]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/ipv4/fib_trie.c
744e5936c10d7ec555d1ca621f5bd4be57f1c72b
[mirror_ubuntu-bionic-kernel.git] / net / ipv4 / fib_trie.c
1 /*
2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
6 *
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
9 *
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
12 *
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
14 *
15 * This work is based on the LPC-trie which is originally described in:
16 *
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
20 *
21 *
22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
24 *
25 *
26 * Code from fib_hash has been reused which includes the following header:
27 *
28 *
29 * INET An implementation of the TCP/IP protocol suite for the LINUX
30 * operating system. INET is implemented using the BSD Socket
31 * interface as the means of communication with the user level.
32 *
33 * IPv4 FIB: lookup engine and maintenance routines.
34 *
35 *
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
37 *
38 * This program is free software; you can redistribute it and/or
39 * modify it under the terms of the GNU General Public License
40 * as published by the Free Software Foundation; either version
41 * 2 of the License, or (at your option) any later version.
42 *
43 * Substantial contributions to this work comes from:
44 *
45 * David S. Miller, <davem@davemloft.net>
46 * Stephen Hemminger <shemminger@osdl.org>
47 * Paul E. McKenney <paulmck@us.ibm.com>
48 * Patrick McHardy <kaber@trash.net>
49 */
50
51 #define VERSION "0.409"
52
53 #include <asm/uaccess.h>
54 #include <linux/bitops.h>
55 #include <linux/types.h>
56 #include <linux/kernel.h>
57 #include <linux/mm.h>
58 #include <linux/string.h>
59 #include <linux/socket.h>
60 #include <linux/sockios.h>
61 #include <linux/errno.h>
62 #include <linux/in.h>
63 #include <linux/inet.h>
64 #include <linux/inetdevice.h>
65 #include <linux/netdevice.h>
66 #include <linux/if_arp.h>
67 #include <linux/proc_fs.h>
68 #include <linux/rcupdate.h>
69 #include <linux/skbuff.h>
70 #include <linux/netlink.h>
71 #include <linux/init.h>
72 #include <linux/list.h>
73 #include <linux/slab.h>
74 #include <linux/export.h>
75 #include <linux/vmalloc.h>
76 #include <net/net_namespace.h>
77 #include <net/ip.h>
78 #include <net/protocol.h>
79 #include <net/route.h>
80 #include <net/tcp.h>
81 #include <net/sock.h>
82 #include <net/ip_fib.h>
83 #include <net/switchdev.h>
84 #include <trace/events/fib.h>
85 #include "fib_lookup.h"
86
87 #define MAX_STAT_DEPTH 32
88
89 #define KEYLENGTH (8*sizeof(t_key))
90 #define KEY_MAX ((t_key)~0)
91
92 typedef unsigned int t_key;
93
94 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
95 #define IS_TNODE(n) ((n)->bits)
96 #define IS_LEAF(n) (!(n)->bits)
97
98 struct key_vector {
99 t_key key;
100 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
101 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
102 unsigned char slen;
103 union {
104 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
105 struct hlist_head leaf;
106 /* This array is valid if (pos | bits) > 0 (TNODE) */
107 struct key_vector __rcu *tnode[0];
108 };
109 };
110
111 struct tnode {
112 struct rcu_head rcu;
113 t_key empty_children; /* KEYLENGTH bits needed */
114 t_key full_children; /* KEYLENGTH bits needed */
115 struct key_vector __rcu *parent;
116 struct key_vector kv[1];
117 #define tn_bits kv[0].bits
118 };
119
120 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
121 #define LEAF_SIZE TNODE_SIZE(1)
122
123 #ifdef CONFIG_IP_FIB_TRIE_STATS
124 struct trie_use_stats {
125 unsigned int gets;
126 unsigned int backtrack;
127 unsigned int semantic_match_passed;
128 unsigned int semantic_match_miss;
129 unsigned int null_node_hit;
130 unsigned int resize_node_skipped;
131 };
132 #endif
133
134 struct trie_stat {
135 unsigned int totdepth;
136 unsigned int maxdepth;
137 unsigned int tnodes;
138 unsigned int leaves;
139 unsigned int nullpointers;
140 unsigned int prefixes;
141 unsigned int nodesizes[MAX_STAT_DEPTH];
142 };
143
144 struct trie {
145 struct key_vector kv[1];
146 #ifdef CONFIG_IP_FIB_TRIE_STATS
147 struct trie_use_stats __percpu *stats;
148 #endif
149 };
150
151 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
152 static size_t tnode_free_size;
153
154 /*
155 * synchronize_rcu after call_rcu for that many pages; it should be especially
156 * useful before resizing the root node with PREEMPT_NONE configs; the value was
157 * obtained experimentally, aiming to avoid visible slowdown.
158 */
159 static const int sync_pages = 128;
160
161 static struct kmem_cache *fn_alias_kmem __read_mostly;
162 static struct kmem_cache *trie_leaf_kmem __read_mostly;
163
164 static inline struct tnode *tn_info(struct key_vector *kv)
165 {
166 return container_of(kv, struct tnode, kv[0]);
167 }
168
169 /* caller must hold RTNL */
170 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
171 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
172
173 /* caller must hold RCU read lock or RTNL */
174 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
175 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
176
177 /* wrapper for rcu_assign_pointer */
178 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
179 {
180 if (n)
181 rcu_assign_pointer(tn_info(n)->parent, tp);
182 }
183
184 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
185
186 /* This provides us with the number of children in this node, in the case of a
187 * leaf this will return 0 meaning none of the children are accessible.
188 */
189 static inline unsigned long child_length(const struct key_vector *tn)
190 {
191 return (1ul << tn->bits) & ~(1ul);
192 }
193
194 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
195
196 static inline unsigned long get_index(t_key key, struct key_vector *kv)
197 {
198 unsigned long index = key ^ kv->key;
199
200 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
201 return 0;
202
203 return index >> kv->pos;
204 }
205
206 /* To understand this stuff, an understanding of keys and all their bits is
207 * necessary. Every node in the trie has a key associated with it, but not
208 * all of the bits in that key are significant.
209 *
210 * Consider a node 'n' and its parent 'tp'.
211 *
212 * If n is a leaf, every bit in its key is significant. Its presence is
213 * necessitated by path compression, since during a tree traversal (when
214 * searching for a leaf - unless we are doing an insertion) we will completely
215 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
216 * a potentially successful search, that we have indeed been walking the
217 * correct key path.
218 *
219 * Note that we can never "miss" the correct key in the tree if present by
220 * following the wrong path. Path compression ensures that segments of the key
221 * that are the same for all keys with a given prefix are skipped, but the
222 * skipped part *is* identical for each node in the subtrie below the skipped
223 * bit! trie_insert() in this implementation takes care of that.
224 *
225 * if n is an internal node - a 'tnode' here, the various parts of its key
226 * have many different meanings.
227 *
228 * Example:
229 * _________________________________________________________________
230 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
231 * -----------------------------------------------------------------
232 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
233 *
234 * _________________________________________________________________
235 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
236 * -----------------------------------------------------------------
237 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
238 *
239 * tp->pos = 22
240 * tp->bits = 3
241 * n->pos = 13
242 * n->bits = 4
243 *
244 * First, let's just ignore the bits that come before the parent tp, that is
245 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
246 * point we do not use them for anything.
247 *
248 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
249 * index into the parent's child array. That is, they will be used to find
250 * 'n' among tp's children.
251 *
252 * The bits from (n->pos + n->bits) to (tn->pos - 1) - "S" - are skipped bits
253 * for the node n.
254 *
255 * All the bits we have seen so far are significant to the node n. The rest
256 * of the bits are really not needed or indeed known in n->key.
257 *
258 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
259 * n's child array, and will of course be different for each child.
260 *
261 * The rest of the bits, from 0 to (n->pos + n->bits), are completely unknown
262 * at this point.
263 */
264
265 static const int halve_threshold = 25;
266 static const int inflate_threshold = 50;
267 static const int halve_threshold_root = 15;
268 static const int inflate_threshold_root = 30;
269
270 static void __alias_free_mem(struct rcu_head *head)
271 {
272 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
273 kmem_cache_free(fn_alias_kmem, fa);
274 }
275
276 static inline void alias_free_mem_rcu(struct fib_alias *fa)
277 {
278 call_rcu(&fa->rcu, __alias_free_mem);
279 }
280
281 #define TNODE_KMALLOC_MAX \
282 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
283 #define TNODE_VMALLOC_MAX \
284 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
285
286 static void __node_free_rcu(struct rcu_head *head)
287 {
288 struct tnode *n = container_of(head, struct tnode, rcu);
289
290 if (!n->tn_bits)
291 kmem_cache_free(trie_leaf_kmem, n);
292 else if (n->tn_bits <= TNODE_KMALLOC_MAX)
293 kfree(n);
294 else
295 vfree(n);
296 }
297
298 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
299
300 static struct tnode *tnode_alloc(int bits)
301 {
302 size_t size;
303
304 /* verify bits is within bounds */
305 if (bits > TNODE_VMALLOC_MAX)
306 return NULL;
307
308 /* determine size and verify it is non-zero and didn't overflow */
309 size = TNODE_SIZE(1ul << bits);
310
311 if (size <= PAGE_SIZE)
312 return kzalloc(size, GFP_KERNEL);
313 else
314 return vzalloc(size);
315 }
316
317 static inline void empty_child_inc(struct key_vector *n)
318 {
319 ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
320 }
321
322 static inline void empty_child_dec(struct key_vector *n)
323 {
324 tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
325 }
326
327 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
328 {
329 struct key_vector *l;
330 struct tnode *kv;
331
332 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
333 if (!kv)
334 return NULL;
335
336 /* initialize key vector */
337 l = kv->kv;
338 l->key = key;
339 l->pos = 0;
340 l->bits = 0;
341 l->slen = fa->fa_slen;
342
343 /* link leaf to fib alias */
344 INIT_HLIST_HEAD(&l->leaf);
345 hlist_add_head(&fa->fa_list, &l->leaf);
346
347 return l;
348 }
349
350 static struct key_vector *tnode_new(t_key key, int pos, int bits)
351 {
352 unsigned int shift = pos + bits;
353 struct key_vector *tn;
354 struct tnode *tnode;
355
356 /* verify bits and pos their msb bits clear and values are valid */
357 BUG_ON(!bits || (shift > KEYLENGTH));
358
359 tnode = tnode_alloc(bits);
360 if (!tnode)
361 return NULL;
362
363 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
364 sizeof(struct key_vector *) << bits);
365
366 if (bits == KEYLENGTH)
367 tnode->full_children = 1;
368 else
369 tnode->empty_children = 1ul << bits;
370
371 tn = tnode->kv;
372 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
373 tn->pos = pos;
374 tn->bits = bits;
375 tn->slen = pos;
376
377 return tn;
378 }
379
380 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
381 * and no bits are skipped. See discussion in dyntree paper p. 6
382 */
383 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
384 {
385 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
386 }
387
388 /* Add a child at position i overwriting the old value.
389 * Update the value of full_children and empty_children.
390 */
391 static void put_child(struct key_vector *tn, unsigned long i,
392 struct key_vector *n)
393 {
394 struct key_vector *chi = get_child(tn, i);
395 int isfull, wasfull;
396
397 BUG_ON(i >= child_length(tn));
398
399 /* update emptyChildren, overflow into fullChildren */
400 if (!n && chi)
401 empty_child_inc(tn);
402 if (n && !chi)
403 empty_child_dec(tn);
404
405 /* update fullChildren */
406 wasfull = tnode_full(tn, chi);
407 isfull = tnode_full(tn, n);
408
409 if (wasfull && !isfull)
410 tn_info(tn)->full_children--;
411 else if (!wasfull && isfull)
412 tn_info(tn)->full_children++;
413
414 if (n && (tn->slen < n->slen))
415 tn->slen = n->slen;
416
417 rcu_assign_pointer(tn->tnode[i], n);
418 }
419
420 static void update_children(struct key_vector *tn)
421 {
422 unsigned long i;
423
424 /* update all of the child parent pointers */
425 for (i = child_length(tn); i;) {
426 struct key_vector *inode = get_child(tn, --i);
427
428 if (!inode)
429 continue;
430
431 /* Either update the children of a tnode that
432 * already belongs to us or update the child
433 * to point to ourselves.
434 */
435 if (node_parent(inode) == tn)
436 update_children(inode);
437 else
438 node_set_parent(inode, tn);
439 }
440 }
441
442 static inline void put_child_root(struct key_vector *tp, t_key key,
443 struct key_vector *n)
444 {
445 if (IS_TRIE(tp))
446 rcu_assign_pointer(tp->tnode[0], n);
447 else
448 put_child(tp, get_index(key, tp), n);
449 }
450
451 static inline void tnode_free_init(struct key_vector *tn)
452 {
453 tn_info(tn)->rcu.next = NULL;
454 }
455
456 static inline void tnode_free_append(struct key_vector *tn,
457 struct key_vector *n)
458 {
459 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
460 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
461 }
462
463 static void tnode_free(struct key_vector *tn)
464 {
465 struct callback_head *head = &tn_info(tn)->rcu;
466
467 while (head) {
468 head = head->next;
469 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
470 node_free(tn);
471
472 tn = container_of(head, struct tnode, rcu)->kv;
473 }
474
475 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
476 tnode_free_size = 0;
477 synchronize_rcu();
478 }
479 }
480
481 static struct key_vector *replace(struct trie *t,
482 struct key_vector *oldtnode,
483 struct key_vector *tn)
484 {
485 struct key_vector *tp = node_parent(oldtnode);
486 unsigned long i;
487
488 /* setup the parent pointer out of and back into this node */
489 NODE_INIT_PARENT(tn, tp);
490 put_child_root(tp, tn->key, tn);
491
492 /* update all of the child parent pointers */
493 update_children(tn);
494
495 /* all pointers should be clean so we are done */
496 tnode_free(oldtnode);
497
498 /* resize children now that oldtnode is freed */
499 for (i = child_length(tn); i;) {
500 struct key_vector *inode = get_child(tn, --i);
501
502 /* resize child node */
503 if (tnode_full(tn, inode))
504 tn = resize(t, inode);
505 }
506
507 return tp;
508 }
509
510 static struct key_vector *inflate(struct trie *t,
511 struct key_vector *oldtnode)
512 {
513 struct key_vector *tn;
514 unsigned long i;
515 t_key m;
516
517 pr_debug("In inflate\n");
518
519 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
520 if (!tn)
521 goto notnode;
522
523 /* prepare oldtnode to be freed */
524 tnode_free_init(oldtnode);
525
526 /* Assemble all of the pointers in our cluster, in this case that
527 * represents all of the pointers out of our allocated nodes that
528 * point to existing tnodes and the links between our allocated
529 * nodes.
530 */
531 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
532 struct key_vector *inode = get_child(oldtnode, --i);
533 struct key_vector *node0, *node1;
534 unsigned long j, k;
535
536 /* An empty child */
537 if (!inode)
538 continue;
539
540 /* A leaf or an internal node with skipped bits */
541 if (!tnode_full(oldtnode, inode)) {
542 put_child(tn, get_index(inode->key, tn), inode);
543 continue;
544 }
545
546 /* drop the node in the old tnode free list */
547 tnode_free_append(oldtnode, inode);
548
549 /* An internal node with two children */
550 if (inode->bits == 1) {
551 put_child(tn, 2 * i + 1, get_child(inode, 1));
552 put_child(tn, 2 * i, get_child(inode, 0));
553 continue;
554 }
555
556 /* We will replace this node 'inode' with two new
557 * ones, 'node0' and 'node1', each with half of the
558 * original children. The two new nodes will have
559 * a position one bit further down the key and this
560 * means that the "significant" part of their keys
561 * (see the discussion near the top of this file)
562 * will differ by one bit, which will be "0" in
563 * node0's key and "1" in node1's key. Since we are
564 * moving the key position by one step, the bit that
565 * we are moving away from - the bit at position
566 * (tn->pos) - is the one that will differ between
567 * node0 and node1. So... we synthesize that bit in the
568 * two new keys.
569 */
570 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
571 if (!node1)
572 goto nomem;
573 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
574
575 tnode_free_append(tn, node1);
576 if (!node0)
577 goto nomem;
578 tnode_free_append(tn, node0);
579
580 /* populate child pointers in new nodes */
581 for (k = child_length(inode), j = k / 2; j;) {
582 put_child(node1, --j, get_child(inode, --k));
583 put_child(node0, j, get_child(inode, j));
584 put_child(node1, --j, get_child(inode, --k));
585 put_child(node0, j, get_child(inode, j));
586 }
587
588 /* link new nodes to parent */
589 NODE_INIT_PARENT(node1, tn);
590 NODE_INIT_PARENT(node0, tn);
591
592 /* link parent to nodes */
593 put_child(tn, 2 * i + 1, node1);
594 put_child(tn, 2 * i, node0);
595 }
596
597 /* setup the parent pointers into and out of this node */
598 return replace(t, oldtnode, tn);
599 nomem:
600 /* all pointers should be clean so we are done */
601 tnode_free(tn);
602 notnode:
603 return NULL;
604 }
605
606 static struct key_vector *halve(struct trie *t,
607 struct key_vector *oldtnode)
608 {
609 struct key_vector *tn;
610 unsigned long i;
611
612 pr_debug("In halve\n");
613
614 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
615 if (!tn)
616 goto notnode;
617
618 /* prepare oldtnode to be freed */
619 tnode_free_init(oldtnode);
620
621 /* Assemble all of the pointers in our cluster, in this case that
622 * represents all of the pointers out of our allocated nodes that
623 * point to existing tnodes and the links between our allocated
624 * nodes.
625 */
626 for (i = child_length(oldtnode); i;) {
627 struct key_vector *node1 = get_child(oldtnode, --i);
628 struct key_vector *node0 = get_child(oldtnode, --i);
629 struct key_vector *inode;
630
631 /* At least one of the children is empty */
632 if (!node1 || !node0) {
633 put_child(tn, i / 2, node1 ? : node0);
634 continue;
635 }
636
637 /* Two nonempty children */
638 inode = tnode_new(node0->key, oldtnode->pos, 1);
639 if (!inode)
640 goto nomem;
641 tnode_free_append(tn, inode);
642
643 /* initialize pointers out of node */
644 put_child(inode, 1, node1);
645 put_child(inode, 0, node0);
646 NODE_INIT_PARENT(inode, tn);
647
648 /* link parent to node */
649 put_child(tn, i / 2, inode);
650 }
651
652 /* setup the parent pointers into and out of this node */
653 return replace(t, oldtnode, tn);
654 nomem:
655 /* all pointers should be clean so we are done */
656 tnode_free(tn);
657 notnode:
658 return NULL;
659 }
660
661 static struct key_vector *collapse(struct trie *t,
662 struct key_vector *oldtnode)
663 {
664 struct key_vector *n, *tp;
665 unsigned long i;
666
667 /* scan the tnode looking for that one child that might still exist */
668 for (n = NULL, i = child_length(oldtnode); !n && i;)
669 n = get_child(oldtnode, --i);
670
671 /* compress one level */
672 tp = node_parent(oldtnode);
673 put_child_root(tp, oldtnode->key, n);
674 node_set_parent(n, tp);
675
676 /* drop dead node */
677 node_free(oldtnode);
678
679 return tp;
680 }
681
682 static unsigned char update_suffix(struct key_vector *tn)
683 {
684 unsigned char slen = tn->pos;
685 unsigned long stride, i;
686
687 /* search though the list of children looking for nodes that might
688 * have a suffix greater than the one we currently have. This is
689 * why we start with a stride of 2 since a stride of 1 would
690 * represent the nodes with suffix length equal to tn->pos
691 */
692 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
693 struct key_vector *n = get_child(tn, i);
694
695 if (!n || (n->slen <= slen))
696 continue;
697
698 /* update stride and slen based on new value */
699 stride <<= (n->slen - slen);
700 slen = n->slen;
701 i &= ~(stride - 1);
702
703 /* if slen covers all but the last bit we can stop here
704 * there will be nothing longer than that since only node
705 * 0 and 1 << (bits - 1) could have that as their suffix
706 * length.
707 */
708 if ((slen + 1) >= (tn->pos + tn->bits))
709 break;
710 }
711
712 tn->slen = slen;
713
714 return slen;
715 }
716
717 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
718 * the Helsinki University of Technology and Matti Tikkanen of Nokia
719 * Telecommunications, page 6:
720 * "A node is doubled if the ratio of non-empty children to all
721 * children in the *doubled* node is at least 'high'."
722 *
723 * 'high' in this instance is the variable 'inflate_threshold'. It
724 * is expressed as a percentage, so we multiply it with
725 * child_length() and instead of multiplying by 2 (since the
726 * child array will be doubled by inflate()) and multiplying
727 * the left-hand side by 100 (to handle the percentage thing) we
728 * multiply the left-hand side by 50.
729 *
730 * The left-hand side may look a bit weird: child_length(tn)
731 * - tn->empty_children is of course the number of non-null children
732 * in the current node. tn->full_children is the number of "full"
733 * children, that is non-null tnodes with a skip value of 0.
734 * All of those will be doubled in the resulting inflated tnode, so
735 * we just count them one extra time here.
736 *
737 * A clearer way to write this would be:
738 *
739 * to_be_doubled = tn->full_children;
740 * not_to_be_doubled = child_length(tn) - tn->empty_children -
741 * tn->full_children;
742 *
743 * new_child_length = child_length(tn) * 2;
744 *
745 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
746 * new_child_length;
747 * if (new_fill_factor >= inflate_threshold)
748 *
749 * ...and so on, tho it would mess up the while () loop.
750 *
751 * anyway,
752 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
753 * inflate_threshold
754 *
755 * avoid a division:
756 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
757 * inflate_threshold * new_child_length
758 *
759 * expand not_to_be_doubled and to_be_doubled, and shorten:
760 * 100 * (child_length(tn) - tn->empty_children +
761 * tn->full_children) >= inflate_threshold * new_child_length
762 *
763 * expand new_child_length:
764 * 100 * (child_length(tn) - tn->empty_children +
765 * tn->full_children) >=
766 * inflate_threshold * child_length(tn) * 2
767 *
768 * shorten again:
769 * 50 * (tn->full_children + child_length(tn) -
770 * tn->empty_children) >= inflate_threshold *
771 * child_length(tn)
772 *
773 */
774 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
775 {
776 unsigned long used = child_length(tn);
777 unsigned long threshold = used;
778
779 /* Keep root node larger */
780 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
781 used -= tn_info(tn)->empty_children;
782 used += tn_info(tn)->full_children;
783
784 /* if bits == KEYLENGTH then pos = 0, and will fail below */
785
786 return (used > 1) && tn->pos && ((50 * used) >= threshold);
787 }
788
789 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
790 {
791 unsigned long used = child_length(tn);
792 unsigned long threshold = used;
793
794 /* Keep root node larger */
795 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
796 used -= tn_info(tn)->empty_children;
797
798 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
799
800 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
801 }
802
803 static inline bool should_collapse(struct key_vector *tn)
804 {
805 unsigned long used = child_length(tn);
806
807 used -= tn_info(tn)->empty_children;
808
809 /* account for bits == KEYLENGTH case */
810 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
811 used -= KEY_MAX;
812
813 /* One child or none, time to drop us from the trie */
814 return used < 2;
815 }
816
817 #define MAX_WORK 10
818 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
819 {
820 #ifdef CONFIG_IP_FIB_TRIE_STATS
821 struct trie_use_stats __percpu *stats = t->stats;
822 #endif
823 struct key_vector *tp = node_parent(tn);
824 unsigned long cindex = get_index(tn->key, tp);
825 int max_work = MAX_WORK;
826
827 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
828 tn, inflate_threshold, halve_threshold);
829
830 /* track the tnode via the pointer from the parent instead of
831 * doing it ourselves. This way we can let RCU fully do its
832 * thing without us interfering
833 */
834 BUG_ON(tn != get_child(tp, cindex));
835
836 /* Double as long as the resulting node has a number of
837 * nonempty nodes that are above the threshold.
838 */
839 while (should_inflate(tp, tn) && max_work) {
840 tp = inflate(t, tn);
841 if (!tp) {
842 #ifdef CONFIG_IP_FIB_TRIE_STATS
843 this_cpu_inc(stats->resize_node_skipped);
844 #endif
845 break;
846 }
847
848 max_work--;
849 tn = get_child(tp, cindex);
850 }
851
852 /* update parent in case inflate failed */
853 tp = node_parent(tn);
854
855 /* Return if at least one inflate is run */
856 if (max_work != MAX_WORK)
857 return tp;
858
859 /* Halve as long as the number of empty children in this
860 * node is above threshold.
861 */
862 while (should_halve(tp, tn) && max_work) {
863 tp = halve(t, tn);
864 if (!tp) {
865 #ifdef CONFIG_IP_FIB_TRIE_STATS
866 this_cpu_inc(stats->resize_node_skipped);
867 #endif
868 break;
869 }
870
871 max_work--;
872 tn = get_child(tp, cindex);
873 }
874
875 /* Only one child remains */
876 if (should_collapse(tn))
877 return collapse(t, tn);
878
879 /* update parent in case halve failed */
880 tp = node_parent(tn);
881
882 /* Return if at least one deflate was run */
883 if (max_work != MAX_WORK)
884 return tp;
885
886 /* push the suffix length to the parent node */
887 if (tn->slen > tn->pos) {
888 unsigned char slen = update_suffix(tn);
889
890 if (slen > tp->slen)
891 tp->slen = slen;
892 }
893
894 return tp;
895 }
896
897 static void leaf_pull_suffix(struct key_vector *tp, struct key_vector *l)
898 {
899 while ((tp->slen > tp->pos) && (tp->slen > l->slen)) {
900 if (update_suffix(tp) > l->slen)
901 break;
902 tp = node_parent(tp);
903 }
904 }
905
906 static void leaf_push_suffix(struct key_vector *tn, struct key_vector *l)
907 {
908 /* if this is a new leaf then tn will be NULL and we can sort
909 * out parent suffix lengths as a part of trie_rebalance
910 */
911 while (tn->slen < l->slen) {
912 tn->slen = l->slen;
913 tn = node_parent(tn);
914 }
915 }
916
917 /* rcu_read_lock needs to be hold by caller from readside */
918 static struct key_vector *fib_find_node(struct trie *t,
919 struct key_vector **tp, u32 key)
920 {
921 struct key_vector *pn, *n = t->kv;
922 unsigned long index = 0;
923
924 do {
925 pn = n;
926 n = get_child_rcu(n, index);
927
928 if (!n)
929 break;
930
931 index = get_cindex(key, n);
932
933 /* This bit of code is a bit tricky but it combines multiple
934 * checks into a single check. The prefix consists of the
935 * prefix plus zeros for the bits in the cindex. The index
936 * is the difference between the key and this value. From
937 * this we can actually derive several pieces of data.
938 * if (index >= (1ul << bits))
939 * we have a mismatch in skip bits and failed
940 * else
941 * we know the value is cindex
942 *
943 * This check is safe even if bits == KEYLENGTH due to the
944 * fact that we can only allocate a node with 32 bits if a
945 * long is greater than 32 bits.
946 */
947 if (index >= (1ul << n->bits)) {
948 n = NULL;
949 break;
950 }
951
952 /* keep searching until we find a perfect match leaf or NULL */
953 } while (IS_TNODE(n));
954
955 *tp = pn;
956
957 return n;
958 }
959
960 /* Return the first fib alias matching TOS with
961 * priority less than or equal to PRIO.
962 */
963 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
964 u8 tos, u32 prio, u32 tb_id)
965 {
966 struct fib_alias *fa;
967
968 if (!fah)
969 return NULL;
970
971 hlist_for_each_entry(fa, fah, fa_list) {
972 if (fa->fa_slen < slen)
973 continue;
974 if (fa->fa_slen != slen)
975 break;
976 if (fa->tb_id > tb_id)
977 continue;
978 if (fa->tb_id != tb_id)
979 break;
980 if (fa->fa_tos > tos)
981 continue;
982 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
983 return fa;
984 }
985
986 return NULL;
987 }
988
989 static void trie_rebalance(struct trie *t, struct key_vector *tn)
990 {
991 while (!IS_TRIE(tn))
992 tn = resize(t, tn);
993 }
994
995 static int fib_insert_node(struct trie *t, struct key_vector *tp,
996 struct fib_alias *new, t_key key)
997 {
998 struct key_vector *n, *l;
999
1000 l = leaf_new(key, new);
1001 if (!l)
1002 goto noleaf;
1003
1004 /* retrieve child from parent node */
1005 n = get_child(tp, get_index(key, tp));
1006
1007 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1008 *
1009 * Add a new tnode here
1010 * first tnode need some special handling
1011 * leaves us in position for handling as case 3
1012 */
1013 if (n) {
1014 struct key_vector *tn;
1015
1016 tn = tnode_new(key, __fls(key ^ n->key), 1);
1017 if (!tn)
1018 goto notnode;
1019
1020 /* initialize routes out of node */
1021 NODE_INIT_PARENT(tn, tp);
1022 put_child(tn, get_index(key, tn) ^ 1, n);
1023
1024 /* start adding routes into the node */
1025 put_child_root(tp, key, tn);
1026 node_set_parent(n, tn);
1027
1028 /* parent now has a NULL spot where the leaf can go */
1029 tp = tn;
1030 }
1031
1032 /* Case 3: n is NULL, and will just insert a new leaf */
1033 NODE_INIT_PARENT(l, tp);
1034 put_child_root(tp, key, l);
1035 trie_rebalance(t, tp);
1036
1037 return 0;
1038 notnode:
1039 node_free(l);
1040 noleaf:
1041 return -ENOMEM;
1042 }
1043
1044 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1045 struct key_vector *l, struct fib_alias *new,
1046 struct fib_alias *fa, t_key key)
1047 {
1048 if (!l)
1049 return fib_insert_node(t, tp, new, key);
1050
1051 if (fa) {
1052 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1053 } else {
1054 struct fib_alias *last;
1055
1056 hlist_for_each_entry(last, &l->leaf, fa_list) {
1057 if (new->fa_slen < last->fa_slen)
1058 break;
1059 if ((new->fa_slen == last->fa_slen) &&
1060 (new->tb_id > last->tb_id))
1061 break;
1062 fa = last;
1063 }
1064
1065 if (fa)
1066 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1067 else
1068 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1069 }
1070
1071 /* if we added to the tail node then we need to update slen */
1072 if (l->slen < new->fa_slen) {
1073 l->slen = new->fa_slen;
1074 leaf_push_suffix(tp, l);
1075 }
1076
1077 return 0;
1078 }
1079
1080 /* Caller must hold RTNL. */
1081 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1082 {
1083 struct trie *t = (struct trie *)tb->tb_data;
1084 struct fib_alias *fa, *new_fa;
1085 struct key_vector *l, *tp;
1086 unsigned int nlflags = 0;
1087 struct fib_info *fi;
1088 u8 plen = cfg->fc_dst_len;
1089 u8 slen = KEYLENGTH - plen;
1090 u8 tos = cfg->fc_tos;
1091 u32 key;
1092 int err;
1093
1094 if (plen > KEYLENGTH)
1095 return -EINVAL;
1096
1097 key = ntohl(cfg->fc_dst);
1098
1099 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1100
1101 if ((plen < KEYLENGTH) && (key << plen))
1102 return -EINVAL;
1103
1104 fi = fib_create_info(cfg);
1105 if (IS_ERR(fi)) {
1106 err = PTR_ERR(fi);
1107 goto err;
1108 }
1109
1110 l = fib_find_node(t, &tp, key);
1111 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1112 tb->tb_id) : NULL;
1113
1114 /* Now fa, if non-NULL, points to the first fib alias
1115 * with the same keys [prefix,tos,priority], if such key already
1116 * exists or to the node before which we will insert new one.
1117 *
1118 * If fa is NULL, we will need to allocate a new one and
1119 * insert to the tail of the section matching the suffix length
1120 * of the new alias.
1121 */
1122
1123 if (fa && fa->fa_tos == tos &&
1124 fa->fa_info->fib_priority == fi->fib_priority) {
1125 struct fib_alias *fa_first, *fa_match;
1126
1127 err = -EEXIST;
1128 if (cfg->fc_nlflags & NLM_F_EXCL)
1129 goto out;
1130
1131 /* We have 2 goals:
1132 * 1. Find exact match for type, scope, fib_info to avoid
1133 * duplicate routes
1134 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1135 */
1136 fa_match = NULL;
1137 fa_first = fa;
1138 hlist_for_each_entry_from(fa, fa_list) {
1139 if ((fa->fa_slen != slen) ||
1140 (fa->tb_id != tb->tb_id) ||
1141 (fa->fa_tos != tos))
1142 break;
1143 if (fa->fa_info->fib_priority != fi->fib_priority)
1144 break;
1145 if (fa->fa_type == cfg->fc_type &&
1146 fa->fa_info == fi) {
1147 fa_match = fa;
1148 break;
1149 }
1150 }
1151
1152 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1153 struct fib_info *fi_drop;
1154 u8 state;
1155
1156 fa = fa_first;
1157 if (fa_match) {
1158 if (fa == fa_match)
1159 err = 0;
1160 goto out;
1161 }
1162 err = -ENOBUFS;
1163 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1164 if (!new_fa)
1165 goto out;
1166
1167 fi_drop = fa->fa_info;
1168 new_fa->fa_tos = fa->fa_tos;
1169 new_fa->fa_info = fi;
1170 new_fa->fa_type = cfg->fc_type;
1171 state = fa->fa_state;
1172 new_fa->fa_state = state & ~FA_S_ACCESSED;
1173 new_fa->fa_slen = fa->fa_slen;
1174 new_fa->tb_id = tb->tb_id;
1175 new_fa->fa_default = -1;
1176
1177 err = switchdev_fib_ipv4_add(key, plen, fi,
1178 new_fa->fa_tos,
1179 cfg->fc_type,
1180 cfg->fc_nlflags,
1181 tb->tb_id);
1182 if (err) {
1183 switchdev_fib_ipv4_abort(fi);
1184 kmem_cache_free(fn_alias_kmem, new_fa);
1185 goto out;
1186 }
1187
1188 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1189
1190 alias_free_mem_rcu(fa);
1191
1192 fib_release_info(fi_drop);
1193 if (state & FA_S_ACCESSED)
1194 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1195 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1196 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1197
1198 goto succeeded;
1199 }
1200 /* Error if we find a perfect match which
1201 * uses the same scope, type, and nexthop
1202 * information.
1203 */
1204 if (fa_match)
1205 goto out;
1206
1207 if (cfg->fc_nlflags & NLM_F_APPEND)
1208 nlflags = NLM_F_APPEND;
1209 else
1210 fa = fa_first;
1211 }
1212 err = -ENOENT;
1213 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1214 goto out;
1215
1216 err = -ENOBUFS;
1217 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1218 if (!new_fa)
1219 goto out;
1220
1221 new_fa->fa_info = fi;
1222 new_fa->fa_tos = tos;
1223 new_fa->fa_type = cfg->fc_type;
1224 new_fa->fa_state = 0;
1225 new_fa->fa_slen = slen;
1226 new_fa->tb_id = tb->tb_id;
1227 new_fa->fa_default = -1;
1228
1229 /* (Optionally) offload fib entry to switch hardware. */
1230 err = switchdev_fib_ipv4_add(key, plen, fi, tos, cfg->fc_type,
1231 cfg->fc_nlflags, tb->tb_id);
1232 if (err) {
1233 switchdev_fib_ipv4_abort(fi);
1234 goto out_free_new_fa;
1235 }
1236
1237 /* Insert new entry to the list. */
1238 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1239 if (err)
1240 goto out_sw_fib_del;
1241
1242 if (!plen)
1243 tb->tb_num_default++;
1244
1245 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1246 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1247 &cfg->fc_nlinfo, nlflags);
1248 succeeded:
1249 return 0;
1250
1251 out_sw_fib_del:
1252 switchdev_fib_ipv4_del(key, plen, fi, tos, cfg->fc_type, tb->tb_id);
1253 out_free_new_fa:
1254 kmem_cache_free(fn_alias_kmem, new_fa);
1255 out:
1256 fib_release_info(fi);
1257 err:
1258 return err;
1259 }
1260
1261 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1262 {
1263 t_key prefix = n->key;
1264
1265 return (key ^ prefix) & (prefix | -prefix);
1266 }
1267
1268 /* should be called with rcu_read_lock */
1269 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1270 struct fib_result *res, int fib_flags)
1271 {
1272 struct trie *t = (struct trie *) tb->tb_data;
1273 #ifdef CONFIG_IP_FIB_TRIE_STATS
1274 struct trie_use_stats __percpu *stats = t->stats;
1275 #endif
1276 const t_key key = ntohl(flp->daddr);
1277 struct key_vector *n, *pn;
1278 struct fib_alias *fa;
1279 unsigned long index;
1280 t_key cindex;
1281
1282 trace_fib_table_lookup(tb->tb_id, flp);
1283
1284 pn = t->kv;
1285 cindex = 0;
1286
1287 n = get_child_rcu(pn, cindex);
1288 if (!n)
1289 return -EAGAIN;
1290
1291 #ifdef CONFIG_IP_FIB_TRIE_STATS
1292 this_cpu_inc(stats->gets);
1293 #endif
1294
1295 /* Step 1: Travel to the longest prefix match in the trie */
1296 for (;;) {
1297 index = get_cindex(key, n);
1298
1299 /* This bit of code is a bit tricky but it combines multiple
1300 * checks into a single check. The prefix consists of the
1301 * prefix plus zeros for the "bits" in the prefix. The index
1302 * is the difference between the key and this value. From
1303 * this we can actually derive several pieces of data.
1304 * if (index >= (1ul << bits))
1305 * we have a mismatch in skip bits and failed
1306 * else
1307 * we know the value is cindex
1308 *
1309 * This check is safe even if bits == KEYLENGTH due to the
1310 * fact that we can only allocate a node with 32 bits if a
1311 * long is greater than 32 bits.
1312 */
1313 if (index >= (1ul << n->bits))
1314 break;
1315
1316 /* we have found a leaf. Prefixes have already been compared */
1317 if (IS_LEAF(n))
1318 goto found;
1319
1320 /* only record pn and cindex if we are going to be chopping
1321 * bits later. Otherwise we are just wasting cycles.
1322 */
1323 if (n->slen > n->pos) {
1324 pn = n;
1325 cindex = index;
1326 }
1327
1328 n = get_child_rcu(n, index);
1329 if (unlikely(!n))
1330 goto backtrace;
1331 }
1332
1333 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1334 for (;;) {
1335 /* record the pointer where our next node pointer is stored */
1336 struct key_vector __rcu **cptr = n->tnode;
1337
1338 /* This test verifies that none of the bits that differ
1339 * between the key and the prefix exist in the region of
1340 * the lsb and higher in the prefix.
1341 */
1342 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1343 goto backtrace;
1344
1345 /* exit out and process leaf */
1346 if (unlikely(IS_LEAF(n)))
1347 break;
1348
1349 /* Don't bother recording parent info. Since we are in
1350 * prefix match mode we will have to come back to wherever
1351 * we started this traversal anyway
1352 */
1353
1354 while ((n = rcu_dereference(*cptr)) == NULL) {
1355 backtrace:
1356 #ifdef CONFIG_IP_FIB_TRIE_STATS
1357 if (!n)
1358 this_cpu_inc(stats->null_node_hit);
1359 #endif
1360 /* If we are at cindex 0 there are no more bits for
1361 * us to strip at this level so we must ascend back
1362 * up one level to see if there are any more bits to
1363 * be stripped there.
1364 */
1365 while (!cindex) {
1366 t_key pkey = pn->key;
1367
1368 /* If we don't have a parent then there is
1369 * nothing for us to do as we do not have any
1370 * further nodes to parse.
1371 */
1372 if (IS_TRIE(pn))
1373 return -EAGAIN;
1374 #ifdef CONFIG_IP_FIB_TRIE_STATS
1375 this_cpu_inc(stats->backtrack);
1376 #endif
1377 /* Get Child's index */
1378 pn = node_parent_rcu(pn);
1379 cindex = get_index(pkey, pn);
1380 }
1381
1382 /* strip the least significant bit from the cindex */
1383 cindex &= cindex - 1;
1384
1385 /* grab pointer for next child node */
1386 cptr = &pn->tnode[cindex];
1387 }
1388 }
1389
1390 found:
1391 /* this line carries forward the xor from earlier in the function */
1392 index = key ^ n->key;
1393
1394 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1395 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1396 struct fib_info *fi = fa->fa_info;
1397 int nhsel, err;
1398
1399 if ((index >= (1ul << fa->fa_slen)) &&
1400 ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen != KEYLENGTH)))
1401 continue;
1402 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1403 continue;
1404 if (fi->fib_dead)
1405 continue;
1406 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1407 continue;
1408 fib_alias_accessed(fa);
1409 err = fib_props[fa->fa_type].error;
1410 if (unlikely(err < 0)) {
1411 #ifdef CONFIG_IP_FIB_TRIE_STATS
1412 this_cpu_inc(stats->semantic_match_passed);
1413 #endif
1414 return err;
1415 }
1416 if (fi->fib_flags & RTNH_F_DEAD)
1417 continue;
1418 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1419 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1420 struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1421
1422 if (nh->nh_flags & RTNH_F_DEAD)
1423 continue;
1424 if (in_dev &&
1425 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1426 nh->nh_flags & RTNH_F_LINKDOWN &&
1427 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1428 continue;
1429 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1430 if (flp->flowi4_oif &&
1431 flp->flowi4_oif != nh->nh_oif)
1432 continue;
1433 }
1434
1435 if (!(fib_flags & FIB_LOOKUP_NOREF))
1436 atomic_inc(&fi->fib_clntref);
1437
1438 res->prefixlen = KEYLENGTH - fa->fa_slen;
1439 res->nh_sel = nhsel;
1440 res->type = fa->fa_type;
1441 res->scope = fi->fib_scope;
1442 res->fi = fi;
1443 res->table = tb;
1444 res->fa_head = &n->leaf;
1445 #ifdef CONFIG_IP_FIB_TRIE_STATS
1446 this_cpu_inc(stats->semantic_match_passed);
1447 #endif
1448 trace_fib_table_lookup_nh(nh);
1449
1450 return err;
1451 }
1452 }
1453 #ifdef CONFIG_IP_FIB_TRIE_STATS
1454 this_cpu_inc(stats->semantic_match_miss);
1455 #endif
1456 goto backtrace;
1457 }
1458 EXPORT_SYMBOL_GPL(fib_table_lookup);
1459
1460 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1461 struct key_vector *l, struct fib_alias *old)
1462 {
1463 /* record the location of the previous list_info entry */
1464 struct hlist_node **pprev = old->fa_list.pprev;
1465 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1466
1467 /* remove the fib_alias from the list */
1468 hlist_del_rcu(&old->fa_list);
1469
1470 /* if we emptied the list this leaf will be freed and we can sort
1471 * out parent suffix lengths as a part of trie_rebalance
1472 */
1473 if (hlist_empty(&l->leaf)) {
1474 put_child_root(tp, l->key, NULL);
1475 node_free(l);
1476 trie_rebalance(t, tp);
1477 return;
1478 }
1479
1480 /* only access fa if it is pointing at the last valid hlist_node */
1481 if (*pprev)
1482 return;
1483
1484 /* update the trie with the latest suffix length */
1485 l->slen = fa->fa_slen;
1486 leaf_pull_suffix(tp, l);
1487 }
1488
1489 /* Caller must hold RTNL. */
1490 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1491 {
1492 struct trie *t = (struct trie *) tb->tb_data;
1493 struct fib_alias *fa, *fa_to_delete;
1494 struct key_vector *l, *tp;
1495 u8 plen = cfg->fc_dst_len;
1496 u8 slen = KEYLENGTH - plen;
1497 u8 tos = cfg->fc_tos;
1498 u32 key;
1499
1500 if (plen > KEYLENGTH)
1501 return -EINVAL;
1502
1503 key = ntohl(cfg->fc_dst);
1504
1505 if ((plen < KEYLENGTH) && (key << plen))
1506 return -EINVAL;
1507
1508 l = fib_find_node(t, &tp, key);
1509 if (!l)
1510 return -ESRCH;
1511
1512 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1513 if (!fa)
1514 return -ESRCH;
1515
1516 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1517
1518 fa_to_delete = NULL;
1519 hlist_for_each_entry_from(fa, fa_list) {
1520 struct fib_info *fi = fa->fa_info;
1521
1522 if ((fa->fa_slen != slen) ||
1523 (fa->tb_id != tb->tb_id) ||
1524 (fa->fa_tos != tos))
1525 break;
1526
1527 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1528 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1529 fa->fa_info->fib_scope == cfg->fc_scope) &&
1530 (!cfg->fc_prefsrc ||
1531 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1532 (!cfg->fc_protocol ||
1533 fi->fib_protocol == cfg->fc_protocol) &&
1534 fib_nh_match(cfg, fi) == 0) {
1535 fa_to_delete = fa;
1536 break;
1537 }
1538 }
1539
1540 if (!fa_to_delete)
1541 return -ESRCH;
1542
1543 switchdev_fib_ipv4_del(key, plen, fa_to_delete->fa_info, tos,
1544 cfg->fc_type, tb->tb_id);
1545
1546 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1547 &cfg->fc_nlinfo, 0);
1548
1549 if (!plen)
1550 tb->tb_num_default--;
1551
1552 fib_remove_alias(t, tp, l, fa_to_delete);
1553
1554 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1555 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1556
1557 fib_release_info(fa_to_delete->fa_info);
1558 alias_free_mem_rcu(fa_to_delete);
1559 return 0;
1560 }
1561
1562 /* Scan for the next leaf starting at the provided key value */
1563 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1564 {
1565 struct key_vector *pn, *n = *tn;
1566 unsigned long cindex;
1567
1568 /* this loop is meant to try and find the key in the trie */
1569 do {
1570 /* record parent and next child index */
1571 pn = n;
1572 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1573
1574 if (cindex >> pn->bits)
1575 break;
1576
1577 /* descend into the next child */
1578 n = get_child_rcu(pn, cindex++);
1579 if (!n)
1580 break;
1581
1582 /* guarantee forward progress on the keys */
1583 if (IS_LEAF(n) && (n->key >= key))
1584 goto found;
1585 } while (IS_TNODE(n));
1586
1587 /* this loop will search for the next leaf with a greater key */
1588 while (!IS_TRIE(pn)) {
1589 /* if we exhausted the parent node we will need to climb */
1590 if (cindex >= (1ul << pn->bits)) {
1591 t_key pkey = pn->key;
1592
1593 pn = node_parent_rcu(pn);
1594 cindex = get_index(pkey, pn) + 1;
1595 continue;
1596 }
1597
1598 /* grab the next available node */
1599 n = get_child_rcu(pn, cindex++);
1600 if (!n)
1601 continue;
1602
1603 /* no need to compare keys since we bumped the index */
1604 if (IS_LEAF(n))
1605 goto found;
1606
1607 /* Rescan start scanning in new node */
1608 pn = n;
1609 cindex = 0;
1610 }
1611
1612 *tn = pn;
1613 return NULL; /* Root of trie */
1614 found:
1615 /* if we are at the limit for keys just return NULL for the tnode */
1616 *tn = pn;
1617 return n;
1618 }
1619
1620 static void fib_trie_free(struct fib_table *tb)
1621 {
1622 struct trie *t = (struct trie *)tb->tb_data;
1623 struct key_vector *pn = t->kv;
1624 unsigned long cindex = 1;
1625 struct hlist_node *tmp;
1626 struct fib_alias *fa;
1627
1628 /* walk trie in reverse order and free everything */
1629 for (;;) {
1630 struct key_vector *n;
1631
1632 if (!(cindex--)) {
1633 t_key pkey = pn->key;
1634
1635 if (IS_TRIE(pn))
1636 break;
1637
1638 n = pn;
1639 pn = node_parent(pn);
1640
1641 /* drop emptied tnode */
1642 put_child_root(pn, n->key, NULL);
1643 node_free(n);
1644
1645 cindex = get_index(pkey, pn);
1646
1647 continue;
1648 }
1649
1650 /* grab the next available node */
1651 n = get_child(pn, cindex);
1652 if (!n)
1653 continue;
1654
1655 if (IS_TNODE(n)) {
1656 /* record pn and cindex for leaf walking */
1657 pn = n;
1658 cindex = 1ul << n->bits;
1659
1660 continue;
1661 }
1662
1663 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1664 hlist_del_rcu(&fa->fa_list);
1665 alias_free_mem_rcu(fa);
1666 }
1667
1668 put_child_root(pn, n->key, NULL);
1669 node_free(n);
1670 }
1671
1672 #ifdef CONFIG_IP_FIB_TRIE_STATS
1673 free_percpu(t->stats);
1674 #endif
1675 kfree(tb);
1676 }
1677
1678 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1679 {
1680 struct trie *ot = (struct trie *)oldtb->tb_data;
1681 struct key_vector *l, *tp = ot->kv;
1682 struct fib_table *local_tb;
1683 struct fib_alias *fa;
1684 struct trie *lt;
1685 t_key key = 0;
1686
1687 if (oldtb->tb_data == oldtb->__data)
1688 return oldtb;
1689
1690 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1691 if (!local_tb)
1692 return NULL;
1693
1694 lt = (struct trie *)local_tb->tb_data;
1695
1696 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1697 struct key_vector *local_l = NULL, *local_tp;
1698
1699 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1700 struct fib_alias *new_fa;
1701
1702 if (local_tb->tb_id != fa->tb_id)
1703 continue;
1704
1705 /* clone fa for new local table */
1706 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1707 if (!new_fa)
1708 goto out;
1709
1710 memcpy(new_fa, fa, sizeof(*fa));
1711
1712 /* insert clone into table */
1713 if (!local_l)
1714 local_l = fib_find_node(lt, &local_tp, l->key);
1715
1716 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1717 NULL, l->key))
1718 goto out;
1719 }
1720
1721 /* stop loop if key wrapped back to 0 */
1722 key = l->key + 1;
1723 if (key < l->key)
1724 break;
1725 }
1726
1727 return local_tb;
1728 out:
1729 fib_trie_free(local_tb);
1730
1731 return NULL;
1732 }
1733
1734 /* Caller must hold RTNL */
1735 void fib_table_flush_external(struct fib_table *tb)
1736 {
1737 struct trie *t = (struct trie *)tb->tb_data;
1738 struct key_vector *pn = t->kv;
1739 unsigned long cindex = 1;
1740 struct hlist_node *tmp;
1741 struct fib_alias *fa;
1742
1743 /* walk trie in reverse order */
1744 for (;;) {
1745 unsigned char slen = 0;
1746 struct key_vector *n;
1747
1748 if (!(cindex--)) {
1749 t_key pkey = pn->key;
1750
1751 /* cannot resize the trie vector */
1752 if (IS_TRIE(pn))
1753 break;
1754
1755 /* resize completed node */
1756 pn = resize(t, pn);
1757 cindex = get_index(pkey, pn);
1758
1759 continue;
1760 }
1761
1762 /* grab the next available node */
1763 n = get_child(pn, cindex);
1764 if (!n)
1765 continue;
1766
1767 if (IS_TNODE(n)) {
1768 /* record pn and cindex for leaf walking */
1769 pn = n;
1770 cindex = 1ul << n->bits;
1771
1772 continue;
1773 }
1774
1775 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1776 struct fib_info *fi = fa->fa_info;
1777
1778 /* if alias was cloned to local then we just
1779 * need to remove the local copy from main
1780 */
1781 if (tb->tb_id != fa->tb_id) {
1782 hlist_del_rcu(&fa->fa_list);
1783 alias_free_mem_rcu(fa);
1784 continue;
1785 }
1786
1787 /* record local slen */
1788 slen = fa->fa_slen;
1789
1790 if (!fi || !(fi->fib_flags & RTNH_F_OFFLOAD))
1791 continue;
1792
1793 switchdev_fib_ipv4_del(n->key, KEYLENGTH - fa->fa_slen,
1794 fi, fa->fa_tos, fa->fa_type,
1795 tb->tb_id);
1796 }
1797
1798 /* update leaf slen */
1799 n->slen = slen;
1800
1801 if (hlist_empty(&n->leaf)) {
1802 put_child_root(pn, n->key, NULL);
1803 node_free(n);
1804 }
1805 }
1806 }
1807
1808 /* Caller must hold RTNL. */
1809 int fib_table_flush(struct fib_table *tb)
1810 {
1811 struct trie *t = (struct trie *)tb->tb_data;
1812 struct key_vector *pn = t->kv;
1813 unsigned long cindex = 1;
1814 struct hlist_node *tmp;
1815 struct fib_alias *fa;
1816 int found = 0;
1817
1818 /* walk trie in reverse order */
1819 for (;;) {
1820 unsigned char slen = 0;
1821 struct key_vector *n;
1822
1823 if (!(cindex--)) {
1824 t_key pkey = pn->key;
1825
1826 /* cannot resize the trie vector */
1827 if (IS_TRIE(pn))
1828 break;
1829
1830 /* resize completed node */
1831 pn = resize(t, pn);
1832 cindex = get_index(pkey, pn);
1833
1834 continue;
1835 }
1836
1837 /* grab the next available node */
1838 n = get_child(pn, cindex);
1839 if (!n)
1840 continue;
1841
1842 if (IS_TNODE(n)) {
1843 /* record pn and cindex for leaf walking */
1844 pn = n;
1845 cindex = 1ul << n->bits;
1846
1847 continue;
1848 }
1849
1850 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1851 struct fib_info *fi = fa->fa_info;
1852
1853 if (!fi || !(fi->fib_flags & RTNH_F_DEAD)) {
1854 slen = fa->fa_slen;
1855 continue;
1856 }
1857
1858 switchdev_fib_ipv4_del(n->key, KEYLENGTH - fa->fa_slen,
1859 fi, fa->fa_tos, fa->fa_type,
1860 tb->tb_id);
1861 hlist_del_rcu(&fa->fa_list);
1862 fib_release_info(fa->fa_info);
1863 alias_free_mem_rcu(fa);
1864 found++;
1865 }
1866
1867 /* update leaf slen */
1868 n->slen = slen;
1869
1870 if (hlist_empty(&n->leaf)) {
1871 put_child_root(pn, n->key, NULL);
1872 node_free(n);
1873 }
1874 }
1875
1876 pr_debug("trie_flush found=%d\n", found);
1877 return found;
1878 }
1879
1880 static void __trie_free_rcu(struct rcu_head *head)
1881 {
1882 struct fib_table *tb = container_of(head, struct fib_table, rcu);
1883 #ifdef CONFIG_IP_FIB_TRIE_STATS
1884 struct trie *t = (struct trie *)tb->tb_data;
1885
1886 if (tb->tb_data == tb->__data)
1887 free_percpu(t->stats);
1888 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1889 kfree(tb);
1890 }
1891
1892 void fib_free_table(struct fib_table *tb)
1893 {
1894 call_rcu(&tb->rcu, __trie_free_rcu);
1895 }
1896
1897 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
1898 struct sk_buff *skb, struct netlink_callback *cb)
1899 {
1900 __be32 xkey = htonl(l->key);
1901 struct fib_alias *fa;
1902 int i, s_i;
1903
1904 s_i = cb->args[4];
1905 i = 0;
1906
1907 /* rcu_read_lock is hold by caller */
1908 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1909 if (i < s_i) {
1910 i++;
1911 continue;
1912 }
1913
1914 if (tb->tb_id != fa->tb_id) {
1915 i++;
1916 continue;
1917 }
1918
1919 if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
1920 cb->nlh->nlmsg_seq,
1921 RTM_NEWROUTE,
1922 tb->tb_id,
1923 fa->fa_type,
1924 xkey,
1925 KEYLENGTH - fa->fa_slen,
1926 fa->fa_tos,
1927 fa->fa_info, NLM_F_MULTI) < 0) {
1928 cb->args[4] = i;
1929 return -1;
1930 }
1931 i++;
1932 }
1933
1934 cb->args[4] = i;
1935 return skb->len;
1936 }
1937
1938 /* rcu_read_lock needs to be hold by caller from readside */
1939 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1940 struct netlink_callback *cb)
1941 {
1942 struct trie *t = (struct trie *)tb->tb_data;
1943 struct key_vector *l, *tp = t->kv;
1944 /* Dump starting at last key.
1945 * Note: 0.0.0.0/0 (ie default) is first key.
1946 */
1947 int count = cb->args[2];
1948 t_key key = cb->args[3];
1949
1950 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1951 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1952 cb->args[3] = key;
1953 cb->args[2] = count;
1954 return -1;
1955 }
1956
1957 ++count;
1958 key = l->key + 1;
1959
1960 memset(&cb->args[4], 0,
1961 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1962
1963 /* stop loop if key wrapped back to 0 */
1964 if (key < l->key)
1965 break;
1966 }
1967
1968 cb->args[3] = key;
1969 cb->args[2] = count;
1970
1971 return skb->len;
1972 }
1973
1974 void __init fib_trie_init(void)
1975 {
1976 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1977 sizeof(struct fib_alias),
1978 0, SLAB_PANIC, NULL);
1979
1980 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1981 LEAF_SIZE,
1982 0, SLAB_PANIC, NULL);
1983 }
1984
1985 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
1986 {
1987 struct fib_table *tb;
1988 struct trie *t;
1989 size_t sz = sizeof(*tb);
1990
1991 if (!alias)
1992 sz += sizeof(struct trie);
1993
1994 tb = kzalloc(sz, GFP_KERNEL);
1995 if (!tb)
1996 return NULL;
1997
1998 tb->tb_id = id;
1999 tb->tb_num_default = 0;
2000 tb->tb_data = (alias ? alias->__data : tb->__data);
2001
2002 if (alias)
2003 return tb;
2004
2005 t = (struct trie *) tb->tb_data;
2006 t->kv[0].pos = KEYLENGTH;
2007 t->kv[0].slen = KEYLENGTH;
2008 #ifdef CONFIG_IP_FIB_TRIE_STATS
2009 t->stats = alloc_percpu(struct trie_use_stats);
2010 if (!t->stats) {
2011 kfree(tb);
2012 tb = NULL;
2013 }
2014 #endif
2015
2016 return tb;
2017 }
2018
2019 #ifdef CONFIG_PROC_FS
2020 /* Depth first Trie walk iterator */
2021 struct fib_trie_iter {
2022 struct seq_net_private p;
2023 struct fib_table *tb;
2024 struct key_vector *tnode;
2025 unsigned int index;
2026 unsigned int depth;
2027 };
2028
2029 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2030 {
2031 unsigned long cindex = iter->index;
2032 struct key_vector *pn = iter->tnode;
2033 t_key pkey;
2034
2035 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2036 iter->tnode, iter->index, iter->depth);
2037
2038 while (!IS_TRIE(pn)) {
2039 while (cindex < child_length(pn)) {
2040 struct key_vector *n = get_child_rcu(pn, cindex++);
2041
2042 if (!n)
2043 continue;
2044
2045 if (IS_LEAF(n)) {
2046 iter->tnode = pn;
2047 iter->index = cindex;
2048 } else {
2049 /* push down one level */
2050 iter->tnode = n;
2051 iter->index = 0;
2052 ++iter->depth;
2053 }
2054
2055 return n;
2056 }
2057
2058 /* Current node exhausted, pop back up */
2059 pkey = pn->key;
2060 pn = node_parent_rcu(pn);
2061 cindex = get_index(pkey, pn) + 1;
2062 --iter->depth;
2063 }
2064
2065 /* record root node so further searches know we are done */
2066 iter->tnode = pn;
2067 iter->index = 0;
2068
2069 return NULL;
2070 }
2071
2072 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2073 struct trie *t)
2074 {
2075 struct key_vector *n, *pn;
2076
2077 if (!t)
2078 return NULL;
2079
2080 pn = t->kv;
2081 n = rcu_dereference(pn->tnode[0]);
2082 if (!n)
2083 return NULL;
2084
2085 if (IS_TNODE(n)) {
2086 iter->tnode = n;
2087 iter->index = 0;
2088 iter->depth = 1;
2089 } else {
2090 iter->tnode = pn;
2091 iter->index = 0;
2092 iter->depth = 0;
2093 }
2094
2095 return n;
2096 }
2097
2098 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2099 {
2100 struct key_vector *n;
2101 struct fib_trie_iter iter;
2102
2103 memset(s, 0, sizeof(*s));
2104
2105 rcu_read_lock();
2106 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2107 if (IS_LEAF(n)) {
2108 struct fib_alias *fa;
2109
2110 s->leaves++;
2111 s->totdepth += iter.depth;
2112 if (iter.depth > s->maxdepth)
2113 s->maxdepth = iter.depth;
2114
2115 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2116 ++s->prefixes;
2117 } else {
2118 s->tnodes++;
2119 if (n->bits < MAX_STAT_DEPTH)
2120 s->nodesizes[n->bits]++;
2121 s->nullpointers += tn_info(n)->empty_children;
2122 }
2123 }
2124 rcu_read_unlock();
2125 }
2126
2127 /*
2128 * This outputs /proc/net/fib_triestats
2129 */
2130 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2131 {
2132 unsigned int i, max, pointers, bytes, avdepth;
2133
2134 if (stat->leaves)
2135 avdepth = stat->totdepth*100 / stat->leaves;
2136 else
2137 avdepth = 0;
2138
2139 seq_printf(seq, "\tAver depth: %u.%02d\n",
2140 avdepth / 100, avdepth % 100);
2141 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2142
2143 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2144 bytes = LEAF_SIZE * stat->leaves;
2145
2146 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2147 bytes += sizeof(struct fib_alias) * stat->prefixes;
2148
2149 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2150 bytes += TNODE_SIZE(0) * stat->tnodes;
2151
2152 max = MAX_STAT_DEPTH;
2153 while (max > 0 && stat->nodesizes[max-1] == 0)
2154 max--;
2155
2156 pointers = 0;
2157 for (i = 1; i < max; i++)
2158 if (stat->nodesizes[i] != 0) {
2159 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2160 pointers += (1<<i) * stat->nodesizes[i];
2161 }
2162 seq_putc(seq, '\n');
2163 seq_printf(seq, "\tPointers: %u\n", pointers);
2164
2165 bytes += sizeof(struct key_vector *) * pointers;
2166 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2167 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2168 }
2169
2170 #ifdef CONFIG_IP_FIB_TRIE_STATS
2171 static void trie_show_usage(struct seq_file *seq,
2172 const struct trie_use_stats __percpu *stats)
2173 {
2174 struct trie_use_stats s = { 0 };
2175 int cpu;
2176
2177 /* loop through all of the CPUs and gather up the stats */
2178 for_each_possible_cpu(cpu) {
2179 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2180
2181 s.gets += pcpu->gets;
2182 s.backtrack += pcpu->backtrack;
2183 s.semantic_match_passed += pcpu->semantic_match_passed;
2184 s.semantic_match_miss += pcpu->semantic_match_miss;
2185 s.null_node_hit += pcpu->null_node_hit;
2186 s.resize_node_skipped += pcpu->resize_node_skipped;
2187 }
2188
2189 seq_printf(seq, "\nCounters:\n---------\n");
2190 seq_printf(seq, "gets = %u\n", s.gets);
2191 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2192 seq_printf(seq, "semantic match passed = %u\n",
2193 s.semantic_match_passed);
2194 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2195 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2196 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2197 }
2198 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2199
2200 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2201 {
2202 if (tb->tb_id == RT_TABLE_LOCAL)
2203 seq_puts(seq, "Local:\n");
2204 else if (tb->tb_id == RT_TABLE_MAIN)
2205 seq_puts(seq, "Main:\n");
2206 else
2207 seq_printf(seq, "Id %d:\n", tb->tb_id);
2208 }
2209
2210
2211 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2212 {
2213 struct net *net = (struct net *)seq->private;
2214 unsigned int h;
2215
2216 seq_printf(seq,
2217 "Basic info: size of leaf:"
2218 " %Zd bytes, size of tnode: %Zd bytes.\n",
2219 LEAF_SIZE, TNODE_SIZE(0));
2220
2221 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2222 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2223 struct fib_table *tb;
2224
2225 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2226 struct trie *t = (struct trie *) tb->tb_data;
2227 struct trie_stat stat;
2228
2229 if (!t)
2230 continue;
2231
2232 fib_table_print(seq, tb);
2233
2234 trie_collect_stats(t, &stat);
2235 trie_show_stats(seq, &stat);
2236 #ifdef CONFIG_IP_FIB_TRIE_STATS
2237 trie_show_usage(seq, t->stats);
2238 #endif
2239 }
2240 }
2241
2242 return 0;
2243 }
2244
2245 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2246 {
2247 return single_open_net(inode, file, fib_triestat_seq_show);
2248 }
2249
2250 static const struct file_operations fib_triestat_fops = {
2251 .owner = THIS_MODULE,
2252 .open = fib_triestat_seq_open,
2253 .read = seq_read,
2254 .llseek = seq_lseek,
2255 .release = single_release_net,
2256 };
2257
2258 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2259 {
2260 struct fib_trie_iter *iter = seq->private;
2261 struct net *net = seq_file_net(seq);
2262 loff_t idx = 0;
2263 unsigned int h;
2264
2265 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2266 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2267 struct fib_table *tb;
2268
2269 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2270 struct key_vector *n;
2271
2272 for (n = fib_trie_get_first(iter,
2273 (struct trie *) tb->tb_data);
2274 n; n = fib_trie_get_next(iter))
2275 if (pos == idx++) {
2276 iter->tb = tb;
2277 return n;
2278 }
2279 }
2280 }
2281
2282 return NULL;
2283 }
2284
2285 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2286 __acquires(RCU)
2287 {
2288 rcu_read_lock();
2289 return fib_trie_get_idx(seq, *pos);
2290 }
2291
2292 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2293 {
2294 struct fib_trie_iter *iter = seq->private;
2295 struct net *net = seq_file_net(seq);
2296 struct fib_table *tb = iter->tb;
2297 struct hlist_node *tb_node;
2298 unsigned int h;
2299 struct key_vector *n;
2300
2301 ++*pos;
2302 /* next node in same table */
2303 n = fib_trie_get_next(iter);
2304 if (n)
2305 return n;
2306
2307 /* walk rest of this hash chain */
2308 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2309 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2310 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2311 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2312 if (n)
2313 goto found;
2314 }
2315
2316 /* new hash chain */
2317 while (++h < FIB_TABLE_HASHSZ) {
2318 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2319 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2320 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2321 if (n)
2322 goto found;
2323 }
2324 }
2325 return NULL;
2326
2327 found:
2328 iter->tb = tb;
2329 return n;
2330 }
2331
2332 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2333 __releases(RCU)
2334 {
2335 rcu_read_unlock();
2336 }
2337
2338 static void seq_indent(struct seq_file *seq, int n)
2339 {
2340 while (n-- > 0)
2341 seq_puts(seq, " ");
2342 }
2343
2344 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2345 {
2346 switch (s) {
2347 case RT_SCOPE_UNIVERSE: return "universe";
2348 case RT_SCOPE_SITE: return "site";
2349 case RT_SCOPE_LINK: return "link";
2350 case RT_SCOPE_HOST: return "host";
2351 case RT_SCOPE_NOWHERE: return "nowhere";
2352 default:
2353 snprintf(buf, len, "scope=%d", s);
2354 return buf;
2355 }
2356 }
2357
2358 static const char *const rtn_type_names[__RTN_MAX] = {
2359 [RTN_UNSPEC] = "UNSPEC",
2360 [RTN_UNICAST] = "UNICAST",
2361 [RTN_LOCAL] = "LOCAL",
2362 [RTN_BROADCAST] = "BROADCAST",
2363 [RTN_ANYCAST] = "ANYCAST",
2364 [RTN_MULTICAST] = "MULTICAST",
2365 [RTN_BLACKHOLE] = "BLACKHOLE",
2366 [RTN_UNREACHABLE] = "UNREACHABLE",
2367 [RTN_PROHIBIT] = "PROHIBIT",
2368 [RTN_THROW] = "THROW",
2369 [RTN_NAT] = "NAT",
2370 [RTN_XRESOLVE] = "XRESOLVE",
2371 };
2372
2373 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2374 {
2375 if (t < __RTN_MAX && rtn_type_names[t])
2376 return rtn_type_names[t];
2377 snprintf(buf, len, "type %u", t);
2378 return buf;
2379 }
2380
2381 /* Pretty print the trie */
2382 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2383 {
2384 const struct fib_trie_iter *iter = seq->private;
2385 struct key_vector *n = v;
2386
2387 if (IS_TRIE(node_parent_rcu(n)))
2388 fib_table_print(seq, iter->tb);
2389
2390 if (IS_TNODE(n)) {
2391 __be32 prf = htonl(n->key);
2392
2393 seq_indent(seq, iter->depth-1);
2394 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2395 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2396 tn_info(n)->full_children,
2397 tn_info(n)->empty_children);
2398 } else {
2399 __be32 val = htonl(n->key);
2400 struct fib_alias *fa;
2401
2402 seq_indent(seq, iter->depth);
2403 seq_printf(seq, " |-- %pI4\n", &val);
2404
2405 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2406 char buf1[32], buf2[32];
2407
2408 seq_indent(seq, iter->depth + 1);
2409 seq_printf(seq, " /%zu %s %s",
2410 KEYLENGTH - fa->fa_slen,
2411 rtn_scope(buf1, sizeof(buf1),
2412 fa->fa_info->fib_scope),
2413 rtn_type(buf2, sizeof(buf2),
2414 fa->fa_type));
2415 if (fa->fa_tos)
2416 seq_printf(seq, " tos=%d", fa->fa_tos);
2417 seq_putc(seq, '\n');
2418 }
2419 }
2420
2421 return 0;
2422 }
2423
2424 static const struct seq_operations fib_trie_seq_ops = {
2425 .start = fib_trie_seq_start,
2426 .next = fib_trie_seq_next,
2427 .stop = fib_trie_seq_stop,
2428 .show = fib_trie_seq_show,
2429 };
2430
2431 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2432 {
2433 return seq_open_net(inode, file, &fib_trie_seq_ops,
2434 sizeof(struct fib_trie_iter));
2435 }
2436
2437 static const struct file_operations fib_trie_fops = {
2438 .owner = THIS_MODULE,
2439 .open = fib_trie_seq_open,
2440 .read = seq_read,
2441 .llseek = seq_lseek,
2442 .release = seq_release_net,
2443 };
2444
2445 struct fib_route_iter {
2446 struct seq_net_private p;
2447 struct fib_table *main_tb;
2448 struct key_vector *tnode;
2449 loff_t pos;
2450 t_key key;
2451 };
2452
2453 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2454 loff_t pos)
2455 {
2456 struct fib_table *tb = iter->main_tb;
2457 struct key_vector *l, **tp = &iter->tnode;
2458 struct trie *t;
2459 t_key key;
2460
2461 /* use cache location of next-to-find key */
2462 if (iter->pos > 0 && pos >= iter->pos) {
2463 pos -= iter->pos;
2464 key = iter->key;
2465 } else {
2466 t = (struct trie *)tb->tb_data;
2467 iter->tnode = t->kv;
2468 iter->pos = 0;
2469 key = 0;
2470 }
2471
2472 while ((l = leaf_walk_rcu(tp, key)) != NULL) {
2473 key = l->key + 1;
2474 iter->pos++;
2475
2476 if (--pos <= 0)
2477 break;
2478
2479 l = NULL;
2480
2481 /* handle unlikely case of a key wrap */
2482 if (!key)
2483 break;
2484 }
2485
2486 if (l)
2487 iter->key = key; /* remember it */
2488 else
2489 iter->pos = 0; /* forget it */
2490
2491 return l;
2492 }
2493
2494 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2495 __acquires(RCU)
2496 {
2497 struct fib_route_iter *iter = seq->private;
2498 struct fib_table *tb;
2499 struct trie *t;
2500
2501 rcu_read_lock();
2502
2503 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2504 if (!tb)
2505 return NULL;
2506
2507 iter->main_tb = tb;
2508
2509 if (*pos != 0)
2510 return fib_route_get_idx(iter, *pos);
2511
2512 t = (struct trie *)tb->tb_data;
2513 iter->tnode = t->kv;
2514 iter->pos = 0;
2515 iter->key = 0;
2516
2517 return SEQ_START_TOKEN;
2518 }
2519
2520 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2521 {
2522 struct fib_route_iter *iter = seq->private;
2523 struct key_vector *l = NULL;
2524 t_key key = iter->key;
2525
2526 ++*pos;
2527
2528 /* only allow key of 0 for start of sequence */
2529 if ((v == SEQ_START_TOKEN) || key)
2530 l = leaf_walk_rcu(&iter->tnode, key);
2531
2532 if (l) {
2533 iter->key = l->key + 1;
2534 iter->pos++;
2535 } else {
2536 iter->pos = 0;
2537 }
2538
2539 return l;
2540 }
2541
2542 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2543 __releases(RCU)
2544 {
2545 rcu_read_unlock();
2546 }
2547
2548 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2549 {
2550 unsigned int flags = 0;
2551
2552 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2553 flags = RTF_REJECT;
2554 if (fi && fi->fib_nh->nh_gw)
2555 flags |= RTF_GATEWAY;
2556 if (mask == htonl(0xFFFFFFFF))
2557 flags |= RTF_HOST;
2558 flags |= RTF_UP;
2559 return flags;
2560 }
2561
2562 /*
2563 * This outputs /proc/net/route.
2564 * The format of the file is not supposed to be changed
2565 * and needs to be same as fib_hash output to avoid breaking
2566 * legacy utilities
2567 */
2568 static int fib_route_seq_show(struct seq_file *seq, void *v)
2569 {
2570 struct fib_route_iter *iter = seq->private;
2571 struct fib_table *tb = iter->main_tb;
2572 struct fib_alias *fa;
2573 struct key_vector *l = v;
2574 __be32 prefix;
2575
2576 if (v == SEQ_START_TOKEN) {
2577 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2578 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2579 "\tWindow\tIRTT");
2580 return 0;
2581 }
2582
2583 prefix = htonl(l->key);
2584
2585 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2586 const struct fib_info *fi = fa->fa_info;
2587 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2588 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2589
2590 if ((fa->fa_type == RTN_BROADCAST) ||
2591 (fa->fa_type == RTN_MULTICAST))
2592 continue;
2593
2594 if (fa->tb_id != tb->tb_id)
2595 continue;
2596
2597 seq_setwidth(seq, 127);
2598
2599 if (fi)
2600 seq_printf(seq,
2601 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2602 "%d\t%08X\t%d\t%u\t%u",
2603 fi->fib_dev ? fi->fib_dev->name : "*",
2604 prefix,
2605 fi->fib_nh->nh_gw, flags, 0, 0,
2606 fi->fib_priority,
2607 mask,
2608 (fi->fib_advmss ?
2609 fi->fib_advmss + 40 : 0),
2610 fi->fib_window,
2611 fi->fib_rtt >> 3);
2612 else
2613 seq_printf(seq,
2614 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2615 "%d\t%08X\t%d\t%u\t%u",
2616 prefix, 0, flags, 0, 0, 0,
2617 mask, 0, 0, 0);
2618
2619 seq_pad(seq, '\n');
2620 }
2621
2622 return 0;
2623 }
2624
2625 static const struct seq_operations fib_route_seq_ops = {
2626 .start = fib_route_seq_start,
2627 .next = fib_route_seq_next,
2628 .stop = fib_route_seq_stop,
2629 .show = fib_route_seq_show,
2630 };
2631
2632 static int fib_route_seq_open(struct inode *inode, struct file *file)
2633 {
2634 return seq_open_net(inode, file, &fib_route_seq_ops,
2635 sizeof(struct fib_route_iter));
2636 }
2637
2638 static const struct file_operations fib_route_fops = {
2639 .owner = THIS_MODULE,
2640 .open = fib_route_seq_open,
2641 .read = seq_read,
2642 .llseek = seq_lseek,
2643 .release = seq_release_net,
2644 };
2645
2646 int __net_init fib_proc_init(struct net *net)
2647 {
2648 if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2649 goto out1;
2650
2651 if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2652 &fib_triestat_fops))
2653 goto out2;
2654
2655 if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2656 goto out3;
2657
2658 return 0;
2659
2660 out3:
2661 remove_proc_entry("fib_triestat", net->proc_net);
2662 out2:
2663 remove_proc_entry("fib_trie", net->proc_net);
2664 out1:
2665 return -ENOMEM;
2666 }
2667
2668 void __net_exit fib_proc_exit(struct net *net)
2669 {
2670 remove_proc_entry("fib_trie", net->proc_net);
2671 remove_proc_entry("fib_triestat", net->proc_net);
2672 remove_proc_entry("route", net->proc_net);
2673 }
2674
2675 #endif /* CONFIG_PROC_FS */