcu_dereference(pn->tnode[0]);
2502 if (!n)
2503 return NULL;
2504
2505 if (IS_TNODE(n)) {
2506 iter->tnode = n;
2507 iter->index = 0;
2508 iter->depth = 1;
2509 } else {
2510 iter->tnode = pn;
2511 iter->index = 0;
2512 iter->depth = 0;
2513 }
2514
2515 return n;
2516}
2517
2518static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2519{
2520 struct key_vector *n;
2521 struct fib_trie_iter iter;
2522
2523 memset(s, 0, sizeof(*s));
2524
2525 rcu_read_lock();
2526 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2527 if (IS_LEAF(n)) {
2528 struct fib_alias *fa;
2529
2530 s->leaves++;
2531 s->totdepth += iter.depth;
2532 if (iter.depth > s->maxdepth)
2533 s->maxdepth = iter.depth;
2534
2535 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2536 ++s->prefixes;
2537 } else {
2538 s->tnodes++;
2539 if (n->bits < MAX_STAT_DEPTH)
2540 s->nodesizes[n->bits]++;
2541 s->nullpointers += tn_info(n)->empty_children;
2542 }
2543 }
2544 rcu_read_unlock();
2545}
2546
2547/*
2548 * This outputs /proc/net/fib_triestats
2549 */
2550static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2551{
2552 unsigned int i, max, pointers, bytes, avdepth;
2553
2554 if (stat->leaves)
2555 avdepth = stat->totdepth*100 / stat->leaves;
2556 else
2557 avdepth = 0;
2558
2559 seq_printf(seq, "\tAver depth: %u.%02d\n",
2560 avdepth / 100, avdepth % 100);
2561 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2562
2563 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2564 bytes = LEAF_SIZE * stat->leaves;
2565
2566 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2567 bytes += sizeof(struct fib_alias) * stat->prefixes;
2568
2569 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2570 bytes += TNODE_SIZE(0) * stat->tnodes;
2571
2572 max = MAX_STAT_DEPTH;
2573 while (max > 0 && stat->nodesizes[max-1] == 0)
2574 max--;
2575
2576 pointers = 0;
2577 for (i = 1; i < max; i++)
2578 if (stat->nodesizes[i] != 0) {
2579 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2580 pointers += (1<<i) * stat->nodesizes[i];
2581 }
2582 seq_putc(seq, '\n');
2583 seq_printf(seq, "\tPointers: %u\n", pointers);
2584
2585 bytes += sizeof(struct key_vector *) * pointers;
2586 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2587 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2588}
2589
2590#ifdef CONFIG_IP_FIB_TRIE_STATS
2591static void trie_show_usage(struct seq_file *seq,
2592 const struct trie_use_stats __percpu *stats)
2593{
2594 struct trie_use_stats s = { 0 };
2595 int cpu;
2596
2597 /* loop through all of the CPUs and gather up the stats */
2598 for_each_possible_cpu(cpu) {
2599 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2600
2601 s.gets += pcpu->gets;
2602 s.backtrack += pcpu->backtrack;
2603 s.semantic_match_passed += pcpu->semantic_match_passed;
2604 s.semantic_match_miss += pcpu->semantic_match_miss;
2605 s.null_node_hit += pcpu->null_node_hit;
2606 s.resize_node_skipped += pcpu->resize_node_skipped;
2607 }
2608
2609 seq_printf(seq, "\nCounters:\n---------\n");
2610 seq_printf(seq, "gets = %u\n", s.gets);
2611 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2612 seq_printf(seq, "semantic match passed = %u\n",
2613 s.semantic_match_passed);
2614 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2615 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2616 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2617}
2618#endif /* CONFIG_IP_FIB_TRIE_STATS */
2619
2620static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2621{
2622 if (tb->tb_id == RT_TABLE_LOCAL)
2623 seq_puts(seq, "Local:\n");
2624 else if (tb->tb_id == RT_TABLE_MAIN)
2625 seq_puts(seq, "Main:\n");
2626 else
2627 seq_printf(seq, "Id %d:\n", tb->tb_id);
2628}
2629
2630
2631static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2632{
2633 struct net *net = seq->private;
2634 unsigned int h;
2635
2636 seq_printf(seq,
2637 "Basic info: size of leaf:"
2638 " %zd bytes, size of tnode: %zd bytes.\n",
2639 LEAF_SIZE, TNODE_SIZE(0));
2640
2641 rcu_read_lock();
2642 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2643 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2644 struct fib_table *tb;
2645
2646 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2647 struct trie *t = (struct trie *) tb->tb_data;
2648 struct trie_stat stat;
2649
2650 if (!t)
2651 continue;
2652
2653 fib_table_print(seq, tb);
2654
2655 trie_collect_stats(t, &stat);
2656 trie_show_stats(seq, &stat);
2657#ifdef CONFIG_IP_FIB_TRIE_STATS
2658 trie_show_usage(seq, t->stats);
2659#endif
2660 }
2661 cond_resched_rcu();
2662 }
2663 rcu_read_unlock();
2664
2665 return 0;
2666}
2667
2668static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2669{
2670 struct fib_trie_iter *iter = seq->private;
2671 struct net *net = seq_file_net(seq);
2672 loff_t idx = 0;
2673 unsigned int h;
2674
2675 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2676 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2677 struct fib_table *tb;
2678
2679 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2680 struct key_vector *n;
2681
2682 for (n = fib_trie_get_first(iter,
2683 (struct trie *) tb->tb_data);
2684 n; n = fib_trie_get_next(iter))
2685 if (pos == idx++) {
2686 iter->tb = tb;
2687 return n;
2688 }
2689 }
2690 }
2691
2692 return NULL;
2693}
2694
2695static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2696 __acquires(RCU)
2697{
2698 rcu_read_lock();
2699 return fib_trie_get_idx(seq, *pos);
2700}
2701
2702static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2703{
2704 struct fib_trie_iter *iter = seq->private;
2705 struct net *net = seq_file_net(seq);
2706 struct fib_table *tb = iter->tb;
2707 struct hlist_node *tb_node;
2708 unsigned int h;
2709 struct key_vector *n;
2710
2711 ++*pos;
2712 /* next node in same table */
2713 n = fib_trie_get_next(iter);
2714 if (n)
2715 return n;
2716
2717 /* walk rest of this hash chain */
2718 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2719 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2720 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2721 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2722 if (n)
2723 goto found;
2724 }
2725
2726 /* new hash chain */
2727 while (++h < FIB_TABLE_HASHSZ) {
2728 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2729 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2730 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2731 if (n)
2732 goto found;
2733 }
2734 }
2735 return NULL;
2736
2737found:
2738 iter->tb = tb;
2739 return n;
2740}
2741
2742static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2743 __releases(RCU)
2744{
2745 rcu_read_unlock();
2746}
2747
2748static void seq_indent(struct seq_file *seq, int n)
2749{
2750 while (n-- > 0)
2751 seq_puts(seq, " ");
2752}
2753
2754static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2755{
2756 switch (s) {
2757 case RT_SCOPE_UNIVERSE: return "universe";
2758 case RT_SCOPE_SITE: return "site";
2759 case RT_SCOPE_LINK: return "link";
2760 case RT_SCOPE_HOST: return "host";
2761 case RT_SCOPE_NOWHERE: return "nowhere";
2762 default:
2763 snprintf(buf, len, "scope=%d", s);
2764 return buf;
2765 }
2766}
2767
2768static const char *const rtn_type_names[__RTN_MAX] = {
2769 [RTN_UNSPEC] = "UNSPEC",
2770 [RTN_UNICAST] = "UNICAST",
2771 [RTN_LOCAL] = "LOCAL",
2772 [RTN_BROADCAST] = "BROADCAST",
2773 [RTN_ANYCAST] = "ANYCAST",
2774 [RTN_MULTICAST] = "MULTICAST",
2775 [RTN_BLACKHOLE] = "BLACKHOLE",
2776 [RTN_UNREACHABLE] = "UNREACHABLE",
2777 [RTN_PROHIBIT] = "PROHIBIT",
2778 [RTN_THROW] = "THROW",
2779 [RTN_NAT] = "NAT",
2780 [RTN_XRESOLVE] = "XRESOLVE",
2781};
2782
2783static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2784{
2785 if (t < __RTN_MAX && rtn_type_names[t])
2786 return rtn_type_names[t];
2787 snprintf(buf, len, "type %u", t);
2788 return buf;
2789}
2790
2791/* Pretty print the trie */
2792static int fib_trie_seq_show(struct seq_file *seq, void *v)
2793{
2794 const struct fib_trie_iter *iter = seq->private;
2795 struct key_vector *n = v;
2796
2797 if (IS_TRIE(node_parent_rcu(n)))
2798 fib_table_print(seq, iter->tb);
2799
2800 if (IS_TNODE(n)) {
2801 __be32 prf = htonl(n->key);
2802
2803 seq_indent(seq, iter->depth-1);
2804 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2805 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2806 tn_info(n)->full_children,
2807 tn_info(n)->empty_children);
2808 } else {
2809 __be32 val = htonl(n->key);
2810 struct fib_alias *fa;
2811
2812 seq_indent(seq, iter->depth);
2813 seq_printf(seq, " |-- %pI4\n", &val);
2814
2815 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2816 char buf1[32], buf2[32];
2817
2818 seq_indent(seq, iter->depth + 1);
2819 seq_printf(seq, " /%zu %s %s",
2820 KEYLENGTH - fa->fa_slen,
2821 rtn_scope(buf1, sizeof(buf1),
2822 fa->fa_info->fib_scope),
2823 rtn_type(buf2, sizeof(buf2),
2824 fa->fa_type));
2825 if (fa->fa_dscp)
2826 seq_printf(seq, " tos=%d",
2827 inet_dscp_to_dsfield(fa->fa_dscp));
2828 seq_putc(seq, '\n');
2829 }
2830 }
2831
2832 return 0;
2833}
2834
2835static const struct seq_operations fib_trie_seq_ops = {
2836 .start = fib_trie_seq_start,
2837 .next = fib_trie_seq_next,
2838 .stop = fib_trie_seq_stop,
2839 .show = fib_trie_seq_show,
2840};
2841
2842struct fib_route_iter {
2843 struct seq_net_private p;
2844 struct fib_table *main_tb;
2845 struct key_vector *tnode;
2846 loff_t pos;
2847 t_key key;
2848};
2849
2850static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2851 loff_t pos)
2852{
2853 struct key_vector *l, **tp = &iter->tnode;
2854 t_key key;
2855
2856 /* use cached location of previously found key */
2857 if (iter->pos > 0 && pos >= iter->pos) {
2858 key = iter->key;
2859 } else {
2860 iter->pos = 1;
2861 key = 0;
2862 }
2863
2864 pos -= iter->pos;
2865
2866 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2867 key = l->key + 1;
2868 iter->pos++;
2869 l = NULL;
2870
2871 /* handle unlikely case of a key wrap */
2872 if (!key)
2873 break;
2874 }
2875
2876 if (l)
2877 iter->key = l->key; /* remember it */
2878 else
2879 iter->pos = 0; /* forget it */
2880
2881 return l;
2882}
2883
2884static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2885 __acquires(RCU)
2886{
2887 struct fib_route_iter *iter = seq->private;
2888 struct fib_table *tb;
2889 struct trie *t;
2890
2891 rcu_read_lock();
2892
2893 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2894 if (!tb)
2895 return NULL;
2896
2897 iter->main_tb = tb;
2898 t = (struct trie *)tb->tb_data;
2899 iter->tnode = t->kv;
2900
2901 if (*pos != 0)
2902 return fib_route_get_idx(iter, *pos);
2903
2904 iter->pos = 0;
2905 iter->key = KEY_MAX;
2906
2907 return SEQ_START_TOKEN;
2908}
2909
2910static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2911{
2912 struct fib_route_iter *iter = seq->private;
2913 struct key_vector *l = NULL;
2914 t_key key = iter->key + 1;
2915
2916 ++*pos;
2917
2918 /* only allow key of 0 for start of sequence */
2919 if ((v == SEQ_START_TOKEN) || key)
2920 l = leaf_walk_rcu(&iter->tnode, key);
2921
2922 if (l) {
2923 iter->key = l->key;
2924 iter->pos++;
2925 } else {
2926 iter->pos = 0;
2927 }
2928
2929 return l;
2930}
2931
2932static void fib_route_seq_stop(struct seq_file *seq, void *v)
2933 __releases(RCU)
2934{
2935 rcu_read_unlock();
2936}
2937
2938static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2939{
2940 unsigned int flags = 0;
2941
2942 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2943 flags = RTF_REJECT;
2944 if (fi) {
2945 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2946
2947 if (nhc->nhc_gw.ipv4)
2948 flags |= RTF_GATEWAY;
2949 }
2950 if (mask == htonl(0xFFFFFFFF))
2951 flags |= RTF_HOST;
2952 flags |= RTF_UP;
2953 return flags;
2954}
2955
2956/*
2957 * This outputs /proc/net/route.
2958 * The format of the file is not supposed to be changed
2959 * and needs to be same as fib_hash output to avoid breaking
2960 * legacy utilities
2961 */
2962static int fib_route_seq_show(struct seq_file *seq, void *v)
2963{
2964 struct fib_route_iter *iter = seq->private;
2965 struct fib_table *tb = iter->main_tb;
2966 struct fib_alias *fa;
2967 struct key_vector *l = v;
2968 __be32 prefix;
2969
2970 if (v == SEQ_START_TOKEN) {
2971 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2972 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2973 "\tWindow\tIRTT");
2974 return 0;
2975 }
2976
2977 prefix = htonl(l->key);
2978
2979 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2980 struct fib_info *fi = fa->fa_info;
2981 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2982 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2983
2984 if ((fa->fa_type == RTN_BROADCAST) ||
2985 (fa->fa_type == RTN_MULTICAST))
2986 continue;
2987
2988 if (fa->tb_id != tb->tb_id)
2989 continue;
2990
2991 seq_setwidth(seq, 127);
2992
2993 if (fi) {
2994 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2995 __be32 gw = 0;
2996
2997 if (nhc->nhc_gw_family == AF_INET)
2998 gw = nhc->nhc_gw.ipv4;
2999
3000 seq_printf(seq,
3001 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
3002 "%d\t%08X\t%d\t%u\t%u",
3003 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
3004 prefix, gw, flags, 0, 0,
3005 fi->fib_priority,
3006 mask,
3007 (fi->fib_advmss ?
3008 fi->fib_advmss + 40 : 0),
3009 fi->fib_window,
3010 fi->fib_rtt >> 3);
3011 } else {
3012 seq_printf(seq,
3013 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
3014 "%d\t%08X\t%d\t%u\t%u",
3015 prefix, 0, flags, 0, 0, 0,
3016 mask, 0, 0, 0);
3017 }
3018 seq_pad(seq, '\n');
3019 }
3020
3021 return 0;
3022}
3023
3024static const struct seq_operations fib_route_seq_ops = {
3025 .start = fib_route_seq_start,
3026 .next = fib_route_seq_next,
3027 .stop = fib_route_seq_stop,
3028 .show = fib_route_seq_show,
3029};
3030
3031int __net_init fib_proc_init(struct net *net)
3032{
3033 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
3034 sizeof(struct fib_trie_iter)))
3035 goto out1;
3036
3037 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
3038 fib_triestat_seq_show, NULL))
3039 goto out2;
3040
3041 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
3042 sizeof(struct fib_route_iter)))
3043 goto out3;
3044
3045 return 0;
3046
3047out3:
3048 remove_proc_entry("fib_triestat", net->proc_net);
3049out2:
3050 remove_proc_entry("fib_trie", net->proc_net);
3051out1:
3052 return -ENOMEM;
3053}
3054
3055void __net_exit fib_proc_exit(struct net *net)
3056{
3057 remove_proc_entry("fib_trie", net->proc_net);
3058 remove_proc_entry("fib_triestat", net->proc_net);
3059 remove_proc_entry("route", net->proc_net);
3060}
3061
3062#endif /* CONFIG_PROC_FS */
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 *
4 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
5 * & Swedish University of Agricultural Sciences.
6 *
7 * Jens Laas <jens.laas@data.slu.se> Swedish University of
8 * Agricultural Sciences.
9 *
10 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
11 *
12 * This work is based on the LPC-trie which is originally described in:
13 *
14 * An experimental study of compression methods for dynamic tries
15 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
16 * https://www.csc.kth.se/~snilsson/software/dyntrie2/
17 *
18 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
19 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
20 *
21 * Code from fib_hash has been reused which includes the following header:
22 *
23 * INET An implementation of the TCP/IP protocol suite for the LINUX
24 * operating system. INET is implemented using the BSD Socket
25 * interface as the means of communication with the user level.
26 *
27 * IPv4 FIB: lookup engine and maintenance routines.
28 *
29 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
30 *
31 * Substantial contributions to this work comes from:
32 *
33 * David S. Miller, <davem@davemloft.net>
34 * Stephen Hemminger <shemminger@osdl.org>
35 * Paul E. McKenney <paulmck@us.ibm.com>
36 * Patrick McHardy <kaber@trash.net>
37 */
38#include <linux/cache.h>
39#include <linux/uaccess.h>
40#include <linux/bitops.h>
41#include <linux/types.h>
42#include <linux/kernel.h>
43#include <linux/mm.h>
44#include <linux/string.h>
45#include <linux/socket.h>
46#include <linux/sockios.h>
47#include <linux/errno.h>
48#include <linux/in.h>
49#include <linux/inet.h>
50#include <linux/inetdevice.h>
51#include <linux/netdevice.h>
52#include <linux/if_arp.h>
53#include <linux/proc_fs.h>
54#include <linux/rcupdate.h>
55#include <linux/skbuff.h>
56#include <linux/netlink.h>
57#include <linux/init.h>
58#include <linux/list.h>
59#include <linux/slab.h>
60#include <linux/export.h>
61#include <linux/vmalloc.h>
62#include <linux/notifier.h>
63#include <net/net_namespace.h>
64#include <net/ip.h>
65#include <net/protocol.h>
66#include <net/route.h>
67#include <net/tcp.h>
68#include <net/sock.h>
69#include <net/ip_fib.h>
70#include <net/fib_notifier.h>
71#include <trace/events/fib.h>
72#include "fib_lookup.h"
73
74static int call_fib_entry_notifier(struct notifier_block *nb,
75 enum fib_event_type event_type, u32 dst,
76 int dst_len, struct fib_alias *fa,
77 struct netlink_ext_ack *extack)
78{
79 struct fib_entry_notifier_info info = {
80 .info.extack = extack,
81 .dst = dst,
82 .dst_len = dst_len,
83 .fi = fa->fa_info,
84 .tos = fa->fa_tos,
85 .type = fa->fa_type,
86 .tb_id = fa->tb_id,
87 };
88 return call_fib4_notifier(nb, event_type, &info.info);
89}
90
91static int call_fib_entry_notifiers(struct net *net,
92 enum fib_event_type event_type, u32 dst,
93 int dst_len, struct fib_alias *fa,
94 struct netlink_ext_ack *extack)
95{
96 struct fib_entry_notifier_info info = {
97 .info.extack = extack,
98 .dst = dst,
99 .dst_len = dst_len,
100 .fi = fa->fa_info,
101 .tos = fa->fa_tos,
102 .type = fa->fa_type,
103 .tb_id = fa->tb_id,
104 };
105 return call_fib4_notifiers(net, event_type, &info.info);
106}
107
108#define MAX_STAT_DEPTH 32
109
110#define KEYLENGTH (8*sizeof(t_key))
111#define KEY_MAX ((t_key)~0)
112
113typedef unsigned int t_key;
114
115#define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
116#define IS_TNODE(n) ((n)->bits)
117#define IS_LEAF(n) (!(n)->bits)
118
119struct key_vector {
120 t_key key;
121 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
122 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
123 unsigned char slen;
124 union {
125 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
126 struct hlist_head leaf;
127 /* This array is valid if (pos | bits) > 0 (TNODE) */
128 struct key_vector __rcu *tnode[0];
129 };
130};
131
132struct tnode {
133 struct rcu_head rcu;
134 t_key empty_children; /* KEYLENGTH bits needed */
135 t_key full_children; /* KEYLENGTH bits needed */
136 struct key_vector __rcu *parent;
137 struct key_vector kv[1];
138#define tn_bits kv[0].bits
139};
140
141#define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
142#define LEAF_SIZE TNODE_SIZE(1)
143
144#ifdef CONFIG_IP_FIB_TRIE_STATS
145struct trie_use_stats {
146 unsigned int gets;
147 unsigned int backtrack;
148 unsigned int semantic_match_passed;
149 unsigned int semantic_match_miss;
150 unsigned int null_node_hit;
151 unsigned int resize_node_skipped;
152};
153#endif
154
155struct trie_stat {
156 unsigned int totdepth;
157 unsigned int maxdepth;
158 unsigned int tnodes;
159 unsigned int leaves;
160 unsigned int nullpointers;
161 unsigned int prefixes;
162 unsigned int nodesizes[MAX_STAT_DEPTH];
163};
164
165struct trie {
166 struct key_vector kv[1];
167#ifdef CONFIG_IP_FIB_TRIE_STATS
168 struct trie_use_stats __percpu *stats;
169#endif
170};
171
172static struct key_vector *resize(struct trie *t, struct key_vector *tn);
173static unsigned int tnode_free_size;
174
175/*
176 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
177 * especially useful before resizing the root node with PREEMPT_NONE configs;
178 * the value was obtained experimentally, aiming to avoid visible slowdown.
179 */
180unsigned int sysctl_fib_sync_mem = 512 * 1024;
181unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
182unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
183
184static struct kmem_cache *fn_alias_kmem __ro_after_init;
185static struct kmem_cache *trie_leaf_kmem __ro_after_init;
186
187static inline struct tnode *tn_info(struct key_vector *kv)
188{
189 return container_of(kv, struct tnode, kv[0]);
190}
191
192/* caller must hold RTNL */
193#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
194#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
195
196/* caller must hold RCU read lock or RTNL */
197#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
198#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
199
200/* wrapper for rcu_assign_pointer */
201static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
202{
203 if (n)
204 rcu_assign_pointer(tn_info(n)->parent, tp);
205}
206
207#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
208
209/* This provides us with the number of children in this node, in the case of a
210 * leaf this will return 0 meaning none of the children are accessible.
211 */
212static inline unsigned long child_length(const struct key_vector *tn)
213{
214 return (1ul << tn->bits) & ~(1ul);
215}
216
217#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
218
219static inline unsigned long get_index(t_key key, struct key_vector *kv)
220{
221 unsigned long index = key ^ kv->key;
222
223 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
224 return 0;
225
226 return index >> kv->pos;
227}
228
229/* To understand this stuff, an understanding of keys and all their bits is
230 * necessary. Every node in the trie has a key associated with it, but not
231 * all of the bits in that key are significant.
232 *
233 * Consider a node 'n' and its parent 'tp'.
234 *
235 * If n is a leaf, every bit in its key is significant. Its presence is
236 * necessitated by path compression, since during a tree traversal (when
237 * searching for a leaf - unless we are doing an insertion) we will completely
238 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
239 * a potentially successful search, that we have indeed been walking the
240 * correct key path.
241 *
242 * Note that we can never "miss" the correct key in the tree if present by
243 * following the wrong path. Path compression ensures that segments of the key
244 * that are the same for all keys with a given prefix are skipped, but the
245 * skipped part *is* identical for each node in the subtrie below the skipped
246 * bit! trie_insert() in this implementation takes care of that.
247 *
248 * if n is an internal node - a 'tnode' here, the various parts of its key
249 * have many different meanings.
250 *
251 * Example:
252 * _________________________________________________________________
253 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
254 * -----------------------------------------------------------------
255 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
256 *
257 * _________________________________________________________________
258 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
259 * -----------------------------------------------------------------
260 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
261 *
262 * tp->pos = 22
263 * tp->bits = 3
264 * n->pos = 13
265 * n->bits = 4
266 *
267 * First, let's just ignore the bits that come before the parent tp, that is
268 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
269 * point we do not use them for anything.
270 *
271 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
272 * index into the parent's child array. That is, they will be used to find
273 * 'n' among tp's children.
274 *
275 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
276 * for the node n.
277 *
278 * All the bits we have seen so far are significant to the node n. The rest
279 * of the bits are really not needed or indeed known in n->key.
280 *
281 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
282 * n's child array, and will of course be different for each child.
283 *
284 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
285 * at this point.
286 */
287
288static const int halve_threshold = 25;
289static const int inflate_threshold = 50;
290static const int halve_threshold_root = 15;
291static const int inflate_threshold_root = 30;
292
293static void __alias_free_mem(struct rcu_head *head)
294{
295 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
296 kmem_cache_free(fn_alias_kmem, fa);
297}
298
299static inline void alias_free_mem_rcu(struct fib_alias *fa)
300{
301 call_rcu(&fa->rcu, __alias_free_mem);
302}
303
304#define TNODE_VMALLOC_MAX \
305 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
306
307static void __node_free_rcu(struct rcu_head *head)
308{
309 struct tnode *n = container_of(head, struct tnode, rcu);
310
311 if (!n->tn_bits)
312 kmem_cache_free(trie_leaf_kmem, n);
313 else
314 kvfree(n);
315}
316
317#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
318
319static struct tnode *tnode_alloc(int bits)
320{
321 size_t size;
322
323 /* verify bits is within bounds */
324 if (bits > TNODE_VMALLOC_MAX)
325 return NULL;
326
327 /* determine size and verify it is non-zero and didn't overflow */
328 size = TNODE_SIZE(1ul << bits);
329
330 if (size <= PAGE_SIZE)
331 return kzalloc(size, GFP_KERNEL);
332 else
333 return vzalloc(size);
334}
335
336static inline void empty_child_inc(struct key_vector *n)
337{
338 tn_info(n)->empty_children++;
339
340 if (!tn_info(n)->empty_children)
341 tn_info(n)->full_children++;
342}
343
344static inline void empty_child_dec(struct key_vector *n)
345{
346 if (!tn_info(n)->empty_children)
347 tn_info(n)->full_children--;
348
349 tn_info(n)->empty_children--;
350}
351
352static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
353{
354 struct key_vector *l;
355 struct tnode *kv;
356
357 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
358 if (!kv)
359 return NULL;
360
361 /* initialize key vector */
362 l = kv->kv;
363 l->key = key;
364 l->pos = 0;
365 l->bits = 0;
366 l->slen = fa->fa_slen;
367
368 /* link leaf to fib alias */
369 INIT_HLIST_HEAD(&l->leaf);
370 hlist_add_head(&fa->fa_list, &l->leaf);
371
372 return l;
373}
374
375static struct key_vector *tnode_new(t_key key, int pos, int bits)
376{
377 unsigned int shift = pos + bits;
378 struct key_vector *tn;
379 struct tnode *tnode;
380
381 /* verify bits and pos their msb bits clear and values are valid */
382 BUG_ON(!bits || (shift > KEYLENGTH));
383
384 tnode = tnode_alloc(bits);
385 if (!tnode)
386 return NULL;
387
388 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
389 sizeof(struct key_vector *) << bits);
390
391 if (bits == KEYLENGTH)
392 tnode->full_children = 1;
393 else
394 tnode->empty_children = 1ul << bits;
395
396 tn = tnode->kv;
397 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
398 tn->pos = pos;
399 tn->bits = bits;
400 tn->slen = pos;
401
402 return tn;
403}
404
405/* Check whether a tnode 'n' is "full", i.e. it is an internal node
406 * and no bits are skipped. See discussion in dyntree paper p. 6
407 */
408static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
409{
410 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
411}
412
413/* Add a child at position i overwriting the old value.
414 * Update the value of full_children and empty_children.
415 */
416static void put_child(struct key_vector *tn, unsigned long i,
417 struct key_vector *n)
418{
419 struct key_vector *chi = get_child(tn, i);
420 int isfull, wasfull;
421
422 BUG_ON(i >= child_length(tn));
423
424 /* update emptyChildren, overflow into fullChildren */
425 if (!n && chi)
426 empty_child_inc(tn);
427 if (n && !chi)
428 empty_child_dec(tn);
429
430 /* update fullChildren */
431 wasfull = tnode_full(tn, chi);
432 isfull = tnode_full(tn, n);
433
434 if (wasfull && !isfull)
435 tn_info(tn)->full_children--;
436 else if (!wasfull && isfull)
437 tn_info(tn)->full_children++;
438
439 if (n && (tn->slen < n->slen))
440 tn->slen = n->slen;
441
442 rcu_assign_pointer(tn->tnode[i], n);
443}
444
445static void update_children(struct key_vector *tn)
446{
447 unsigned long i;
448
449 /* update all of the child parent pointers */
450 for (i = child_length(tn); i;) {
451 struct key_vector *inode = get_child(tn, --i);
452
453 if (!inode)
454 continue;
455
456 /* Either update the children of a tnode that
457 * already belongs to us or update the child
458 * to point to ourselves.
459 */
460 if (node_parent(inode) == tn)
461 update_children(inode);
462 else
463 node_set_parent(inode, tn);
464 }
465}
466
467static inline void put_child_root(struct key_vector *tp, t_key key,
468 struct key_vector *n)
469{
470 if (IS_TRIE(tp))
471 rcu_assign_pointer(tp->tnode[0], n);
472 else
473 put_child(tp, get_index(key, tp), n);
474}
475
476static inline void tnode_free_init(struct key_vector *tn)
477{
478 tn_info(tn)->rcu.next = NULL;
479}
480
481static inline void tnode_free_append(struct key_vector *tn,
482 struct key_vector *n)
483{
484 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
485 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
486}
487
488static void tnode_free(struct key_vector *tn)
489{
490 struct callback_head *head = &tn_info(tn)->rcu;
491
492 while (head) {
493 head = head->next;
494 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
495 node_free(tn);
496
497 tn = container_of(head, struct tnode, rcu)->kv;
498 }
499
500 if (tnode_free_size >= sysctl_fib_sync_mem) {
501 tnode_free_size = 0;
502 synchronize_rcu();
503 }
504}
505
506static struct key_vector *replace(struct trie *t,
507 struct key_vector *oldtnode,
508 struct key_vector *tn)
509{
510 struct key_vector *tp = node_parent(oldtnode);
511 unsigned long i;
512
513 /* setup the parent pointer out of and back into this node */
514 NODE_INIT_PARENT(tn, tp);
515 put_child_root(tp, tn->key, tn);
516
517 /* update all of the child parent pointers */
518 update_children(tn);
519
520 /* all pointers should be clean so we are done */
521 tnode_free(oldtnode);
522
523 /* resize children now that oldtnode is freed */
524 for (i = child_length(tn); i;) {
525 struct key_vector *inode = get_child(tn, --i);
526
527 /* resize child node */
528 if (tnode_full(tn, inode))
529 tn = resize(t, inode);
530 }
531
532 return tp;
533}
534
535static struct key_vector *inflate(struct trie *t,
536 struct key_vector *oldtnode)
537{
538 struct key_vector *tn;
539 unsigned long i;
540 t_key m;
541
542 pr_debug("In inflate\n");
543
544 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
545 if (!tn)
546 goto notnode;
547
548 /* prepare oldtnode to be freed */
549 tnode_free_init(oldtnode);
550
551 /* Assemble all of the pointers in our cluster, in this case that
552 * represents all of the pointers out of our allocated nodes that
553 * point to existing tnodes and the links between our allocated
554 * nodes.
555 */
556 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
557 struct key_vector *inode = get_child(oldtnode, --i);
558 struct key_vector *node0, *node1;
559 unsigned long j, k;
560
561 /* An empty child */
562 if (!inode)
563 continue;
564
565 /* A leaf or an internal node with skipped bits */
566 if (!tnode_full(oldtnode, inode)) {
567 put_child(tn, get_index(inode->key, tn), inode);
568 continue;
569 }
570
571 /* drop the node in the old tnode free list */
572 tnode_free_append(oldtnode, inode);
573
574 /* An internal node with two children */
575 if (inode->bits == 1) {
576 put_child(tn, 2 * i + 1, get_child(inode, 1));
577 put_child(tn, 2 * i, get_child(inode, 0));
578 continue;
579 }
580
581 /* We will replace this node 'inode' with two new
582 * ones, 'node0' and 'node1', each with half of the
583 * original children. The two new nodes will have
584 * a position one bit further down the key and this
585 * means that the "significant" part of their keys
586 * (see the discussion near the top of this file)
587 * will differ by one bit, which will be "0" in
588 * node0's key and "1" in node1's key. Since we are
589 * moving the key position by one step, the bit that
590 * we are moving away from - the bit at position
591 * (tn->pos) - is the one that will differ between
592 * node0 and node1. So... we synthesize that bit in the
593 * two new keys.
594 */
595 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
596 if (!node1)
597 goto nomem;
598 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
599
600 tnode_free_append(tn, node1);
601 if (!node0)
602 goto nomem;
603 tnode_free_append(tn, node0);
604
605 /* populate child pointers in new nodes */
606 for (k = child_length(inode), j = k / 2; j;) {
607 put_child(node1, --j, get_child(inode, --k));
608 put_child(node0, j, get_child(inode, j));
609 put_child(node1, --j, get_child(inode, --k));
610 put_child(node0, j, get_child(inode, j));
611 }
612
613 /* link new nodes to parent */
614 NODE_INIT_PARENT(node1, tn);
615 NODE_INIT_PARENT(node0, tn);
616
617 /* link parent to nodes */
618 put_child(tn, 2 * i + 1, node1);
619 put_child(tn, 2 * i, node0);
620 }
621
622 /* setup the parent pointers into and out of this node */
623 return replace(t, oldtnode, tn);
624nomem:
625 /* all pointers should be clean so we are done */
626 tnode_free(tn);
627notnode:
628 return NULL;
629}
630
631static struct key_vector *halve(struct trie *t,
632 struct key_vector *oldtnode)
633{
634 struct key_vector *tn;
635 unsigned long i;
636
637 pr_debug("In halve\n");
638
639 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
640 if (!tn)
641 goto notnode;
642
643 /* prepare oldtnode to be freed */
644 tnode_free_init(oldtnode);
645
646 /* Assemble all of the pointers in our cluster, in this case that
647 * represents all of the pointers out of our allocated nodes that
648 * point to existing tnodes and the links between our allocated
649 * nodes.
650 */
651 for (i = child_length(oldtnode); i;) {
652 struct key_vector *node1 = get_child(oldtnode, --i);
653 struct key_vector *node0 = get_child(oldtnode, --i);
654 struct key_vector *inode;
655
656 /* At least one of the children is empty */
657 if (!node1 || !node0) {
658 put_child(tn, i / 2, node1 ? : node0);
659 continue;
660 }
661
662 /* Two nonempty children */
663 inode = tnode_new(node0->key, oldtnode->pos, 1);
664 if (!inode)
665 goto nomem;
666 tnode_free_append(tn, inode);
667
668 /* initialize pointers out of node */
669 put_child(inode, 1, node1);
670 put_child(inode, 0, node0);
671 NODE_INIT_PARENT(inode, tn);
672
673 /* link parent to node */
674 put_child(tn, i / 2, inode);
675 }
676
677 /* setup the parent pointers into and out of this node */
678 return replace(t, oldtnode, tn);
679nomem:
680 /* all pointers should be clean so we are done */
681 tnode_free(tn);
682notnode:
683 return NULL;
684}
685
686static struct key_vector *collapse(struct trie *t,
687 struct key_vector *oldtnode)
688{
689 struct key_vector *n, *tp;
690 unsigned long i;
691
692 /* scan the tnode looking for that one child that might still exist */
693 for (n = NULL, i = child_length(oldtnode); !n && i;)
694 n = get_child(oldtnode, --i);
695
696 /* compress one level */
697 tp = node_parent(oldtnode);
698 put_child_root(tp, oldtnode->key, n);
699 node_set_parent(n, tp);
700
701 /* drop dead node */
702 node_free(oldtnode);
703
704 return tp;
705}
706
707static unsigned char update_suffix(struct key_vector *tn)
708{
709 unsigned char slen = tn->pos;
710 unsigned long stride, i;
711 unsigned char slen_max;
712
713 /* only vector 0 can have a suffix length greater than or equal to
714 * tn->pos + tn->bits, the second highest node will have a suffix
715 * length at most of tn->pos + tn->bits - 1
716 */
717 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
718
719 /* search though the list of children looking for nodes that might
720 * have a suffix greater than the one we currently have. This is
721 * why we start with a stride of 2 since a stride of 1 would
722 * represent the nodes with suffix length equal to tn->pos
723 */
724 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
725 struct key_vector *n = get_child(tn, i);
726
727 if (!n || (n->slen <= slen))
728 continue;
729
730 /* update stride and slen based on new value */
731 stride <<= (n->slen - slen);
732 slen = n->slen;
733 i &= ~(stride - 1);
734
735 /* stop searching if we have hit the maximum possible value */
736 if (slen >= slen_max)
737 break;
738 }
739
740 tn->slen = slen;
741
742 return slen;
743}
744
745/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
746 * the Helsinki University of Technology and Matti Tikkanen of Nokia
747 * Telecommunications, page 6:
748 * "A node is doubled if the ratio of non-empty children to all
749 * children in the *doubled* node is at least 'high'."
750 *
751 * 'high' in this instance is the variable 'inflate_threshold'. It
752 * is expressed as a percentage, so we multiply it with
753 * child_length() and instead of multiplying by 2 (since the
754 * child array will be doubled by inflate()) and multiplying
755 * the left-hand side by 100 (to handle the percentage thing) we
756 * multiply the left-hand side by 50.
757 *
758 * The left-hand side may look a bit weird: child_length(tn)
759 * - tn->empty_children is of course the number of non-null children
760 * in the current node. tn->full_children is the number of "full"
761 * children, that is non-null tnodes with a skip value of 0.
762 * All of those will be doubled in the resulting inflated tnode, so
763 * we just count them one extra time here.
764 *
765 * A clearer way to write this would be:
766 *
767 * to_be_doubled = tn->full_children;
768 * not_to_be_doubled = child_length(tn) - tn->empty_children -
769 * tn->full_children;
770 *
771 * new_child_length = child_length(tn) * 2;
772 *
773 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
774 * new_child_length;
775 * if (new_fill_factor >= inflate_threshold)
776 *
777 * ...and so on, tho it would mess up the while () loop.
778 *
779 * anyway,
780 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
781 * inflate_threshold
782 *
783 * avoid a division:
784 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
785 * inflate_threshold * new_child_length
786 *
787 * expand not_to_be_doubled and to_be_doubled, and shorten:
788 * 100 * (child_length(tn) - tn->empty_children +
789 * tn->full_children) >= inflate_threshold * new_child_length
790 *
791 * expand new_child_length:
792 * 100 * (child_length(tn) - tn->empty_children +
793 * tn->full_children) >=
794 * inflate_threshold * child_length(tn) * 2
795 *
796 * shorten again:
797 * 50 * (tn->full_children + child_length(tn) -
798 * tn->empty_children) >= inflate_threshold *
799 * child_length(tn)
800 *
801 */
802static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
803{
804 unsigned long used = child_length(tn);
805 unsigned long threshold = used;
806
807 /* Keep root node larger */
808 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
809 used -= tn_info(tn)->empty_children;
810 used += tn_info(tn)->full_children;
811
812 /* if bits == KEYLENGTH then pos = 0, and will fail below */
813
814 return (used > 1) && tn->pos && ((50 * used) >= threshold);
815}
816
817static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
818{
819 unsigned long used = child_length(tn);
820 unsigned long threshold = used;
821
822 /* Keep root node larger */
823 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
824 used -= tn_info(tn)->empty_children;
825
826 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
827
828 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
829}
830
831static inline bool should_collapse(struct key_vector *tn)
832{
833 unsigned long used = child_length(tn);
834
835 used -= tn_info(tn)->empty_children;
836
837 /* account for bits == KEYLENGTH case */
838 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
839 used -= KEY_MAX;
840
841 /* One child or none, time to drop us from the trie */
842 return used < 2;
843}
844
845#define MAX_WORK 10
846static struct key_vector *resize(struct trie *t, struct key_vector *tn)
847{
848#ifdef CONFIG_IP_FIB_TRIE_STATS
849 struct trie_use_stats __percpu *stats = t->stats;
850#endif
851 struct key_vector *tp = node_parent(tn);
852 unsigned long cindex = get_index(tn->key, tp);
853 int max_work = MAX_WORK;
854
855 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
856 tn, inflate_threshold, halve_threshold);
857
858 /* track the tnode via the pointer from the parent instead of
859 * doing it ourselves. This way we can let RCU fully do its
860 * thing without us interfering
861 */
862 BUG_ON(tn != get_child(tp, cindex));
863
864 /* Double as long as the resulting node has a number of
865 * nonempty nodes that are above the threshold.
866 */
867 while (should_inflate(tp, tn) && max_work) {
868 tp = inflate(t, tn);
869 if (!tp) {
870#ifdef CONFIG_IP_FIB_TRIE_STATS
871 this_cpu_inc(stats->resize_node_skipped);
872#endif
873 break;
874 }
875
876 max_work--;
877 tn = get_child(tp, cindex);
878 }
879
880 /* update parent in case inflate failed */
881 tp = node_parent(tn);
882
883 /* Return if at least one inflate is run */
884 if (max_work != MAX_WORK)
885 return tp;
886
887 /* Halve as long as the number of empty children in this
888 * node is above threshold.
889 */
890 while (should_halve(tp, tn) && max_work) {
891 tp = halve(t, tn);
892 if (!tp) {
893#ifdef CONFIG_IP_FIB_TRIE_STATS
894 this_cpu_inc(stats->resize_node_skipped);
895#endif
896 break;
897 }
898
899 max_work--;
900 tn = get_child(tp, cindex);
901 }
902
903 /* Only one child remains */
904 if (should_collapse(tn))
905 return collapse(t, tn);
906
907 /* update parent in case halve failed */
908 return node_parent(tn);
909}
910
911static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
912{
913 unsigned char node_slen = tn->slen;
914
915 while ((node_slen > tn->pos) && (node_slen > slen)) {
916 slen = update_suffix(tn);
917 if (node_slen == slen)
918 break;
919
920 tn = node_parent(tn);
921 node_slen = tn->slen;
922 }
923}
924
925static void node_push_suffix(struct key_vector *tn, unsigned char slen)
926{
927 while (tn->slen < slen) {
928 tn->slen = slen;
929 tn = node_parent(tn);
930 }
931}
932
933/* rcu_read_lock needs to be hold by caller from readside */
934static struct key_vector *fib_find_node(struct trie *t,
935 struct key_vector **tp, u32 key)
936{
937 struct key_vector *pn, *n = t->kv;
938 unsigned long index = 0;
939
940 do {
941 pn = n;
942 n = get_child_rcu(n, index);
943
944 if (!n)
945 break;
946
947 index = get_cindex(key, n);
948
949 /* This bit of code is a bit tricky but it combines multiple
950 * checks into a single check. The prefix consists of the
951 * prefix plus zeros for the bits in the cindex. The index
952 * is the difference between the key and this value. From
953 * this we can actually derive several pieces of data.
954 * if (index >= (1ul << bits))
955 * we have a mismatch in skip bits and failed
956 * else
957 * we know the value is cindex
958 *
959 * This check is safe even if bits == KEYLENGTH due to the
960 * fact that we can only allocate a node with 32 bits if a
961 * long is greater than 32 bits.
962 */
963 if (index >= (1ul << n->bits)) {
964 n = NULL;
965 break;
966 }
967
968 /* keep searching until we find a perfect match leaf or NULL */
969 } while (IS_TNODE(n));
970
971 *tp = pn;
972
973 return n;
974}
975
976/* Return the first fib alias matching TOS with
977 * priority less than or equal to PRIO.
978 * If 'find_first' is set, return the first matching
979 * fib alias, regardless of TOS and priority.
980 */
981static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
982 u8 tos, u32 prio, u32 tb_id,
983 bool find_first)
984{
985 struct fib_alias *fa;
986
987 if (!fah)
988 return NULL;
989
990 hlist_for_each_entry(fa, fah, fa_list) {
991 if (fa->fa_slen < slen)
992 continue;
993 if (fa->fa_slen != slen)
994 break;
995 if (fa->tb_id > tb_id)
996 continue;
997 if (fa->tb_id != tb_id)
998 break;
999 if (find_first)
1000 return fa;
1001 if (fa->fa_tos > tos)
1002 continue;
1003 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1004 return fa;
1005 }
1006
1007 return NULL;
1008}
1009
1010static struct fib_alias *
1011fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1012{
1013 u8 slen = KEYLENGTH - fri->dst_len;
1014 struct key_vector *l, *tp;
1015 struct fib_table *tb;
1016 struct fib_alias *fa;
1017 struct trie *t;
1018
1019 tb = fib_get_table(net, fri->tb_id);
1020 if (!tb)
1021 return NULL;
1022
1023 t = (struct trie *)tb->tb_data;
1024 l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1025 if (!l)
1026 return NULL;
1027
1028 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1029 if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1030 fa->fa_tos == fri->tos && fa->fa_info == fri->fi &&
1031 fa->fa_type == fri->type)
1032 return fa;
1033 }
1034
1035 return NULL;
1036}
1037
1038void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1039{
1040 struct fib_alias *fa_match;
1041 struct sk_buff *skb;
1042 int err;
1043
1044 rcu_read_lock();
1045
1046 fa_match = fib_find_matching_alias(net, fri);
1047 if (!fa_match)
1048 goto out;
1049
1050 if (fa_match->offload == fri->offload && fa_match->trap == fri->trap &&
1051 fa_match->offload_failed == fri->offload_failed)
1052 goto out;
1053
1054 fa_match->offload = fri->offload;
1055 fa_match->trap = fri->trap;
1056
1057 /* 2 means send notifications only if offload_failed was changed. */
1058 if (net->ipv4.sysctl_fib_notify_on_flag_change == 2 &&
1059 fa_match->offload_failed == fri->offload_failed)
1060 goto out;
1061
1062 fa_match->offload_failed = fri->offload_failed;
1063
1064 if (!net->ipv4.sysctl_fib_notify_on_flag_change)
1065 goto out;
1066
1067 skb = nlmsg_new(fib_nlmsg_size(fa_match->fa_info), GFP_ATOMIC);
1068 if (!skb) {
1069 err = -ENOBUFS;
1070 goto errout;
1071 }
1072
1073 err = fib_dump_info(skb, 0, 0, RTM_NEWROUTE, fri, 0);
1074 if (err < 0) {
1075 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
1076 WARN_ON(err == -EMSGSIZE);
1077 kfree_skb(skb);
1078 goto errout;
1079 }
1080
1081 rtnl_notify(skb, net, 0, RTNLGRP_IPV4_ROUTE, NULL, GFP_ATOMIC);
1082 goto out;
1083
1084errout:
1085 rtnl_set_sk_err(net, RTNLGRP_IPV4_ROUTE, err);
1086out:
1087 rcu_read_unlock();
1088}
1089EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1090
1091static void trie_rebalance(struct trie *t, struct key_vector *tn)
1092{
1093 while (!IS_TRIE(tn))
1094 tn = resize(t, tn);
1095}
1096
1097static int fib_insert_node(struct trie *t, struct key_vector *tp,
1098 struct fib_alias *new, t_key key)
1099{
1100 struct key_vector *n, *l;
1101
1102 l = leaf_new(key, new);
1103 if (!l)
1104 goto noleaf;
1105
1106 /* retrieve child from parent node */
1107 n = get_child(tp, get_index(key, tp));
1108
1109 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1110 *
1111 * Add a new tnode here
1112 * first tnode need some special handling
1113 * leaves us in position for handling as case 3
1114 */
1115 if (n) {
1116 struct key_vector *tn;
1117
1118 tn = tnode_new(key, __fls(key ^ n->key), 1);
1119 if (!tn)
1120 goto notnode;
1121
1122 /* initialize routes out of node */
1123 NODE_INIT_PARENT(tn, tp);
1124 put_child(tn, get_index(key, tn) ^ 1, n);
1125
1126 /* start adding routes into the node */
1127 put_child_root(tp, key, tn);
1128 node_set_parent(n, tn);
1129
1130 /* parent now has a NULL spot where the leaf can go */
1131 tp = tn;
1132 }
1133
1134 /* Case 3: n is NULL, and will just insert a new leaf */
1135 node_push_suffix(tp, new->fa_slen);
1136 NODE_INIT_PARENT(l, tp);
1137 put_child_root(tp, key, l);
1138 trie_rebalance(t, tp);
1139
1140 return 0;
1141notnode:
1142 node_free(l);
1143noleaf:
1144 return -ENOMEM;
1145}
1146
1147static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1148 struct key_vector *l, struct fib_alias *new,
1149 struct fib_alias *fa, t_key key)
1150{
1151 if (!l)
1152 return fib_insert_node(t, tp, new, key);
1153
1154 if (fa) {
1155 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1156 } else {
1157 struct fib_alias *last;
1158
1159 hlist_for_each_entry(last, &l->leaf, fa_list) {
1160 if (new->fa_slen < last->fa_slen)
1161 break;
1162 if ((new->fa_slen == last->fa_slen) &&
1163 (new->tb_id > last->tb_id))
1164 break;
1165 fa = last;
1166 }
1167
1168 if (fa)
1169 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1170 else
1171 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1172 }
1173
1174 /* if we added to the tail node then we need to update slen */
1175 if (l->slen < new->fa_slen) {
1176 l->slen = new->fa_slen;
1177 node_push_suffix(tp, new->fa_slen);
1178 }
1179
1180 return 0;
1181}
1182
1183static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1184{
1185 if (plen > KEYLENGTH) {
1186 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1187 return false;
1188 }
1189
1190 if ((plen < KEYLENGTH) && (key << plen)) {
1191 NL_SET_ERR_MSG(extack,
1192 "Invalid prefix for given prefix length");
1193 return false;
1194 }
1195
1196 return true;
1197}
1198
1199static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1200 struct key_vector *l, struct fib_alias *old);
1201
1202/* Caller must hold RTNL. */
1203int fib_table_insert(struct net *net, struct fib_table *tb,
1204 struct fib_config *cfg, struct netlink_ext_ack *extack)
1205{
1206 struct trie *t = (struct trie *)tb->tb_data;
1207 struct fib_alias *fa, *new_fa;
1208 struct key_vector *l, *tp;
1209 u16 nlflags = NLM_F_EXCL;
1210 struct fib_info *fi;
1211 u8 plen = cfg->fc_dst_len;
1212 u8 slen = KEYLENGTH - plen;
1213 u8 tos = cfg->fc_tos;
1214 u32 key;
1215 int err;
1216
1217 key = ntohl(cfg->fc_dst);
1218
1219 if (!fib_valid_key_len(key, plen, extack))
1220 return -EINVAL;
1221
1222 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1223
1224 fi = fib_create_info(cfg, extack);
1225 if (IS_ERR(fi)) {
1226 err = PTR_ERR(fi);
1227 goto err;
1228 }
1229
1230 l = fib_find_node(t, &tp, key);
1231 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1232 tb->tb_id, false) : NULL;
1233
1234 /* Now fa, if non-NULL, points to the first fib alias
1235 * with the same keys [prefix,tos,priority], if such key already
1236 * exists or to the node before which we will insert new one.
1237 *
1238 * If fa is NULL, we will need to allocate a new one and
1239 * insert to the tail of the section matching the suffix length
1240 * of the new alias.
1241 */
1242
1243 if (fa && fa->fa_tos == tos &&
1244 fa->fa_info->fib_priority == fi->fib_priority) {
1245 struct fib_alias *fa_first, *fa_match;
1246
1247 err = -EEXIST;
1248 if (cfg->fc_nlflags & NLM_F_EXCL)
1249 goto out;
1250
1251 nlflags &= ~NLM_F_EXCL;
1252
1253 /* We have 2 goals:
1254 * 1. Find exact match for type, scope, fib_info to avoid
1255 * duplicate routes
1256 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1257 */
1258 fa_match = NULL;
1259 fa_first = fa;
1260 hlist_for_each_entry_from(fa, fa_list) {
1261 if ((fa->fa_slen != slen) ||
1262 (fa->tb_id != tb->tb_id) ||
1263 (fa->fa_tos != tos))
1264 break;
1265 if (fa->fa_info->fib_priority != fi->fib_priority)
1266 break;
1267 if (fa->fa_type == cfg->fc_type &&
1268 fa->fa_info == fi) {
1269 fa_match = fa;
1270 break;
1271 }
1272 }
1273
1274 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1275 struct fib_info *fi_drop;
1276 u8 state;
1277
1278 nlflags |= NLM_F_REPLACE;
1279 fa = fa_first;
1280 if (fa_match) {
1281 if (fa == fa_match)
1282 err = 0;
1283 goto out;
1284 }
1285 err = -ENOBUFS;
1286 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1287 if (!new_fa)
1288 goto out;
1289
1290 fi_drop = fa->fa_info;
1291 new_fa->fa_tos = fa->fa_tos;
1292 new_fa->fa_info = fi;
1293 new_fa->fa_type = cfg->fc_type;
1294 state = fa->fa_state;
1295 new_fa->fa_state = state & ~FA_S_ACCESSED;
1296 new_fa->fa_slen = fa->fa_slen;
1297 new_fa->tb_id = tb->tb_id;
1298 new_fa->fa_default = -1;
1299 new_fa->offload = 0;
1300 new_fa->trap = 0;
1301 new_fa->offload_failed = 0;
1302
1303 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1304
1305 if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1306 tb->tb_id, true) == new_fa) {
1307 enum fib_event_type fib_event;
1308
1309 fib_event = FIB_EVENT_ENTRY_REPLACE;
1310 err = call_fib_entry_notifiers(net, fib_event,
1311 key, plen,
1312 new_fa, extack);
1313 if (err) {
1314 hlist_replace_rcu(&new_fa->fa_list,
1315 &fa->fa_list);
1316 goto out_free_new_fa;
1317 }
1318 }
1319
1320 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1321 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1322
1323 alias_free_mem_rcu(fa);
1324
1325 fib_release_info(fi_drop);
1326 if (state & FA_S_ACCESSED)
1327 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1328
1329 goto succeeded;
1330 }
1331 /* Error if we find a perfect match which
1332 * uses the same scope, type, and nexthop
1333 * information.
1334 */
1335 if (fa_match)
1336 goto out;
1337
1338 if (cfg->fc_nlflags & NLM_F_APPEND)
1339 nlflags |= NLM_F_APPEND;
1340 else
1341 fa = fa_first;
1342 }
1343 err = -ENOENT;
1344 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1345 goto out;
1346
1347 nlflags |= NLM_F_CREATE;
1348 err = -ENOBUFS;
1349 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1350 if (!new_fa)
1351 goto out;
1352
1353 new_fa->fa_info = fi;
1354 new_fa->fa_tos = tos;
1355 new_fa->fa_type = cfg->fc_type;
1356 new_fa->fa_state = 0;
1357 new_fa->fa_slen = slen;
1358 new_fa->tb_id = tb->tb_id;
1359 new_fa->fa_default = -1;
1360 new_fa->offload = 0;
1361 new_fa->trap = 0;
1362 new_fa->offload_failed = 0;
1363
1364 /* Insert new entry to the list. */
1365 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1366 if (err)
1367 goto out_free_new_fa;
1368
1369 /* The alias was already inserted, so the node must exist. */
1370 l = l ? l : fib_find_node(t, &tp, key);
1371 if (WARN_ON_ONCE(!l))
1372 goto out_free_new_fa;
1373
1374 if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1375 new_fa) {
1376 enum fib_event_type fib_event;
1377
1378 fib_event = FIB_EVENT_ENTRY_REPLACE;
1379 err = call_fib_entry_notifiers(net, fib_event, key, plen,
1380 new_fa, extack);
1381 if (err)
1382 goto out_remove_new_fa;
1383 }
1384
1385 if (!plen)
1386 tb->tb_num_default++;
1387
1388 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1389 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1390 &cfg->fc_nlinfo, nlflags);
1391succeeded:
1392 return 0;
1393
1394out_remove_new_fa:
1395 fib_remove_alias(t, tp, l, new_fa);
1396out_free_new_fa:
1397 kmem_cache_free(fn_alias_kmem, new_fa);
1398out:
1399 fib_release_info(fi);
1400err:
1401 return err;
1402}
1403
1404static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1405{
1406 t_key prefix = n->key;
1407
1408 return (key ^ prefix) & (prefix | -prefix);
1409}
1410
1411bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1412 const struct flowi4 *flp)
1413{
1414 if (nhc->nhc_flags & RTNH_F_DEAD)
1415 return false;
1416
1417 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1418 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1419 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1420 return false;
1421
1422 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1423 if (flp->flowi4_oif &&
1424 flp->flowi4_oif != nhc->nhc_oif)
1425 return false;
1426 }
1427
1428 return true;
1429}
1430
1431/* should be called with rcu_read_lock */
1432int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1433 struct fib_result *res, int fib_flags)
1434{
1435 struct trie *t = (struct trie *) tb->tb_data;
1436#ifdef CONFIG_IP_FIB_TRIE_STATS
1437 struct trie_use_stats __percpu *stats = t->stats;
1438#endif
1439 const t_key key = ntohl(flp->daddr);
1440 struct key_vector *n, *pn;
1441 struct fib_alias *fa;
1442 unsigned long index;
1443 t_key cindex;
1444
1445 pn = t->kv;
1446 cindex = 0;
1447
1448 n = get_child_rcu(pn, cindex);
1449 if (!n) {
1450 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1451 return -EAGAIN;
1452 }
1453
1454#ifdef CONFIG_IP_FIB_TRIE_STATS
1455 this_cpu_inc(stats->gets);
1456#endif
1457
1458 /* Step 1: Travel to the longest prefix match in the trie */
1459 for (;;) {
1460 index = get_cindex(key, n);
1461
1462 /* This bit of code is a bit tricky but it combines multiple
1463 * checks into a single check. The prefix consists of the
1464 * prefix plus zeros for the "bits" in the prefix. The index
1465 * is the difference between the key and this value. From
1466 * this we can actually derive several pieces of data.
1467 * if (index >= (1ul << bits))
1468 * we have a mismatch in skip bits and failed
1469 * else
1470 * we know the value is cindex
1471 *
1472 * This check is safe even if bits == KEYLENGTH due to the
1473 * fact that we can only allocate a node with 32 bits if a
1474 * long is greater than 32 bits.
1475 */
1476 if (index >= (1ul << n->bits))
1477 break;
1478
1479 /* we have found a leaf. Prefixes have already been compared */
1480 if (IS_LEAF(n))
1481 goto found;
1482
1483 /* only record pn and cindex if we are going to be chopping
1484 * bits later. Otherwise we are just wasting cycles.
1485 */
1486 if (n->slen > n->pos) {
1487 pn = n;
1488 cindex = index;
1489 }
1490
1491 n = get_child_rcu(n, index);
1492 if (unlikely(!n))
1493 goto backtrace;
1494 }
1495
1496 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1497 for (;;) {
1498 /* record the pointer where our next node pointer is stored */
1499 struct key_vector __rcu **cptr = n->tnode;
1500
1501 /* This test verifies that none of the bits that differ
1502 * between the key and the prefix exist in the region of
1503 * the lsb and higher in the prefix.
1504 */
1505 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1506 goto backtrace;
1507
1508 /* exit out and process leaf */
1509 if (unlikely(IS_LEAF(n)))
1510 break;
1511
1512 /* Don't bother recording parent info. Since we are in
1513 * prefix match mode we will have to come back to wherever
1514 * we started this traversal anyway
1515 */
1516
1517 while ((n = rcu_dereference(*cptr)) == NULL) {
1518backtrace:
1519#ifdef CONFIG_IP_FIB_TRIE_STATS
1520 if (!n)
1521 this_cpu_inc(stats->null_node_hit);
1522#endif
1523 /* If we are at cindex 0 there are no more bits for
1524 * us to strip at this level so we must ascend back
1525 * up one level to see if there are any more bits to
1526 * be stripped there.
1527 */
1528 while (!cindex) {
1529 t_key pkey = pn->key;
1530
1531 /* If we don't have a parent then there is
1532 * nothing for us to do as we do not have any
1533 * further nodes to parse.
1534 */
1535 if (IS_TRIE(pn)) {
1536 trace_fib_table_lookup(tb->tb_id, flp,
1537 NULL, -EAGAIN);
1538 return -EAGAIN;
1539 }
1540#ifdef CONFIG_IP_FIB_TRIE_STATS
1541 this_cpu_inc(stats->backtrack);
1542#endif
1543 /* Get Child's index */
1544 pn = node_parent_rcu(pn);
1545 cindex = get_index(pkey, pn);
1546 }
1547
1548 /* strip the least significant bit from the cindex */
1549 cindex &= cindex - 1;
1550
1551 /* grab pointer for next child node */
1552 cptr = &pn->tnode[cindex];
1553 }
1554 }
1555
1556found:
1557 /* this line carries forward the xor from earlier in the function */
1558 index = key ^ n->key;
1559
1560 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1561 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1562 struct fib_info *fi = fa->fa_info;
1563 struct fib_nh_common *nhc;
1564 int nhsel, err;
1565
1566 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1567 if (index >= (1ul << fa->fa_slen))
1568 continue;
1569 }
1570 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1571 continue;
1572 if (fi->fib_dead)
1573 continue;
1574 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1575 continue;
1576 fib_alias_accessed(fa);
1577 err = fib_props[fa->fa_type].error;
1578 if (unlikely(err < 0)) {
1579out_reject:
1580#ifdef CONFIG_IP_FIB_TRIE_STATS
1581 this_cpu_inc(stats->semantic_match_passed);
1582#endif
1583 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1584 return err;
1585 }
1586 if (fi->fib_flags & RTNH_F_DEAD)
1587 continue;
1588
1589 if (unlikely(fi->nh)) {
1590 if (nexthop_is_blackhole(fi->nh)) {
1591 err = fib_props[RTN_BLACKHOLE].error;
1592 goto out_reject;
1593 }
1594
1595 nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1596 &nhsel);
1597 if (nhc)
1598 goto set_result;
1599 goto miss;
1600 }
1601
1602 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1603 nhc = fib_info_nhc(fi, nhsel);
1604
1605 if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1606 continue;
1607set_result:
1608 if (!(fib_flags & FIB_LOOKUP_NOREF))
1609 refcount_inc(&fi->fib_clntref);
1610
1611 res->prefix = htonl(n->key);
1612 res->prefixlen = KEYLENGTH - fa->fa_slen;
1613 res->nh_sel = nhsel;
1614 res->nhc = nhc;
1615 res->type = fa->fa_type;
1616 res->scope = fi->fib_scope;
1617 res->fi = fi;
1618 res->table = tb;
1619 res->fa_head = &n->leaf;
1620#ifdef CONFIG_IP_FIB_TRIE_STATS
1621 this_cpu_inc(stats->semantic_match_passed);
1622#endif
1623 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1624
1625 return err;
1626 }
1627 }
1628miss:
1629#ifdef CONFIG_IP_FIB_TRIE_STATS
1630 this_cpu_inc(stats->semantic_match_miss);
1631#endif
1632 goto backtrace;
1633}
1634EXPORT_SYMBOL_GPL(fib_table_lookup);
1635
1636static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1637 struct key_vector *l, struct fib_alias *old)
1638{
1639 /* record the location of the previous list_info entry */
1640 struct hlist_node **pprev = old->fa_list.pprev;
1641 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1642
1643 /* remove the fib_alias from the list */
1644 hlist_del_rcu(&old->fa_list);
1645
1646 /* if we emptied the list this leaf will be freed and we can sort
1647 * out parent suffix lengths as a part of trie_rebalance
1648 */
1649 if (hlist_empty(&l->leaf)) {
1650 if (tp->slen == l->slen)
1651 node_pull_suffix(tp, tp->pos);
1652 put_child_root(tp, l->key, NULL);
1653 node_free(l);
1654 trie_rebalance(t, tp);
1655 return;
1656 }
1657
1658 /* only access fa if it is pointing at the last valid hlist_node */
1659 if (*pprev)
1660 return;
1661
1662 /* update the trie with the latest suffix length */
1663 l->slen = fa->fa_slen;
1664 node_pull_suffix(tp, fa->fa_slen);
1665}
1666
1667static void fib_notify_alias_delete(struct net *net, u32 key,
1668 struct hlist_head *fah,
1669 struct fib_alias *fa_to_delete,
1670 struct netlink_ext_ack *extack)
1671{
1672 struct fib_alias *fa_next, *fa_to_notify;
1673 u32 tb_id = fa_to_delete->tb_id;
1674 u8 slen = fa_to_delete->fa_slen;
1675 enum fib_event_type fib_event;
1676
1677 /* Do not notify if we do not care about the route. */
1678 if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1679 return;
1680
1681 /* Determine if the route should be replaced by the next route in the
1682 * list.
1683 */
1684 fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1685 struct fib_alias, fa_list);
1686 if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1687 fib_event = FIB_EVENT_ENTRY_REPLACE;
1688 fa_to_notify = fa_next;
1689 } else {
1690 fib_event = FIB_EVENT_ENTRY_DEL;
1691 fa_to_notify = fa_to_delete;
1692 }
1693 call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1694 fa_to_notify, extack);
1695}
1696
1697/* Caller must hold RTNL. */
1698int fib_table_delete(struct net *net, struct fib_table *tb,
1699 struct fib_config *cfg, struct netlink_ext_ack *extack)
1700{
1701 struct trie *t = (struct trie *) tb->tb_data;
1702 struct fib_alias *fa, *fa_to_delete;
1703 struct key_vector *l, *tp;
1704 u8 plen = cfg->fc_dst_len;
1705 u8 slen = KEYLENGTH - plen;
1706 u8 tos = cfg->fc_tos;
1707 u32 key;
1708
1709 key = ntohl(cfg->fc_dst);
1710
1711 if (!fib_valid_key_len(key, plen, extack))
1712 return -EINVAL;
1713
1714 l = fib_find_node(t, &tp, key);
1715 if (!l)
1716 return -ESRCH;
1717
1718 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id, false);
1719 if (!fa)
1720 return -ESRCH;
1721
1722 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1723
1724 fa_to_delete = NULL;
1725 hlist_for_each_entry_from(fa, fa_list) {
1726 struct fib_info *fi = fa->fa_info;
1727
1728 if ((fa->fa_slen != slen) ||
1729 (fa->tb_id != tb->tb_id) ||
1730 (fa->fa_tos != tos))
1731 break;
1732
1733 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1734 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1735 fa->fa_info->fib_scope == cfg->fc_scope) &&
1736 (!cfg->fc_prefsrc ||
1737 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1738 (!cfg->fc_protocol ||
1739 fi->fib_protocol == cfg->fc_protocol) &&
1740 fib_nh_match(net, cfg, fi, extack) == 0 &&
1741 fib_metrics_match(cfg, fi)) {
1742 fa_to_delete = fa;
1743 break;
1744 }
1745 }
1746
1747 if (!fa_to_delete)
1748 return -ESRCH;
1749
1750 fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1751 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1752 &cfg->fc_nlinfo, 0);
1753
1754 if (!plen)
1755 tb->tb_num_default--;
1756
1757 fib_remove_alias(t, tp, l, fa_to_delete);
1758
1759 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1760 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1761
1762 fib_release_info(fa_to_delete->fa_info);
1763 alias_free_mem_rcu(fa_to_delete);
1764 return 0;
1765}
1766
1767/* Scan for the next leaf starting at the provided key value */
1768static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1769{
1770 struct key_vector *pn, *n = *tn;
1771 unsigned long cindex;
1772
1773 /* this loop is meant to try and find the key in the trie */
1774 do {
1775 /* record parent and next child index */
1776 pn = n;
1777 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1778
1779 if (cindex >> pn->bits)
1780 break;
1781
1782 /* descend into the next child */
1783 n = get_child_rcu(pn, cindex++);
1784 if (!n)
1785 break;
1786
1787 /* guarantee forward progress on the keys */
1788 if (IS_LEAF(n) && (n->key >= key))
1789 goto found;
1790 } while (IS_TNODE(n));
1791
1792 /* this loop will search for the next leaf with a greater key */
1793 while (!IS_TRIE(pn)) {
1794 /* if we exhausted the parent node we will need to climb */
1795 if (cindex >= (1ul << pn->bits)) {
1796 t_key pkey = pn->key;
1797
1798 pn = node_parent_rcu(pn);
1799 cindex = get_index(pkey, pn) + 1;
1800 continue;
1801 }
1802
1803 /* grab the next available node */
1804 n = get_child_rcu(pn, cindex++);
1805 if (!n)
1806 continue;
1807
1808 /* no need to compare keys since we bumped the index */
1809 if (IS_LEAF(n))
1810 goto found;
1811
1812 /* Rescan start scanning in new node */
1813 pn = n;
1814 cindex = 0;
1815 }
1816
1817 *tn = pn;
1818 return NULL; /* Root of trie */
1819found:
1820 /* if we are at the limit for keys just return NULL for the tnode */
1821 *tn = pn;
1822 return n;
1823}
1824
1825static void fib_trie_free(struct fib_table *tb)
1826{
1827 struct trie *t = (struct trie *)tb->tb_data;
1828 struct key_vector *pn = t->kv;
1829 unsigned long cindex = 1;
1830 struct hlist_node *tmp;
1831 struct fib_alias *fa;
1832
1833 /* walk trie in reverse order and free everything */
1834 for (;;) {
1835 struct key_vector *n;
1836
1837 if (!(cindex--)) {
1838 t_key pkey = pn->key;
1839
1840 if (IS_TRIE(pn))
1841 break;
1842
1843 n = pn;
1844 pn = node_parent(pn);
1845
1846 /* drop emptied tnode */
1847 put_child_root(pn, n->key, NULL);
1848 node_free(n);
1849
1850 cindex = get_index(pkey, pn);
1851
1852 continue;
1853 }
1854
1855 /* grab the next available node */
1856 n = get_child(pn, cindex);
1857 if (!n)
1858 continue;
1859
1860 if (IS_TNODE(n)) {
1861 /* record pn and cindex for leaf walking */
1862 pn = n;
1863 cindex = 1ul << n->bits;
1864
1865 continue;
1866 }
1867
1868 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1869 hlist_del_rcu(&fa->fa_list);
1870 alias_free_mem_rcu(fa);
1871 }
1872
1873 put_child_root(pn, n->key, NULL);
1874 node_free(n);
1875 }
1876
1877#ifdef CONFIG_IP_FIB_TRIE_STATS
1878 free_percpu(t->stats);
1879#endif
1880 kfree(tb);
1881}
1882
1883struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1884{
1885 struct trie *ot = (struct trie *)oldtb->tb_data;
1886 struct key_vector *l, *tp = ot->kv;
1887 struct fib_table *local_tb;
1888 struct fib_alias *fa;
1889 struct trie *lt;
1890 t_key key = 0;
1891
1892 if (oldtb->tb_data == oldtb->__data)
1893 return oldtb;
1894
1895 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1896 if (!local_tb)
1897 return NULL;
1898
1899 lt = (struct trie *)local_tb->tb_data;
1900
1901 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1902 struct key_vector *local_l = NULL, *local_tp;
1903
1904 hlist_for_each_entry(fa, &l->leaf, fa_list) {
1905 struct fib_alias *new_fa;
1906
1907 if (local_tb->tb_id != fa->tb_id)
1908 continue;
1909
1910 /* clone fa for new local table */
1911 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1912 if (!new_fa)
1913 goto out;
1914
1915 memcpy(new_fa, fa, sizeof(*fa));
1916
1917 /* insert clone into table */
1918 if (!local_l)
1919 local_l = fib_find_node(lt, &local_tp, l->key);
1920
1921 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1922 NULL, l->key)) {
1923 kmem_cache_free(fn_alias_kmem, new_fa);
1924 goto out;
1925 }
1926 }
1927
1928 /* stop loop if key wrapped back to 0 */
1929 key = l->key + 1;
1930 if (key < l->key)
1931 break;
1932 }
1933
1934 return local_tb;
1935out:
1936 fib_trie_free(local_tb);
1937
1938 return NULL;
1939}
1940
1941/* Caller must hold RTNL */
1942void fib_table_flush_external(struct fib_table *tb)
1943{
1944 struct trie *t = (struct trie *)tb->tb_data;
1945 struct key_vector *pn = t->kv;
1946 unsigned long cindex = 1;
1947 struct hlist_node *tmp;
1948 struct fib_alias *fa;
1949
1950 /* walk trie in reverse order */
1951 for (;;) {
1952 unsigned char slen = 0;
1953 struct key_vector *n;
1954
1955 if (!(cindex--)) {
1956 t_key pkey = pn->key;
1957
1958 /* cannot resize the trie vector */
1959 if (IS_TRIE(pn))
1960 break;
1961
1962 /* update the suffix to address pulled leaves */
1963 if (pn->slen > pn->pos)
1964 update_suffix(pn);
1965
1966 /* resize completed node */
1967 pn = resize(t, pn);
1968 cindex = get_index(pkey, pn);
1969
1970 continue;
1971 }
1972
1973 /* grab the next available node */
1974 n = get_child(pn, cindex);
1975 if (!n)
1976 continue;
1977
1978 if (IS_TNODE(n)) {
1979 /* record pn and cindex for leaf walking */
1980 pn = n;
1981 cindex = 1ul << n->bits;
1982
1983 continue;
1984 }
1985
1986 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1987 /* if alias was cloned to local then we just
1988 * need to remove the local copy from main
1989 */
1990 if (tb->tb_id != fa->tb_id) {
1991 hlist_del_rcu(&fa->fa_list);
1992 alias_free_mem_rcu(fa);
1993 continue;
1994 }
1995
1996 /* record local slen */
1997 slen = fa->fa_slen;
1998 }
1999
2000 /* update leaf slen */
2001 n->slen = slen;
2002
2003 if (hlist_empty(&n->leaf)) {
2004 put_child_root(pn, n->key, NULL);
2005 node_free(n);
2006 }
2007 }
2008}
2009
2010/* Caller must hold RTNL. */
2011int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
2012{
2013 struct trie *t = (struct trie *)tb->tb_data;
2014 struct key_vector *pn = t->kv;
2015 unsigned long cindex = 1;
2016 struct hlist_node *tmp;
2017 struct fib_alias *fa;
2018 int found = 0;
2019
2020 /* walk trie in reverse order */
2021 for (;;) {
2022 unsigned char slen = 0;
2023 struct key_vector *n;
2024
2025 if (!(cindex--)) {
2026 t_key pkey = pn->key;
2027
2028 /* cannot resize the trie vector */
2029 if (IS_TRIE(pn))
2030 break;
2031
2032 /* update the suffix to address pulled leaves */
2033 if (pn->slen > pn->pos)
2034 update_suffix(pn);
2035
2036 /* resize completed node */
2037 pn = resize(t, pn);
2038 cindex = get_index(pkey, pn);
2039
2040 continue;
2041 }
2042
2043 /* grab the next available node */
2044 n = get_child(pn, cindex);
2045 if (!n)
2046 continue;
2047
2048 if (IS_TNODE(n)) {
2049 /* record pn and cindex for leaf walking */
2050 pn = n;
2051 cindex = 1ul << n->bits;
2052
2053 continue;
2054 }
2055
2056 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2057 struct fib_info *fi = fa->fa_info;
2058
2059 if (!fi || tb->tb_id != fa->tb_id ||
2060 (!(fi->fib_flags & RTNH_F_DEAD) &&
2061 !fib_props[fa->fa_type].error)) {
2062 slen = fa->fa_slen;
2063 continue;
2064 }
2065
2066 /* Do not flush error routes if network namespace is
2067 * not being dismantled
2068 */
2069 if (!flush_all && fib_props[fa->fa_type].error) {
2070 slen = fa->fa_slen;
2071 continue;
2072 }
2073
2074 fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2075 NULL);
2076 hlist_del_rcu(&fa->fa_list);
2077 fib_release_info(fa->fa_info);
2078 alias_free_mem_rcu(fa);
2079 found++;
2080 }
2081
2082 /* update leaf slen */
2083 n->slen = slen;
2084
2085 if (hlist_empty(&n->leaf)) {
2086 put_child_root(pn, n->key, NULL);
2087 node_free(n);
2088 }
2089 }
2090
2091 pr_debug("trie_flush found=%d\n", found);
2092 return found;
2093}
2094
2095/* derived from fib_trie_free */
2096static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2097 struct nl_info *info)
2098{
2099 struct trie *t = (struct trie *)tb->tb_data;
2100 struct key_vector *pn = t->kv;
2101 unsigned long cindex = 1;
2102 struct fib_alias *fa;
2103
2104 for (;;) {
2105 struct key_vector *n;
2106
2107 if (!(cindex--)) {
2108 t_key pkey = pn->key;
2109
2110 if (IS_TRIE(pn))
2111 break;
2112
2113 pn = node_parent(pn);
2114 cindex = get_index(pkey, pn);
2115 continue;
2116 }
2117
2118 /* grab the next available node */
2119 n = get_child(pn, cindex);
2120 if (!n)
2121 continue;
2122
2123 if (IS_TNODE(n)) {
2124 /* record pn and cindex for leaf walking */
2125 pn = n;
2126 cindex = 1ul << n->bits;
2127
2128 continue;
2129 }
2130
2131 hlist_for_each_entry(fa, &n->leaf, fa_list) {
2132 struct fib_info *fi = fa->fa_info;
2133
2134 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2135 continue;
2136
2137 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2138 KEYLENGTH - fa->fa_slen, tb->tb_id,
2139 info, NLM_F_REPLACE);
2140 }
2141 }
2142}
2143
2144void fib_info_notify_update(struct net *net, struct nl_info *info)
2145{
2146 unsigned int h;
2147
2148 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2149 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2150 struct fib_table *tb;
2151
2152 hlist_for_each_entry_rcu(tb, head, tb_hlist,
2153 lockdep_rtnl_is_held())
2154 __fib_info_notify_update(net, tb, info);
2155 }
2156}
2157
2158static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2159 struct notifier_block *nb,
2160 struct netlink_ext_ack *extack)
2161{
2162 struct fib_alias *fa;
2163 int last_slen = -1;
2164 int err;
2165
2166 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2167 struct fib_info *fi = fa->fa_info;
2168
2169 if (!fi)
2170 continue;
2171
2172 /* local and main table can share the same trie,
2173 * so don't notify twice for the same entry.
2174 */
2175 if (tb->tb_id != fa->tb_id)
2176 continue;
2177
2178 if (fa->fa_slen == last_slen)
2179 continue;
2180
2181 last_slen = fa->fa_slen;
2182 err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2183 l->key, KEYLENGTH - fa->fa_slen,
2184 fa, extack);
2185 if (err)
2186 return err;
2187 }
2188 return 0;
2189}
2190
2191static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2192 struct netlink_ext_ack *extack)
2193{
2194 struct trie *t = (struct trie *)tb->tb_data;
2195 struct key_vector *l, *tp = t->kv;
2196 t_key key = 0;
2197 int err;
2198
2199 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2200 err = fib_leaf_notify(l, tb, nb, extack);
2201 if (err)
2202 return err;
2203
2204 key = l->key + 1;
2205 /* stop in case of wrap around */
2206 if (key < l->key)
2207 break;
2208 }
2209 return 0;
2210}
2211
2212int fib_notify(struct net *net, struct notifier_block *nb,
2213 struct netlink_ext_ack *extack)
2214{
2215 unsigned int h;
2216 int err;
2217
2218 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2219 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2220 struct fib_table *tb;
2221
2222 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2223 err = fib_table_notify(tb, nb, extack);
2224 if (err)
2225 return err;
2226 }
2227 }
2228 return 0;
2229}
2230
2231static void __trie_free_rcu(struct rcu_head *head)
2232{
2233 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2234#ifdef CONFIG_IP_FIB_TRIE_STATS
2235 struct trie *t = (struct trie *)tb->tb_data;
2236
2237 if (tb->tb_data == tb->__data)
2238 free_percpu(t->stats);
2239#endif /* CONFIG_IP_FIB_TRIE_STATS */
2240 kfree(tb);
2241}
2242
2243void fib_free_table(struct fib_table *tb)
2244{
2245 call_rcu(&tb->rcu, __trie_free_rcu);
2246}
2247
2248static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2249 struct sk_buff *skb, struct netlink_callback *cb,
2250 struct fib_dump_filter *filter)
2251{
2252 unsigned int flags = NLM_F_MULTI;
2253 __be32 xkey = htonl(l->key);
2254 int i, s_i, i_fa, s_fa, err;
2255 struct fib_alias *fa;
2256
2257 if (filter->filter_set ||
2258 !filter->dump_exceptions || !filter->dump_routes)
2259 flags |= NLM_F_DUMP_FILTERED;
2260
2261 s_i = cb->args[4];
2262 s_fa = cb->args[5];
2263 i = 0;
2264
2265 /* rcu_read_lock is hold by caller */
2266 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2267 struct fib_info *fi = fa->fa_info;
2268
2269 if (i < s_i)
2270 goto next;
2271
2272 i_fa = 0;
2273
2274 if (tb->tb_id != fa->tb_id)
2275 goto next;
2276
2277 if (filter->filter_set) {
2278 if (filter->rt_type && fa->fa_type != filter->rt_type)
2279 goto next;
2280
2281 if ((filter->protocol &&
2282 fi->fib_protocol != filter->protocol))
2283 goto next;
2284
2285 if (filter->dev &&
2286 !fib_info_nh_uses_dev(fi, filter->dev))
2287 goto next;
2288 }
2289
2290 if (filter->dump_routes) {
2291 if (!s_fa) {
2292 struct fib_rt_info fri;
2293
2294 fri.fi = fi;
2295 fri.tb_id = tb->tb_id;
2296 fri.dst = xkey;
2297 fri.dst_len = KEYLENGTH - fa->fa_slen;
2298 fri.tos = fa->fa_tos;
2299 fri.type = fa->fa_type;
2300 fri.offload = fa->offload;
2301 fri.trap = fa->trap;
2302 fri.offload_failed = fa->offload_failed;
2303 err = fib_dump_info(skb,
2304 NETLINK_CB(cb->skb).portid,
2305 cb->nlh->nlmsg_seq,
2306 RTM_NEWROUTE, &fri, flags);
2307 if (err < 0)
2308 goto stop;
2309 }
2310
2311 i_fa++;
2312 }
2313
2314 if (filter->dump_exceptions) {
2315 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2316 &i_fa, s_fa, flags);
2317 if (err < 0)
2318 goto stop;
2319 }
2320
2321next:
2322 i++;
2323 }
2324
2325 cb->args[4] = i;
2326 return skb->len;
2327
2328stop:
2329 cb->args[4] = i;
2330 cb->args[5] = i_fa;
2331 return err;
2332}
2333
2334/* rcu_read_lock needs to be hold by caller from readside */
2335int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2336 struct netlink_callback *cb, struct fib_dump_filter *filter)
2337{
2338 struct trie *t = (struct trie *)tb->tb_data;
2339 struct key_vector *l, *tp = t->kv;
2340 /* Dump starting at last key.
2341 * Note: 0.0.0.0/0 (ie default) is first key.
2342 */
2343 int count = cb->args[2];
2344 t_key key = cb->args[3];
2345
2346 /* First time here, count and key are both always 0. Count > 0
2347 * and key == 0 means the dump has wrapped around and we are done.
2348 */
2349 if (count && !key)
2350 return skb->len;
2351
2352 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2353 int err;
2354
2355 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2356 if (err < 0) {
2357 cb->args[3] = key;
2358 cb->args[2] = count;
2359 return err;
2360 }
2361
2362 ++count;
2363 key = l->key + 1;
2364
2365 memset(&cb->args[4], 0,
2366 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2367
2368 /* stop loop if key wrapped back to 0 */
2369 if (key < l->key)
2370 break;
2371 }
2372
2373 cb->args[3] = key;
2374 cb->args[2] = count;
2375
2376 return skb->len;
2377}
2378
2379void __init fib_trie_init(void)
2380{
2381 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2382 sizeof(struct fib_alias),
2383 0, SLAB_PANIC, NULL);
2384
2385 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2386 LEAF_SIZE,
2387 0, SLAB_PANIC, NULL);
2388}
2389
2390struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2391{
2392 struct fib_table *tb;
2393 struct trie *t;
2394 size_t sz = sizeof(*tb);
2395
2396 if (!alias)
2397 sz += sizeof(struct trie);
2398
2399 tb = kzalloc(sz, GFP_KERNEL);
2400 if (!tb)
2401 return NULL;
2402
2403 tb->tb_id = id;
2404 tb->tb_num_default = 0;
2405 tb->tb_data = (alias ? alias->__data : tb->__data);
2406
2407 if (alias)
2408 return tb;
2409
2410 t = (struct trie *) tb->tb_data;
2411 t->kv[0].pos = KEYLENGTH;
2412 t->kv[0].slen = KEYLENGTH;
2413#ifdef CONFIG_IP_FIB_TRIE_STATS
2414 t->stats = alloc_percpu(struct trie_use_stats);
2415 if (!t->stats) {
2416 kfree(tb);
2417 tb = NULL;
2418 }
2419#endif
2420
2421 return tb;
2422}
2423
2424#ifdef CONFIG_PROC_FS
2425/* Depth first Trie walk iterator */
2426struct fib_trie_iter {
2427 struct seq_net_private p;
2428 struct fib_table *tb;
2429 struct key_vector *tnode;
2430 unsigned int index;
2431 unsigned int depth;
2432};
2433
2434static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2435{
2436 unsigned long cindex = iter->index;
2437 struct key_vector *pn = iter->tnode;
2438 t_key pkey;
2439
2440 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2441 iter->tnode, iter->index, iter->depth);
2442
2443 while (!IS_TRIE(pn)) {
2444 while (cindex < child_length(pn)) {
2445 struct key_vector *n = get_child_rcu(pn, cindex++);
2446
2447 if (!n)
2448 continue;
2449
2450 if (IS_LEAF(n)) {
2451 iter->tnode = pn;
2452 iter->index = cindex;
2453 } else {
2454 /* push down one level */
2455 iter->tnode = n;
2456 iter->index = 0;
2457 ++iter->depth;
2458 }
2459
2460 return n;
2461 }
2462
2463 /* Current node exhausted, pop back up */
2464 pkey = pn->key;
2465 pn = node_parent_rcu(pn);
2466 cindex = get_index(pkey, pn) + 1;
2467 --iter->depth;
2468 }
2469
2470 /* record root node so further searches know we are done */
2471 iter->tnode = pn;
2472 iter->index = 0;
2473
2474 return NULL;
2475}
2476
2477static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2478 struct trie *t)
2479{
2480 struct key_vector *n, *pn;
2481
2482 if (!t)
2483 return NULL;
2484
2485 pn = t->kv;
2486 n = rcu_dereference(pn->tnode[0]);
2487 if (!n)
2488 return NULL;
2489
2490 if (IS_TNODE(n)) {
2491 iter->tnode = n;
2492 iter->index = 0;
2493 iter->depth = 1;
2494 } else {
2495 iter->tnode = pn;
2496 iter->index = 0;
2497 iter->depth = 0;
2498 }
2499
2500 return n;
2501}
2502
2503static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2504{
2505 struct key_vector *n;
2506 struct fib_trie_iter iter;
2507
2508 memset(s, 0, sizeof(*s));
2509
2510 rcu_read_lock();
2511 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2512 if (IS_LEAF(n)) {
2513 struct fib_alias *fa;
2514
2515 s->leaves++;
2516 s->totdepth += iter.depth;
2517 if (iter.depth > s->maxdepth)
2518 s->maxdepth = iter.depth;
2519
2520 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2521 ++s->prefixes;
2522 } else {
2523 s->tnodes++;
2524 if (n->bits < MAX_STAT_DEPTH)
2525 s->nodesizes[n->bits]++;
2526 s->nullpointers += tn_info(n)->empty_children;
2527 }
2528 }
2529 rcu_read_unlock();
2530}
2531
2532/*
2533 * This outputs /proc/net/fib_triestats
2534 */
2535static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2536{
2537 unsigned int i, max, pointers, bytes, avdepth;
2538
2539 if (stat->leaves)
2540 avdepth = stat->totdepth*100 / stat->leaves;
2541 else
2542 avdepth = 0;
2543
2544 seq_printf(seq, "\tAver depth: %u.%02d\n",
2545 avdepth / 100, avdepth % 100);
2546 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2547
2548 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2549 bytes = LEAF_SIZE * stat->leaves;
2550
2551 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2552 bytes += sizeof(struct fib_alias) * stat->prefixes;
2553
2554 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2555 bytes += TNODE_SIZE(0) * stat->tnodes;
2556
2557 max = MAX_STAT_DEPTH;
2558 while (max > 0 && stat->nodesizes[max-1] == 0)
2559 max--;
2560
2561 pointers = 0;
2562 for (i = 1; i < max; i++)
2563 if (stat->nodesizes[i] != 0) {
2564 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2565 pointers += (1<<i) * stat->nodesizes[i];
2566 }
2567 seq_putc(seq, '\n');
2568 seq_printf(seq, "\tPointers: %u\n", pointers);
2569
2570 bytes += sizeof(struct key_vector *) * pointers;
2571 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2572 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2573}
2574
2575#ifdef CONFIG_IP_FIB_TRIE_STATS
2576static void trie_show_usage(struct seq_file *seq,
2577 const struct trie_use_stats __percpu *stats)
2578{
2579 struct trie_use_stats s = { 0 };
2580 int cpu;
2581
2582 /* loop through all of the CPUs and gather up the stats */
2583 for_each_possible_cpu(cpu) {
2584 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2585
2586 s.gets += pcpu->gets;
2587 s.backtrack += pcpu->backtrack;
2588 s.semantic_match_passed += pcpu->semantic_match_passed;
2589 s.semantic_match_miss += pcpu->semantic_match_miss;
2590 s.null_node_hit += pcpu->null_node_hit;
2591 s.resize_node_skipped += pcpu->resize_node_skipped;
2592 }
2593
2594 seq_printf(seq, "\nCounters:\n---------\n");
2595 seq_printf(seq, "gets = %u\n", s.gets);
2596 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2597 seq_printf(seq, "semantic match passed = %u\n",
2598 s.semantic_match_passed);
2599 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2600 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2601 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2602}
2603#endif /* CONFIG_IP_FIB_TRIE_STATS */
2604
2605static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2606{
2607 if (tb->tb_id == RT_TABLE_LOCAL)
2608 seq_puts(seq, "Local:\n");
2609 else if (tb->tb_id == RT_TABLE_MAIN)
2610 seq_puts(seq, "Main:\n");
2611 else
2612 seq_printf(seq, "Id %d:\n", tb->tb_id);
2613}
2614
2615
2616static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2617{
2618 struct net *net = (struct net *)seq->private;
2619 unsigned int h;
2620
2621 seq_printf(seq,
2622 "Basic info: size of leaf:"
2623 " %zd bytes, size of tnode: %zd bytes.\n",
2624 LEAF_SIZE, TNODE_SIZE(0));
2625
2626 rcu_read_lock();
2627 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2628 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2629 struct fib_table *tb;
2630
2631 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2632 struct trie *t = (struct trie *) tb->tb_data;
2633 struct trie_stat stat;
2634
2635 if (!t)
2636 continue;
2637
2638 fib_table_print(seq, tb);
2639
2640 trie_collect_stats(t, &stat);
2641 trie_show_stats(seq, &stat);
2642#ifdef CONFIG_IP_FIB_TRIE_STATS
2643 trie_show_usage(seq, t->stats);
2644#endif
2645 }
2646 cond_resched_rcu();
2647 }
2648 rcu_read_unlock();
2649
2650 return 0;
2651}
2652
2653static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2654{
2655 struct fib_trie_iter *iter = seq->private;
2656 struct net *net = seq_file_net(seq);
2657 loff_t idx = 0;
2658 unsigned int h;
2659
2660 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2661 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2662 struct fib_table *tb;
2663
2664 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2665 struct key_vector *n;
2666
2667 for (n = fib_trie_get_first(iter,
2668 (struct trie *) tb->tb_data);
2669 n; n = fib_trie_get_next(iter))
2670 if (pos == idx++) {
2671 iter->tb = tb;
2672 return n;
2673 }
2674 }
2675 }
2676
2677 return NULL;
2678}
2679
2680static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2681 __acquires(RCU)
2682{
2683 rcu_read_lock();
2684 return fib_trie_get_idx(seq, *pos);
2685}
2686
2687static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2688{
2689 struct fib_trie_iter *iter = seq->private;
2690 struct net *net = seq_file_net(seq);
2691 struct fib_table *tb = iter->tb;
2692 struct hlist_node *tb_node;
2693 unsigned int h;
2694 struct key_vector *n;
2695
2696 ++*pos;
2697 /* next node in same table */
2698 n = fib_trie_get_next(iter);
2699 if (n)
2700 return n;
2701
2702 /* walk rest of this hash chain */
2703 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2704 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2705 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2706 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2707 if (n)
2708 goto found;
2709 }
2710
2711 /* new hash chain */
2712 while (++h < FIB_TABLE_HASHSZ) {
2713 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2714 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2715 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2716 if (n)
2717 goto found;
2718 }
2719 }
2720 return NULL;
2721
2722found:
2723 iter->tb = tb;
2724 return n;
2725}
2726
2727static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2728 __releases(RCU)
2729{
2730 rcu_read_unlock();
2731}
2732
2733static void seq_indent(struct seq_file *seq, int n)
2734{
2735 while (n-- > 0)
2736 seq_puts(seq, " ");
2737}
2738
2739static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2740{
2741 switch (s) {
2742 case RT_SCOPE_UNIVERSE: return "universe";
2743 case RT_SCOPE_SITE: return "site";
2744 case RT_SCOPE_LINK: return "link";
2745 case RT_SCOPE_HOST: return "host";
2746 case RT_SCOPE_NOWHERE: return "nowhere";
2747 default:
2748 snprintf(buf, len, "scope=%d", s);
2749 return buf;
2750 }
2751}
2752
2753static const char *const rtn_type_names[__RTN_MAX] = {
2754 [RTN_UNSPEC] = "UNSPEC",
2755 [RTN_UNICAST] = "UNICAST",
2756 [RTN_LOCAL] = "LOCAL",
2757 [RTN_BROADCAST] = "BROADCAST",
2758 [RTN_ANYCAST] = "ANYCAST",
2759 [RTN_MULTICAST] = "MULTICAST",
2760 [RTN_BLACKHOLE] = "BLACKHOLE",
2761 [RTN_UNREACHABLE] = "UNREACHABLE",
2762 [RTN_PROHIBIT] = "PROHIBIT",
2763 [RTN_THROW] = "THROW",
2764 [RTN_NAT] = "NAT",
2765 [RTN_XRESOLVE] = "XRESOLVE",
2766};
2767
2768static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2769{
2770 if (t < __RTN_MAX && rtn_type_names[t])
2771 return rtn_type_names[t];
2772 snprintf(buf, len, "type %u", t);
2773 return buf;
2774}
2775
2776/* Pretty print the trie */
2777static int fib_trie_seq_show(struct seq_file *seq, void *v)
2778{
2779 const struct fib_trie_iter *iter = seq->private;
2780 struct key_vector *n = v;
2781
2782 if (IS_TRIE(node_parent_rcu(n)))
2783 fib_table_print(seq, iter->tb);
2784
2785 if (IS_TNODE(n)) {
2786 __be32 prf = htonl(n->key);
2787
2788 seq_indent(seq, iter->depth-1);
2789 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2790 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2791 tn_info(n)->full_children,
2792 tn_info(n)->empty_children);
2793 } else {
2794 __be32 val = htonl(n->key);
2795 struct fib_alias *fa;
2796
2797 seq_indent(seq, iter->depth);
2798 seq_printf(seq, " |-- %pI4\n", &val);
2799
2800 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2801 char buf1[32], buf2[32];
2802
2803 seq_indent(seq, iter->depth + 1);
2804 seq_printf(seq, " /%zu %s %s",
2805 KEYLENGTH - fa->fa_slen,
2806 rtn_scope(buf1, sizeof(buf1),
2807 fa->fa_info->fib_scope),
2808 rtn_type(buf2, sizeof(buf2),
2809 fa->fa_type));
2810 if (fa->fa_tos)
2811 seq_printf(seq, " tos=%d", fa->fa_tos);
2812 seq_putc(seq, '\n');
2813 }
2814 }
2815
2816 return 0;
2817}
2818
2819static const struct seq_operations fib_trie_seq_ops = {
2820 .start = fib_trie_seq_start,
2821 .next = fib_trie_seq_next,
2822 .stop = fib_trie_seq_stop,
2823 .show = fib_trie_seq_show,
2824};
2825
2826struct fib_route_iter {
2827 struct seq_net_private p;
2828 struct fib_table *main_tb;
2829 struct key_vector *tnode;
2830 loff_t pos;
2831 t_key key;
2832};
2833
2834static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2835 loff_t pos)
2836{
2837 struct key_vector *l, **tp = &iter->tnode;
2838 t_key key;
2839
2840 /* use cached location of previously found key */
2841 if (iter->pos > 0 && pos >= iter->pos) {
2842 key = iter->key;
2843 } else {
2844 iter->pos = 1;
2845 key = 0;
2846 }
2847
2848 pos -= iter->pos;
2849
2850 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2851 key = l->key + 1;
2852 iter->pos++;
2853 l = NULL;
2854
2855 /* handle unlikely case of a key wrap */
2856 if (!key)
2857 break;
2858 }
2859
2860 if (l)
2861 iter->key = l->key; /* remember it */
2862 else
2863 iter->pos = 0; /* forget it */
2864
2865 return l;
2866}
2867
2868static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2869 __acquires(RCU)
2870{
2871 struct fib_route_iter *iter = seq->private;
2872 struct fib_table *tb;
2873 struct trie *t;
2874
2875 rcu_read_lock();
2876
2877 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2878 if (!tb)
2879 return NULL;
2880
2881 iter->main_tb = tb;
2882 t = (struct trie *)tb->tb_data;
2883 iter->tnode = t->kv;
2884
2885 if (*pos != 0)
2886 return fib_route_get_idx(iter, *pos);
2887
2888 iter->pos = 0;
2889 iter->key = KEY_MAX;
2890
2891 return SEQ_START_TOKEN;
2892}
2893
2894static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2895{
2896 struct fib_route_iter *iter = seq->private;
2897 struct key_vector *l = NULL;
2898 t_key key = iter->key + 1;
2899
2900 ++*pos;
2901
2902 /* only allow key of 0 for start of sequence */
2903 if ((v == SEQ_START_TOKEN) || key)
2904 l = leaf_walk_rcu(&iter->tnode, key);
2905
2906 if (l) {
2907 iter->key = l->key;
2908 iter->pos++;
2909 } else {
2910 iter->pos = 0;
2911 }
2912
2913 return l;
2914}
2915
2916static void fib_route_seq_stop(struct seq_file *seq, void *v)
2917 __releases(RCU)
2918{
2919 rcu_read_unlock();
2920}
2921
2922static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2923{
2924 unsigned int flags = 0;
2925
2926 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2927 flags = RTF_REJECT;
2928 if (fi) {
2929 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2930
2931 if (nhc->nhc_gw.ipv4)
2932 flags |= RTF_GATEWAY;
2933 }
2934 if (mask == htonl(0xFFFFFFFF))
2935 flags |= RTF_HOST;
2936 flags |= RTF_UP;
2937 return flags;
2938}
2939
2940/*
2941 * This outputs /proc/net/route.
2942 * The format of the file is not supposed to be changed
2943 * and needs to be same as fib_hash output to avoid breaking
2944 * legacy utilities
2945 */
2946static int fib_route_seq_show(struct seq_file *seq, void *v)
2947{
2948 struct fib_route_iter *iter = seq->private;
2949 struct fib_table *tb = iter->main_tb;
2950 struct fib_alias *fa;
2951 struct key_vector *l = v;
2952 __be32 prefix;
2953
2954 if (v == SEQ_START_TOKEN) {
2955 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2956 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2957 "\tWindow\tIRTT");
2958 return 0;
2959 }
2960
2961 prefix = htonl(l->key);
2962
2963 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2964 struct fib_info *fi = fa->fa_info;
2965 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2966 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2967
2968 if ((fa->fa_type == RTN_BROADCAST) ||
2969 (fa->fa_type == RTN_MULTICAST))
2970 continue;
2971
2972 if (fa->tb_id != tb->tb_id)
2973 continue;
2974
2975 seq_setwidth(seq, 127);
2976
2977 if (fi) {
2978 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2979 __be32 gw = 0;
2980
2981 if (nhc->nhc_gw_family == AF_INET)
2982 gw = nhc->nhc_gw.ipv4;
2983
2984 seq_printf(seq,
2985 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2986 "%d\t%08X\t%d\t%u\t%u",
2987 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2988 prefix, gw, flags, 0, 0,
2989 fi->fib_priority,
2990 mask,
2991 (fi->fib_advmss ?
2992 fi->fib_advmss + 40 : 0),
2993 fi->fib_window,
2994 fi->fib_rtt >> 3);
2995 } else {
2996 seq_printf(seq,
2997 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2998 "%d\t%08X\t%d\t%u\t%u",
2999 prefix, 0, flags, 0, 0, 0,
3000 mask, 0, 0, 0);
3001 }
3002 seq_pad(seq, '\n');
3003 }
3004
3005 return 0;
3006}
3007
3008static const struct seq_operations fib_route_seq_ops = {
3009 .start = fib_route_seq_start,
3010 .next = fib_route_seq_next,
3011 .stop = fib_route_seq_stop,
3012 .show = fib_route_seq_show,
3013};
3014
3015int __net_init fib_proc_init(struct net *net)
3016{
3017 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
3018 sizeof(struct fib_trie_iter)))
3019 goto out1;
3020
3021 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
3022 fib_triestat_seq_show, NULL))
3023 goto out2;
3024
3025 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
3026 sizeof(struct fib_route_iter)))
3027 goto out3;
3028
3029 return 0;
3030
3031out3:
3032 remove_proc_entry("fib_triestat", net->proc_net);
3033out2:
3034 remove_proc_entry("fib_trie", net->proc_net);
3035out1:
3036 return -ENOMEM;
3037}
3038
3039void __net_exit fib_proc_exit(struct net *net)
3040{
3041 remove_proc_entry("fib_trie", net->proc_net);
3042 remove_proc_entry("fib_triestat", net->proc_net);
3043 remove_proc_entry("route", net->proc_net);
3044}
3045
3046#endif /* CONFIG_PROC_FS */