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 * http://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
39#define VERSION "0.409"
40
41#include <linux/cache.h>
42#include <linux/uaccess.h>
43#include <linux/bitops.h>
44#include <linux/types.h>
45#include <linux/kernel.h>
46#include <linux/mm.h>
47#include <linux/string.h>
48#include <linux/socket.h>
49#include <linux/sockios.h>
50#include <linux/errno.h>
51#include <linux/in.h>
52#include <linux/inet.h>
53#include <linux/inetdevice.h>
54#include <linux/netdevice.h>
55#include <linux/if_arp.h>
56#include <linux/proc_fs.h>
57#include <linux/rcupdate.h>
58#include <linux/skbuff.h>
59#include <linux/netlink.h>
60#include <linux/init.h>
61#include <linux/list.h>
62#include <linux/slab.h>
63#include <linux/export.h>
64#include <linux/vmalloc.h>
65#include <linux/notifier.h>
66#include <net/net_namespace.h>
67#include <net/ip.h>
68#include <net/protocol.h>
69#include <net/route.h>
70#include <net/tcp.h>
71#include <net/sock.h>
72#include <net/ip_fib.h>
73#include <net/fib_notifier.h>
74#include <trace/events/fib.h>
75#include "fib_lookup.h"
76
77static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
78 enum fib_event_type event_type, u32 dst,
79 int dst_len, struct fib_alias *fa)
80{
81 struct fib_entry_notifier_info info = {
82 .dst = dst,
83 .dst_len = dst_len,
84 .fi = fa->fa_info,
85 .tos = fa->fa_tos,
86 .type = fa->fa_type,
87 .tb_id = fa->tb_id,
88 };
89 return call_fib4_notifier(nb, net, event_type, &info.info);
90}
91
92static int call_fib_entry_notifiers(struct net *net,
93 enum fib_event_type event_type, u32 dst,
94 int dst_len, struct fib_alias *fa,
95 struct netlink_ext_ack *extack)
96{
97 struct fib_entry_notifier_info info = {
98 .info.extack = extack,
99 .dst = dst,
100 .dst_len = dst_len,
101 .fi = fa->fa_info,
102 .tos = fa->fa_tos,
103 .type = fa->fa_type,
104 .tb_id = fa->tb_id,
105 };
106 return call_fib4_notifiers(net, event_type, &info.info);
107}
108
109#define MAX_STAT_DEPTH 32
110
111#define KEYLENGTH (8*sizeof(t_key))
112#define KEY_MAX ((t_key)~0)
113
114typedef unsigned int t_key;
115
116#define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
117#define IS_TNODE(n) ((n)->bits)
118#define IS_LEAF(n) (!(n)->bits)
119
120struct key_vector {
121 t_key key;
122 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
123 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
124 unsigned char slen;
125 union {
126 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
127 struct hlist_head leaf;
128 /* This array is valid if (pos | bits) > 0 (TNODE) */
129 struct key_vector __rcu *tnode[0];
130 };
131};
132
133struct tnode {
134 struct rcu_head rcu;
135 t_key empty_children; /* KEYLENGTH bits needed */
136 t_key full_children; /* KEYLENGTH bits needed */
137 struct key_vector __rcu *parent;
138 struct key_vector kv[1];
139#define tn_bits kv[0].bits
140};
141
142#define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
143#define LEAF_SIZE TNODE_SIZE(1)
144
145#ifdef CONFIG_IP_FIB_TRIE_STATS
146struct trie_use_stats {
147 unsigned int gets;
148 unsigned int backtrack;
149 unsigned int semantic_match_passed;
150 unsigned int semantic_match_miss;
151 unsigned int null_node_hit;
152 unsigned int resize_node_skipped;
153};
154#endif
155
156struct trie_stat {
157 unsigned int totdepth;
158 unsigned int maxdepth;
159 unsigned int tnodes;
160 unsigned int leaves;
161 unsigned int nullpointers;
162 unsigned int prefixes;
163 unsigned int nodesizes[MAX_STAT_DEPTH];
164};
165
166struct trie {
167 struct key_vector kv[1];
168#ifdef CONFIG_IP_FIB_TRIE_STATS
169 struct trie_use_stats __percpu *stats;
170#endif
171};
172
173static struct key_vector *resize(struct trie *t, struct key_vector *tn);
174static unsigned int tnode_free_size;
175
176/*
177 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
178 * especially useful before resizing the root node with PREEMPT_NONE configs;
179 * the value was obtained experimentally, aiming to avoid visible slowdown.
180 */
181unsigned int sysctl_fib_sync_mem = 512 * 1024;
182unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
183unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
184
185static struct kmem_cache *fn_alias_kmem __ro_after_init;
186static struct kmem_cache *trie_leaf_kmem __ro_after_init;
187
188static inline struct tnode *tn_info(struct key_vector *kv)
189{
190 return container_of(kv, struct tnode, kv[0]);
191}
192
193/* caller must hold RTNL */
194#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
195#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
196
197/* caller must hold RCU read lock or RTNL */
198#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
199#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
200
201/* wrapper for rcu_assign_pointer */
202static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
203{
204 if (n)
205 rcu_assign_pointer(tn_info(n)->parent, tp);
206}
207
208#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
209
210/* This provides us with the number of children in this node, in the case of a
211 * leaf this will return 0 meaning none of the children are accessible.
212 */
213static inline unsigned long child_length(const struct key_vector *tn)
214{
215 return (1ul << tn->bits) & ~(1ul);
216}
217
218#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
219
220static inline unsigned long get_index(t_key key, struct key_vector *kv)
221{
222 unsigned long index = key ^ kv->key;
223
224 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
225 return 0;
226
227 return index >> kv->pos;
228}
229
230/* To understand this stuff, an understanding of keys and all their bits is
231 * necessary. Every node in the trie has a key associated with it, but not
232 * all of the bits in that key are significant.
233 *
234 * Consider a node 'n' and its parent 'tp'.
235 *
236 * If n is a leaf, every bit in its key is significant. Its presence is
237 * necessitated by path compression, since during a tree traversal (when
238 * searching for a leaf - unless we are doing an insertion) we will completely
239 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
240 * a potentially successful search, that we have indeed been walking the
241 * correct key path.
242 *
243 * Note that we can never "miss" the correct key in the tree if present by
244 * following the wrong path. Path compression ensures that segments of the key
245 * that are the same for all keys with a given prefix are skipped, but the
246 * skipped part *is* identical for each node in the subtrie below the skipped
247 * bit! trie_insert() in this implementation takes care of that.
248 *
249 * if n is an internal node - a 'tnode' here, the various parts of its key
250 * have many different meanings.
251 *
252 * Example:
253 * _________________________________________________________________
254 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
255 * -----------------------------------------------------------------
256 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
257 *
258 * _________________________________________________________________
259 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
260 * -----------------------------------------------------------------
261 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
262 *
263 * tp->pos = 22
264 * tp->bits = 3
265 * n->pos = 13
266 * n->bits = 4
267 *
268 * First, let's just ignore the bits that come before the parent tp, that is
269 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
270 * point we do not use them for anything.
271 *
272 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
273 * index into the parent's child array. That is, they will be used to find
274 * 'n' among tp's children.
275 *
276 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
277 * for the node n.
278 *
279 * All the bits we have seen so far are significant to the node n. The rest
280 * of the bits are really not needed or indeed known in n->key.
281 *
282 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
283 * n's child array, and will of course be different for each child.
284 *
285 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
286 * at this point.
287 */
288
289static const int halve_threshold = 25;
290static const int inflate_threshold = 50;
291static const int halve_threshold_root = 15;
292static const int inflate_threshold_root = 30;
293
294static void __alias_free_mem(struct rcu_head *head)
295{
296 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
297 kmem_cache_free(fn_alias_kmem, fa);
298}
299
300static inline void alias_free_mem_rcu(struct fib_alias *fa)
301{
302 call_rcu(&fa->rcu, __alias_free_mem);
303}
304
305#define TNODE_KMALLOC_MAX \
306 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
307#define TNODE_VMALLOC_MAX \
308 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
309
310static void __node_free_rcu(struct rcu_head *head)
311{
312 struct tnode *n = container_of(head, struct tnode, rcu);
313
314 if (!n->tn_bits)
315 kmem_cache_free(trie_leaf_kmem, n);
316 else
317 kvfree(n);
318}
319
320#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
321
322static struct tnode *tnode_alloc(int bits)
323{
324 size_t size;
325
326 /* verify bits is within bounds */
327 if (bits > TNODE_VMALLOC_MAX)
328 return NULL;
329
330 /* determine size and verify it is non-zero and didn't overflow */
331 size = TNODE_SIZE(1ul << bits);
332
333 if (size <= PAGE_SIZE)
334 return kzalloc(size, GFP_KERNEL);
335 else
336 return vzalloc(size);
337}
338
339static inline void empty_child_inc(struct key_vector *n)
340{
341 tn_info(n)->empty_children++;
342
343 if (!tn_info(n)->empty_children)
344 tn_info(n)->full_children++;
345}
346
347static inline void empty_child_dec(struct key_vector *n)
348{
349 if (!tn_info(n)->empty_children)
350 tn_info(n)->full_children--;
351
352 tn_info(n)->empty_children--;
353}
354
355static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
356{
357 struct key_vector *l;
358 struct tnode *kv;
359
360 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
361 if (!kv)
362 return NULL;
363
364 /* initialize key vector */
365 l = kv->kv;
366 l->key = key;
367 l->pos = 0;
368 l->bits = 0;
369 l->slen = fa->fa_slen;
370
371 /* link leaf to fib alias */
372 INIT_HLIST_HEAD(&l->leaf);
373 hlist_add_head(&fa->fa_list, &l->leaf);
374
375 return l;
376}
377
378static struct key_vector *tnode_new(t_key key, int pos, int bits)
379{
380 unsigned int shift = pos + bits;
381 struct key_vector *tn;
382 struct tnode *tnode;
383
384 /* verify bits and pos their msb bits clear and values are valid */
385 BUG_ON(!bits || (shift > KEYLENGTH));
386
387 tnode = tnode_alloc(bits);
388 if (!tnode)
389 return NULL;
390
391 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
392 sizeof(struct key_vector *) << bits);
393
394 if (bits == KEYLENGTH)
395 tnode->full_children = 1;
396 else
397 tnode->empty_children = 1ul << bits;
398
399 tn = tnode->kv;
400 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
401 tn->pos = pos;
402 tn->bits = bits;
403 tn->slen = pos;
404
405 return tn;
406}
407
408/* Check whether a tnode 'n' is "full", i.e. it is an internal node
409 * and no bits are skipped. See discussion in dyntree paper p. 6
410 */
411static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
412{
413 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
414}
415
416/* Add a child at position i overwriting the old value.
417 * Update the value of full_children and empty_children.
418 */
419static void put_child(struct key_vector *tn, unsigned long i,
420 struct key_vector *n)
421{
422 struct key_vector *chi = get_child(tn, i);
423 int isfull, wasfull;
424
425 BUG_ON(i >= child_length(tn));
426
427 /* update emptyChildren, overflow into fullChildren */
428 if (!n && chi)
429 empty_child_inc(tn);
430 if (n && !chi)
431 empty_child_dec(tn);
432
433 /* update fullChildren */
434 wasfull = tnode_full(tn, chi);
435 isfull = tnode_full(tn, n);
436
437 if (wasfull && !isfull)
438 tn_info(tn)->full_children--;
439 else if (!wasfull && isfull)
440 tn_info(tn)->full_children++;
441
442 if (n && (tn->slen < n->slen))
443 tn->slen = n->slen;
444
445 rcu_assign_pointer(tn->tnode[i], n);
446}
447
448static void update_children(struct key_vector *tn)
449{
450 unsigned long i;
451
452 /* update all of the child parent pointers */
453 for (i = child_length(tn); i;) {
454 struct key_vector *inode = get_child(tn, --i);
455
456 if (!inode)
457 continue;
458
459 /* Either update the children of a tnode that
460 * already belongs to us or update the child
461 * to point to ourselves.
462 */
463 if (node_parent(inode) == tn)
464 update_children(inode);
465 else
466 node_set_parent(inode, tn);
467 }
468}
469
470static inline void put_child_root(struct key_vector *tp, t_key key,
471 struct key_vector *n)
472{
473 if (IS_TRIE(tp))
474 rcu_assign_pointer(tp->tnode[0], n);
475 else
476 put_child(tp, get_index(key, tp), n);
477}
478
479static inline void tnode_free_init(struct key_vector *tn)
480{
481 tn_info(tn)->rcu.next = NULL;
482}
483
484static inline void tnode_free_append(struct key_vector *tn,
485 struct key_vector *n)
486{
487 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
488 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
489}
490
491static void tnode_free(struct key_vector *tn)
492{
493 struct callback_head *head = &tn_info(tn)->rcu;
494
495 while (head) {
496 head = head->next;
497 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
498 node_free(tn);
499
500 tn = container_of(head, struct tnode, rcu)->kv;
501 }
502
503 if (tnode_free_size >= sysctl_fib_sync_mem) {
504 tnode_free_size = 0;
505 synchronize_rcu();
506 }
507}
508
509static struct key_vector *replace(struct trie *t,
510 struct key_vector *oldtnode,
511 struct key_vector *tn)
512{
513 struct key_vector *tp = node_parent(oldtnode);
514 unsigned long i;
515
516 /* setup the parent pointer out of and back into this node */
517 NODE_INIT_PARENT(tn, tp);
518 put_child_root(tp, tn->key, tn);
519
520 /* update all of the child parent pointers */
521 update_children(tn);
522
523 /* all pointers should be clean so we are done */
524 tnode_free(oldtnode);
525
526 /* resize children now that oldtnode is freed */
527 for (i = child_length(tn); i;) {
528 struct key_vector *inode = get_child(tn, --i);
529
530 /* resize child node */
531 if (tnode_full(tn, inode))
532 tn = resize(t, inode);
533 }
534
535 return tp;
536}
537
538static struct key_vector *inflate(struct trie *t,
539 struct key_vector *oldtnode)
540{
541 struct key_vector *tn;
542 unsigned long i;
543 t_key m;
544
545 pr_debug("In inflate\n");
546
547 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
548 if (!tn)
549 goto notnode;
550
551 /* prepare oldtnode to be freed */
552 tnode_free_init(oldtnode);
553
554 /* Assemble all of the pointers in our cluster, in this case that
555 * represents all of the pointers out of our allocated nodes that
556 * point to existing tnodes and the links between our allocated
557 * nodes.
558 */
559 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
560 struct key_vector *inode = get_child(oldtnode, --i);
561 struct key_vector *node0, *node1;
562 unsigned long j, k;
563
564 /* An empty child */
565 if (!inode)
566 continue;
567
568 /* A leaf or an internal node with skipped bits */
569 if (!tnode_full(oldtnode, inode)) {
570 put_child(tn, get_index(inode->key, tn), inode);
571 continue;
572 }
573
574 /* drop the node in the old tnode free list */
575 tnode_free_append(oldtnode, inode);
576
577 /* An internal node with two children */
578 if (inode->bits == 1) {
579 put_child(tn, 2 * i + 1, get_child(inode, 1));
580 put_child(tn, 2 * i, get_child(inode, 0));
581 continue;
582 }
583
584 /* We will replace this node 'inode' with two new
585 * ones, 'node0' and 'node1', each with half of the
586 * original children. The two new nodes will have
587 * a position one bit further down the key and this
588 * means that the "significant" part of their keys
589 * (see the discussion near the top of this file)
590 * will differ by one bit, which will be "0" in
591 * node0's key and "1" in node1's key. Since we are
592 * moving the key position by one step, the bit that
593 * we are moving away from - the bit at position
594 * (tn->pos) - is the one that will differ between
595 * node0 and node1. So... we synthesize that bit in the
596 * two new keys.
597 */
598 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
599 if (!node1)
600 goto nomem;
601 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
602
603 tnode_free_append(tn, node1);
604 if (!node0)
605 goto nomem;
606 tnode_free_append(tn, node0);
607
608 /* populate child pointers in new nodes */
609 for (k = child_length(inode), j = k / 2; j;) {
610 put_child(node1, --j, get_child(inode, --k));
611 put_child(node0, j, get_child(inode, j));
612 put_child(node1, --j, get_child(inode, --k));
613 put_child(node0, j, get_child(inode, j));
614 }
615
616 /* link new nodes to parent */
617 NODE_INIT_PARENT(node1, tn);
618 NODE_INIT_PARENT(node0, tn);
619
620 /* link parent to nodes */
621 put_child(tn, 2 * i + 1, node1);
622 put_child(tn, 2 * i, node0);
623 }
624
625 /* setup the parent pointers into and out of this node */
626 return replace(t, oldtnode, tn);
627nomem:
628 /* all pointers should be clean so we are done */
629 tnode_free(tn);
630notnode:
631 return NULL;
632}
633
634static struct key_vector *halve(struct trie *t,
635 struct key_vector *oldtnode)
636{
637 struct key_vector *tn;
638 unsigned long i;
639
640 pr_debug("In halve\n");
641
642 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
643 if (!tn)
644 goto notnode;
645
646 /* prepare oldtnode to be freed */
647 tnode_free_init(oldtnode);
648
649 /* Assemble all of the pointers in our cluster, in this case that
650 * represents all of the pointers out of our allocated nodes that
651 * point to existing tnodes and the links between our allocated
652 * nodes.
653 */
654 for (i = child_length(oldtnode); i;) {
655 struct key_vector *node1 = get_child(oldtnode, --i);
656 struct key_vector *node0 = get_child(oldtnode, --i);
657 struct key_vector *inode;
658
659 /* At least one of the children is empty */
660 if (!node1 || !node0) {
661 put_child(tn, i / 2, node1 ? : node0);
662 continue;
663 }
664
665 /* Two nonempty children */
666 inode = tnode_new(node0->key, oldtnode->pos, 1);
667 if (!inode)
668 goto nomem;
669 tnode_free_append(tn, inode);
670
671 /* initialize pointers out of node */
672 put_child(inode, 1, node1);
673 put_child(inode, 0, node0);
674 NODE_INIT_PARENT(inode, tn);
675
676 /* link parent to node */
677 put_child(tn, i / 2, inode);
678 }
679
680 /* setup the parent pointers into and out of this node */
681 return replace(t, oldtnode, tn);
682nomem:
683 /* all pointers should be clean so we are done */
684 tnode_free(tn);
685notnode:
686 return NULL;
687}
688
689static struct key_vector *collapse(struct trie *t,
690 struct key_vector *oldtnode)
691{
692 struct key_vector *n, *tp;
693 unsigned long i;
694
695 /* scan the tnode looking for that one child that might still exist */
696 for (n = NULL, i = child_length(oldtnode); !n && i;)
697 n = get_child(oldtnode, --i);
698
699 /* compress one level */
700 tp = node_parent(oldtnode);
701 put_child_root(tp, oldtnode->key, n);
702 node_set_parent(n, tp);
703
704 /* drop dead node */
705 node_free(oldtnode);
706
707 return tp;
708}
709
710static unsigned char update_suffix(struct key_vector *tn)
711{
712 unsigned char slen = tn->pos;
713 unsigned long stride, i;
714 unsigned char slen_max;
715
716 /* only vector 0 can have a suffix length greater than or equal to
717 * tn->pos + tn->bits, the second highest node will have a suffix
718 * length at most of tn->pos + tn->bits - 1
719 */
720 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
721
722 /* search though the list of children looking for nodes that might
723 * have a suffix greater than the one we currently have. This is
724 * why we start with a stride of 2 since a stride of 1 would
725 * represent the nodes with suffix length equal to tn->pos
726 */
727 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
728 struct key_vector *n = get_child(tn, i);
729
730 if (!n || (n->slen <= slen))
731 continue;
732
733 /* update stride and slen based on new value */
734 stride <<= (n->slen - slen);
735 slen = n->slen;
736 i &= ~(stride - 1);
737
738 /* stop searching if we have hit the maximum possible value */
739 if (slen >= slen_max)
740 break;
741 }
742
743 tn->slen = slen;
744
745 return slen;
746}
747
748/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
749 * the Helsinki University of Technology and Matti Tikkanen of Nokia
750 * Telecommunications, page 6:
751 * "A node is doubled if the ratio of non-empty children to all
752 * children in the *doubled* node is at least 'high'."
753 *
754 * 'high' in this instance is the variable 'inflate_threshold'. It
755 * is expressed as a percentage, so we multiply it with
756 * child_length() and instead of multiplying by 2 (since the
757 * child array will be doubled by inflate()) and multiplying
758 * the left-hand side by 100 (to handle the percentage thing) we
759 * multiply the left-hand side by 50.
760 *
761 * The left-hand side may look a bit weird: child_length(tn)
762 * - tn->empty_children is of course the number of non-null children
763 * in the current node. tn->full_children is the number of "full"
764 * children, that is non-null tnodes with a skip value of 0.
765 * All of those will be doubled in the resulting inflated tnode, so
766 * we just count them one extra time here.
767 *
768 * A clearer way to write this would be:
769 *
770 * to_be_doubled = tn->full_children;
771 * not_to_be_doubled = child_length(tn) - tn->empty_children -
772 * tn->full_children;
773 *
774 * new_child_length = child_length(tn) * 2;
775 *
776 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
777 * new_child_length;
778 * if (new_fill_factor >= inflate_threshold)
779 *
780 * ...and so on, tho it would mess up the while () loop.
781 *
782 * anyway,
783 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
784 * inflate_threshold
785 *
786 * avoid a division:
787 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
788 * inflate_threshold * new_child_length
789 *
790 * expand not_to_be_doubled and to_be_doubled, and shorten:
791 * 100 * (child_length(tn) - tn->empty_children +
792 * tn->full_children) >= inflate_threshold * new_child_length
793 *
794 * expand new_child_length:
795 * 100 * (child_length(tn) - tn->empty_children +
796 * tn->full_children) >=
797 * inflate_threshold * child_length(tn) * 2
798 *
799 * shorten again:
800 * 50 * (tn->full_children + child_length(tn) -
801 * tn->empty_children) >= inflate_threshold *
802 * child_length(tn)
803 *
804 */
805static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
806{
807 unsigned long used = child_length(tn);
808 unsigned long threshold = used;
809
810 /* Keep root node larger */
811 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
812 used -= tn_info(tn)->empty_children;
813 used += tn_info(tn)->full_children;
814
815 /* if bits == KEYLENGTH then pos = 0, and will fail below */
816
817 return (used > 1) && tn->pos && ((50 * used) >= threshold);
818}
819
820static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
821{
822 unsigned long used = child_length(tn);
823 unsigned long threshold = used;
824
825 /* Keep root node larger */
826 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
827 used -= tn_info(tn)->empty_children;
828
829 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
830
831 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
832}
833
834static inline bool should_collapse(struct key_vector *tn)
835{
836 unsigned long used = child_length(tn);
837
838 used -= tn_info(tn)->empty_children;
839
840 /* account for bits == KEYLENGTH case */
841 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
842 used -= KEY_MAX;
843
844 /* One child or none, time to drop us from the trie */
845 return used < 2;
846}
847
848#define MAX_WORK 10
849static struct key_vector *resize(struct trie *t, struct key_vector *tn)
850{
851#ifdef CONFIG_IP_FIB_TRIE_STATS
852 struct trie_use_stats __percpu *stats = t->stats;
853#endif
854 struct key_vector *tp = node_parent(tn);
855 unsigned long cindex = get_index(tn->key, tp);
856 int max_work = MAX_WORK;
857
858 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
859 tn, inflate_threshold, halve_threshold);
860
861 /* track the tnode via the pointer from the parent instead of
862 * doing it ourselves. This way we can let RCU fully do its
863 * thing without us interfering
864 */
865 BUG_ON(tn != get_child(tp, cindex));
866
867 /* Double as long as the resulting node has a number of
868 * nonempty nodes that are above the threshold.
869 */
870 while (should_inflate(tp, tn) && max_work) {
871 tp = inflate(t, tn);
872 if (!tp) {
873#ifdef CONFIG_IP_FIB_TRIE_STATS
874 this_cpu_inc(stats->resize_node_skipped);
875#endif
876 break;
877 }
878
879 max_work--;
880 tn = get_child(tp, cindex);
881 }
882
883 /* update parent in case inflate failed */
884 tp = node_parent(tn);
885
886 /* Return if at least one inflate is run */
887 if (max_work != MAX_WORK)
888 return tp;
889
890 /* Halve as long as the number of empty children in this
891 * node is above threshold.
892 */
893 while (should_halve(tp, tn) && max_work) {
894 tp = halve(t, tn);
895 if (!tp) {
896#ifdef CONFIG_IP_FIB_TRIE_STATS
897 this_cpu_inc(stats->resize_node_skipped);
898#endif
899 break;
900 }
901
902 max_work--;
903 tn = get_child(tp, cindex);
904 }
905
906 /* Only one child remains */
907 if (should_collapse(tn))
908 return collapse(t, tn);
909
910 /* update parent in case halve failed */
911 return node_parent(tn);
912}
913
914static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
915{
916 unsigned char node_slen = tn->slen;
917
918 while ((node_slen > tn->pos) && (node_slen > slen)) {
919 slen = update_suffix(tn);
920 if (node_slen == slen)
921 break;
922
923 tn = node_parent(tn);
924 node_slen = tn->slen;
925 }
926}
927
928static void node_push_suffix(struct key_vector *tn, unsigned char slen)
929{
930 while (tn->slen < slen) {
931 tn->slen = slen;
932 tn = node_parent(tn);
933 }
934}
935
936/* rcu_read_lock needs to be hold by caller from readside */
937static struct key_vector *fib_find_node(struct trie *t,
938 struct key_vector **tp, u32 key)
939{
940 struct key_vector *pn, *n = t->kv;
941 unsigned long index = 0;
942
943 do {
944 pn = n;
945 n = get_child_rcu(n, index);
946
947 if (!n)
948 break;
949
950 index = get_cindex(key, n);
951
952 /* This bit of code is a bit tricky but it combines multiple
953 * checks into a single check. The prefix consists of the
954 * prefix plus zeros for the bits in the cindex. The index
955 * is the difference between the key and this value. From
956 * this we can actually derive several pieces of data.
957 * if (index >= (1ul << bits))
958 * we have a mismatch in skip bits and failed
959 * else
960 * we know the value is cindex
961 *
962 * This check is safe even if bits == KEYLENGTH due to the
963 * fact that we can only allocate a node with 32 bits if a
964 * long is greater than 32 bits.
965 */
966 if (index >= (1ul << n->bits)) {
967 n = NULL;
968 break;
969 }
970
971 /* keep searching until we find a perfect match leaf or NULL */
972 } while (IS_TNODE(n));
973
974 *tp = pn;
975
976 return n;
977}
978
979/* Return the first fib alias matching TOS with
980 * priority less than or equal to PRIO.
981 */
982static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
983 u8 tos, u32 prio, u32 tb_id)
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 (fa->fa_tos > tos)
1000 continue;
1001 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1002 return fa;
1003 }
1004
1005 return NULL;
1006}
1007
1008static void trie_rebalance(struct trie *t, struct key_vector *tn)
1009{
1010 while (!IS_TRIE(tn))
1011 tn = resize(t, tn);
1012}
1013
1014static int fib_insert_node(struct trie *t, struct key_vector *tp,
1015 struct fib_alias *new, t_key key)
1016{
1017 struct key_vector *n, *l;
1018
1019 l = leaf_new(key, new);
1020 if (!l)
1021 goto noleaf;
1022
1023 /* retrieve child from parent node */
1024 n = get_child(tp, get_index(key, tp));
1025
1026 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1027 *
1028 * Add a new tnode here
1029 * first tnode need some special handling
1030 * leaves us in position for handling as case 3
1031 */
1032 if (n) {
1033 struct key_vector *tn;
1034
1035 tn = tnode_new(key, __fls(key ^ n->key), 1);
1036 if (!tn)
1037 goto notnode;
1038
1039 /* initialize routes out of node */
1040 NODE_INIT_PARENT(tn, tp);
1041 put_child(tn, get_index(key, tn) ^ 1, n);
1042
1043 /* start adding routes into the node */
1044 put_child_root(tp, key, tn);
1045 node_set_parent(n, tn);
1046
1047 /* parent now has a NULL spot where the leaf can go */
1048 tp = tn;
1049 }
1050
1051 /* Case 3: n is NULL, and will just insert a new leaf */
1052 node_push_suffix(tp, new->fa_slen);
1053 NODE_INIT_PARENT(l, tp);
1054 put_child_root(tp, key, l);
1055 trie_rebalance(t, tp);
1056
1057 return 0;
1058notnode:
1059 node_free(l);
1060noleaf:
1061 return -ENOMEM;
1062}
1063
1064/* fib notifier for ADD is sent before calling fib_insert_alias with
1065 * the expectation that the only possible failure ENOMEM
1066 */
1067static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1068 struct key_vector *l, struct fib_alias *new,
1069 struct fib_alias *fa, t_key key)
1070{
1071 if (!l)
1072 return fib_insert_node(t, tp, new, key);
1073
1074 if (fa) {
1075 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1076 } else {
1077 struct fib_alias *last;
1078
1079 hlist_for_each_entry(last, &l->leaf, fa_list) {
1080 if (new->fa_slen < last->fa_slen)
1081 break;
1082 if ((new->fa_slen == last->fa_slen) &&
1083 (new->tb_id > last->tb_id))
1084 break;
1085 fa = last;
1086 }
1087
1088 if (fa)
1089 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1090 else
1091 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1092 }
1093
1094 /* if we added to the tail node then we need to update slen */
1095 if (l->slen < new->fa_slen) {
1096 l->slen = new->fa_slen;
1097 node_push_suffix(tp, new->fa_slen);
1098 }
1099
1100 return 0;
1101}
1102
1103static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1104{
1105 if (plen > KEYLENGTH) {
1106 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1107 return false;
1108 }
1109
1110 if ((plen < KEYLENGTH) && (key << plen)) {
1111 NL_SET_ERR_MSG(extack,
1112 "Invalid prefix for given prefix length");
1113 return false;
1114 }
1115
1116 return true;
1117}
1118
1119/* Caller must hold RTNL. */
1120int fib_table_insert(struct net *net, struct fib_table *tb,
1121 struct fib_config *cfg, struct netlink_ext_ack *extack)
1122{
1123 enum fib_event_type event = FIB_EVENT_ENTRY_ADD;
1124 struct trie *t = (struct trie *)tb->tb_data;
1125 struct fib_alias *fa, *new_fa;
1126 struct key_vector *l, *tp;
1127 u16 nlflags = NLM_F_EXCL;
1128 struct fib_info *fi;
1129 u8 plen = cfg->fc_dst_len;
1130 u8 slen = KEYLENGTH - plen;
1131 u8 tos = cfg->fc_tos;
1132 u32 key;
1133 int err;
1134
1135 key = ntohl(cfg->fc_dst);
1136
1137 if (!fib_valid_key_len(key, plen, extack))
1138 return -EINVAL;
1139
1140 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1141
1142 fi = fib_create_info(cfg, extack);
1143 if (IS_ERR(fi)) {
1144 err = PTR_ERR(fi);
1145 goto err;
1146 }
1147
1148 l = fib_find_node(t, &tp, key);
1149 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1150 tb->tb_id) : NULL;
1151
1152 /* Now fa, if non-NULL, points to the first fib alias
1153 * with the same keys [prefix,tos,priority], if such key already
1154 * exists or to the node before which we will insert new one.
1155 *
1156 * If fa is NULL, we will need to allocate a new one and
1157 * insert to the tail of the section matching the suffix length
1158 * of the new alias.
1159 */
1160
1161 if (fa && fa->fa_tos == tos &&
1162 fa->fa_info->fib_priority == fi->fib_priority) {
1163 struct fib_alias *fa_first, *fa_match;
1164
1165 err = -EEXIST;
1166 if (cfg->fc_nlflags & NLM_F_EXCL)
1167 goto out;
1168
1169 nlflags &= ~NLM_F_EXCL;
1170
1171 /* We have 2 goals:
1172 * 1. Find exact match for type, scope, fib_info to avoid
1173 * duplicate routes
1174 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1175 */
1176 fa_match = NULL;
1177 fa_first = fa;
1178 hlist_for_each_entry_from(fa, fa_list) {
1179 if ((fa->fa_slen != slen) ||
1180 (fa->tb_id != tb->tb_id) ||
1181 (fa->fa_tos != tos))
1182 break;
1183 if (fa->fa_info->fib_priority != fi->fib_priority)
1184 break;
1185 if (fa->fa_type == cfg->fc_type &&
1186 fa->fa_info == fi) {
1187 fa_match = fa;
1188 break;
1189 }
1190 }
1191
1192 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1193 struct fib_info *fi_drop;
1194 u8 state;
1195
1196 nlflags |= NLM_F_REPLACE;
1197 fa = fa_first;
1198 if (fa_match) {
1199 if (fa == fa_match)
1200 err = 0;
1201 goto out;
1202 }
1203 err = -ENOBUFS;
1204 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1205 if (!new_fa)
1206 goto out;
1207
1208 fi_drop = fa->fa_info;
1209 new_fa->fa_tos = fa->fa_tos;
1210 new_fa->fa_info = fi;
1211 new_fa->fa_type = cfg->fc_type;
1212 state = fa->fa_state;
1213 new_fa->fa_state = state & ~FA_S_ACCESSED;
1214 new_fa->fa_slen = fa->fa_slen;
1215 new_fa->tb_id = tb->tb_id;
1216 new_fa->fa_default = -1;
1217
1218 err = call_fib_entry_notifiers(net,
1219 FIB_EVENT_ENTRY_REPLACE,
1220 key, plen, new_fa,
1221 extack);
1222 if (err)
1223 goto out_free_new_fa;
1224
1225 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1226 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1227
1228 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1229
1230 alias_free_mem_rcu(fa);
1231
1232 fib_release_info(fi_drop);
1233 if (state & FA_S_ACCESSED)
1234 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1235
1236 goto succeeded;
1237 }
1238 /* Error if we find a perfect match which
1239 * uses the same scope, type, and nexthop
1240 * information.
1241 */
1242 if (fa_match)
1243 goto out;
1244
1245 if (cfg->fc_nlflags & NLM_F_APPEND) {
1246 event = FIB_EVENT_ENTRY_APPEND;
1247 nlflags |= NLM_F_APPEND;
1248 } else {
1249 fa = fa_first;
1250 }
1251 }
1252 err = -ENOENT;
1253 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1254 goto out;
1255
1256 nlflags |= NLM_F_CREATE;
1257 err = -ENOBUFS;
1258 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1259 if (!new_fa)
1260 goto out;
1261
1262 new_fa->fa_info = fi;
1263 new_fa->fa_tos = tos;
1264 new_fa->fa_type = cfg->fc_type;
1265 new_fa->fa_state = 0;
1266 new_fa->fa_slen = slen;
1267 new_fa->tb_id = tb->tb_id;
1268 new_fa->fa_default = -1;
1269
1270 err = call_fib_entry_notifiers(net, event, key, plen, new_fa, extack);
1271 if (err)
1272 goto out_free_new_fa;
1273
1274 /* Insert new entry to the list. */
1275 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1276 if (err)
1277 goto out_fib_notif;
1278
1279 if (!plen)
1280 tb->tb_num_default++;
1281
1282 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1283 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1284 &cfg->fc_nlinfo, nlflags);
1285succeeded:
1286 return 0;
1287
1288out_fib_notif:
1289 /* notifier was sent that entry would be added to trie, but
1290 * the add failed and need to recover. Only failure for
1291 * fib_insert_alias is ENOMEM.
1292 */
1293 NL_SET_ERR_MSG(extack, "Failed to insert route into trie");
1294 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key,
1295 plen, new_fa, NULL);
1296out_free_new_fa:
1297 kmem_cache_free(fn_alias_kmem, new_fa);
1298out:
1299 fib_release_info(fi);
1300err:
1301 return err;
1302}
1303
1304static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1305{
1306 t_key prefix = n->key;
1307
1308 return (key ^ prefix) & (prefix | -prefix);
1309}
1310
1311/* should be called with rcu_read_lock */
1312int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1313 struct fib_result *res, int fib_flags)
1314{
1315 struct trie *t = (struct trie *) tb->tb_data;
1316#ifdef CONFIG_IP_FIB_TRIE_STATS
1317 struct trie_use_stats __percpu *stats = t->stats;
1318#endif
1319 const t_key key = ntohl(flp->daddr);
1320 struct key_vector *n, *pn;
1321 struct fib_alias *fa;
1322 unsigned long index;
1323 t_key cindex;
1324
1325 pn = t->kv;
1326 cindex = 0;
1327
1328 n = get_child_rcu(pn, cindex);
1329 if (!n) {
1330 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1331 return -EAGAIN;
1332 }
1333
1334#ifdef CONFIG_IP_FIB_TRIE_STATS
1335 this_cpu_inc(stats->gets);
1336#endif
1337
1338 /* Step 1: Travel to the longest prefix match in the trie */
1339 for (;;) {
1340 index = get_cindex(key, n);
1341
1342 /* This bit of code is a bit tricky but it combines multiple
1343 * checks into a single check. The prefix consists of the
1344 * prefix plus zeros for the "bits" in the prefix. The index
1345 * is the difference between the key and this value. From
1346 * this we can actually derive several pieces of data.
1347 * if (index >= (1ul << bits))
1348 * we have a mismatch in skip bits and failed
1349 * else
1350 * we know the value is cindex
1351 *
1352 * This check is safe even if bits == KEYLENGTH due to the
1353 * fact that we can only allocate a node with 32 bits if a
1354 * long is greater than 32 bits.
1355 */
1356 if (index >= (1ul << n->bits))
1357 break;
1358
1359 /* we have found a leaf. Prefixes have already been compared */
1360 if (IS_LEAF(n))
1361 goto found;
1362
1363 /* only record pn and cindex if we are going to be chopping
1364 * bits later. Otherwise we are just wasting cycles.
1365 */
1366 if (n->slen > n->pos) {
1367 pn = n;
1368 cindex = index;
1369 }
1370
1371 n = get_child_rcu(n, index);
1372 if (unlikely(!n))
1373 goto backtrace;
1374 }
1375
1376 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1377 for (;;) {
1378 /* record the pointer where our next node pointer is stored */
1379 struct key_vector __rcu **cptr = n->tnode;
1380
1381 /* This test verifies that none of the bits that differ
1382 * between the key and the prefix exist in the region of
1383 * the lsb and higher in the prefix.
1384 */
1385 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1386 goto backtrace;
1387
1388 /* exit out and process leaf */
1389 if (unlikely(IS_LEAF(n)))
1390 break;
1391
1392 /* Don't bother recording parent info. Since we are in
1393 * prefix match mode we will have to come back to wherever
1394 * we started this traversal anyway
1395 */
1396
1397 while ((n = rcu_dereference(*cptr)) == NULL) {
1398backtrace:
1399#ifdef CONFIG_IP_FIB_TRIE_STATS
1400 if (!n)
1401 this_cpu_inc(stats->null_node_hit);
1402#endif
1403 /* If we are at cindex 0 there are no more bits for
1404 * us to strip at this level so we must ascend back
1405 * up one level to see if there are any more bits to
1406 * be stripped there.
1407 */
1408 while (!cindex) {
1409 t_key pkey = pn->key;
1410
1411 /* If we don't have a parent then there is
1412 * nothing for us to do as we do not have any
1413 * further nodes to parse.
1414 */
1415 if (IS_TRIE(pn)) {
1416 trace_fib_table_lookup(tb->tb_id, flp,
1417 NULL, -EAGAIN);
1418 return -EAGAIN;
1419 }
1420#ifdef CONFIG_IP_FIB_TRIE_STATS
1421 this_cpu_inc(stats->backtrack);
1422#endif
1423 /* Get Child's index */
1424 pn = node_parent_rcu(pn);
1425 cindex = get_index(pkey, pn);
1426 }
1427
1428 /* strip the least significant bit from the cindex */
1429 cindex &= cindex - 1;
1430
1431 /* grab pointer for next child node */
1432 cptr = &pn->tnode[cindex];
1433 }
1434 }
1435
1436found:
1437 /* this line carries forward the xor from earlier in the function */
1438 index = key ^ n->key;
1439
1440 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1441 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1442 struct fib_info *fi = fa->fa_info;
1443 int nhsel, err;
1444
1445 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1446 if (index >= (1ul << fa->fa_slen))
1447 continue;
1448 }
1449 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1450 continue;
1451 if (fi->fib_dead)
1452 continue;
1453 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1454 continue;
1455 fib_alias_accessed(fa);
1456 err = fib_props[fa->fa_type].error;
1457 if (unlikely(err < 0)) {
1458out_reject:
1459#ifdef CONFIG_IP_FIB_TRIE_STATS
1460 this_cpu_inc(stats->semantic_match_passed);
1461#endif
1462 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1463 return err;
1464 }
1465 if (fi->fib_flags & RTNH_F_DEAD)
1466 continue;
1467
1468 if (unlikely(fi->nh && nexthop_is_blackhole(fi->nh))) {
1469 err = fib_props[RTN_BLACKHOLE].error;
1470 goto out_reject;
1471 }
1472
1473 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1474 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
1475
1476 if (nhc->nhc_flags & RTNH_F_DEAD)
1477 continue;
1478 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1479 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1480 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1481 continue;
1482 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1483 if (flp->flowi4_oif &&
1484 flp->flowi4_oif != nhc->nhc_oif)
1485 continue;
1486 }
1487
1488 if (!(fib_flags & FIB_LOOKUP_NOREF))
1489 refcount_inc(&fi->fib_clntref);
1490
1491 res->prefix = htonl(n->key);
1492 res->prefixlen = KEYLENGTH - fa->fa_slen;
1493 res->nh_sel = nhsel;
1494 res->nhc = nhc;
1495 res->type = fa->fa_type;
1496 res->scope = fi->fib_scope;
1497 res->fi = fi;
1498 res->table = tb;
1499 res->fa_head = &n->leaf;
1500#ifdef CONFIG_IP_FIB_TRIE_STATS
1501 this_cpu_inc(stats->semantic_match_passed);
1502#endif
1503 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1504
1505 return err;
1506 }
1507 }
1508#ifdef CONFIG_IP_FIB_TRIE_STATS
1509 this_cpu_inc(stats->semantic_match_miss);
1510#endif
1511 goto backtrace;
1512}
1513EXPORT_SYMBOL_GPL(fib_table_lookup);
1514
1515static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1516 struct key_vector *l, struct fib_alias *old)
1517{
1518 /* record the location of the previous list_info entry */
1519 struct hlist_node **pprev = old->fa_list.pprev;
1520 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1521
1522 /* remove the fib_alias from the list */
1523 hlist_del_rcu(&old->fa_list);
1524
1525 /* if we emptied the list this leaf will be freed and we can sort
1526 * out parent suffix lengths as a part of trie_rebalance
1527 */
1528 if (hlist_empty(&l->leaf)) {
1529 if (tp->slen == l->slen)
1530 node_pull_suffix(tp, tp->pos);
1531 put_child_root(tp, l->key, NULL);
1532 node_free(l);
1533 trie_rebalance(t, tp);
1534 return;
1535 }
1536
1537 /* only access fa if it is pointing at the last valid hlist_node */
1538 if (*pprev)
1539 return;
1540
1541 /* update the trie with the latest suffix length */
1542 l->slen = fa->fa_slen;
1543 node_pull_suffix(tp, fa->fa_slen);
1544}
1545
1546/* Caller must hold RTNL. */
1547int fib_table_delete(struct net *net, struct fib_table *tb,
1548 struct fib_config *cfg, struct netlink_ext_ack *extack)
1549{
1550 struct trie *t = (struct trie *) tb->tb_data;
1551 struct fib_alias *fa, *fa_to_delete;
1552 struct key_vector *l, *tp;
1553 u8 plen = cfg->fc_dst_len;
1554 u8 slen = KEYLENGTH - plen;
1555 u8 tos = cfg->fc_tos;
1556 u32 key;
1557
1558 key = ntohl(cfg->fc_dst);
1559
1560 if (!fib_valid_key_len(key, plen, extack))
1561 return -EINVAL;
1562
1563 l = fib_find_node(t, &tp, key);
1564 if (!l)
1565 return -ESRCH;
1566
1567 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1568 if (!fa)
1569 return -ESRCH;
1570
1571 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1572
1573 fa_to_delete = NULL;
1574 hlist_for_each_entry_from(fa, fa_list) {
1575 struct fib_info *fi = fa->fa_info;
1576
1577 if ((fa->fa_slen != slen) ||
1578 (fa->tb_id != tb->tb_id) ||
1579 (fa->fa_tos != tos))
1580 break;
1581
1582 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1583 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1584 fa->fa_info->fib_scope == cfg->fc_scope) &&
1585 (!cfg->fc_prefsrc ||
1586 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1587 (!cfg->fc_protocol ||
1588 fi->fib_protocol == cfg->fc_protocol) &&
1589 fib_nh_match(cfg, fi, extack) == 0 &&
1590 fib_metrics_match(cfg, fi)) {
1591 fa_to_delete = fa;
1592 break;
1593 }
1594 }
1595
1596 if (!fa_to_delete)
1597 return -ESRCH;
1598
1599 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1600 fa_to_delete, extack);
1601 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1602 &cfg->fc_nlinfo, 0);
1603
1604 if (!plen)
1605 tb->tb_num_default--;
1606
1607 fib_remove_alias(t, tp, l, fa_to_delete);
1608
1609 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1610 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1611
1612 fib_release_info(fa_to_delete->fa_info);
1613 alias_free_mem_rcu(fa_to_delete);
1614 return 0;
1615}
1616
1617/* Scan for the next leaf starting at the provided key value */
1618static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1619{
1620 struct key_vector *pn, *n = *tn;
1621 unsigned long cindex;
1622
1623 /* this loop is meant to try and find the key in the trie */
1624 do {
1625 /* record parent and next child index */
1626 pn = n;
1627 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1628
1629 if (cindex >> pn->bits)
1630 break;
1631
1632 /* descend into the next child */
1633 n = get_child_rcu(pn, cindex++);
1634 if (!n)
1635 break;
1636
1637 /* guarantee forward progress on the keys */
1638 if (IS_LEAF(n) && (n->key >= key))
1639 goto found;
1640 } while (IS_TNODE(n));
1641
1642 /* this loop will search for the next leaf with a greater key */
1643 while (!IS_TRIE(pn)) {
1644 /* if we exhausted the parent node we will need to climb */
1645 if (cindex >= (1ul << pn->bits)) {
1646 t_key pkey = pn->key;
1647
1648 pn = node_parent_rcu(pn);
1649 cindex = get_index(pkey, pn) + 1;
1650 continue;
1651 }
1652
1653 /* grab the next available node */
1654 n = get_child_rcu(pn, cindex++);
1655 if (!n)
1656 continue;
1657
1658 /* no need to compare keys since we bumped the index */
1659 if (IS_LEAF(n))
1660 goto found;
1661
1662 /* Rescan start scanning in new node */
1663 pn = n;
1664 cindex = 0;
1665 }
1666
1667 *tn = pn;
1668 return NULL; /* Root of trie */
1669found:
1670 /* if we are at the limit for keys just return NULL for the tnode */
1671 *tn = pn;
1672 return n;
1673}
1674
1675static void fib_trie_free(struct fib_table *tb)
1676{
1677 struct trie *t = (struct trie *)tb->tb_data;
1678 struct key_vector *pn = t->kv;
1679 unsigned long cindex = 1;
1680 struct hlist_node *tmp;
1681 struct fib_alias *fa;
1682
1683 /* walk trie in reverse order and free everything */
1684 for (;;) {
1685 struct key_vector *n;
1686
1687 if (!(cindex--)) {
1688 t_key pkey = pn->key;
1689
1690 if (IS_TRIE(pn))
1691 break;
1692
1693 n = pn;
1694 pn = node_parent(pn);
1695
1696 /* drop emptied tnode */
1697 put_child_root(pn, n->key, NULL);
1698 node_free(n);
1699
1700 cindex = get_index(pkey, pn);
1701
1702 continue;
1703 }
1704
1705 /* grab the next available node */
1706 n = get_child(pn, cindex);
1707 if (!n)
1708 continue;
1709
1710 if (IS_TNODE(n)) {
1711 /* record pn and cindex for leaf walking */
1712 pn = n;
1713 cindex = 1ul << n->bits;
1714
1715 continue;
1716 }
1717
1718 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1719 hlist_del_rcu(&fa->fa_list);
1720 alias_free_mem_rcu(fa);
1721 }
1722
1723 put_child_root(pn, n->key, NULL);
1724 node_free(n);
1725 }
1726
1727#ifdef CONFIG_IP_FIB_TRIE_STATS
1728 free_percpu(t->stats);
1729#endif
1730 kfree(tb);
1731}
1732
1733struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1734{
1735 struct trie *ot = (struct trie *)oldtb->tb_data;
1736 struct key_vector *l, *tp = ot->kv;
1737 struct fib_table *local_tb;
1738 struct fib_alias *fa;
1739 struct trie *lt;
1740 t_key key = 0;
1741
1742 if (oldtb->tb_data == oldtb->__data)
1743 return oldtb;
1744
1745 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1746 if (!local_tb)
1747 return NULL;
1748
1749 lt = (struct trie *)local_tb->tb_data;
1750
1751 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1752 struct key_vector *local_l = NULL, *local_tp;
1753
1754 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1755 struct fib_alias *new_fa;
1756
1757 if (local_tb->tb_id != fa->tb_id)
1758 continue;
1759
1760 /* clone fa for new local table */
1761 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1762 if (!new_fa)
1763 goto out;
1764
1765 memcpy(new_fa, fa, sizeof(*fa));
1766
1767 /* insert clone into table */
1768 if (!local_l)
1769 local_l = fib_find_node(lt, &local_tp, l->key);
1770
1771 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1772 NULL, l->key)) {
1773 kmem_cache_free(fn_alias_kmem, new_fa);
1774 goto out;
1775 }
1776 }
1777
1778 /* stop loop if key wrapped back to 0 */
1779 key = l->key + 1;
1780 if (key < l->key)
1781 break;
1782 }
1783
1784 return local_tb;
1785out:
1786 fib_trie_free(local_tb);
1787
1788 return NULL;
1789}
1790
1791/* Caller must hold RTNL */
1792void fib_table_flush_external(struct fib_table *tb)
1793{
1794 struct trie *t = (struct trie *)tb->tb_data;
1795 struct key_vector *pn = t->kv;
1796 unsigned long cindex = 1;
1797 struct hlist_node *tmp;
1798 struct fib_alias *fa;
1799
1800 /* walk trie in reverse order */
1801 for (;;) {
1802 unsigned char slen = 0;
1803 struct key_vector *n;
1804
1805 if (!(cindex--)) {
1806 t_key pkey = pn->key;
1807
1808 /* cannot resize the trie vector */
1809 if (IS_TRIE(pn))
1810 break;
1811
1812 /* update the suffix to address pulled leaves */
1813 if (pn->slen > pn->pos)
1814 update_suffix(pn);
1815
1816 /* resize completed node */
1817 pn = resize(t, pn);
1818 cindex = get_index(pkey, pn);
1819
1820 continue;
1821 }
1822
1823 /* grab the next available node */
1824 n = get_child(pn, cindex);
1825 if (!n)
1826 continue;
1827
1828 if (IS_TNODE(n)) {
1829 /* record pn and cindex for leaf walking */
1830 pn = n;
1831 cindex = 1ul << n->bits;
1832
1833 continue;
1834 }
1835
1836 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1837 /* if alias was cloned to local then we just
1838 * need to remove the local copy from main
1839 */
1840 if (tb->tb_id != fa->tb_id) {
1841 hlist_del_rcu(&fa->fa_list);
1842 alias_free_mem_rcu(fa);
1843 continue;
1844 }
1845
1846 /* record local slen */
1847 slen = fa->fa_slen;
1848 }
1849
1850 /* update leaf slen */
1851 n->slen = slen;
1852
1853 if (hlist_empty(&n->leaf)) {
1854 put_child_root(pn, n->key, NULL);
1855 node_free(n);
1856 }
1857 }
1858}
1859
1860/* Caller must hold RTNL. */
1861int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1862{
1863 struct trie *t = (struct trie *)tb->tb_data;
1864 struct key_vector *pn = t->kv;
1865 unsigned long cindex = 1;
1866 struct hlist_node *tmp;
1867 struct fib_alias *fa;
1868 int found = 0;
1869
1870 /* walk trie in reverse order */
1871 for (;;) {
1872 unsigned char slen = 0;
1873 struct key_vector *n;
1874
1875 if (!(cindex--)) {
1876 t_key pkey = pn->key;
1877
1878 /* cannot resize the trie vector */
1879 if (IS_TRIE(pn))
1880 break;
1881
1882 /* update the suffix to address pulled leaves */
1883 if (pn->slen > pn->pos)
1884 update_suffix(pn);
1885
1886 /* resize completed node */
1887 pn = resize(t, pn);
1888 cindex = get_index(pkey, pn);
1889
1890 continue;
1891 }
1892
1893 /* grab the next available node */
1894 n = get_child(pn, cindex);
1895 if (!n)
1896 continue;
1897
1898 if (IS_TNODE(n)) {
1899 /* record pn and cindex for leaf walking */
1900 pn = n;
1901 cindex = 1ul << n->bits;
1902
1903 continue;
1904 }
1905
1906 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1907 struct fib_info *fi = fa->fa_info;
1908
1909 if (!fi || tb->tb_id != fa->tb_id ||
1910 (!(fi->fib_flags & RTNH_F_DEAD) &&
1911 !fib_props[fa->fa_type].error)) {
1912 slen = fa->fa_slen;
1913 continue;
1914 }
1915
1916 /* Do not flush error routes if network namespace is
1917 * not being dismantled
1918 */
1919 if (!flush_all && fib_props[fa->fa_type].error) {
1920 slen = fa->fa_slen;
1921 continue;
1922 }
1923
1924 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1925 n->key,
1926 KEYLENGTH - fa->fa_slen, fa,
1927 NULL);
1928 hlist_del_rcu(&fa->fa_list);
1929 fib_release_info(fa->fa_info);
1930 alias_free_mem_rcu(fa);
1931 found++;
1932 }
1933
1934 /* update leaf slen */
1935 n->slen = slen;
1936
1937 if (hlist_empty(&n->leaf)) {
1938 put_child_root(pn, n->key, NULL);
1939 node_free(n);
1940 }
1941 }
1942
1943 pr_debug("trie_flush found=%d\n", found);
1944 return found;
1945}
1946
1947/* derived from fib_trie_free */
1948static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
1949 struct nl_info *info)
1950{
1951 struct trie *t = (struct trie *)tb->tb_data;
1952 struct key_vector *pn = t->kv;
1953 unsigned long cindex = 1;
1954 struct fib_alias *fa;
1955
1956 for (;;) {
1957 struct key_vector *n;
1958
1959 if (!(cindex--)) {
1960 t_key pkey = pn->key;
1961
1962 if (IS_TRIE(pn))
1963 break;
1964
1965 pn = node_parent(pn);
1966 cindex = get_index(pkey, pn);
1967 continue;
1968 }
1969
1970 /* grab the next available node */
1971 n = get_child(pn, cindex);
1972 if (!n)
1973 continue;
1974
1975 if (IS_TNODE(n)) {
1976 /* record pn and cindex for leaf walking */
1977 pn = n;
1978 cindex = 1ul << n->bits;
1979
1980 continue;
1981 }
1982
1983 hlist_for_each_entry(fa, &n->leaf, fa_list) {
1984 struct fib_info *fi = fa->fa_info;
1985
1986 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
1987 continue;
1988
1989 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
1990 KEYLENGTH - fa->fa_slen, tb->tb_id,
1991 info, NLM_F_REPLACE);
1992
1993 /* call_fib_entry_notifiers will be removed when
1994 * in-kernel notifier is implemented and supported
1995 * for nexthop objects
1996 */
1997 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
1998 n->key,
1999 KEYLENGTH - fa->fa_slen, fa,
2000 NULL);
2001 }
2002 }
2003}
2004
2005void fib_info_notify_update(struct net *net, struct nl_info *info)
2006{
2007 unsigned int h;
2008
2009 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2010 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2011 struct fib_table *tb;
2012
2013 hlist_for_each_entry_rcu(tb, head, tb_hlist)
2014 __fib_info_notify_update(net, tb, info);
2015 }
2016}
2017
2018static void fib_leaf_notify(struct net *net, struct key_vector *l,
2019 struct fib_table *tb, struct notifier_block *nb)
2020{
2021 struct fib_alias *fa;
2022
2023 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2024 struct fib_info *fi = fa->fa_info;
2025
2026 if (!fi)
2027 continue;
2028
2029 /* local and main table can share the same trie,
2030 * so don't notify twice for the same entry.
2031 */
2032 if (tb->tb_id != fa->tb_id)
2033 continue;
2034
2035 call_fib_entry_notifier(nb, net, FIB_EVENT_ENTRY_ADD, l->key,
2036 KEYLENGTH - fa->fa_slen, fa);
2037 }
2038}
2039
2040static void fib_table_notify(struct net *net, struct fib_table *tb,
2041 struct notifier_block *nb)
2042{
2043 struct trie *t = (struct trie *)tb->tb_data;
2044 struct key_vector *l, *tp = t->kv;
2045 t_key key = 0;
2046
2047 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2048 fib_leaf_notify(net, l, tb, nb);
2049
2050 key = l->key + 1;
2051 /* stop in case of wrap around */
2052 if (key < l->key)
2053 break;
2054 }
2055}
2056
2057void fib_notify(struct net *net, struct notifier_block *nb)
2058{
2059 unsigned int h;
2060
2061 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2062 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2063 struct fib_table *tb;
2064
2065 hlist_for_each_entry_rcu(tb, head, tb_hlist)
2066 fib_table_notify(net, tb, nb);
2067 }
2068}
2069
2070static void __trie_free_rcu(struct rcu_head *head)
2071{
2072 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2073#ifdef CONFIG_IP_FIB_TRIE_STATS
2074 struct trie *t = (struct trie *)tb->tb_data;
2075
2076 if (tb->tb_data == tb->__data)
2077 free_percpu(t->stats);
2078#endif /* CONFIG_IP_FIB_TRIE_STATS */
2079 kfree(tb);
2080}
2081
2082void fib_free_table(struct fib_table *tb)
2083{
2084 call_rcu(&tb->rcu, __trie_free_rcu);
2085}
2086
2087static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2088 struct sk_buff *skb, struct netlink_callback *cb,
2089 struct fib_dump_filter *filter)
2090{
2091 unsigned int flags = NLM_F_MULTI;
2092 __be32 xkey = htonl(l->key);
2093 int i, s_i, i_fa, s_fa, err;
2094 struct fib_alias *fa;
2095
2096 if (filter->filter_set ||
2097 !filter->dump_exceptions || !filter->dump_routes)
2098 flags |= NLM_F_DUMP_FILTERED;
2099
2100 s_i = cb->args[4];
2101 s_fa = cb->args[5];
2102 i = 0;
2103
2104 /* rcu_read_lock is hold by caller */
2105 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2106 struct fib_info *fi = fa->fa_info;
2107
2108 if (i < s_i)
2109 goto next;
2110
2111 i_fa = 0;
2112
2113 if (tb->tb_id != fa->tb_id)
2114 goto next;
2115
2116 if (filter->filter_set) {
2117 if (filter->rt_type && fa->fa_type != filter->rt_type)
2118 goto next;
2119
2120 if ((filter->protocol &&
2121 fi->fib_protocol != filter->protocol))
2122 goto next;
2123
2124 if (filter->dev &&
2125 !fib_info_nh_uses_dev(fi, filter->dev))
2126 goto next;
2127 }
2128
2129 if (filter->dump_routes) {
2130 if (!s_fa) {
2131 err = fib_dump_info(skb,
2132 NETLINK_CB(cb->skb).portid,
2133 cb->nlh->nlmsg_seq,
2134 RTM_NEWROUTE,
2135 tb->tb_id, fa->fa_type,
2136 xkey,
2137 KEYLENGTH - fa->fa_slen,
2138 fa->fa_tos, fi, flags);
2139 if (err < 0)
2140 goto stop;
2141 }
2142
2143 i_fa++;
2144 }
2145
2146 if (filter->dump_exceptions) {
2147 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2148 &i_fa, s_fa, flags);
2149 if (err < 0)
2150 goto stop;
2151 }
2152
2153next:
2154 i++;
2155 }
2156
2157 cb->args[4] = i;
2158 return skb->len;
2159
2160stop:
2161 cb->args[4] = i;
2162 cb->args[5] = i_fa;
2163 return err;
2164}
2165
2166/* rcu_read_lock needs to be hold by caller from readside */
2167int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2168 struct netlink_callback *cb, struct fib_dump_filter *filter)
2169{
2170 struct trie *t = (struct trie *)tb->tb_data;
2171 struct key_vector *l, *tp = t->kv;
2172 /* Dump starting at last key.
2173 * Note: 0.0.0.0/0 (ie default) is first key.
2174 */
2175 int count = cb->args[2];
2176 t_key key = cb->args[3];
2177
2178 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2179 int err;
2180
2181 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2182 if (err < 0) {
2183 cb->args[3] = key;
2184 cb->args[2] = count;
2185 return err;
2186 }
2187
2188 ++count;
2189 key = l->key + 1;
2190
2191 memset(&cb->args[4], 0,
2192 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2193
2194 /* stop loop if key wrapped back to 0 */
2195 if (key < l->key)
2196 break;
2197 }
2198
2199 cb->args[3] = key;
2200 cb->args[2] = count;
2201
2202 return skb->len;
2203}
2204
2205void __init fib_trie_init(void)
2206{
2207 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2208 sizeof(struct fib_alias),
2209 0, SLAB_PANIC, NULL);
2210
2211 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2212 LEAF_SIZE,
2213 0, SLAB_PANIC, NULL);
2214}
2215
2216struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2217{
2218 struct fib_table *tb;
2219 struct trie *t;
2220 size_t sz = sizeof(*tb);
2221
2222 if (!alias)
2223 sz += sizeof(struct trie);
2224
2225 tb = kzalloc(sz, GFP_KERNEL);
2226 if (!tb)
2227 return NULL;
2228
2229 tb->tb_id = id;
2230 tb->tb_num_default = 0;
2231 tb->tb_data = (alias ? alias->__data : tb->__data);
2232
2233 if (alias)
2234 return tb;
2235
2236 t = (struct trie *) tb->tb_data;
2237 t->kv[0].pos = KEYLENGTH;
2238 t->kv[0].slen = KEYLENGTH;
2239#ifdef CONFIG_IP_FIB_TRIE_STATS
2240 t->stats = alloc_percpu(struct trie_use_stats);
2241 if (!t->stats) {
2242 kfree(tb);
2243 tb = NULL;
2244 }
2245#endif
2246
2247 return tb;
2248}
2249
2250#ifdef CONFIG_PROC_FS
2251/* Depth first Trie walk iterator */
2252struct fib_trie_iter {
2253 struct seq_net_private p;
2254 struct fib_table *tb;
2255 struct key_vector *tnode;
2256 unsigned int index;
2257 unsigned int depth;
2258};
2259
2260static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2261{
2262 unsigned long cindex = iter->index;
2263 struct key_vector *pn = iter->tnode;
2264 t_key pkey;
2265
2266 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2267 iter->tnode, iter->index, iter->depth);
2268
2269 while (!IS_TRIE(pn)) {
2270 while (cindex < child_length(pn)) {
2271 struct key_vector *n = get_child_rcu(pn, cindex++);
2272
2273 if (!n)
2274 continue;
2275
2276 if (IS_LEAF(n)) {
2277 iter->tnode = pn;
2278 iter->index = cindex;
2279 } else {
2280 /* push down one level */
2281 iter->tnode = n;
2282 iter->index = 0;
2283 ++iter->depth;
2284 }
2285
2286 return n;
2287 }
2288
2289 /* Current node exhausted, pop back up */
2290 pkey = pn->key;
2291 pn = node_parent_rcu(pn);
2292 cindex = get_index(pkey, pn) + 1;
2293 --iter->depth;
2294 }
2295
2296 /* record root node so further searches know we are done */
2297 iter->tnode = pn;
2298 iter->index = 0;
2299
2300 return NULL;
2301}
2302
2303static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2304 struct trie *t)
2305{
2306 struct key_vector *n, *pn;
2307
2308 if (!t)
2309 return NULL;
2310
2311 pn = t->kv;
2312 n = rcu_dereference(pn->tnode[0]);
2313 if (!n)
2314 return NULL;
2315
2316 if (IS_TNODE(n)) {
2317 iter->tnode = n;
2318 iter->index = 0;
2319 iter->depth = 1;
2320 } else {
2321 iter->tnode = pn;
2322 iter->index = 0;
2323 iter->depth = 0;
2324 }
2325
2326 return n;
2327}
2328
2329static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2330{
2331 struct key_vector *n;
2332 struct fib_trie_iter iter;
2333
2334 memset(s, 0, sizeof(*s));
2335
2336 rcu_read_lock();
2337 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2338 if (IS_LEAF(n)) {
2339 struct fib_alias *fa;
2340
2341 s->leaves++;
2342 s->totdepth += iter.depth;
2343 if (iter.depth > s->maxdepth)
2344 s->maxdepth = iter.depth;
2345
2346 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2347 ++s->prefixes;
2348 } else {
2349 s->tnodes++;
2350 if (n->bits < MAX_STAT_DEPTH)
2351 s->nodesizes[n->bits]++;
2352 s->nullpointers += tn_info(n)->empty_children;
2353 }
2354 }
2355 rcu_read_unlock();
2356}
2357
2358/*
2359 * This outputs /proc/net/fib_triestats
2360 */
2361static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2362{
2363 unsigned int i, max, pointers, bytes, avdepth;
2364
2365 if (stat->leaves)
2366 avdepth = stat->totdepth*100 / stat->leaves;
2367 else
2368 avdepth = 0;
2369
2370 seq_printf(seq, "\tAver depth: %u.%02d\n",
2371 avdepth / 100, avdepth % 100);
2372 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2373
2374 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2375 bytes = LEAF_SIZE * stat->leaves;
2376
2377 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2378 bytes += sizeof(struct fib_alias) * stat->prefixes;
2379
2380 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2381 bytes += TNODE_SIZE(0) * stat->tnodes;
2382
2383 max = MAX_STAT_DEPTH;
2384 while (max > 0 && stat->nodesizes[max-1] == 0)
2385 max--;
2386
2387 pointers = 0;
2388 for (i = 1; i < max; i++)
2389 if (stat->nodesizes[i] != 0) {
2390 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2391 pointers += (1<<i) * stat->nodesizes[i];
2392 }
2393 seq_putc(seq, '\n');
2394 seq_printf(seq, "\tPointers: %u\n", pointers);
2395
2396 bytes += sizeof(struct key_vector *) * pointers;
2397 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2398 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2399}
2400
2401#ifdef CONFIG_IP_FIB_TRIE_STATS
2402static void trie_show_usage(struct seq_file *seq,
2403 const struct trie_use_stats __percpu *stats)
2404{
2405 struct trie_use_stats s = { 0 };
2406 int cpu;
2407
2408 /* loop through all of the CPUs and gather up the stats */
2409 for_each_possible_cpu(cpu) {
2410 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2411
2412 s.gets += pcpu->gets;
2413 s.backtrack += pcpu->backtrack;
2414 s.semantic_match_passed += pcpu->semantic_match_passed;
2415 s.semantic_match_miss += pcpu->semantic_match_miss;
2416 s.null_node_hit += pcpu->null_node_hit;
2417 s.resize_node_skipped += pcpu->resize_node_skipped;
2418 }
2419
2420 seq_printf(seq, "\nCounters:\n---------\n");
2421 seq_printf(seq, "gets = %u\n", s.gets);
2422 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2423 seq_printf(seq, "semantic match passed = %u\n",
2424 s.semantic_match_passed);
2425 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2426 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2427 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2428}
2429#endif /* CONFIG_IP_FIB_TRIE_STATS */
2430
2431static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2432{
2433 if (tb->tb_id == RT_TABLE_LOCAL)
2434 seq_puts(seq, "Local:\n");
2435 else if (tb->tb_id == RT_TABLE_MAIN)
2436 seq_puts(seq, "Main:\n");
2437 else
2438 seq_printf(seq, "Id %d:\n", tb->tb_id);
2439}
2440
2441
2442static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2443{
2444 struct net *net = (struct net *)seq->private;
2445 unsigned int h;
2446
2447 seq_printf(seq,
2448 "Basic info: size of leaf:"
2449 " %zd bytes, size of tnode: %zd bytes.\n",
2450 LEAF_SIZE, TNODE_SIZE(0));
2451
2452 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2453 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2454 struct fib_table *tb;
2455
2456 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2457 struct trie *t = (struct trie *) tb->tb_data;
2458 struct trie_stat stat;
2459
2460 if (!t)
2461 continue;
2462
2463 fib_table_print(seq, tb);
2464
2465 trie_collect_stats(t, &stat);
2466 trie_show_stats(seq, &stat);
2467#ifdef CONFIG_IP_FIB_TRIE_STATS
2468 trie_show_usage(seq, t->stats);
2469#endif
2470 }
2471 }
2472
2473 return 0;
2474}
2475
2476static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2477{
2478 struct fib_trie_iter *iter = seq->private;
2479 struct net *net = seq_file_net(seq);
2480 loff_t idx = 0;
2481 unsigned int h;
2482
2483 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2484 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2485 struct fib_table *tb;
2486
2487 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2488 struct key_vector *n;
2489
2490 for (n = fib_trie_get_first(iter,
2491 (struct trie *) tb->tb_data);
2492 n; n = fib_trie_get_next(iter))
2493 if (pos == idx++) {
2494 iter->tb = tb;
2495 return n;
2496 }
2497 }
2498 }
2499
2500 return NULL;
2501}
2502
2503static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2504 __acquires(RCU)
2505{
2506 rcu_read_lock();
2507 return fib_trie_get_idx(seq, *pos);
2508}
2509
2510static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2511{
2512 struct fib_trie_iter *iter = seq->private;
2513 struct net *net = seq_file_net(seq);
2514 struct fib_table *tb = iter->tb;
2515 struct hlist_node *tb_node;
2516 unsigned int h;
2517 struct key_vector *n;
2518
2519 ++*pos;
2520 /* next node in same table */
2521 n = fib_trie_get_next(iter);
2522 if (n)
2523 return n;
2524
2525 /* walk rest of this hash chain */
2526 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2527 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2528 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2529 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2530 if (n)
2531 goto found;
2532 }
2533
2534 /* new hash chain */
2535 while (++h < FIB_TABLE_HASHSZ) {
2536 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2537 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2538 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2539 if (n)
2540 goto found;
2541 }
2542 }
2543 return NULL;
2544
2545found:
2546 iter->tb = tb;
2547 return n;
2548}
2549
2550static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2551 __releases(RCU)
2552{
2553 rcu_read_unlock();
2554}
2555
2556static void seq_indent(struct seq_file *seq, int n)
2557{
2558 while (n-- > 0)
2559 seq_puts(seq, " ");
2560}
2561
2562static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2563{
2564 switch (s) {
2565 case RT_SCOPE_UNIVERSE: return "universe";
2566 case RT_SCOPE_SITE: return "site";
2567 case RT_SCOPE_LINK: return "link";
2568 case RT_SCOPE_HOST: return "host";
2569 case RT_SCOPE_NOWHERE: return "nowhere";
2570 default:
2571 snprintf(buf, len, "scope=%d", s);
2572 return buf;
2573 }
2574}
2575
2576static const char *const rtn_type_names[__RTN_MAX] = {
2577 [RTN_UNSPEC] = "UNSPEC",
2578 [RTN_UNICAST] = "UNICAST",
2579 [RTN_LOCAL] = "LOCAL",
2580 [RTN_BROADCAST] = "BROADCAST",
2581 [RTN_ANYCAST] = "ANYCAST",
2582 [RTN_MULTICAST] = "MULTICAST",
2583 [RTN_BLACKHOLE] = "BLACKHOLE",
2584 [RTN_UNREACHABLE] = "UNREACHABLE",
2585 [RTN_PROHIBIT] = "PROHIBIT",
2586 [RTN_THROW] = "THROW",
2587 [RTN_NAT] = "NAT",
2588 [RTN_XRESOLVE] = "XRESOLVE",
2589};
2590
2591static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2592{
2593 if (t < __RTN_MAX && rtn_type_names[t])
2594 return rtn_type_names[t];
2595 snprintf(buf, len, "type %u", t);
2596 return buf;
2597}
2598
2599/* Pretty print the trie */
2600static int fib_trie_seq_show(struct seq_file *seq, void *v)
2601{
2602 const struct fib_trie_iter *iter = seq->private;
2603 struct key_vector *n = v;
2604
2605 if (IS_TRIE(node_parent_rcu(n)))
2606 fib_table_print(seq, iter->tb);
2607
2608 if (IS_TNODE(n)) {
2609 __be32 prf = htonl(n->key);
2610
2611 seq_indent(seq, iter->depth-1);
2612 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2613 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2614 tn_info(n)->full_children,
2615 tn_info(n)->empty_children);
2616 } else {
2617 __be32 val = htonl(n->key);
2618 struct fib_alias *fa;
2619
2620 seq_indent(seq, iter->depth);
2621 seq_printf(seq, " |-- %pI4\n", &val);
2622
2623 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2624 char buf1[32], buf2[32];
2625
2626 seq_indent(seq, iter->depth + 1);
2627 seq_printf(seq, " /%zu %s %s",
2628 KEYLENGTH - fa->fa_slen,
2629 rtn_scope(buf1, sizeof(buf1),
2630 fa->fa_info->fib_scope),
2631 rtn_type(buf2, sizeof(buf2),
2632 fa->fa_type));
2633 if (fa->fa_tos)
2634 seq_printf(seq, " tos=%d", fa->fa_tos);
2635 seq_putc(seq, '\n');
2636 }
2637 }
2638
2639 return 0;
2640}
2641
2642static const struct seq_operations fib_trie_seq_ops = {
2643 .start = fib_trie_seq_start,
2644 .next = fib_trie_seq_next,
2645 .stop = fib_trie_seq_stop,
2646 .show = fib_trie_seq_show,
2647};
2648
2649struct fib_route_iter {
2650 struct seq_net_private p;
2651 struct fib_table *main_tb;
2652 struct key_vector *tnode;
2653 loff_t pos;
2654 t_key key;
2655};
2656
2657static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2658 loff_t pos)
2659{
2660 struct key_vector *l, **tp = &iter->tnode;
2661 t_key key;
2662
2663 /* use cached location of previously found key */
2664 if (iter->pos > 0 && pos >= iter->pos) {
2665 key = iter->key;
2666 } else {
2667 iter->pos = 1;
2668 key = 0;
2669 }
2670
2671 pos -= iter->pos;
2672
2673 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2674 key = l->key + 1;
2675 iter->pos++;
2676 l = NULL;
2677
2678 /* handle unlikely case of a key wrap */
2679 if (!key)
2680 break;
2681 }
2682
2683 if (l)
2684 iter->key = l->key; /* remember it */
2685 else
2686 iter->pos = 0; /* forget it */
2687
2688 return l;
2689}
2690
2691static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2692 __acquires(RCU)
2693{
2694 struct fib_route_iter *iter = seq->private;
2695 struct fib_table *tb;
2696 struct trie *t;
2697
2698 rcu_read_lock();
2699
2700 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2701 if (!tb)
2702 return NULL;
2703
2704 iter->main_tb = tb;
2705 t = (struct trie *)tb->tb_data;
2706 iter->tnode = t->kv;
2707
2708 if (*pos != 0)
2709 return fib_route_get_idx(iter, *pos);
2710
2711 iter->pos = 0;
2712 iter->key = KEY_MAX;
2713
2714 return SEQ_START_TOKEN;
2715}
2716
2717static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2718{
2719 struct fib_route_iter *iter = seq->private;
2720 struct key_vector *l = NULL;
2721 t_key key = iter->key + 1;
2722
2723 ++*pos;
2724
2725 /* only allow key of 0 for start of sequence */
2726 if ((v == SEQ_START_TOKEN) || key)
2727 l = leaf_walk_rcu(&iter->tnode, key);
2728
2729 if (l) {
2730 iter->key = l->key;
2731 iter->pos++;
2732 } else {
2733 iter->pos = 0;
2734 }
2735
2736 return l;
2737}
2738
2739static void fib_route_seq_stop(struct seq_file *seq, void *v)
2740 __releases(RCU)
2741{
2742 rcu_read_unlock();
2743}
2744
2745static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2746{
2747 unsigned int flags = 0;
2748
2749 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2750 flags = RTF_REJECT;
2751 if (fi) {
2752 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2753
2754 if (nhc->nhc_gw.ipv4)
2755 flags |= RTF_GATEWAY;
2756 }
2757 if (mask == htonl(0xFFFFFFFF))
2758 flags |= RTF_HOST;
2759 flags |= RTF_UP;
2760 return flags;
2761}
2762
2763/*
2764 * This outputs /proc/net/route.
2765 * The format of the file is not supposed to be changed
2766 * and needs to be same as fib_hash output to avoid breaking
2767 * legacy utilities
2768 */
2769static int fib_route_seq_show(struct seq_file *seq, void *v)
2770{
2771 struct fib_route_iter *iter = seq->private;
2772 struct fib_table *tb = iter->main_tb;
2773 struct fib_alias *fa;
2774 struct key_vector *l = v;
2775 __be32 prefix;
2776
2777 if (v == SEQ_START_TOKEN) {
2778 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2779 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2780 "\tWindow\tIRTT");
2781 return 0;
2782 }
2783
2784 prefix = htonl(l->key);
2785
2786 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2787 struct fib_info *fi = fa->fa_info;
2788 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2789 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2790
2791 if ((fa->fa_type == RTN_BROADCAST) ||
2792 (fa->fa_type == RTN_MULTICAST))
2793 continue;
2794
2795 if (fa->tb_id != tb->tb_id)
2796 continue;
2797
2798 seq_setwidth(seq, 127);
2799
2800 if (fi) {
2801 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2802 __be32 gw = 0;
2803
2804 if (nhc->nhc_gw_family == AF_INET)
2805 gw = nhc->nhc_gw.ipv4;
2806
2807 seq_printf(seq,
2808 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2809 "%d\t%08X\t%d\t%u\t%u",
2810 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2811 prefix, gw, flags, 0, 0,
2812 fi->fib_priority,
2813 mask,
2814 (fi->fib_advmss ?
2815 fi->fib_advmss + 40 : 0),
2816 fi->fib_window,
2817 fi->fib_rtt >> 3);
2818 } else {
2819 seq_printf(seq,
2820 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2821 "%d\t%08X\t%d\t%u\t%u",
2822 prefix, 0, flags, 0, 0, 0,
2823 mask, 0, 0, 0);
2824 }
2825 seq_pad(seq, '\n');
2826 }
2827
2828 return 0;
2829}
2830
2831static const struct seq_operations fib_route_seq_ops = {
2832 .start = fib_route_seq_start,
2833 .next = fib_route_seq_next,
2834 .stop = fib_route_seq_stop,
2835 .show = fib_route_seq_show,
2836};
2837
2838int __net_init fib_proc_init(struct net *net)
2839{
2840 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2841 sizeof(struct fib_trie_iter)))
2842 goto out1;
2843
2844 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
2845 fib_triestat_seq_show, NULL))
2846 goto out2;
2847
2848 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
2849 sizeof(struct fib_route_iter)))
2850 goto out3;
2851
2852 return 0;
2853
2854out3:
2855 remove_proc_entry("fib_triestat", net->proc_net);
2856out2:
2857 remove_proc_entry("fib_trie", net->proc_net);
2858out1:
2859 return -ENOMEM;
2860}
2861
2862void __net_exit fib_proc_exit(struct net *net)
2863{
2864 remove_proc_entry("fib_trie", net->proc_net);
2865 remove_proc_entry("fib_triestat", net->proc_net);
2866 remove_proc_entry("route", net->proc_net);
2867}
2868
2869#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/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 void __alias_free_mem(struct rcu_head *head)
296{
297 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
298 kmem_cache_free(fn_alias_kmem, fa);
299}
300
301static inline void alias_free_mem_rcu(struct fib_alias *fa)
302{
303 call_rcu(&fa->rcu, __alias_free_mem);
304}
305
306#define TNODE_VMALLOC_MAX \
307 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
308
309static void __node_free_rcu(struct rcu_head *head)
310{
311 struct tnode *n = container_of(head, struct tnode, rcu);
312
313 if (!n->tn_bits)
314 kmem_cache_free(trie_leaf_kmem, n);
315 else
316 kvfree(n);
317}
318
319#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
320
321static struct tnode *tnode_alloc(int bits)
322{
323 size_t size;
324
325 /* verify bits is within bounds */
326 if (bits > TNODE_VMALLOC_MAX)
327 return NULL;
328
329 /* determine size and verify it is non-zero and didn't overflow */
330 size = TNODE_SIZE(1ul << bits);
331
332 if (size <= PAGE_SIZE)
333 return kzalloc(size, GFP_KERNEL);
334 else
335 return vzalloc(size);
336}
337
338static inline void empty_child_inc(struct key_vector *n)
339{
340 tn_info(n)->empty_children++;
341
342 if (!tn_info(n)->empty_children)
343 tn_info(n)->full_children++;
344}
345
346static inline void empty_child_dec(struct key_vector *n)
347{
348 if (!tn_info(n)->empty_children)
349 tn_info(n)->full_children--;
350
351 tn_info(n)->empty_children--;
352}
353
354static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
355{
356 struct key_vector *l;
357 struct tnode *kv;
358
359 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
360 if (!kv)
361 return NULL;
362
363 /* initialize key vector */
364 l = kv->kv;
365 l->key = key;
366 l->pos = 0;
367 l->bits = 0;
368 l->slen = fa->fa_slen;
369
370 /* link leaf to fib alias */
371 INIT_HLIST_HEAD(&l->leaf);
372 hlist_add_head(&fa->fa_list, &l->leaf);
373
374 return l;
375}
376
377static struct key_vector *tnode_new(t_key key, int pos, int bits)
378{
379 unsigned int shift = pos + bits;
380 struct key_vector *tn;
381 struct tnode *tnode;
382
383 /* verify bits and pos their msb bits clear and values are valid */
384 BUG_ON(!bits || (shift > KEYLENGTH));
385
386 tnode = tnode_alloc(bits);
387 if (!tnode)
388 return NULL;
389
390 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
391 sizeof(struct key_vector *) << bits);
392
393 if (bits == KEYLENGTH)
394 tnode->full_children = 1;
395 else
396 tnode->empty_children = 1ul << bits;
397
398 tn = tnode->kv;
399 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
400 tn->pos = pos;
401 tn->bits = bits;
402 tn->slen = pos;
403
404 return tn;
405}
406
407/* Check whether a tnode 'n' is "full", i.e. it is an internal node
408 * and no bits are skipped. See discussion in dyntree paper p. 6
409 */
410static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
411{
412 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
413}
414
415/* Add a child at position i overwriting the old value.
416 * Update the value of full_children and empty_children.
417 */
418static void put_child(struct key_vector *tn, unsigned long i,
419 struct key_vector *n)
420{
421 struct key_vector *chi = get_child(tn, i);
422 int isfull, wasfull;
423
424 BUG_ON(i >= child_length(tn));
425
426 /* update emptyChildren, overflow into fullChildren */
427 if (!n && chi)
428 empty_child_inc(tn);
429 if (n && !chi)
430 empty_child_dec(tn);
431
432 /* update fullChildren */
433 wasfull = tnode_full(tn, chi);
434 isfull = tnode_full(tn, n);
435
436 if (wasfull && !isfull)
437 tn_info(tn)->full_children--;
438 else if (!wasfull && isfull)
439 tn_info(tn)->full_children++;
440
441 if (n && (tn->slen < n->slen))
442 tn->slen = n->slen;
443
444 rcu_assign_pointer(tn->tnode[i], n);
445}
446
447static void update_children(struct key_vector *tn)
448{
449 unsigned long i;
450
451 /* update all of the child parent pointers */
452 for (i = child_length(tn); i;) {
453 struct key_vector *inode = get_child(tn, --i);
454
455 if (!inode)
456 continue;
457
458 /* Either update the children of a tnode that
459 * already belongs to us or update the child
460 * to point to ourselves.
461 */
462 if (node_parent(inode) == tn)
463 update_children(inode);
464 else
465 node_set_parent(inode, tn);
466 }
467}
468
469static inline void put_child_root(struct key_vector *tp, t_key key,
470 struct key_vector *n)
471{
472 if (IS_TRIE(tp))
473 rcu_assign_pointer(tp->tnode[0], n);
474 else
475 put_child(tp, get_index(key, tp), n);
476}
477
478static inline void tnode_free_init(struct key_vector *tn)
479{
480 tn_info(tn)->rcu.next = NULL;
481}
482
483static inline void tnode_free_append(struct key_vector *tn,
484 struct key_vector *n)
485{
486 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
487 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
488}
489
490static void tnode_free(struct key_vector *tn)
491{
492 struct callback_head *head = &tn_info(tn)->rcu;
493
494 while (head) {
495 head = head->next;
496 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
497 node_free(tn);
498
499 tn = container_of(head, struct tnode, rcu)->kv;
500 }
501
502 if (tnode_free_size >= READ_ONCE(sysctl_fib_sync_mem)) {
503 tnode_free_size = 0;
504 synchronize_rcu();
505 }
506}
507
508static struct key_vector *replace(struct trie *t,
509 struct key_vector *oldtnode,
510 struct key_vector *tn)
511{
512 struct key_vector *tp = node_parent(oldtnode);
513 unsigned long i;
514
515 /* setup the parent pointer out of and back into this node */
516 NODE_INIT_PARENT(tn, tp);
517 put_child_root(tp, tn->key, tn);
518
519 /* update all of the child parent pointers */
520 update_children(tn);
521
522 /* all pointers should be clean so we are done */
523 tnode_free(oldtnode);
524
525 /* resize children now that oldtnode is freed */
526 for (i = child_length(tn); i;) {
527 struct key_vector *inode = get_child(tn, --i);
528
529 /* resize child node */
530 if (tnode_full(tn, inode))
531 tn = resize(t, inode);
532 }
533
534 return tp;
535}
536
537static struct key_vector *inflate(struct trie *t,
538 struct key_vector *oldtnode)
539{
540 struct key_vector *tn;
541 unsigned long i;
542 t_key m;
543
544 pr_debug("In inflate\n");
545
546 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
547 if (!tn)
548 goto notnode;
549
550 /* prepare oldtnode to be freed */
551 tnode_free_init(oldtnode);
552
553 /* Assemble all of the pointers in our cluster, in this case that
554 * represents all of the pointers out of our allocated nodes that
555 * point to existing tnodes and the links between our allocated
556 * nodes.
557 */
558 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
559 struct key_vector *inode = get_child(oldtnode, --i);
560 struct key_vector *node0, *node1;
561 unsigned long j, k;
562
563 /* An empty child */
564 if (!inode)
565 continue;
566
567 /* A leaf or an internal node with skipped bits */
568 if (!tnode_full(oldtnode, inode)) {
569 put_child(tn, get_index(inode->key, tn), inode);
570 continue;
571 }
572
573 /* drop the node in the old tnode free list */
574 tnode_free_append(oldtnode, inode);
575
576 /* An internal node with two children */
577 if (inode->bits == 1) {
578 put_child(tn, 2 * i + 1, get_child(inode, 1));
579 put_child(tn, 2 * i, get_child(inode, 0));
580 continue;
581 }
582
583 /* We will replace this node 'inode' with two new
584 * ones, 'node0' and 'node1', each with half of the
585 * original children. The two new nodes will have
586 * a position one bit further down the key and this
587 * means that the "significant" part of their keys
588 * (see the discussion near the top of this file)
589 * will differ by one bit, which will be "0" in
590 * node0's key and "1" in node1's key. Since we are
591 * moving the key position by one step, the bit that
592 * we are moving away from - the bit at position
593 * (tn->pos) - is the one that will differ between
594 * node0 and node1. So... we synthesize that bit in the
595 * two new keys.
596 */
597 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
598 if (!node1)
599 goto nomem;
600 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
601
602 tnode_free_append(tn, node1);
603 if (!node0)
604 goto nomem;
605 tnode_free_append(tn, node0);
606
607 /* populate child pointers in new nodes */
608 for (k = child_length(inode), j = k / 2; j;) {
609 put_child(node1, --j, get_child(inode, --k));
610 put_child(node0, j, get_child(inode, j));
611 put_child(node1, --j, get_child(inode, --k));
612 put_child(node0, j, get_child(inode, j));
613 }
614
615 /* link new nodes to parent */
616 NODE_INIT_PARENT(node1, tn);
617 NODE_INIT_PARENT(node0, tn);
618
619 /* link parent to nodes */
620 put_child(tn, 2 * i + 1, node1);
621 put_child(tn, 2 * i, node0);
622 }
623
624 /* setup the parent pointers into and out of this node */
625 return replace(t, oldtnode, tn);
626nomem:
627 /* all pointers should be clean so we are done */
628 tnode_free(tn);
629notnode:
630 return NULL;
631}
632
633static struct key_vector *halve(struct trie *t,
634 struct key_vector *oldtnode)
635{
636 struct key_vector *tn;
637 unsigned long i;
638
639 pr_debug("In halve\n");
640
641 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
642 if (!tn)
643 goto notnode;
644
645 /* prepare oldtnode to be freed */
646 tnode_free_init(oldtnode);
647
648 /* Assemble all of the pointers in our cluster, in this case that
649 * represents all of the pointers out of our allocated nodes that
650 * point to existing tnodes and the links between our allocated
651 * nodes.
652 */
653 for (i = child_length(oldtnode); i;) {
654 struct key_vector *node1 = get_child(oldtnode, --i);
655 struct key_vector *node0 = get_child(oldtnode, --i);
656 struct key_vector *inode;
657
658 /* At least one of the children is empty */
659 if (!node1 || !node0) {
660 put_child(tn, i / 2, node1 ? : node0);
661 continue;
662 }
663
664 /* Two nonempty children */
665 inode = tnode_new(node0->key, oldtnode->pos, 1);
666 if (!inode)
667 goto nomem;
668 tnode_free_append(tn, inode);
669
670 /* initialize pointers out of node */
671 put_child(inode, 1, node1);
672 put_child(inode, 0, node0);
673 NODE_INIT_PARENT(inode, tn);
674
675 /* link parent to node */
676 put_child(tn, i / 2, inode);
677 }
678
679 /* setup the parent pointers into and out of this node */
680 return replace(t, oldtnode, tn);
681nomem:
682 /* all pointers should be clean so we are done */
683 tnode_free(tn);
684notnode:
685 return NULL;
686}
687
688static struct key_vector *collapse(struct trie *t,
689 struct key_vector *oldtnode)
690{
691 struct key_vector *n, *tp;
692 unsigned long i;
693
694 /* scan the tnode looking for that one child that might still exist */
695 for (n = NULL, i = child_length(oldtnode); !n && i;)
696 n = get_child(oldtnode, --i);
697
698 /* compress one level */
699 tp = node_parent(oldtnode);
700 put_child_root(tp, oldtnode->key, n);
701 node_set_parent(n, tp);
702
703 /* drop dead node */
704 node_free(oldtnode);
705
706 return tp;
707}
708
709static unsigned char update_suffix(struct key_vector *tn)
710{
711 unsigned char slen = tn->pos;
712 unsigned long stride, i;
713 unsigned char slen_max;
714
715 /* only vector 0 can have a suffix length greater than or equal to
716 * tn->pos + tn->bits, the second highest node will have a suffix
717 * length at most of tn->pos + tn->bits - 1
718 */
719 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
720
721 /* search though the list of children looking for nodes that might
722 * have a suffix greater than the one we currently have. This is
723 * why we start with a stride of 2 since a stride of 1 would
724 * represent the nodes with suffix length equal to tn->pos
725 */
726 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
727 struct key_vector *n = get_child(tn, i);
728
729 if (!n || (n->slen <= slen))
730 continue;
731
732 /* update stride and slen based on new value */
733 stride <<= (n->slen - slen);
734 slen = n->slen;
735 i &= ~(stride - 1);
736
737 /* stop searching if we have hit the maximum possible value */
738 if (slen >= slen_max)
739 break;
740 }
741
742 tn->slen = slen;
743
744 return slen;
745}
746
747/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
748 * the Helsinki University of Technology and Matti Tikkanen of Nokia
749 * Telecommunications, page 6:
750 * "A node is doubled if the ratio of non-empty children to all
751 * children in the *doubled* node is at least 'high'."
752 *
753 * 'high' in this instance is the variable 'inflate_threshold'. It
754 * is expressed as a percentage, so we multiply it with
755 * child_length() and instead of multiplying by 2 (since the
756 * child array will be doubled by inflate()) and multiplying
757 * the left-hand side by 100 (to handle the percentage thing) we
758 * multiply the left-hand side by 50.
759 *
760 * The left-hand side may look a bit weird: child_length(tn)
761 * - tn->empty_children is of course the number of non-null children
762 * in the current node. tn->full_children is the number of "full"
763 * children, that is non-null tnodes with a skip value of 0.
764 * All of those will be doubled in the resulting inflated tnode, so
765 * we just count them one extra time here.
766 *
767 * A clearer way to write this would be:
768 *
769 * to_be_doubled = tn->full_children;
770 * not_to_be_doubled = child_length(tn) - tn->empty_children -
771 * tn->full_children;
772 *
773 * new_child_length = child_length(tn) * 2;
774 *
775 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
776 * new_child_length;
777 * if (new_fill_factor >= inflate_threshold)
778 *
779 * ...and so on, tho it would mess up the while () loop.
780 *
781 * anyway,
782 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
783 * inflate_threshold
784 *
785 * avoid a division:
786 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
787 * inflate_threshold * new_child_length
788 *
789 * expand not_to_be_doubled and to_be_doubled, and shorten:
790 * 100 * (child_length(tn) - tn->empty_children +
791 * tn->full_children) >= inflate_threshold * new_child_length
792 *
793 * expand new_child_length:
794 * 100 * (child_length(tn) - tn->empty_children +
795 * tn->full_children) >=
796 * inflate_threshold * child_length(tn) * 2
797 *
798 * shorten again:
799 * 50 * (tn->full_children + child_length(tn) -
800 * tn->empty_children) >= inflate_threshold *
801 * child_length(tn)
802 *
803 */
804static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
805{
806 unsigned long used = child_length(tn);
807 unsigned long threshold = used;
808
809 /* Keep root node larger */
810 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
811 used -= tn_info(tn)->empty_children;
812 used += tn_info(tn)->full_children;
813
814 /* if bits == KEYLENGTH then pos = 0, and will fail below */
815
816 return (used > 1) && tn->pos && ((50 * used) >= threshold);
817}
818
819static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
820{
821 unsigned long used = child_length(tn);
822 unsigned long threshold = used;
823
824 /* Keep root node larger */
825 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
826 used -= tn_info(tn)->empty_children;
827
828 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
829
830 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
831}
832
833static inline bool should_collapse(struct key_vector *tn)
834{
835 unsigned long used = child_length(tn);
836
837 used -= tn_info(tn)->empty_children;
838
839 /* account for bits == KEYLENGTH case */
840 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
841 used -= KEY_MAX;
842
843 /* One child or none, time to drop us from the trie */
844 return used < 2;
845}
846
847#define MAX_WORK 10
848static struct key_vector *resize(struct trie *t, struct key_vector *tn)
849{
850#ifdef CONFIG_IP_FIB_TRIE_STATS
851 struct trie_use_stats __percpu *stats = t->stats;
852#endif
853 struct key_vector *tp = node_parent(tn);
854 unsigned long cindex = get_index(tn->key, tp);
855 int max_work = MAX_WORK;
856
857 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
858 tn, inflate_threshold, halve_threshold);
859
860 /* track the tnode via the pointer from the parent instead of
861 * doing it ourselves. This way we can let RCU fully do its
862 * thing without us interfering
863 */
864 BUG_ON(tn != get_child(tp, cindex));
865
866 /* Double as long as the resulting node has a number of
867 * nonempty nodes that are above the threshold.
868 */
869 while (should_inflate(tp, tn) && max_work) {
870 tp = inflate(t, tn);
871 if (!tp) {
872#ifdef CONFIG_IP_FIB_TRIE_STATS
873 this_cpu_inc(stats->resize_node_skipped);
874#endif
875 break;
876 }
877
878 max_work--;
879 tn = get_child(tp, cindex);
880 }
881
882 /* update parent in case inflate failed */
883 tp = node_parent(tn);
884
885 /* Return if at least one inflate is run */
886 if (max_work != MAX_WORK)
887 return tp;
888
889 /* Halve as long as the number of empty children in this
890 * node is above threshold.
891 */
892 while (should_halve(tp, tn) && max_work) {
893 tp = halve(t, tn);
894 if (!tp) {
895#ifdef CONFIG_IP_FIB_TRIE_STATS
896 this_cpu_inc(stats->resize_node_skipped);
897#endif
898 break;
899 }
900
901 max_work--;
902 tn = get_child(tp, cindex);
903 }
904
905 /* Only one child remains */
906 if (should_collapse(tn))
907 return collapse(t, tn);
908
909 /* update parent in case halve failed */
910 return node_parent(tn);
911}
912
913static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
914{
915 unsigned char node_slen = tn->slen;
916
917 while ((node_slen > tn->pos) && (node_slen > slen)) {
918 slen = update_suffix(tn);
919 if (node_slen == slen)
920 break;
921
922 tn = node_parent(tn);
923 node_slen = tn->slen;
924 }
925}
926
927static void node_push_suffix(struct key_vector *tn, unsigned char slen)
928{
929 while (tn->slen < slen) {
930 tn->slen = slen;
931 tn = node_parent(tn);
932 }
933}
934
935/* rcu_read_lock needs to be hold by caller from readside */
936static struct key_vector *fib_find_node(struct trie *t,
937 struct key_vector **tp, u32 key)
938{
939 struct key_vector *pn, *n = t->kv;
940 unsigned long index = 0;
941
942 do {
943 pn = n;
944 n = get_child_rcu(n, index);
945
946 if (!n)
947 break;
948
949 index = get_cindex(key, n);
950
951 /* This bit of code is a bit tricky but it combines multiple
952 * checks into a single check. The prefix consists of the
953 * prefix plus zeros for the bits in the cindex. The index
954 * is the difference between the key and this value. From
955 * this we can actually derive several pieces of data.
956 * if (index >= (1ul << bits))
957 * we have a mismatch in skip bits and failed
958 * else
959 * we know the value is cindex
960 *
961 * This check is safe even if bits == KEYLENGTH due to the
962 * fact that we can only allocate a node with 32 bits if a
963 * long is greater than 32 bits.
964 */
965 if (index >= (1ul << n->bits)) {
966 n = NULL;
967 break;
968 }
969
970 /* keep searching until we find a perfect match leaf or NULL */
971 } while (IS_TNODE(n));
972
973 *tp = pn;
974
975 return n;
976}
977
978/* Return the first fib alias matching DSCP with
979 * priority less than or equal to PRIO.
980 * If 'find_first' is set, return the first matching
981 * fib alias, regardless of DSCP and priority.
982 */
983static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
984 dscp_t dscp, u32 prio, u32 tb_id,
985 bool find_first)
986{
987 struct fib_alias *fa;
988
989 if (!fah)
990 return NULL;
991
992 hlist_for_each_entry(fa, fah, fa_list) {
993 /* Avoid Sparse warning when using dscp_t in inequalities */
994 u8 __fa_dscp = inet_dscp_to_dsfield(fa->fa_dscp);
995 u8 __dscp = inet_dscp_to_dsfield(dscp);
996
997 if (fa->fa_slen < slen)
998 continue;
999 if (fa->fa_slen != slen)
1000 break;
1001 if (fa->tb_id > tb_id)
1002 continue;
1003 if (fa->tb_id != tb_id)
1004 break;
1005 if (find_first)
1006 return fa;
1007 if (__fa_dscp > __dscp)
1008 continue;
1009 if (fa->fa_info->fib_priority >= prio || __fa_dscp < __dscp)
1010 return fa;
1011 }
1012
1013 return NULL;
1014}
1015
1016static struct fib_alias *
1017fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1018{
1019 u8 slen = KEYLENGTH - fri->dst_len;
1020 struct key_vector *l, *tp;
1021 struct fib_table *tb;
1022 struct fib_alias *fa;
1023 struct trie *t;
1024
1025 tb = fib_get_table(net, fri->tb_id);
1026 if (!tb)
1027 return NULL;
1028
1029 t = (struct trie *)tb->tb_data;
1030 l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1031 if (!l)
1032 return NULL;
1033
1034 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1035 if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1036 fa->fa_dscp == fri->dscp && fa->fa_info == fri->fi &&
1037 fa->fa_type == fri->type)
1038 return fa;
1039 }
1040
1041 return NULL;
1042}
1043
1044void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1045{
1046 u8 fib_notify_on_flag_change;
1047 struct fib_alias *fa_match;
1048 struct sk_buff *skb;
1049 int err;
1050
1051 rcu_read_lock();
1052
1053 fa_match = fib_find_matching_alias(net, fri);
1054 if (!fa_match)
1055 goto out;
1056
1057 /* These are paired with the WRITE_ONCE() happening in this function.
1058 * The reason is that we are only protected by RCU at this point.
1059 */
1060 if (READ_ONCE(fa_match->offload) == fri->offload &&
1061 READ_ONCE(fa_match->trap) == fri->trap &&
1062 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1063 goto out;
1064
1065 WRITE_ONCE(fa_match->offload, fri->offload);
1066 WRITE_ONCE(fa_match->trap, fri->trap);
1067
1068 fib_notify_on_flag_change = READ_ONCE(net->ipv4.sysctl_fib_notify_on_flag_change);
1069
1070 /* 2 means send notifications only if offload_failed was changed. */
1071 if (fib_notify_on_flag_change == 2 &&
1072 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1073 goto out;
1074
1075 WRITE_ONCE(fa_match->offload_failed, fri->offload_failed);
1076
1077 if (!fib_notify_on_flag_change)
1078 goto out;
1079
1080 skb = nlmsg_new(fib_nlmsg_size(fa_match->fa_info), GFP_ATOMIC);
1081 if (!skb) {
1082 err = -ENOBUFS;
1083 goto errout;
1084 }
1085
1086 err = fib_dump_info(skb, 0, 0, RTM_NEWROUTE, fri, 0);
1087 if (err < 0) {
1088 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
1089 WARN_ON(err == -EMSGSIZE);
1090 kfree_skb(skb);
1091 goto errout;
1092 }
1093
1094 rtnl_notify(skb, net, 0, RTNLGRP_IPV4_ROUTE, NULL, GFP_ATOMIC);
1095 goto out;
1096
1097errout:
1098 rtnl_set_sk_err(net, RTNLGRP_IPV4_ROUTE, err);
1099out:
1100 rcu_read_unlock();
1101}
1102EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1103
1104static void trie_rebalance(struct trie *t, struct key_vector *tn)
1105{
1106 while (!IS_TRIE(tn))
1107 tn = resize(t, tn);
1108}
1109
1110static int fib_insert_node(struct trie *t, struct key_vector *tp,
1111 struct fib_alias *new, t_key key)
1112{
1113 struct key_vector *n, *l;
1114
1115 l = leaf_new(key, new);
1116 if (!l)
1117 goto noleaf;
1118
1119 /* retrieve child from parent node */
1120 n = get_child(tp, get_index(key, tp));
1121
1122 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1123 *
1124 * Add a new tnode here
1125 * first tnode need some special handling
1126 * leaves us in position for handling as case 3
1127 */
1128 if (n) {
1129 struct key_vector *tn;
1130
1131 tn = tnode_new(key, __fls(key ^ n->key), 1);
1132 if (!tn)
1133 goto notnode;
1134
1135 /* initialize routes out of node */
1136 NODE_INIT_PARENT(tn, tp);
1137 put_child(tn, get_index(key, tn) ^ 1, n);
1138
1139 /* start adding routes into the node */
1140 put_child_root(tp, key, tn);
1141 node_set_parent(n, tn);
1142
1143 /* parent now has a NULL spot where the leaf can go */
1144 tp = tn;
1145 }
1146
1147 /* Case 3: n is NULL, and will just insert a new leaf */
1148 node_push_suffix(tp, new->fa_slen);
1149 NODE_INIT_PARENT(l, tp);
1150 put_child_root(tp, key, l);
1151 trie_rebalance(t, tp);
1152
1153 return 0;
1154notnode:
1155 node_free(l);
1156noleaf:
1157 return -ENOMEM;
1158}
1159
1160static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1161 struct key_vector *l, struct fib_alias *new,
1162 struct fib_alias *fa, t_key key)
1163{
1164 if (!l)
1165 return fib_insert_node(t, tp, new, key);
1166
1167 if (fa) {
1168 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1169 } else {
1170 struct fib_alias *last;
1171
1172 hlist_for_each_entry(last, &l->leaf, fa_list) {
1173 if (new->fa_slen < last->fa_slen)
1174 break;
1175 if ((new->fa_slen == last->fa_slen) &&
1176 (new->tb_id > last->tb_id))
1177 break;
1178 fa = last;
1179 }
1180
1181 if (fa)
1182 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1183 else
1184 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1185 }
1186
1187 /* if we added to the tail node then we need to update slen */
1188 if (l->slen < new->fa_slen) {
1189 l->slen = new->fa_slen;
1190 node_push_suffix(tp, new->fa_slen);
1191 }
1192
1193 return 0;
1194}
1195
1196static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1197{
1198 if (plen > KEYLENGTH) {
1199 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1200 return false;
1201 }
1202
1203 if ((plen < KEYLENGTH) && (key << plen)) {
1204 NL_SET_ERR_MSG(extack,
1205 "Invalid prefix for given prefix length");
1206 return false;
1207 }
1208
1209 return true;
1210}
1211
1212static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1213 struct key_vector *l, struct fib_alias *old);
1214
1215/* Caller must hold RTNL. */
1216int fib_table_insert(struct net *net, struct fib_table *tb,
1217 struct fib_config *cfg, struct netlink_ext_ack *extack)
1218{
1219 struct trie *t = (struct trie *)tb->tb_data;
1220 struct fib_alias *fa, *new_fa;
1221 struct key_vector *l, *tp;
1222 u16 nlflags = NLM_F_EXCL;
1223 struct fib_info *fi;
1224 u8 plen = cfg->fc_dst_len;
1225 u8 slen = KEYLENGTH - plen;
1226 dscp_t dscp;
1227 u32 key;
1228 int err;
1229
1230 key = ntohl(cfg->fc_dst);
1231
1232 if (!fib_valid_key_len(key, plen, extack))
1233 return -EINVAL;
1234
1235 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1236
1237 fi = fib_create_info(cfg, extack);
1238 if (IS_ERR(fi)) {
1239 err = PTR_ERR(fi);
1240 goto err;
1241 }
1242
1243 dscp = cfg->fc_dscp;
1244 l = fib_find_node(t, &tp, key);
1245 fa = l ? fib_find_alias(&l->leaf, slen, dscp, fi->fib_priority,
1246 tb->tb_id, false) : NULL;
1247
1248 /* Now fa, if non-NULL, points to the first fib alias
1249 * with the same keys [prefix,dscp,priority], if such key already
1250 * exists or to the node before which we will insert new one.
1251 *
1252 * If fa is NULL, we will need to allocate a new one and
1253 * insert to the tail of the section matching the suffix length
1254 * of the new alias.
1255 */
1256
1257 if (fa && fa->fa_dscp == dscp &&
1258 fa->fa_info->fib_priority == fi->fib_priority) {
1259 struct fib_alias *fa_first, *fa_match;
1260
1261 err = -EEXIST;
1262 if (cfg->fc_nlflags & NLM_F_EXCL)
1263 goto out;
1264
1265 nlflags &= ~NLM_F_EXCL;
1266
1267 /* We have 2 goals:
1268 * 1. Find exact match for type, scope, fib_info to avoid
1269 * duplicate routes
1270 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1271 */
1272 fa_match = NULL;
1273 fa_first = fa;
1274 hlist_for_each_entry_from(fa, fa_list) {
1275 if ((fa->fa_slen != slen) ||
1276 (fa->tb_id != tb->tb_id) ||
1277 (fa->fa_dscp != dscp))
1278 break;
1279 if (fa->fa_info->fib_priority != fi->fib_priority)
1280 break;
1281 if (fa->fa_type == cfg->fc_type &&
1282 fa->fa_info == fi) {
1283 fa_match = fa;
1284 break;
1285 }
1286 }
1287
1288 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1289 struct fib_info *fi_drop;
1290 u8 state;
1291
1292 nlflags |= NLM_F_REPLACE;
1293 fa = fa_first;
1294 if (fa_match) {
1295 if (fa == fa_match)
1296 err = 0;
1297 goto out;
1298 }
1299 err = -ENOBUFS;
1300 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1301 if (!new_fa)
1302 goto out;
1303
1304 fi_drop = fa->fa_info;
1305 new_fa->fa_dscp = fa->fa_dscp;
1306 new_fa->fa_info = fi;
1307 new_fa->fa_type = cfg->fc_type;
1308 state = fa->fa_state;
1309 new_fa->fa_state = state & ~FA_S_ACCESSED;
1310 new_fa->fa_slen = fa->fa_slen;
1311 new_fa->tb_id = tb->tb_id;
1312 new_fa->fa_default = -1;
1313 new_fa->offload = 0;
1314 new_fa->trap = 0;
1315 new_fa->offload_failed = 0;
1316
1317 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1318
1319 if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1320 tb->tb_id, true) == new_fa) {
1321 enum fib_event_type fib_event;
1322
1323 fib_event = FIB_EVENT_ENTRY_REPLACE;
1324 err = call_fib_entry_notifiers(net, fib_event,
1325 key, plen,
1326 new_fa, extack);
1327 if (err) {
1328 hlist_replace_rcu(&new_fa->fa_list,
1329 &fa->fa_list);
1330 goto out_free_new_fa;
1331 }
1332 }
1333
1334 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1335 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1336
1337 alias_free_mem_rcu(fa);
1338
1339 fib_release_info(fi_drop);
1340 if (state & FA_S_ACCESSED)
1341 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1342
1343 goto succeeded;
1344 }
1345 /* Error if we find a perfect match which
1346 * uses the same scope, type, and nexthop
1347 * information.
1348 */
1349 if (fa_match)
1350 goto out;
1351
1352 if (cfg->fc_nlflags & NLM_F_APPEND)
1353 nlflags |= NLM_F_APPEND;
1354 else
1355 fa = fa_first;
1356 }
1357 err = -ENOENT;
1358 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1359 goto out;
1360
1361 nlflags |= NLM_F_CREATE;
1362 err = -ENOBUFS;
1363 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1364 if (!new_fa)
1365 goto out;
1366
1367 new_fa->fa_info = fi;
1368 new_fa->fa_dscp = dscp;
1369 new_fa->fa_type = cfg->fc_type;
1370 new_fa->fa_state = 0;
1371 new_fa->fa_slen = slen;
1372 new_fa->tb_id = tb->tb_id;
1373 new_fa->fa_default = -1;
1374 new_fa->offload = 0;
1375 new_fa->trap = 0;
1376 new_fa->offload_failed = 0;
1377
1378 /* Insert new entry to the list. */
1379 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1380 if (err)
1381 goto out_free_new_fa;
1382
1383 /* The alias was already inserted, so the node must exist. */
1384 l = l ? l : fib_find_node(t, &tp, key);
1385 if (WARN_ON_ONCE(!l)) {
1386 err = -ENOENT;
1387 goto out_free_new_fa;
1388 }
1389
1390 if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1391 new_fa) {
1392 enum fib_event_type fib_event;
1393
1394 fib_event = FIB_EVENT_ENTRY_REPLACE;
1395 err = call_fib_entry_notifiers(net, fib_event, key, plen,
1396 new_fa, extack);
1397 if (err)
1398 goto out_remove_new_fa;
1399 }
1400
1401 if (!plen)
1402 tb->tb_num_default++;
1403
1404 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1405 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1406 &cfg->fc_nlinfo, nlflags);
1407succeeded:
1408 return 0;
1409
1410out_remove_new_fa:
1411 fib_remove_alias(t, tp, l, new_fa);
1412out_free_new_fa:
1413 kmem_cache_free(fn_alias_kmem, new_fa);
1414out:
1415 fib_release_info(fi);
1416err:
1417 return err;
1418}
1419
1420static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1421{
1422 t_key prefix = n->key;
1423
1424 return (key ^ prefix) & (prefix | -prefix);
1425}
1426
1427bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1428 const struct flowi4 *flp)
1429{
1430 if (nhc->nhc_flags & RTNH_F_DEAD)
1431 return false;
1432
1433 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1434 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1435 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1436 return false;
1437
1438 if (flp->flowi4_oif && flp->flowi4_oif != nhc->nhc_oif)
1439 return false;
1440
1441 return true;
1442}
1443
1444/* should be called with rcu_read_lock */
1445int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1446 struct fib_result *res, int fib_flags)
1447{
1448 struct trie *t = (struct trie *) tb->tb_data;
1449#ifdef CONFIG_IP_FIB_TRIE_STATS
1450 struct trie_use_stats __percpu *stats = t->stats;
1451#endif
1452 const t_key key = ntohl(flp->daddr);
1453 struct key_vector *n, *pn;
1454 struct fib_alias *fa;
1455 unsigned long index;
1456 t_key cindex;
1457
1458 pn = t->kv;
1459 cindex = 0;
1460
1461 n = get_child_rcu(pn, cindex);
1462 if (!n) {
1463 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1464 return -EAGAIN;
1465 }
1466
1467#ifdef CONFIG_IP_FIB_TRIE_STATS
1468 this_cpu_inc(stats->gets);
1469#endif
1470
1471 /* Step 1: Travel to the longest prefix match in the trie */
1472 for (;;) {
1473 index = get_cindex(key, n);
1474
1475 /* This bit of code is a bit tricky but it combines multiple
1476 * checks into a single check. The prefix consists of the
1477 * prefix plus zeros for the "bits" in the prefix. The index
1478 * is the difference between the key and this value. From
1479 * this we can actually derive several pieces of data.
1480 * if (index >= (1ul << bits))
1481 * we have a mismatch in skip bits and failed
1482 * else
1483 * we know the value is cindex
1484 *
1485 * This check is safe even if bits == KEYLENGTH due to the
1486 * fact that we can only allocate a node with 32 bits if a
1487 * long is greater than 32 bits.
1488 */
1489 if (index >= (1ul << n->bits))
1490 break;
1491
1492 /* we have found a leaf. Prefixes have already been compared */
1493 if (IS_LEAF(n))
1494 goto found;
1495
1496 /* only record pn and cindex if we are going to be chopping
1497 * bits later. Otherwise we are just wasting cycles.
1498 */
1499 if (n->slen > n->pos) {
1500 pn = n;
1501 cindex = index;
1502 }
1503
1504 n = get_child_rcu(n, index);
1505 if (unlikely(!n))
1506 goto backtrace;
1507 }
1508
1509 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1510 for (;;) {
1511 /* record the pointer where our next node pointer is stored */
1512 struct key_vector __rcu **cptr = n->tnode;
1513
1514 /* This test verifies that none of the bits that differ
1515 * between the key and the prefix exist in the region of
1516 * the lsb and higher in the prefix.
1517 */
1518 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1519 goto backtrace;
1520
1521 /* exit out and process leaf */
1522 if (unlikely(IS_LEAF(n)))
1523 break;
1524
1525 /* Don't bother recording parent info. Since we are in
1526 * prefix match mode we will have to come back to wherever
1527 * we started this traversal anyway
1528 */
1529
1530 while ((n = rcu_dereference(*cptr)) == NULL) {
1531backtrace:
1532#ifdef CONFIG_IP_FIB_TRIE_STATS
1533 if (!n)
1534 this_cpu_inc(stats->null_node_hit);
1535#endif
1536 /* If we are at cindex 0 there are no more bits for
1537 * us to strip at this level so we must ascend back
1538 * up one level to see if there are any more bits to
1539 * be stripped there.
1540 */
1541 while (!cindex) {
1542 t_key pkey = pn->key;
1543
1544 /* If we don't have a parent then there is
1545 * nothing for us to do as we do not have any
1546 * further nodes to parse.
1547 */
1548 if (IS_TRIE(pn)) {
1549 trace_fib_table_lookup(tb->tb_id, flp,
1550 NULL, -EAGAIN);
1551 return -EAGAIN;
1552 }
1553#ifdef CONFIG_IP_FIB_TRIE_STATS
1554 this_cpu_inc(stats->backtrack);
1555#endif
1556 /* Get Child's index */
1557 pn = node_parent_rcu(pn);
1558 cindex = get_index(pkey, pn);
1559 }
1560
1561 /* strip the least significant bit from the cindex */
1562 cindex &= cindex - 1;
1563
1564 /* grab pointer for next child node */
1565 cptr = &pn->tnode[cindex];
1566 }
1567 }
1568
1569found:
1570 /* this line carries forward the xor from earlier in the function */
1571 index = key ^ n->key;
1572
1573 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1574 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1575 struct fib_info *fi = fa->fa_info;
1576 struct fib_nh_common *nhc;
1577 int nhsel, err;
1578
1579 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1580 if (index >= (1ul << fa->fa_slen))
1581 continue;
1582 }
1583 if (fa->fa_dscp &&
1584 inet_dscp_to_dsfield(fa->fa_dscp) != flp->flowi4_tos)
1585 continue;
1586 /* Paired with WRITE_ONCE() in fib_release_info() */
1587 if (READ_ONCE(fi->fib_dead))
1588 continue;
1589 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1590 continue;
1591 fib_alias_accessed(fa);
1592 err = fib_props[fa->fa_type].error;
1593 if (unlikely(err < 0)) {
1594out_reject:
1595#ifdef CONFIG_IP_FIB_TRIE_STATS
1596 this_cpu_inc(stats->semantic_match_passed);
1597#endif
1598 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1599 return err;
1600 }
1601 if (fi->fib_flags & RTNH_F_DEAD)
1602 continue;
1603
1604 if (unlikely(fi->nh)) {
1605 if (nexthop_is_blackhole(fi->nh)) {
1606 err = fib_props[RTN_BLACKHOLE].error;
1607 goto out_reject;
1608 }
1609
1610 nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1611 &nhsel);
1612 if (nhc)
1613 goto set_result;
1614 goto miss;
1615 }
1616
1617 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1618 nhc = fib_info_nhc(fi, nhsel);
1619
1620 if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1621 continue;
1622set_result:
1623 if (!(fib_flags & FIB_LOOKUP_NOREF))
1624 refcount_inc(&fi->fib_clntref);
1625
1626 res->prefix = htonl(n->key);
1627 res->prefixlen = KEYLENGTH - fa->fa_slen;
1628 res->nh_sel = nhsel;
1629 res->nhc = nhc;
1630 res->type = fa->fa_type;
1631 res->scope = fi->fib_scope;
1632 res->fi = fi;
1633 res->table = tb;
1634 res->fa_head = &n->leaf;
1635#ifdef CONFIG_IP_FIB_TRIE_STATS
1636 this_cpu_inc(stats->semantic_match_passed);
1637#endif
1638 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1639
1640 return err;
1641 }
1642 }
1643miss:
1644#ifdef CONFIG_IP_FIB_TRIE_STATS
1645 this_cpu_inc(stats->semantic_match_miss);
1646#endif
1647 goto backtrace;
1648}
1649EXPORT_SYMBOL_GPL(fib_table_lookup);
1650
1651static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1652 struct key_vector *l, struct fib_alias *old)
1653{
1654 /* record the location of the previous list_info entry */
1655 struct hlist_node **pprev = old->fa_list.pprev;
1656 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1657
1658 /* remove the fib_alias from the list */
1659 hlist_del_rcu(&old->fa_list);
1660
1661 /* if we emptied the list this leaf will be freed and we can sort
1662 * out parent suffix lengths as a part of trie_rebalance
1663 */
1664 if (hlist_empty(&l->leaf)) {
1665 if (tp->slen == l->slen)
1666 node_pull_suffix(tp, tp->pos);
1667 put_child_root(tp, l->key, NULL);
1668 node_free(l);
1669 trie_rebalance(t, tp);
1670 return;
1671 }
1672
1673 /* only access fa if it is pointing at the last valid hlist_node */
1674 if (*pprev)
1675 return;
1676
1677 /* update the trie with the latest suffix length */
1678 l->slen = fa->fa_slen;
1679 node_pull_suffix(tp, fa->fa_slen);
1680}
1681
1682static void fib_notify_alias_delete(struct net *net, u32 key,
1683 struct hlist_head *fah,
1684 struct fib_alias *fa_to_delete,
1685 struct netlink_ext_ack *extack)
1686{
1687 struct fib_alias *fa_next, *fa_to_notify;
1688 u32 tb_id = fa_to_delete->tb_id;
1689 u8 slen = fa_to_delete->fa_slen;
1690 enum fib_event_type fib_event;
1691
1692 /* Do not notify if we do not care about the route. */
1693 if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1694 return;
1695
1696 /* Determine if the route should be replaced by the next route in the
1697 * list.
1698 */
1699 fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1700 struct fib_alias, fa_list);
1701 if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1702 fib_event = FIB_EVENT_ENTRY_REPLACE;
1703 fa_to_notify = fa_next;
1704 } else {
1705 fib_event = FIB_EVENT_ENTRY_DEL;
1706 fa_to_notify = fa_to_delete;
1707 }
1708 call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1709 fa_to_notify, extack);
1710}
1711
1712/* Caller must hold RTNL. */
1713int fib_table_delete(struct net *net, struct fib_table *tb,
1714 struct fib_config *cfg, struct netlink_ext_ack *extack)
1715{
1716 struct trie *t = (struct trie *) tb->tb_data;
1717 struct fib_alias *fa, *fa_to_delete;
1718 struct key_vector *l, *tp;
1719 u8 plen = cfg->fc_dst_len;
1720 u8 slen = KEYLENGTH - plen;
1721 dscp_t dscp;
1722 u32 key;
1723
1724 key = ntohl(cfg->fc_dst);
1725
1726 if (!fib_valid_key_len(key, plen, extack))
1727 return -EINVAL;
1728
1729 l = fib_find_node(t, &tp, key);
1730 if (!l)
1731 return -ESRCH;
1732
1733 dscp = cfg->fc_dscp;
1734 fa = fib_find_alias(&l->leaf, slen, dscp, 0, tb->tb_id, false);
1735 if (!fa)
1736 return -ESRCH;
1737
1738 pr_debug("Deleting %08x/%d dsfield=0x%02x t=%p\n", key, plen,
1739 inet_dscp_to_dsfield(dscp), t);
1740
1741 fa_to_delete = NULL;
1742 hlist_for_each_entry_from(fa, fa_list) {
1743 struct fib_info *fi = fa->fa_info;
1744
1745 if ((fa->fa_slen != slen) ||
1746 (fa->tb_id != tb->tb_id) ||
1747 (fa->fa_dscp != dscp))
1748 break;
1749
1750 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1751 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1752 fa->fa_info->fib_scope == cfg->fc_scope) &&
1753 (!cfg->fc_prefsrc ||
1754 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1755 (!cfg->fc_protocol ||
1756 fi->fib_protocol == cfg->fc_protocol) &&
1757 fib_nh_match(net, cfg, fi, extack) == 0 &&
1758 fib_metrics_match(cfg, fi)) {
1759 fa_to_delete = fa;
1760 break;
1761 }
1762 }
1763
1764 if (!fa_to_delete)
1765 return -ESRCH;
1766
1767 fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1768 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1769 &cfg->fc_nlinfo, 0);
1770
1771 if (!plen)
1772 tb->tb_num_default--;
1773
1774 fib_remove_alias(t, tp, l, fa_to_delete);
1775
1776 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1777 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1778
1779 fib_release_info(fa_to_delete->fa_info);
1780 alias_free_mem_rcu(fa_to_delete);
1781 return 0;
1782}
1783
1784/* Scan for the next leaf starting at the provided key value */
1785static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1786{
1787 struct key_vector *pn, *n = *tn;
1788 unsigned long cindex;
1789
1790 /* this loop is meant to try and find the key in the trie */
1791 do {
1792 /* record parent and next child index */
1793 pn = n;
1794 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1795
1796 if (cindex >> pn->bits)
1797 break;
1798
1799 /* descend into the next child */
1800 n = get_child_rcu(pn, cindex++);
1801 if (!n)
1802 break;
1803
1804 /* guarantee forward progress on the keys */
1805 if (IS_LEAF(n) && (n->key >= key))
1806 goto found;
1807 } while (IS_TNODE(n));
1808
1809 /* this loop will search for the next leaf with a greater key */
1810 while (!IS_TRIE(pn)) {
1811 /* if we exhausted the parent node we will need to climb */
1812 if (cindex >= (1ul << pn->bits)) {
1813 t_key pkey = pn->key;
1814
1815 pn = node_parent_rcu(pn);
1816 cindex = get_index(pkey, pn) + 1;
1817 continue;
1818 }
1819
1820 /* grab the next available node */
1821 n = get_child_rcu(pn, cindex++);
1822 if (!n)
1823 continue;
1824
1825 /* no need to compare keys since we bumped the index */
1826 if (IS_LEAF(n))
1827 goto found;
1828
1829 /* Rescan start scanning in new node */
1830 pn = n;
1831 cindex = 0;
1832 }
1833
1834 *tn = pn;
1835 return NULL; /* Root of trie */
1836found:
1837 /* if we are at the limit for keys just return NULL for the tnode */
1838 *tn = pn;
1839 return n;
1840}
1841
1842static void fib_trie_free(struct fib_table *tb)
1843{
1844 struct trie *t = (struct trie *)tb->tb_data;
1845 struct key_vector *pn = t->kv;
1846 unsigned long cindex = 1;
1847 struct hlist_node *tmp;
1848 struct fib_alias *fa;
1849
1850 /* walk trie in reverse order and free everything */
1851 for (;;) {
1852 struct key_vector *n;
1853
1854 if (!(cindex--)) {
1855 t_key pkey = pn->key;
1856
1857 if (IS_TRIE(pn))
1858 break;
1859
1860 n = pn;
1861 pn = node_parent(pn);
1862
1863 /* drop emptied tnode */
1864 put_child_root(pn, n->key, NULL);
1865 node_free(n);
1866
1867 cindex = get_index(pkey, pn);
1868
1869 continue;
1870 }
1871
1872 /* grab the next available node */
1873 n = get_child(pn, cindex);
1874 if (!n)
1875 continue;
1876
1877 if (IS_TNODE(n)) {
1878 /* record pn and cindex for leaf walking */
1879 pn = n;
1880 cindex = 1ul << n->bits;
1881
1882 continue;
1883 }
1884
1885 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1886 hlist_del_rcu(&fa->fa_list);
1887 alias_free_mem_rcu(fa);
1888 }
1889
1890 put_child_root(pn, n->key, NULL);
1891 node_free(n);
1892 }
1893
1894#ifdef CONFIG_IP_FIB_TRIE_STATS
1895 free_percpu(t->stats);
1896#endif
1897 kfree(tb);
1898}
1899
1900struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1901{
1902 struct trie *ot = (struct trie *)oldtb->tb_data;
1903 struct key_vector *l, *tp = ot->kv;
1904 struct fib_table *local_tb;
1905 struct fib_alias *fa;
1906 struct trie *lt;
1907 t_key key = 0;
1908
1909 if (oldtb->tb_data == oldtb->__data)
1910 return oldtb;
1911
1912 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1913 if (!local_tb)
1914 return NULL;
1915
1916 lt = (struct trie *)local_tb->tb_data;
1917
1918 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1919 struct key_vector *local_l = NULL, *local_tp;
1920
1921 hlist_for_each_entry(fa, &l->leaf, fa_list) {
1922 struct fib_alias *new_fa;
1923
1924 if (local_tb->tb_id != fa->tb_id)
1925 continue;
1926
1927 /* clone fa for new local table */
1928 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1929 if (!new_fa)
1930 goto out;
1931
1932 memcpy(new_fa, fa, sizeof(*fa));
1933
1934 /* insert clone into table */
1935 if (!local_l)
1936 local_l = fib_find_node(lt, &local_tp, l->key);
1937
1938 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1939 NULL, l->key)) {
1940 kmem_cache_free(fn_alias_kmem, new_fa);
1941 goto out;
1942 }
1943 }
1944
1945 /* stop loop if key wrapped back to 0 */
1946 key = l->key + 1;
1947 if (key < l->key)
1948 break;
1949 }
1950
1951 return local_tb;
1952out:
1953 fib_trie_free(local_tb);
1954
1955 return NULL;
1956}
1957
1958/* Caller must hold RTNL */
1959void fib_table_flush_external(struct fib_table *tb)
1960{
1961 struct trie *t = (struct trie *)tb->tb_data;
1962 struct key_vector *pn = t->kv;
1963 unsigned long cindex = 1;
1964 struct hlist_node *tmp;
1965 struct fib_alias *fa;
1966
1967 /* walk trie in reverse order */
1968 for (;;) {
1969 unsigned char slen = 0;
1970 struct key_vector *n;
1971
1972 if (!(cindex--)) {
1973 t_key pkey = pn->key;
1974
1975 /* cannot resize the trie vector */
1976 if (IS_TRIE(pn))
1977 break;
1978
1979 /* update the suffix to address pulled leaves */
1980 if (pn->slen > pn->pos)
1981 update_suffix(pn);
1982
1983 /* resize completed node */
1984 pn = resize(t, pn);
1985 cindex = get_index(pkey, pn);
1986
1987 continue;
1988 }
1989
1990 /* grab the next available node */
1991 n = get_child(pn, cindex);
1992 if (!n)
1993 continue;
1994
1995 if (IS_TNODE(n)) {
1996 /* record pn and cindex for leaf walking */
1997 pn = n;
1998 cindex = 1ul << n->bits;
1999
2000 continue;
2001 }
2002
2003 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2004 /* if alias was cloned to local then we just
2005 * need to remove the local copy from main
2006 */
2007 if (tb->tb_id != fa->tb_id) {
2008 hlist_del_rcu(&fa->fa_list);
2009 alias_free_mem_rcu(fa);
2010 continue;
2011 }
2012
2013 /* record local slen */
2014 slen = fa->fa_slen;
2015 }
2016
2017 /* update leaf slen */
2018 n->slen = slen;
2019
2020 if (hlist_empty(&n->leaf)) {
2021 put_child_root(pn, n->key, NULL);
2022 node_free(n);
2023 }
2024 }
2025}
2026
2027/* Caller must hold RTNL. */
2028int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
2029{
2030 struct trie *t = (struct trie *)tb->tb_data;
2031 struct nl_info info = { .nl_net = net };
2032 struct key_vector *pn = t->kv;
2033 unsigned long cindex = 1;
2034 struct hlist_node *tmp;
2035 struct fib_alias *fa;
2036 int found = 0;
2037
2038 /* walk trie in reverse order */
2039 for (;;) {
2040 unsigned char slen = 0;
2041 struct key_vector *n;
2042
2043 if (!(cindex--)) {
2044 t_key pkey = pn->key;
2045
2046 /* cannot resize the trie vector */
2047 if (IS_TRIE(pn))
2048 break;
2049
2050 /* update the suffix to address pulled leaves */
2051 if (pn->slen > pn->pos)
2052 update_suffix(pn);
2053
2054 /* resize completed node */
2055 pn = resize(t, pn);
2056 cindex = get_index(pkey, pn);
2057
2058 continue;
2059 }
2060
2061 /* grab the next available node */
2062 n = get_child(pn, cindex);
2063 if (!n)
2064 continue;
2065
2066 if (IS_TNODE(n)) {
2067 /* record pn and cindex for leaf walking */
2068 pn = n;
2069 cindex = 1ul << n->bits;
2070
2071 continue;
2072 }
2073
2074 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2075 struct fib_info *fi = fa->fa_info;
2076
2077 if (!fi || tb->tb_id != fa->tb_id ||
2078 (!(fi->fib_flags & RTNH_F_DEAD) &&
2079 !fib_props[fa->fa_type].error)) {
2080 slen = fa->fa_slen;
2081 continue;
2082 }
2083
2084 /* Do not flush error routes if network namespace is
2085 * not being dismantled
2086 */
2087 if (!flush_all && fib_props[fa->fa_type].error) {
2088 slen = fa->fa_slen;
2089 continue;
2090 }
2091
2092 fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2093 NULL);
2094 if (fi->pfsrc_removed)
2095 rtmsg_fib(RTM_DELROUTE, htonl(n->key), fa,
2096 KEYLENGTH - fa->fa_slen, tb->tb_id, &info, 0);
2097 hlist_del_rcu(&fa->fa_list);
2098 fib_release_info(fa->fa_info);
2099 alias_free_mem_rcu(fa);
2100 found++;
2101 }
2102
2103 /* update leaf slen */
2104 n->slen = slen;
2105
2106 if (hlist_empty(&n->leaf)) {
2107 put_child_root(pn, n->key, NULL);
2108 node_free(n);
2109 }
2110 }
2111
2112 pr_debug("trie_flush found=%d\n", found);
2113 return found;
2114}
2115
2116/* derived from fib_trie_free */
2117static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2118 struct nl_info *info)
2119{
2120 struct trie *t = (struct trie *)tb->tb_data;
2121 struct key_vector *pn = t->kv;
2122 unsigned long cindex = 1;
2123 struct fib_alias *fa;
2124
2125 for (;;) {
2126 struct key_vector *n;
2127
2128 if (!(cindex--)) {
2129 t_key pkey = pn->key;
2130
2131 if (IS_TRIE(pn))
2132 break;
2133
2134 pn = node_parent(pn);
2135 cindex = get_index(pkey, pn);
2136 continue;
2137 }
2138
2139 /* grab the next available node */
2140 n = get_child(pn, cindex);
2141 if (!n)
2142 continue;
2143
2144 if (IS_TNODE(n)) {
2145 /* record pn and cindex for leaf walking */
2146 pn = n;
2147 cindex = 1ul << n->bits;
2148
2149 continue;
2150 }
2151
2152 hlist_for_each_entry(fa, &n->leaf, fa_list) {
2153 struct fib_info *fi = fa->fa_info;
2154
2155 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2156 continue;
2157
2158 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2159 KEYLENGTH - fa->fa_slen, tb->tb_id,
2160 info, NLM_F_REPLACE);
2161 }
2162 }
2163}
2164
2165void fib_info_notify_update(struct net *net, struct nl_info *info)
2166{
2167 unsigned int h;
2168
2169 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2170 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2171 struct fib_table *tb;
2172
2173 hlist_for_each_entry_rcu(tb, head, tb_hlist,
2174 lockdep_rtnl_is_held())
2175 __fib_info_notify_update(net, tb, info);
2176 }
2177}
2178
2179static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2180 struct notifier_block *nb,
2181 struct netlink_ext_ack *extack)
2182{
2183 struct fib_alias *fa;
2184 int last_slen = -1;
2185 int err;
2186
2187 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2188 struct fib_info *fi = fa->fa_info;
2189
2190 if (!fi)
2191 continue;
2192
2193 /* local and main table can share the same trie,
2194 * so don't notify twice for the same entry.
2195 */
2196 if (tb->tb_id != fa->tb_id)
2197 continue;
2198
2199 if (fa->fa_slen == last_slen)
2200 continue;
2201
2202 last_slen = fa->fa_slen;
2203 err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2204 l->key, KEYLENGTH - fa->fa_slen,
2205 fa, extack);
2206 if (err)
2207 return err;
2208 }
2209 return 0;
2210}
2211
2212static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2213 struct netlink_ext_ack *extack)
2214{
2215 struct trie *t = (struct trie *)tb->tb_data;
2216 struct key_vector *l, *tp = t->kv;
2217 t_key key = 0;
2218 int err;
2219
2220 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2221 err = fib_leaf_notify(l, tb, nb, extack);
2222 if (err)
2223 return err;
2224
2225 key = l->key + 1;
2226 /* stop in case of wrap around */
2227 if (key < l->key)
2228 break;
2229 }
2230 return 0;
2231}
2232
2233int fib_notify(struct net *net, struct notifier_block *nb,
2234 struct netlink_ext_ack *extack)
2235{
2236 unsigned int h;
2237 int err;
2238
2239 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2240 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2241 struct fib_table *tb;
2242
2243 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2244 err = fib_table_notify(tb, nb, extack);
2245 if (err)
2246 return err;
2247 }
2248 }
2249 return 0;
2250}
2251
2252static void __trie_free_rcu(struct rcu_head *head)
2253{
2254 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2255#ifdef CONFIG_IP_FIB_TRIE_STATS
2256 struct trie *t = (struct trie *)tb->tb_data;
2257
2258 if (tb->tb_data == tb->__data)
2259 free_percpu(t->stats);
2260#endif /* CONFIG_IP_FIB_TRIE_STATS */
2261 kfree(tb);
2262}
2263
2264void fib_free_table(struct fib_table *tb)
2265{
2266 call_rcu(&tb->rcu, __trie_free_rcu);
2267}
2268
2269static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2270 struct sk_buff *skb, struct netlink_callback *cb,
2271 struct fib_dump_filter *filter)
2272{
2273 unsigned int flags = NLM_F_MULTI;
2274 __be32 xkey = htonl(l->key);
2275 int i, s_i, i_fa, s_fa, err;
2276 struct fib_alias *fa;
2277
2278 if (filter->filter_set ||
2279 !filter->dump_exceptions || !filter->dump_routes)
2280 flags |= NLM_F_DUMP_FILTERED;
2281
2282 s_i = cb->args[4];
2283 s_fa = cb->args[5];
2284 i = 0;
2285
2286 /* rcu_read_lock is hold by caller */
2287 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2288 struct fib_info *fi = fa->fa_info;
2289
2290 if (i < s_i)
2291 goto next;
2292
2293 i_fa = 0;
2294
2295 if (tb->tb_id != fa->tb_id)
2296 goto next;
2297
2298 if (filter->filter_set) {
2299 if (filter->rt_type && fa->fa_type != filter->rt_type)
2300 goto next;
2301
2302 if ((filter->protocol &&
2303 fi->fib_protocol != filter->protocol))
2304 goto next;
2305
2306 if (filter->dev &&
2307 !fib_info_nh_uses_dev(fi, filter->dev))
2308 goto next;
2309 }
2310
2311 if (filter->dump_routes) {
2312 if (!s_fa) {
2313 struct fib_rt_info fri;
2314
2315 fri.fi = fi;
2316 fri.tb_id = tb->tb_id;
2317 fri.dst = xkey;
2318 fri.dst_len = KEYLENGTH - fa->fa_slen;
2319 fri.dscp = fa->fa_dscp;
2320 fri.type = fa->fa_type;
2321 fri.offload = READ_ONCE(fa->offload);
2322 fri.trap = READ_ONCE(fa->trap);
2323 fri.offload_failed = READ_ONCE(fa->offload_failed);
2324 err = fib_dump_info(skb,
2325 NETLINK_CB(cb->skb).portid,
2326 cb->nlh->nlmsg_seq,
2327 RTM_NEWROUTE, &fri, flags);
2328 if (err < 0)
2329 goto stop;
2330 }
2331
2332 i_fa++;
2333 }
2334
2335 if (filter->dump_exceptions) {
2336 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2337 &i_fa, s_fa, flags);
2338 if (err < 0)
2339 goto stop;
2340 }
2341
2342next:
2343 i++;
2344 }
2345
2346 cb->args[4] = i;
2347 return skb->len;
2348
2349stop:
2350 cb->args[4] = i;
2351 cb->args[5] = i_fa;
2352 return err;
2353}
2354
2355/* rcu_read_lock needs to be hold by caller from readside */
2356int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2357 struct netlink_callback *cb, struct fib_dump_filter *filter)
2358{
2359 struct trie *t = (struct trie *)tb->tb_data;
2360 struct key_vector *l, *tp = t->kv;
2361 /* Dump starting at last key.
2362 * Note: 0.0.0.0/0 (ie default) is first key.
2363 */
2364 int count = cb->args[2];
2365 t_key key = cb->args[3];
2366
2367 /* First time here, count and key are both always 0. Count > 0
2368 * and key == 0 means the dump has wrapped around and we are done.
2369 */
2370 if (count && !key)
2371 return skb->len;
2372
2373 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2374 int err;
2375
2376 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2377 if (err < 0) {
2378 cb->args[3] = key;
2379 cb->args[2] = count;
2380 return err;
2381 }
2382
2383 ++count;
2384 key = l->key + 1;
2385
2386 memset(&cb->args[4], 0,
2387 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2388
2389 /* stop loop if key wrapped back to 0 */
2390 if (key < l->key)
2391 break;
2392 }
2393
2394 cb->args[3] = key;
2395 cb->args[2] = count;
2396
2397 return skb->len;
2398}
2399
2400void __init fib_trie_init(void)
2401{
2402 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2403 sizeof(struct fib_alias),
2404 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2405
2406 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2407 LEAF_SIZE,
2408 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2409}
2410
2411struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2412{
2413 struct fib_table *tb;
2414 struct trie *t;
2415 size_t sz = sizeof(*tb);
2416
2417 if (!alias)
2418 sz += sizeof(struct trie);
2419
2420 tb = kzalloc(sz, GFP_KERNEL);
2421 if (!tb)
2422 return NULL;
2423
2424 tb->tb_id = id;
2425 tb->tb_num_default = 0;
2426 tb->tb_data = (alias ? alias->__data : tb->__data);
2427
2428 if (alias)
2429 return tb;
2430
2431 t = (struct trie *) tb->tb_data;
2432 t->kv[0].pos = KEYLENGTH;
2433 t->kv[0].slen = KEYLENGTH;
2434#ifdef CONFIG_IP_FIB_TRIE_STATS
2435 t->stats = alloc_percpu(struct trie_use_stats);
2436 if (!t->stats) {
2437 kfree(tb);
2438 tb = NULL;
2439 }
2440#endif
2441
2442 return tb;
2443}
2444
2445#ifdef CONFIG_PROC_FS
2446/* Depth first Trie walk iterator */
2447struct fib_trie_iter {
2448 struct seq_net_private p;
2449 struct fib_table *tb;
2450 struct key_vector *tnode;
2451 unsigned int index;
2452 unsigned int depth;
2453};
2454
2455static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2456{
2457 unsigned long cindex = iter->index;
2458 struct key_vector *pn = iter->tnode;
2459 t_key pkey;
2460
2461 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2462 iter->tnode, iter->index, iter->depth);
2463
2464 while (!IS_TRIE(pn)) {
2465 while (cindex < child_length(pn)) {
2466 struct key_vector *n = get_child_rcu(pn, cindex++);
2467
2468 if (!n)
2469 continue;
2470
2471 if (IS_LEAF(n)) {
2472 iter->tnode = pn;
2473 iter->index = cindex;
2474 } else {
2475 /* push down one level */
2476 iter->tnode = n;
2477 iter->index = 0;
2478 ++iter->depth;
2479 }
2480
2481 return n;
2482 }
2483
2484 /* Current node exhausted, pop back up */
2485 pkey = pn->key;
2486 pn = node_parent_rcu(pn);
2487 cindex = get_index(pkey, pn) + 1;
2488 --iter->depth;
2489 }
2490
2491 /* record root node so further searches know we are done */
2492 iter->tnode = pn;
2493 iter->index = 0;
2494
2495 return NULL;
2496}
2497
2498static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2499 struct trie *t)
2500{
2501 struct key_vector *n, *pn;
2502
2503 if (!t)
2504 return NULL;
2505
2506 pn = t->kv;
2507 n = rcu_dereference(pn->tnode[0]);
2508 if (!n)
2509 return NULL;
2510
2511 if (IS_TNODE(n)) {
2512 iter->tnode = n;
2513 iter->index = 0;
2514 iter->depth = 1;
2515 } else {
2516 iter->tnode = pn;
2517 iter->index = 0;
2518 iter->depth = 0;
2519 }
2520
2521 return n;
2522}
2523
2524static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2525{
2526 struct key_vector *n;
2527 struct fib_trie_iter iter;
2528
2529 memset(s, 0, sizeof(*s));
2530
2531 rcu_read_lock();
2532 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2533 if (IS_LEAF(n)) {
2534 struct fib_alias *fa;
2535
2536 s->leaves++;
2537 s->totdepth += iter.depth;
2538 if (iter.depth > s->maxdepth)
2539 s->maxdepth = iter.depth;
2540
2541 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2542 ++s->prefixes;
2543 } else {
2544 s->tnodes++;
2545 if (n->bits < MAX_STAT_DEPTH)
2546 s->nodesizes[n->bits]++;
2547 s->nullpointers += tn_info(n)->empty_children;
2548 }
2549 }
2550 rcu_read_unlock();
2551}
2552
2553/*
2554 * This outputs /proc/net/fib_triestats
2555 */
2556static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2557{
2558 unsigned int i, max, pointers, bytes, avdepth;
2559
2560 if (stat->leaves)
2561 avdepth = stat->totdepth*100 / stat->leaves;
2562 else
2563 avdepth = 0;
2564
2565 seq_printf(seq, "\tAver depth: %u.%02d\n",
2566 avdepth / 100, avdepth % 100);
2567 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2568
2569 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2570 bytes = LEAF_SIZE * stat->leaves;
2571
2572 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2573 bytes += sizeof(struct fib_alias) * stat->prefixes;
2574
2575 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2576 bytes += TNODE_SIZE(0) * stat->tnodes;
2577
2578 max = MAX_STAT_DEPTH;
2579 while (max > 0 && stat->nodesizes[max-1] == 0)
2580 max--;
2581
2582 pointers = 0;
2583 for (i = 1; i < max; i++)
2584 if (stat->nodesizes[i] != 0) {
2585 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2586 pointers += (1<<i) * stat->nodesizes[i];
2587 }
2588 seq_putc(seq, '\n');
2589 seq_printf(seq, "\tPointers: %u\n", pointers);
2590
2591 bytes += sizeof(struct key_vector *) * pointers;
2592 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2593 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2594}
2595
2596#ifdef CONFIG_IP_FIB_TRIE_STATS
2597static void trie_show_usage(struct seq_file *seq,
2598 const struct trie_use_stats __percpu *stats)
2599{
2600 struct trie_use_stats s = { 0 };
2601 int cpu;
2602
2603 /* loop through all of the CPUs and gather up the stats */
2604 for_each_possible_cpu(cpu) {
2605 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2606
2607 s.gets += pcpu->gets;
2608 s.backtrack += pcpu->backtrack;
2609 s.semantic_match_passed += pcpu->semantic_match_passed;
2610 s.semantic_match_miss += pcpu->semantic_match_miss;
2611 s.null_node_hit += pcpu->null_node_hit;
2612 s.resize_node_skipped += pcpu->resize_node_skipped;
2613 }
2614
2615 seq_printf(seq, "\nCounters:\n---------\n");
2616 seq_printf(seq, "gets = %u\n", s.gets);
2617 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2618 seq_printf(seq, "semantic match passed = %u\n",
2619 s.semantic_match_passed);
2620 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2621 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2622 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2623}
2624#endif /* CONFIG_IP_FIB_TRIE_STATS */
2625
2626static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2627{
2628 if (tb->tb_id == RT_TABLE_LOCAL)
2629 seq_puts(seq, "Local:\n");
2630 else if (tb->tb_id == RT_TABLE_MAIN)
2631 seq_puts(seq, "Main:\n");
2632 else
2633 seq_printf(seq, "Id %d:\n", tb->tb_id);
2634}
2635
2636
2637static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2638{
2639 struct net *net = seq->private;
2640 unsigned int h;
2641
2642 seq_printf(seq,
2643 "Basic info: size of leaf:"
2644 " %zd bytes, size of tnode: %zd bytes.\n",
2645 LEAF_SIZE, TNODE_SIZE(0));
2646
2647 rcu_read_lock();
2648 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2649 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2650 struct fib_table *tb;
2651
2652 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2653 struct trie *t = (struct trie *) tb->tb_data;
2654 struct trie_stat stat;
2655
2656 if (!t)
2657 continue;
2658
2659 fib_table_print(seq, tb);
2660
2661 trie_collect_stats(t, &stat);
2662 trie_show_stats(seq, &stat);
2663#ifdef CONFIG_IP_FIB_TRIE_STATS
2664 trie_show_usage(seq, t->stats);
2665#endif
2666 }
2667 cond_resched_rcu();
2668 }
2669 rcu_read_unlock();
2670
2671 return 0;
2672}
2673
2674static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2675{
2676 struct fib_trie_iter *iter = seq->private;
2677 struct net *net = seq_file_net(seq);
2678 loff_t idx = 0;
2679 unsigned int h;
2680
2681 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2682 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2683 struct fib_table *tb;
2684
2685 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2686 struct key_vector *n;
2687
2688 for (n = fib_trie_get_first(iter,
2689 (struct trie *) tb->tb_data);
2690 n; n = fib_trie_get_next(iter))
2691 if (pos == idx++) {
2692 iter->tb = tb;
2693 return n;
2694 }
2695 }
2696 }
2697
2698 return NULL;
2699}
2700
2701static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2702 __acquires(RCU)
2703{
2704 rcu_read_lock();
2705 return fib_trie_get_idx(seq, *pos);
2706}
2707
2708static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2709{
2710 struct fib_trie_iter *iter = seq->private;
2711 struct net *net = seq_file_net(seq);
2712 struct fib_table *tb = iter->tb;
2713 struct hlist_node *tb_node;
2714 unsigned int h;
2715 struct key_vector *n;
2716
2717 ++*pos;
2718 /* next node in same table */
2719 n = fib_trie_get_next(iter);
2720 if (n)
2721 return n;
2722
2723 /* walk rest of this hash chain */
2724 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2725 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2726 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2727 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2728 if (n)
2729 goto found;
2730 }
2731
2732 /* new hash chain */
2733 while (++h < FIB_TABLE_HASHSZ) {
2734 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2735 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2736 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2737 if (n)
2738 goto found;
2739 }
2740 }
2741 return NULL;
2742
2743found:
2744 iter->tb = tb;
2745 return n;
2746}
2747
2748static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2749 __releases(RCU)
2750{
2751 rcu_read_unlock();
2752}
2753
2754static void seq_indent(struct seq_file *seq, int n)
2755{
2756 while (n-- > 0)
2757 seq_puts(seq, " ");
2758}
2759
2760static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2761{
2762 switch (s) {
2763 case RT_SCOPE_UNIVERSE: return "universe";
2764 case RT_SCOPE_SITE: return "site";
2765 case RT_SCOPE_LINK: return "link";
2766 case RT_SCOPE_HOST: return "host";
2767 case RT_SCOPE_NOWHERE: return "nowhere";
2768 default:
2769 snprintf(buf, len, "scope=%d", s);
2770 return buf;
2771 }
2772}
2773
2774static const char *const rtn_type_names[__RTN_MAX] = {
2775 [RTN_UNSPEC] = "UNSPEC",
2776 [RTN_UNICAST] = "UNICAST",
2777 [RTN_LOCAL] = "LOCAL",
2778 [RTN_BROADCAST] = "BROADCAST",
2779 [RTN_ANYCAST] = "ANYCAST",
2780 [RTN_MULTICAST] = "MULTICAST",
2781 [RTN_BLACKHOLE] = "BLACKHOLE",
2782 [RTN_UNREACHABLE] = "UNREACHABLE",
2783 [RTN_PROHIBIT] = "PROHIBIT",
2784 [RTN_THROW] = "THROW",
2785 [RTN_NAT] = "NAT",
2786 [RTN_XRESOLVE] = "XRESOLVE",
2787};
2788
2789static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2790{
2791 if (t < __RTN_MAX && rtn_type_names[t])
2792 return rtn_type_names[t];
2793 snprintf(buf, len, "type %u", t);
2794 return buf;
2795}
2796
2797/* Pretty print the trie */
2798static int fib_trie_seq_show(struct seq_file *seq, void *v)
2799{
2800 const struct fib_trie_iter *iter = seq->private;
2801 struct key_vector *n = v;
2802
2803 if (IS_TRIE(node_parent_rcu(n)))
2804 fib_table_print(seq, iter->tb);
2805
2806 if (IS_TNODE(n)) {
2807 __be32 prf = htonl(n->key);
2808
2809 seq_indent(seq, iter->depth-1);
2810 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2811 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2812 tn_info(n)->full_children,
2813 tn_info(n)->empty_children);
2814 } else {
2815 __be32 val = htonl(n->key);
2816 struct fib_alias *fa;
2817
2818 seq_indent(seq, iter->depth);
2819 seq_printf(seq, " |-- %pI4\n", &val);
2820
2821 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2822 char buf1[32], buf2[32];
2823
2824 seq_indent(seq, iter->depth + 1);
2825 seq_printf(seq, " /%zu %s %s",
2826 KEYLENGTH - fa->fa_slen,
2827 rtn_scope(buf1, sizeof(buf1),
2828 fa->fa_info->fib_scope),
2829 rtn_type(buf2, sizeof(buf2),
2830 fa->fa_type));
2831 if (fa->fa_dscp)
2832 seq_printf(seq, " tos=%d",
2833 inet_dscp_to_dsfield(fa->fa_dscp));
2834 seq_putc(seq, '\n');
2835 }
2836 }
2837
2838 return 0;
2839}
2840
2841static const struct seq_operations fib_trie_seq_ops = {
2842 .start = fib_trie_seq_start,
2843 .next = fib_trie_seq_next,
2844 .stop = fib_trie_seq_stop,
2845 .show = fib_trie_seq_show,
2846};
2847
2848struct fib_route_iter {
2849 struct seq_net_private p;
2850 struct fib_table *main_tb;
2851 struct key_vector *tnode;
2852 loff_t pos;
2853 t_key key;
2854};
2855
2856static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2857 loff_t pos)
2858{
2859 struct key_vector *l, **tp = &iter->tnode;
2860 t_key key;
2861
2862 /* use cached location of previously found key */
2863 if (iter->pos > 0 && pos >= iter->pos) {
2864 key = iter->key;
2865 } else {
2866 iter->pos = 1;
2867 key = 0;
2868 }
2869
2870 pos -= iter->pos;
2871
2872 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2873 key = l->key + 1;
2874 iter->pos++;
2875 l = NULL;
2876
2877 /* handle unlikely case of a key wrap */
2878 if (!key)
2879 break;
2880 }
2881
2882 if (l)
2883 iter->key = l->key; /* remember it */
2884 else
2885 iter->pos = 0; /* forget it */
2886
2887 return l;
2888}
2889
2890static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2891 __acquires(RCU)
2892{
2893 struct fib_route_iter *iter = seq->private;
2894 struct fib_table *tb;
2895 struct trie *t;
2896
2897 rcu_read_lock();
2898
2899 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2900 if (!tb)
2901 return NULL;
2902
2903 iter->main_tb = tb;
2904 t = (struct trie *)tb->tb_data;
2905 iter->tnode = t->kv;
2906
2907 if (*pos != 0)
2908 return fib_route_get_idx(iter, *pos);
2909
2910 iter->pos = 0;
2911 iter->key = KEY_MAX;
2912
2913 return SEQ_START_TOKEN;
2914}
2915
2916static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2917{
2918 struct fib_route_iter *iter = seq->private;
2919 struct key_vector *l = NULL;
2920 t_key key = iter->key + 1;
2921
2922 ++*pos;
2923
2924 /* only allow key of 0 for start of sequence */
2925 if ((v == SEQ_START_TOKEN) || key)
2926 l = leaf_walk_rcu(&iter->tnode, key);
2927
2928 if (l) {
2929 iter->key = l->key;
2930 iter->pos++;
2931 } else {
2932 iter->pos = 0;
2933 }
2934
2935 return l;
2936}
2937
2938static void fib_route_seq_stop(struct seq_file *seq, void *v)
2939 __releases(RCU)
2940{
2941 rcu_read_unlock();
2942}
2943
2944static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2945{
2946 unsigned int flags = 0;
2947
2948 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2949 flags = RTF_REJECT;
2950 if (fi) {
2951 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2952
2953 if (nhc->nhc_gw.ipv4)
2954 flags |= RTF_GATEWAY;
2955 }
2956 if (mask == htonl(0xFFFFFFFF))
2957 flags |= RTF_HOST;
2958 flags |= RTF_UP;
2959 return flags;
2960}
2961
2962/*
2963 * This outputs /proc/net/route.
2964 * The format of the file is not supposed to be changed
2965 * and needs to be same as fib_hash output to avoid breaking
2966 * legacy utilities
2967 */
2968static int fib_route_seq_show(struct seq_file *seq, void *v)
2969{
2970 struct fib_route_iter *iter = seq->private;
2971 struct fib_table *tb = iter->main_tb;
2972 struct fib_alias *fa;
2973 struct key_vector *l = v;
2974 __be32 prefix;
2975
2976 if (v == SEQ_START_TOKEN) {
2977 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2978 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2979 "\tWindow\tIRTT");
2980 return 0;
2981 }
2982
2983 prefix = htonl(l->key);
2984
2985 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2986 struct fib_info *fi = fa->fa_info;
2987 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2988 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2989
2990 if ((fa->fa_type == RTN_BROADCAST) ||
2991 (fa->fa_type == RTN_MULTICAST))
2992 continue;
2993
2994 if (fa->tb_id != tb->tb_id)
2995 continue;
2996
2997 seq_setwidth(seq, 127);
2998
2999 if (fi) {
3000 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
3001 __be32 gw = 0;
3002
3003 if (nhc->nhc_gw_family == AF_INET)
3004 gw = nhc->nhc_gw.ipv4;
3005
3006 seq_printf(seq,
3007 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
3008 "%d\t%08X\t%d\t%u\t%u",
3009 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
3010 prefix, gw, flags, 0, 0,
3011 fi->fib_priority,
3012 mask,
3013 (fi->fib_advmss ?
3014 fi->fib_advmss + 40 : 0),
3015 fi->fib_window,
3016 fi->fib_rtt >> 3);
3017 } else {
3018 seq_printf(seq,
3019 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
3020 "%d\t%08X\t%d\t%u\t%u",
3021 prefix, 0, flags, 0, 0, 0,
3022 mask, 0, 0, 0);
3023 }
3024 seq_pad(seq, '\n');
3025 }
3026
3027 return 0;
3028}
3029
3030static const struct seq_operations fib_route_seq_ops = {
3031 .start = fib_route_seq_start,
3032 .next = fib_route_seq_next,
3033 .stop = fib_route_seq_stop,
3034 .show = fib_route_seq_show,
3035};
3036
3037int __net_init fib_proc_init(struct net *net)
3038{
3039 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
3040 sizeof(struct fib_trie_iter)))
3041 goto out1;
3042
3043 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
3044 fib_triestat_seq_show, NULL))
3045 goto out2;
3046
3047 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
3048 sizeof(struct fib_route_iter)))
3049 goto out3;
3050
3051 return 0;
3052
3053out3:
3054 remove_proc_entry("fib_triestat", net->proc_net);
3055out2:
3056 remove_proc_entry("fib_trie", net->proc_net);
3057out1:
3058 return -ENOMEM;
3059}
3060
3061void __net_exit fib_proc_exit(struct net *net)
3062{
3063 remove_proc_entry("fib_trie", net->proc_net);
3064 remove_proc_entry("fib_triestat", net->proc_net);
3065 remove_proc_entry("route", net->proc_net);
3066}
3067
3068#endif /* CONFIG_PROC_FS */