octeontx2-pf: Fix error return code in otx2_probe()
[linux/fpc-iii.git] / net / ipv4 / fib_trie.c
blob4f334b42553853b2cc1de001bd99a53a71de5030
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
4 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
5 * & Swedish University of Agricultural Sciences.
7 * Jens Laas <jens.laas@data.slu.se> Swedish University of
8 * Agricultural Sciences.
10 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
12 * This work is based on the LPC-trie which is originally described in:
14 * An experimental study of compression methods for dynamic tries
15 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
16 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
18 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
19 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
21 * Code from fib_hash has been reused which includes the following header:
23 * INET An implementation of the TCP/IP protocol suite for the LINUX
24 * operating system. INET is implemented using the BSD Socket
25 * interface as the means of communication with the user level.
27 * IPv4 FIB: lookup engine and maintenance routines.
29 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
31 * Substantial contributions to this work comes from:
33 * David S. Miller, <davem@davemloft.net>
34 * Stephen Hemminger <shemminger@osdl.org>
35 * Paul E. McKenney <paulmck@us.ibm.com>
36 * Patrick McHardy <kaber@trash.net>
38 #include <linux/cache.h>
39 #include <linux/uaccess.h>
40 #include <linux/bitops.h>
41 #include <linux/types.h>
42 #include <linux/kernel.h>
43 #include <linux/mm.h>
44 #include <linux/string.h>
45 #include <linux/socket.h>
46 #include <linux/sockios.h>
47 #include <linux/errno.h>
48 #include <linux/in.h>
49 #include <linux/inet.h>
50 #include <linux/inetdevice.h>
51 #include <linux/netdevice.h>
52 #include <linux/if_arp.h>
53 #include <linux/proc_fs.h>
54 #include <linux/rcupdate.h>
55 #include <linux/skbuff.h>
56 #include <linux/netlink.h>
57 #include <linux/init.h>
58 #include <linux/list.h>
59 #include <linux/slab.h>
60 #include <linux/export.h>
61 #include <linux/vmalloc.h>
62 #include <linux/notifier.h>
63 #include <net/net_namespace.h>
64 #include <net/ip.h>
65 #include <net/protocol.h>
66 #include <net/route.h>
67 #include <net/tcp.h>
68 #include <net/sock.h>
69 #include <net/ip_fib.h>
70 #include <net/fib_notifier.h>
71 #include <trace/events/fib.h>
72 #include "fib_lookup.h"
74 static int call_fib_entry_notifier(struct notifier_block *nb,
75 enum fib_event_type event_type, u32 dst,
76 int dst_len, struct fib_alias *fa,
77 struct netlink_ext_ack *extack)
79 struct fib_entry_notifier_info info = {
80 .info.extack = extack,
81 .dst = dst,
82 .dst_len = dst_len,
83 .fi = fa->fa_info,
84 .tos = fa->fa_tos,
85 .type = fa->fa_type,
86 .tb_id = fa->tb_id,
88 return call_fib4_notifier(nb, event_type, &info.info);
91 static int call_fib_entry_notifiers(struct net *net,
92 enum fib_event_type event_type, u32 dst,
93 int dst_len, struct fib_alias *fa,
94 struct netlink_ext_ack *extack)
96 struct fib_entry_notifier_info info = {
97 .info.extack = extack,
98 .dst = dst,
99 .dst_len = dst_len,
100 .fi = fa->fa_info,
101 .tos = fa->fa_tos,
102 .type = fa->fa_type,
103 .tb_id = fa->tb_id,
105 return call_fib4_notifiers(net, event_type, &info.info);
108 #define MAX_STAT_DEPTH 32
110 #define KEYLENGTH (8*sizeof(t_key))
111 #define KEY_MAX ((t_key)~0)
113 typedef unsigned int t_key;
115 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
116 #define IS_TNODE(n) ((n)->bits)
117 #define IS_LEAF(n) (!(n)->bits)
119 struct key_vector {
120 t_key key;
121 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
122 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
123 unsigned char slen;
124 union {
125 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
126 struct hlist_head leaf;
127 /* This array is valid if (pos | bits) > 0 (TNODE) */
128 struct key_vector __rcu *tnode[0];
132 struct tnode {
133 struct rcu_head rcu;
134 t_key empty_children; /* KEYLENGTH bits needed */
135 t_key full_children; /* KEYLENGTH bits needed */
136 struct key_vector __rcu *parent;
137 struct key_vector kv[1];
138 #define tn_bits kv[0].bits
141 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
142 #define LEAF_SIZE TNODE_SIZE(1)
144 #ifdef CONFIG_IP_FIB_TRIE_STATS
145 struct trie_use_stats {
146 unsigned int gets;
147 unsigned int backtrack;
148 unsigned int semantic_match_passed;
149 unsigned int semantic_match_miss;
150 unsigned int null_node_hit;
151 unsigned int resize_node_skipped;
153 #endif
155 struct trie_stat {
156 unsigned int totdepth;
157 unsigned int maxdepth;
158 unsigned int tnodes;
159 unsigned int leaves;
160 unsigned int nullpointers;
161 unsigned int prefixes;
162 unsigned int nodesizes[MAX_STAT_DEPTH];
165 struct trie {
166 struct key_vector kv[1];
167 #ifdef CONFIG_IP_FIB_TRIE_STATS
168 struct trie_use_stats __percpu *stats;
169 #endif
172 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
173 static unsigned int tnode_free_size;
176 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
177 * especially useful before resizing the root node with PREEMPT_NONE configs;
178 * the value was obtained experimentally, aiming to avoid visible slowdown.
180 unsigned int sysctl_fib_sync_mem = 512 * 1024;
181 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
182 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
184 static struct kmem_cache *fn_alias_kmem __ro_after_init;
185 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
187 static inline struct tnode *tn_info(struct key_vector *kv)
189 return container_of(kv, struct tnode, kv[0]);
192 /* caller must hold RTNL */
193 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
194 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
196 /* caller must hold RCU read lock or RTNL */
197 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
198 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
200 /* wrapper for rcu_assign_pointer */
201 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
203 if (n)
204 rcu_assign_pointer(tn_info(n)->parent, tp);
207 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
209 /* This provides us with the number of children in this node, in the case of a
210 * leaf this will return 0 meaning none of the children are accessible.
212 static inline unsigned long child_length(const struct key_vector *tn)
214 return (1ul << tn->bits) & ~(1ul);
217 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
219 static inline unsigned long get_index(t_key key, struct key_vector *kv)
221 unsigned long index = key ^ kv->key;
223 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
224 return 0;
226 return index >> kv->pos;
229 /* To understand this stuff, an understanding of keys and all their bits is
230 * necessary. Every node in the trie has a key associated with it, but not
231 * all of the bits in that key are significant.
233 * Consider a node 'n' and its parent 'tp'.
235 * If n is a leaf, every bit in its key is significant. Its presence is
236 * necessitated by path compression, since during a tree traversal (when
237 * searching for a leaf - unless we are doing an insertion) we will completely
238 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
239 * a potentially successful search, that we have indeed been walking the
240 * correct key path.
242 * Note that we can never "miss" the correct key in the tree if present by
243 * following the wrong path. Path compression ensures that segments of the key
244 * that are the same for all keys with a given prefix are skipped, but the
245 * skipped part *is* identical for each node in the subtrie below the skipped
246 * bit! trie_insert() in this implementation takes care of that.
248 * if n is an internal node - a 'tnode' here, the various parts of its key
249 * have many different meanings.
251 * Example:
252 * _________________________________________________________________
253 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
254 * -----------------------------------------------------------------
255 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
257 * _________________________________________________________________
258 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
259 * -----------------------------------------------------------------
260 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
262 * tp->pos = 22
263 * tp->bits = 3
264 * n->pos = 13
265 * n->bits = 4
267 * First, let's just ignore the bits that come before the parent tp, that is
268 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
269 * point we do not use them for anything.
271 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
272 * index into the parent's child array. That is, they will be used to find
273 * 'n' among tp's children.
275 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
276 * for the node n.
278 * All the bits we have seen so far are significant to the node n. The rest
279 * of the bits are really not needed or indeed known in n->key.
281 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
282 * n's child array, and will of course be different for each child.
284 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
285 * at this point.
288 static const int halve_threshold = 25;
289 static const int inflate_threshold = 50;
290 static const int halve_threshold_root = 15;
291 static const int inflate_threshold_root = 30;
293 static void __alias_free_mem(struct rcu_head *head)
295 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
296 kmem_cache_free(fn_alias_kmem, fa);
299 static inline void alias_free_mem_rcu(struct fib_alias *fa)
301 call_rcu(&fa->rcu, __alias_free_mem);
304 #define TNODE_VMALLOC_MAX \
305 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
307 static void __node_free_rcu(struct rcu_head *head)
309 struct tnode *n = container_of(head, struct tnode, rcu);
311 if (!n->tn_bits)
312 kmem_cache_free(trie_leaf_kmem, n);
313 else
314 kvfree(n);
317 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
319 static struct tnode *tnode_alloc(int bits)
321 size_t size;
323 /* verify bits is within bounds */
324 if (bits > TNODE_VMALLOC_MAX)
325 return NULL;
327 /* determine size and verify it is non-zero and didn't overflow */
328 size = TNODE_SIZE(1ul << bits);
330 if (size <= PAGE_SIZE)
331 return kzalloc(size, GFP_KERNEL);
332 else
333 return vzalloc(size);
336 static inline void empty_child_inc(struct key_vector *n)
338 tn_info(n)->empty_children++;
340 if (!tn_info(n)->empty_children)
341 tn_info(n)->full_children++;
344 static inline void empty_child_dec(struct key_vector *n)
346 if (!tn_info(n)->empty_children)
347 tn_info(n)->full_children--;
349 tn_info(n)->empty_children--;
352 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
354 struct key_vector *l;
355 struct tnode *kv;
357 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
358 if (!kv)
359 return NULL;
361 /* initialize key vector */
362 l = kv->kv;
363 l->key = key;
364 l->pos = 0;
365 l->bits = 0;
366 l->slen = fa->fa_slen;
368 /* link leaf to fib alias */
369 INIT_HLIST_HEAD(&l->leaf);
370 hlist_add_head(&fa->fa_list, &l->leaf);
372 return l;
375 static struct key_vector *tnode_new(t_key key, int pos, int bits)
377 unsigned int shift = pos + bits;
378 struct key_vector *tn;
379 struct tnode *tnode;
381 /* verify bits and pos their msb bits clear and values are valid */
382 BUG_ON(!bits || (shift > KEYLENGTH));
384 tnode = tnode_alloc(bits);
385 if (!tnode)
386 return NULL;
388 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
389 sizeof(struct key_vector *) << bits);
391 if (bits == KEYLENGTH)
392 tnode->full_children = 1;
393 else
394 tnode->empty_children = 1ul << bits;
396 tn = tnode->kv;
397 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
398 tn->pos = pos;
399 tn->bits = bits;
400 tn->slen = pos;
402 return tn;
405 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
406 * and no bits are skipped. See discussion in dyntree paper p. 6
408 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
410 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
413 /* Add a child at position i overwriting the old value.
414 * Update the value of full_children and empty_children.
416 static void put_child(struct key_vector *tn, unsigned long i,
417 struct key_vector *n)
419 struct key_vector *chi = get_child(tn, i);
420 int isfull, wasfull;
422 BUG_ON(i >= child_length(tn));
424 /* update emptyChildren, overflow into fullChildren */
425 if (!n && chi)
426 empty_child_inc(tn);
427 if (n && !chi)
428 empty_child_dec(tn);
430 /* update fullChildren */
431 wasfull = tnode_full(tn, chi);
432 isfull = tnode_full(tn, n);
434 if (wasfull && !isfull)
435 tn_info(tn)->full_children--;
436 else if (!wasfull && isfull)
437 tn_info(tn)->full_children++;
439 if (n && (tn->slen < n->slen))
440 tn->slen = n->slen;
442 rcu_assign_pointer(tn->tnode[i], n);
445 static void update_children(struct key_vector *tn)
447 unsigned long i;
449 /* update all of the child parent pointers */
450 for (i = child_length(tn); i;) {
451 struct key_vector *inode = get_child(tn, --i);
453 if (!inode)
454 continue;
456 /* Either update the children of a tnode that
457 * already belongs to us or update the child
458 * to point to ourselves.
460 if (node_parent(inode) == tn)
461 update_children(inode);
462 else
463 node_set_parent(inode, tn);
467 static inline void put_child_root(struct key_vector *tp, t_key key,
468 struct key_vector *n)
470 if (IS_TRIE(tp))
471 rcu_assign_pointer(tp->tnode[0], n);
472 else
473 put_child(tp, get_index(key, tp), n);
476 static inline void tnode_free_init(struct key_vector *tn)
478 tn_info(tn)->rcu.next = NULL;
481 static inline void tnode_free_append(struct key_vector *tn,
482 struct key_vector *n)
484 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
485 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
488 static void tnode_free(struct key_vector *tn)
490 struct callback_head *head = &tn_info(tn)->rcu;
492 while (head) {
493 head = head->next;
494 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
495 node_free(tn);
497 tn = container_of(head, struct tnode, rcu)->kv;
500 if (tnode_free_size >= sysctl_fib_sync_mem) {
501 tnode_free_size = 0;
502 synchronize_rcu();
506 static struct key_vector *replace(struct trie *t,
507 struct key_vector *oldtnode,
508 struct key_vector *tn)
510 struct key_vector *tp = node_parent(oldtnode);
511 unsigned long i;
513 /* setup the parent pointer out of and back into this node */
514 NODE_INIT_PARENT(tn, tp);
515 put_child_root(tp, tn->key, tn);
517 /* update all of the child parent pointers */
518 update_children(tn);
520 /* all pointers should be clean so we are done */
521 tnode_free(oldtnode);
523 /* resize children now that oldtnode is freed */
524 for (i = child_length(tn); i;) {
525 struct key_vector *inode = get_child(tn, --i);
527 /* resize child node */
528 if (tnode_full(tn, inode))
529 tn = resize(t, inode);
532 return tp;
535 static struct key_vector *inflate(struct trie *t,
536 struct key_vector *oldtnode)
538 struct key_vector *tn;
539 unsigned long i;
540 t_key m;
542 pr_debug("In inflate\n");
544 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
545 if (!tn)
546 goto notnode;
548 /* prepare oldtnode to be freed */
549 tnode_free_init(oldtnode);
551 /* Assemble all of the pointers in our cluster, in this case that
552 * represents all of the pointers out of our allocated nodes that
553 * point to existing tnodes and the links between our allocated
554 * nodes.
556 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
557 struct key_vector *inode = get_child(oldtnode, --i);
558 struct key_vector *node0, *node1;
559 unsigned long j, k;
561 /* An empty child */
562 if (!inode)
563 continue;
565 /* A leaf or an internal node with skipped bits */
566 if (!tnode_full(oldtnode, inode)) {
567 put_child(tn, get_index(inode->key, tn), inode);
568 continue;
571 /* drop the node in the old tnode free list */
572 tnode_free_append(oldtnode, inode);
574 /* An internal node with two children */
575 if (inode->bits == 1) {
576 put_child(tn, 2 * i + 1, get_child(inode, 1));
577 put_child(tn, 2 * i, get_child(inode, 0));
578 continue;
581 /* We will replace this node 'inode' with two new
582 * ones, 'node0' and 'node1', each with half of the
583 * original children. The two new nodes will have
584 * a position one bit further down the key and this
585 * means that the "significant" part of their keys
586 * (see the discussion near the top of this file)
587 * will differ by one bit, which will be "0" in
588 * node0's key and "1" in node1's key. Since we are
589 * moving the key position by one step, the bit that
590 * we are moving away from - the bit at position
591 * (tn->pos) - is the one that will differ between
592 * node0 and node1. So... we synthesize that bit in the
593 * two new keys.
595 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
596 if (!node1)
597 goto nomem;
598 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
600 tnode_free_append(tn, node1);
601 if (!node0)
602 goto nomem;
603 tnode_free_append(tn, node0);
605 /* populate child pointers in new nodes */
606 for (k = child_length(inode), j = k / 2; j;) {
607 put_child(node1, --j, get_child(inode, --k));
608 put_child(node0, j, get_child(inode, j));
609 put_child(node1, --j, get_child(inode, --k));
610 put_child(node0, j, get_child(inode, j));
613 /* link new nodes to parent */
614 NODE_INIT_PARENT(node1, tn);
615 NODE_INIT_PARENT(node0, tn);
617 /* link parent to nodes */
618 put_child(tn, 2 * i + 1, node1);
619 put_child(tn, 2 * i, node0);
622 /* setup the parent pointers into and out of this node */
623 return replace(t, oldtnode, tn);
624 nomem:
625 /* all pointers should be clean so we are done */
626 tnode_free(tn);
627 notnode:
628 return NULL;
631 static struct key_vector *halve(struct trie *t,
632 struct key_vector *oldtnode)
634 struct key_vector *tn;
635 unsigned long i;
637 pr_debug("In halve\n");
639 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
640 if (!tn)
641 goto notnode;
643 /* prepare oldtnode to be freed */
644 tnode_free_init(oldtnode);
646 /* Assemble all of the pointers in our cluster, in this case that
647 * represents all of the pointers out of our allocated nodes that
648 * point to existing tnodes and the links between our allocated
649 * nodes.
651 for (i = child_length(oldtnode); i;) {
652 struct key_vector *node1 = get_child(oldtnode, --i);
653 struct key_vector *node0 = get_child(oldtnode, --i);
654 struct key_vector *inode;
656 /* At least one of the children is empty */
657 if (!node1 || !node0) {
658 put_child(tn, i / 2, node1 ? : node0);
659 continue;
662 /* Two nonempty children */
663 inode = tnode_new(node0->key, oldtnode->pos, 1);
664 if (!inode)
665 goto nomem;
666 tnode_free_append(tn, inode);
668 /* initialize pointers out of node */
669 put_child(inode, 1, node1);
670 put_child(inode, 0, node0);
671 NODE_INIT_PARENT(inode, tn);
673 /* link parent to node */
674 put_child(tn, i / 2, inode);
677 /* setup the parent pointers into and out of this node */
678 return replace(t, oldtnode, tn);
679 nomem:
680 /* all pointers should be clean so we are done */
681 tnode_free(tn);
682 notnode:
683 return NULL;
686 static struct key_vector *collapse(struct trie *t,
687 struct key_vector *oldtnode)
689 struct key_vector *n, *tp;
690 unsigned long i;
692 /* scan the tnode looking for that one child that might still exist */
693 for (n = NULL, i = child_length(oldtnode); !n && i;)
694 n = get_child(oldtnode, --i);
696 /* compress one level */
697 tp = node_parent(oldtnode);
698 put_child_root(tp, oldtnode->key, n);
699 node_set_parent(n, tp);
701 /* drop dead node */
702 node_free(oldtnode);
704 return tp;
707 static unsigned char update_suffix(struct key_vector *tn)
709 unsigned char slen = tn->pos;
710 unsigned long stride, i;
711 unsigned char slen_max;
713 /* only vector 0 can have a suffix length greater than or equal to
714 * tn->pos + tn->bits, the second highest node will have a suffix
715 * length at most of tn->pos + tn->bits - 1
717 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
719 /* search though the list of children looking for nodes that might
720 * have a suffix greater than the one we currently have. This is
721 * why we start with a stride of 2 since a stride of 1 would
722 * represent the nodes with suffix length equal to tn->pos
724 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
725 struct key_vector *n = get_child(tn, i);
727 if (!n || (n->slen <= slen))
728 continue;
730 /* update stride and slen based on new value */
731 stride <<= (n->slen - slen);
732 slen = n->slen;
733 i &= ~(stride - 1);
735 /* stop searching if we have hit the maximum possible value */
736 if (slen >= slen_max)
737 break;
740 tn->slen = slen;
742 return slen;
745 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
746 * the Helsinki University of Technology and Matti Tikkanen of Nokia
747 * Telecommunications, page 6:
748 * "A node is doubled if the ratio of non-empty children to all
749 * children in the *doubled* node is at least 'high'."
751 * 'high' in this instance is the variable 'inflate_threshold'. It
752 * is expressed as a percentage, so we multiply it with
753 * child_length() and instead of multiplying by 2 (since the
754 * child array will be doubled by inflate()) and multiplying
755 * the left-hand side by 100 (to handle the percentage thing) we
756 * multiply the left-hand side by 50.
758 * The left-hand side may look a bit weird: child_length(tn)
759 * - tn->empty_children is of course the number of non-null children
760 * in the current node. tn->full_children is the number of "full"
761 * children, that is non-null tnodes with a skip value of 0.
762 * All of those will be doubled in the resulting inflated tnode, so
763 * we just count them one extra time here.
765 * A clearer way to write this would be:
767 * to_be_doubled = tn->full_children;
768 * not_to_be_doubled = child_length(tn) - tn->empty_children -
769 * tn->full_children;
771 * new_child_length = child_length(tn) * 2;
773 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
774 * new_child_length;
775 * if (new_fill_factor >= inflate_threshold)
777 * ...and so on, tho it would mess up the while () loop.
779 * anyway,
780 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
781 * inflate_threshold
783 * avoid a division:
784 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
785 * inflate_threshold * new_child_length
787 * expand not_to_be_doubled and to_be_doubled, and shorten:
788 * 100 * (child_length(tn) - tn->empty_children +
789 * tn->full_children) >= inflate_threshold * new_child_length
791 * expand new_child_length:
792 * 100 * (child_length(tn) - tn->empty_children +
793 * tn->full_children) >=
794 * inflate_threshold * child_length(tn) * 2
796 * shorten again:
797 * 50 * (tn->full_children + child_length(tn) -
798 * tn->empty_children) >= inflate_threshold *
799 * child_length(tn)
802 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
804 unsigned long used = child_length(tn);
805 unsigned long threshold = used;
807 /* Keep root node larger */
808 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
809 used -= tn_info(tn)->empty_children;
810 used += tn_info(tn)->full_children;
812 /* if bits == KEYLENGTH then pos = 0, and will fail below */
814 return (used > 1) && tn->pos && ((50 * used) >= threshold);
817 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
819 unsigned long used = child_length(tn);
820 unsigned long threshold = used;
822 /* Keep root node larger */
823 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
824 used -= tn_info(tn)->empty_children;
826 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
828 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
831 static inline bool should_collapse(struct key_vector *tn)
833 unsigned long used = child_length(tn);
835 used -= tn_info(tn)->empty_children;
837 /* account for bits == KEYLENGTH case */
838 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
839 used -= KEY_MAX;
841 /* One child or none, time to drop us from the trie */
842 return used < 2;
845 #define MAX_WORK 10
846 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
848 #ifdef CONFIG_IP_FIB_TRIE_STATS
849 struct trie_use_stats __percpu *stats = t->stats;
850 #endif
851 struct key_vector *tp = node_parent(tn);
852 unsigned long cindex = get_index(tn->key, tp);
853 int max_work = MAX_WORK;
855 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
856 tn, inflate_threshold, halve_threshold);
858 /* track the tnode via the pointer from the parent instead of
859 * doing it ourselves. This way we can let RCU fully do its
860 * thing without us interfering
862 BUG_ON(tn != get_child(tp, cindex));
864 /* Double as long as the resulting node has a number of
865 * nonempty nodes that are above the threshold.
867 while (should_inflate(tp, tn) && max_work) {
868 tp = inflate(t, tn);
869 if (!tp) {
870 #ifdef CONFIG_IP_FIB_TRIE_STATS
871 this_cpu_inc(stats->resize_node_skipped);
872 #endif
873 break;
876 max_work--;
877 tn = get_child(tp, cindex);
880 /* update parent in case inflate failed */
881 tp = node_parent(tn);
883 /* Return if at least one inflate is run */
884 if (max_work != MAX_WORK)
885 return tp;
887 /* Halve as long as the number of empty children in this
888 * node is above threshold.
890 while (should_halve(tp, tn) && max_work) {
891 tp = halve(t, tn);
892 if (!tp) {
893 #ifdef CONFIG_IP_FIB_TRIE_STATS
894 this_cpu_inc(stats->resize_node_skipped);
895 #endif
896 break;
899 max_work--;
900 tn = get_child(tp, cindex);
903 /* Only one child remains */
904 if (should_collapse(tn))
905 return collapse(t, tn);
907 /* update parent in case halve failed */
908 return node_parent(tn);
911 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
913 unsigned char node_slen = tn->slen;
915 while ((node_slen > tn->pos) && (node_slen > slen)) {
916 slen = update_suffix(tn);
917 if (node_slen == slen)
918 break;
920 tn = node_parent(tn);
921 node_slen = tn->slen;
925 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
927 while (tn->slen < slen) {
928 tn->slen = slen;
929 tn = node_parent(tn);
933 /* rcu_read_lock needs to be hold by caller from readside */
934 static struct key_vector *fib_find_node(struct trie *t,
935 struct key_vector **tp, u32 key)
937 struct key_vector *pn, *n = t->kv;
938 unsigned long index = 0;
940 do {
941 pn = n;
942 n = get_child_rcu(n, index);
944 if (!n)
945 break;
947 index = get_cindex(key, n);
949 /* This bit of code is a bit tricky but it combines multiple
950 * checks into a single check. The prefix consists of the
951 * prefix plus zeros for the bits in the cindex. The index
952 * is the difference between the key and this value. From
953 * this we can actually derive several pieces of data.
954 * if (index >= (1ul << bits))
955 * we have a mismatch in skip bits and failed
956 * else
957 * we know the value is cindex
959 * This check is safe even if bits == KEYLENGTH due to the
960 * fact that we can only allocate a node with 32 bits if a
961 * long is greater than 32 bits.
963 if (index >= (1ul << n->bits)) {
964 n = NULL;
965 break;
968 /* keep searching until we find a perfect match leaf or NULL */
969 } while (IS_TNODE(n));
971 *tp = pn;
973 return n;
976 /* Return the first fib alias matching TOS with
977 * priority less than or equal to PRIO.
978 * If 'find_first' is set, return the first matching
979 * fib alias, regardless of TOS and priority.
981 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
982 u8 tos, u32 prio, u32 tb_id,
983 bool find_first)
985 struct fib_alias *fa;
987 if (!fah)
988 return NULL;
990 hlist_for_each_entry(fa, fah, fa_list) {
991 if (fa->fa_slen < slen)
992 continue;
993 if (fa->fa_slen != slen)
994 break;
995 if (fa->tb_id > tb_id)
996 continue;
997 if (fa->tb_id != tb_id)
998 break;
999 if (find_first)
1000 return fa;
1001 if (fa->fa_tos > tos)
1002 continue;
1003 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1004 return fa;
1007 return NULL;
1010 static struct fib_alias *
1011 fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1013 u8 slen = KEYLENGTH - fri->dst_len;
1014 struct key_vector *l, *tp;
1015 struct fib_table *tb;
1016 struct fib_alias *fa;
1017 struct trie *t;
1019 tb = fib_get_table(net, fri->tb_id);
1020 if (!tb)
1021 return NULL;
1023 t = (struct trie *)tb->tb_data;
1024 l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1025 if (!l)
1026 return NULL;
1028 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1029 if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1030 fa->fa_tos == fri->tos && fa->fa_info == fri->fi &&
1031 fa->fa_type == fri->type)
1032 return fa;
1035 return NULL;
1038 void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1040 struct fib_alias *fa_match;
1042 rcu_read_lock();
1044 fa_match = fib_find_matching_alias(net, fri);
1045 if (!fa_match)
1046 goto out;
1048 fa_match->offload = fri->offload;
1049 fa_match->trap = fri->trap;
1051 out:
1052 rcu_read_unlock();
1054 EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1056 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1058 while (!IS_TRIE(tn))
1059 tn = resize(t, tn);
1062 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1063 struct fib_alias *new, t_key key)
1065 struct key_vector *n, *l;
1067 l = leaf_new(key, new);
1068 if (!l)
1069 goto noleaf;
1071 /* retrieve child from parent node */
1072 n = get_child(tp, get_index(key, tp));
1074 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1076 * Add a new tnode here
1077 * first tnode need some special handling
1078 * leaves us in position for handling as case 3
1080 if (n) {
1081 struct key_vector *tn;
1083 tn = tnode_new(key, __fls(key ^ n->key), 1);
1084 if (!tn)
1085 goto notnode;
1087 /* initialize routes out of node */
1088 NODE_INIT_PARENT(tn, tp);
1089 put_child(tn, get_index(key, tn) ^ 1, n);
1091 /* start adding routes into the node */
1092 put_child_root(tp, key, tn);
1093 node_set_parent(n, tn);
1095 /* parent now has a NULL spot where the leaf can go */
1096 tp = tn;
1099 /* Case 3: n is NULL, and will just insert a new leaf */
1100 node_push_suffix(tp, new->fa_slen);
1101 NODE_INIT_PARENT(l, tp);
1102 put_child_root(tp, key, l);
1103 trie_rebalance(t, tp);
1105 return 0;
1106 notnode:
1107 node_free(l);
1108 noleaf:
1109 return -ENOMEM;
1112 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1113 struct key_vector *l, struct fib_alias *new,
1114 struct fib_alias *fa, t_key key)
1116 if (!l)
1117 return fib_insert_node(t, tp, new, key);
1119 if (fa) {
1120 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1121 } else {
1122 struct fib_alias *last;
1124 hlist_for_each_entry(last, &l->leaf, fa_list) {
1125 if (new->fa_slen < last->fa_slen)
1126 break;
1127 if ((new->fa_slen == last->fa_slen) &&
1128 (new->tb_id > last->tb_id))
1129 break;
1130 fa = last;
1133 if (fa)
1134 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1135 else
1136 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1139 /* if we added to the tail node then we need to update slen */
1140 if (l->slen < new->fa_slen) {
1141 l->slen = new->fa_slen;
1142 node_push_suffix(tp, new->fa_slen);
1145 return 0;
1148 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1150 if (plen > KEYLENGTH) {
1151 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1152 return false;
1155 if ((plen < KEYLENGTH) && (key << plen)) {
1156 NL_SET_ERR_MSG(extack,
1157 "Invalid prefix for given prefix length");
1158 return false;
1161 return true;
1164 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1165 struct key_vector *l, struct fib_alias *old);
1167 /* Caller must hold RTNL. */
1168 int fib_table_insert(struct net *net, struct fib_table *tb,
1169 struct fib_config *cfg, struct netlink_ext_ack *extack)
1171 struct trie *t = (struct trie *)tb->tb_data;
1172 struct fib_alias *fa, *new_fa;
1173 struct key_vector *l, *tp;
1174 u16 nlflags = NLM_F_EXCL;
1175 struct fib_info *fi;
1176 u8 plen = cfg->fc_dst_len;
1177 u8 slen = KEYLENGTH - plen;
1178 u8 tos = cfg->fc_tos;
1179 u32 key;
1180 int err;
1182 key = ntohl(cfg->fc_dst);
1184 if (!fib_valid_key_len(key, plen, extack))
1185 return -EINVAL;
1187 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1189 fi = fib_create_info(cfg, extack);
1190 if (IS_ERR(fi)) {
1191 err = PTR_ERR(fi);
1192 goto err;
1195 l = fib_find_node(t, &tp, key);
1196 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1197 tb->tb_id, false) : NULL;
1199 /* Now fa, if non-NULL, points to the first fib alias
1200 * with the same keys [prefix,tos,priority], if such key already
1201 * exists or to the node before which we will insert new one.
1203 * If fa is NULL, we will need to allocate a new one and
1204 * insert to the tail of the section matching the suffix length
1205 * of the new alias.
1208 if (fa && fa->fa_tos == tos &&
1209 fa->fa_info->fib_priority == fi->fib_priority) {
1210 struct fib_alias *fa_first, *fa_match;
1212 err = -EEXIST;
1213 if (cfg->fc_nlflags & NLM_F_EXCL)
1214 goto out;
1216 nlflags &= ~NLM_F_EXCL;
1218 /* We have 2 goals:
1219 * 1. Find exact match for type, scope, fib_info to avoid
1220 * duplicate routes
1221 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1223 fa_match = NULL;
1224 fa_first = fa;
1225 hlist_for_each_entry_from(fa, fa_list) {
1226 if ((fa->fa_slen != slen) ||
1227 (fa->tb_id != tb->tb_id) ||
1228 (fa->fa_tos != tos))
1229 break;
1230 if (fa->fa_info->fib_priority != fi->fib_priority)
1231 break;
1232 if (fa->fa_type == cfg->fc_type &&
1233 fa->fa_info == fi) {
1234 fa_match = fa;
1235 break;
1239 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1240 struct fib_info *fi_drop;
1241 u8 state;
1243 nlflags |= NLM_F_REPLACE;
1244 fa = fa_first;
1245 if (fa_match) {
1246 if (fa == fa_match)
1247 err = 0;
1248 goto out;
1250 err = -ENOBUFS;
1251 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1252 if (!new_fa)
1253 goto out;
1255 fi_drop = fa->fa_info;
1256 new_fa->fa_tos = fa->fa_tos;
1257 new_fa->fa_info = fi;
1258 new_fa->fa_type = cfg->fc_type;
1259 state = fa->fa_state;
1260 new_fa->fa_state = state & ~FA_S_ACCESSED;
1261 new_fa->fa_slen = fa->fa_slen;
1262 new_fa->tb_id = tb->tb_id;
1263 new_fa->fa_default = -1;
1264 new_fa->offload = 0;
1265 new_fa->trap = 0;
1267 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1269 if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1270 tb->tb_id, true) == new_fa) {
1271 enum fib_event_type fib_event;
1273 fib_event = FIB_EVENT_ENTRY_REPLACE;
1274 err = call_fib_entry_notifiers(net, fib_event,
1275 key, plen,
1276 new_fa, extack);
1277 if (err) {
1278 hlist_replace_rcu(&new_fa->fa_list,
1279 &fa->fa_list);
1280 goto out_free_new_fa;
1284 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1285 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1287 alias_free_mem_rcu(fa);
1289 fib_release_info(fi_drop);
1290 if (state & FA_S_ACCESSED)
1291 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1293 goto succeeded;
1295 /* Error if we find a perfect match which
1296 * uses the same scope, type, and nexthop
1297 * information.
1299 if (fa_match)
1300 goto out;
1302 if (cfg->fc_nlflags & NLM_F_APPEND)
1303 nlflags |= NLM_F_APPEND;
1304 else
1305 fa = fa_first;
1307 err = -ENOENT;
1308 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1309 goto out;
1311 nlflags |= NLM_F_CREATE;
1312 err = -ENOBUFS;
1313 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1314 if (!new_fa)
1315 goto out;
1317 new_fa->fa_info = fi;
1318 new_fa->fa_tos = tos;
1319 new_fa->fa_type = cfg->fc_type;
1320 new_fa->fa_state = 0;
1321 new_fa->fa_slen = slen;
1322 new_fa->tb_id = tb->tb_id;
1323 new_fa->fa_default = -1;
1324 new_fa->offload = 0;
1325 new_fa->trap = 0;
1327 /* Insert new entry to the list. */
1328 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1329 if (err)
1330 goto out_free_new_fa;
1332 /* The alias was already inserted, so the node must exist. */
1333 l = l ? l : fib_find_node(t, &tp, key);
1334 if (WARN_ON_ONCE(!l))
1335 goto out_free_new_fa;
1337 if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1338 new_fa) {
1339 enum fib_event_type fib_event;
1341 fib_event = FIB_EVENT_ENTRY_REPLACE;
1342 err = call_fib_entry_notifiers(net, fib_event, key, plen,
1343 new_fa, extack);
1344 if (err)
1345 goto out_remove_new_fa;
1348 if (!plen)
1349 tb->tb_num_default++;
1351 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1352 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1353 &cfg->fc_nlinfo, nlflags);
1354 succeeded:
1355 return 0;
1357 out_remove_new_fa:
1358 fib_remove_alias(t, tp, l, new_fa);
1359 out_free_new_fa:
1360 kmem_cache_free(fn_alias_kmem, new_fa);
1361 out:
1362 fib_release_info(fi);
1363 err:
1364 return err;
1367 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1369 t_key prefix = n->key;
1371 return (key ^ prefix) & (prefix | -prefix);
1374 /* should be called with rcu_read_lock */
1375 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1376 struct fib_result *res, int fib_flags)
1378 struct trie *t = (struct trie *) tb->tb_data;
1379 #ifdef CONFIG_IP_FIB_TRIE_STATS
1380 struct trie_use_stats __percpu *stats = t->stats;
1381 #endif
1382 const t_key key = ntohl(flp->daddr);
1383 struct key_vector *n, *pn;
1384 struct fib_alias *fa;
1385 unsigned long index;
1386 t_key cindex;
1388 pn = t->kv;
1389 cindex = 0;
1391 n = get_child_rcu(pn, cindex);
1392 if (!n) {
1393 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1394 return -EAGAIN;
1397 #ifdef CONFIG_IP_FIB_TRIE_STATS
1398 this_cpu_inc(stats->gets);
1399 #endif
1401 /* Step 1: Travel to the longest prefix match in the trie */
1402 for (;;) {
1403 index = get_cindex(key, n);
1405 /* This bit of code is a bit tricky but it combines multiple
1406 * checks into a single check. The prefix consists of the
1407 * prefix plus zeros for the "bits" in the prefix. The index
1408 * is the difference between the key and this value. From
1409 * this we can actually derive several pieces of data.
1410 * if (index >= (1ul << bits))
1411 * we have a mismatch in skip bits and failed
1412 * else
1413 * we know the value is cindex
1415 * This check is safe even if bits == KEYLENGTH due to the
1416 * fact that we can only allocate a node with 32 bits if a
1417 * long is greater than 32 bits.
1419 if (index >= (1ul << n->bits))
1420 break;
1422 /* we have found a leaf. Prefixes have already been compared */
1423 if (IS_LEAF(n))
1424 goto found;
1426 /* only record pn and cindex if we are going to be chopping
1427 * bits later. Otherwise we are just wasting cycles.
1429 if (n->slen > n->pos) {
1430 pn = n;
1431 cindex = index;
1434 n = get_child_rcu(n, index);
1435 if (unlikely(!n))
1436 goto backtrace;
1439 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1440 for (;;) {
1441 /* record the pointer where our next node pointer is stored */
1442 struct key_vector __rcu **cptr = n->tnode;
1444 /* This test verifies that none of the bits that differ
1445 * between the key and the prefix exist in the region of
1446 * the lsb and higher in the prefix.
1448 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1449 goto backtrace;
1451 /* exit out and process leaf */
1452 if (unlikely(IS_LEAF(n)))
1453 break;
1455 /* Don't bother recording parent info. Since we are in
1456 * prefix match mode we will have to come back to wherever
1457 * we started this traversal anyway
1460 while ((n = rcu_dereference(*cptr)) == NULL) {
1461 backtrace:
1462 #ifdef CONFIG_IP_FIB_TRIE_STATS
1463 if (!n)
1464 this_cpu_inc(stats->null_node_hit);
1465 #endif
1466 /* If we are at cindex 0 there are no more bits for
1467 * us to strip at this level so we must ascend back
1468 * up one level to see if there are any more bits to
1469 * be stripped there.
1471 while (!cindex) {
1472 t_key pkey = pn->key;
1474 /* If we don't have a parent then there is
1475 * nothing for us to do as we do not have any
1476 * further nodes to parse.
1478 if (IS_TRIE(pn)) {
1479 trace_fib_table_lookup(tb->tb_id, flp,
1480 NULL, -EAGAIN);
1481 return -EAGAIN;
1483 #ifdef CONFIG_IP_FIB_TRIE_STATS
1484 this_cpu_inc(stats->backtrack);
1485 #endif
1486 /* Get Child's index */
1487 pn = node_parent_rcu(pn);
1488 cindex = get_index(pkey, pn);
1491 /* strip the least significant bit from the cindex */
1492 cindex &= cindex - 1;
1494 /* grab pointer for next child node */
1495 cptr = &pn->tnode[cindex];
1499 found:
1500 /* this line carries forward the xor from earlier in the function */
1501 index = key ^ n->key;
1503 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1504 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1505 struct fib_info *fi = fa->fa_info;
1506 int nhsel, err;
1508 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1509 if (index >= (1ul << fa->fa_slen))
1510 continue;
1512 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1513 continue;
1514 if (fi->fib_dead)
1515 continue;
1516 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1517 continue;
1518 fib_alias_accessed(fa);
1519 err = fib_props[fa->fa_type].error;
1520 if (unlikely(err < 0)) {
1521 out_reject:
1522 #ifdef CONFIG_IP_FIB_TRIE_STATS
1523 this_cpu_inc(stats->semantic_match_passed);
1524 #endif
1525 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1526 return err;
1528 if (fi->fib_flags & RTNH_F_DEAD)
1529 continue;
1531 if (unlikely(fi->nh && nexthop_is_blackhole(fi->nh))) {
1532 err = fib_props[RTN_BLACKHOLE].error;
1533 goto out_reject;
1536 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1537 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
1539 if (nhc->nhc_flags & RTNH_F_DEAD)
1540 continue;
1541 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1542 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1543 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1544 continue;
1545 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1546 if (flp->flowi4_oif &&
1547 flp->flowi4_oif != nhc->nhc_oif)
1548 continue;
1551 if (!(fib_flags & FIB_LOOKUP_NOREF))
1552 refcount_inc(&fi->fib_clntref);
1554 res->prefix = htonl(n->key);
1555 res->prefixlen = KEYLENGTH - fa->fa_slen;
1556 res->nh_sel = nhsel;
1557 res->nhc = nhc;
1558 res->type = fa->fa_type;
1559 res->scope = fi->fib_scope;
1560 res->fi = fi;
1561 res->table = tb;
1562 res->fa_head = &n->leaf;
1563 #ifdef CONFIG_IP_FIB_TRIE_STATS
1564 this_cpu_inc(stats->semantic_match_passed);
1565 #endif
1566 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1568 return err;
1571 #ifdef CONFIG_IP_FIB_TRIE_STATS
1572 this_cpu_inc(stats->semantic_match_miss);
1573 #endif
1574 goto backtrace;
1576 EXPORT_SYMBOL_GPL(fib_table_lookup);
1578 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1579 struct key_vector *l, struct fib_alias *old)
1581 /* record the location of the previous list_info entry */
1582 struct hlist_node **pprev = old->fa_list.pprev;
1583 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1585 /* remove the fib_alias from the list */
1586 hlist_del_rcu(&old->fa_list);
1588 /* if we emptied the list this leaf will be freed and we can sort
1589 * out parent suffix lengths as a part of trie_rebalance
1591 if (hlist_empty(&l->leaf)) {
1592 if (tp->slen == l->slen)
1593 node_pull_suffix(tp, tp->pos);
1594 put_child_root(tp, l->key, NULL);
1595 node_free(l);
1596 trie_rebalance(t, tp);
1597 return;
1600 /* only access fa if it is pointing at the last valid hlist_node */
1601 if (*pprev)
1602 return;
1604 /* update the trie with the latest suffix length */
1605 l->slen = fa->fa_slen;
1606 node_pull_suffix(tp, fa->fa_slen);
1609 static void fib_notify_alias_delete(struct net *net, u32 key,
1610 struct hlist_head *fah,
1611 struct fib_alias *fa_to_delete,
1612 struct netlink_ext_ack *extack)
1614 struct fib_alias *fa_next, *fa_to_notify;
1615 u32 tb_id = fa_to_delete->tb_id;
1616 u8 slen = fa_to_delete->fa_slen;
1617 enum fib_event_type fib_event;
1619 /* Do not notify if we do not care about the route. */
1620 if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1621 return;
1623 /* Determine if the route should be replaced by the next route in the
1624 * list.
1626 fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1627 struct fib_alias, fa_list);
1628 if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1629 fib_event = FIB_EVENT_ENTRY_REPLACE;
1630 fa_to_notify = fa_next;
1631 } else {
1632 fib_event = FIB_EVENT_ENTRY_DEL;
1633 fa_to_notify = fa_to_delete;
1635 call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1636 fa_to_notify, extack);
1639 /* Caller must hold RTNL. */
1640 int fib_table_delete(struct net *net, struct fib_table *tb,
1641 struct fib_config *cfg, struct netlink_ext_ack *extack)
1643 struct trie *t = (struct trie *) tb->tb_data;
1644 struct fib_alias *fa, *fa_to_delete;
1645 struct key_vector *l, *tp;
1646 u8 plen = cfg->fc_dst_len;
1647 u8 slen = KEYLENGTH - plen;
1648 u8 tos = cfg->fc_tos;
1649 u32 key;
1651 key = ntohl(cfg->fc_dst);
1653 if (!fib_valid_key_len(key, plen, extack))
1654 return -EINVAL;
1656 l = fib_find_node(t, &tp, key);
1657 if (!l)
1658 return -ESRCH;
1660 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id, false);
1661 if (!fa)
1662 return -ESRCH;
1664 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1666 fa_to_delete = NULL;
1667 hlist_for_each_entry_from(fa, fa_list) {
1668 struct fib_info *fi = fa->fa_info;
1670 if ((fa->fa_slen != slen) ||
1671 (fa->tb_id != tb->tb_id) ||
1672 (fa->fa_tos != tos))
1673 break;
1675 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1676 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1677 fa->fa_info->fib_scope == cfg->fc_scope) &&
1678 (!cfg->fc_prefsrc ||
1679 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1680 (!cfg->fc_protocol ||
1681 fi->fib_protocol == cfg->fc_protocol) &&
1682 fib_nh_match(net, cfg, fi, extack) == 0 &&
1683 fib_metrics_match(cfg, fi)) {
1684 fa_to_delete = fa;
1685 break;
1689 if (!fa_to_delete)
1690 return -ESRCH;
1692 fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1693 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1694 &cfg->fc_nlinfo, 0);
1696 if (!plen)
1697 tb->tb_num_default--;
1699 fib_remove_alias(t, tp, l, fa_to_delete);
1701 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1702 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1704 fib_release_info(fa_to_delete->fa_info);
1705 alias_free_mem_rcu(fa_to_delete);
1706 return 0;
1709 /* Scan for the next leaf starting at the provided key value */
1710 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1712 struct key_vector *pn, *n = *tn;
1713 unsigned long cindex;
1715 /* this loop is meant to try and find the key in the trie */
1716 do {
1717 /* record parent and next child index */
1718 pn = n;
1719 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1721 if (cindex >> pn->bits)
1722 break;
1724 /* descend into the next child */
1725 n = get_child_rcu(pn, cindex++);
1726 if (!n)
1727 break;
1729 /* guarantee forward progress on the keys */
1730 if (IS_LEAF(n) && (n->key >= key))
1731 goto found;
1732 } while (IS_TNODE(n));
1734 /* this loop will search for the next leaf with a greater key */
1735 while (!IS_TRIE(pn)) {
1736 /* if we exhausted the parent node we will need to climb */
1737 if (cindex >= (1ul << pn->bits)) {
1738 t_key pkey = pn->key;
1740 pn = node_parent_rcu(pn);
1741 cindex = get_index(pkey, pn) + 1;
1742 continue;
1745 /* grab the next available node */
1746 n = get_child_rcu(pn, cindex++);
1747 if (!n)
1748 continue;
1750 /* no need to compare keys since we bumped the index */
1751 if (IS_LEAF(n))
1752 goto found;
1754 /* Rescan start scanning in new node */
1755 pn = n;
1756 cindex = 0;
1759 *tn = pn;
1760 return NULL; /* Root of trie */
1761 found:
1762 /* if we are at the limit for keys just return NULL for the tnode */
1763 *tn = pn;
1764 return n;
1767 static void fib_trie_free(struct fib_table *tb)
1769 struct trie *t = (struct trie *)tb->tb_data;
1770 struct key_vector *pn = t->kv;
1771 unsigned long cindex = 1;
1772 struct hlist_node *tmp;
1773 struct fib_alias *fa;
1775 /* walk trie in reverse order and free everything */
1776 for (;;) {
1777 struct key_vector *n;
1779 if (!(cindex--)) {
1780 t_key pkey = pn->key;
1782 if (IS_TRIE(pn))
1783 break;
1785 n = pn;
1786 pn = node_parent(pn);
1788 /* drop emptied tnode */
1789 put_child_root(pn, n->key, NULL);
1790 node_free(n);
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 hlist_del_rcu(&fa->fa_list);
1812 alias_free_mem_rcu(fa);
1815 put_child_root(pn, n->key, NULL);
1816 node_free(n);
1819 #ifdef CONFIG_IP_FIB_TRIE_STATS
1820 free_percpu(t->stats);
1821 #endif
1822 kfree(tb);
1825 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1827 struct trie *ot = (struct trie *)oldtb->tb_data;
1828 struct key_vector *l, *tp = ot->kv;
1829 struct fib_table *local_tb;
1830 struct fib_alias *fa;
1831 struct trie *lt;
1832 t_key key = 0;
1834 if (oldtb->tb_data == oldtb->__data)
1835 return oldtb;
1837 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1838 if (!local_tb)
1839 return NULL;
1841 lt = (struct trie *)local_tb->tb_data;
1843 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1844 struct key_vector *local_l = NULL, *local_tp;
1846 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1847 struct fib_alias *new_fa;
1849 if (local_tb->tb_id != fa->tb_id)
1850 continue;
1852 /* clone fa for new local table */
1853 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1854 if (!new_fa)
1855 goto out;
1857 memcpy(new_fa, fa, sizeof(*fa));
1859 /* insert clone into table */
1860 if (!local_l)
1861 local_l = fib_find_node(lt, &local_tp, l->key);
1863 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1864 NULL, l->key)) {
1865 kmem_cache_free(fn_alias_kmem, new_fa);
1866 goto out;
1870 /* stop loop if key wrapped back to 0 */
1871 key = l->key + 1;
1872 if (key < l->key)
1873 break;
1876 return local_tb;
1877 out:
1878 fib_trie_free(local_tb);
1880 return NULL;
1883 /* Caller must hold RTNL */
1884 void fib_table_flush_external(struct fib_table *tb)
1886 struct trie *t = (struct trie *)tb->tb_data;
1887 struct key_vector *pn = t->kv;
1888 unsigned long cindex = 1;
1889 struct hlist_node *tmp;
1890 struct fib_alias *fa;
1892 /* walk trie in reverse order */
1893 for (;;) {
1894 unsigned char slen = 0;
1895 struct key_vector *n;
1897 if (!(cindex--)) {
1898 t_key pkey = pn->key;
1900 /* cannot resize the trie vector */
1901 if (IS_TRIE(pn))
1902 break;
1904 /* update the suffix to address pulled leaves */
1905 if (pn->slen > pn->pos)
1906 update_suffix(pn);
1908 /* resize completed node */
1909 pn = resize(t, pn);
1910 cindex = get_index(pkey, pn);
1912 continue;
1915 /* grab the next available node */
1916 n = get_child(pn, cindex);
1917 if (!n)
1918 continue;
1920 if (IS_TNODE(n)) {
1921 /* record pn and cindex for leaf walking */
1922 pn = n;
1923 cindex = 1ul << n->bits;
1925 continue;
1928 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1929 /* if alias was cloned to local then we just
1930 * need to remove the local copy from main
1932 if (tb->tb_id != fa->tb_id) {
1933 hlist_del_rcu(&fa->fa_list);
1934 alias_free_mem_rcu(fa);
1935 continue;
1938 /* record local slen */
1939 slen = fa->fa_slen;
1942 /* update leaf slen */
1943 n->slen = slen;
1945 if (hlist_empty(&n->leaf)) {
1946 put_child_root(pn, n->key, NULL);
1947 node_free(n);
1952 /* Caller must hold RTNL. */
1953 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1955 struct trie *t = (struct trie *)tb->tb_data;
1956 struct key_vector *pn = t->kv;
1957 unsigned long cindex = 1;
1958 struct hlist_node *tmp;
1959 struct fib_alias *fa;
1960 int found = 0;
1962 /* walk trie in reverse order */
1963 for (;;) {
1964 unsigned char slen = 0;
1965 struct key_vector *n;
1967 if (!(cindex--)) {
1968 t_key pkey = pn->key;
1970 /* cannot resize the trie vector */
1971 if (IS_TRIE(pn))
1972 break;
1974 /* update the suffix to address pulled leaves */
1975 if (pn->slen > pn->pos)
1976 update_suffix(pn);
1978 /* resize completed node */
1979 pn = resize(t, pn);
1980 cindex = get_index(pkey, pn);
1982 continue;
1985 /* grab the next available node */
1986 n = get_child(pn, cindex);
1987 if (!n)
1988 continue;
1990 if (IS_TNODE(n)) {
1991 /* record pn and cindex for leaf walking */
1992 pn = n;
1993 cindex = 1ul << n->bits;
1995 continue;
1998 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1999 struct fib_info *fi = fa->fa_info;
2001 if (!fi || tb->tb_id != fa->tb_id ||
2002 (!(fi->fib_flags & RTNH_F_DEAD) &&
2003 !fib_props[fa->fa_type].error)) {
2004 slen = fa->fa_slen;
2005 continue;
2008 /* Do not flush error routes if network namespace is
2009 * not being dismantled
2011 if (!flush_all && fib_props[fa->fa_type].error) {
2012 slen = fa->fa_slen;
2013 continue;
2016 fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2017 NULL);
2018 hlist_del_rcu(&fa->fa_list);
2019 fib_release_info(fa->fa_info);
2020 alias_free_mem_rcu(fa);
2021 found++;
2024 /* update leaf slen */
2025 n->slen = slen;
2027 if (hlist_empty(&n->leaf)) {
2028 put_child_root(pn, n->key, NULL);
2029 node_free(n);
2033 pr_debug("trie_flush found=%d\n", found);
2034 return found;
2037 /* derived from fib_trie_free */
2038 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2039 struct nl_info *info)
2041 struct trie *t = (struct trie *)tb->tb_data;
2042 struct key_vector *pn = t->kv;
2043 unsigned long cindex = 1;
2044 struct fib_alias *fa;
2046 for (;;) {
2047 struct key_vector *n;
2049 if (!(cindex--)) {
2050 t_key pkey = pn->key;
2052 if (IS_TRIE(pn))
2053 break;
2055 pn = node_parent(pn);
2056 cindex = get_index(pkey, pn);
2057 continue;
2060 /* grab the next available node */
2061 n = get_child(pn, cindex);
2062 if (!n)
2063 continue;
2065 if (IS_TNODE(n)) {
2066 /* record pn and cindex for leaf walking */
2067 pn = n;
2068 cindex = 1ul << n->bits;
2070 continue;
2073 hlist_for_each_entry(fa, &n->leaf, fa_list) {
2074 struct fib_info *fi = fa->fa_info;
2076 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2077 continue;
2079 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2080 KEYLENGTH - fa->fa_slen, tb->tb_id,
2081 info, NLM_F_REPLACE);
2083 /* call_fib_entry_notifiers will be removed when
2084 * in-kernel notifier is implemented and supported
2085 * for nexthop objects
2087 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
2088 n->key,
2089 KEYLENGTH - fa->fa_slen, fa,
2090 NULL);
2095 void fib_info_notify_update(struct net *net, struct nl_info *info)
2097 unsigned int h;
2099 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2100 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2101 struct fib_table *tb;
2103 hlist_for_each_entry_rcu(tb, head, tb_hlist)
2104 __fib_info_notify_update(net, tb, info);
2108 static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2109 struct notifier_block *nb,
2110 struct netlink_ext_ack *extack)
2112 struct fib_alias *fa;
2113 int last_slen = -1;
2114 int err;
2116 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2117 struct fib_info *fi = fa->fa_info;
2119 if (!fi)
2120 continue;
2122 /* local and main table can share the same trie,
2123 * so don't notify twice for the same entry.
2125 if (tb->tb_id != fa->tb_id)
2126 continue;
2128 if (fa->fa_slen == last_slen)
2129 continue;
2131 last_slen = fa->fa_slen;
2132 err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2133 l->key, KEYLENGTH - fa->fa_slen,
2134 fa, extack);
2135 if (err)
2136 return err;
2138 return 0;
2141 static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2142 struct netlink_ext_ack *extack)
2144 struct trie *t = (struct trie *)tb->tb_data;
2145 struct key_vector *l, *tp = t->kv;
2146 t_key key = 0;
2147 int err;
2149 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2150 err = fib_leaf_notify(l, tb, nb, extack);
2151 if (err)
2152 return err;
2154 key = l->key + 1;
2155 /* stop in case of wrap around */
2156 if (key < l->key)
2157 break;
2159 return 0;
2162 int fib_notify(struct net *net, struct notifier_block *nb,
2163 struct netlink_ext_ack *extack)
2165 unsigned int h;
2166 int err;
2168 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2169 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2170 struct fib_table *tb;
2172 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2173 err = fib_table_notify(tb, nb, extack);
2174 if (err)
2175 return err;
2178 return 0;
2181 static void __trie_free_rcu(struct rcu_head *head)
2183 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2184 #ifdef CONFIG_IP_FIB_TRIE_STATS
2185 struct trie *t = (struct trie *)tb->tb_data;
2187 if (tb->tb_data == tb->__data)
2188 free_percpu(t->stats);
2189 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2190 kfree(tb);
2193 void fib_free_table(struct fib_table *tb)
2195 call_rcu(&tb->rcu, __trie_free_rcu);
2198 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2199 struct sk_buff *skb, struct netlink_callback *cb,
2200 struct fib_dump_filter *filter)
2202 unsigned int flags = NLM_F_MULTI;
2203 __be32 xkey = htonl(l->key);
2204 int i, s_i, i_fa, s_fa, err;
2205 struct fib_alias *fa;
2207 if (filter->filter_set ||
2208 !filter->dump_exceptions || !filter->dump_routes)
2209 flags |= NLM_F_DUMP_FILTERED;
2211 s_i = cb->args[4];
2212 s_fa = cb->args[5];
2213 i = 0;
2215 /* rcu_read_lock is hold by caller */
2216 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2217 struct fib_info *fi = fa->fa_info;
2219 if (i < s_i)
2220 goto next;
2222 i_fa = 0;
2224 if (tb->tb_id != fa->tb_id)
2225 goto next;
2227 if (filter->filter_set) {
2228 if (filter->rt_type && fa->fa_type != filter->rt_type)
2229 goto next;
2231 if ((filter->protocol &&
2232 fi->fib_protocol != filter->protocol))
2233 goto next;
2235 if (filter->dev &&
2236 !fib_info_nh_uses_dev(fi, filter->dev))
2237 goto next;
2240 if (filter->dump_routes) {
2241 if (!s_fa) {
2242 struct fib_rt_info fri;
2244 fri.fi = fi;
2245 fri.tb_id = tb->tb_id;
2246 fri.dst = xkey;
2247 fri.dst_len = KEYLENGTH - fa->fa_slen;
2248 fri.tos = fa->fa_tos;
2249 fri.type = fa->fa_type;
2250 fri.offload = fa->offload;
2251 fri.trap = fa->trap;
2252 err = fib_dump_info(skb,
2253 NETLINK_CB(cb->skb).portid,
2254 cb->nlh->nlmsg_seq,
2255 RTM_NEWROUTE, &fri, flags);
2256 if (err < 0)
2257 goto stop;
2260 i_fa++;
2263 if (filter->dump_exceptions) {
2264 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2265 &i_fa, s_fa, flags);
2266 if (err < 0)
2267 goto stop;
2270 next:
2271 i++;
2274 cb->args[4] = i;
2275 return skb->len;
2277 stop:
2278 cb->args[4] = i;
2279 cb->args[5] = i_fa;
2280 return err;
2283 /* rcu_read_lock needs to be hold by caller from readside */
2284 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2285 struct netlink_callback *cb, struct fib_dump_filter *filter)
2287 struct trie *t = (struct trie *)tb->tb_data;
2288 struct key_vector *l, *tp = t->kv;
2289 /* Dump starting at last key.
2290 * Note: 0.0.0.0/0 (ie default) is first key.
2292 int count = cb->args[2];
2293 t_key key = cb->args[3];
2295 /* First time here, count and key are both always 0. Count > 0
2296 * and key == 0 means the dump has wrapped around and we are done.
2298 if (count && !key)
2299 return skb->len;
2301 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2302 int err;
2304 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2305 if (err < 0) {
2306 cb->args[3] = key;
2307 cb->args[2] = count;
2308 return err;
2311 ++count;
2312 key = l->key + 1;
2314 memset(&cb->args[4], 0,
2315 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2317 /* stop loop if key wrapped back to 0 */
2318 if (key < l->key)
2319 break;
2322 cb->args[3] = key;
2323 cb->args[2] = count;
2325 return skb->len;
2328 void __init fib_trie_init(void)
2330 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2331 sizeof(struct fib_alias),
2332 0, SLAB_PANIC, NULL);
2334 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2335 LEAF_SIZE,
2336 0, SLAB_PANIC, NULL);
2339 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2341 struct fib_table *tb;
2342 struct trie *t;
2343 size_t sz = sizeof(*tb);
2345 if (!alias)
2346 sz += sizeof(struct trie);
2348 tb = kzalloc(sz, GFP_KERNEL);
2349 if (!tb)
2350 return NULL;
2352 tb->tb_id = id;
2353 tb->tb_num_default = 0;
2354 tb->tb_data = (alias ? alias->__data : tb->__data);
2356 if (alias)
2357 return tb;
2359 t = (struct trie *) tb->tb_data;
2360 t->kv[0].pos = KEYLENGTH;
2361 t->kv[0].slen = KEYLENGTH;
2362 #ifdef CONFIG_IP_FIB_TRIE_STATS
2363 t->stats = alloc_percpu(struct trie_use_stats);
2364 if (!t->stats) {
2365 kfree(tb);
2366 tb = NULL;
2368 #endif
2370 return tb;
2373 #ifdef CONFIG_PROC_FS
2374 /* Depth first Trie walk iterator */
2375 struct fib_trie_iter {
2376 struct seq_net_private p;
2377 struct fib_table *tb;
2378 struct key_vector *tnode;
2379 unsigned int index;
2380 unsigned int depth;
2383 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2385 unsigned long cindex = iter->index;
2386 struct key_vector *pn = iter->tnode;
2387 t_key pkey;
2389 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2390 iter->tnode, iter->index, iter->depth);
2392 while (!IS_TRIE(pn)) {
2393 while (cindex < child_length(pn)) {
2394 struct key_vector *n = get_child_rcu(pn, cindex++);
2396 if (!n)
2397 continue;
2399 if (IS_LEAF(n)) {
2400 iter->tnode = pn;
2401 iter->index = cindex;
2402 } else {
2403 /* push down one level */
2404 iter->tnode = n;
2405 iter->index = 0;
2406 ++iter->depth;
2409 return n;
2412 /* Current node exhausted, pop back up */
2413 pkey = pn->key;
2414 pn = node_parent_rcu(pn);
2415 cindex = get_index(pkey, pn) + 1;
2416 --iter->depth;
2419 /* record root node so further searches know we are done */
2420 iter->tnode = pn;
2421 iter->index = 0;
2423 return NULL;
2426 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2427 struct trie *t)
2429 struct key_vector *n, *pn;
2431 if (!t)
2432 return NULL;
2434 pn = t->kv;
2435 n = rcu_dereference(pn->tnode[0]);
2436 if (!n)
2437 return NULL;
2439 if (IS_TNODE(n)) {
2440 iter->tnode = n;
2441 iter->index = 0;
2442 iter->depth = 1;
2443 } else {
2444 iter->tnode = pn;
2445 iter->index = 0;
2446 iter->depth = 0;
2449 return n;
2452 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2454 struct key_vector *n;
2455 struct fib_trie_iter iter;
2457 memset(s, 0, sizeof(*s));
2459 rcu_read_lock();
2460 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2461 if (IS_LEAF(n)) {
2462 struct fib_alias *fa;
2464 s->leaves++;
2465 s->totdepth += iter.depth;
2466 if (iter.depth > s->maxdepth)
2467 s->maxdepth = iter.depth;
2469 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2470 ++s->prefixes;
2471 } else {
2472 s->tnodes++;
2473 if (n->bits < MAX_STAT_DEPTH)
2474 s->nodesizes[n->bits]++;
2475 s->nullpointers += tn_info(n)->empty_children;
2478 rcu_read_unlock();
2482 * This outputs /proc/net/fib_triestats
2484 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2486 unsigned int i, max, pointers, bytes, avdepth;
2488 if (stat->leaves)
2489 avdepth = stat->totdepth*100 / stat->leaves;
2490 else
2491 avdepth = 0;
2493 seq_printf(seq, "\tAver depth: %u.%02d\n",
2494 avdepth / 100, avdepth % 100);
2495 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2497 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2498 bytes = LEAF_SIZE * stat->leaves;
2500 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2501 bytes += sizeof(struct fib_alias) * stat->prefixes;
2503 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2504 bytes += TNODE_SIZE(0) * stat->tnodes;
2506 max = MAX_STAT_DEPTH;
2507 while (max > 0 && stat->nodesizes[max-1] == 0)
2508 max--;
2510 pointers = 0;
2511 for (i = 1; i < max; i++)
2512 if (stat->nodesizes[i] != 0) {
2513 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2514 pointers += (1<<i) * stat->nodesizes[i];
2516 seq_putc(seq, '\n');
2517 seq_printf(seq, "\tPointers: %u\n", pointers);
2519 bytes += sizeof(struct key_vector *) * pointers;
2520 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2521 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2524 #ifdef CONFIG_IP_FIB_TRIE_STATS
2525 static void trie_show_usage(struct seq_file *seq,
2526 const struct trie_use_stats __percpu *stats)
2528 struct trie_use_stats s = { 0 };
2529 int cpu;
2531 /* loop through all of the CPUs and gather up the stats */
2532 for_each_possible_cpu(cpu) {
2533 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2535 s.gets += pcpu->gets;
2536 s.backtrack += pcpu->backtrack;
2537 s.semantic_match_passed += pcpu->semantic_match_passed;
2538 s.semantic_match_miss += pcpu->semantic_match_miss;
2539 s.null_node_hit += pcpu->null_node_hit;
2540 s.resize_node_skipped += pcpu->resize_node_skipped;
2543 seq_printf(seq, "\nCounters:\n---------\n");
2544 seq_printf(seq, "gets = %u\n", s.gets);
2545 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2546 seq_printf(seq, "semantic match passed = %u\n",
2547 s.semantic_match_passed);
2548 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2549 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2550 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2552 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2554 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2556 if (tb->tb_id == RT_TABLE_LOCAL)
2557 seq_puts(seq, "Local:\n");
2558 else if (tb->tb_id == RT_TABLE_MAIN)
2559 seq_puts(seq, "Main:\n");
2560 else
2561 seq_printf(seq, "Id %d:\n", tb->tb_id);
2565 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2567 struct net *net = (struct net *)seq->private;
2568 unsigned int h;
2570 seq_printf(seq,
2571 "Basic info: size of leaf:"
2572 " %zd bytes, size of tnode: %zd bytes.\n",
2573 LEAF_SIZE, TNODE_SIZE(0));
2575 rcu_read_lock();
2576 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2577 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2578 struct fib_table *tb;
2580 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2581 struct trie *t = (struct trie *) tb->tb_data;
2582 struct trie_stat stat;
2584 if (!t)
2585 continue;
2587 fib_table_print(seq, tb);
2589 trie_collect_stats(t, &stat);
2590 trie_show_stats(seq, &stat);
2591 #ifdef CONFIG_IP_FIB_TRIE_STATS
2592 trie_show_usage(seq, t->stats);
2593 #endif
2595 cond_resched_rcu();
2597 rcu_read_unlock();
2599 return 0;
2602 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2604 struct fib_trie_iter *iter = seq->private;
2605 struct net *net = seq_file_net(seq);
2606 loff_t idx = 0;
2607 unsigned int h;
2609 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2610 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2611 struct fib_table *tb;
2613 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2614 struct key_vector *n;
2616 for (n = fib_trie_get_first(iter,
2617 (struct trie *) tb->tb_data);
2618 n; n = fib_trie_get_next(iter))
2619 if (pos == idx++) {
2620 iter->tb = tb;
2621 return n;
2626 return NULL;
2629 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2630 __acquires(RCU)
2632 rcu_read_lock();
2633 return fib_trie_get_idx(seq, *pos);
2636 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2638 struct fib_trie_iter *iter = seq->private;
2639 struct net *net = seq_file_net(seq);
2640 struct fib_table *tb = iter->tb;
2641 struct hlist_node *tb_node;
2642 unsigned int h;
2643 struct key_vector *n;
2645 ++*pos;
2646 /* next node in same table */
2647 n = fib_trie_get_next(iter);
2648 if (n)
2649 return n;
2651 /* walk rest of this hash chain */
2652 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2653 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2654 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2655 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2656 if (n)
2657 goto found;
2660 /* new hash chain */
2661 while (++h < FIB_TABLE_HASHSZ) {
2662 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2663 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2664 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2665 if (n)
2666 goto found;
2669 return NULL;
2671 found:
2672 iter->tb = tb;
2673 return n;
2676 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2677 __releases(RCU)
2679 rcu_read_unlock();
2682 static void seq_indent(struct seq_file *seq, int n)
2684 while (n-- > 0)
2685 seq_puts(seq, " ");
2688 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2690 switch (s) {
2691 case RT_SCOPE_UNIVERSE: return "universe";
2692 case RT_SCOPE_SITE: return "site";
2693 case RT_SCOPE_LINK: return "link";
2694 case RT_SCOPE_HOST: return "host";
2695 case RT_SCOPE_NOWHERE: return "nowhere";
2696 default:
2697 snprintf(buf, len, "scope=%d", s);
2698 return buf;
2702 static const char *const rtn_type_names[__RTN_MAX] = {
2703 [RTN_UNSPEC] = "UNSPEC",
2704 [RTN_UNICAST] = "UNICAST",
2705 [RTN_LOCAL] = "LOCAL",
2706 [RTN_BROADCAST] = "BROADCAST",
2707 [RTN_ANYCAST] = "ANYCAST",
2708 [RTN_MULTICAST] = "MULTICAST",
2709 [RTN_BLACKHOLE] = "BLACKHOLE",
2710 [RTN_UNREACHABLE] = "UNREACHABLE",
2711 [RTN_PROHIBIT] = "PROHIBIT",
2712 [RTN_THROW] = "THROW",
2713 [RTN_NAT] = "NAT",
2714 [RTN_XRESOLVE] = "XRESOLVE",
2717 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2719 if (t < __RTN_MAX && rtn_type_names[t])
2720 return rtn_type_names[t];
2721 snprintf(buf, len, "type %u", t);
2722 return buf;
2725 /* Pretty print the trie */
2726 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2728 const struct fib_trie_iter *iter = seq->private;
2729 struct key_vector *n = v;
2731 if (IS_TRIE(node_parent_rcu(n)))
2732 fib_table_print(seq, iter->tb);
2734 if (IS_TNODE(n)) {
2735 __be32 prf = htonl(n->key);
2737 seq_indent(seq, iter->depth-1);
2738 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2739 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2740 tn_info(n)->full_children,
2741 tn_info(n)->empty_children);
2742 } else {
2743 __be32 val = htonl(n->key);
2744 struct fib_alias *fa;
2746 seq_indent(seq, iter->depth);
2747 seq_printf(seq, " |-- %pI4\n", &val);
2749 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2750 char buf1[32], buf2[32];
2752 seq_indent(seq, iter->depth + 1);
2753 seq_printf(seq, " /%zu %s %s",
2754 KEYLENGTH - fa->fa_slen,
2755 rtn_scope(buf1, sizeof(buf1),
2756 fa->fa_info->fib_scope),
2757 rtn_type(buf2, sizeof(buf2),
2758 fa->fa_type));
2759 if (fa->fa_tos)
2760 seq_printf(seq, " tos=%d", fa->fa_tos);
2761 seq_putc(seq, '\n');
2765 return 0;
2768 static const struct seq_operations fib_trie_seq_ops = {
2769 .start = fib_trie_seq_start,
2770 .next = fib_trie_seq_next,
2771 .stop = fib_trie_seq_stop,
2772 .show = fib_trie_seq_show,
2775 struct fib_route_iter {
2776 struct seq_net_private p;
2777 struct fib_table *main_tb;
2778 struct key_vector *tnode;
2779 loff_t pos;
2780 t_key key;
2783 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2784 loff_t pos)
2786 struct key_vector *l, **tp = &iter->tnode;
2787 t_key key;
2789 /* use cached location of previously found key */
2790 if (iter->pos > 0 && pos >= iter->pos) {
2791 key = iter->key;
2792 } else {
2793 iter->pos = 1;
2794 key = 0;
2797 pos -= iter->pos;
2799 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2800 key = l->key + 1;
2801 iter->pos++;
2802 l = NULL;
2804 /* handle unlikely case of a key wrap */
2805 if (!key)
2806 break;
2809 if (l)
2810 iter->key = l->key; /* remember it */
2811 else
2812 iter->pos = 0; /* forget it */
2814 return l;
2817 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2818 __acquires(RCU)
2820 struct fib_route_iter *iter = seq->private;
2821 struct fib_table *tb;
2822 struct trie *t;
2824 rcu_read_lock();
2826 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2827 if (!tb)
2828 return NULL;
2830 iter->main_tb = tb;
2831 t = (struct trie *)tb->tb_data;
2832 iter->tnode = t->kv;
2834 if (*pos != 0)
2835 return fib_route_get_idx(iter, *pos);
2837 iter->pos = 0;
2838 iter->key = KEY_MAX;
2840 return SEQ_START_TOKEN;
2843 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2845 struct fib_route_iter *iter = seq->private;
2846 struct key_vector *l = NULL;
2847 t_key key = iter->key + 1;
2849 ++*pos;
2851 /* only allow key of 0 for start of sequence */
2852 if ((v == SEQ_START_TOKEN) || key)
2853 l = leaf_walk_rcu(&iter->tnode, key);
2855 if (l) {
2856 iter->key = l->key;
2857 iter->pos++;
2858 } else {
2859 iter->pos = 0;
2862 return l;
2865 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2866 __releases(RCU)
2868 rcu_read_unlock();
2871 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2873 unsigned int flags = 0;
2875 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2876 flags = RTF_REJECT;
2877 if (fi) {
2878 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2880 if (nhc->nhc_gw.ipv4)
2881 flags |= RTF_GATEWAY;
2883 if (mask == htonl(0xFFFFFFFF))
2884 flags |= RTF_HOST;
2885 flags |= RTF_UP;
2886 return flags;
2890 * This outputs /proc/net/route.
2891 * The format of the file is not supposed to be changed
2892 * and needs to be same as fib_hash output to avoid breaking
2893 * legacy utilities
2895 static int fib_route_seq_show(struct seq_file *seq, void *v)
2897 struct fib_route_iter *iter = seq->private;
2898 struct fib_table *tb = iter->main_tb;
2899 struct fib_alias *fa;
2900 struct key_vector *l = v;
2901 __be32 prefix;
2903 if (v == SEQ_START_TOKEN) {
2904 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2905 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2906 "\tWindow\tIRTT");
2907 return 0;
2910 prefix = htonl(l->key);
2912 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2913 struct fib_info *fi = fa->fa_info;
2914 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2915 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2917 if ((fa->fa_type == RTN_BROADCAST) ||
2918 (fa->fa_type == RTN_MULTICAST))
2919 continue;
2921 if (fa->tb_id != tb->tb_id)
2922 continue;
2924 seq_setwidth(seq, 127);
2926 if (fi) {
2927 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2928 __be32 gw = 0;
2930 if (nhc->nhc_gw_family == AF_INET)
2931 gw = nhc->nhc_gw.ipv4;
2933 seq_printf(seq,
2934 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2935 "%d\t%08X\t%d\t%u\t%u",
2936 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2937 prefix, gw, flags, 0, 0,
2938 fi->fib_priority,
2939 mask,
2940 (fi->fib_advmss ?
2941 fi->fib_advmss + 40 : 0),
2942 fi->fib_window,
2943 fi->fib_rtt >> 3);
2944 } else {
2945 seq_printf(seq,
2946 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2947 "%d\t%08X\t%d\t%u\t%u",
2948 prefix, 0, flags, 0, 0, 0,
2949 mask, 0, 0, 0);
2951 seq_pad(seq, '\n');
2954 return 0;
2957 static const struct seq_operations fib_route_seq_ops = {
2958 .start = fib_route_seq_start,
2959 .next = fib_route_seq_next,
2960 .stop = fib_route_seq_stop,
2961 .show = fib_route_seq_show,
2964 int __net_init fib_proc_init(struct net *net)
2966 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2967 sizeof(struct fib_trie_iter)))
2968 goto out1;
2970 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
2971 fib_triestat_seq_show, NULL))
2972 goto out2;
2974 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
2975 sizeof(struct fib_route_iter)))
2976 goto out3;
2978 return 0;
2980 out3:
2981 remove_proc_entry("fib_triestat", net->proc_net);
2982 out2:
2983 remove_proc_entry("fib_trie", net->proc_net);
2984 out1:
2985 return -ENOMEM;
2988 void __net_exit fib_proc_exit(struct net *net)
2990 remove_proc_entry("fib_trie", net->proc_net);
2991 remove_proc_entry("fib_triestat", net->proc_net);
2992 remove_proc_entry("route", net->proc_net);
2995 #endif /* CONFIG_PROC_FS */