mwifiex: remove_bss_prio_lock
[deliverable/linux.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 <net/net_namespace.h>
76 #include <net/ip.h>
77 #include <net/protocol.h>
78 #include <net/route.h>
79 #include <net/tcp.h>
80 #include <net/sock.h>
81 #include <net/ip_fib.h>
82 #include <net/switchdev.h>
83 #include "fib_lookup.h"
84
85 #define MAX_STAT_DEPTH 32
86
87 #define KEYLENGTH (8*sizeof(t_key))
88 #define KEY_MAX ((t_key)~0)
89
90 typedef unsigned int t_key;
91
92 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
93 #define IS_TNODE(n) ((n)->bits)
94 #define IS_LEAF(n) (!(n)->bits)
95
96 struct key_vector {
97 t_key key;
98 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
99 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
100 unsigned char slen;
101 union {
102 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
103 struct hlist_head leaf;
104 /* This array is valid if (pos | bits) > 0 (TNODE) */
105 struct key_vector __rcu *tnode[0];
106 };
107 };
108
109 struct tnode {
110 struct rcu_head rcu;
111 t_key empty_children; /* KEYLENGTH bits needed */
112 t_key full_children; /* KEYLENGTH bits needed */
113 struct key_vector __rcu *parent;
114 struct key_vector kv[1];
115 #define tn_bits kv[0].bits
116 };
117
118 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
119 #define LEAF_SIZE TNODE_SIZE(1)
120
121 #ifdef CONFIG_IP_FIB_TRIE_STATS
122 struct trie_use_stats {
123 unsigned int gets;
124 unsigned int backtrack;
125 unsigned int semantic_match_passed;
126 unsigned int semantic_match_miss;
127 unsigned int null_node_hit;
128 unsigned int resize_node_skipped;
129 };
130 #endif
131
132 struct trie_stat {
133 unsigned int totdepth;
134 unsigned int maxdepth;
135 unsigned int tnodes;
136 unsigned int leaves;
137 unsigned int nullpointers;
138 unsigned int prefixes;
139 unsigned int nodesizes[MAX_STAT_DEPTH];
140 };
141
142 struct trie {
143 struct key_vector kv[1];
144 #ifdef CONFIG_IP_FIB_TRIE_STATS
145 struct trie_use_stats __percpu *stats;
146 #endif
147 };
148
149 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
150 static size_t tnode_free_size;
151
152 /*
153 * synchronize_rcu after call_rcu for that many pages; it should be especially
154 * useful before resizing the root node with PREEMPT_NONE configs; the value was
155 * obtained experimentally, aiming to avoid visible slowdown.
156 */
157 static const int sync_pages = 128;
158
159 static struct kmem_cache *fn_alias_kmem __read_mostly;
160 static struct kmem_cache *trie_leaf_kmem __read_mostly;
161
162 static inline struct tnode *tn_info(struct key_vector *kv)
163 {
164 return container_of(kv, struct tnode, kv[0]);
165 }
166
167 /* caller must hold RTNL */
168 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
169 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
170
171 /* caller must hold RCU read lock or RTNL */
172 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
173 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
174
175 /* wrapper for rcu_assign_pointer */
176 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
177 {
178 if (n)
179 rcu_assign_pointer(tn_info(n)->parent, tp);
180 }
181
182 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
183
184 /* This provides us with the number of children in this node, in the case of a
185 * leaf this will return 0 meaning none of the children are accessible.
186 */
187 static inline unsigned long child_length(const struct key_vector *tn)
188 {
189 return (1ul << tn->bits) & ~(1ul);
190 }
191
192 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
193
194 static inline unsigned long get_index(t_key key, struct key_vector *kv)
195 {
196 unsigned long index = key ^ kv->key;
197
198 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
199 return 0;
200
201 return index >> kv->pos;
202 }
203
204 /* To understand this stuff, an understanding of keys and all their bits is
205 * necessary. Every node in the trie has a key associated with it, but not
206 * all of the bits in that key are significant.
207 *
208 * Consider a node 'n' and its parent 'tp'.
209 *
210 * If n is a leaf, every bit in its key is significant. Its presence is
211 * necessitated by path compression, since during a tree traversal (when
212 * searching for a leaf - unless we are doing an insertion) we will completely
213 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
214 * a potentially successful search, that we have indeed been walking the
215 * correct key path.
216 *
217 * Note that we can never "miss" the correct key in the tree if present by
218 * following the wrong path. Path compression ensures that segments of the key
219 * that are the same for all keys with a given prefix are skipped, but the
220 * skipped part *is* identical for each node in the subtrie below the skipped
221 * bit! trie_insert() in this implementation takes care of that.
222 *
223 * if n is an internal node - a 'tnode' here, the various parts of its key
224 * have many different meanings.
225 *
226 * Example:
227 * _________________________________________________________________
228 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
229 * -----------------------------------------------------------------
230 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
231 *
232 * _________________________________________________________________
233 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
234 * -----------------------------------------------------------------
235 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
236 *
237 * tp->pos = 22
238 * tp->bits = 3
239 * n->pos = 13
240 * n->bits = 4
241 *
242 * First, let's just ignore the bits that come before the parent tp, that is
243 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
244 * point we do not use them for anything.
245 *
246 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
247 * index into the parent's child array. That is, they will be used to find
248 * 'n' among tp's children.
249 *
250 * The bits from (n->pos + n->bits) to (tn->pos - 1) - "S" - are skipped bits
251 * for the node n.
252 *
253 * All the bits we have seen so far are significant to the node n. The rest
254 * of the bits are really not needed or indeed known in n->key.
255 *
256 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
257 * n's child array, and will of course be different for each child.
258 *
259 * The rest of the bits, from 0 to (n->pos + n->bits), are completely unknown
260 * at this point.
261 */
262
263 static const int halve_threshold = 25;
264 static const int inflate_threshold = 50;
265 static const int halve_threshold_root = 15;
266 static const int inflate_threshold_root = 30;
267
268 static void __alias_free_mem(struct rcu_head *head)
269 {
270 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
271 kmem_cache_free(fn_alias_kmem, fa);
272 }
273
274 static inline void alias_free_mem_rcu(struct fib_alias *fa)
275 {
276 call_rcu(&fa->rcu, __alias_free_mem);
277 }
278
279 #define TNODE_KMALLOC_MAX \
280 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
281 #define TNODE_VMALLOC_MAX \
282 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
283
284 static void __node_free_rcu(struct rcu_head *head)
285 {
286 struct tnode *n = container_of(head, struct tnode, rcu);
287
288 if (!n->tn_bits)
289 kmem_cache_free(trie_leaf_kmem, n);
290 else if (n->tn_bits <= TNODE_KMALLOC_MAX)
291 kfree(n);
292 else
293 vfree(n);
294 }
295
296 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
297
298 static struct tnode *tnode_alloc(int bits)
299 {
300 size_t size;
301
302 /* verify bits is within bounds */
303 if (bits > TNODE_VMALLOC_MAX)
304 return NULL;
305
306 /* determine size and verify it is non-zero and didn't overflow */
307 size = TNODE_SIZE(1ul << bits);
308
309 if (size <= PAGE_SIZE)
310 return kzalloc(size, GFP_KERNEL);
311 else
312 return vzalloc(size);
313 }
314
315 static inline void empty_child_inc(struct key_vector *n)
316 {
317 ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
318 }
319
320 static inline void empty_child_dec(struct key_vector *n)
321 {
322 tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
323 }
324
325 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
326 {
327 struct tnode *kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
328 struct key_vector *l = kv->kv;
329
330 if (!kv)
331 return NULL;
332
333 /* initialize key vector */
334 l->key = key;
335 l->pos = 0;
336 l->bits = 0;
337 l->slen = fa->fa_slen;
338
339 /* link leaf to fib alias */
340 INIT_HLIST_HEAD(&l->leaf);
341 hlist_add_head(&fa->fa_list, &l->leaf);
342
343 return l;
344 }
345
346 static struct key_vector *tnode_new(t_key key, int pos, int bits)
347 {
348 struct tnode *tnode = tnode_alloc(bits);
349 unsigned int shift = pos + bits;
350 struct key_vector *tn = tnode->kv;
351
352 /* verify bits and pos their msb bits clear and values are valid */
353 BUG_ON(!bits || (shift > KEYLENGTH));
354
355 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
356 sizeof(struct key_vector *) << bits);
357
358 if (!tnode)
359 return NULL;
360
361 if (bits == KEYLENGTH)
362 tnode->full_children = 1;
363 else
364 tnode->empty_children = 1ul << bits;
365
366 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
367 tn->pos = pos;
368 tn->bits = bits;
369 tn->slen = pos;
370
371 return tn;
372 }
373
374 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
375 * and no bits are skipped. See discussion in dyntree paper p. 6
376 */
377 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
378 {
379 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
380 }
381
382 /* Add a child at position i overwriting the old value.
383 * Update the value of full_children and empty_children.
384 */
385 static void put_child(struct key_vector *tn, unsigned long i,
386 struct key_vector *n)
387 {
388 struct key_vector *chi = get_child(tn, i);
389 int isfull, wasfull;
390
391 BUG_ON(i >= child_length(tn));
392
393 /* update emptyChildren, overflow into fullChildren */
394 if (n == NULL && chi != NULL)
395 empty_child_inc(tn);
396 if (n != NULL && chi == NULL)
397 empty_child_dec(tn);
398
399 /* update fullChildren */
400 wasfull = tnode_full(tn, chi);
401 isfull = tnode_full(tn, n);
402
403 if (wasfull && !isfull)
404 tn_info(tn)->full_children--;
405 else if (!wasfull && isfull)
406 tn_info(tn)->full_children++;
407
408 if (n && (tn->slen < n->slen))
409 tn->slen = n->slen;
410
411 rcu_assign_pointer(tn->tnode[i], n);
412 }
413
414 static void update_children(struct key_vector *tn)
415 {
416 unsigned long i;
417
418 /* update all of the child parent pointers */
419 for (i = child_length(tn); i;) {
420 struct key_vector *inode = get_child(tn, --i);
421
422 if (!inode)
423 continue;
424
425 /* Either update the children of a tnode that
426 * already belongs to us or update the child
427 * to point to ourselves.
428 */
429 if (node_parent(inode) == tn)
430 update_children(inode);
431 else
432 node_set_parent(inode, tn);
433 }
434 }
435
436 static inline void put_child_root(struct key_vector *tp, t_key key,
437 struct key_vector *n)
438 {
439 if (IS_TRIE(tp))
440 rcu_assign_pointer(tp->tnode[0], n);
441 else
442 put_child(tp, get_index(key, tp), n);
443 }
444
445 static inline void tnode_free_init(struct key_vector *tn)
446 {
447 tn_info(tn)->rcu.next = NULL;
448 }
449
450 static inline void tnode_free_append(struct key_vector *tn,
451 struct key_vector *n)
452 {
453 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
454 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
455 }
456
457 static void tnode_free(struct key_vector *tn)
458 {
459 struct callback_head *head = &tn_info(tn)->rcu;
460
461 while (head) {
462 head = head->next;
463 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
464 node_free(tn);
465
466 tn = container_of(head, struct tnode, rcu)->kv;
467 }
468
469 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
470 tnode_free_size = 0;
471 synchronize_rcu();
472 }
473 }
474
475 static struct key_vector *replace(struct trie *t,
476 struct key_vector *oldtnode,
477 struct key_vector *tn)
478 {
479 struct key_vector *tp = node_parent(oldtnode);
480 unsigned long i;
481
482 /* setup the parent pointer out of and back into this node */
483 NODE_INIT_PARENT(tn, tp);
484 put_child_root(tp, tn->key, tn);
485
486 /* update all of the child parent pointers */
487 update_children(tn);
488
489 /* all pointers should be clean so we are done */
490 tnode_free(oldtnode);
491
492 /* resize children now that oldtnode is freed */
493 for (i = child_length(tn); i;) {
494 struct key_vector *inode = get_child(tn, --i);
495
496 /* resize child node */
497 if (tnode_full(tn, inode))
498 tn = resize(t, inode);
499 }
500
501 return tp;
502 }
503
504 static struct key_vector *inflate(struct trie *t,
505 struct key_vector *oldtnode)
506 {
507 struct key_vector *tn;
508 unsigned long i;
509 t_key m;
510
511 pr_debug("In inflate\n");
512
513 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
514 if (!tn)
515 goto notnode;
516
517 /* prepare oldtnode to be freed */
518 tnode_free_init(oldtnode);
519
520 /* Assemble all of the pointers in our cluster, in this case that
521 * represents all of the pointers out of our allocated nodes that
522 * point to existing tnodes and the links between our allocated
523 * nodes.
524 */
525 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
526 struct key_vector *inode = get_child(oldtnode, --i);
527 struct key_vector *node0, *node1;
528 unsigned long j, k;
529
530 /* An empty child */
531 if (inode == NULL)
532 continue;
533
534 /* A leaf or an internal node with skipped bits */
535 if (!tnode_full(oldtnode, inode)) {
536 put_child(tn, get_index(inode->key, tn), inode);
537 continue;
538 }
539
540 /* drop the node in the old tnode free list */
541 tnode_free_append(oldtnode, inode);
542
543 /* An internal node with two children */
544 if (inode->bits == 1) {
545 put_child(tn, 2 * i + 1, get_child(inode, 1));
546 put_child(tn, 2 * i, get_child(inode, 0));
547 continue;
548 }
549
550 /* We will replace this node 'inode' with two new
551 * ones, 'node0' and 'node1', each with half of the
552 * original children. The two new nodes will have
553 * a position one bit further down the key and this
554 * means that the "significant" part of their keys
555 * (see the discussion near the top of this file)
556 * will differ by one bit, which will be "0" in
557 * node0's key and "1" in node1's key. Since we are
558 * moving the key position by one step, the bit that
559 * we are moving away from - the bit at position
560 * (tn->pos) - is the one that will differ between
561 * node0 and node1. So... we synthesize that bit in the
562 * two new keys.
563 */
564 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
565 if (!node1)
566 goto nomem;
567 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
568
569 tnode_free_append(tn, node1);
570 if (!node0)
571 goto nomem;
572 tnode_free_append(tn, node0);
573
574 /* populate child pointers in new nodes */
575 for (k = child_length(inode), j = k / 2; j;) {
576 put_child(node1, --j, get_child(inode, --k));
577 put_child(node0, j, get_child(inode, j));
578 put_child(node1, --j, get_child(inode, --k));
579 put_child(node0, j, get_child(inode, j));
580 }
581
582 /* link new nodes to parent */
583 NODE_INIT_PARENT(node1, tn);
584 NODE_INIT_PARENT(node0, tn);
585
586 /* link parent to nodes */
587 put_child(tn, 2 * i + 1, node1);
588 put_child(tn, 2 * i, node0);
589 }
590
591 /* setup the parent pointers into and out of this node */
592 return replace(t, oldtnode, tn);
593 nomem:
594 /* all pointers should be clean so we are done */
595 tnode_free(tn);
596 notnode:
597 return NULL;
598 }
599
600 static struct key_vector *halve(struct trie *t,
601 struct key_vector *oldtnode)
602 {
603 struct key_vector *tn;
604 unsigned long i;
605
606 pr_debug("In halve\n");
607
608 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
609 if (!tn)
610 goto notnode;
611
612 /* prepare oldtnode to be freed */
613 tnode_free_init(oldtnode);
614
615 /* Assemble all of the pointers in our cluster, in this case that
616 * represents all of the pointers out of our allocated nodes that
617 * point to existing tnodes and the links between our allocated
618 * nodes.
619 */
620 for (i = child_length(oldtnode); i;) {
621 struct key_vector *node1 = get_child(oldtnode, --i);
622 struct key_vector *node0 = get_child(oldtnode, --i);
623 struct key_vector *inode;
624
625 /* At least one of the children is empty */
626 if (!node1 || !node0) {
627 put_child(tn, i / 2, node1 ? : node0);
628 continue;
629 }
630
631 /* Two nonempty children */
632 inode = tnode_new(node0->key, oldtnode->pos, 1);
633 if (!inode)
634 goto nomem;
635 tnode_free_append(tn, inode);
636
637 /* initialize pointers out of node */
638 put_child(inode, 1, node1);
639 put_child(inode, 0, node0);
640 NODE_INIT_PARENT(inode, tn);
641
642 /* link parent to node */
643 put_child(tn, i / 2, inode);
644 }
645
646 /* setup the parent pointers into and out of this node */
647 return replace(t, oldtnode, tn);
648 nomem:
649 /* all pointers should be clean so we are done */
650 tnode_free(tn);
651 notnode:
652 return NULL;
653 }
654
655 static struct key_vector *collapse(struct trie *t,
656 struct key_vector *oldtnode)
657 {
658 struct key_vector *n, *tp;
659 unsigned long i;
660
661 /* scan the tnode looking for that one child that might still exist */
662 for (n = NULL, i = child_length(oldtnode); !n && i;)
663 n = get_child(oldtnode, --i);
664
665 /* compress one level */
666 tp = node_parent(oldtnode);
667 put_child_root(tp, oldtnode->key, n);
668 node_set_parent(n, tp);
669
670 /* drop dead node */
671 node_free(oldtnode);
672
673 return tp;
674 }
675
676 static unsigned char update_suffix(struct key_vector *tn)
677 {
678 unsigned char slen = tn->pos;
679 unsigned long stride, i;
680
681 /* search though the list of children looking for nodes that might
682 * have a suffix greater than the one we currently have. This is
683 * why we start with a stride of 2 since a stride of 1 would
684 * represent the nodes with suffix length equal to tn->pos
685 */
686 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
687 struct key_vector *n = get_child(tn, i);
688
689 if (!n || (n->slen <= slen))
690 continue;
691
692 /* update stride and slen based on new value */
693 stride <<= (n->slen - slen);
694 slen = n->slen;
695 i &= ~(stride - 1);
696
697 /* if slen covers all but the last bit we can stop here
698 * there will be nothing longer than that since only node
699 * 0 and 1 << (bits - 1) could have that as their suffix
700 * length.
701 */
702 if ((slen + 1) >= (tn->pos + tn->bits))
703 break;
704 }
705
706 tn->slen = slen;
707
708 return slen;
709 }
710
711 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
712 * the Helsinki University of Technology and Matti Tikkanen of Nokia
713 * Telecommunications, page 6:
714 * "A node is doubled if the ratio of non-empty children to all
715 * children in the *doubled* node is at least 'high'."
716 *
717 * 'high' in this instance is the variable 'inflate_threshold'. It
718 * is expressed as a percentage, so we multiply it with
719 * child_length() and instead of multiplying by 2 (since the
720 * child array will be doubled by inflate()) and multiplying
721 * the left-hand side by 100 (to handle the percentage thing) we
722 * multiply the left-hand side by 50.
723 *
724 * The left-hand side may look a bit weird: child_length(tn)
725 * - tn->empty_children is of course the number of non-null children
726 * in the current node. tn->full_children is the number of "full"
727 * children, that is non-null tnodes with a skip value of 0.
728 * All of those will be doubled in the resulting inflated tnode, so
729 * we just count them one extra time here.
730 *
731 * A clearer way to write this would be:
732 *
733 * to_be_doubled = tn->full_children;
734 * not_to_be_doubled = child_length(tn) - tn->empty_children -
735 * tn->full_children;
736 *
737 * new_child_length = child_length(tn) * 2;
738 *
739 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
740 * new_child_length;
741 * if (new_fill_factor >= inflate_threshold)
742 *
743 * ...and so on, tho it would mess up the while () loop.
744 *
745 * anyway,
746 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
747 * inflate_threshold
748 *
749 * avoid a division:
750 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
751 * inflate_threshold * new_child_length
752 *
753 * expand not_to_be_doubled and to_be_doubled, and shorten:
754 * 100 * (child_length(tn) - tn->empty_children +
755 * tn->full_children) >= inflate_threshold * new_child_length
756 *
757 * expand new_child_length:
758 * 100 * (child_length(tn) - tn->empty_children +
759 * tn->full_children) >=
760 * inflate_threshold * child_length(tn) * 2
761 *
762 * shorten again:
763 * 50 * (tn->full_children + child_length(tn) -
764 * tn->empty_children) >= inflate_threshold *
765 * child_length(tn)
766 *
767 */
768 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
769 {
770 unsigned long used = child_length(tn);
771 unsigned long threshold = used;
772
773 /* Keep root node larger */
774 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
775 used -= tn_info(tn)->empty_children;
776 used += tn_info(tn)->full_children;
777
778 /* if bits == KEYLENGTH then pos = 0, and will fail below */
779
780 return (used > 1) && tn->pos && ((50 * used) >= threshold);
781 }
782
783 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
784 {
785 unsigned long used = child_length(tn);
786 unsigned long threshold = used;
787
788 /* Keep root node larger */
789 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
790 used -= tn_info(tn)->empty_children;
791
792 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
793
794 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
795 }
796
797 static inline bool should_collapse(struct key_vector *tn)
798 {
799 unsigned long used = child_length(tn);
800
801 used -= tn_info(tn)->empty_children;
802
803 /* account for bits == KEYLENGTH case */
804 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
805 used -= KEY_MAX;
806
807 /* One child or none, time to drop us from the trie */
808 return used < 2;
809 }
810
811 #define MAX_WORK 10
812 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
813 {
814 #ifdef CONFIG_IP_FIB_TRIE_STATS
815 struct trie_use_stats __percpu *stats = t->stats;
816 #endif
817 struct key_vector *tp = node_parent(tn);
818 unsigned long cindex = get_index(tn->key, tp);
819 int max_work = MAX_WORK;
820
821 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
822 tn, inflate_threshold, halve_threshold);
823
824 /* track the tnode via the pointer from the parent instead of
825 * doing it ourselves. This way we can let RCU fully do its
826 * thing without us interfering
827 */
828 BUG_ON(tn != get_child(tp, cindex));
829
830 /* Double as long as the resulting node has a number of
831 * nonempty nodes that are above the threshold.
832 */
833 while (should_inflate(tp, tn) && max_work--) {
834 tp = inflate(t, tn);
835 if (!tp) {
836 #ifdef CONFIG_IP_FIB_TRIE_STATS
837 this_cpu_inc(stats->resize_node_skipped);
838 #endif
839 break;
840 }
841
842 tn = get_child(tp, cindex);
843 }
844
845 /* Return if at least one inflate is run */
846 if (max_work != MAX_WORK)
847 return node_parent(tn);
848
849 /* Halve as long as the number of empty children in this
850 * node is above threshold.
851 */
852 while (should_halve(tp, tn) && max_work--) {
853 tp = halve(t, tn);
854 if (!tp) {
855 #ifdef CONFIG_IP_FIB_TRIE_STATS
856 this_cpu_inc(stats->resize_node_skipped);
857 #endif
858 break;
859 }
860
861 tn = get_child(tp, cindex);
862 }
863
864 /* Only one child remains */
865 if (should_collapse(tn))
866 return collapse(t, tn);
867
868 /* update parent in case inflate or halve failed */
869 tp = node_parent(tn);
870
871 /* Return if at least one deflate was run */
872 if (max_work != MAX_WORK)
873 return tp;
874
875 /* push the suffix length to the parent node */
876 if (tn->slen > tn->pos) {
877 unsigned char slen = update_suffix(tn);
878
879 if (slen > tp->slen)
880 tp->slen = slen;
881 }
882
883 return tp;
884 }
885
886 static void leaf_pull_suffix(struct key_vector *tp, struct key_vector *l)
887 {
888 while ((tp->slen > tp->pos) && (tp->slen > l->slen)) {
889 if (update_suffix(tp) > l->slen)
890 break;
891 tp = node_parent(tp);
892 }
893 }
894
895 static void leaf_push_suffix(struct key_vector *tn, struct key_vector *l)
896 {
897 /* if this is a new leaf then tn will be NULL and we can sort
898 * out parent suffix lengths as a part of trie_rebalance
899 */
900 while (tn->slen < l->slen) {
901 tn->slen = l->slen;
902 tn = node_parent(tn);
903 }
904 }
905
906 /* rcu_read_lock needs to be hold by caller from readside */
907 static struct key_vector *fib_find_node(struct trie *t,
908 struct key_vector **tp, u32 key)
909 {
910 struct key_vector *pn, *n = t->kv;
911 unsigned long index = 0;
912
913 do {
914 pn = n;
915 n = get_child_rcu(n, index);
916
917 if (!n)
918 break;
919
920 index = get_cindex(key, n);
921
922 /* This bit of code is a bit tricky but it combines multiple
923 * checks into a single check. The prefix consists of the
924 * prefix plus zeros for the bits in the cindex. The index
925 * is the difference between the key and this value. From
926 * this we can actually derive several pieces of data.
927 * if (index >= (1ul << bits))
928 * we have a mismatch in skip bits and failed
929 * else
930 * we know the value is cindex
931 *
932 * This check is safe even if bits == KEYLENGTH due to the
933 * fact that we can only allocate a node with 32 bits if a
934 * long is greater than 32 bits.
935 */
936 if (index >= (1ul << n->bits)) {
937 n = NULL;
938 break;
939 }
940
941 /* keep searching until we find a perfect match leaf or NULL */
942 } while (IS_TNODE(n));
943
944 *tp = pn;
945
946 return n;
947 }
948
949 /* Return the first fib alias matching TOS with
950 * priority less than or equal to PRIO.
951 */
952 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
953 u8 tos, u32 prio, u32 tb_id)
954 {
955 struct fib_alias *fa;
956
957 if (!fah)
958 return NULL;
959
960 hlist_for_each_entry(fa, fah, fa_list) {
961 if (fa->fa_slen < slen)
962 continue;
963 if (fa->fa_slen != slen)
964 break;
965 if (fa->tb_id > tb_id)
966 continue;
967 if (fa->tb_id != tb_id)
968 break;
969 if (fa->fa_tos > tos)
970 continue;
971 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
972 return fa;
973 }
974
975 return NULL;
976 }
977
978 static void trie_rebalance(struct trie *t, struct key_vector *tn)
979 {
980 while (!IS_TRIE(tn))
981 tn = resize(t, tn);
982 }
983
984 static int fib_insert_node(struct trie *t, struct key_vector *tp,
985 struct fib_alias *new, t_key key)
986 {
987 struct key_vector *n, *l;
988
989 l = leaf_new(key, new);
990 if (!l)
991 goto noleaf;
992
993 /* retrieve child from parent node */
994 n = get_child(tp, get_index(key, tp));
995
996 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
997 *
998 * Add a new tnode here
999 * first tnode need some special handling
1000 * leaves us in position for handling as case 3
1001 */
1002 if (n) {
1003 struct key_vector *tn;
1004
1005 tn = tnode_new(key, __fls(key ^ n->key), 1);
1006 if (!tn)
1007 goto notnode;
1008
1009 /* initialize routes out of node */
1010 NODE_INIT_PARENT(tn, tp);
1011 put_child(tn, get_index(key, tn) ^ 1, n);
1012
1013 /* start adding routes into the node */
1014 put_child_root(tp, key, tn);
1015 node_set_parent(n, tn);
1016
1017 /* parent now has a NULL spot where the leaf can go */
1018 tp = tn;
1019 }
1020
1021 /* Case 3: n is NULL, and will just insert a new leaf */
1022 NODE_INIT_PARENT(l, tp);
1023 put_child_root(tp, key, l);
1024 trie_rebalance(t, tp);
1025
1026 return 0;
1027 notnode:
1028 node_free(l);
1029 noleaf:
1030 return -ENOMEM;
1031 }
1032
1033 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1034 struct key_vector *l, struct fib_alias *new,
1035 struct fib_alias *fa, t_key key)
1036 {
1037 if (!l)
1038 return fib_insert_node(t, tp, new, key);
1039
1040 if (fa) {
1041 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1042 } else {
1043 struct fib_alias *last;
1044
1045 hlist_for_each_entry(last, &l->leaf, fa_list) {
1046 if (new->fa_slen < last->fa_slen)
1047 break;
1048 if ((new->fa_slen == last->fa_slen) &&
1049 (new->tb_id > last->tb_id))
1050 break;
1051 fa = last;
1052 }
1053
1054 if (fa)
1055 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1056 else
1057 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1058 }
1059
1060 /* if we added to the tail node then we need to update slen */
1061 if (l->slen < new->fa_slen) {
1062 l->slen = new->fa_slen;
1063 leaf_push_suffix(tp, l);
1064 }
1065
1066 return 0;
1067 }
1068
1069 /* Caller must hold RTNL. */
1070 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1071 {
1072 struct trie *t = (struct trie *)tb->tb_data;
1073 struct fib_alias *fa, *new_fa;
1074 struct key_vector *l, *tp;
1075 struct fib_info *fi;
1076 u8 plen = cfg->fc_dst_len;
1077 u8 slen = KEYLENGTH - plen;
1078 u8 tos = cfg->fc_tos;
1079 u32 key;
1080 int err;
1081
1082 if (plen > KEYLENGTH)
1083 return -EINVAL;
1084
1085 key = ntohl(cfg->fc_dst);
1086
1087 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1088
1089 if ((plen < KEYLENGTH) && (key << plen))
1090 return -EINVAL;
1091
1092 fi = fib_create_info(cfg);
1093 if (IS_ERR(fi)) {
1094 err = PTR_ERR(fi);
1095 goto err;
1096 }
1097
1098 l = fib_find_node(t, &tp, key);
1099 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1100 tb->tb_id) : NULL;
1101
1102 /* Now fa, if non-NULL, points to the first fib alias
1103 * with the same keys [prefix,tos,priority], if such key already
1104 * exists or to the node before which we will insert new one.
1105 *
1106 * If fa is NULL, we will need to allocate a new one and
1107 * insert to the tail of the section matching the suffix length
1108 * of the new alias.
1109 */
1110
1111 if (fa && fa->fa_tos == tos &&
1112 fa->fa_info->fib_priority == fi->fib_priority) {
1113 struct fib_alias *fa_first, *fa_match;
1114
1115 err = -EEXIST;
1116 if (cfg->fc_nlflags & NLM_F_EXCL)
1117 goto out;
1118
1119 /* We have 2 goals:
1120 * 1. Find exact match for type, scope, fib_info to avoid
1121 * duplicate routes
1122 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1123 */
1124 fa_match = NULL;
1125 fa_first = fa;
1126 hlist_for_each_entry_from(fa, fa_list) {
1127 if ((fa->fa_slen != slen) ||
1128 (fa->tb_id != tb->tb_id) ||
1129 (fa->fa_tos != tos))
1130 break;
1131 if (fa->fa_info->fib_priority != fi->fib_priority)
1132 break;
1133 if (fa->fa_type == cfg->fc_type &&
1134 fa->fa_info == fi) {
1135 fa_match = fa;
1136 break;
1137 }
1138 }
1139
1140 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1141 struct fib_info *fi_drop;
1142 u8 state;
1143
1144 fa = fa_first;
1145 if (fa_match) {
1146 if (fa == fa_match)
1147 err = 0;
1148 goto out;
1149 }
1150 err = -ENOBUFS;
1151 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1152 if (new_fa == NULL)
1153 goto out;
1154
1155 fi_drop = fa->fa_info;
1156 new_fa->fa_tos = fa->fa_tos;
1157 new_fa->fa_info = fi;
1158 new_fa->fa_type = cfg->fc_type;
1159 state = fa->fa_state;
1160 new_fa->fa_state = state & ~FA_S_ACCESSED;
1161 new_fa->fa_slen = fa->fa_slen;
1162
1163 err = netdev_switch_fib_ipv4_add(key, plen, fi,
1164 new_fa->fa_tos,
1165 cfg->fc_type,
1166 cfg->fc_nlflags,
1167 tb->tb_id);
1168 if (err) {
1169 netdev_switch_fib_ipv4_abort(fi);
1170 kmem_cache_free(fn_alias_kmem, new_fa);
1171 goto out;
1172 }
1173
1174 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1175
1176 alias_free_mem_rcu(fa);
1177
1178 fib_release_info(fi_drop);
1179 if (state & FA_S_ACCESSED)
1180 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1181 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1182 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1183
1184 goto succeeded;
1185 }
1186 /* Error if we find a perfect match which
1187 * uses the same scope, type, and nexthop
1188 * information.
1189 */
1190 if (fa_match)
1191 goto out;
1192
1193 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1194 fa = fa_first;
1195 }
1196 err = -ENOENT;
1197 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1198 goto out;
1199
1200 err = -ENOBUFS;
1201 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1202 if (new_fa == NULL)
1203 goto out;
1204
1205 new_fa->fa_info = fi;
1206 new_fa->fa_tos = tos;
1207 new_fa->fa_type = cfg->fc_type;
1208 new_fa->fa_state = 0;
1209 new_fa->fa_slen = slen;
1210 new_fa->tb_id = tb->tb_id;
1211
1212 /* (Optionally) offload fib entry to switch hardware. */
1213 err = netdev_switch_fib_ipv4_add(key, plen, fi, tos,
1214 cfg->fc_type,
1215 cfg->fc_nlflags,
1216 tb->tb_id);
1217 if (err) {
1218 netdev_switch_fib_ipv4_abort(fi);
1219 goto out_free_new_fa;
1220 }
1221
1222 /* Insert new entry to the list. */
1223 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1224 if (err)
1225 goto out_sw_fib_del;
1226
1227 if (!plen)
1228 tb->tb_num_default++;
1229
1230 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1231 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1232 &cfg->fc_nlinfo, 0);
1233 succeeded:
1234 return 0;
1235
1236 out_sw_fib_del:
1237 netdev_switch_fib_ipv4_del(key, plen, fi, tos, cfg->fc_type, tb->tb_id);
1238 out_free_new_fa:
1239 kmem_cache_free(fn_alias_kmem, new_fa);
1240 out:
1241 fib_release_info(fi);
1242 err:
1243 return err;
1244 }
1245
1246 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1247 {
1248 t_key prefix = n->key;
1249
1250 return (key ^ prefix) & (prefix | -prefix);
1251 }
1252
1253 /* should be called with rcu_read_lock */
1254 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1255 struct fib_result *res, int fib_flags)
1256 {
1257 struct trie *t = (struct trie *) tb->tb_data;
1258 #ifdef CONFIG_IP_FIB_TRIE_STATS
1259 struct trie_use_stats __percpu *stats = t->stats;
1260 #endif
1261 const t_key key = ntohl(flp->daddr);
1262 struct key_vector *n, *pn;
1263 struct fib_alias *fa;
1264 unsigned long index;
1265 t_key cindex;
1266
1267 pn = t->kv;
1268 cindex = 0;
1269
1270 n = get_child_rcu(pn, cindex);
1271 if (!n)
1272 return -EAGAIN;
1273
1274 #ifdef CONFIG_IP_FIB_TRIE_STATS
1275 this_cpu_inc(stats->gets);
1276 #endif
1277
1278 /* Step 1: Travel to the longest prefix match in the trie */
1279 for (;;) {
1280 index = get_cindex(key, n);
1281
1282 /* This bit of code is a bit tricky but it combines multiple
1283 * checks into a single check. The prefix consists of the
1284 * prefix plus zeros for the "bits" in the prefix. The index
1285 * is the difference between the key and this value. From
1286 * this we can actually derive several pieces of data.
1287 * if (index >= (1ul << bits))
1288 * we have a mismatch in skip bits and failed
1289 * else
1290 * we know the value is cindex
1291 *
1292 * This check is safe even if bits == KEYLENGTH due to the
1293 * fact that we can only allocate a node with 32 bits if a
1294 * long is greater than 32 bits.
1295 */
1296 if (index >= (1ul << n->bits))
1297 break;
1298
1299 /* we have found a leaf. Prefixes have already been compared */
1300 if (IS_LEAF(n))
1301 goto found;
1302
1303 /* only record pn and cindex if we are going to be chopping
1304 * bits later. Otherwise we are just wasting cycles.
1305 */
1306 if (n->slen > n->pos) {
1307 pn = n;
1308 cindex = index;
1309 }
1310
1311 n = get_child_rcu(n, index);
1312 if (unlikely(!n))
1313 goto backtrace;
1314 }
1315
1316 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1317 for (;;) {
1318 /* record the pointer where our next node pointer is stored */
1319 struct key_vector __rcu **cptr = n->tnode;
1320
1321 /* This test verifies that none of the bits that differ
1322 * between the key and the prefix exist in the region of
1323 * the lsb and higher in the prefix.
1324 */
1325 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1326 goto backtrace;
1327
1328 /* exit out and process leaf */
1329 if (unlikely(IS_LEAF(n)))
1330 break;
1331
1332 /* Don't bother recording parent info. Since we are in
1333 * prefix match mode we will have to come back to wherever
1334 * we started this traversal anyway
1335 */
1336
1337 while ((n = rcu_dereference(*cptr)) == NULL) {
1338 backtrace:
1339 #ifdef CONFIG_IP_FIB_TRIE_STATS
1340 if (!n)
1341 this_cpu_inc(stats->null_node_hit);
1342 #endif
1343 /* If we are at cindex 0 there are no more bits for
1344 * us to strip at this level so we must ascend back
1345 * up one level to see if there are any more bits to
1346 * be stripped there.
1347 */
1348 while (!cindex) {
1349 t_key pkey = pn->key;
1350
1351 /* If we don't have a parent then there is
1352 * nothing for us to do as we do not have any
1353 * further nodes to parse.
1354 */
1355 if (IS_TRIE(pn))
1356 return -EAGAIN;
1357 #ifdef CONFIG_IP_FIB_TRIE_STATS
1358 this_cpu_inc(stats->backtrack);
1359 #endif
1360 /* Get Child's index */
1361 pn = node_parent_rcu(pn);
1362 cindex = get_index(pkey, pn);
1363 }
1364
1365 /* strip the least significant bit from the cindex */
1366 cindex &= cindex - 1;
1367
1368 /* grab pointer for next child node */
1369 cptr = &pn->tnode[cindex];
1370 }
1371 }
1372
1373 found:
1374 /* this line carries forward the xor from earlier in the function */
1375 index = key ^ n->key;
1376
1377 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1378 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1379 struct fib_info *fi = fa->fa_info;
1380 int nhsel, err;
1381
1382 if ((index >= (1ul << fa->fa_slen)) &&
1383 ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen != KEYLENGTH)))
1384 continue;
1385 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1386 continue;
1387 if (fi->fib_dead)
1388 continue;
1389 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1390 continue;
1391 fib_alias_accessed(fa);
1392 err = fib_props[fa->fa_type].error;
1393 if (unlikely(err < 0)) {
1394 #ifdef CONFIG_IP_FIB_TRIE_STATS
1395 this_cpu_inc(stats->semantic_match_passed);
1396 #endif
1397 return err;
1398 }
1399 if (fi->fib_flags & RTNH_F_DEAD)
1400 continue;
1401 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1402 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1403
1404 if (nh->nh_flags & RTNH_F_DEAD)
1405 continue;
1406 if (flp->flowi4_oif && flp->flowi4_oif != nh->nh_oif)
1407 continue;
1408
1409 if (!(fib_flags & FIB_LOOKUP_NOREF))
1410 atomic_inc(&fi->fib_clntref);
1411
1412 res->prefixlen = KEYLENGTH - fa->fa_slen;
1413 res->nh_sel = nhsel;
1414 res->type = fa->fa_type;
1415 res->scope = fi->fib_scope;
1416 res->fi = fi;
1417 res->table = tb;
1418 res->fa_head = &n->leaf;
1419 #ifdef CONFIG_IP_FIB_TRIE_STATS
1420 this_cpu_inc(stats->semantic_match_passed);
1421 #endif
1422 return err;
1423 }
1424 }
1425 #ifdef CONFIG_IP_FIB_TRIE_STATS
1426 this_cpu_inc(stats->semantic_match_miss);
1427 #endif
1428 goto backtrace;
1429 }
1430 EXPORT_SYMBOL_GPL(fib_table_lookup);
1431
1432 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1433 struct key_vector *l, struct fib_alias *old)
1434 {
1435 /* record the location of the previous list_info entry */
1436 struct hlist_node **pprev = old->fa_list.pprev;
1437 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1438
1439 /* remove the fib_alias from the list */
1440 hlist_del_rcu(&old->fa_list);
1441
1442 /* if we emptied the list this leaf will be freed and we can sort
1443 * out parent suffix lengths as a part of trie_rebalance
1444 */
1445 if (hlist_empty(&l->leaf)) {
1446 put_child_root(tp, l->key, NULL);
1447 node_free(l);
1448 trie_rebalance(t, tp);
1449 return;
1450 }
1451
1452 /* only access fa if it is pointing at the last valid hlist_node */
1453 if (*pprev)
1454 return;
1455
1456 /* update the trie with the latest suffix length */
1457 l->slen = fa->fa_slen;
1458 leaf_pull_suffix(tp, l);
1459 }
1460
1461 /* Caller must hold RTNL. */
1462 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1463 {
1464 struct trie *t = (struct trie *) tb->tb_data;
1465 struct fib_alias *fa, *fa_to_delete;
1466 struct key_vector *l, *tp;
1467 u8 plen = cfg->fc_dst_len;
1468 u8 slen = KEYLENGTH - plen;
1469 u8 tos = cfg->fc_tos;
1470 u32 key;
1471
1472 if (plen > KEYLENGTH)
1473 return -EINVAL;
1474
1475 key = ntohl(cfg->fc_dst);
1476
1477 if ((plen < KEYLENGTH) && (key << plen))
1478 return -EINVAL;
1479
1480 l = fib_find_node(t, &tp, key);
1481 if (!l)
1482 return -ESRCH;
1483
1484 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1485 if (!fa)
1486 return -ESRCH;
1487
1488 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1489
1490 fa_to_delete = NULL;
1491 hlist_for_each_entry_from(fa, fa_list) {
1492 struct fib_info *fi = fa->fa_info;
1493
1494 if ((fa->fa_slen != slen) ||
1495 (fa->tb_id != tb->tb_id) ||
1496 (fa->fa_tos != tos))
1497 break;
1498
1499 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1500 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1501 fa->fa_info->fib_scope == cfg->fc_scope) &&
1502 (!cfg->fc_prefsrc ||
1503 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1504 (!cfg->fc_protocol ||
1505 fi->fib_protocol == cfg->fc_protocol) &&
1506 fib_nh_match(cfg, fi) == 0) {
1507 fa_to_delete = fa;
1508 break;
1509 }
1510 }
1511
1512 if (!fa_to_delete)
1513 return -ESRCH;
1514
1515 netdev_switch_fib_ipv4_del(key, plen, fa_to_delete->fa_info, tos,
1516 cfg->fc_type, tb->tb_id);
1517
1518 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1519 &cfg->fc_nlinfo, 0);
1520
1521 if (!plen)
1522 tb->tb_num_default--;
1523
1524 fib_remove_alias(t, tp, l, fa_to_delete);
1525
1526 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1527 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1528
1529 fib_release_info(fa_to_delete->fa_info);
1530 alias_free_mem_rcu(fa_to_delete);
1531 return 0;
1532 }
1533
1534 /* Scan for the next leaf starting at the provided key value */
1535 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1536 {
1537 struct key_vector *pn, *n = *tn;
1538 unsigned long cindex;
1539
1540 /* this loop is meant to try and find the key in the trie */
1541 do {
1542 /* record parent and next child index */
1543 pn = n;
1544 cindex = key ? get_index(key, pn) : 0;
1545
1546 if (cindex >> pn->bits)
1547 break;
1548
1549 /* descend into the next child */
1550 n = get_child_rcu(pn, cindex++);
1551 if (!n)
1552 break;
1553
1554 /* guarantee forward progress on the keys */
1555 if (IS_LEAF(n) && (n->key >= key))
1556 goto found;
1557 } while (IS_TNODE(n));
1558
1559 /* this loop will search for the next leaf with a greater key */
1560 while (!IS_TRIE(pn)) {
1561 /* if we exhausted the parent node we will need to climb */
1562 if (cindex >= (1ul << pn->bits)) {
1563 t_key pkey = pn->key;
1564
1565 pn = node_parent_rcu(pn);
1566 cindex = get_index(pkey, pn) + 1;
1567 continue;
1568 }
1569
1570 /* grab the next available node */
1571 n = get_child_rcu(pn, cindex++);
1572 if (!n)
1573 continue;
1574
1575 /* no need to compare keys since we bumped the index */
1576 if (IS_LEAF(n))
1577 goto found;
1578
1579 /* Rescan start scanning in new node */
1580 pn = n;
1581 cindex = 0;
1582 }
1583
1584 *tn = pn;
1585 return NULL; /* Root of trie */
1586 found:
1587 /* if we are at the limit for keys just return NULL for the tnode */
1588 *tn = pn;
1589 return n;
1590 }
1591
1592 static void fib_trie_free(struct fib_table *tb)
1593 {
1594 struct trie *t = (struct trie *)tb->tb_data;
1595 struct key_vector *pn = t->kv;
1596 unsigned long cindex = 1;
1597 struct hlist_node *tmp;
1598 struct fib_alias *fa;
1599
1600 /* walk trie in reverse order and free everything */
1601 for (;;) {
1602 struct key_vector *n;
1603
1604 if (!(cindex--)) {
1605 t_key pkey = pn->key;
1606
1607 if (IS_TRIE(pn))
1608 break;
1609
1610 n = pn;
1611 pn = node_parent(pn);
1612
1613 /* drop emptied tnode */
1614 put_child_root(pn, n->key, NULL);
1615 node_free(n);
1616
1617 cindex = get_index(pkey, pn);
1618
1619 continue;
1620 }
1621
1622 /* grab the next available node */
1623 n = get_child(pn, cindex);
1624 if (!n)
1625 continue;
1626
1627 if (IS_TNODE(n)) {
1628 /* record pn and cindex for leaf walking */
1629 pn = n;
1630 cindex = 1ul << n->bits;
1631
1632 continue;
1633 }
1634
1635 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1636 hlist_del_rcu(&fa->fa_list);
1637 alias_free_mem_rcu(fa);
1638 }
1639
1640 put_child_root(pn, n->key, NULL);
1641 node_free(n);
1642 }
1643
1644 #ifdef CONFIG_IP_FIB_TRIE_STATS
1645 free_percpu(t->stats);
1646 #endif
1647 kfree(tb);
1648 }
1649
1650 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1651 {
1652 struct trie *ot = (struct trie *)oldtb->tb_data;
1653 struct key_vector *l, *tp = ot->kv;
1654 struct fib_table *local_tb;
1655 struct fib_alias *fa;
1656 struct trie *lt;
1657 t_key key = 0;
1658
1659 if (oldtb->tb_data == oldtb->__data)
1660 return oldtb;
1661
1662 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1663 if (!local_tb)
1664 return NULL;
1665
1666 lt = (struct trie *)local_tb->tb_data;
1667
1668 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1669 struct key_vector *local_l = NULL, *local_tp;
1670
1671 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1672 struct fib_alias *new_fa;
1673
1674 if (local_tb->tb_id != fa->tb_id)
1675 continue;
1676
1677 /* clone fa for new local table */
1678 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1679 if (!new_fa)
1680 goto out;
1681
1682 memcpy(new_fa, fa, sizeof(*fa));
1683
1684 /* insert clone into table */
1685 if (!local_l)
1686 local_l = fib_find_node(lt, &local_tp, l->key);
1687
1688 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1689 NULL, l->key))
1690 goto out;
1691 }
1692
1693 /* stop loop if key wrapped back to 0 */
1694 key = l->key + 1;
1695 if (key < l->key)
1696 break;
1697 }
1698
1699 return local_tb;
1700 out:
1701 fib_trie_free(local_tb);
1702
1703 return NULL;
1704 }
1705
1706 /* Caller must hold RTNL */
1707 void fib_table_flush_external(struct fib_table *tb)
1708 {
1709 struct trie *t = (struct trie *)tb->tb_data;
1710 struct key_vector *pn = t->kv;
1711 unsigned long cindex = 1;
1712 struct hlist_node *tmp;
1713 struct fib_alias *fa;
1714
1715 /* walk trie in reverse order */
1716 for (;;) {
1717 unsigned char slen = 0;
1718 struct key_vector *n;
1719
1720 if (!(cindex--)) {
1721 t_key pkey = pn->key;
1722
1723 /* cannot resize the trie vector */
1724 if (IS_TRIE(pn))
1725 break;
1726
1727 /* resize completed node */
1728 pn = resize(t, pn);
1729 cindex = get_index(pkey, pn);
1730
1731 continue;
1732 }
1733
1734 /* grab the next available node */
1735 n = get_child(pn, cindex);
1736 if (!n)
1737 continue;
1738
1739 if (IS_TNODE(n)) {
1740 /* record pn and cindex for leaf walking */
1741 pn = n;
1742 cindex = 1ul << n->bits;
1743
1744 continue;
1745 }
1746
1747 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1748 struct fib_info *fi = fa->fa_info;
1749
1750 /* if alias was cloned to local then we just
1751 * need to remove the local copy from main
1752 */
1753 if (tb->tb_id != fa->tb_id) {
1754 hlist_del_rcu(&fa->fa_list);
1755 alias_free_mem_rcu(fa);
1756 continue;
1757 }
1758
1759 /* record local slen */
1760 slen = fa->fa_slen;
1761
1762 if (!fi || !(fi->fib_flags & RTNH_F_EXTERNAL))
1763 continue;
1764
1765 netdev_switch_fib_ipv4_del(n->key,
1766 KEYLENGTH - fa->fa_slen,
1767 fi, fa->fa_tos,
1768 fa->fa_type, tb->tb_id);
1769 }
1770
1771 /* update leaf slen */
1772 n->slen = slen;
1773
1774 if (hlist_empty(&n->leaf)) {
1775 put_child_root(pn, n->key, NULL);
1776 node_free(n);
1777 } else {
1778 leaf_pull_suffix(pn, n);
1779 }
1780 }
1781 }
1782
1783 /* Caller must hold RTNL. */
1784 int fib_table_flush(struct fib_table *tb)
1785 {
1786 struct trie *t = (struct trie *)tb->tb_data;
1787 struct key_vector *pn = t->kv;
1788 unsigned long cindex = 1;
1789 struct hlist_node *tmp;
1790 struct fib_alias *fa;
1791 int found = 0;
1792
1793 /* walk trie in reverse order */
1794 for (;;) {
1795 unsigned char slen = 0;
1796 struct key_vector *n;
1797
1798 if (!(cindex--)) {
1799 t_key pkey = pn->key;
1800
1801 /* cannot resize the trie vector */
1802 if (IS_TRIE(pn))
1803 break;
1804
1805 /* resize completed node */
1806 pn = resize(t, pn);
1807 cindex = get_index(pkey, pn);
1808
1809 continue;
1810 }
1811
1812 /* grab the next available node */
1813 n = get_child(pn, cindex);
1814 if (!n)
1815 continue;
1816
1817 if (IS_TNODE(n)) {
1818 /* record pn and cindex for leaf walking */
1819 pn = n;
1820 cindex = 1ul << n->bits;
1821
1822 continue;
1823 }
1824
1825 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1826 struct fib_info *fi = fa->fa_info;
1827
1828 if (!fi || !(fi->fib_flags & RTNH_F_DEAD)) {
1829 slen = fa->fa_slen;
1830 continue;
1831 }
1832
1833 netdev_switch_fib_ipv4_del(n->key,
1834 KEYLENGTH - fa->fa_slen,
1835 fi, fa->fa_tos,
1836 fa->fa_type, tb->tb_id);
1837 hlist_del_rcu(&fa->fa_list);
1838 fib_release_info(fa->fa_info);
1839 alias_free_mem_rcu(fa);
1840 found++;
1841 }
1842
1843 /* update leaf slen */
1844 n->slen = slen;
1845
1846 if (hlist_empty(&n->leaf)) {
1847 put_child_root(pn, n->key, NULL);
1848 node_free(n);
1849 } else {
1850 leaf_pull_suffix(pn, n);
1851 }
1852 }
1853
1854 pr_debug("trie_flush found=%d\n", found);
1855 return found;
1856 }
1857
1858 static void __trie_free_rcu(struct rcu_head *head)
1859 {
1860 struct fib_table *tb = container_of(head, struct fib_table, rcu);
1861 #ifdef CONFIG_IP_FIB_TRIE_STATS
1862 struct trie *t = (struct trie *)tb->tb_data;
1863
1864 if (tb->tb_data == tb->__data)
1865 free_percpu(t->stats);
1866 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1867 kfree(tb);
1868 }
1869
1870 void fib_free_table(struct fib_table *tb)
1871 {
1872 call_rcu(&tb->rcu, __trie_free_rcu);
1873 }
1874
1875 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
1876 struct sk_buff *skb, struct netlink_callback *cb)
1877 {
1878 __be32 xkey = htonl(l->key);
1879 struct fib_alias *fa;
1880 int i, s_i;
1881
1882 s_i = cb->args[4];
1883 i = 0;
1884
1885 /* rcu_read_lock is hold by caller */
1886 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1887 if (i < s_i) {
1888 i++;
1889 continue;
1890 }
1891
1892 if (tb->tb_id != fa->tb_id) {
1893 i++;
1894 continue;
1895 }
1896
1897 if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
1898 cb->nlh->nlmsg_seq,
1899 RTM_NEWROUTE,
1900 tb->tb_id,
1901 fa->fa_type,
1902 xkey,
1903 KEYLENGTH - fa->fa_slen,
1904 fa->fa_tos,
1905 fa->fa_info, NLM_F_MULTI) < 0) {
1906 cb->args[4] = i;
1907 return -1;
1908 }
1909 i++;
1910 }
1911
1912 cb->args[4] = i;
1913 return skb->len;
1914 }
1915
1916 /* rcu_read_lock needs to be hold by caller from readside */
1917 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1918 struct netlink_callback *cb)
1919 {
1920 struct trie *t = (struct trie *)tb->tb_data;
1921 struct key_vector *l, *tp = t->kv;
1922 /* Dump starting at last key.
1923 * Note: 0.0.0.0/0 (ie default) is first key.
1924 */
1925 int count = cb->args[2];
1926 t_key key = cb->args[3];
1927
1928 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1929 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1930 cb->args[3] = key;
1931 cb->args[2] = count;
1932 return -1;
1933 }
1934
1935 ++count;
1936 key = l->key + 1;
1937
1938 memset(&cb->args[4], 0,
1939 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1940
1941 /* stop loop if key wrapped back to 0 */
1942 if (key < l->key)
1943 break;
1944 }
1945
1946 cb->args[3] = key;
1947 cb->args[2] = count;
1948
1949 return skb->len;
1950 }
1951
1952 void __init fib_trie_init(void)
1953 {
1954 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1955 sizeof(struct fib_alias),
1956 0, SLAB_PANIC, NULL);
1957
1958 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1959 LEAF_SIZE,
1960 0, SLAB_PANIC, NULL);
1961 }
1962
1963 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
1964 {
1965 struct fib_table *tb;
1966 struct trie *t;
1967 size_t sz = sizeof(*tb);
1968
1969 if (!alias)
1970 sz += sizeof(struct trie);
1971
1972 tb = kzalloc(sz, GFP_KERNEL);
1973 if (tb == NULL)
1974 return NULL;
1975
1976 tb->tb_id = id;
1977 tb->tb_default = -1;
1978 tb->tb_num_default = 0;
1979 tb->tb_data = (alias ? alias->__data : tb->__data);
1980
1981 if (alias)
1982 return tb;
1983
1984 t = (struct trie *) tb->tb_data;
1985 t->kv[0].pos = KEYLENGTH;
1986 t->kv[0].slen = KEYLENGTH;
1987 #ifdef CONFIG_IP_FIB_TRIE_STATS
1988 t->stats = alloc_percpu(struct trie_use_stats);
1989 if (!t->stats) {
1990 kfree(tb);
1991 tb = NULL;
1992 }
1993 #endif
1994
1995 return tb;
1996 }
1997
1998 #ifdef CONFIG_PROC_FS
1999 /* Depth first Trie walk iterator */
2000 struct fib_trie_iter {
2001 struct seq_net_private p;
2002 struct fib_table *tb;
2003 struct key_vector *tnode;
2004 unsigned int index;
2005 unsigned int depth;
2006 };
2007
2008 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2009 {
2010 unsigned long cindex = iter->index;
2011 struct key_vector *pn = iter->tnode;
2012 t_key pkey;
2013
2014 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2015 iter->tnode, iter->index, iter->depth);
2016
2017 while (!IS_TRIE(pn)) {
2018 while (cindex < child_length(pn)) {
2019 struct key_vector *n = get_child_rcu(pn, cindex++);
2020
2021 if (!n)
2022 continue;
2023
2024 if (IS_LEAF(n)) {
2025 iter->tnode = pn;
2026 iter->index = cindex;
2027 } else {
2028 /* push down one level */
2029 iter->tnode = n;
2030 iter->index = 0;
2031 ++iter->depth;
2032 }
2033
2034 return n;
2035 }
2036
2037 /* Current node exhausted, pop back up */
2038 pkey = pn->key;
2039 pn = node_parent_rcu(pn);
2040 cindex = get_index(pkey, pn) + 1;
2041 --iter->depth;
2042 }
2043
2044 /* record root node so further searches know we are done */
2045 iter->tnode = pn;
2046 iter->index = 0;
2047
2048 return NULL;
2049 }
2050
2051 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2052 struct trie *t)
2053 {
2054 struct key_vector *n, *pn = t->kv;
2055
2056 if (!t)
2057 return NULL;
2058
2059 n = rcu_dereference(pn->tnode[0]);
2060 if (!n)
2061 return NULL;
2062
2063 if (IS_TNODE(n)) {
2064 iter->tnode = n;
2065 iter->index = 0;
2066 iter->depth = 1;
2067 } else {
2068 iter->tnode = pn;
2069 iter->index = 0;
2070 iter->depth = 0;
2071 }
2072
2073 return n;
2074 }
2075
2076 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2077 {
2078 struct key_vector *n;
2079 struct fib_trie_iter iter;
2080
2081 memset(s, 0, sizeof(*s));
2082
2083 rcu_read_lock();
2084 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2085 if (IS_LEAF(n)) {
2086 struct fib_alias *fa;
2087
2088 s->leaves++;
2089 s->totdepth += iter.depth;
2090 if (iter.depth > s->maxdepth)
2091 s->maxdepth = iter.depth;
2092
2093 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2094 ++s->prefixes;
2095 } else {
2096 s->tnodes++;
2097 if (n->bits < MAX_STAT_DEPTH)
2098 s->nodesizes[n->bits]++;
2099 s->nullpointers += tn_info(n)->empty_children;
2100 }
2101 }
2102 rcu_read_unlock();
2103 }
2104
2105 /*
2106 * This outputs /proc/net/fib_triestats
2107 */
2108 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2109 {
2110 unsigned int i, max, pointers, bytes, avdepth;
2111
2112 if (stat->leaves)
2113 avdepth = stat->totdepth*100 / stat->leaves;
2114 else
2115 avdepth = 0;
2116
2117 seq_printf(seq, "\tAver depth: %u.%02d\n",
2118 avdepth / 100, avdepth % 100);
2119 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2120
2121 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2122 bytes = LEAF_SIZE * stat->leaves;
2123
2124 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2125 bytes += sizeof(struct fib_alias) * stat->prefixes;
2126
2127 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2128 bytes += TNODE_SIZE(0) * stat->tnodes;
2129
2130 max = MAX_STAT_DEPTH;
2131 while (max > 0 && stat->nodesizes[max-1] == 0)
2132 max--;
2133
2134 pointers = 0;
2135 for (i = 1; i < max; i++)
2136 if (stat->nodesizes[i] != 0) {
2137 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2138 pointers += (1<<i) * stat->nodesizes[i];
2139 }
2140 seq_putc(seq, '\n');
2141 seq_printf(seq, "\tPointers: %u\n", pointers);
2142
2143 bytes += sizeof(struct key_vector *) * pointers;
2144 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2145 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2146 }
2147
2148 #ifdef CONFIG_IP_FIB_TRIE_STATS
2149 static void trie_show_usage(struct seq_file *seq,
2150 const struct trie_use_stats __percpu *stats)
2151 {
2152 struct trie_use_stats s = { 0 };
2153 int cpu;
2154
2155 /* loop through all of the CPUs and gather up the stats */
2156 for_each_possible_cpu(cpu) {
2157 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2158
2159 s.gets += pcpu->gets;
2160 s.backtrack += pcpu->backtrack;
2161 s.semantic_match_passed += pcpu->semantic_match_passed;
2162 s.semantic_match_miss += pcpu->semantic_match_miss;
2163 s.null_node_hit += pcpu->null_node_hit;
2164 s.resize_node_skipped += pcpu->resize_node_skipped;
2165 }
2166
2167 seq_printf(seq, "\nCounters:\n---------\n");
2168 seq_printf(seq, "gets = %u\n", s.gets);
2169 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2170 seq_printf(seq, "semantic match passed = %u\n",
2171 s.semantic_match_passed);
2172 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2173 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2174 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2175 }
2176 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2177
2178 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2179 {
2180 if (tb->tb_id == RT_TABLE_LOCAL)
2181 seq_puts(seq, "Local:\n");
2182 else if (tb->tb_id == RT_TABLE_MAIN)
2183 seq_puts(seq, "Main:\n");
2184 else
2185 seq_printf(seq, "Id %d:\n", tb->tb_id);
2186 }
2187
2188
2189 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2190 {
2191 struct net *net = (struct net *)seq->private;
2192 unsigned int h;
2193
2194 seq_printf(seq,
2195 "Basic info: size of leaf:"
2196 " %Zd bytes, size of tnode: %Zd bytes.\n",
2197 LEAF_SIZE, TNODE_SIZE(0));
2198
2199 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2200 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2201 struct fib_table *tb;
2202
2203 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2204 struct trie *t = (struct trie *) tb->tb_data;
2205 struct trie_stat stat;
2206
2207 if (!t)
2208 continue;
2209
2210 fib_table_print(seq, tb);
2211
2212 trie_collect_stats(t, &stat);
2213 trie_show_stats(seq, &stat);
2214 #ifdef CONFIG_IP_FIB_TRIE_STATS
2215 trie_show_usage(seq, t->stats);
2216 #endif
2217 }
2218 }
2219
2220 return 0;
2221 }
2222
2223 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2224 {
2225 return single_open_net(inode, file, fib_triestat_seq_show);
2226 }
2227
2228 static const struct file_operations fib_triestat_fops = {
2229 .owner = THIS_MODULE,
2230 .open = fib_triestat_seq_open,
2231 .read = seq_read,
2232 .llseek = seq_lseek,
2233 .release = single_release_net,
2234 };
2235
2236 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2237 {
2238 struct fib_trie_iter *iter = seq->private;
2239 struct net *net = seq_file_net(seq);
2240 loff_t idx = 0;
2241 unsigned int h;
2242
2243 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2244 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2245 struct fib_table *tb;
2246
2247 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2248 struct key_vector *n;
2249
2250 for (n = fib_trie_get_first(iter,
2251 (struct trie *) tb->tb_data);
2252 n; n = fib_trie_get_next(iter))
2253 if (pos == idx++) {
2254 iter->tb = tb;
2255 return n;
2256 }
2257 }
2258 }
2259
2260 return NULL;
2261 }
2262
2263 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2264 __acquires(RCU)
2265 {
2266 rcu_read_lock();
2267 return fib_trie_get_idx(seq, *pos);
2268 }
2269
2270 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2271 {
2272 struct fib_trie_iter *iter = seq->private;
2273 struct net *net = seq_file_net(seq);
2274 struct fib_table *tb = iter->tb;
2275 struct hlist_node *tb_node;
2276 unsigned int h;
2277 struct key_vector *n;
2278
2279 ++*pos;
2280 /* next node in same table */
2281 n = fib_trie_get_next(iter);
2282 if (n)
2283 return n;
2284
2285 /* walk rest of this hash chain */
2286 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2287 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2288 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2289 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2290 if (n)
2291 goto found;
2292 }
2293
2294 /* new hash chain */
2295 while (++h < FIB_TABLE_HASHSZ) {
2296 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2297 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2298 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2299 if (n)
2300 goto found;
2301 }
2302 }
2303 return NULL;
2304
2305 found:
2306 iter->tb = tb;
2307 return n;
2308 }
2309
2310 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2311 __releases(RCU)
2312 {
2313 rcu_read_unlock();
2314 }
2315
2316 static void seq_indent(struct seq_file *seq, int n)
2317 {
2318 while (n-- > 0)
2319 seq_puts(seq, " ");
2320 }
2321
2322 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2323 {
2324 switch (s) {
2325 case RT_SCOPE_UNIVERSE: return "universe";
2326 case RT_SCOPE_SITE: return "site";
2327 case RT_SCOPE_LINK: return "link";
2328 case RT_SCOPE_HOST: return "host";
2329 case RT_SCOPE_NOWHERE: return "nowhere";
2330 default:
2331 snprintf(buf, len, "scope=%d", s);
2332 return buf;
2333 }
2334 }
2335
2336 static const char *const rtn_type_names[__RTN_MAX] = {
2337 [RTN_UNSPEC] = "UNSPEC",
2338 [RTN_UNICAST] = "UNICAST",
2339 [RTN_LOCAL] = "LOCAL",
2340 [RTN_BROADCAST] = "BROADCAST",
2341 [RTN_ANYCAST] = "ANYCAST",
2342 [RTN_MULTICAST] = "MULTICAST",
2343 [RTN_BLACKHOLE] = "BLACKHOLE",
2344 [RTN_UNREACHABLE] = "UNREACHABLE",
2345 [RTN_PROHIBIT] = "PROHIBIT",
2346 [RTN_THROW] = "THROW",
2347 [RTN_NAT] = "NAT",
2348 [RTN_XRESOLVE] = "XRESOLVE",
2349 };
2350
2351 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2352 {
2353 if (t < __RTN_MAX && rtn_type_names[t])
2354 return rtn_type_names[t];
2355 snprintf(buf, len, "type %u", t);
2356 return buf;
2357 }
2358
2359 /* Pretty print the trie */
2360 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2361 {
2362 const struct fib_trie_iter *iter = seq->private;
2363 struct key_vector *n = v;
2364
2365 if (IS_TRIE(node_parent_rcu(n)))
2366 fib_table_print(seq, iter->tb);
2367
2368 if (IS_TNODE(n)) {
2369 __be32 prf = htonl(n->key);
2370
2371 seq_indent(seq, iter->depth-1);
2372 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2373 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2374 tn_info(n)->full_children,
2375 tn_info(n)->empty_children);
2376 } else {
2377 __be32 val = htonl(n->key);
2378 struct fib_alias *fa;
2379
2380 seq_indent(seq, iter->depth);
2381 seq_printf(seq, " |-- %pI4\n", &val);
2382
2383 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2384 char buf1[32], buf2[32];
2385
2386 seq_indent(seq, iter->depth + 1);
2387 seq_printf(seq, " /%zu %s %s",
2388 KEYLENGTH - fa->fa_slen,
2389 rtn_scope(buf1, sizeof(buf1),
2390 fa->fa_info->fib_scope),
2391 rtn_type(buf2, sizeof(buf2),
2392 fa->fa_type));
2393 if (fa->fa_tos)
2394 seq_printf(seq, " tos=%d", fa->fa_tos);
2395 seq_putc(seq, '\n');
2396 }
2397 }
2398
2399 return 0;
2400 }
2401
2402 static const struct seq_operations fib_trie_seq_ops = {
2403 .start = fib_trie_seq_start,
2404 .next = fib_trie_seq_next,
2405 .stop = fib_trie_seq_stop,
2406 .show = fib_trie_seq_show,
2407 };
2408
2409 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2410 {
2411 return seq_open_net(inode, file, &fib_trie_seq_ops,
2412 sizeof(struct fib_trie_iter));
2413 }
2414
2415 static const struct file_operations fib_trie_fops = {
2416 .owner = THIS_MODULE,
2417 .open = fib_trie_seq_open,
2418 .read = seq_read,
2419 .llseek = seq_lseek,
2420 .release = seq_release_net,
2421 };
2422
2423 struct fib_route_iter {
2424 struct seq_net_private p;
2425 struct fib_table *main_tb;
2426 struct key_vector *tnode;
2427 loff_t pos;
2428 t_key key;
2429 };
2430
2431 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2432 loff_t pos)
2433 {
2434 struct fib_table *tb = iter->main_tb;
2435 struct key_vector *l, **tp = &iter->tnode;
2436 struct trie *t;
2437 t_key key;
2438
2439 /* use cache location of next-to-find key */
2440 if (iter->pos > 0 && pos >= iter->pos) {
2441 pos -= iter->pos;
2442 key = iter->key;
2443 } else {
2444 t = (struct trie *)tb->tb_data;
2445 iter->tnode = t->kv;
2446 iter->pos = 0;
2447 key = 0;
2448 }
2449
2450 while ((l = leaf_walk_rcu(tp, key)) != NULL) {
2451 key = l->key + 1;
2452 iter->pos++;
2453
2454 if (pos-- <= 0)
2455 break;
2456
2457 l = NULL;
2458
2459 /* handle unlikely case of a key wrap */
2460 if (!key)
2461 break;
2462 }
2463
2464 if (l)
2465 iter->key = key; /* remember it */
2466 else
2467 iter->pos = 0; /* forget it */
2468
2469 return l;
2470 }
2471
2472 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2473 __acquires(RCU)
2474 {
2475 struct fib_route_iter *iter = seq->private;
2476 struct fib_table *tb;
2477 struct trie *t;
2478
2479 rcu_read_lock();
2480
2481 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2482 if (!tb)
2483 return NULL;
2484
2485 iter->main_tb = tb;
2486
2487 if (*pos != 0)
2488 return fib_route_get_idx(iter, *pos);
2489
2490 t = (struct trie *)tb->tb_data;
2491 iter->tnode = t->kv;
2492 iter->pos = 0;
2493 iter->key = 0;
2494
2495 return SEQ_START_TOKEN;
2496 }
2497
2498 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2499 {
2500 struct fib_route_iter *iter = seq->private;
2501 struct key_vector *l = NULL;
2502 t_key key = iter->key;
2503
2504 ++*pos;
2505
2506 /* only allow key of 0 for start of sequence */
2507 if ((v == SEQ_START_TOKEN) || key)
2508 l = leaf_walk_rcu(&iter->tnode, key);
2509
2510 if (l) {
2511 iter->key = l->key + 1;
2512 iter->pos++;
2513 } else {
2514 iter->pos = 0;
2515 }
2516
2517 return l;
2518 }
2519
2520 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2521 __releases(RCU)
2522 {
2523 rcu_read_unlock();
2524 }
2525
2526 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2527 {
2528 unsigned int flags = 0;
2529
2530 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2531 flags = RTF_REJECT;
2532 if (fi && fi->fib_nh->nh_gw)
2533 flags |= RTF_GATEWAY;
2534 if (mask == htonl(0xFFFFFFFF))
2535 flags |= RTF_HOST;
2536 flags |= RTF_UP;
2537 return flags;
2538 }
2539
2540 /*
2541 * This outputs /proc/net/route.
2542 * The format of the file is not supposed to be changed
2543 * and needs to be same as fib_hash output to avoid breaking
2544 * legacy utilities
2545 */
2546 static int fib_route_seq_show(struct seq_file *seq, void *v)
2547 {
2548 struct fib_route_iter *iter = seq->private;
2549 struct fib_table *tb = iter->main_tb;
2550 struct fib_alias *fa;
2551 struct key_vector *l = v;
2552 __be32 prefix;
2553
2554 if (v == SEQ_START_TOKEN) {
2555 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2556 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2557 "\tWindow\tIRTT");
2558 return 0;
2559 }
2560
2561 prefix = htonl(l->key);
2562
2563 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2564 const struct fib_info *fi = fa->fa_info;
2565 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2566 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2567
2568 if ((fa->fa_type == RTN_BROADCAST) ||
2569 (fa->fa_type == RTN_MULTICAST))
2570 continue;
2571
2572 if (fa->tb_id != tb->tb_id)
2573 continue;
2574
2575 seq_setwidth(seq, 127);
2576
2577 if (fi)
2578 seq_printf(seq,
2579 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2580 "%d\t%08X\t%d\t%u\t%u",
2581 fi->fib_dev ? fi->fib_dev->name : "*",
2582 prefix,
2583 fi->fib_nh->nh_gw, flags, 0, 0,
2584 fi->fib_priority,
2585 mask,
2586 (fi->fib_advmss ?
2587 fi->fib_advmss + 40 : 0),
2588 fi->fib_window,
2589 fi->fib_rtt >> 3);
2590 else
2591 seq_printf(seq,
2592 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2593 "%d\t%08X\t%d\t%u\t%u",
2594 prefix, 0, flags, 0, 0, 0,
2595 mask, 0, 0, 0);
2596
2597 seq_pad(seq, '\n');
2598 }
2599
2600 return 0;
2601 }
2602
2603 static const struct seq_operations fib_route_seq_ops = {
2604 .start = fib_route_seq_start,
2605 .next = fib_route_seq_next,
2606 .stop = fib_route_seq_stop,
2607 .show = fib_route_seq_show,
2608 };
2609
2610 static int fib_route_seq_open(struct inode *inode, struct file *file)
2611 {
2612 return seq_open_net(inode, file, &fib_route_seq_ops,
2613 sizeof(struct fib_route_iter));
2614 }
2615
2616 static const struct file_operations fib_route_fops = {
2617 .owner = THIS_MODULE,
2618 .open = fib_route_seq_open,
2619 .read = seq_read,
2620 .llseek = seq_lseek,
2621 .release = seq_release_net,
2622 };
2623
2624 int __net_init fib_proc_init(struct net *net)
2625 {
2626 if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2627 goto out1;
2628
2629 if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2630 &fib_triestat_fops))
2631 goto out2;
2632
2633 if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2634 goto out3;
2635
2636 return 0;
2637
2638 out3:
2639 remove_proc_entry("fib_triestat", net->proc_net);
2640 out2:
2641 remove_proc_entry("fib_trie", net->proc_net);
2642 out1:
2643 return -ENOMEM;
2644 }
2645
2646 void __net_exit fib_proc_exit(struct net *net)
2647 {
2648 remove_proc_entry("fib_trie", net->proc_net);
2649 remove_proc_entry("fib_triestat", net->proc_net);
2650 remove_proc_entry("route", net->proc_net);
2651 }
2652
2653 #endif /* CONFIG_PROC_FS */
This page took 0.124677 seconds and 5 git commands to generate.