1 // SPDX-License-Identifier: GPL-2.0-or-later
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>
39 #define VERSION "0.409"
41 #include <linux/cache.h>
42 #include <linux/uaccess.h>
43 #include <linux/bitops.h>
44 #include <linux/types.h>
45 #include <linux/kernel.h>
47 #include <linux/string.h>
48 #include <linux/socket.h>
49 #include <linux/sockios.h>
50 #include <linux/errno.h>
52 #include <linux/inet.h>
53 #include <linux/inetdevice.h>
54 #include <linux/netdevice.h>
55 #include <linux/if_arp.h>
56 #include <linux/proc_fs.h>
57 #include <linux/rcupdate.h>
58 #include <linux/skbuff.h>
59 #include <linux/netlink.h>
60 #include <linux/init.h>
61 #include <linux/list.h>
62 #include <linux/slab.h>
63 #include <linux/export.h>
64 #include <linux/vmalloc.h>
65 #include <linux/notifier.h>
66 #include <net/net_namespace.h>
68 #include <net/protocol.h>
69 #include <net/route.h>
72 #include <net/ip_fib.h>
73 #include <net/fib_notifier.h>
74 #include <trace/events/fib.h>
75 #include "fib_lookup.h"
77 static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
78 enum fib_event_type event_type, u32 dst,
79 int dst_len, struct fib_alias *fa)
81 struct fib_entry_notifier_info info = {
89 return call_fib4_notifier(nb, net, event_type, &info.info);
92 static int call_fib_entry_notifiers(struct net *net,
93 enum fib_event_type event_type, u32 dst,
94 int dst_len, struct fib_alias *fa,
95 struct netlink_ext_ack *extack)
97 struct fib_entry_notifier_info info = {
98 .info.extack = extack,
106 return call_fib4_notifiers(net, event_type, &info.info);
109 #define MAX_STAT_DEPTH 32
111 #define KEYLENGTH (8*sizeof(t_key))
112 #define KEY_MAX ((t_key)~0)
114 typedef unsigned int t_key;
116 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
117 #define IS_TNODE(n) ((n)->bits)
118 #define IS_LEAF(n) (!(n)->bits)
122 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
123 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
126 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
127 struct hlist_head leaf;
128 /* This array is valid if (pos | bits) > 0 (TNODE) */
129 struct key_vector __rcu *tnode[0];
135 t_key empty_children; /* KEYLENGTH bits needed */
136 t_key full_children; /* KEYLENGTH bits needed */
137 struct key_vector __rcu *parent;
138 struct key_vector kv[1];
139 #define tn_bits kv[0].bits
142 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
143 #define LEAF_SIZE TNODE_SIZE(1)
145 #ifdef CONFIG_IP_FIB_TRIE_STATS
146 struct trie_use_stats {
148 unsigned int backtrack;
149 unsigned int semantic_match_passed;
150 unsigned int semantic_match_miss;
151 unsigned int null_node_hit;
152 unsigned int resize_node_skipped;
157 unsigned int totdepth;
158 unsigned int maxdepth;
161 unsigned int nullpointers;
162 unsigned int prefixes;
163 unsigned int nodesizes[MAX_STAT_DEPTH];
167 struct key_vector kv[1];
168 #ifdef CONFIG_IP_FIB_TRIE_STATS
169 struct trie_use_stats __percpu *stats;
173 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
174 static unsigned int tnode_free_size;
177 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
178 * especially useful before resizing the root node with PREEMPT_NONE configs;
179 * the value was obtained experimentally, aiming to avoid visible slowdown.
181 unsigned int sysctl_fib_sync_mem = 512 * 1024;
182 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
183 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
185 static struct kmem_cache *fn_alias_kmem __ro_after_init;
186 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
188 static inline struct tnode *tn_info(struct key_vector *kv)
190 return container_of(kv, struct tnode, kv[0]);
193 /* caller must hold RTNL */
194 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
195 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
197 /* caller must hold RCU read lock or RTNL */
198 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
199 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
201 /* wrapper for rcu_assign_pointer */
202 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
205 rcu_assign_pointer(tn_info(n)->parent, tp);
208 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
210 /* This provides us with the number of children in this node, in the case of a
211 * leaf this will return 0 meaning none of the children are accessible.
213 static inline unsigned long child_length(const struct key_vector *tn)
215 return (1ul << tn->bits) & ~(1ul);
218 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
220 static inline unsigned long get_index(t_key key, struct key_vector *kv)
222 unsigned long index = key ^ kv->key;
224 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
227 return index >> kv->pos;
230 /* To understand this stuff, an understanding of keys and all their bits is
231 * necessary. Every node in the trie has a key associated with it, but not
232 * all of the bits in that key are significant.
234 * Consider a node 'n' and its parent 'tp'.
236 * If n is a leaf, every bit in its key is significant. Its presence is
237 * necessitated by path compression, since during a tree traversal (when
238 * searching for a leaf - unless we are doing an insertion) we will completely
239 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
240 * a potentially successful search, that we have indeed been walking the
243 * Note that we can never "miss" the correct key in the tree if present by
244 * following the wrong path. Path compression ensures that segments of the key
245 * that are the same for all keys with a given prefix are skipped, but the
246 * skipped part *is* identical for each node in the subtrie below the skipped
247 * bit! trie_insert() in this implementation takes care of that.
249 * if n is an internal node - a 'tnode' here, the various parts of its key
250 * have many different meanings.
253 * _________________________________________________________________
254 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
255 * -----------------------------------------------------------------
256 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
258 * _________________________________________________________________
259 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
260 * -----------------------------------------------------------------
261 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
268 * First, let's just ignore the bits that come before the parent tp, that is
269 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
270 * point we do not use them for anything.
272 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
273 * index into the parent's child array. That is, they will be used to find
274 * 'n' among tp's children.
276 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
279 * All the bits we have seen so far are significant to the node n. The rest
280 * of the bits are really not needed or indeed known in n->key.
282 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
283 * n's child array, and will of course be different for each child.
285 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
289 static const int halve_threshold = 25;
290 static const int inflate_threshold = 50;
291 static const int halve_threshold_root = 15;
292 static const int inflate_threshold_root = 30;
294 static void __alias_free_mem(struct rcu_head *head)
296 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
297 kmem_cache_free(fn_alias_kmem, fa);
300 static inline void alias_free_mem_rcu(struct fib_alias *fa)
302 call_rcu(&fa->rcu, __alias_free_mem);
305 #define TNODE_KMALLOC_MAX \
306 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
307 #define TNODE_VMALLOC_MAX \
308 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
310 static void __node_free_rcu(struct rcu_head *head)
312 struct tnode *n = container_of(head, struct tnode, rcu);
315 kmem_cache_free(trie_leaf_kmem, n);
320 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
322 static struct tnode *tnode_alloc(int bits)
326 /* verify bits is within bounds */
327 if (bits > TNODE_VMALLOC_MAX)
330 /* determine size and verify it is non-zero and didn't overflow */
331 size = TNODE_SIZE(1ul << bits);
333 if (size <= PAGE_SIZE)
334 return kzalloc(size, GFP_KERNEL);
336 return vzalloc(size);
339 static inline void empty_child_inc(struct key_vector *n)
341 tn_info(n)->empty_children++;
343 if (!tn_info(n)->empty_children)
344 tn_info(n)->full_children++;
347 static inline void empty_child_dec(struct key_vector *n)
349 if (!tn_info(n)->empty_children)
350 tn_info(n)->full_children--;
352 tn_info(n)->empty_children--;
355 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
357 struct key_vector *l;
360 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
364 /* initialize key vector */
369 l->slen = fa->fa_slen;
371 /* link leaf to fib alias */
372 INIT_HLIST_HEAD(&l->leaf);
373 hlist_add_head(&fa->fa_list, &l->leaf);
378 static struct key_vector *tnode_new(t_key key, int pos, int bits)
380 unsigned int shift = pos + bits;
381 struct key_vector *tn;
384 /* verify bits and pos their msb bits clear and values are valid */
385 BUG_ON(!bits || (shift > KEYLENGTH));
387 tnode = tnode_alloc(bits);
391 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
392 sizeof(struct key_vector *) << bits);
394 if (bits == KEYLENGTH)
395 tnode->full_children = 1;
397 tnode->empty_children = 1ul << bits;
400 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
408 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
409 * and no bits are skipped. See discussion in dyntree paper p. 6
411 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
413 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
416 /* Add a child at position i overwriting the old value.
417 * Update the value of full_children and empty_children.
419 static void put_child(struct key_vector *tn, unsigned long i,
420 struct key_vector *n)
422 struct key_vector *chi = get_child(tn, i);
425 BUG_ON(i >= child_length(tn));
427 /* update emptyChildren, overflow into fullChildren */
433 /* update fullChildren */
434 wasfull = tnode_full(tn, chi);
435 isfull = tnode_full(tn, n);
437 if (wasfull && !isfull)
438 tn_info(tn)->full_children--;
439 else if (!wasfull && isfull)
440 tn_info(tn)->full_children++;
442 if (n && (tn->slen < n->slen))
445 rcu_assign_pointer(tn->tnode[i], n);
448 static void update_children(struct key_vector *tn)
452 /* update all of the child parent pointers */
453 for (i = child_length(tn); i;) {
454 struct key_vector *inode = get_child(tn, --i);
459 /* Either update the children of a tnode that
460 * already belongs to us or update the child
461 * to point to ourselves.
463 if (node_parent(inode) == tn)
464 update_children(inode);
466 node_set_parent(inode, tn);
470 static inline void put_child_root(struct key_vector *tp, t_key key,
471 struct key_vector *n)
474 rcu_assign_pointer(tp->tnode[0], n);
476 put_child(tp, get_index(key, tp), n);
479 static inline void tnode_free_init(struct key_vector *tn)
481 tn_info(tn)->rcu.next = NULL;
484 static inline void tnode_free_append(struct key_vector *tn,
485 struct key_vector *n)
487 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
488 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
491 static void tnode_free(struct key_vector *tn)
493 struct callback_head *head = &tn_info(tn)->rcu;
497 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
500 tn = container_of(head, struct tnode, rcu)->kv;
503 if (tnode_free_size >= sysctl_fib_sync_mem) {
509 static struct key_vector *replace(struct trie *t,
510 struct key_vector *oldtnode,
511 struct key_vector *tn)
513 struct key_vector *tp = node_parent(oldtnode);
516 /* setup the parent pointer out of and back into this node */
517 NODE_INIT_PARENT(tn, tp);
518 put_child_root(tp, tn->key, tn);
520 /* update all of the child parent pointers */
523 /* all pointers should be clean so we are done */
524 tnode_free(oldtnode);
526 /* resize children now that oldtnode is freed */
527 for (i = child_length(tn); i;) {
528 struct key_vector *inode = get_child(tn, --i);
530 /* resize child node */
531 if (tnode_full(tn, inode))
532 tn = resize(t, inode);
538 static struct key_vector *inflate(struct trie *t,
539 struct key_vector *oldtnode)
541 struct key_vector *tn;
545 pr_debug("In inflate\n");
547 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
551 /* prepare oldtnode to be freed */
552 tnode_free_init(oldtnode);
554 /* Assemble all of the pointers in our cluster, in this case that
555 * represents all of the pointers out of our allocated nodes that
556 * point to existing tnodes and the links between our allocated
559 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
560 struct key_vector *inode = get_child(oldtnode, --i);
561 struct key_vector *node0, *node1;
568 /* A leaf or an internal node with skipped bits */
569 if (!tnode_full(oldtnode, inode)) {
570 put_child(tn, get_index(inode->key, tn), inode);
574 /* drop the node in the old tnode free list */
575 tnode_free_append(oldtnode, inode);
577 /* An internal node with two children */
578 if (inode->bits == 1) {
579 put_child(tn, 2 * i + 1, get_child(inode, 1));
580 put_child(tn, 2 * i, get_child(inode, 0));
584 /* We will replace this node 'inode' with two new
585 * ones, 'node0' and 'node1', each with half of the
586 * original children. The two new nodes will have
587 * a position one bit further down the key and this
588 * means that the "significant" part of their keys
589 * (see the discussion near the top of this file)
590 * will differ by one bit, which will be "0" in
591 * node0's key and "1" in node1's key. Since we are
592 * moving the key position by one step, the bit that
593 * we are moving away from - the bit at position
594 * (tn->pos) - is the one that will differ between
595 * node0 and node1. So... we synthesize that bit in the
598 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
601 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
603 tnode_free_append(tn, node1);
606 tnode_free_append(tn, node0);
608 /* populate child pointers in new nodes */
609 for (k = child_length(inode), j = k / 2; j;) {
610 put_child(node1, --j, get_child(inode, --k));
611 put_child(node0, j, get_child(inode, j));
612 put_child(node1, --j, get_child(inode, --k));
613 put_child(node0, j, get_child(inode, j));
616 /* link new nodes to parent */
617 NODE_INIT_PARENT(node1, tn);
618 NODE_INIT_PARENT(node0, tn);
620 /* link parent to nodes */
621 put_child(tn, 2 * i + 1, node1);
622 put_child(tn, 2 * i, node0);
625 /* setup the parent pointers into and out of this node */
626 return replace(t, oldtnode, tn);
628 /* all pointers should be clean so we are done */
634 static struct key_vector *halve(struct trie *t,
635 struct key_vector *oldtnode)
637 struct key_vector *tn;
640 pr_debug("In halve\n");
642 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
646 /* prepare oldtnode to be freed */
647 tnode_free_init(oldtnode);
649 /* Assemble all of the pointers in our cluster, in this case that
650 * represents all of the pointers out of our allocated nodes that
651 * point to existing tnodes and the links between our allocated
654 for (i = child_length(oldtnode); i;) {
655 struct key_vector *node1 = get_child(oldtnode, --i);
656 struct key_vector *node0 = get_child(oldtnode, --i);
657 struct key_vector *inode;
659 /* At least one of the children is empty */
660 if (!node1 || !node0) {
661 put_child(tn, i / 2, node1 ? : node0);
665 /* Two nonempty children */
666 inode = tnode_new(node0->key, oldtnode->pos, 1);
669 tnode_free_append(tn, inode);
671 /* initialize pointers out of node */
672 put_child(inode, 1, node1);
673 put_child(inode, 0, node0);
674 NODE_INIT_PARENT(inode, tn);
676 /* link parent to node */
677 put_child(tn, i / 2, inode);
680 /* setup the parent pointers into and out of this node */
681 return replace(t, oldtnode, tn);
683 /* all pointers should be clean so we are done */
689 static struct key_vector *collapse(struct trie *t,
690 struct key_vector *oldtnode)
692 struct key_vector *n, *tp;
695 /* scan the tnode looking for that one child that might still exist */
696 for (n = NULL, i = child_length(oldtnode); !n && i;)
697 n = get_child(oldtnode, --i);
699 /* compress one level */
700 tp = node_parent(oldtnode);
701 put_child_root(tp, oldtnode->key, n);
702 node_set_parent(n, tp);
710 static unsigned char update_suffix(struct key_vector *tn)
712 unsigned char slen = tn->pos;
713 unsigned long stride, i;
714 unsigned char slen_max;
716 /* only vector 0 can have a suffix length greater than or equal to
717 * tn->pos + tn->bits, the second highest node will have a suffix
718 * length at most of tn->pos + tn->bits - 1
720 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
722 /* search though the list of children looking for nodes that might
723 * have a suffix greater than the one we currently have. This is
724 * why we start with a stride of 2 since a stride of 1 would
725 * represent the nodes with suffix length equal to tn->pos
727 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
728 struct key_vector *n = get_child(tn, i);
730 if (!n || (n->slen <= slen))
733 /* update stride and slen based on new value */
734 stride <<= (n->slen - slen);
738 /* stop searching if we have hit the maximum possible value */
739 if (slen >= slen_max)
748 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
749 * the Helsinki University of Technology and Matti Tikkanen of Nokia
750 * Telecommunications, page 6:
751 * "A node is doubled if the ratio of non-empty children to all
752 * children in the *doubled* node is at least 'high'."
754 * 'high' in this instance is the variable 'inflate_threshold'. It
755 * is expressed as a percentage, so we multiply it with
756 * child_length() and instead of multiplying by 2 (since the
757 * child array will be doubled by inflate()) and multiplying
758 * the left-hand side by 100 (to handle the percentage thing) we
759 * multiply the left-hand side by 50.
761 * The left-hand side may look a bit weird: child_length(tn)
762 * - tn->empty_children is of course the number of non-null children
763 * in the current node. tn->full_children is the number of "full"
764 * children, that is non-null tnodes with a skip value of 0.
765 * All of those will be doubled in the resulting inflated tnode, so
766 * we just count them one extra time here.
768 * A clearer way to write this would be:
770 * to_be_doubled = tn->full_children;
771 * not_to_be_doubled = child_length(tn) - tn->empty_children -
774 * new_child_length = child_length(tn) * 2;
776 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
778 * if (new_fill_factor >= inflate_threshold)
780 * ...and so on, tho it would mess up the while () loop.
783 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
787 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
788 * inflate_threshold * new_child_length
790 * expand not_to_be_doubled and to_be_doubled, and shorten:
791 * 100 * (child_length(tn) - tn->empty_children +
792 * tn->full_children) >= inflate_threshold * new_child_length
794 * expand new_child_length:
795 * 100 * (child_length(tn) - tn->empty_children +
796 * tn->full_children) >=
797 * inflate_threshold * child_length(tn) * 2
800 * 50 * (tn->full_children + child_length(tn) -
801 * tn->empty_children) >= inflate_threshold *
805 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
807 unsigned long used = child_length(tn);
808 unsigned long threshold = used;
810 /* Keep root node larger */
811 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
812 used -= tn_info(tn)->empty_children;
813 used += tn_info(tn)->full_children;
815 /* if bits == KEYLENGTH then pos = 0, and will fail below */
817 return (used > 1) && tn->pos && ((50 * used) >= threshold);
820 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
822 unsigned long used = child_length(tn);
823 unsigned long threshold = used;
825 /* Keep root node larger */
826 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
827 used -= tn_info(tn)->empty_children;
829 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
831 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
834 static inline bool should_collapse(struct key_vector *tn)
836 unsigned long used = child_length(tn);
838 used -= tn_info(tn)->empty_children;
840 /* account for bits == KEYLENGTH case */
841 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
844 /* One child or none, time to drop us from the trie */
849 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
851 #ifdef CONFIG_IP_FIB_TRIE_STATS
852 struct trie_use_stats __percpu *stats = t->stats;
854 struct key_vector *tp = node_parent(tn);
855 unsigned long cindex = get_index(tn->key, tp);
856 int max_work = MAX_WORK;
858 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
859 tn, inflate_threshold, halve_threshold);
861 /* track the tnode via the pointer from the parent instead of
862 * doing it ourselves. This way we can let RCU fully do its
863 * thing without us interfering
865 BUG_ON(tn != get_child(tp, cindex));
867 /* Double as long as the resulting node has a number of
868 * nonempty nodes that are above the threshold.
870 while (should_inflate(tp, tn) && max_work) {
873 #ifdef CONFIG_IP_FIB_TRIE_STATS
874 this_cpu_inc(stats->resize_node_skipped);
880 tn = get_child(tp, cindex);
883 /* update parent in case inflate failed */
884 tp = node_parent(tn);
886 /* Return if at least one inflate is run */
887 if (max_work != MAX_WORK)
890 /* Halve as long as the number of empty children in this
891 * node is above threshold.
893 while (should_halve(tp, tn) && max_work) {
896 #ifdef CONFIG_IP_FIB_TRIE_STATS
897 this_cpu_inc(stats->resize_node_skipped);
903 tn = get_child(tp, cindex);
906 /* Only one child remains */
907 if (should_collapse(tn))
908 return collapse(t, tn);
910 /* update parent in case halve failed */
911 return node_parent(tn);
914 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
916 unsigned char node_slen = tn->slen;
918 while ((node_slen > tn->pos) && (node_slen > slen)) {
919 slen = update_suffix(tn);
920 if (node_slen == slen)
923 tn = node_parent(tn);
924 node_slen = tn->slen;
928 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
930 while (tn->slen < slen) {
932 tn = node_parent(tn);
936 /* rcu_read_lock needs to be hold by caller from readside */
937 static struct key_vector *fib_find_node(struct trie *t,
938 struct key_vector **tp, u32 key)
940 struct key_vector *pn, *n = t->kv;
941 unsigned long index = 0;
945 n = get_child_rcu(n, index);
950 index = get_cindex(key, n);
952 /* This bit of code is a bit tricky but it combines multiple
953 * checks into a single check. The prefix consists of the
954 * prefix plus zeros for the bits in the cindex. The index
955 * is the difference between the key and this value. From
956 * this we can actually derive several pieces of data.
957 * if (index >= (1ul << bits))
958 * we have a mismatch in skip bits and failed
960 * we know the value is cindex
962 * This check is safe even if bits == KEYLENGTH due to the
963 * fact that we can only allocate a node with 32 bits if a
964 * long is greater than 32 bits.
966 if (index >= (1ul << n->bits)) {
971 /* keep searching until we find a perfect match leaf or NULL */
972 } while (IS_TNODE(n));
979 /* Return the first fib alias matching TOS with
980 * priority less than or equal to PRIO.
982 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
983 u8 tos, u32 prio, u32 tb_id)
985 struct fib_alias *fa;
990 hlist_for_each_entry(fa, fah, fa_list) {
991 if (fa->fa_slen < slen)
993 if (fa->fa_slen != slen)
995 if (fa->tb_id > tb_id)
997 if (fa->tb_id != tb_id)
999 if (fa->fa_tos > tos)
1001 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1008 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1010 while (!IS_TRIE(tn))
1014 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1015 struct fib_alias *new, t_key key)
1017 struct key_vector *n, *l;
1019 l = leaf_new(key, new);
1023 /* retrieve child from parent node */
1024 n = get_child(tp, get_index(key, tp));
1026 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1028 * Add a new tnode here
1029 * first tnode need some special handling
1030 * leaves us in position for handling as case 3
1033 struct key_vector *tn;
1035 tn = tnode_new(key, __fls(key ^ n->key), 1);
1039 /* initialize routes out of node */
1040 NODE_INIT_PARENT(tn, tp);
1041 put_child(tn, get_index(key, tn) ^ 1, n);
1043 /* start adding routes into the node */
1044 put_child_root(tp, key, tn);
1045 node_set_parent(n, tn);
1047 /* parent now has a NULL spot where the leaf can go */
1051 /* Case 3: n is NULL, and will just insert a new leaf */
1052 node_push_suffix(tp, new->fa_slen);
1053 NODE_INIT_PARENT(l, tp);
1054 put_child_root(tp, key, l);
1055 trie_rebalance(t, tp);
1064 /* fib notifier for ADD is sent before calling fib_insert_alias with
1065 * the expectation that the only possible failure 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)
1072 return fib_insert_node(t, tp, new, key);
1075 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1077 struct fib_alias *last;
1079 hlist_for_each_entry(last, &l->leaf, fa_list) {
1080 if (new->fa_slen < last->fa_slen)
1082 if ((new->fa_slen == last->fa_slen) &&
1083 (new->tb_id > last->tb_id))
1089 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
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);
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");
1110 if ((plen < KEYLENGTH) && (key << plen)) {
1111 NL_SET_ERR_MSG(extack,
1112 "Invalid prefix for given prefix length");
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;
1135 key = ntohl(cfg->fc_dst);
1137 if (!fib_valid_key_len(key, plen, extack))
1140 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1142 fi = fib_create_info(cfg, extack);
1148 l = fib_find_node(t, &tp, key);
1149 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
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
1161 if (fa && fa->fa_tos == tos &&
1162 fa->fa_info->fib_priority == fi->fib_priority) {
1163 struct fib_alias *fa_first, *fa_match;
1166 if (cfg->fc_nlflags & NLM_F_EXCL)
1169 nlflags &= ~NLM_F_EXCL;
1172 * 1. Find exact match for type, scope, fib_info to avoid
1174 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
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))
1183 if (fa->fa_info->fib_priority != fi->fib_priority)
1185 if (fa->fa_type == cfg->fc_type &&
1186 fa->fa_info == fi) {
1192 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1193 struct fib_info *fi_drop;
1196 nlflags |= NLM_F_REPLACE;
1204 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
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 err = call_fib_entry_notifiers(net,
1219 FIB_EVENT_ENTRY_REPLACE,
1223 goto out_free_new_fa;
1225 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1226 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1228 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1230 alias_free_mem_rcu(fa);
1232 fib_release_info(fi_drop);
1233 if (state & FA_S_ACCESSED)
1234 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1238 /* Error if we find a perfect match which
1239 * uses the same scope, type, and nexthop
1245 if (cfg->fc_nlflags & NLM_F_APPEND) {
1246 event = FIB_EVENT_ENTRY_APPEND;
1247 nlflags |= NLM_F_APPEND;
1253 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1256 nlflags |= NLM_F_CREATE;
1258 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1262 new_fa->fa_info = fi;
1263 new_fa->fa_tos = tos;
1264 new_fa->fa_type = cfg->fc_type;
1265 new_fa->fa_state = 0;
1266 new_fa->fa_slen = slen;
1267 new_fa->tb_id = tb->tb_id;
1268 new_fa->fa_default = -1;
1270 err = call_fib_entry_notifiers(net, event, key, plen, new_fa, extack);
1272 goto out_free_new_fa;
1274 /* Insert new entry to the list. */
1275 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1280 tb->tb_num_default++;
1282 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1283 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1284 &cfg->fc_nlinfo, nlflags);
1289 /* notifier was sent that entry would be added to trie, but
1290 * the add failed and need to recover. Only failure for
1291 * fib_insert_alias is ENOMEM.
1293 NL_SET_ERR_MSG(extack, "Failed to insert route into trie");
1294 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key,
1295 plen, new_fa, NULL);
1297 kmem_cache_free(fn_alias_kmem, new_fa);
1299 fib_release_info(fi);
1304 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1306 t_key prefix = n->key;
1308 return (key ^ prefix) & (prefix | -prefix);
1311 /* should be called with rcu_read_lock */
1312 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1313 struct fib_result *res, int fib_flags)
1315 struct trie *t = (struct trie *) tb->tb_data;
1316 #ifdef CONFIG_IP_FIB_TRIE_STATS
1317 struct trie_use_stats __percpu *stats = t->stats;
1319 const t_key key = ntohl(flp->daddr);
1320 struct key_vector *n, *pn;
1321 struct fib_alias *fa;
1322 unsigned long index;
1328 n = get_child_rcu(pn, cindex);
1330 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1334 #ifdef CONFIG_IP_FIB_TRIE_STATS
1335 this_cpu_inc(stats->gets);
1338 /* Step 1: Travel to the longest prefix match in the trie */
1340 index = get_cindex(key, n);
1342 /* This bit of code is a bit tricky but it combines multiple
1343 * checks into a single check. The prefix consists of the
1344 * prefix plus zeros for the "bits" in the prefix. The index
1345 * is the difference between the key and this value. From
1346 * this we can actually derive several pieces of data.
1347 * if (index >= (1ul << bits))
1348 * we have a mismatch in skip bits and failed
1350 * we know the value is cindex
1352 * This check is safe even if bits == KEYLENGTH due to the
1353 * fact that we can only allocate a node with 32 bits if a
1354 * long is greater than 32 bits.
1356 if (index >= (1ul << n->bits))
1359 /* we have found a leaf. Prefixes have already been compared */
1363 /* only record pn and cindex if we are going to be chopping
1364 * bits later. Otherwise we are just wasting cycles.
1366 if (n->slen > n->pos) {
1371 n = get_child_rcu(n, index);
1376 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1378 /* record the pointer where our next node pointer is stored */
1379 struct key_vector __rcu **cptr = n->tnode;
1381 /* This test verifies that none of the bits that differ
1382 * between the key and the prefix exist in the region of
1383 * the lsb and higher in the prefix.
1385 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1388 /* exit out and process leaf */
1389 if (unlikely(IS_LEAF(n)))
1392 /* Don't bother recording parent info. Since we are in
1393 * prefix match mode we will have to come back to wherever
1394 * we started this traversal anyway
1397 while ((n = rcu_dereference(*cptr)) == NULL) {
1399 #ifdef CONFIG_IP_FIB_TRIE_STATS
1401 this_cpu_inc(stats->null_node_hit);
1403 /* If we are at cindex 0 there are no more bits for
1404 * us to strip at this level so we must ascend back
1405 * up one level to see if there are any more bits to
1406 * be stripped there.
1409 t_key pkey = pn->key;
1411 /* If we don't have a parent then there is
1412 * nothing for us to do as we do not have any
1413 * further nodes to parse.
1416 trace_fib_table_lookup(tb->tb_id, flp,
1420 #ifdef CONFIG_IP_FIB_TRIE_STATS
1421 this_cpu_inc(stats->backtrack);
1423 /* Get Child's index */
1424 pn = node_parent_rcu(pn);
1425 cindex = get_index(pkey, pn);
1428 /* strip the least significant bit from the cindex */
1429 cindex &= cindex - 1;
1431 /* grab pointer for next child node */
1432 cptr = &pn->tnode[cindex];
1437 /* this line carries forward the xor from earlier in the function */
1438 index = key ^ n->key;
1440 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1441 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1442 struct fib_info *fi = fa->fa_info;
1445 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1446 if (index >= (1ul << fa->fa_slen))
1449 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1453 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1455 fib_alias_accessed(fa);
1456 err = fib_props[fa->fa_type].error;
1457 if (unlikely(err < 0)) {
1459 #ifdef CONFIG_IP_FIB_TRIE_STATS
1460 this_cpu_inc(stats->semantic_match_passed);
1462 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1465 if (fi->fib_flags & RTNH_F_DEAD)
1468 if (unlikely(fi->nh && nexthop_is_blackhole(fi->nh))) {
1469 err = fib_props[RTN_BLACKHOLE].error;
1473 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1474 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
1476 if (nhc->nhc_flags & RTNH_F_DEAD)
1478 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1479 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1480 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1482 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1483 if (flp->flowi4_oif &&
1484 flp->flowi4_oif != nhc->nhc_oif)
1488 if (!(fib_flags & FIB_LOOKUP_NOREF))
1489 refcount_inc(&fi->fib_clntref);
1491 res->prefix = htonl(n->key);
1492 res->prefixlen = KEYLENGTH - fa->fa_slen;
1493 res->nh_sel = nhsel;
1495 res->type = fa->fa_type;
1496 res->scope = fi->fib_scope;
1499 res->fa_head = &n->leaf;
1500 #ifdef CONFIG_IP_FIB_TRIE_STATS
1501 this_cpu_inc(stats->semantic_match_passed);
1503 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1508 #ifdef CONFIG_IP_FIB_TRIE_STATS
1509 this_cpu_inc(stats->semantic_match_miss);
1513 EXPORT_SYMBOL_GPL(fib_table_lookup);
1515 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1516 struct key_vector *l, struct fib_alias *old)
1518 /* record the location of the previous list_info entry */
1519 struct hlist_node **pprev = old->fa_list.pprev;
1520 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1522 /* remove the fib_alias from the list */
1523 hlist_del_rcu(&old->fa_list);
1525 /* if we emptied the list this leaf will be freed and we can sort
1526 * out parent suffix lengths as a part of trie_rebalance
1528 if (hlist_empty(&l->leaf)) {
1529 if (tp->slen == l->slen)
1530 node_pull_suffix(tp, tp->pos);
1531 put_child_root(tp, l->key, NULL);
1533 trie_rebalance(t, tp);
1537 /* only access fa if it is pointing at the last valid hlist_node */
1541 /* update the trie with the latest suffix length */
1542 l->slen = fa->fa_slen;
1543 node_pull_suffix(tp, fa->fa_slen);
1546 /* Caller must hold RTNL. */
1547 int fib_table_delete(struct net *net, struct fib_table *tb,
1548 struct fib_config *cfg, struct netlink_ext_ack *extack)
1550 struct trie *t = (struct trie *) tb->tb_data;
1551 struct fib_alias *fa, *fa_to_delete;
1552 struct key_vector *l, *tp;
1553 u8 plen = cfg->fc_dst_len;
1554 u8 slen = KEYLENGTH - plen;
1555 u8 tos = cfg->fc_tos;
1558 key = ntohl(cfg->fc_dst);
1560 if (!fib_valid_key_len(key, plen, extack))
1563 l = fib_find_node(t, &tp, key);
1567 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1571 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1573 fa_to_delete = NULL;
1574 hlist_for_each_entry_from(fa, fa_list) {
1575 struct fib_info *fi = fa->fa_info;
1577 if ((fa->fa_slen != slen) ||
1578 (fa->tb_id != tb->tb_id) ||
1579 (fa->fa_tos != tos))
1582 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1583 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1584 fa->fa_info->fib_scope == cfg->fc_scope) &&
1585 (!cfg->fc_prefsrc ||
1586 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1587 (!cfg->fc_protocol ||
1588 fi->fib_protocol == cfg->fc_protocol) &&
1589 fib_nh_match(cfg, fi, extack) == 0 &&
1590 fib_metrics_match(cfg, fi)) {
1599 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1600 fa_to_delete, extack);
1601 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1602 &cfg->fc_nlinfo, 0);
1605 tb->tb_num_default--;
1607 fib_remove_alias(t, tp, l, fa_to_delete);
1609 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1610 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1612 fib_release_info(fa_to_delete->fa_info);
1613 alias_free_mem_rcu(fa_to_delete);
1617 /* Scan for the next leaf starting at the provided key value */
1618 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1620 struct key_vector *pn, *n = *tn;
1621 unsigned long cindex;
1623 /* this loop is meant to try and find the key in the trie */
1625 /* record parent and next child index */
1627 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1629 if (cindex >> pn->bits)
1632 /* descend into the next child */
1633 n = get_child_rcu(pn, cindex++);
1637 /* guarantee forward progress on the keys */
1638 if (IS_LEAF(n) && (n->key >= key))
1640 } while (IS_TNODE(n));
1642 /* this loop will search for the next leaf with a greater key */
1643 while (!IS_TRIE(pn)) {
1644 /* if we exhausted the parent node we will need to climb */
1645 if (cindex >= (1ul << pn->bits)) {
1646 t_key pkey = pn->key;
1648 pn = node_parent_rcu(pn);
1649 cindex = get_index(pkey, pn) + 1;
1653 /* grab the next available node */
1654 n = get_child_rcu(pn, cindex++);
1658 /* no need to compare keys since we bumped the index */
1662 /* Rescan start scanning in new node */
1668 return NULL; /* Root of trie */
1670 /* if we are at the limit for keys just return NULL for the tnode */
1675 static void fib_trie_free(struct fib_table *tb)
1677 struct trie *t = (struct trie *)tb->tb_data;
1678 struct key_vector *pn = t->kv;
1679 unsigned long cindex = 1;
1680 struct hlist_node *tmp;
1681 struct fib_alias *fa;
1683 /* walk trie in reverse order and free everything */
1685 struct key_vector *n;
1688 t_key pkey = pn->key;
1694 pn = node_parent(pn);
1696 /* drop emptied tnode */
1697 put_child_root(pn, n->key, NULL);
1700 cindex = get_index(pkey, pn);
1705 /* grab the next available node */
1706 n = get_child(pn, cindex);
1711 /* record pn and cindex for leaf walking */
1713 cindex = 1ul << n->bits;
1718 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1719 hlist_del_rcu(&fa->fa_list);
1720 alias_free_mem_rcu(fa);
1723 put_child_root(pn, n->key, NULL);
1727 #ifdef CONFIG_IP_FIB_TRIE_STATS
1728 free_percpu(t->stats);
1733 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1735 struct trie *ot = (struct trie *)oldtb->tb_data;
1736 struct key_vector *l, *tp = ot->kv;
1737 struct fib_table *local_tb;
1738 struct fib_alias *fa;
1742 if (oldtb->tb_data == oldtb->__data)
1745 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1749 lt = (struct trie *)local_tb->tb_data;
1751 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1752 struct key_vector *local_l = NULL, *local_tp;
1754 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1755 struct fib_alias *new_fa;
1757 if (local_tb->tb_id != fa->tb_id)
1760 /* clone fa for new local table */
1761 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1765 memcpy(new_fa, fa, sizeof(*fa));
1767 /* insert clone into table */
1769 local_l = fib_find_node(lt, &local_tp, l->key);
1771 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1773 kmem_cache_free(fn_alias_kmem, new_fa);
1778 /* stop loop if key wrapped back to 0 */
1786 fib_trie_free(local_tb);
1791 /* Caller must hold RTNL */
1792 void fib_table_flush_external(struct fib_table *tb)
1794 struct trie *t = (struct trie *)tb->tb_data;
1795 struct key_vector *pn = t->kv;
1796 unsigned long cindex = 1;
1797 struct hlist_node *tmp;
1798 struct fib_alias *fa;
1800 /* walk trie in reverse order */
1802 unsigned char slen = 0;
1803 struct key_vector *n;
1806 t_key pkey = pn->key;
1808 /* cannot resize the trie vector */
1812 /* update the suffix to address pulled leaves */
1813 if (pn->slen > pn->pos)
1816 /* resize completed node */
1818 cindex = get_index(pkey, pn);
1823 /* grab the next available node */
1824 n = get_child(pn, cindex);
1829 /* record pn and cindex for leaf walking */
1831 cindex = 1ul << n->bits;
1836 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1837 /* if alias was cloned to local then we just
1838 * need to remove the local copy from main
1840 if (tb->tb_id != fa->tb_id) {
1841 hlist_del_rcu(&fa->fa_list);
1842 alias_free_mem_rcu(fa);
1846 /* record local slen */
1850 /* update leaf slen */
1853 if (hlist_empty(&n->leaf)) {
1854 put_child_root(pn, n->key, NULL);
1860 /* Caller must hold RTNL. */
1861 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1863 struct trie *t = (struct trie *)tb->tb_data;
1864 struct key_vector *pn = t->kv;
1865 unsigned long cindex = 1;
1866 struct hlist_node *tmp;
1867 struct fib_alias *fa;
1870 /* walk trie in reverse order */
1872 unsigned char slen = 0;
1873 struct key_vector *n;
1876 t_key pkey = pn->key;
1878 /* cannot resize the trie vector */
1882 /* update the suffix to address pulled leaves */
1883 if (pn->slen > pn->pos)
1886 /* resize completed node */
1888 cindex = get_index(pkey, pn);
1893 /* grab the next available node */
1894 n = get_child(pn, cindex);
1899 /* record pn and cindex for leaf walking */
1901 cindex = 1ul << n->bits;
1906 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1907 struct fib_info *fi = fa->fa_info;
1909 if (!fi || tb->tb_id != fa->tb_id ||
1910 (!(fi->fib_flags & RTNH_F_DEAD) &&
1911 !fib_props[fa->fa_type].error)) {
1916 /* Do not flush error routes if network namespace is
1917 * not being dismantled
1919 if (!flush_all && fib_props[fa->fa_type].error) {
1924 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1926 KEYLENGTH - fa->fa_slen, fa,
1928 hlist_del_rcu(&fa->fa_list);
1929 fib_release_info(fa->fa_info);
1930 alias_free_mem_rcu(fa);
1934 /* update leaf slen */
1937 if (hlist_empty(&n->leaf)) {
1938 put_child_root(pn, n->key, NULL);
1943 pr_debug("trie_flush found=%d\n", found);
1947 /* derived from fib_trie_free */
1948 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
1949 struct nl_info *info)
1951 struct trie *t = (struct trie *)tb->tb_data;
1952 struct key_vector *pn = t->kv;
1953 unsigned long cindex = 1;
1954 struct fib_alias *fa;
1957 struct key_vector *n;
1960 t_key pkey = pn->key;
1965 pn = node_parent(pn);
1966 cindex = get_index(pkey, pn);
1970 /* grab the next available node */
1971 n = get_child(pn, cindex);
1976 /* record pn and cindex for leaf walking */
1978 cindex = 1ul << n->bits;
1983 hlist_for_each_entry(fa, &n->leaf, fa_list) {
1984 struct fib_info *fi = fa->fa_info;
1986 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
1989 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
1990 KEYLENGTH - fa->fa_slen, tb->tb_id,
1991 info, NLM_F_REPLACE);
1993 /* call_fib_entry_notifiers will be removed when
1994 * in-kernel notifier is implemented and supported
1995 * for nexthop objects
1997 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
1999 KEYLENGTH - fa->fa_slen, fa,
2005 void fib_info_notify_update(struct net *net, struct nl_info *info)
2009 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2010 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2011 struct fib_table *tb;
2013 hlist_for_each_entry_rcu(tb, head, tb_hlist)
2014 __fib_info_notify_update(net, tb, info);
2018 static void fib_leaf_notify(struct net *net, struct key_vector *l,
2019 struct fib_table *tb, struct notifier_block *nb)
2021 struct fib_alias *fa;
2023 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2024 struct fib_info *fi = fa->fa_info;
2029 /* local and main table can share the same trie,
2030 * so don't notify twice for the same entry.
2032 if (tb->tb_id != fa->tb_id)
2035 call_fib_entry_notifier(nb, net, FIB_EVENT_ENTRY_ADD, l->key,
2036 KEYLENGTH - fa->fa_slen, fa);
2040 static void fib_table_notify(struct net *net, struct fib_table *tb,
2041 struct notifier_block *nb)
2043 struct trie *t = (struct trie *)tb->tb_data;
2044 struct key_vector *l, *tp = t->kv;
2047 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2048 fib_leaf_notify(net, l, tb, nb);
2051 /* stop in case of wrap around */
2057 void fib_notify(struct net *net, struct notifier_block *nb)
2061 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2062 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2063 struct fib_table *tb;
2065 hlist_for_each_entry_rcu(tb, head, tb_hlist)
2066 fib_table_notify(net, tb, nb);
2070 static void __trie_free_rcu(struct rcu_head *head)
2072 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2073 #ifdef CONFIG_IP_FIB_TRIE_STATS
2074 struct trie *t = (struct trie *)tb->tb_data;
2076 if (tb->tb_data == tb->__data)
2077 free_percpu(t->stats);
2078 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2082 void fib_free_table(struct fib_table *tb)
2084 call_rcu(&tb->rcu, __trie_free_rcu);
2087 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2088 struct sk_buff *skb, struct netlink_callback *cb,
2089 struct fib_dump_filter *filter)
2091 unsigned int flags = NLM_F_MULTI;
2092 __be32 xkey = htonl(l->key);
2093 int i, s_i, i_fa, s_fa, err;
2094 struct fib_alias *fa;
2096 if (filter->filter_set ||
2097 !filter->dump_exceptions || !filter->dump_routes)
2098 flags |= NLM_F_DUMP_FILTERED;
2104 /* rcu_read_lock is hold by caller */
2105 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2106 struct fib_info *fi = fa->fa_info;
2113 if (tb->tb_id != fa->tb_id)
2116 if (filter->filter_set) {
2117 if (filter->rt_type && fa->fa_type != filter->rt_type)
2120 if ((filter->protocol &&
2121 fi->fib_protocol != filter->protocol))
2125 !fib_info_nh_uses_dev(fi, filter->dev))
2129 if (filter->dump_routes) {
2131 err = fib_dump_info(skb,
2132 NETLINK_CB(cb->skb).portid,
2135 tb->tb_id, fa->fa_type,
2137 KEYLENGTH - fa->fa_slen,
2138 fa->fa_tos, fi, flags);
2146 if (filter->dump_exceptions) {
2147 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2166 /* rcu_read_lock needs to be hold by caller from readside */
2167 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2168 struct netlink_callback *cb, struct fib_dump_filter *filter)
2170 struct trie *t = (struct trie *)tb->tb_data;
2171 struct key_vector *l, *tp = t->kv;
2172 /* Dump starting at last key.
2173 * Note: 0.0.0.0/0 (ie default) is first key.
2175 int count = cb->args[2];
2176 t_key key = cb->args[3];
2178 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2181 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2184 cb->args[2] = count;
2191 memset(&cb->args[4], 0,
2192 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2194 /* stop loop if key wrapped back to 0 */
2200 cb->args[2] = count;
2205 void __init fib_trie_init(void)
2207 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2208 sizeof(struct fib_alias),
2209 0, SLAB_PANIC, NULL);
2211 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2213 0, SLAB_PANIC, NULL);
2216 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2218 struct fib_table *tb;
2220 size_t sz = sizeof(*tb);
2223 sz += sizeof(struct trie);
2225 tb = kzalloc(sz, GFP_KERNEL);
2230 tb->tb_num_default = 0;
2231 tb->tb_data = (alias ? alias->__data : tb->__data);
2236 t = (struct trie *) tb->tb_data;
2237 t->kv[0].pos = KEYLENGTH;
2238 t->kv[0].slen = KEYLENGTH;
2239 #ifdef CONFIG_IP_FIB_TRIE_STATS
2240 t->stats = alloc_percpu(struct trie_use_stats);
2250 #ifdef CONFIG_PROC_FS
2251 /* Depth first Trie walk iterator */
2252 struct fib_trie_iter {
2253 struct seq_net_private p;
2254 struct fib_table *tb;
2255 struct key_vector *tnode;
2260 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2262 unsigned long cindex = iter->index;
2263 struct key_vector *pn = iter->tnode;
2266 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2267 iter->tnode, iter->index, iter->depth);
2269 while (!IS_TRIE(pn)) {
2270 while (cindex < child_length(pn)) {
2271 struct key_vector *n = get_child_rcu(pn, cindex++);
2278 iter->index = cindex;
2280 /* push down one level */
2289 /* Current node exhausted, pop back up */
2291 pn = node_parent_rcu(pn);
2292 cindex = get_index(pkey, pn) + 1;
2296 /* record root node so further searches know we are done */
2303 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2306 struct key_vector *n, *pn;
2312 n = rcu_dereference(pn->tnode[0]);
2329 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2331 struct key_vector *n;
2332 struct fib_trie_iter iter;
2334 memset(s, 0, sizeof(*s));
2337 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2339 struct fib_alias *fa;
2342 s->totdepth += iter.depth;
2343 if (iter.depth > s->maxdepth)
2344 s->maxdepth = iter.depth;
2346 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2350 if (n->bits < MAX_STAT_DEPTH)
2351 s->nodesizes[n->bits]++;
2352 s->nullpointers += tn_info(n)->empty_children;
2359 * This outputs /proc/net/fib_triestats
2361 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2363 unsigned int i, max, pointers, bytes, avdepth;
2366 avdepth = stat->totdepth*100 / stat->leaves;
2370 seq_printf(seq, "\tAver depth: %u.%02d\n",
2371 avdepth / 100, avdepth % 100);
2372 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2374 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2375 bytes = LEAF_SIZE * stat->leaves;
2377 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2378 bytes += sizeof(struct fib_alias) * stat->prefixes;
2380 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2381 bytes += TNODE_SIZE(0) * stat->tnodes;
2383 max = MAX_STAT_DEPTH;
2384 while (max > 0 && stat->nodesizes[max-1] == 0)
2388 for (i = 1; i < max; i++)
2389 if (stat->nodesizes[i] != 0) {
2390 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2391 pointers += (1<<i) * stat->nodesizes[i];
2393 seq_putc(seq, '\n');
2394 seq_printf(seq, "\tPointers: %u\n", pointers);
2396 bytes += sizeof(struct key_vector *) * pointers;
2397 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2398 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2401 #ifdef CONFIG_IP_FIB_TRIE_STATS
2402 static void trie_show_usage(struct seq_file *seq,
2403 const struct trie_use_stats __percpu *stats)
2405 struct trie_use_stats s = { 0 };
2408 /* loop through all of the CPUs and gather up the stats */
2409 for_each_possible_cpu(cpu) {
2410 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2412 s.gets += pcpu->gets;
2413 s.backtrack += pcpu->backtrack;
2414 s.semantic_match_passed += pcpu->semantic_match_passed;
2415 s.semantic_match_miss += pcpu->semantic_match_miss;
2416 s.null_node_hit += pcpu->null_node_hit;
2417 s.resize_node_skipped += pcpu->resize_node_skipped;
2420 seq_printf(seq, "\nCounters:\n---------\n");
2421 seq_printf(seq, "gets = %u\n", s.gets);
2422 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2423 seq_printf(seq, "semantic match passed = %u\n",
2424 s.semantic_match_passed);
2425 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2426 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2427 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2429 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2431 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2433 if (tb->tb_id == RT_TABLE_LOCAL)
2434 seq_puts(seq, "Local:\n");
2435 else if (tb->tb_id == RT_TABLE_MAIN)
2436 seq_puts(seq, "Main:\n");
2438 seq_printf(seq, "Id %d:\n", tb->tb_id);
2442 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2444 struct net *net = (struct net *)seq->private;
2448 "Basic info: size of leaf:"
2449 " %zd bytes, size of tnode: %zd bytes.\n",
2450 LEAF_SIZE, TNODE_SIZE(0));
2452 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2453 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2454 struct fib_table *tb;
2456 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2457 struct trie *t = (struct trie *) tb->tb_data;
2458 struct trie_stat stat;
2463 fib_table_print(seq, tb);
2465 trie_collect_stats(t, &stat);
2466 trie_show_stats(seq, &stat);
2467 #ifdef CONFIG_IP_FIB_TRIE_STATS
2468 trie_show_usage(seq, t->stats);
2476 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2478 struct fib_trie_iter *iter = seq->private;
2479 struct net *net = seq_file_net(seq);
2483 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2484 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2485 struct fib_table *tb;
2487 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2488 struct key_vector *n;
2490 for (n = fib_trie_get_first(iter,
2491 (struct trie *) tb->tb_data);
2492 n; n = fib_trie_get_next(iter))
2503 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2507 return fib_trie_get_idx(seq, *pos);
2510 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2512 struct fib_trie_iter *iter = seq->private;
2513 struct net *net = seq_file_net(seq);
2514 struct fib_table *tb = iter->tb;
2515 struct hlist_node *tb_node;
2517 struct key_vector *n;
2520 /* next node in same table */
2521 n = fib_trie_get_next(iter);
2525 /* walk rest of this hash chain */
2526 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2527 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2528 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2529 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2534 /* new hash chain */
2535 while (++h < FIB_TABLE_HASHSZ) {
2536 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2537 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2538 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2550 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2556 static void seq_indent(struct seq_file *seq, int n)
2562 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2565 case RT_SCOPE_UNIVERSE: return "universe";
2566 case RT_SCOPE_SITE: return "site";
2567 case RT_SCOPE_LINK: return "link";
2568 case RT_SCOPE_HOST: return "host";
2569 case RT_SCOPE_NOWHERE: return "nowhere";
2571 snprintf(buf, len, "scope=%d", s);
2576 static const char *const rtn_type_names[__RTN_MAX] = {
2577 [RTN_UNSPEC] = "UNSPEC",
2578 [RTN_UNICAST] = "UNICAST",
2579 [RTN_LOCAL] = "LOCAL",
2580 [RTN_BROADCAST] = "BROADCAST",
2581 [RTN_ANYCAST] = "ANYCAST",
2582 [RTN_MULTICAST] = "MULTICAST",
2583 [RTN_BLACKHOLE] = "BLACKHOLE",
2584 [RTN_UNREACHABLE] = "UNREACHABLE",
2585 [RTN_PROHIBIT] = "PROHIBIT",
2586 [RTN_THROW] = "THROW",
2588 [RTN_XRESOLVE] = "XRESOLVE",
2591 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2593 if (t < __RTN_MAX && rtn_type_names[t])
2594 return rtn_type_names[t];
2595 snprintf(buf, len, "type %u", t);
2599 /* Pretty print the trie */
2600 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2602 const struct fib_trie_iter *iter = seq->private;
2603 struct key_vector *n = v;
2605 if (IS_TRIE(node_parent_rcu(n)))
2606 fib_table_print(seq, iter->tb);
2609 __be32 prf = htonl(n->key);
2611 seq_indent(seq, iter->depth-1);
2612 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2613 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2614 tn_info(n)->full_children,
2615 tn_info(n)->empty_children);
2617 __be32 val = htonl(n->key);
2618 struct fib_alias *fa;
2620 seq_indent(seq, iter->depth);
2621 seq_printf(seq, " |-- %pI4\n", &val);
2623 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2624 char buf1[32], buf2[32];
2626 seq_indent(seq, iter->depth + 1);
2627 seq_printf(seq, " /%zu %s %s",
2628 KEYLENGTH - fa->fa_slen,
2629 rtn_scope(buf1, sizeof(buf1),
2630 fa->fa_info->fib_scope),
2631 rtn_type(buf2, sizeof(buf2),
2634 seq_printf(seq, " tos=%d", fa->fa_tos);
2635 seq_putc(seq, '\n');
2642 static const struct seq_operations fib_trie_seq_ops = {
2643 .start = fib_trie_seq_start,
2644 .next = fib_trie_seq_next,
2645 .stop = fib_trie_seq_stop,
2646 .show = fib_trie_seq_show,
2649 struct fib_route_iter {
2650 struct seq_net_private p;
2651 struct fib_table *main_tb;
2652 struct key_vector *tnode;
2657 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2660 struct key_vector *l, **tp = &iter->tnode;
2663 /* use cached location of previously found key */
2664 if (iter->pos > 0 && pos >= iter->pos) {
2673 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2678 /* handle unlikely case of a key wrap */
2684 iter->key = l->key; /* remember it */
2686 iter->pos = 0; /* forget it */
2691 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2694 struct fib_route_iter *iter = seq->private;
2695 struct fib_table *tb;
2700 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2705 t = (struct trie *)tb->tb_data;
2706 iter->tnode = t->kv;
2709 return fib_route_get_idx(iter, *pos);
2712 iter->key = KEY_MAX;
2714 return SEQ_START_TOKEN;
2717 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2719 struct fib_route_iter *iter = seq->private;
2720 struct key_vector *l = NULL;
2721 t_key key = iter->key + 1;
2725 /* only allow key of 0 for start of sequence */
2726 if ((v == SEQ_START_TOKEN) || key)
2727 l = leaf_walk_rcu(&iter->tnode, key);
2739 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2745 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2747 unsigned int flags = 0;
2749 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2752 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2754 if (nhc->nhc_gw.ipv4)
2755 flags |= RTF_GATEWAY;
2757 if (mask == htonl(0xFFFFFFFF))
2764 * This outputs /proc/net/route.
2765 * The format of the file is not supposed to be changed
2766 * and needs to be same as fib_hash output to avoid breaking
2769 static int fib_route_seq_show(struct seq_file *seq, void *v)
2771 struct fib_route_iter *iter = seq->private;
2772 struct fib_table *tb = iter->main_tb;
2773 struct fib_alias *fa;
2774 struct key_vector *l = v;
2777 if (v == SEQ_START_TOKEN) {
2778 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2779 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2784 prefix = htonl(l->key);
2786 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2787 struct fib_info *fi = fa->fa_info;
2788 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2789 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2791 if ((fa->fa_type == RTN_BROADCAST) ||
2792 (fa->fa_type == RTN_MULTICAST))
2795 if (fa->tb_id != tb->tb_id)
2798 seq_setwidth(seq, 127);
2801 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2804 if (nhc->nhc_gw_family == AF_INET)
2805 gw = nhc->nhc_gw.ipv4;
2808 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2809 "%d\t%08X\t%d\t%u\t%u",
2810 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2811 prefix, gw, flags, 0, 0,
2815 fi->fib_advmss + 40 : 0),
2820 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2821 "%d\t%08X\t%d\t%u\t%u",
2822 prefix, 0, flags, 0, 0, 0,
2831 static const struct seq_operations fib_route_seq_ops = {
2832 .start = fib_route_seq_start,
2833 .next = fib_route_seq_next,
2834 .stop = fib_route_seq_stop,
2835 .show = fib_route_seq_show,
2838 int __net_init fib_proc_init(struct net *net)
2840 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2841 sizeof(struct fib_trie_iter)))
2844 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
2845 fib_triestat_seq_show, NULL))
2848 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
2849 sizeof(struct fib_route_iter)))
2855 remove_proc_entry("fib_triestat", net->proc_net);
2857 remove_proc_entry("fib_trie", net->proc_net);
2862 void __net_exit fib_proc_exit(struct net *net)
2864 remove_proc_entry("fib_trie", net->proc_net);
2865 remove_proc_entry("fib_triestat", net->proc_net);
2866 remove_proc_entry("route", net->proc_net);
2869 #endif /* CONFIG_PROC_FS */