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,
78 enum fib_event_type event_type, u32 dst,
79 int dst_len, struct fib_alias *fa,
80 struct netlink_ext_ack *extack)
82 struct fib_entry_notifier_info info = {
83 .info.extack = extack,
91 return call_fib4_notifier(nb, event_type, &info.info);
94 static int call_fib_entry_notifiers(struct net *net,
95 enum fib_event_type event_type, u32 dst,
96 int dst_len, struct fib_alias *fa,
97 struct netlink_ext_ack *extack)
99 struct fib_entry_notifier_info info = {
100 .info.extack = extack,
108 return call_fib4_notifiers(net, event_type, &info.info);
111 #define MAX_STAT_DEPTH 32
113 #define KEYLENGTH (8*sizeof(t_key))
114 #define KEY_MAX ((t_key)~0)
116 typedef unsigned int t_key;
118 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
119 #define IS_TNODE(n) ((n)->bits)
120 #define IS_LEAF(n) (!(n)->bits)
124 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
125 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
128 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
129 struct hlist_head leaf;
130 /* This array is valid if (pos | bits) > 0 (TNODE) */
131 struct key_vector __rcu *tnode[0];
137 t_key empty_children; /* KEYLENGTH bits needed */
138 t_key full_children; /* KEYLENGTH bits needed */
139 struct key_vector __rcu *parent;
140 struct key_vector kv[1];
141 #define tn_bits kv[0].bits
144 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
145 #define LEAF_SIZE TNODE_SIZE(1)
147 #ifdef CONFIG_IP_FIB_TRIE_STATS
148 struct trie_use_stats {
150 unsigned int backtrack;
151 unsigned int semantic_match_passed;
152 unsigned int semantic_match_miss;
153 unsigned int null_node_hit;
154 unsigned int resize_node_skipped;
159 unsigned int totdepth;
160 unsigned int maxdepth;
163 unsigned int nullpointers;
164 unsigned int prefixes;
165 unsigned int nodesizes[MAX_STAT_DEPTH];
169 struct key_vector kv[1];
170 #ifdef CONFIG_IP_FIB_TRIE_STATS
171 struct trie_use_stats __percpu *stats;
175 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
176 static unsigned int tnode_free_size;
179 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
180 * especially useful before resizing the root node with PREEMPT_NONE configs;
181 * the value was obtained experimentally, aiming to avoid visible slowdown.
183 unsigned int sysctl_fib_sync_mem = 512 * 1024;
184 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
185 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
187 static struct kmem_cache *fn_alias_kmem __ro_after_init;
188 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
190 static inline struct tnode *tn_info(struct key_vector *kv)
192 return container_of(kv, struct tnode, kv[0]);
195 /* caller must hold RTNL */
196 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
197 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
199 /* caller must hold RCU read lock or RTNL */
200 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
201 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
203 /* wrapper for rcu_assign_pointer */
204 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
207 rcu_assign_pointer(tn_info(n)->parent, tp);
210 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
212 /* This provides us with the number of children in this node, in the case of a
213 * leaf this will return 0 meaning none of the children are accessible.
215 static inline unsigned long child_length(const struct key_vector *tn)
217 return (1ul << tn->bits) & ~(1ul);
220 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
222 static inline unsigned long get_index(t_key key, struct key_vector *kv)
224 unsigned long index = key ^ kv->key;
226 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
229 return index >> kv->pos;
232 /* To understand this stuff, an understanding of keys and all their bits is
233 * necessary. Every node in the trie has a key associated with it, but not
234 * all of the bits in that key are significant.
236 * Consider a node 'n' and its parent 'tp'.
238 * If n is a leaf, every bit in its key is significant. Its presence is
239 * necessitated by path compression, since during a tree traversal (when
240 * searching for a leaf - unless we are doing an insertion) we will completely
241 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
242 * a potentially successful search, that we have indeed been walking the
245 * Note that we can never "miss" the correct key in the tree if present by
246 * following the wrong path. Path compression ensures that segments of the key
247 * that are the same for all keys with a given prefix are skipped, but the
248 * skipped part *is* identical for each node in the subtrie below the skipped
249 * bit! trie_insert() in this implementation takes care of that.
251 * if n is an internal node - a 'tnode' here, the various parts of its key
252 * have many different meanings.
255 * _________________________________________________________________
256 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
257 * -----------------------------------------------------------------
258 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
260 * _________________________________________________________________
261 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
262 * -----------------------------------------------------------------
263 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
270 * First, let's just ignore the bits that come before the parent tp, that is
271 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
272 * point we do not use them for anything.
274 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
275 * index into the parent's child array. That is, they will be used to find
276 * 'n' among tp's children.
278 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
281 * All the bits we have seen so far are significant to the node n. The rest
282 * of the bits are really not needed or indeed known in n->key.
284 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
285 * n's child array, and will of course be different for each child.
287 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
291 static const int halve_threshold = 25;
292 static const int inflate_threshold = 50;
293 static const int halve_threshold_root = 15;
294 static const int inflate_threshold_root = 30;
296 static void __alias_free_mem(struct rcu_head *head)
298 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
299 kmem_cache_free(fn_alias_kmem, fa);
302 static inline void alias_free_mem_rcu(struct fib_alias *fa)
304 call_rcu(&fa->rcu, __alias_free_mem);
307 #define TNODE_KMALLOC_MAX \
308 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
309 #define TNODE_VMALLOC_MAX \
310 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
312 static void __node_free_rcu(struct rcu_head *head)
314 struct tnode *n = container_of(head, struct tnode, rcu);
317 kmem_cache_free(trie_leaf_kmem, n);
322 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
324 static struct tnode *tnode_alloc(int bits)
328 /* verify bits is within bounds */
329 if (bits > TNODE_VMALLOC_MAX)
332 /* determine size and verify it is non-zero and didn't overflow */
333 size = TNODE_SIZE(1ul << bits);
335 if (size <= PAGE_SIZE)
336 return kzalloc(size, GFP_KERNEL);
338 return vzalloc(size);
341 static inline void empty_child_inc(struct key_vector *n)
343 tn_info(n)->empty_children++;
345 if (!tn_info(n)->empty_children)
346 tn_info(n)->full_children++;
349 static inline void empty_child_dec(struct key_vector *n)
351 if (!tn_info(n)->empty_children)
352 tn_info(n)->full_children--;
354 tn_info(n)->empty_children--;
357 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
359 struct key_vector *l;
362 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
366 /* initialize key vector */
371 l->slen = fa->fa_slen;
373 /* link leaf to fib alias */
374 INIT_HLIST_HEAD(&l->leaf);
375 hlist_add_head(&fa->fa_list, &l->leaf);
380 static struct key_vector *tnode_new(t_key key, int pos, int bits)
382 unsigned int shift = pos + bits;
383 struct key_vector *tn;
386 /* verify bits and pos their msb bits clear and values are valid */
387 BUG_ON(!bits || (shift > KEYLENGTH));
389 tnode = tnode_alloc(bits);
393 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
394 sizeof(struct key_vector *) << bits);
396 if (bits == KEYLENGTH)
397 tnode->full_children = 1;
399 tnode->empty_children = 1ul << bits;
402 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
410 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
411 * and no bits are skipped. See discussion in dyntree paper p. 6
413 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
415 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
418 /* Add a child at position i overwriting the old value.
419 * Update the value of full_children and empty_children.
421 static void put_child(struct key_vector *tn, unsigned long i,
422 struct key_vector *n)
424 struct key_vector *chi = get_child(tn, i);
427 BUG_ON(i >= child_length(tn));
429 /* update emptyChildren, overflow into fullChildren */
435 /* update fullChildren */
436 wasfull = tnode_full(tn, chi);
437 isfull = tnode_full(tn, n);
439 if (wasfull && !isfull)
440 tn_info(tn)->full_children--;
441 else if (!wasfull && isfull)
442 tn_info(tn)->full_children++;
444 if (n && (tn->slen < n->slen))
447 rcu_assign_pointer(tn->tnode[i], n);
450 static void update_children(struct key_vector *tn)
454 /* update all of the child parent pointers */
455 for (i = child_length(tn); i;) {
456 struct key_vector *inode = get_child(tn, --i);
461 /* Either update the children of a tnode that
462 * already belongs to us or update the child
463 * to point to ourselves.
465 if (node_parent(inode) == tn)
466 update_children(inode);
468 node_set_parent(inode, tn);
472 static inline void put_child_root(struct key_vector *tp, t_key key,
473 struct key_vector *n)
476 rcu_assign_pointer(tp->tnode[0], n);
478 put_child(tp, get_index(key, tp), n);
481 static inline void tnode_free_init(struct key_vector *tn)
483 tn_info(tn)->rcu.next = NULL;
486 static inline void tnode_free_append(struct key_vector *tn,
487 struct key_vector *n)
489 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
490 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
493 static void tnode_free(struct key_vector *tn)
495 struct callback_head *head = &tn_info(tn)->rcu;
499 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
502 tn = container_of(head, struct tnode, rcu)->kv;
505 if (tnode_free_size >= sysctl_fib_sync_mem) {
511 static struct key_vector *replace(struct trie *t,
512 struct key_vector *oldtnode,
513 struct key_vector *tn)
515 struct key_vector *tp = node_parent(oldtnode);
518 /* setup the parent pointer out of and back into this node */
519 NODE_INIT_PARENT(tn, tp);
520 put_child_root(tp, tn->key, tn);
522 /* update all of the child parent pointers */
525 /* all pointers should be clean so we are done */
526 tnode_free(oldtnode);
528 /* resize children now that oldtnode is freed */
529 for (i = child_length(tn); i;) {
530 struct key_vector *inode = get_child(tn, --i);
532 /* resize child node */
533 if (tnode_full(tn, inode))
534 tn = resize(t, inode);
540 static struct key_vector *inflate(struct trie *t,
541 struct key_vector *oldtnode)
543 struct key_vector *tn;
547 pr_debug("In inflate\n");
549 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
553 /* prepare oldtnode to be freed */
554 tnode_free_init(oldtnode);
556 /* Assemble all of the pointers in our cluster, in this case that
557 * represents all of the pointers out of our allocated nodes that
558 * point to existing tnodes and the links between our allocated
561 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
562 struct key_vector *inode = get_child(oldtnode, --i);
563 struct key_vector *node0, *node1;
570 /* A leaf or an internal node with skipped bits */
571 if (!tnode_full(oldtnode, inode)) {
572 put_child(tn, get_index(inode->key, tn), inode);
576 /* drop the node in the old tnode free list */
577 tnode_free_append(oldtnode, inode);
579 /* An internal node with two children */
580 if (inode->bits == 1) {
581 put_child(tn, 2 * i + 1, get_child(inode, 1));
582 put_child(tn, 2 * i, get_child(inode, 0));
586 /* We will replace this node 'inode' with two new
587 * ones, 'node0' and 'node1', each with half of the
588 * original children. The two new nodes will have
589 * a position one bit further down the key and this
590 * means that the "significant" part of their keys
591 * (see the discussion near the top of this file)
592 * will differ by one bit, which will be "0" in
593 * node0's key and "1" in node1's key. Since we are
594 * moving the key position by one step, the bit that
595 * we are moving away from - the bit at position
596 * (tn->pos) - is the one that will differ between
597 * node0 and node1. So... we synthesize that bit in the
600 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
603 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
605 tnode_free_append(tn, node1);
608 tnode_free_append(tn, node0);
610 /* populate child pointers in new nodes */
611 for (k = child_length(inode), j = k / 2; j;) {
612 put_child(node1, --j, get_child(inode, --k));
613 put_child(node0, j, get_child(inode, j));
614 put_child(node1, --j, get_child(inode, --k));
615 put_child(node0, j, get_child(inode, j));
618 /* link new nodes to parent */
619 NODE_INIT_PARENT(node1, tn);
620 NODE_INIT_PARENT(node0, tn);
622 /* link parent to nodes */
623 put_child(tn, 2 * i + 1, node1);
624 put_child(tn, 2 * i, node0);
627 /* setup the parent pointers into and out of this node */
628 return replace(t, oldtnode, tn);
630 /* all pointers should be clean so we are done */
636 static struct key_vector *halve(struct trie *t,
637 struct key_vector *oldtnode)
639 struct key_vector *tn;
642 pr_debug("In halve\n");
644 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
648 /* prepare oldtnode to be freed */
649 tnode_free_init(oldtnode);
651 /* Assemble all of the pointers in our cluster, in this case that
652 * represents all of the pointers out of our allocated nodes that
653 * point to existing tnodes and the links between our allocated
656 for (i = child_length(oldtnode); i;) {
657 struct key_vector *node1 = get_child(oldtnode, --i);
658 struct key_vector *node0 = get_child(oldtnode, --i);
659 struct key_vector *inode;
661 /* At least one of the children is empty */
662 if (!node1 || !node0) {
663 put_child(tn, i / 2, node1 ? : node0);
667 /* Two nonempty children */
668 inode = tnode_new(node0->key, oldtnode->pos, 1);
671 tnode_free_append(tn, inode);
673 /* initialize pointers out of node */
674 put_child(inode, 1, node1);
675 put_child(inode, 0, node0);
676 NODE_INIT_PARENT(inode, tn);
678 /* link parent to node */
679 put_child(tn, i / 2, inode);
682 /* setup the parent pointers into and out of this node */
683 return replace(t, oldtnode, tn);
685 /* all pointers should be clean so we are done */
691 static struct key_vector *collapse(struct trie *t,
692 struct key_vector *oldtnode)
694 struct key_vector *n, *tp;
697 /* scan the tnode looking for that one child that might still exist */
698 for (n = NULL, i = child_length(oldtnode); !n && i;)
699 n = get_child(oldtnode, --i);
701 /* compress one level */
702 tp = node_parent(oldtnode);
703 put_child_root(tp, oldtnode->key, n);
704 node_set_parent(n, tp);
712 static unsigned char update_suffix(struct key_vector *tn)
714 unsigned char slen = tn->pos;
715 unsigned long stride, i;
716 unsigned char slen_max;
718 /* only vector 0 can have a suffix length greater than or equal to
719 * tn->pos + tn->bits, the second highest node will have a suffix
720 * length at most of tn->pos + tn->bits - 1
722 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
724 /* search though the list of children looking for nodes that might
725 * have a suffix greater than the one we currently have. This is
726 * why we start with a stride of 2 since a stride of 1 would
727 * represent the nodes with suffix length equal to tn->pos
729 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
730 struct key_vector *n = get_child(tn, i);
732 if (!n || (n->slen <= slen))
735 /* update stride and slen based on new value */
736 stride <<= (n->slen - slen);
740 /* stop searching if we have hit the maximum possible value */
741 if (slen >= slen_max)
750 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
751 * the Helsinki University of Technology and Matti Tikkanen of Nokia
752 * Telecommunications, page 6:
753 * "A node is doubled if the ratio of non-empty children to all
754 * children in the *doubled* node is at least 'high'."
756 * 'high' in this instance is the variable 'inflate_threshold'. It
757 * is expressed as a percentage, so we multiply it with
758 * child_length() and instead of multiplying by 2 (since the
759 * child array will be doubled by inflate()) and multiplying
760 * the left-hand side by 100 (to handle the percentage thing) we
761 * multiply the left-hand side by 50.
763 * The left-hand side may look a bit weird: child_length(tn)
764 * - tn->empty_children is of course the number of non-null children
765 * in the current node. tn->full_children is the number of "full"
766 * children, that is non-null tnodes with a skip value of 0.
767 * All of those will be doubled in the resulting inflated tnode, so
768 * we just count them one extra time here.
770 * A clearer way to write this would be:
772 * to_be_doubled = tn->full_children;
773 * not_to_be_doubled = child_length(tn) - tn->empty_children -
776 * new_child_length = child_length(tn) * 2;
778 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
780 * if (new_fill_factor >= inflate_threshold)
782 * ...and so on, tho it would mess up the while () loop.
785 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
789 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
790 * inflate_threshold * new_child_length
792 * expand not_to_be_doubled and to_be_doubled, and shorten:
793 * 100 * (child_length(tn) - tn->empty_children +
794 * tn->full_children) >= inflate_threshold * new_child_length
796 * expand new_child_length:
797 * 100 * (child_length(tn) - tn->empty_children +
798 * tn->full_children) >=
799 * inflate_threshold * child_length(tn) * 2
802 * 50 * (tn->full_children + child_length(tn) -
803 * tn->empty_children) >= inflate_threshold *
807 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
809 unsigned long used = child_length(tn);
810 unsigned long threshold = used;
812 /* Keep root node larger */
813 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
814 used -= tn_info(tn)->empty_children;
815 used += tn_info(tn)->full_children;
817 /* if bits == KEYLENGTH then pos = 0, and will fail below */
819 return (used > 1) && tn->pos && ((50 * used) >= threshold);
822 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
824 unsigned long used = child_length(tn);
825 unsigned long threshold = used;
827 /* Keep root node larger */
828 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
829 used -= tn_info(tn)->empty_children;
831 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
833 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
836 static inline bool should_collapse(struct key_vector *tn)
838 unsigned long used = child_length(tn);
840 used -= tn_info(tn)->empty_children;
842 /* account for bits == KEYLENGTH case */
843 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
846 /* One child or none, time to drop us from the trie */
851 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
853 #ifdef CONFIG_IP_FIB_TRIE_STATS
854 struct trie_use_stats __percpu *stats = t->stats;
856 struct key_vector *tp = node_parent(tn);
857 unsigned long cindex = get_index(tn->key, tp);
858 int max_work = MAX_WORK;
860 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
861 tn, inflate_threshold, halve_threshold);
863 /* track the tnode via the pointer from the parent instead of
864 * doing it ourselves. This way we can let RCU fully do its
865 * thing without us interfering
867 BUG_ON(tn != get_child(tp, cindex));
869 /* Double as long as the resulting node has a number of
870 * nonempty nodes that are above the threshold.
872 while (should_inflate(tp, tn) && max_work) {
875 #ifdef CONFIG_IP_FIB_TRIE_STATS
876 this_cpu_inc(stats->resize_node_skipped);
882 tn = get_child(tp, cindex);
885 /* update parent in case inflate failed */
886 tp = node_parent(tn);
888 /* Return if at least one inflate is run */
889 if (max_work != MAX_WORK)
892 /* Halve as long as the number of empty children in this
893 * node is above threshold.
895 while (should_halve(tp, tn) && max_work) {
898 #ifdef CONFIG_IP_FIB_TRIE_STATS
899 this_cpu_inc(stats->resize_node_skipped);
905 tn = get_child(tp, cindex);
908 /* Only one child remains */
909 if (should_collapse(tn))
910 return collapse(t, tn);
912 /* update parent in case halve failed */
913 return node_parent(tn);
916 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
918 unsigned char node_slen = tn->slen;
920 while ((node_slen > tn->pos) && (node_slen > slen)) {
921 slen = update_suffix(tn);
922 if (node_slen == slen)
925 tn = node_parent(tn);
926 node_slen = tn->slen;
930 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
932 while (tn->slen < slen) {
934 tn = node_parent(tn);
938 /* rcu_read_lock needs to be hold by caller from readside */
939 static struct key_vector *fib_find_node(struct trie *t,
940 struct key_vector **tp, u32 key)
942 struct key_vector *pn, *n = t->kv;
943 unsigned long index = 0;
947 n = get_child_rcu(n, index);
952 index = get_cindex(key, n);
954 /* This bit of code is a bit tricky but it combines multiple
955 * checks into a single check. The prefix consists of the
956 * prefix plus zeros for the bits in the cindex. The index
957 * is the difference between the key and this value. From
958 * this we can actually derive several pieces of data.
959 * if (index >= (1ul << bits))
960 * we have a mismatch in skip bits and failed
962 * we know the value is cindex
964 * This check is safe even if bits == KEYLENGTH due to the
965 * fact that we can only allocate a node with 32 bits if a
966 * long is greater than 32 bits.
968 if (index >= (1ul << n->bits)) {
973 /* keep searching until we find a perfect match leaf or NULL */
974 } while (IS_TNODE(n));
981 /* Return the first fib alias matching TOS with
982 * priority less than or equal to PRIO.
983 * If 'find_first' is set, return the first matching
984 * fib alias, regardless of TOS and priority.
986 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
987 u8 tos, u32 prio, u32 tb_id,
990 struct fib_alias *fa;
995 hlist_for_each_entry(fa, fah, fa_list) {
996 if (fa->fa_slen < slen)
998 if (fa->fa_slen != slen)
1000 if (fa->tb_id > tb_id)
1002 if (fa->tb_id != tb_id)
1006 if (fa->fa_tos > tos)
1008 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1015 static struct fib_alias *
1016 fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1018 u8 slen = KEYLENGTH - fri->dst_len;
1019 struct key_vector *l, *tp;
1020 struct fib_table *tb;
1021 struct fib_alias *fa;
1024 tb = fib_get_table(net, fri->tb_id);
1028 t = (struct trie *)tb->tb_data;
1029 l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1033 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1034 if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1035 fa->fa_tos == fri->tos && fa->fa_info == fri->fi &&
1036 fa->fa_type == fri->type)
1043 void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1045 struct fib_alias *fa_match;
1049 fa_match = fib_find_matching_alias(net, fri);
1053 fa_match->offload = fri->offload;
1054 fa_match->trap = fri->trap;
1059 EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1061 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1063 while (!IS_TRIE(tn))
1067 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1068 struct fib_alias *new, t_key key)
1070 struct key_vector *n, *l;
1072 l = leaf_new(key, new);
1076 /* retrieve child from parent node */
1077 n = get_child(tp, get_index(key, tp));
1079 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1081 * Add a new tnode here
1082 * first tnode need some special handling
1083 * leaves us in position for handling as case 3
1086 struct key_vector *tn;
1088 tn = tnode_new(key, __fls(key ^ n->key), 1);
1092 /* initialize routes out of node */
1093 NODE_INIT_PARENT(tn, tp);
1094 put_child(tn, get_index(key, tn) ^ 1, n);
1096 /* start adding routes into the node */
1097 put_child_root(tp, key, tn);
1098 node_set_parent(n, tn);
1100 /* parent now has a NULL spot where the leaf can go */
1104 /* Case 3: n is NULL, and will just insert a new leaf */
1105 node_push_suffix(tp, new->fa_slen);
1106 NODE_INIT_PARENT(l, tp);
1107 put_child_root(tp, key, l);
1108 trie_rebalance(t, tp);
1117 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1118 struct key_vector *l, struct fib_alias *new,
1119 struct fib_alias *fa, t_key key)
1122 return fib_insert_node(t, tp, new, key);
1125 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1127 struct fib_alias *last;
1129 hlist_for_each_entry(last, &l->leaf, fa_list) {
1130 if (new->fa_slen < last->fa_slen)
1132 if ((new->fa_slen == last->fa_slen) &&
1133 (new->tb_id > last->tb_id))
1139 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1141 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1144 /* if we added to the tail node then we need to update slen */
1145 if (l->slen < new->fa_slen) {
1146 l->slen = new->fa_slen;
1147 node_push_suffix(tp, new->fa_slen);
1153 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1155 if (plen > KEYLENGTH) {
1156 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1160 if ((plen < KEYLENGTH) && (key << plen)) {
1161 NL_SET_ERR_MSG(extack,
1162 "Invalid prefix for given prefix length");
1169 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1170 struct key_vector *l, struct fib_alias *old);
1172 /* Caller must hold RTNL. */
1173 int fib_table_insert(struct net *net, struct fib_table *tb,
1174 struct fib_config *cfg, struct netlink_ext_ack *extack)
1176 struct trie *t = (struct trie *)tb->tb_data;
1177 struct fib_alias *fa, *new_fa;
1178 struct key_vector *l, *tp;
1179 u16 nlflags = NLM_F_EXCL;
1180 struct fib_info *fi;
1181 u8 plen = cfg->fc_dst_len;
1182 u8 slen = KEYLENGTH - plen;
1183 u8 tos = cfg->fc_tos;
1187 key = ntohl(cfg->fc_dst);
1189 if (!fib_valid_key_len(key, plen, extack))
1192 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1194 fi = fib_create_info(cfg, extack);
1200 l = fib_find_node(t, &tp, key);
1201 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1202 tb->tb_id, false) : NULL;
1204 /* Now fa, if non-NULL, points to the first fib alias
1205 * with the same keys [prefix,tos,priority], if such key already
1206 * exists or to the node before which we will insert new one.
1208 * If fa is NULL, we will need to allocate a new one and
1209 * insert to the tail of the section matching the suffix length
1213 if (fa && fa->fa_tos == tos &&
1214 fa->fa_info->fib_priority == fi->fib_priority) {
1215 struct fib_alias *fa_first, *fa_match;
1218 if (cfg->fc_nlflags & NLM_F_EXCL)
1221 nlflags &= ~NLM_F_EXCL;
1224 * 1. Find exact match for type, scope, fib_info to avoid
1226 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1230 hlist_for_each_entry_from(fa, fa_list) {
1231 if ((fa->fa_slen != slen) ||
1232 (fa->tb_id != tb->tb_id) ||
1233 (fa->fa_tos != tos))
1235 if (fa->fa_info->fib_priority != fi->fib_priority)
1237 if (fa->fa_type == cfg->fc_type &&
1238 fa->fa_info == fi) {
1244 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1245 struct fib_info *fi_drop;
1248 nlflags |= NLM_F_REPLACE;
1256 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1260 fi_drop = fa->fa_info;
1261 new_fa->fa_tos = fa->fa_tos;
1262 new_fa->fa_info = fi;
1263 new_fa->fa_type = cfg->fc_type;
1264 state = fa->fa_state;
1265 new_fa->fa_state = state & ~FA_S_ACCESSED;
1266 new_fa->fa_slen = fa->fa_slen;
1267 new_fa->tb_id = tb->tb_id;
1268 new_fa->fa_default = -1;
1269 new_fa->offload = 0;
1272 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1274 if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1275 tb->tb_id, true) == new_fa) {
1276 enum fib_event_type fib_event;
1278 fib_event = FIB_EVENT_ENTRY_REPLACE;
1279 err = call_fib_entry_notifiers(net, fib_event,
1283 hlist_replace_rcu(&new_fa->fa_list,
1285 goto out_free_new_fa;
1289 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1290 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1292 alias_free_mem_rcu(fa);
1294 fib_release_info(fi_drop);
1295 if (state & FA_S_ACCESSED)
1296 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1300 /* Error if we find a perfect match which
1301 * uses the same scope, type, and nexthop
1307 if (cfg->fc_nlflags & NLM_F_APPEND)
1308 nlflags |= NLM_F_APPEND;
1313 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1316 nlflags |= NLM_F_CREATE;
1318 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1322 new_fa->fa_info = fi;
1323 new_fa->fa_tos = tos;
1324 new_fa->fa_type = cfg->fc_type;
1325 new_fa->fa_state = 0;
1326 new_fa->fa_slen = slen;
1327 new_fa->tb_id = tb->tb_id;
1328 new_fa->fa_default = -1;
1329 new_fa->offload = 0;
1332 /* Insert new entry to the list. */
1333 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1335 goto out_free_new_fa;
1337 /* The alias was already inserted, so the node must exist. */
1338 l = l ? l : fib_find_node(t, &tp, key);
1339 if (WARN_ON_ONCE(!l))
1340 goto out_free_new_fa;
1342 if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1344 enum fib_event_type fib_event;
1346 fib_event = FIB_EVENT_ENTRY_REPLACE;
1347 err = call_fib_entry_notifiers(net, fib_event, key, plen,
1350 goto out_remove_new_fa;
1354 tb->tb_num_default++;
1356 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1357 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1358 &cfg->fc_nlinfo, nlflags);
1363 fib_remove_alias(t, tp, l, new_fa);
1365 kmem_cache_free(fn_alias_kmem, new_fa);
1367 fib_release_info(fi);
1372 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1374 t_key prefix = n->key;
1376 return (key ^ prefix) & (prefix | -prefix);
1379 /* should be called with rcu_read_lock */
1380 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1381 struct fib_result *res, int fib_flags)
1383 struct trie *t = (struct trie *) tb->tb_data;
1384 #ifdef CONFIG_IP_FIB_TRIE_STATS
1385 struct trie_use_stats __percpu *stats = t->stats;
1387 const t_key key = ntohl(flp->daddr);
1388 struct key_vector *n, *pn;
1389 struct fib_alias *fa;
1390 unsigned long index;
1396 n = get_child_rcu(pn, cindex);
1398 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1402 #ifdef CONFIG_IP_FIB_TRIE_STATS
1403 this_cpu_inc(stats->gets);
1406 /* Step 1: Travel to the longest prefix match in the trie */
1408 index = get_cindex(key, n);
1410 /* This bit of code is a bit tricky but it combines multiple
1411 * checks into a single check. The prefix consists of the
1412 * prefix plus zeros for the "bits" in the prefix. The index
1413 * is the difference between the key and this value. From
1414 * this we can actually derive several pieces of data.
1415 * if (index >= (1ul << bits))
1416 * we have a mismatch in skip bits and failed
1418 * we know the value is cindex
1420 * This check is safe even if bits == KEYLENGTH due to the
1421 * fact that we can only allocate a node with 32 bits if a
1422 * long is greater than 32 bits.
1424 if (index >= (1ul << n->bits))
1427 /* we have found a leaf. Prefixes have already been compared */
1431 /* only record pn and cindex if we are going to be chopping
1432 * bits later. Otherwise we are just wasting cycles.
1434 if (n->slen > n->pos) {
1439 n = get_child_rcu(n, index);
1444 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1446 /* record the pointer where our next node pointer is stored */
1447 struct key_vector __rcu **cptr = n->tnode;
1449 /* This test verifies that none of the bits that differ
1450 * between the key and the prefix exist in the region of
1451 * the lsb and higher in the prefix.
1453 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1456 /* exit out and process leaf */
1457 if (unlikely(IS_LEAF(n)))
1460 /* Don't bother recording parent info. Since we are in
1461 * prefix match mode we will have to come back to wherever
1462 * we started this traversal anyway
1465 while ((n = rcu_dereference(*cptr)) == NULL) {
1467 #ifdef CONFIG_IP_FIB_TRIE_STATS
1469 this_cpu_inc(stats->null_node_hit);
1471 /* If we are at cindex 0 there are no more bits for
1472 * us to strip at this level so we must ascend back
1473 * up one level to see if there are any more bits to
1474 * be stripped there.
1477 t_key pkey = pn->key;
1479 /* If we don't have a parent then there is
1480 * nothing for us to do as we do not have any
1481 * further nodes to parse.
1484 trace_fib_table_lookup(tb->tb_id, flp,
1488 #ifdef CONFIG_IP_FIB_TRIE_STATS
1489 this_cpu_inc(stats->backtrack);
1491 /* Get Child's index */
1492 pn = node_parent_rcu(pn);
1493 cindex = get_index(pkey, pn);
1496 /* strip the least significant bit from the cindex */
1497 cindex &= cindex - 1;
1499 /* grab pointer for next child node */
1500 cptr = &pn->tnode[cindex];
1505 /* this line carries forward the xor from earlier in the function */
1506 index = key ^ n->key;
1508 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1509 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1510 struct fib_info *fi = fa->fa_info;
1513 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1514 if (index >= (1ul << fa->fa_slen))
1517 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1521 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1523 fib_alias_accessed(fa);
1524 err = fib_props[fa->fa_type].error;
1525 if (unlikely(err < 0)) {
1527 #ifdef CONFIG_IP_FIB_TRIE_STATS
1528 this_cpu_inc(stats->semantic_match_passed);
1530 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1533 if (fi->fib_flags & RTNH_F_DEAD)
1536 if (unlikely(fi->nh && nexthop_is_blackhole(fi->nh))) {
1537 err = fib_props[RTN_BLACKHOLE].error;
1541 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1542 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
1544 if (nhc->nhc_flags & RTNH_F_DEAD)
1546 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1547 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1548 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1550 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1551 if (flp->flowi4_oif &&
1552 flp->flowi4_oif != nhc->nhc_oif)
1556 if (!(fib_flags & FIB_LOOKUP_NOREF))
1557 refcount_inc(&fi->fib_clntref);
1559 res->prefix = htonl(n->key);
1560 res->prefixlen = KEYLENGTH - fa->fa_slen;
1561 res->nh_sel = nhsel;
1563 res->type = fa->fa_type;
1564 res->scope = fi->fib_scope;
1567 res->fa_head = &n->leaf;
1568 #ifdef CONFIG_IP_FIB_TRIE_STATS
1569 this_cpu_inc(stats->semantic_match_passed);
1571 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1576 #ifdef CONFIG_IP_FIB_TRIE_STATS
1577 this_cpu_inc(stats->semantic_match_miss);
1581 EXPORT_SYMBOL_GPL(fib_table_lookup);
1583 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1584 struct key_vector *l, struct fib_alias *old)
1586 /* record the location of the previous list_info entry */
1587 struct hlist_node **pprev = old->fa_list.pprev;
1588 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1590 /* remove the fib_alias from the list */
1591 hlist_del_rcu(&old->fa_list);
1593 /* if we emptied the list this leaf will be freed and we can sort
1594 * out parent suffix lengths as a part of trie_rebalance
1596 if (hlist_empty(&l->leaf)) {
1597 if (tp->slen == l->slen)
1598 node_pull_suffix(tp, tp->pos);
1599 put_child_root(tp, l->key, NULL);
1601 trie_rebalance(t, tp);
1605 /* only access fa if it is pointing at the last valid hlist_node */
1609 /* update the trie with the latest suffix length */
1610 l->slen = fa->fa_slen;
1611 node_pull_suffix(tp, fa->fa_slen);
1614 static void fib_notify_alias_delete(struct net *net, u32 key,
1615 struct hlist_head *fah,
1616 struct fib_alias *fa_to_delete,
1617 struct netlink_ext_ack *extack)
1619 struct fib_alias *fa_next, *fa_to_notify;
1620 u32 tb_id = fa_to_delete->tb_id;
1621 u8 slen = fa_to_delete->fa_slen;
1622 enum fib_event_type fib_event;
1624 /* Do not notify if we do not care about the route. */
1625 if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1628 /* Determine if the route should be replaced by the next route in the
1631 fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1632 struct fib_alias, fa_list);
1633 if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1634 fib_event = FIB_EVENT_ENTRY_REPLACE;
1635 fa_to_notify = fa_next;
1637 fib_event = FIB_EVENT_ENTRY_DEL;
1638 fa_to_notify = fa_to_delete;
1640 call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1641 fa_to_notify, extack);
1644 /* Caller must hold RTNL. */
1645 int fib_table_delete(struct net *net, struct fib_table *tb,
1646 struct fib_config *cfg, struct netlink_ext_ack *extack)
1648 struct trie *t = (struct trie *) tb->tb_data;
1649 struct fib_alias *fa, *fa_to_delete;
1650 struct key_vector *l, *tp;
1651 u8 plen = cfg->fc_dst_len;
1652 u8 slen = KEYLENGTH - plen;
1653 u8 tos = cfg->fc_tos;
1656 key = ntohl(cfg->fc_dst);
1658 if (!fib_valid_key_len(key, plen, extack))
1661 l = fib_find_node(t, &tp, key);
1665 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id, false);
1669 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1671 fa_to_delete = NULL;
1672 hlist_for_each_entry_from(fa, fa_list) {
1673 struct fib_info *fi = fa->fa_info;
1675 if ((fa->fa_slen != slen) ||
1676 (fa->tb_id != tb->tb_id) ||
1677 (fa->fa_tos != tos))
1680 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1681 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1682 fa->fa_info->fib_scope == cfg->fc_scope) &&
1683 (!cfg->fc_prefsrc ||
1684 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1685 (!cfg->fc_protocol ||
1686 fi->fib_protocol == cfg->fc_protocol) &&
1687 fib_nh_match(cfg, fi, extack) == 0 &&
1688 fib_metrics_match(cfg, fi)) {
1697 fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1698 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1699 &cfg->fc_nlinfo, 0);
1702 tb->tb_num_default--;
1704 fib_remove_alias(t, tp, l, fa_to_delete);
1706 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1707 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1709 fib_release_info(fa_to_delete->fa_info);
1710 alias_free_mem_rcu(fa_to_delete);
1714 /* Scan for the next leaf starting at the provided key value */
1715 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1717 struct key_vector *pn, *n = *tn;
1718 unsigned long cindex;
1720 /* this loop is meant to try and find the key in the trie */
1722 /* record parent and next child index */
1724 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1726 if (cindex >> pn->bits)
1729 /* descend into the next child */
1730 n = get_child_rcu(pn, cindex++);
1734 /* guarantee forward progress on the keys */
1735 if (IS_LEAF(n) && (n->key >= key))
1737 } while (IS_TNODE(n));
1739 /* this loop will search for the next leaf with a greater key */
1740 while (!IS_TRIE(pn)) {
1741 /* if we exhausted the parent node we will need to climb */
1742 if (cindex >= (1ul << pn->bits)) {
1743 t_key pkey = pn->key;
1745 pn = node_parent_rcu(pn);
1746 cindex = get_index(pkey, pn) + 1;
1750 /* grab the next available node */
1751 n = get_child_rcu(pn, cindex++);
1755 /* no need to compare keys since we bumped the index */
1759 /* Rescan start scanning in new node */
1765 return NULL; /* Root of trie */
1767 /* if we are at the limit for keys just return NULL for the tnode */
1772 static void fib_trie_free(struct fib_table *tb)
1774 struct trie *t = (struct trie *)tb->tb_data;
1775 struct key_vector *pn = t->kv;
1776 unsigned long cindex = 1;
1777 struct hlist_node *tmp;
1778 struct fib_alias *fa;
1780 /* walk trie in reverse order and free everything */
1782 struct key_vector *n;
1785 t_key pkey = pn->key;
1791 pn = node_parent(pn);
1793 /* drop emptied tnode */
1794 put_child_root(pn, n->key, NULL);
1797 cindex = get_index(pkey, pn);
1802 /* grab the next available node */
1803 n = get_child(pn, cindex);
1808 /* record pn and cindex for leaf walking */
1810 cindex = 1ul << n->bits;
1815 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1816 hlist_del_rcu(&fa->fa_list);
1817 alias_free_mem_rcu(fa);
1820 put_child_root(pn, n->key, NULL);
1824 #ifdef CONFIG_IP_FIB_TRIE_STATS
1825 free_percpu(t->stats);
1830 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1832 struct trie *ot = (struct trie *)oldtb->tb_data;
1833 struct key_vector *l, *tp = ot->kv;
1834 struct fib_table *local_tb;
1835 struct fib_alias *fa;
1839 if (oldtb->tb_data == oldtb->__data)
1842 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1846 lt = (struct trie *)local_tb->tb_data;
1848 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1849 struct key_vector *local_l = NULL, *local_tp;
1851 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1852 struct fib_alias *new_fa;
1854 if (local_tb->tb_id != fa->tb_id)
1857 /* clone fa for new local table */
1858 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1862 memcpy(new_fa, fa, sizeof(*fa));
1864 /* insert clone into table */
1866 local_l = fib_find_node(lt, &local_tp, l->key);
1868 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1870 kmem_cache_free(fn_alias_kmem, new_fa);
1875 /* stop loop if key wrapped back to 0 */
1883 fib_trie_free(local_tb);
1888 /* Caller must hold RTNL */
1889 void fib_table_flush_external(struct fib_table *tb)
1891 struct trie *t = (struct trie *)tb->tb_data;
1892 struct key_vector *pn = t->kv;
1893 unsigned long cindex = 1;
1894 struct hlist_node *tmp;
1895 struct fib_alias *fa;
1897 /* walk trie in reverse order */
1899 unsigned char slen = 0;
1900 struct key_vector *n;
1903 t_key pkey = pn->key;
1905 /* cannot resize the trie vector */
1909 /* update the suffix to address pulled leaves */
1910 if (pn->slen > pn->pos)
1913 /* resize completed node */
1915 cindex = get_index(pkey, pn);
1920 /* grab the next available node */
1921 n = get_child(pn, cindex);
1926 /* record pn and cindex for leaf walking */
1928 cindex = 1ul << n->bits;
1933 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1934 /* if alias was cloned to local then we just
1935 * need to remove the local copy from main
1937 if (tb->tb_id != fa->tb_id) {
1938 hlist_del_rcu(&fa->fa_list);
1939 alias_free_mem_rcu(fa);
1943 /* record local slen */
1947 /* update leaf slen */
1950 if (hlist_empty(&n->leaf)) {
1951 put_child_root(pn, n->key, NULL);
1957 /* Caller must hold RTNL. */
1958 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1960 struct trie *t = (struct trie *)tb->tb_data;
1961 struct key_vector *pn = t->kv;
1962 unsigned long cindex = 1;
1963 struct hlist_node *tmp;
1964 struct fib_alias *fa;
1967 /* walk trie in reverse order */
1969 unsigned char slen = 0;
1970 struct key_vector *n;
1973 t_key pkey = pn->key;
1975 /* cannot resize the trie vector */
1979 /* update the suffix to address pulled leaves */
1980 if (pn->slen > pn->pos)
1983 /* resize completed node */
1985 cindex = get_index(pkey, pn);
1990 /* grab the next available node */
1991 n = get_child(pn, cindex);
1996 /* record pn and cindex for leaf walking */
1998 cindex = 1ul << n->bits;
2003 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2004 struct fib_info *fi = fa->fa_info;
2006 if (!fi || tb->tb_id != fa->tb_id ||
2007 (!(fi->fib_flags & RTNH_F_DEAD) &&
2008 !fib_props[fa->fa_type].error)) {
2013 /* Do not flush error routes if network namespace is
2014 * not being dismantled
2016 if (!flush_all && fib_props[fa->fa_type].error) {
2021 fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2023 hlist_del_rcu(&fa->fa_list);
2024 fib_release_info(fa->fa_info);
2025 alias_free_mem_rcu(fa);
2029 /* update leaf slen */
2032 if (hlist_empty(&n->leaf)) {
2033 put_child_root(pn, n->key, NULL);
2038 pr_debug("trie_flush found=%d\n", found);
2042 /* derived from fib_trie_free */
2043 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2044 struct nl_info *info)
2046 struct trie *t = (struct trie *)tb->tb_data;
2047 struct key_vector *pn = t->kv;
2048 unsigned long cindex = 1;
2049 struct fib_alias *fa;
2052 struct key_vector *n;
2055 t_key pkey = pn->key;
2060 pn = node_parent(pn);
2061 cindex = get_index(pkey, pn);
2065 /* grab the next available node */
2066 n = get_child(pn, cindex);
2071 /* record pn and cindex for leaf walking */
2073 cindex = 1ul << n->bits;
2078 hlist_for_each_entry(fa, &n->leaf, fa_list) {
2079 struct fib_info *fi = fa->fa_info;
2081 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2084 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2085 KEYLENGTH - fa->fa_slen, tb->tb_id,
2086 info, NLM_F_REPLACE);
2088 /* call_fib_entry_notifiers will be removed when
2089 * in-kernel notifier is implemented and supported
2090 * for nexthop objects
2092 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
2094 KEYLENGTH - fa->fa_slen, fa,
2100 void fib_info_notify_update(struct net *net, struct nl_info *info)
2104 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2105 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2106 struct fib_table *tb;
2108 hlist_for_each_entry_rcu(tb, head, tb_hlist)
2109 __fib_info_notify_update(net, tb, info);
2113 static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2114 struct notifier_block *nb,
2115 struct netlink_ext_ack *extack)
2117 struct fib_alias *fa;
2121 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2122 struct fib_info *fi = fa->fa_info;
2127 /* local and main table can share the same trie,
2128 * so don't notify twice for the same entry.
2130 if (tb->tb_id != fa->tb_id)
2133 if (fa->fa_slen == last_slen)
2136 last_slen = fa->fa_slen;
2137 err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2138 l->key, KEYLENGTH - fa->fa_slen,
2146 static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2147 struct netlink_ext_ack *extack)
2149 struct trie *t = (struct trie *)tb->tb_data;
2150 struct key_vector *l, *tp = t->kv;
2154 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2155 err = fib_leaf_notify(l, tb, nb, extack);
2160 /* stop in case of wrap around */
2167 int fib_notify(struct net *net, struct notifier_block *nb,
2168 struct netlink_ext_ack *extack)
2173 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2174 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2175 struct fib_table *tb;
2177 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2178 err = fib_table_notify(tb, nb, extack);
2186 static void __trie_free_rcu(struct rcu_head *head)
2188 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2189 #ifdef CONFIG_IP_FIB_TRIE_STATS
2190 struct trie *t = (struct trie *)tb->tb_data;
2192 if (tb->tb_data == tb->__data)
2193 free_percpu(t->stats);
2194 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2198 void fib_free_table(struct fib_table *tb)
2200 call_rcu(&tb->rcu, __trie_free_rcu);
2203 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2204 struct sk_buff *skb, struct netlink_callback *cb,
2205 struct fib_dump_filter *filter)
2207 unsigned int flags = NLM_F_MULTI;
2208 __be32 xkey = htonl(l->key);
2209 int i, s_i, i_fa, s_fa, err;
2210 struct fib_alias *fa;
2212 if (filter->filter_set ||
2213 !filter->dump_exceptions || !filter->dump_routes)
2214 flags |= NLM_F_DUMP_FILTERED;
2220 /* rcu_read_lock is hold by caller */
2221 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2222 struct fib_info *fi = fa->fa_info;
2229 if (tb->tb_id != fa->tb_id)
2232 if (filter->filter_set) {
2233 if (filter->rt_type && fa->fa_type != filter->rt_type)
2236 if ((filter->protocol &&
2237 fi->fib_protocol != filter->protocol))
2241 !fib_info_nh_uses_dev(fi, filter->dev))
2245 if (filter->dump_routes) {
2247 struct fib_rt_info fri;
2250 fri.tb_id = tb->tb_id;
2252 fri.dst_len = KEYLENGTH - fa->fa_slen;
2253 fri.tos = fa->fa_tos;
2254 fri.type = fa->fa_type;
2255 fri.offload = fa->offload;
2256 fri.trap = fa->trap;
2257 err = fib_dump_info(skb,
2258 NETLINK_CB(cb->skb).portid,
2260 RTM_NEWROUTE, &fri, flags);
2268 if (filter->dump_exceptions) {
2269 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2270 &i_fa, s_fa, flags);
2288 /* rcu_read_lock needs to be hold by caller from readside */
2289 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2290 struct netlink_callback *cb, struct fib_dump_filter *filter)
2292 struct trie *t = (struct trie *)tb->tb_data;
2293 struct key_vector *l, *tp = t->kv;
2294 /* Dump starting at last key.
2295 * Note: 0.0.0.0/0 (ie default) is first key.
2297 int count = cb->args[2];
2298 t_key key = cb->args[3];
2300 /* First time here, count and key are both always 0. Count > 0
2301 * and key == 0 means the dump has wrapped around and we are done.
2306 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2309 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2312 cb->args[2] = count;
2319 memset(&cb->args[4], 0,
2320 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2322 /* stop loop if key wrapped back to 0 */
2328 cb->args[2] = count;
2333 void __init fib_trie_init(void)
2335 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2336 sizeof(struct fib_alias),
2337 0, SLAB_PANIC, NULL);
2339 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2341 0, SLAB_PANIC, NULL);
2344 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2346 struct fib_table *tb;
2348 size_t sz = sizeof(*tb);
2351 sz += sizeof(struct trie);
2353 tb = kzalloc(sz, GFP_KERNEL);
2358 tb->tb_num_default = 0;
2359 tb->tb_data = (alias ? alias->__data : tb->__data);
2364 t = (struct trie *) tb->tb_data;
2365 t->kv[0].pos = KEYLENGTH;
2366 t->kv[0].slen = KEYLENGTH;
2367 #ifdef CONFIG_IP_FIB_TRIE_STATS
2368 t->stats = alloc_percpu(struct trie_use_stats);
2378 #ifdef CONFIG_PROC_FS
2379 /* Depth first Trie walk iterator */
2380 struct fib_trie_iter {
2381 struct seq_net_private p;
2382 struct fib_table *tb;
2383 struct key_vector *tnode;
2388 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2390 unsigned long cindex = iter->index;
2391 struct key_vector *pn = iter->tnode;
2394 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2395 iter->tnode, iter->index, iter->depth);
2397 while (!IS_TRIE(pn)) {
2398 while (cindex < child_length(pn)) {
2399 struct key_vector *n = get_child_rcu(pn, cindex++);
2406 iter->index = cindex;
2408 /* push down one level */
2417 /* Current node exhausted, pop back up */
2419 pn = node_parent_rcu(pn);
2420 cindex = get_index(pkey, pn) + 1;
2424 /* record root node so further searches know we are done */
2431 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2434 struct key_vector *n, *pn;
2440 n = rcu_dereference(pn->tnode[0]);
2457 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2459 struct key_vector *n;
2460 struct fib_trie_iter iter;
2462 memset(s, 0, sizeof(*s));
2465 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2467 struct fib_alias *fa;
2470 s->totdepth += iter.depth;
2471 if (iter.depth > s->maxdepth)
2472 s->maxdepth = iter.depth;
2474 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2478 if (n->bits < MAX_STAT_DEPTH)
2479 s->nodesizes[n->bits]++;
2480 s->nullpointers += tn_info(n)->empty_children;
2487 * This outputs /proc/net/fib_triestats
2489 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2491 unsigned int i, max, pointers, bytes, avdepth;
2494 avdepth = stat->totdepth*100 / stat->leaves;
2498 seq_printf(seq, "\tAver depth: %u.%02d\n",
2499 avdepth / 100, avdepth % 100);
2500 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2502 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2503 bytes = LEAF_SIZE * stat->leaves;
2505 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2506 bytes += sizeof(struct fib_alias) * stat->prefixes;
2508 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2509 bytes += TNODE_SIZE(0) * stat->tnodes;
2511 max = MAX_STAT_DEPTH;
2512 while (max > 0 && stat->nodesizes[max-1] == 0)
2516 for (i = 1; i < max; i++)
2517 if (stat->nodesizes[i] != 0) {
2518 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2519 pointers += (1<<i) * stat->nodesizes[i];
2521 seq_putc(seq, '\n');
2522 seq_printf(seq, "\tPointers: %u\n", pointers);
2524 bytes += sizeof(struct key_vector *) * pointers;
2525 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2526 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2529 #ifdef CONFIG_IP_FIB_TRIE_STATS
2530 static void trie_show_usage(struct seq_file *seq,
2531 const struct trie_use_stats __percpu *stats)
2533 struct trie_use_stats s = { 0 };
2536 /* loop through all of the CPUs and gather up the stats */
2537 for_each_possible_cpu(cpu) {
2538 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2540 s.gets += pcpu->gets;
2541 s.backtrack += pcpu->backtrack;
2542 s.semantic_match_passed += pcpu->semantic_match_passed;
2543 s.semantic_match_miss += pcpu->semantic_match_miss;
2544 s.null_node_hit += pcpu->null_node_hit;
2545 s.resize_node_skipped += pcpu->resize_node_skipped;
2548 seq_printf(seq, "\nCounters:\n---------\n");
2549 seq_printf(seq, "gets = %u\n", s.gets);
2550 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2551 seq_printf(seq, "semantic match passed = %u\n",
2552 s.semantic_match_passed);
2553 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2554 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2555 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2557 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2559 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2561 if (tb->tb_id == RT_TABLE_LOCAL)
2562 seq_puts(seq, "Local:\n");
2563 else if (tb->tb_id == RT_TABLE_MAIN)
2564 seq_puts(seq, "Main:\n");
2566 seq_printf(seq, "Id %d:\n", tb->tb_id);
2570 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2572 struct net *net = (struct net *)seq->private;
2576 "Basic info: size of leaf:"
2577 " %zd bytes, size of tnode: %zd bytes.\n",
2578 LEAF_SIZE, TNODE_SIZE(0));
2580 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2581 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2582 struct fib_table *tb;
2584 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2585 struct trie *t = (struct trie *) tb->tb_data;
2586 struct trie_stat stat;
2591 fib_table_print(seq, tb);
2593 trie_collect_stats(t, &stat);
2594 trie_show_stats(seq, &stat);
2595 #ifdef CONFIG_IP_FIB_TRIE_STATS
2596 trie_show_usage(seq, t->stats);
2604 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2606 struct fib_trie_iter *iter = seq->private;
2607 struct net *net = seq_file_net(seq);
2611 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2612 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2613 struct fib_table *tb;
2615 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2616 struct key_vector *n;
2618 for (n = fib_trie_get_first(iter,
2619 (struct trie *) tb->tb_data);
2620 n; n = fib_trie_get_next(iter))
2631 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2635 return fib_trie_get_idx(seq, *pos);
2638 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2640 struct fib_trie_iter *iter = seq->private;
2641 struct net *net = seq_file_net(seq);
2642 struct fib_table *tb = iter->tb;
2643 struct hlist_node *tb_node;
2645 struct key_vector *n;
2648 /* next node in same table */
2649 n = fib_trie_get_next(iter);
2653 /* walk rest of this hash chain */
2654 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2655 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2656 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2657 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2662 /* new hash chain */
2663 while (++h < FIB_TABLE_HASHSZ) {
2664 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2665 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2666 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2678 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2684 static void seq_indent(struct seq_file *seq, int n)
2690 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2693 case RT_SCOPE_UNIVERSE: return "universe";
2694 case RT_SCOPE_SITE: return "site";
2695 case RT_SCOPE_LINK: return "link";
2696 case RT_SCOPE_HOST: return "host";
2697 case RT_SCOPE_NOWHERE: return "nowhere";
2699 snprintf(buf, len, "scope=%d", s);
2704 static const char *const rtn_type_names[__RTN_MAX] = {
2705 [RTN_UNSPEC] = "UNSPEC",
2706 [RTN_UNICAST] = "UNICAST",
2707 [RTN_LOCAL] = "LOCAL",
2708 [RTN_BROADCAST] = "BROADCAST",
2709 [RTN_ANYCAST] = "ANYCAST",
2710 [RTN_MULTICAST] = "MULTICAST",
2711 [RTN_BLACKHOLE] = "BLACKHOLE",
2712 [RTN_UNREACHABLE] = "UNREACHABLE",
2713 [RTN_PROHIBIT] = "PROHIBIT",
2714 [RTN_THROW] = "THROW",
2716 [RTN_XRESOLVE] = "XRESOLVE",
2719 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2721 if (t < __RTN_MAX && rtn_type_names[t])
2722 return rtn_type_names[t];
2723 snprintf(buf, len, "type %u", t);
2727 /* Pretty print the trie */
2728 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2730 const struct fib_trie_iter *iter = seq->private;
2731 struct key_vector *n = v;
2733 if (IS_TRIE(node_parent_rcu(n)))
2734 fib_table_print(seq, iter->tb);
2737 __be32 prf = htonl(n->key);
2739 seq_indent(seq, iter->depth-1);
2740 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2741 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2742 tn_info(n)->full_children,
2743 tn_info(n)->empty_children);
2745 __be32 val = htonl(n->key);
2746 struct fib_alias *fa;
2748 seq_indent(seq, iter->depth);
2749 seq_printf(seq, " |-- %pI4\n", &val);
2751 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2752 char buf1[32], buf2[32];
2754 seq_indent(seq, iter->depth + 1);
2755 seq_printf(seq, " /%zu %s %s",
2756 KEYLENGTH - fa->fa_slen,
2757 rtn_scope(buf1, sizeof(buf1),
2758 fa->fa_info->fib_scope),
2759 rtn_type(buf2, sizeof(buf2),
2762 seq_printf(seq, " tos=%d", fa->fa_tos);
2763 seq_putc(seq, '\n');
2770 static const struct seq_operations fib_trie_seq_ops = {
2771 .start = fib_trie_seq_start,
2772 .next = fib_trie_seq_next,
2773 .stop = fib_trie_seq_stop,
2774 .show = fib_trie_seq_show,
2777 struct fib_route_iter {
2778 struct seq_net_private p;
2779 struct fib_table *main_tb;
2780 struct key_vector *tnode;
2785 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2788 struct key_vector *l, **tp = &iter->tnode;
2791 /* use cached location of previously found key */
2792 if (iter->pos > 0 && pos >= iter->pos) {
2801 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2806 /* handle unlikely case of a key wrap */
2812 iter->key = l->key; /* remember it */
2814 iter->pos = 0; /* forget it */
2819 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2822 struct fib_route_iter *iter = seq->private;
2823 struct fib_table *tb;
2828 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2833 t = (struct trie *)tb->tb_data;
2834 iter->tnode = t->kv;
2837 return fib_route_get_idx(iter, *pos);
2840 iter->key = KEY_MAX;
2842 return SEQ_START_TOKEN;
2845 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2847 struct fib_route_iter *iter = seq->private;
2848 struct key_vector *l = NULL;
2849 t_key key = iter->key + 1;
2853 /* only allow key of 0 for start of sequence */
2854 if ((v == SEQ_START_TOKEN) || key)
2855 l = leaf_walk_rcu(&iter->tnode, key);
2867 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2873 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2875 unsigned int flags = 0;
2877 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2880 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2882 if (nhc->nhc_gw.ipv4)
2883 flags |= RTF_GATEWAY;
2885 if (mask == htonl(0xFFFFFFFF))
2892 * This outputs /proc/net/route.
2893 * The format of the file is not supposed to be changed
2894 * and needs to be same as fib_hash output to avoid breaking
2897 static int fib_route_seq_show(struct seq_file *seq, void *v)
2899 struct fib_route_iter *iter = seq->private;
2900 struct fib_table *tb = iter->main_tb;
2901 struct fib_alias *fa;
2902 struct key_vector *l = v;
2905 if (v == SEQ_START_TOKEN) {
2906 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2907 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2912 prefix = htonl(l->key);
2914 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2915 struct fib_info *fi = fa->fa_info;
2916 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2917 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2919 if ((fa->fa_type == RTN_BROADCAST) ||
2920 (fa->fa_type == RTN_MULTICAST))
2923 if (fa->tb_id != tb->tb_id)
2926 seq_setwidth(seq, 127);
2929 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2932 if (nhc->nhc_gw_family == AF_INET)
2933 gw = nhc->nhc_gw.ipv4;
2936 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2937 "%d\t%08X\t%d\t%u\t%u",
2938 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2939 prefix, gw, flags, 0, 0,
2943 fi->fib_advmss + 40 : 0),
2948 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2949 "%d\t%08X\t%d\t%u\t%u",
2950 prefix, 0, flags, 0, 0, 0,
2959 static const struct seq_operations fib_route_seq_ops = {
2960 .start = fib_route_seq_start,
2961 .next = fib_route_seq_next,
2962 .stop = fib_route_seq_stop,
2963 .show = fib_route_seq_show,
2966 int __net_init fib_proc_init(struct net *net)
2968 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2969 sizeof(struct fib_trie_iter)))
2972 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
2973 fib_triestat_seq_show, NULL))
2976 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
2977 sizeof(struct fib_route_iter)))
2983 remove_proc_entry("fib_triestat", net->proc_net);
2985 remove_proc_entry("fib_trie", net->proc_net);
2990 void __net_exit fib_proc_exit(struct net *net)
2992 remove_proc_entry("fib_trie", net->proc_net);
2993 remove_proc_entry("fib_triestat", net->proc_net);
2994 remove_proc_entry("route", net->proc_net);
2997 #endif /* CONFIG_PROC_FS */