Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[cris-mirror.git] / net / ipv4 / fib_trie.c
blob5530cd6fdbc7d44dc259653e0b26d2065d007f1b
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.
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
15 * This work is based on the LPC-trie which is originally described in:
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/
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
26 * Code from fib_hash has been reused which includes the following header:
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.
33 * IPv4 FIB: lookup engine and maintenance routines.
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
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.
43 * Substantial contributions to this work comes from:
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>
51 #define VERSION "0.409"
53 #include <linux/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 <linux/notifier.h>
77 #include <net/net_namespace.h>
78 #include <net/ip.h>
79 #include <net/protocol.h>
80 #include <net/route.h>
81 #include <net/tcp.h>
82 #include <net/sock.h>
83 #include <net/ip_fib.h>
84 #include <net/fib_notifier.h>
85 #include <trace/events/fib.h>
86 #include "fib_lookup.h"
88 static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
89 enum fib_event_type event_type, u32 dst,
90 int dst_len, struct fib_alias *fa)
92 struct fib_entry_notifier_info info = {
93 .dst = dst,
94 .dst_len = dst_len,
95 .fi = fa->fa_info,
96 .tos = fa->fa_tos,
97 .type = fa->fa_type,
98 .tb_id = fa->tb_id,
100 return call_fib4_notifier(nb, net, event_type, &info.info);
103 static int call_fib_entry_notifiers(struct net *net,
104 enum fib_event_type event_type, u32 dst,
105 int dst_len, struct fib_alias *fa,
106 struct netlink_ext_ack *extack)
108 struct fib_entry_notifier_info info = {
109 .info.extack = extack,
110 .dst = dst,
111 .dst_len = dst_len,
112 .fi = fa->fa_info,
113 .tos = fa->fa_tos,
114 .type = fa->fa_type,
115 .tb_id = fa->tb_id,
117 return call_fib4_notifiers(net, event_type, &info.info);
120 #define MAX_STAT_DEPTH 32
122 #define KEYLENGTH (8*sizeof(t_key))
123 #define KEY_MAX ((t_key)~0)
125 typedef unsigned int t_key;
127 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
128 #define IS_TNODE(n) ((n)->bits)
129 #define IS_LEAF(n) (!(n)->bits)
131 struct key_vector {
132 t_key key;
133 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
134 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
135 unsigned char slen;
136 union {
137 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
138 struct hlist_head leaf;
139 /* This array is valid if (pos | bits) > 0 (TNODE) */
140 struct key_vector __rcu *tnode[0];
144 struct tnode {
145 struct rcu_head rcu;
146 t_key empty_children; /* KEYLENGTH bits needed */
147 t_key full_children; /* KEYLENGTH bits needed */
148 struct key_vector __rcu *parent;
149 struct key_vector kv[1];
150 #define tn_bits kv[0].bits
153 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
154 #define LEAF_SIZE TNODE_SIZE(1)
156 #ifdef CONFIG_IP_FIB_TRIE_STATS
157 struct trie_use_stats {
158 unsigned int gets;
159 unsigned int backtrack;
160 unsigned int semantic_match_passed;
161 unsigned int semantic_match_miss;
162 unsigned int null_node_hit;
163 unsigned int resize_node_skipped;
165 #endif
167 struct trie_stat {
168 unsigned int totdepth;
169 unsigned int maxdepth;
170 unsigned int tnodes;
171 unsigned int leaves;
172 unsigned int nullpointers;
173 unsigned int prefixes;
174 unsigned int nodesizes[MAX_STAT_DEPTH];
177 struct trie {
178 struct key_vector kv[1];
179 #ifdef CONFIG_IP_FIB_TRIE_STATS
180 struct trie_use_stats __percpu *stats;
181 #endif
184 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
185 static size_t tnode_free_size;
188 * synchronize_rcu after call_rcu for that many pages; it should be especially
189 * useful before resizing the root node with PREEMPT_NONE configs; the value was
190 * obtained experimentally, aiming to avoid visible slowdown.
192 static const int sync_pages = 128;
194 static struct kmem_cache *fn_alias_kmem __read_mostly;
195 static struct kmem_cache *trie_leaf_kmem __read_mostly;
197 static inline struct tnode *tn_info(struct key_vector *kv)
199 return container_of(kv, struct tnode, kv[0]);
202 /* caller must hold RTNL */
203 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
204 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
206 /* caller must hold RCU read lock or RTNL */
207 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
208 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
210 /* wrapper for rcu_assign_pointer */
211 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
213 if (n)
214 rcu_assign_pointer(tn_info(n)->parent, tp);
217 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
219 /* This provides us with the number of children in this node, in the case of a
220 * leaf this will return 0 meaning none of the children are accessible.
222 static inline unsigned long child_length(const struct key_vector *tn)
224 return (1ul << tn->bits) & ~(1ul);
227 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
229 static inline unsigned long get_index(t_key key, struct key_vector *kv)
231 unsigned long index = key ^ kv->key;
233 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
234 return 0;
236 return index >> kv->pos;
239 /* To understand this stuff, an understanding of keys and all their bits is
240 * necessary. Every node in the trie has a key associated with it, but not
241 * all of the bits in that key are significant.
243 * Consider a node 'n' and its parent 'tp'.
245 * If n is a leaf, every bit in its key is significant. Its presence is
246 * necessitated by path compression, since during a tree traversal (when
247 * searching for a leaf - unless we are doing an insertion) we will completely
248 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
249 * a potentially successful search, that we have indeed been walking the
250 * correct key path.
252 * Note that we can never "miss" the correct key in the tree if present by
253 * following the wrong path. Path compression ensures that segments of the key
254 * that are the same for all keys with a given prefix are skipped, but the
255 * skipped part *is* identical for each node in the subtrie below the skipped
256 * bit! trie_insert() in this implementation takes care of that.
258 * if n is an internal node - a 'tnode' here, the various parts of its key
259 * have many different meanings.
261 * Example:
262 * _________________________________________________________________
263 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
264 * -----------------------------------------------------------------
265 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
267 * _________________________________________________________________
268 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
269 * -----------------------------------------------------------------
270 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
272 * tp->pos = 22
273 * tp->bits = 3
274 * n->pos = 13
275 * n->bits = 4
277 * First, let's just ignore the bits that come before the parent tp, that is
278 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
279 * point we do not use them for anything.
281 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
282 * index into the parent's child array. That is, they will be used to find
283 * 'n' among tp's children.
285 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
286 * for the node n.
288 * All the bits we have seen so far are significant to the node n. The rest
289 * of the bits are really not needed or indeed known in n->key.
291 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
292 * n's child array, and will of course be different for each child.
294 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
295 * at this point.
298 static const int halve_threshold = 25;
299 static const int inflate_threshold = 50;
300 static const int halve_threshold_root = 15;
301 static const int inflate_threshold_root = 30;
303 static void __alias_free_mem(struct rcu_head *head)
305 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
306 kmem_cache_free(fn_alias_kmem, fa);
309 static inline void alias_free_mem_rcu(struct fib_alias *fa)
311 call_rcu(&fa->rcu, __alias_free_mem);
314 #define TNODE_KMALLOC_MAX \
315 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
316 #define TNODE_VMALLOC_MAX \
317 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
319 static void __node_free_rcu(struct rcu_head *head)
321 struct tnode *n = container_of(head, struct tnode, rcu);
323 if (!n->tn_bits)
324 kmem_cache_free(trie_leaf_kmem, n);
325 else
326 kvfree(n);
329 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
331 static struct tnode *tnode_alloc(int bits)
333 size_t size;
335 /* verify bits is within bounds */
336 if (bits > TNODE_VMALLOC_MAX)
337 return NULL;
339 /* determine size and verify it is non-zero and didn't overflow */
340 size = TNODE_SIZE(1ul << bits);
342 if (size <= PAGE_SIZE)
343 return kzalloc(size, GFP_KERNEL);
344 else
345 return vzalloc(size);
348 static inline void empty_child_inc(struct key_vector *n)
350 ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
353 static inline void empty_child_dec(struct key_vector *n)
355 tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
358 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
360 struct key_vector *l;
361 struct tnode *kv;
363 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
364 if (!kv)
365 return NULL;
367 /* initialize key vector */
368 l = kv->kv;
369 l->key = key;
370 l->pos = 0;
371 l->bits = 0;
372 l->slen = fa->fa_slen;
374 /* link leaf to fib alias */
375 INIT_HLIST_HEAD(&l->leaf);
376 hlist_add_head(&fa->fa_list, &l->leaf);
378 return l;
381 static struct key_vector *tnode_new(t_key key, int pos, int bits)
383 unsigned int shift = pos + bits;
384 struct key_vector *tn;
385 struct tnode *tnode;
387 /* verify bits and pos their msb bits clear and values are valid */
388 BUG_ON(!bits || (shift > KEYLENGTH));
390 tnode = tnode_alloc(bits);
391 if (!tnode)
392 return NULL;
394 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
395 sizeof(struct key_vector *) << bits);
397 if (bits == KEYLENGTH)
398 tnode->full_children = 1;
399 else
400 tnode->empty_children = 1ul << bits;
402 tn = tnode->kv;
403 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
404 tn->pos = pos;
405 tn->bits = bits;
406 tn->slen = pos;
408 return tn;
411 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
412 * and no bits are skipped. See discussion in dyntree paper p. 6
414 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
416 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
419 /* Add a child at position i overwriting the old value.
420 * Update the value of full_children and empty_children.
422 static void put_child(struct key_vector *tn, unsigned long i,
423 struct key_vector *n)
425 struct key_vector *chi = get_child(tn, i);
426 int isfull, wasfull;
428 BUG_ON(i >= child_length(tn));
430 /* update emptyChildren, overflow into fullChildren */
431 if (!n && chi)
432 empty_child_inc(tn);
433 if (n && !chi)
434 empty_child_dec(tn);
436 /* update fullChildren */
437 wasfull = tnode_full(tn, chi);
438 isfull = tnode_full(tn, n);
440 if (wasfull && !isfull)
441 tn_info(tn)->full_children--;
442 else if (!wasfull && isfull)
443 tn_info(tn)->full_children++;
445 if (n && (tn->slen < n->slen))
446 tn->slen = n->slen;
448 rcu_assign_pointer(tn->tnode[i], n);
451 static void update_children(struct key_vector *tn)
453 unsigned long i;
455 /* update all of the child parent pointers */
456 for (i = child_length(tn); i;) {
457 struct key_vector *inode = get_child(tn, --i);
459 if (!inode)
460 continue;
462 /* Either update the children of a tnode that
463 * already belongs to us or update the child
464 * to point to ourselves.
466 if (node_parent(inode) == tn)
467 update_children(inode);
468 else
469 node_set_parent(inode, tn);
473 static inline void put_child_root(struct key_vector *tp, t_key key,
474 struct key_vector *n)
476 if (IS_TRIE(tp))
477 rcu_assign_pointer(tp->tnode[0], n);
478 else
479 put_child(tp, get_index(key, tp), n);
482 static inline void tnode_free_init(struct key_vector *tn)
484 tn_info(tn)->rcu.next = NULL;
487 static inline void tnode_free_append(struct key_vector *tn,
488 struct key_vector *n)
490 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
491 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
494 static void tnode_free(struct key_vector *tn)
496 struct callback_head *head = &tn_info(tn)->rcu;
498 while (head) {
499 head = head->next;
500 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
501 node_free(tn);
503 tn = container_of(head, struct tnode, rcu)->kv;
506 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
507 tnode_free_size = 0;
508 synchronize_rcu();
512 static struct key_vector *replace(struct trie *t,
513 struct key_vector *oldtnode,
514 struct key_vector *tn)
516 struct key_vector *tp = node_parent(oldtnode);
517 unsigned long i;
519 /* setup the parent pointer out of and back into this node */
520 NODE_INIT_PARENT(tn, tp);
521 put_child_root(tp, tn->key, tn);
523 /* update all of the child parent pointers */
524 update_children(tn);
526 /* all pointers should be clean so we are done */
527 tnode_free(oldtnode);
529 /* resize children now that oldtnode is freed */
530 for (i = child_length(tn); i;) {
531 struct key_vector *inode = get_child(tn, --i);
533 /* resize child node */
534 if (tnode_full(tn, inode))
535 tn = resize(t, inode);
538 return tp;
541 static struct key_vector *inflate(struct trie *t,
542 struct key_vector *oldtnode)
544 struct key_vector *tn;
545 unsigned long i;
546 t_key m;
548 pr_debug("In inflate\n");
550 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
551 if (!tn)
552 goto notnode;
554 /* prepare oldtnode to be freed */
555 tnode_free_init(oldtnode);
557 /* Assemble all of the pointers in our cluster, in this case that
558 * represents all of the pointers out of our allocated nodes that
559 * point to existing tnodes and the links between our allocated
560 * nodes.
562 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
563 struct key_vector *inode = get_child(oldtnode, --i);
564 struct key_vector *node0, *node1;
565 unsigned long j, k;
567 /* An empty child */
568 if (!inode)
569 continue;
571 /* A leaf or an internal node with skipped bits */
572 if (!tnode_full(oldtnode, inode)) {
573 put_child(tn, get_index(inode->key, tn), inode);
574 continue;
577 /* drop the node in the old tnode free list */
578 tnode_free_append(oldtnode, inode);
580 /* An internal node with two children */
581 if (inode->bits == 1) {
582 put_child(tn, 2 * i + 1, get_child(inode, 1));
583 put_child(tn, 2 * i, get_child(inode, 0));
584 continue;
587 /* We will replace this node 'inode' with two new
588 * ones, 'node0' and 'node1', each with half of the
589 * original children. The two new nodes will have
590 * a position one bit further down the key and this
591 * means that the "significant" part of their keys
592 * (see the discussion near the top of this file)
593 * will differ by one bit, which will be "0" in
594 * node0's key and "1" in node1's key. Since we are
595 * moving the key position by one step, the bit that
596 * we are moving away from - the bit at position
597 * (tn->pos) - is the one that will differ between
598 * node0 and node1. So... we synthesize that bit in the
599 * two new keys.
601 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
602 if (!node1)
603 goto nomem;
604 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
606 tnode_free_append(tn, node1);
607 if (!node0)
608 goto nomem;
609 tnode_free_append(tn, node0);
611 /* populate child pointers in new nodes */
612 for (k = child_length(inode), j = k / 2; j;) {
613 put_child(node1, --j, get_child(inode, --k));
614 put_child(node0, j, get_child(inode, j));
615 put_child(node1, --j, get_child(inode, --k));
616 put_child(node0, j, get_child(inode, j));
619 /* link new nodes to parent */
620 NODE_INIT_PARENT(node1, tn);
621 NODE_INIT_PARENT(node0, tn);
623 /* link parent to nodes */
624 put_child(tn, 2 * i + 1, node1);
625 put_child(tn, 2 * i, node0);
628 /* setup the parent pointers into and out of this node */
629 return replace(t, oldtnode, tn);
630 nomem:
631 /* all pointers should be clean so we are done */
632 tnode_free(tn);
633 notnode:
634 return NULL;
637 static struct key_vector *halve(struct trie *t,
638 struct key_vector *oldtnode)
640 struct key_vector *tn;
641 unsigned long i;
643 pr_debug("In halve\n");
645 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
646 if (!tn)
647 goto notnode;
649 /* prepare oldtnode to be freed */
650 tnode_free_init(oldtnode);
652 /* Assemble all of the pointers in our cluster, in this case that
653 * represents all of the pointers out of our allocated nodes that
654 * point to existing tnodes and the links between our allocated
655 * nodes.
657 for (i = child_length(oldtnode); i;) {
658 struct key_vector *node1 = get_child(oldtnode, --i);
659 struct key_vector *node0 = get_child(oldtnode, --i);
660 struct key_vector *inode;
662 /* At least one of the children is empty */
663 if (!node1 || !node0) {
664 put_child(tn, i / 2, node1 ? : node0);
665 continue;
668 /* Two nonempty children */
669 inode = tnode_new(node0->key, oldtnode->pos, 1);
670 if (!inode)
671 goto nomem;
672 tnode_free_append(tn, inode);
674 /* initialize pointers out of node */
675 put_child(inode, 1, node1);
676 put_child(inode, 0, node0);
677 NODE_INIT_PARENT(inode, tn);
679 /* link parent to node */
680 put_child(tn, i / 2, inode);
683 /* setup the parent pointers into and out of this node */
684 return replace(t, oldtnode, tn);
685 nomem:
686 /* all pointers should be clean so we are done */
687 tnode_free(tn);
688 notnode:
689 return NULL;
692 static struct key_vector *collapse(struct trie *t,
693 struct key_vector *oldtnode)
695 struct key_vector *n, *tp;
696 unsigned long i;
698 /* scan the tnode looking for that one child that might still exist */
699 for (n = NULL, i = child_length(oldtnode); !n && i;)
700 n = get_child(oldtnode, --i);
702 /* compress one level */
703 tp = node_parent(oldtnode);
704 put_child_root(tp, oldtnode->key, n);
705 node_set_parent(n, tp);
707 /* drop dead node */
708 node_free(oldtnode);
710 return tp;
713 static unsigned char update_suffix(struct key_vector *tn)
715 unsigned char slen = tn->pos;
716 unsigned long stride, i;
717 unsigned char slen_max;
719 /* only vector 0 can have a suffix length greater than or equal to
720 * tn->pos + tn->bits, the second highest node will have a suffix
721 * length at most of tn->pos + tn->bits - 1
723 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
725 /* search though the list of children looking for nodes that might
726 * have a suffix greater than the one we currently have. This is
727 * why we start with a stride of 2 since a stride of 1 would
728 * represent the nodes with suffix length equal to tn->pos
730 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
731 struct key_vector *n = get_child(tn, i);
733 if (!n || (n->slen <= slen))
734 continue;
736 /* update stride and slen based on new value */
737 stride <<= (n->slen - slen);
738 slen = n->slen;
739 i &= ~(stride - 1);
741 /* stop searching if we have hit the maximum possible value */
742 if (slen >= slen_max)
743 break;
746 tn->slen = slen;
748 return slen;
751 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
752 * the Helsinki University of Technology and Matti Tikkanen of Nokia
753 * Telecommunications, page 6:
754 * "A node is doubled if the ratio of non-empty children to all
755 * children in the *doubled* node is at least 'high'."
757 * 'high' in this instance is the variable 'inflate_threshold'. It
758 * is expressed as a percentage, so we multiply it with
759 * child_length() and instead of multiplying by 2 (since the
760 * child array will be doubled by inflate()) and multiplying
761 * the left-hand side by 100 (to handle the percentage thing) we
762 * multiply the left-hand side by 50.
764 * The left-hand side may look a bit weird: child_length(tn)
765 * - tn->empty_children is of course the number of non-null children
766 * in the current node. tn->full_children is the number of "full"
767 * children, that is non-null tnodes with a skip value of 0.
768 * All of those will be doubled in the resulting inflated tnode, so
769 * we just count them one extra time here.
771 * A clearer way to write this would be:
773 * to_be_doubled = tn->full_children;
774 * not_to_be_doubled = child_length(tn) - tn->empty_children -
775 * tn->full_children;
777 * new_child_length = child_length(tn) * 2;
779 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
780 * new_child_length;
781 * if (new_fill_factor >= inflate_threshold)
783 * ...and so on, tho it would mess up the while () loop.
785 * anyway,
786 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
787 * inflate_threshold
789 * avoid a division:
790 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
791 * inflate_threshold * new_child_length
793 * expand not_to_be_doubled and to_be_doubled, and shorten:
794 * 100 * (child_length(tn) - tn->empty_children +
795 * tn->full_children) >= inflate_threshold * new_child_length
797 * expand new_child_length:
798 * 100 * (child_length(tn) - tn->empty_children +
799 * tn->full_children) >=
800 * inflate_threshold * child_length(tn) * 2
802 * shorten again:
803 * 50 * (tn->full_children + child_length(tn) -
804 * tn->empty_children) >= inflate_threshold *
805 * child_length(tn)
808 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
810 unsigned long used = child_length(tn);
811 unsigned long threshold = used;
813 /* Keep root node larger */
814 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
815 used -= tn_info(tn)->empty_children;
816 used += tn_info(tn)->full_children;
818 /* if bits == KEYLENGTH then pos = 0, and will fail below */
820 return (used > 1) && tn->pos && ((50 * used) >= threshold);
823 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
825 unsigned long used = child_length(tn);
826 unsigned long threshold = used;
828 /* Keep root node larger */
829 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
830 used -= tn_info(tn)->empty_children;
832 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
834 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
837 static inline bool should_collapse(struct key_vector *tn)
839 unsigned long used = child_length(tn);
841 used -= tn_info(tn)->empty_children;
843 /* account for bits == KEYLENGTH case */
844 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
845 used -= KEY_MAX;
847 /* One child or none, time to drop us from the trie */
848 return used < 2;
851 #define MAX_WORK 10
852 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
854 #ifdef CONFIG_IP_FIB_TRIE_STATS
855 struct trie_use_stats __percpu *stats = t->stats;
856 #endif
857 struct key_vector *tp = node_parent(tn);
858 unsigned long cindex = get_index(tn->key, tp);
859 int max_work = MAX_WORK;
861 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
862 tn, inflate_threshold, halve_threshold);
864 /* track the tnode via the pointer from the parent instead of
865 * doing it ourselves. This way we can let RCU fully do its
866 * thing without us interfering
868 BUG_ON(tn != get_child(tp, cindex));
870 /* Double as long as the resulting node has a number of
871 * nonempty nodes that are above the threshold.
873 while (should_inflate(tp, tn) && max_work) {
874 tp = inflate(t, tn);
875 if (!tp) {
876 #ifdef CONFIG_IP_FIB_TRIE_STATS
877 this_cpu_inc(stats->resize_node_skipped);
878 #endif
879 break;
882 max_work--;
883 tn = get_child(tp, cindex);
886 /* update parent in case inflate failed */
887 tp = node_parent(tn);
889 /* Return if at least one inflate is run */
890 if (max_work != MAX_WORK)
891 return tp;
893 /* Halve as long as the number of empty children in this
894 * node is above threshold.
896 while (should_halve(tp, tn) && max_work) {
897 tp = halve(t, tn);
898 if (!tp) {
899 #ifdef CONFIG_IP_FIB_TRIE_STATS
900 this_cpu_inc(stats->resize_node_skipped);
901 #endif
902 break;
905 max_work--;
906 tn = get_child(tp, cindex);
909 /* Only one child remains */
910 if (should_collapse(tn))
911 return collapse(t, tn);
913 /* update parent in case halve failed */
914 return node_parent(tn);
917 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
919 unsigned char node_slen = tn->slen;
921 while ((node_slen > tn->pos) && (node_slen > slen)) {
922 slen = update_suffix(tn);
923 if (node_slen == slen)
924 break;
926 tn = node_parent(tn);
927 node_slen = tn->slen;
931 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
933 while (tn->slen < slen) {
934 tn->slen = slen;
935 tn = node_parent(tn);
939 /* rcu_read_lock needs to be hold by caller from readside */
940 static struct key_vector *fib_find_node(struct trie *t,
941 struct key_vector **tp, u32 key)
943 struct key_vector *pn, *n = t->kv;
944 unsigned long index = 0;
946 do {
947 pn = n;
948 n = get_child_rcu(n, index);
950 if (!n)
951 break;
953 index = get_cindex(key, n);
955 /* This bit of code is a bit tricky but it combines multiple
956 * checks into a single check. The prefix consists of the
957 * prefix plus zeros for the bits in the cindex. The index
958 * is the difference between the key and this value. From
959 * this we can actually derive several pieces of data.
960 * if (index >= (1ul << bits))
961 * we have a mismatch in skip bits and failed
962 * else
963 * we know the value is cindex
965 * This check is safe even if bits == KEYLENGTH due to the
966 * fact that we can only allocate a node with 32 bits if a
967 * long is greater than 32 bits.
969 if (index >= (1ul << n->bits)) {
970 n = NULL;
971 break;
974 /* keep searching until we find a perfect match leaf or NULL */
975 } while (IS_TNODE(n));
977 *tp = pn;
979 return n;
982 /* Return the first fib alias matching TOS with
983 * priority less than or equal to PRIO.
985 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
986 u8 tos, u32 prio, u32 tb_id)
988 struct fib_alias *fa;
990 if (!fah)
991 return NULL;
993 hlist_for_each_entry(fa, fah, fa_list) {
994 if (fa->fa_slen < slen)
995 continue;
996 if (fa->fa_slen != slen)
997 break;
998 if (fa->tb_id > tb_id)
999 continue;
1000 if (fa->tb_id != tb_id)
1001 break;
1002 if (fa->fa_tos > tos)
1003 continue;
1004 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1005 return fa;
1008 return NULL;
1011 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1013 while (!IS_TRIE(tn))
1014 tn = resize(t, tn);
1017 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1018 struct fib_alias *new, t_key key)
1020 struct key_vector *n, *l;
1022 l = leaf_new(key, new);
1023 if (!l)
1024 goto noleaf;
1026 /* retrieve child from parent node */
1027 n = get_child(tp, get_index(key, tp));
1029 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1031 * Add a new tnode here
1032 * first tnode need some special handling
1033 * leaves us in position for handling as case 3
1035 if (n) {
1036 struct key_vector *tn;
1038 tn = tnode_new(key, __fls(key ^ n->key), 1);
1039 if (!tn)
1040 goto notnode;
1042 /* initialize routes out of node */
1043 NODE_INIT_PARENT(tn, tp);
1044 put_child(tn, get_index(key, tn) ^ 1, n);
1046 /* start adding routes into the node */
1047 put_child_root(tp, key, tn);
1048 node_set_parent(n, tn);
1050 /* parent now has a NULL spot where the leaf can go */
1051 tp = tn;
1054 /* Case 3: n is NULL, and will just insert a new leaf */
1055 node_push_suffix(tp, new->fa_slen);
1056 NODE_INIT_PARENT(l, tp);
1057 put_child_root(tp, key, l);
1058 trie_rebalance(t, tp);
1060 return 0;
1061 notnode:
1062 node_free(l);
1063 noleaf:
1064 return -ENOMEM;
1067 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1068 struct key_vector *l, struct fib_alias *new,
1069 struct fib_alias *fa, t_key key)
1071 if (!l)
1072 return fib_insert_node(t, tp, new, key);
1074 if (fa) {
1075 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1076 } else {
1077 struct fib_alias *last;
1079 hlist_for_each_entry(last, &l->leaf, fa_list) {
1080 if (new->fa_slen < last->fa_slen)
1081 break;
1082 if ((new->fa_slen == last->fa_slen) &&
1083 (new->tb_id > last->tb_id))
1084 break;
1085 fa = last;
1088 if (fa)
1089 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1090 else
1091 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1094 /* if we added to the tail node then we need to update slen */
1095 if (l->slen < new->fa_slen) {
1096 l->slen = new->fa_slen;
1097 node_push_suffix(tp, new->fa_slen);
1100 return 0;
1103 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1105 if (plen > KEYLENGTH) {
1106 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1107 return false;
1110 if ((plen < KEYLENGTH) && (key << plen)) {
1111 NL_SET_ERR_MSG(extack,
1112 "Invalid prefix for given prefix length");
1113 return false;
1116 return true;
1119 /* Caller must hold RTNL. */
1120 int fib_table_insert(struct net *net, struct fib_table *tb,
1121 struct fib_config *cfg, struct netlink_ext_ack *extack)
1123 enum fib_event_type event = FIB_EVENT_ENTRY_ADD;
1124 struct trie *t = (struct trie *)tb->tb_data;
1125 struct fib_alias *fa, *new_fa;
1126 struct key_vector *l, *tp;
1127 u16 nlflags = NLM_F_EXCL;
1128 struct fib_info *fi;
1129 u8 plen = cfg->fc_dst_len;
1130 u8 slen = KEYLENGTH - plen;
1131 u8 tos = cfg->fc_tos;
1132 u32 key;
1133 int err;
1135 key = ntohl(cfg->fc_dst);
1137 if (!fib_valid_key_len(key, plen, extack))
1138 return -EINVAL;
1140 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1142 fi = fib_create_info(cfg, extack);
1143 if (IS_ERR(fi)) {
1144 err = PTR_ERR(fi);
1145 goto err;
1148 l = fib_find_node(t, &tp, key);
1149 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1150 tb->tb_id) : NULL;
1152 /* Now fa, if non-NULL, points to the first fib alias
1153 * with the same keys [prefix,tos,priority], if such key already
1154 * exists or to the node before which we will insert new one.
1156 * If fa is NULL, we will need to allocate a new one and
1157 * insert to the tail of the section matching the suffix length
1158 * of the new alias.
1161 if (fa && fa->fa_tos == tos &&
1162 fa->fa_info->fib_priority == fi->fib_priority) {
1163 struct fib_alias *fa_first, *fa_match;
1165 err = -EEXIST;
1166 if (cfg->fc_nlflags & NLM_F_EXCL)
1167 goto out;
1169 nlflags &= ~NLM_F_EXCL;
1171 /* We have 2 goals:
1172 * 1. Find exact match for type, scope, fib_info to avoid
1173 * duplicate routes
1174 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1176 fa_match = NULL;
1177 fa_first = fa;
1178 hlist_for_each_entry_from(fa, fa_list) {
1179 if ((fa->fa_slen != slen) ||
1180 (fa->tb_id != tb->tb_id) ||
1181 (fa->fa_tos != tos))
1182 break;
1183 if (fa->fa_info->fib_priority != fi->fib_priority)
1184 break;
1185 if (fa->fa_type == cfg->fc_type &&
1186 fa->fa_info == fi) {
1187 fa_match = fa;
1188 break;
1192 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1193 struct fib_info *fi_drop;
1194 u8 state;
1196 nlflags |= NLM_F_REPLACE;
1197 fa = fa_first;
1198 if (fa_match) {
1199 if (fa == fa_match)
1200 err = 0;
1201 goto out;
1203 err = -ENOBUFS;
1204 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1205 if (!new_fa)
1206 goto out;
1208 fi_drop = fa->fa_info;
1209 new_fa->fa_tos = fa->fa_tos;
1210 new_fa->fa_info = fi;
1211 new_fa->fa_type = cfg->fc_type;
1212 state = fa->fa_state;
1213 new_fa->fa_state = state & ~FA_S_ACCESSED;
1214 new_fa->fa_slen = fa->fa_slen;
1215 new_fa->tb_id = tb->tb_id;
1216 new_fa->fa_default = -1;
1218 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
1219 key, plen, new_fa, extack);
1220 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1221 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1223 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1225 alias_free_mem_rcu(fa);
1227 fib_release_info(fi_drop);
1228 if (state & FA_S_ACCESSED)
1229 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1231 goto succeeded;
1233 /* Error if we find a perfect match which
1234 * uses the same scope, type, and nexthop
1235 * information.
1237 if (fa_match)
1238 goto out;
1240 if (cfg->fc_nlflags & NLM_F_APPEND) {
1241 event = FIB_EVENT_ENTRY_APPEND;
1242 nlflags |= NLM_F_APPEND;
1243 } else {
1244 fa = fa_first;
1247 err = -ENOENT;
1248 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1249 goto out;
1251 nlflags |= NLM_F_CREATE;
1252 err = -ENOBUFS;
1253 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1254 if (!new_fa)
1255 goto out;
1257 new_fa->fa_info = fi;
1258 new_fa->fa_tos = tos;
1259 new_fa->fa_type = cfg->fc_type;
1260 new_fa->fa_state = 0;
1261 new_fa->fa_slen = slen;
1262 new_fa->tb_id = tb->tb_id;
1263 new_fa->fa_default = -1;
1265 /* Insert new entry to the list. */
1266 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1267 if (err)
1268 goto out_free_new_fa;
1270 if (!plen)
1271 tb->tb_num_default++;
1273 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1274 call_fib_entry_notifiers(net, event, key, plen, new_fa, extack);
1275 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1276 &cfg->fc_nlinfo, nlflags);
1277 succeeded:
1278 return 0;
1280 out_free_new_fa:
1281 kmem_cache_free(fn_alias_kmem, new_fa);
1282 out:
1283 fib_release_info(fi);
1284 err:
1285 return err;
1288 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1290 t_key prefix = n->key;
1292 return (key ^ prefix) & (prefix | -prefix);
1295 /* should be called with rcu_read_lock */
1296 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1297 struct fib_result *res, int fib_flags)
1299 struct trie *t = (struct trie *) tb->tb_data;
1300 #ifdef CONFIG_IP_FIB_TRIE_STATS
1301 struct trie_use_stats __percpu *stats = t->stats;
1302 #endif
1303 const t_key key = ntohl(flp->daddr);
1304 struct key_vector *n, *pn;
1305 struct fib_alias *fa;
1306 unsigned long index;
1307 t_key cindex;
1309 trace_fib_table_lookup(tb->tb_id, flp);
1311 pn = t->kv;
1312 cindex = 0;
1314 n = get_child_rcu(pn, cindex);
1315 if (!n)
1316 return -EAGAIN;
1318 #ifdef CONFIG_IP_FIB_TRIE_STATS
1319 this_cpu_inc(stats->gets);
1320 #endif
1322 /* Step 1: Travel to the longest prefix match in the trie */
1323 for (;;) {
1324 index = get_cindex(key, n);
1326 /* This bit of code is a bit tricky but it combines multiple
1327 * checks into a single check. The prefix consists of the
1328 * prefix plus zeros for the "bits" in the prefix. The index
1329 * is the difference between the key and this value. From
1330 * this we can actually derive several pieces of data.
1331 * if (index >= (1ul << bits))
1332 * we have a mismatch in skip bits and failed
1333 * else
1334 * we know the value is cindex
1336 * This check is safe even if bits == KEYLENGTH due to the
1337 * fact that we can only allocate a node with 32 bits if a
1338 * long is greater than 32 bits.
1340 if (index >= (1ul << n->bits))
1341 break;
1343 /* we have found a leaf. Prefixes have already been compared */
1344 if (IS_LEAF(n))
1345 goto found;
1347 /* only record pn and cindex if we are going to be chopping
1348 * bits later. Otherwise we are just wasting cycles.
1350 if (n->slen > n->pos) {
1351 pn = n;
1352 cindex = index;
1355 n = get_child_rcu(n, index);
1356 if (unlikely(!n))
1357 goto backtrace;
1360 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1361 for (;;) {
1362 /* record the pointer where our next node pointer is stored */
1363 struct key_vector __rcu **cptr = n->tnode;
1365 /* This test verifies that none of the bits that differ
1366 * between the key and the prefix exist in the region of
1367 * the lsb and higher in the prefix.
1369 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1370 goto backtrace;
1372 /* exit out and process leaf */
1373 if (unlikely(IS_LEAF(n)))
1374 break;
1376 /* Don't bother recording parent info. Since we are in
1377 * prefix match mode we will have to come back to wherever
1378 * we started this traversal anyway
1381 while ((n = rcu_dereference(*cptr)) == NULL) {
1382 backtrace:
1383 #ifdef CONFIG_IP_FIB_TRIE_STATS
1384 if (!n)
1385 this_cpu_inc(stats->null_node_hit);
1386 #endif
1387 /* If we are at cindex 0 there are no more bits for
1388 * us to strip at this level so we must ascend back
1389 * up one level to see if there are any more bits to
1390 * be stripped there.
1392 while (!cindex) {
1393 t_key pkey = pn->key;
1395 /* If we don't have a parent then there is
1396 * nothing for us to do as we do not have any
1397 * further nodes to parse.
1399 if (IS_TRIE(pn))
1400 return -EAGAIN;
1401 #ifdef CONFIG_IP_FIB_TRIE_STATS
1402 this_cpu_inc(stats->backtrack);
1403 #endif
1404 /* Get Child's index */
1405 pn = node_parent_rcu(pn);
1406 cindex = get_index(pkey, pn);
1409 /* strip the least significant bit from the cindex */
1410 cindex &= cindex - 1;
1412 /* grab pointer for next child node */
1413 cptr = &pn->tnode[cindex];
1417 found:
1418 /* this line carries forward the xor from earlier in the function */
1419 index = key ^ n->key;
1421 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1422 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1423 struct fib_info *fi = fa->fa_info;
1424 int nhsel, err;
1426 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1427 if (index >= (1ul << fa->fa_slen))
1428 continue;
1430 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1431 continue;
1432 if (fi->fib_dead)
1433 continue;
1434 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1435 continue;
1436 fib_alias_accessed(fa);
1437 err = fib_props[fa->fa_type].error;
1438 if (unlikely(err < 0)) {
1439 #ifdef CONFIG_IP_FIB_TRIE_STATS
1440 this_cpu_inc(stats->semantic_match_passed);
1441 #endif
1442 return err;
1444 if (fi->fib_flags & RTNH_F_DEAD)
1445 continue;
1446 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1447 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1448 struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1450 if (nh->nh_flags & RTNH_F_DEAD)
1451 continue;
1452 if (in_dev &&
1453 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1454 nh->nh_flags & RTNH_F_LINKDOWN &&
1455 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1456 continue;
1457 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1458 if (flp->flowi4_oif &&
1459 flp->flowi4_oif != nh->nh_oif)
1460 continue;
1463 if (!(fib_flags & FIB_LOOKUP_NOREF))
1464 refcount_inc(&fi->fib_clntref);
1466 res->prefix = htonl(n->key);
1467 res->prefixlen = KEYLENGTH - fa->fa_slen;
1468 res->nh_sel = nhsel;
1469 res->type = fa->fa_type;
1470 res->scope = fi->fib_scope;
1471 res->fi = fi;
1472 res->table = tb;
1473 res->fa_head = &n->leaf;
1474 #ifdef CONFIG_IP_FIB_TRIE_STATS
1475 this_cpu_inc(stats->semantic_match_passed);
1476 #endif
1477 trace_fib_table_lookup_nh(nh);
1479 return err;
1482 #ifdef CONFIG_IP_FIB_TRIE_STATS
1483 this_cpu_inc(stats->semantic_match_miss);
1484 #endif
1485 goto backtrace;
1487 EXPORT_SYMBOL_GPL(fib_table_lookup);
1489 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1490 struct key_vector *l, struct fib_alias *old)
1492 /* record the location of the previous list_info entry */
1493 struct hlist_node **pprev = old->fa_list.pprev;
1494 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1496 /* remove the fib_alias from the list */
1497 hlist_del_rcu(&old->fa_list);
1499 /* if we emptied the list this leaf will be freed and we can sort
1500 * out parent suffix lengths as a part of trie_rebalance
1502 if (hlist_empty(&l->leaf)) {
1503 if (tp->slen == l->slen)
1504 node_pull_suffix(tp, tp->pos);
1505 put_child_root(tp, l->key, NULL);
1506 node_free(l);
1507 trie_rebalance(t, tp);
1508 return;
1511 /* only access fa if it is pointing at the last valid hlist_node */
1512 if (*pprev)
1513 return;
1515 /* update the trie with the latest suffix length */
1516 l->slen = fa->fa_slen;
1517 node_pull_suffix(tp, fa->fa_slen);
1520 /* Caller must hold RTNL. */
1521 int fib_table_delete(struct net *net, struct fib_table *tb,
1522 struct fib_config *cfg, struct netlink_ext_ack *extack)
1524 struct trie *t = (struct trie *) tb->tb_data;
1525 struct fib_alias *fa, *fa_to_delete;
1526 struct key_vector *l, *tp;
1527 u8 plen = cfg->fc_dst_len;
1528 u8 slen = KEYLENGTH - plen;
1529 u8 tos = cfg->fc_tos;
1530 u32 key;
1532 key = ntohl(cfg->fc_dst);
1534 if (!fib_valid_key_len(key, plen, extack))
1535 return -EINVAL;
1537 l = fib_find_node(t, &tp, key);
1538 if (!l)
1539 return -ESRCH;
1541 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1542 if (!fa)
1543 return -ESRCH;
1545 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1547 fa_to_delete = NULL;
1548 hlist_for_each_entry_from(fa, fa_list) {
1549 struct fib_info *fi = fa->fa_info;
1551 if ((fa->fa_slen != slen) ||
1552 (fa->tb_id != tb->tb_id) ||
1553 (fa->fa_tos != tos))
1554 break;
1556 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1557 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1558 fa->fa_info->fib_scope == cfg->fc_scope) &&
1559 (!cfg->fc_prefsrc ||
1560 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1561 (!cfg->fc_protocol ||
1562 fi->fib_protocol == cfg->fc_protocol) &&
1563 fib_nh_match(cfg, fi, extack) == 0 &&
1564 fib_metrics_match(cfg, fi)) {
1565 fa_to_delete = fa;
1566 break;
1570 if (!fa_to_delete)
1571 return -ESRCH;
1573 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1574 fa_to_delete, extack);
1575 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1576 &cfg->fc_nlinfo, 0);
1578 if (!plen)
1579 tb->tb_num_default--;
1581 fib_remove_alias(t, tp, l, fa_to_delete);
1583 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1584 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1586 fib_release_info(fa_to_delete->fa_info);
1587 alias_free_mem_rcu(fa_to_delete);
1588 return 0;
1591 /* Scan for the next leaf starting at the provided key value */
1592 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1594 struct key_vector *pn, *n = *tn;
1595 unsigned long cindex;
1597 /* this loop is meant to try and find the key in the trie */
1598 do {
1599 /* record parent and next child index */
1600 pn = n;
1601 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1603 if (cindex >> pn->bits)
1604 break;
1606 /* descend into the next child */
1607 n = get_child_rcu(pn, cindex++);
1608 if (!n)
1609 break;
1611 /* guarantee forward progress on the keys */
1612 if (IS_LEAF(n) && (n->key >= key))
1613 goto found;
1614 } while (IS_TNODE(n));
1616 /* this loop will search for the next leaf with a greater key */
1617 while (!IS_TRIE(pn)) {
1618 /* if we exhausted the parent node we will need to climb */
1619 if (cindex >= (1ul << pn->bits)) {
1620 t_key pkey = pn->key;
1622 pn = node_parent_rcu(pn);
1623 cindex = get_index(pkey, pn) + 1;
1624 continue;
1627 /* grab the next available node */
1628 n = get_child_rcu(pn, cindex++);
1629 if (!n)
1630 continue;
1632 /* no need to compare keys since we bumped the index */
1633 if (IS_LEAF(n))
1634 goto found;
1636 /* Rescan start scanning in new node */
1637 pn = n;
1638 cindex = 0;
1641 *tn = pn;
1642 return NULL; /* Root of trie */
1643 found:
1644 /* if we are at the limit for keys just return NULL for the tnode */
1645 *tn = pn;
1646 return n;
1649 static void fib_trie_free(struct fib_table *tb)
1651 struct trie *t = (struct trie *)tb->tb_data;
1652 struct key_vector *pn = t->kv;
1653 unsigned long cindex = 1;
1654 struct hlist_node *tmp;
1655 struct fib_alias *fa;
1657 /* walk trie in reverse order and free everything */
1658 for (;;) {
1659 struct key_vector *n;
1661 if (!(cindex--)) {
1662 t_key pkey = pn->key;
1664 if (IS_TRIE(pn))
1665 break;
1667 n = pn;
1668 pn = node_parent(pn);
1670 /* drop emptied tnode */
1671 put_child_root(pn, n->key, NULL);
1672 node_free(n);
1674 cindex = get_index(pkey, pn);
1676 continue;
1679 /* grab the next available node */
1680 n = get_child(pn, cindex);
1681 if (!n)
1682 continue;
1684 if (IS_TNODE(n)) {
1685 /* record pn and cindex for leaf walking */
1686 pn = n;
1687 cindex = 1ul << n->bits;
1689 continue;
1692 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1693 hlist_del_rcu(&fa->fa_list);
1694 alias_free_mem_rcu(fa);
1697 put_child_root(pn, n->key, NULL);
1698 node_free(n);
1701 #ifdef CONFIG_IP_FIB_TRIE_STATS
1702 free_percpu(t->stats);
1703 #endif
1704 kfree(tb);
1707 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1709 struct trie *ot = (struct trie *)oldtb->tb_data;
1710 struct key_vector *l, *tp = ot->kv;
1711 struct fib_table *local_tb;
1712 struct fib_alias *fa;
1713 struct trie *lt;
1714 t_key key = 0;
1716 if (oldtb->tb_data == oldtb->__data)
1717 return oldtb;
1719 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1720 if (!local_tb)
1721 return NULL;
1723 lt = (struct trie *)local_tb->tb_data;
1725 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1726 struct key_vector *local_l = NULL, *local_tp;
1728 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1729 struct fib_alias *new_fa;
1731 if (local_tb->tb_id != fa->tb_id)
1732 continue;
1734 /* clone fa for new local table */
1735 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1736 if (!new_fa)
1737 goto out;
1739 memcpy(new_fa, fa, sizeof(*fa));
1741 /* insert clone into table */
1742 if (!local_l)
1743 local_l = fib_find_node(lt, &local_tp, l->key);
1745 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1746 NULL, l->key)) {
1747 kmem_cache_free(fn_alias_kmem, new_fa);
1748 goto out;
1752 /* stop loop if key wrapped back to 0 */
1753 key = l->key + 1;
1754 if (key < l->key)
1755 break;
1758 return local_tb;
1759 out:
1760 fib_trie_free(local_tb);
1762 return NULL;
1765 /* Caller must hold RTNL */
1766 void fib_table_flush_external(struct fib_table *tb)
1768 struct trie *t = (struct trie *)tb->tb_data;
1769 struct key_vector *pn = t->kv;
1770 unsigned long cindex = 1;
1771 struct hlist_node *tmp;
1772 struct fib_alias *fa;
1774 /* walk trie in reverse order */
1775 for (;;) {
1776 unsigned char slen = 0;
1777 struct key_vector *n;
1779 if (!(cindex--)) {
1780 t_key pkey = pn->key;
1782 /* cannot resize the trie vector */
1783 if (IS_TRIE(pn))
1784 break;
1786 /* update the suffix to address pulled leaves */
1787 if (pn->slen > pn->pos)
1788 update_suffix(pn);
1790 /* resize completed node */
1791 pn = resize(t, pn);
1792 cindex = get_index(pkey, pn);
1794 continue;
1797 /* grab the next available node */
1798 n = get_child(pn, cindex);
1799 if (!n)
1800 continue;
1802 if (IS_TNODE(n)) {
1803 /* record pn and cindex for leaf walking */
1804 pn = n;
1805 cindex = 1ul << n->bits;
1807 continue;
1810 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1811 /* if alias was cloned to local then we just
1812 * need to remove the local copy from main
1814 if (tb->tb_id != fa->tb_id) {
1815 hlist_del_rcu(&fa->fa_list);
1816 alias_free_mem_rcu(fa);
1817 continue;
1820 /* record local slen */
1821 slen = fa->fa_slen;
1824 /* update leaf slen */
1825 n->slen = slen;
1827 if (hlist_empty(&n->leaf)) {
1828 put_child_root(pn, n->key, NULL);
1829 node_free(n);
1834 /* Caller must hold RTNL. */
1835 int fib_table_flush(struct net *net, struct fib_table *tb)
1837 struct trie *t = (struct trie *)tb->tb_data;
1838 struct key_vector *pn = t->kv;
1839 unsigned long cindex = 1;
1840 struct hlist_node *tmp;
1841 struct fib_alias *fa;
1842 int found = 0;
1844 /* walk trie in reverse order */
1845 for (;;) {
1846 unsigned char slen = 0;
1847 struct key_vector *n;
1849 if (!(cindex--)) {
1850 t_key pkey = pn->key;
1852 /* cannot resize the trie vector */
1853 if (IS_TRIE(pn))
1854 break;
1856 /* update the suffix to address pulled leaves */
1857 if (pn->slen > pn->pos)
1858 update_suffix(pn);
1860 /* resize completed node */
1861 pn = resize(t, pn);
1862 cindex = get_index(pkey, pn);
1864 continue;
1867 /* grab the next available node */
1868 n = get_child(pn, cindex);
1869 if (!n)
1870 continue;
1872 if (IS_TNODE(n)) {
1873 /* record pn and cindex for leaf walking */
1874 pn = n;
1875 cindex = 1ul << n->bits;
1877 continue;
1880 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1881 struct fib_info *fi = fa->fa_info;
1883 if (!fi || !(fi->fib_flags & RTNH_F_DEAD) ||
1884 tb->tb_id != fa->tb_id) {
1885 slen = fa->fa_slen;
1886 continue;
1889 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1890 n->key,
1891 KEYLENGTH - fa->fa_slen, fa,
1892 NULL);
1893 hlist_del_rcu(&fa->fa_list);
1894 fib_release_info(fa->fa_info);
1895 alias_free_mem_rcu(fa);
1896 found++;
1899 /* update leaf slen */
1900 n->slen = slen;
1902 if (hlist_empty(&n->leaf)) {
1903 put_child_root(pn, n->key, NULL);
1904 node_free(n);
1908 pr_debug("trie_flush found=%d\n", found);
1909 return found;
1912 static void fib_leaf_notify(struct net *net, struct key_vector *l,
1913 struct fib_table *tb, struct notifier_block *nb)
1915 struct fib_alias *fa;
1917 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1918 struct fib_info *fi = fa->fa_info;
1920 if (!fi)
1921 continue;
1923 /* local and main table can share the same trie,
1924 * so don't notify twice for the same entry.
1926 if (tb->tb_id != fa->tb_id)
1927 continue;
1929 call_fib_entry_notifier(nb, net, FIB_EVENT_ENTRY_ADD, l->key,
1930 KEYLENGTH - fa->fa_slen, fa);
1934 static void fib_table_notify(struct net *net, struct fib_table *tb,
1935 struct notifier_block *nb)
1937 struct trie *t = (struct trie *)tb->tb_data;
1938 struct key_vector *l, *tp = t->kv;
1939 t_key key = 0;
1941 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1942 fib_leaf_notify(net, l, tb, nb);
1944 key = l->key + 1;
1945 /* stop in case of wrap around */
1946 if (key < l->key)
1947 break;
1951 void fib_notify(struct net *net, struct notifier_block *nb)
1953 unsigned int h;
1955 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1956 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
1957 struct fib_table *tb;
1959 hlist_for_each_entry_rcu(tb, head, tb_hlist)
1960 fib_table_notify(net, tb, nb);
1964 static void __trie_free_rcu(struct rcu_head *head)
1966 struct fib_table *tb = container_of(head, struct fib_table, rcu);
1967 #ifdef CONFIG_IP_FIB_TRIE_STATS
1968 struct trie *t = (struct trie *)tb->tb_data;
1970 if (tb->tb_data == tb->__data)
1971 free_percpu(t->stats);
1972 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1973 kfree(tb);
1976 void fib_free_table(struct fib_table *tb)
1978 call_rcu(&tb->rcu, __trie_free_rcu);
1981 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
1982 struct sk_buff *skb, struct netlink_callback *cb)
1984 __be32 xkey = htonl(l->key);
1985 struct fib_alias *fa;
1986 int i, s_i;
1988 s_i = cb->args[4];
1989 i = 0;
1991 /* rcu_read_lock is hold by caller */
1992 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1993 int err;
1995 if (i < s_i) {
1996 i++;
1997 continue;
2000 if (tb->tb_id != fa->tb_id) {
2001 i++;
2002 continue;
2005 err = fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
2006 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2007 tb->tb_id, fa->fa_type,
2008 xkey, KEYLENGTH - fa->fa_slen,
2009 fa->fa_tos, fa->fa_info, NLM_F_MULTI);
2010 if (err < 0) {
2011 cb->args[4] = i;
2012 return err;
2014 i++;
2017 cb->args[4] = i;
2018 return skb->len;
2021 /* rcu_read_lock needs to be hold by caller from readside */
2022 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2023 struct netlink_callback *cb)
2025 struct trie *t = (struct trie *)tb->tb_data;
2026 struct key_vector *l, *tp = t->kv;
2027 /* Dump starting at last key.
2028 * Note: 0.0.0.0/0 (ie default) is first key.
2030 int count = cb->args[2];
2031 t_key key = cb->args[3];
2033 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2034 int err;
2036 err = fn_trie_dump_leaf(l, tb, skb, cb);
2037 if (err < 0) {
2038 cb->args[3] = key;
2039 cb->args[2] = count;
2040 return err;
2043 ++count;
2044 key = l->key + 1;
2046 memset(&cb->args[4], 0,
2047 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2049 /* stop loop if key wrapped back to 0 */
2050 if (key < l->key)
2051 break;
2054 cb->args[3] = key;
2055 cb->args[2] = count;
2057 return skb->len;
2060 void __init fib_trie_init(void)
2062 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2063 sizeof(struct fib_alias),
2064 0, SLAB_PANIC, NULL);
2066 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2067 LEAF_SIZE,
2068 0, SLAB_PANIC, NULL);
2071 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2073 struct fib_table *tb;
2074 struct trie *t;
2075 size_t sz = sizeof(*tb);
2077 if (!alias)
2078 sz += sizeof(struct trie);
2080 tb = kzalloc(sz, GFP_KERNEL);
2081 if (!tb)
2082 return NULL;
2084 tb->tb_id = id;
2085 tb->tb_num_default = 0;
2086 tb->tb_data = (alias ? alias->__data : tb->__data);
2088 if (alias)
2089 return tb;
2091 t = (struct trie *) tb->tb_data;
2092 t->kv[0].pos = KEYLENGTH;
2093 t->kv[0].slen = KEYLENGTH;
2094 #ifdef CONFIG_IP_FIB_TRIE_STATS
2095 t->stats = alloc_percpu(struct trie_use_stats);
2096 if (!t->stats) {
2097 kfree(tb);
2098 tb = NULL;
2100 #endif
2102 return tb;
2105 #ifdef CONFIG_PROC_FS
2106 /* Depth first Trie walk iterator */
2107 struct fib_trie_iter {
2108 struct seq_net_private p;
2109 struct fib_table *tb;
2110 struct key_vector *tnode;
2111 unsigned int index;
2112 unsigned int depth;
2115 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2117 unsigned long cindex = iter->index;
2118 struct key_vector *pn = iter->tnode;
2119 t_key pkey;
2121 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2122 iter->tnode, iter->index, iter->depth);
2124 while (!IS_TRIE(pn)) {
2125 while (cindex < child_length(pn)) {
2126 struct key_vector *n = get_child_rcu(pn, cindex++);
2128 if (!n)
2129 continue;
2131 if (IS_LEAF(n)) {
2132 iter->tnode = pn;
2133 iter->index = cindex;
2134 } else {
2135 /* push down one level */
2136 iter->tnode = n;
2137 iter->index = 0;
2138 ++iter->depth;
2141 return n;
2144 /* Current node exhausted, pop back up */
2145 pkey = pn->key;
2146 pn = node_parent_rcu(pn);
2147 cindex = get_index(pkey, pn) + 1;
2148 --iter->depth;
2151 /* record root node so further searches know we are done */
2152 iter->tnode = pn;
2153 iter->index = 0;
2155 return NULL;
2158 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2159 struct trie *t)
2161 struct key_vector *n, *pn;
2163 if (!t)
2164 return NULL;
2166 pn = t->kv;
2167 n = rcu_dereference(pn->tnode[0]);
2168 if (!n)
2169 return NULL;
2171 if (IS_TNODE(n)) {
2172 iter->tnode = n;
2173 iter->index = 0;
2174 iter->depth = 1;
2175 } else {
2176 iter->tnode = pn;
2177 iter->index = 0;
2178 iter->depth = 0;
2181 return n;
2184 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2186 struct key_vector *n;
2187 struct fib_trie_iter iter;
2189 memset(s, 0, sizeof(*s));
2191 rcu_read_lock();
2192 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2193 if (IS_LEAF(n)) {
2194 struct fib_alias *fa;
2196 s->leaves++;
2197 s->totdepth += iter.depth;
2198 if (iter.depth > s->maxdepth)
2199 s->maxdepth = iter.depth;
2201 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2202 ++s->prefixes;
2203 } else {
2204 s->tnodes++;
2205 if (n->bits < MAX_STAT_DEPTH)
2206 s->nodesizes[n->bits]++;
2207 s->nullpointers += tn_info(n)->empty_children;
2210 rcu_read_unlock();
2214 * This outputs /proc/net/fib_triestats
2216 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2218 unsigned int i, max, pointers, bytes, avdepth;
2220 if (stat->leaves)
2221 avdepth = stat->totdepth*100 / stat->leaves;
2222 else
2223 avdepth = 0;
2225 seq_printf(seq, "\tAver depth: %u.%02d\n",
2226 avdepth / 100, avdepth % 100);
2227 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2229 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2230 bytes = LEAF_SIZE * stat->leaves;
2232 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2233 bytes += sizeof(struct fib_alias) * stat->prefixes;
2235 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2236 bytes += TNODE_SIZE(0) * stat->tnodes;
2238 max = MAX_STAT_DEPTH;
2239 while (max > 0 && stat->nodesizes[max-1] == 0)
2240 max--;
2242 pointers = 0;
2243 for (i = 1; i < max; i++)
2244 if (stat->nodesizes[i] != 0) {
2245 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2246 pointers += (1<<i) * stat->nodesizes[i];
2248 seq_putc(seq, '\n');
2249 seq_printf(seq, "\tPointers: %u\n", pointers);
2251 bytes += sizeof(struct key_vector *) * pointers;
2252 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2253 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2256 #ifdef CONFIG_IP_FIB_TRIE_STATS
2257 static void trie_show_usage(struct seq_file *seq,
2258 const struct trie_use_stats __percpu *stats)
2260 struct trie_use_stats s = { 0 };
2261 int cpu;
2263 /* loop through all of the CPUs and gather up the stats */
2264 for_each_possible_cpu(cpu) {
2265 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2267 s.gets += pcpu->gets;
2268 s.backtrack += pcpu->backtrack;
2269 s.semantic_match_passed += pcpu->semantic_match_passed;
2270 s.semantic_match_miss += pcpu->semantic_match_miss;
2271 s.null_node_hit += pcpu->null_node_hit;
2272 s.resize_node_skipped += pcpu->resize_node_skipped;
2275 seq_printf(seq, "\nCounters:\n---------\n");
2276 seq_printf(seq, "gets = %u\n", s.gets);
2277 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2278 seq_printf(seq, "semantic match passed = %u\n",
2279 s.semantic_match_passed);
2280 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2281 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2282 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2284 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2286 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2288 if (tb->tb_id == RT_TABLE_LOCAL)
2289 seq_puts(seq, "Local:\n");
2290 else if (tb->tb_id == RT_TABLE_MAIN)
2291 seq_puts(seq, "Main:\n");
2292 else
2293 seq_printf(seq, "Id %d:\n", tb->tb_id);
2297 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2299 struct net *net = (struct net *)seq->private;
2300 unsigned int h;
2302 seq_printf(seq,
2303 "Basic info: size of leaf:"
2304 " %zd bytes, size of tnode: %zd bytes.\n",
2305 LEAF_SIZE, TNODE_SIZE(0));
2307 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2308 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2309 struct fib_table *tb;
2311 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2312 struct trie *t = (struct trie *) tb->tb_data;
2313 struct trie_stat stat;
2315 if (!t)
2316 continue;
2318 fib_table_print(seq, tb);
2320 trie_collect_stats(t, &stat);
2321 trie_show_stats(seq, &stat);
2322 #ifdef CONFIG_IP_FIB_TRIE_STATS
2323 trie_show_usage(seq, t->stats);
2324 #endif
2328 return 0;
2331 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2333 return single_open_net(inode, file, fib_triestat_seq_show);
2336 static const struct file_operations fib_triestat_fops = {
2337 .open = fib_triestat_seq_open,
2338 .read = seq_read,
2339 .llseek = seq_lseek,
2340 .release = single_release_net,
2343 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2345 struct fib_trie_iter *iter = seq->private;
2346 struct net *net = seq_file_net(seq);
2347 loff_t idx = 0;
2348 unsigned int h;
2350 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2351 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2352 struct fib_table *tb;
2354 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2355 struct key_vector *n;
2357 for (n = fib_trie_get_first(iter,
2358 (struct trie *) tb->tb_data);
2359 n; n = fib_trie_get_next(iter))
2360 if (pos == idx++) {
2361 iter->tb = tb;
2362 return n;
2367 return NULL;
2370 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2371 __acquires(RCU)
2373 rcu_read_lock();
2374 return fib_trie_get_idx(seq, *pos);
2377 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2379 struct fib_trie_iter *iter = seq->private;
2380 struct net *net = seq_file_net(seq);
2381 struct fib_table *tb = iter->tb;
2382 struct hlist_node *tb_node;
2383 unsigned int h;
2384 struct key_vector *n;
2386 ++*pos;
2387 /* next node in same table */
2388 n = fib_trie_get_next(iter);
2389 if (n)
2390 return n;
2392 /* walk rest of this hash chain */
2393 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2394 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2395 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2396 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2397 if (n)
2398 goto found;
2401 /* new hash chain */
2402 while (++h < FIB_TABLE_HASHSZ) {
2403 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2404 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2405 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2406 if (n)
2407 goto found;
2410 return NULL;
2412 found:
2413 iter->tb = tb;
2414 return n;
2417 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2418 __releases(RCU)
2420 rcu_read_unlock();
2423 static void seq_indent(struct seq_file *seq, int n)
2425 while (n-- > 0)
2426 seq_puts(seq, " ");
2429 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2431 switch (s) {
2432 case RT_SCOPE_UNIVERSE: return "universe";
2433 case RT_SCOPE_SITE: return "site";
2434 case RT_SCOPE_LINK: return "link";
2435 case RT_SCOPE_HOST: return "host";
2436 case RT_SCOPE_NOWHERE: return "nowhere";
2437 default:
2438 snprintf(buf, len, "scope=%d", s);
2439 return buf;
2443 static const char *const rtn_type_names[__RTN_MAX] = {
2444 [RTN_UNSPEC] = "UNSPEC",
2445 [RTN_UNICAST] = "UNICAST",
2446 [RTN_LOCAL] = "LOCAL",
2447 [RTN_BROADCAST] = "BROADCAST",
2448 [RTN_ANYCAST] = "ANYCAST",
2449 [RTN_MULTICAST] = "MULTICAST",
2450 [RTN_BLACKHOLE] = "BLACKHOLE",
2451 [RTN_UNREACHABLE] = "UNREACHABLE",
2452 [RTN_PROHIBIT] = "PROHIBIT",
2453 [RTN_THROW] = "THROW",
2454 [RTN_NAT] = "NAT",
2455 [RTN_XRESOLVE] = "XRESOLVE",
2458 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2460 if (t < __RTN_MAX && rtn_type_names[t])
2461 return rtn_type_names[t];
2462 snprintf(buf, len, "type %u", t);
2463 return buf;
2466 /* Pretty print the trie */
2467 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2469 const struct fib_trie_iter *iter = seq->private;
2470 struct key_vector *n = v;
2472 if (IS_TRIE(node_parent_rcu(n)))
2473 fib_table_print(seq, iter->tb);
2475 if (IS_TNODE(n)) {
2476 __be32 prf = htonl(n->key);
2478 seq_indent(seq, iter->depth-1);
2479 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2480 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2481 tn_info(n)->full_children,
2482 tn_info(n)->empty_children);
2483 } else {
2484 __be32 val = htonl(n->key);
2485 struct fib_alias *fa;
2487 seq_indent(seq, iter->depth);
2488 seq_printf(seq, " |-- %pI4\n", &val);
2490 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2491 char buf1[32], buf2[32];
2493 seq_indent(seq, iter->depth + 1);
2494 seq_printf(seq, " /%zu %s %s",
2495 KEYLENGTH - fa->fa_slen,
2496 rtn_scope(buf1, sizeof(buf1),
2497 fa->fa_info->fib_scope),
2498 rtn_type(buf2, sizeof(buf2),
2499 fa->fa_type));
2500 if (fa->fa_tos)
2501 seq_printf(seq, " tos=%d", fa->fa_tos);
2502 seq_putc(seq, '\n');
2506 return 0;
2509 static const struct seq_operations fib_trie_seq_ops = {
2510 .start = fib_trie_seq_start,
2511 .next = fib_trie_seq_next,
2512 .stop = fib_trie_seq_stop,
2513 .show = fib_trie_seq_show,
2516 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2518 return seq_open_net(inode, file, &fib_trie_seq_ops,
2519 sizeof(struct fib_trie_iter));
2522 static const struct file_operations fib_trie_fops = {
2523 .open = fib_trie_seq_open,
2524 .read = seq_read,
2525 .llseek = seq_lseek,
2526 .release = seq_release_net,
2529 struct fib_route_iter {
2530 struct seq_net_private p;
2531 struct fib_table *main_tb;
2532 struct key_vector *tnode;
2533 loff_t pos;
2534 t_key key;
2537 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2538 loff_t pos)
2540 struct key_vector *l, **tp = &iter->tnode;
2541 t_key key;
2543 /* use cached location of previously found key */
2544 if (iter->pos > 0 && pos >= iter->pos) {
2545 key = iter->key;
2546 } else {
2547 iter->pos = 1;
2548 key = 0;
2551 pos -= iter->pos;
2553 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2554 key = l->key + 1;
2555 iter->pos++;
2556 l = NULL;
2558 /* handle unlikely case of a key wrap */
2559 if (!key)
2560 break;
2563 if (l)
2564 iter->key = l->key; /* remember it */
2565 else
2566 iter->pos = 0; /* forget it */
2568 return l;
2571 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2572 __acquires(RCU)
2574 struct fib_route_iter *iter = seq->private;
2575 struct fib_table *tb;
2576 struct trie *t;
2578 rcu_read_lock();
2580 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2581 if (!tb)
2582 return NULL;
2584 iter->main_tb = tb;
2585 t = (struct trie *)tb->tb_data;
2586 iter->tnode = t->kv;
2588 if (*pos != 0)
2589 return fib_route_get_idx(iter, *pos);
2591 iter->pos = 0;
2592 iter->key = KEY_MAX;
2594 return SEQ_START_TOKEN;
2597 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2599 struct fib_route_iter *iter = seq->private;
2600 struct key_vector *l = NULL;
2601 t_key key = iter->key + 1;
2603 ++*pos;
2605 /* only allow key of 0 for start of sequence */
2606 if ((v == SEQ_START_TOKEN) || key)
2607 l = leaf_walk_rcu(&iter->tnode, key);
2609 if (l) {
2610 iter->key = l->key;
2611 iter->pos++;
2612 } else {
2613 iter->pos = 0;
2616 return l;
2619 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2620 __releases(RCU)
2622 rcu_read_unlock();
2625 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2627 unsigned int flags = 0;
2629 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2630 flags = RTF_REJECT;
2631 if (fi && fi->fib_nh->nh_gw)
2632 flags |= RTF_GATEWAY;
2633 if (mask == htonl(0xFFFFFFFF))
2634 flags |= RTF_HOST;
2635 flags |= RTF_UP;
2636 return flags;
2640 * This outputs /proc/net/route.
2641 * The format of the file is not supposed to be changed
2642 * and needs to be same as fib_hash output to avoid breaking
2643 * legacy utilities
2645 static int fib_route_seq_show(struct seq_file *seq, void *v)
2647 struct fib_route_iter *iter = seq->private;
2648 struct fib_table *tb = iter->main_tb;
2649 struct fib_alias *fa;
2650 struct key_vector *l = v;
2651 __be32 prefix;
2653 if (v == SEQ_START_TOKEN) {
2654 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2655 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2656 "\tWindow\tIRTT");
2657 return 0;
2660 prefix = htonl(l->key);
2662 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2663 const struct fib_info *fi = fa->fa_info;
2664 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2665 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2667 if ((fa->fa_type == RTN_BROADCAST) ||
2668 (fa->fa_type == RTN_MULTICAST))
2669 continue;
2671 if (fa->tb_id != tb->tb_id)
2672 continue;
2674 seq_setwidth(seq, 127);
2676 if (fi)
2677 seq_printf(seq,
2678 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2679 "%d\t%08X\t%d\t%u\t%u",
2680 fi->fib_dev ? fi->fib_dev->name : "*",
2681 prefix,
2682 fi->fib_nh->nh_gw, flags, 0, 0,
2683 fi->fib_priority,
2684 mask,
2685 (fi->fib_advmss ?
2686 fi->fib_advmss + 40 : 0),
2687 fi->fib_window,
2688 fi->fib_rtt >> 3);
2689 else
2690 seq_printf(seq,
2691 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2692 "%d\t%08X\t%d\t%u\t%u",
2693 prefix, 0, flags, 0, 0, 0,
2694 mask, 0, 0, 0);
2696 seq_pad(seq, '\n');
2699 return 0;
2702 static const struct seq_operations fib_route_seq_ops = {
2703 .start = fib_route_seq_start,
2704 .next = fib_route_seq_next,
2705 .stop = fib_route_seq_stop,
2706 .show = fib_route_seq_show,
2709 static int fib_route_seq_open(struct inode *inode, struct file *file)
2711 return seq_open_net(inode, file, &fib_route_seq_ops,
2712 sizeof(struct fib_route_iter));
2715 static const struct file_operations fib_route_fops = {
2716 .open = fib_route_seq_open,
2717 .read = seq_read,
2718 .llseek = seq_lseek,
2719 .release = seq_release_net,
2722 int __net_init fib_proc_init(struct net *net)
2724 if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2725 goto out1;
2727 if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2728 &fib_triestat_fops))
2729 goto out2;
2731 if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2732 goto out3;
2734 return 0;
2736 out3:
2737 remove_proc_entry("fib_triestat", net->proc_net);
2738 out2:
2739 remove_proc_entry("fib_trie", net->proc_net);
2740 out1:
2741 return -ENOMEM;
2744 void __net_exit fib_proc_exit(struct net *net)
2746 remove_proc_entry("fib_trie", net->proc_net);
2747 remove_proc_entry("fib_triestat", net->proc_net);
2748 remove_proc_entry("route", net->proc_net);
2751 #endif /* CONFIG_PROC_FS */