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 * https://www.csc.kth.se/~snilsson/software/dyntrie2/
18 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
19 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
21 * Code from fib_hash has been reused which includes the following header:
23 * INET An implementation of the TCP/IP protocol suite for the LINUX
24 * operating system. INET is implemented using the BSD Socket
25 * interface as the means of communication with the user level.
27 * IPv4 FIB: lookup engine and maintenance routines.
29 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
31 * Substantial contributions to this work comes from:
33 * David S. Miller, <davem@davemloft.net>
34 * Stephen Hemminger <shemminger@osdl.org>
35 * Paul E. McKenney <paulmck@us.ibm.com>
36 * Patrick McHardy <kaber@trash.net>
38 #include <linux/cache.h>
39 #include <linux/uaccess.h>
40 #include <linux/bitops.h>
41 #include <linux/types.h>
42 #include <linux/kernel.h>
44 #include <linux/string.h>
45 #include <linux/socket.h>
46 #include <linux/sockios.h>
47 #include <linux/errno.h>
49 #include <linux/inet.h>
50 #include <linux/inetdevice.h>
51 #include <linux/netdevice.h>
52 #include <linux/if_arp.h>
53 #include <linux/proc_fs.h>
54 #include <linux/rcupdate.h>
55 #include <linux/skbuff.h>
56 #include <linux/netlink.h>
57 #include <linux/init.h>
58 #include <linux/list.h>
59 #include <linux/slab.h>
60 #include <linux/export.h>
61 #include <linux/vmalloc.h>
62 #include <linux/notifier.h>
63 #include <net/net_namespace.h>
65 #include <net/protocol.h>
66 #include <net/route.h>
69 #include <net/ip_fib.h>
70 #include <net/fib_notifier.h>
71 #include <trace/events/fib.h>
72 #include "fib_lookup.h"
74 static int call_fib_entry_notifier(struct notifier_block *nb,
75 enum fib_event_type event_type, u32 dst,
76 int dst_len, struct fib_alias *fa,
77 struct netlink_ext_ack *extack)
79 struct fib_entry_notifier_info info = {
80 .info.extack = extack,
88 return call_fib4_notifier(nb, event_type, &info.info);
91 static int call_fib_entry_notifiers(struct net *net,
92 enum fib_event_type event_type, u32 dst,
93 int dst_len, struct fib_alias *fa,
94 struct netlink_ext_ack *extack)
96 struct fib_entry_notifier_info info = {
97 .info.extack = extack,
105 return call_fib4_notifiers(net, event_type, &info.info);
108 #define MAX_STAT_DEPTH 32
110 #define KEYLENGTH (8*sizeof(t_key))
111 #define KEY_MAX ((t_key)~0)
113 typedef unsigned int t_key;
115 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
116 #define IS_TNODE(n) ((n)->bits)
117 #define IS_LEAF(n) (!(n)->bits)
121 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
122 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
125 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
126 struct hlist_head leaf;
127 /* This array is valid if (pos | bits) > 0 (TNODE) */
128 struct key_vector __rcu *tnode[0];
134 t_key empty_children; /* KEYLENGTH bits needed */
135 t_key full_children; /* KEYLENGTH bits needed */
136 struct key_vector __rcu *parent;
137 struct key_vector kv[1];
138 #define tn_bits kv[0].bits
141 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
142 #define LEAF_SIZE TNODE_SIZE(1)
144 #ifdef CONFIG_IP_FIB_TRIE_STATS
145 struct trie_use_stats {
147 unsigned int backtrack;
148 unsigned int semantic_match_passed;
149 unsigned int semantic_match_miss;
150 unsigned int null_node_hit;
151 unsigned int resize_node_skipped;
156 unsigned int totdepth;
157 unsigned int maxdepth;
160 unsigned int nullpointers;
161 unsigned int prefixes;
162 unsigned int nodesizes[MAX_STAT_DEPTH];
166 struct key_vector kv[1];
167 #ifdef CONFIG_IP_FIB_TRIE_STATS
168 struct trie_use_stats __percpu *stats;
172 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
173 static unsigned int tnode_free_size;
176 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
177 * especially useful before resizing the root node with PREEMPT_NONE configs;
178 * the value was obtained experimentally, aiming to avoid visible slowdown.
180 unsigned int sysctl_fib_sync_mem = 512 * 1024;
181 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
182 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
184 static struct kmem_cache *fn_alias_kmem __ro_after_init;
185 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
187 static inline struct tnode *tn_info(struct key_vector *kv)
189 return container_of(kv, struct tnode, kv[0]);
192 /* caller must hold RTNL */
193 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
194 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
196 /* caller must hold RCU read lock or RTNL */
197 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
198 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
200 /* wrapper for rcu_assign_pointer */
201 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
204 rcu_assign_pointer(tn_info(n)->parent, tp);
207 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
209 /* This provides us with the number of children in this node, in the case of a
210 * leaf this will return 0 meaning none of the children are accessible.
212 static inline unsigned long child_length(const struct key_vector *tn)
214 return (1ul << tn->bits) & ~(1ul);
217 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
219 static inline unsigned long get_index(t_key key, struct key_vector *kv)
221 unsigned long index = key ^ kv->key;
223 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
226 return index >> kv->pos;
229 /* To understand this stuff, an understanding of keys and all their bits is
230 * necessary. Every node in the trie has a key associated with it, but not
231 * all of the bits in that key are significant.
233 * Consider a node 'n' and its parent 'tp'.
235 * If n is a leaf, every bit in its key is significant. Its presence is
236 * necessitated by path compression, since during a tree traversal (when
237 * searching for a leaf - unless we are doing an insertion) we will completely
238 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
239 * a potentially successful search, that we have indeed been walking the
242 * Note that we can never "miss" the correct key in the tree if present by
243 * following the wrong path. Path compression ensures that segments of the key
244 * that are the same for all keys with a given prefix are skipped, but the
245 * skipped part *is* identical for each node in the subtrie below the skipped
246 * bit! trie_insert() in this implementation takes care of that.
248 * if n is an internal node - a 'tnode' here, the various parts of its key
249 * have many different meanings.
252 * _________________________________________________________________
253 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
254 * -----------------------------------------------------------------
255 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
257 * _________________________________________________________________
258 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
259 * -----------------------------------------------------------------
260 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
267 * First, let's just ignore the bits that come before the parent tp, that is
268 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
269 * point we do not use them for anything.
271 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
272 * index into the parent's child array. That is, they will be used to find
273 * 'n' among tp's children.
275 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
278 * All the bits we have seen so far are significant to the node n. The rest
279 * of the bits are really not needed or indeed known in n->key.
281 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
282 * n's child array, and will of course be different for each child.
284 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
288 static const int halve_threshold = 25;
289 static const int inflate_threshold = 50;
290 static const int halve_threshold_root = 15;
291 static const int inflate_threshold_root = 30;
293 static void __alias_free_mem(struct rcu_head *head)
295 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
296 kmem_cache_free(fn_alias_kmem, fa);
299 static inline void alias_free_mem_rcu(struct fib_alias *fa)
301 call_rcu(&fa->rcu, __alias_free_mem);
304 #define TNODE_VMALLOC_MAX \
305 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
307 static void __node_free_rcu(struct rcu_head *head)
309 struct tnode *n = container_of(head, struct tnode, rcu);
312 kmem_cache_free(trie_leaf_kmem, n);
317 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
319 static struct tnode *tnode_alloc(int bits)
323 /* verify bits is within bounds */
324 if (bits > TNODE_VMALLOC_MAX)
327 /* determine size and verify it is non-zero and didn't overflow */
328 size = TNODE_SIZE(1ul << bits);
330 if (size <= PAGE_SIZE)
331 return kzalloc(size, GFP_KERNEL);
333 return vzalloc(size);
336 static inline void empty_child_inc(struct key_vector *n)
338 tn_info(n)->empty_children++;
340 if (!tn_info(n)->empty_children)
341 tn_info(n)->full_children++;
344 static inline void empty_child_dec(struct key_vector *n)
346 if (!tn_info(n)->empty_children)
347 tn_info(n)->full_children--;
349 tn_info(n)->empty_children--;
352 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
354 struct key_vector *l;
357 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
361 /* initialize key vector */
366 l->slen = fa->fa_slen;
368 /* link leaf to fib alias */
369 INIT_HLIST_HEAD(&l->leaf);
370 hlist_add_head(&fa->fa_list, &l->leaf);
375 static struct key_vector *tnode_new(t_key key, int pos, int bits)
377 unsigned int shift = pos + bits;
378 struct key_vector *tn;
381 /* verify bits and pos their msb bits clear and values are valid */
382 BUG_ON(!bits || (shift > KEYLENGTH));
384 tnode = tnode_alloc(bits);
388 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
389 sizeof(struct key_vector *) << bits);
391 if (bits == KEYLENGTH)
392 tnode->full_children = 1;
394 tnode->empty_children = 1ul << bits;
397 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
405 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
406 * and no bits are skipped. See discussion in dyntree paper p. 6
408 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
410 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
413 /* Add a child at position i overwriting the old value.
414 * Update the value of full_children and empty_children.
416 static void put_child(struct key_vector *tn, unsigned long i,
417 struct key_vector *n)
419 struct key_vector *chi = get_child(tn, i);
422 BUG_ON(i >= child_length(tn));
424 /* update emptyChildren, overflow into fullChildren */
430 /* update fullChildren */
431 wasfull = tnode_full(tn, chi);
432 isfull = tnode_full(tn, n);
434 if (wasfull && !isfull)
435 tn_info(tn)->full_children--;
436 else if (!wasfull && isfull)
437 tn_info(tn)->full_children++;
439 if (n && (tn->slen < n->slen))
442 rcu_assign_pointer(tn->tnode[i], n);
445 static void update_children(struct key_vector *tn)
449 /* update all of the child parent pointers */
450 for (i = child_length(tn); i;) {
451 struct key_vector *inode = get_child(tn, --i);
456 /* Either update the children of a tnode that
457 * already belongs to us or update the child
458 * to point to ourselves.
460 if (node_parent(inode) == tn)
461 update_children(inode);
463 node_set_parent(inode, tn);
467 static inline void put_child_root(struct key_vector *tp, t_key key,
468 struct key_vector *n)
471 rcu_assign_pointer(tp->tnode[0], n);
473 put_child(tp, get_index(key, tp), n);
476 static inline void tnode_free_init(struct key_vector *tn)
478 tn_info(tn)->rcu.next = NULL;
481 static inline void tnode_free_append(struct key_vector *tn,
482 struct key_vector *n)
484 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
485 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
488 static void tnode_free(struct key_vector *tn)
490 struct callback_head *head = &tn_info(tn)->rcu;
494 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
497 tn = container_of(head, struct tnode, rcu)->kv;
500 if (tnode_free_size >= sysctl_fib_sync_mem) {
506 static struct key_vector *replace(struct trie *t,
507 struct key_vector *oldtnode,
508 struct key_vector *tn)
510 struct key_vector *tp = node_parent(oldtnode);
513 /* setup the parent pointer out of and back into this node */
514 NODE_INIT_PARENT(tn, tp);
515 put_child_root(tp, tn->key, tn);
517 /* update all of the child parent pointers */
520 /* all pointers should be clean so we are done */
521 tnode_free(oldtnode);
523 /* resize children now that oldtnode is freed */
524 for (i = child_length(tn); i;) {
525 struct key_vector *inode = get_child(tn, --i);
527 /* resize child node */
528 if (tnode_full(tn, inode))
529 tn = resize(t, inode);
535 static struct key_vector *inflate(struct trie *t,
536 struct key_vector *oldtnode)
538 struct key_vector *tn;
542 pr_debug("In inflate\n");
544 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
548 /* prepare oldtnode to be freed */
549 tnode_free_init(oldtnode);
551 /* Assemble all of the pointers in our cluster, in this case that
552 * represents all of the pointers out of our allocated nodes that
553 * point to existing tnodes and the links between our allocated
556 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
557 struct key_vector *inode = get_child(oldtnode, --i);
558 struct key_vector *node0, *node1;
565 /* A leaf or an internal node with skipped bits */
566 if (!tnode_full(oldtnode, inode)) {
567 put_child(tn, get_index(inode->key, tn), inode);
571 /* drop the node in the old tnode free list */
572 tnode_free_append(oldtnode, inode);
574 /* An internal node with two children */
575 if (inode->bits == 1) {
576 put_child(tn, 2 * i + 1, get_child(inode, 1));
577 put_child(tn, 2 * i, get_child(inode, 0));
581 /* We will replace this node 'inode' with two new
582 * ones, 'node0' and 'node1', each with half of the
583 * original children. The two new nodes will have
584 * a position one bit further down the key and this
585 * means that the "significant" part of their keys
586 * (see the discussion near the top of this file)
587 * will differ by one bit, which will be "0" in
588 * node0's key and "1" in node1's key. Since we are
589 * moving the key position by one step, the bit that
590 * we are moving away from - the bit at position
591 * (tn->pos) - is the one that will differ between
592 * node0 and node1. So... we synthesize that bit in the
595 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
598 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
600 tnode_free_append(tn, node1);
603 tnode_free_append(tn, node0);
605 /* populate child pointers in new nodes */
606 for (k = child_length(inode), j = k / 2; j;) {
607 put_child(node1, --j, get_child(inode, --k));
608 put_child(node0, j, get_child(inode, j));
609 put_child(node1, --j, get_child(inode, --k));
610 put_child(node0, j, get_child(inode, j));
613 /* link new nodes to parent */
614 NODE_INIT_PARENT(node1, tn);
615 NODE_INIT_PARENT(node0, tn);
617 /* link parent to nodes */
618 put_child(tn, 2 * i + 1, node1);
619 put_child(tn, 2 * i, node0);
622 /* setup the parent pointers into and out of this node */
623 return replace(t, oldtnode, tn);
625 /* all pointers should be clean so we are done */
631 static struct key_vector *halve(struct trie *t,
632 struct key_vector *oldtnode)
634 struct key_vector *tn;
637 pr_debug("In halve\n");
639 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
643 /* prepare oldtnode to be freed */
644 tnode_free_init(oldtnode);
646 /* Assemble all of the pointers in our cluster, in this case that
647 * represents all of the pointers out of our allocated nodes that
648 * point to existing tnodes and the links between our allocated
651 for (i = child_length(oldtnode); i;) {
652 struct key_vector *node1 = get_child(oldtnode, --i);
653 struct key_vector *node0 = get_child(oldtnode, --i);
654 struct key_vector *inode;
656 /* At least one of the children is empty */
657 if (!node1 || !node0) {
658 put_child(tn, i / 2, node1 ? : node0);
662 /* Two nonempty children */
663 inode = tnode_new(node0->key, oldtnode->pos, 1);
666 tnode_free_append(tn, inode);
668 /* initialize pointers out of node */
669 put_child(inode, 1, node1);
670 put_child(inode, 0, node0);
671 NODE_INIT_PARENT(inode, tn);
673 /* link parent to node */
674 put_child(tn, i / 2, inode);
677 /* setup the parent pointers into and out of this node */
678 return replace(t, oldtnode, tn);
680 /* all pointers should be clean so we are done */
686 static struct key_vector *collapse(struct trie *t,
687 struct key_vector *oldtnode)
689 struct key_vector *n, *tp;
692 /* scan the tnode looking for that one child that might still exist */
693 for (n = NULL, i = child_length(oldtnode); !n && i;)
694 n = get_child(oldtnode, --i);
696 /* compress one level */
697 tp = node_parent(oldtnode);
698 put_child_root(tp, oldtnode->key, n);
699 node_set_parent(n, tp);
707 static unsigned char update_suffix(struct key_vector *tn)
709 unsigned char slen = tn->pos;
710 unsigned long stride, i;
711 unsigned char slen_max;
713 /* only vector 0 can have a suffix length greater than or equal to
714 * tn->pos + tn->bits, the second highest node will have a suffix
715 * length at most of tn->pos + tn->bits - 1
717 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
719 /* search though the list of children looking for nodes that might
720 * have a suffix greater than the one we currently have. This is
721 * why we start with a stride of 2 since a stride of 1 would
722 * represent the nodes with suffix length equal to tn->pos
724 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
725 struct key_vector *n = get_child(tn, i);
727 if (!n || (n->slen <= slen))
730 /* update stride and slen based on new value */
731 stride <<= (n->slen - slen);
735 /* stop searching if we have hit the maximum possible value */
736 if (slen >= slen_max)
745 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
746 * the Helsinki University of Technology and Matti Tikkanen of Nokia
747 * Telecommunications, page 6:
748 * "A node is doubled if the ratio of non-empty children to all
749 * children in the *doubled* node is at least 'high'."
751 * 'high' in this instance is the variable 'inflate_threshold'. It
752 * is expressed as a percentage, so we multiply it with
753 * child_length() and instead of multiplying by 2 (since the
754 * child array will be doubled by inflate()) and multiplying
755 * the left-hand side by 100 (to handle the percentage thing) we
756 * multiply the left-hand side by 50.
758 * The left-hand side may look a bit weird: child_length(tn)
759 * - tn->empty_children is of course the number of non-null children
760 * in the current node. tn->full_children is the number of "full"
761 * children, that is non-null tnodes with a skip value of 0.
762 * All of those will be doubled in the resulting inflated tnode, so
763 * we just count them one extra time here.
765 * A clearer way to write this would be:
767 * to_be_doubled = tn->full_children;
768 * not_to_be_doubled = child_length(tn) - tn->empty_children -
771 * new_child_length = child_length(tn) * 2;
773 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
775 * if (new_fill_factor >= inflate_threshold)
777 * ...and so on, tho it would mess up the while () loop.
780 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
784 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
785 * inflate_threshold * new_child_length
787 * expand not_to_be_doubled and to_be_doubled, and shorten:
788 * 100 * (child_length(tn) - tn->empty_children +
789 * tn->full_children) >= inflate_threshold * new_child_length
791 * expand new_child_length:
792 * 100 * (child_length(tn) - tn->empty_children +
793 * tn->full_children) >=
794 * inflate_threshold * child_length(tn) * 2
797 * 50 * (tn->full_children + child_length(tn) -
798 * tn->empty_children) >= inflate_threshold *
802 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
804 unsigned long used = child_length(tn);
805 unsigned long threshold = used;
807 /* Keep root node larger */
808 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
809 used -= tn_info(tn)->empty_children;
810 used += tn_info(tn)->full_children;
812 /* if bits == KEYLENGTH then pos = 0, and will fail below */
814 return (used > 1) && tn->pos && ((50 * used) >= threshold);
817 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
819 unsigned long used = child_length(tn);
820 unsigned long threshold = used;
822 /* Keep root node larger */
823 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
824 used -= tn_info(tn)->empty_children;
826 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
828 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
831 static inline bool should_collapse(struct key_vector *tn)
833 unsigned long used = child_length(tn);
835 used -= tn_info(tn)->empty_children;
837 /* account for bits == KEYLENGTH case */
838 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
841 /* One child or none, time to drop us from the trie */
846 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
848 #ifdef CONFIG_IP_FIB_TRIE_STATS
849 struct trie_use_stats __percpu *stats = t->stats;
851 struct key_vector *tp = node_parent(tn);
852 unsigned long cindex = get_index(tn->key, tp);
853 int max_work = MAX_WORK;
855 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
856 tn, inflate_threshold, halve_threshold);
858 /* track the tnode via the pointer from the parent instead of
859 * doing it ourselves. This way we can let RCU fully do its
860 * thing without us interfering
862 BUG_ON(tn != get_child(tp, cindex));
864 /* Double as long as the resulting node has a number of
865 * nonempty nodes that are above the threshold.
867 while (should_inflate(tp, tn) && max_work) {
870 #ifdef CONFIG_IP_FIB_TRIE_STATS
871 this_cpu_inc(stats->resize_node_skipped);
877 tn = get_child(tp, cindex);
880 /* update parent in case inflate failed */
881 tp = node_parent(tn);
883 /* Return if at least one inflate is run */
884 if (max_work != MAX_WORK)
887 /* Halve as long as the number of empty children in this
888 * node is above threshold.
890 while (should_halve(tp, tn) && max_work) {
893 #ifdef CONFIG_IP_FIB_TRIE_STATS
894 this_cpu_inc(stats->resize_node_skipped);
900 tn = get_child(tp, cindex);
903 /* Only one child remains */
904 if (should_collapse(tn))
905 return collapse(t, tn);
907 /* update parent in case halve failed */
908 return node_parent(tn);
911 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
913 unsigned char node_slen = tn->slen;
915 while ((node_slen > tn->pos) && (node_slen > slen)) {
916 slen = update_suffix(tn);
917 if (node_slen == slen)
920 tn = node_parent(tn);
921 node_slen = tn->slen;
925 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
927 while (tn->slen < slen) {
929 tn = node_parent(tn);
933 /* rcu_read_lock needs to be hold by caller from readside */
934 static struct key_vector *fib_find_node(struct trie *t,
935 struct key_vector **tp, u32 key)
937 struct key_vector *pn, *n = t->kv;
938 unsigned long index = 0;
942 n = get_child_rcu(n, index);
947 index = get_cindex(key, n);
949 /* This bit of code is a bit tricky but it combines multiple
950 * checks into a single check. The prefix consists of the
951 * prefix plus zeros for the bits in the cindex. The index
952 * is the difference between the key and this value. From
953 * this we can actually derive several pieces of data.
954 * if (index >= (1ul << bits))
955 * we have a mismatch in skip bits and failed
957 * we know the value is cindex
959 * This check is safe even if bits == KEYLENGTH due to the
960 * fact that we can only allocate a node with 32 bits if a
961 * long is greater than 32 bits.
963 if (index >= (1ul << n->bits)) {
968 /* keep searching until we find a perfect match leaf or NULL */
969 } while (IS_TNODE(n));
976 /* Return the first fib alias matching TOS with
977 * priority less than or equal to PRIO.
978 * If 'find_first' is set, return the first matching
979 * fib alias, regardless of TOS and priority.
981 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
982 u8 tos, u32 prio, u32 tb_id,
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)
1001 if (fa->fa_tos > tos)
1003 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1010 static struct fib_alias *
1011 fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1013 u8 slen = KEYLENGTH - fri->dst_len;
1014 struct key_vector *l, *tp;
1015 struct fib_table *tb;
1016 struct fib_alias *fa;
1019 tb = fib_get_table(net, fri->tb_id);
1023 t = (struct trie *)tb->tb_data;
1024 l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1028 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1029 if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1030 fa->fa_tos == fri->tos && fa->fa_info == fri->fi &&
1031 fa->fa_type == fri->type)
1038 void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1040 struct fib_alias *fa_match;
1041 struct sk_buff *skb;
1046 fa_match = fib_find_matching_alias(net, fri);
1050 /* These are paired with the WRITE_ONCE() happening in this function.
1051 * The reason is that we are only protected by RCU at this point.
1053 if (READ_ONCE(fa_match->offload) == fri->offload &&
1054 READ_ONCE(fa_match->trap) == fri->trap &&
1055 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1058 WRITE_ONCE(fa_match->offload, fri->offload);
1059 WRITE_ONCE(fa_match->trap, fri->trap);
1061 /* 2 means send notifications only if offload_failed was changed. */
1062 if (net->ipv4.sysctl_fib_notify_on_flag_change == 2 &&
1063 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1066 WRITE_ONCE(fa_match->offload_failed, fri->offload_failed);
1068 if (!net->ipv4.sysctl_fib_notify_on_flag_change)
1071 skb = nlmsg_new(fib_nlmsg_size(fa_match->fa_info), GFP_ATOMIC);
1077 err = fib_dump_info(skb, 0, 0, RTM_NEWROUTE, fri, 0);
1079 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
1080 WARN_ON(err == -EMSGSIZE);
1085 rtnl_notify(skb, net, 0, RTNLGRP_IPV4_ROUTE, NULL, GFP_ATOMIC);
1089 rtnl_set_sk_err(net, RTNLGRP_IPV4_ROUTE, err);
1093 EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1095 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1097 while (!IS_TRIE(tn))
1101 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1102 struct fib_alias *new, t_key key)
1104 struct key_vector *n, *l;
1106 l = leaf_new(key, new);
1110 /* retrieve child from parent node */
1111 n = get_child(tp, get_index(key, tp));
1113 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1115 * Add a new tnode here
1116 * first tnode need some special handling
1117 * leaves us in position for handling as case 3
1120 struct key_vector *tn;
1122 tn = tnode_new(key, __fls(key ^ n->key), 1);
1126 /* initialize routes out of node */
1127 NODE_INIT_PARENT(tn, tp);
1128 put_child(tn, get_index(key, tn) ^ 1, n);
1130 /* start adding routes into the node */
1131 put_child_root(tp, key, tn);
1132 node_set_parent(n, tn);
1134 /* parent now has a NULL spot where the leaf can go */
1138 /* Case 3: n is NULL, and will just insert a new leaf */
1139 node_push_suffix(tp, new->fa_slen);
1140 NODE_INIT_PARENT(l, tp);
1141 put_child_root(tp, key, l);
1142 trie_rebalance(t, tp);
1151 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1152 struct key_vector *l, struct fib_alias *new,
1153 struct fib_alias *fa, t_key key)
1156 return fib_insert_node(t, tp, new, key);
1159 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1161 struct fib_alias *last;
1163 hlist_for_each_entry(last, &l->leaf, fa_list) {
1164 if (new->fa_slen < last->fa_slen)
1166 if ((new->fa_slen == last->fa_slen) &&
1167 (new->tb_id > last->tb_id))
1173 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1175 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1178 /* if we added to the tail node then we need to update slen */
1179 if (l->slen < new->fa_slen) {
1180 l->slen = new->fa_slen;
1181 node_push_suffix(tp, new->fa_slen);
1187 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1189 if (plen > KEYLENGTH) {
1190 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1194 if ((plen < KEYLENGTH) && (key << plen)) {
1195 NL_SET_ERR_MSG(extack,
1196 "Invalid prefix for given prefix length");
1203 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1204 struct key_vector *l, struct fib_alias *old);
1206 /* Caller must hold RTNL. */
1207 int fib_table_insert(struct net *net, struct fib_table *tb,
1208 struct fib_config *cfg, struct netlink_ext_ack *extack)
1210 struct trie *t = (struct trie *)tb->tb_data;
1211 struct fib_alias *fa, *new_fa;
1212 struct key_vector *l, *tp;
1213 u16 nlflags = NLM_F_EXCL;
1214 struct fib_info *fi;
1215 u8 plen = cfg->fc_dst_len;
1216 u8 slen = KEYLENGTH - plen;
1217 u8 tos = cfg->fc_tos;
1221 key = ntohl(cfg->fc_dst);
1223 if (!fib_valid_key_len(key, plen, extack))
1226 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1228 fi = fib_create_info(cfg, extack);
1234 l = fib_find_node(t, &tp, key);
1235 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1236 tb->tb_id, false) : NULL;
1238 /* Now fa, if non-NULL, points to the first fib alias
1239 * with the same keys [prefix,tos,priority], if such key already
1240 * exists or to the node before which we will insert new one.
1242 * If fa is NULL, we will need to allocate a new one and
1243 * insert to the tail of the section matching the suffix length
1247 if (fa && fa->fa_tos == tos &&
1248 fa->fa_info->fib_priority == fi->fib_priority) {
1249 struct fib_alias *fa_first, *fa_match;
1252 if (cfg->fc_nlflags & NLM_F_EXCL)
1255 nlflags &= ~NLM_F_EXCL;
1258 * 1. Find exact match for type, scope, fib_info to avoid
1260 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1264 hlist_for_each_entry_from(fa, fa_list) {
1265 if ((fa->fa_slen != slen) ||
1266 (fa->tb_id != tb->tb_id) ||
1267 (fa->fa_tos != tos))
1269 if (fa->fa_info->fib_priority != fi->fib_priority)
1271 if (fa->fa_type == cfg->fc_type &&
1272 fa->fa_info == fi) {
1278 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1279 struct fib_info *fi_drop;
1282 nlflags |= NLM_F_REPLACE;
1290 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1294 fi_drop = fa->fa_info;
1295 new_fa->fa_tos = fa->fa_tos;
1296 new_fa->fa_info = fi;
1297 new_fa->fa_type = cfg->fc_type;
1298 state = fa->fa_state;
1299 new_fa->fa_state = state & ~FA_S_ACCESSED;
1300 new_fa->fa_slen = fa->fa_slen;
1301 new_fa->tb_id = tb->tb_id;
1302 new_fa->fa_default = -1;
1303 new_fa->offload = 0;
1305 new_fa->offload_failed = 0;
1307 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1309 if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1310 tb->tb_id, true) == new_fa) {
1311 enum fib_event_type fib_event;
1313 fib_event = FIB_EVENT_ENTRY_REPLACE;
1314 err = call_fib_entry_notifiers(net, fib_event,
1318 hlist_replace_rcu(&new_fa->fa_list,
1320 goto out_free_new_fa;
1324 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1325 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1327 alias_free_mem_rcu(fa);
1329 fib_release_info(fi_drop);
1330 if (state & FA_S_ACCESSED)
1331 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1335 /* Error if we find a perfect match which
1336 * uses the same scope, type, and nexthop
1342 if (cfg->fc_nlflags & NLM_F_APPEND)
1343 nlflags |= NLM_F_APPEND;
1348 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1351 nlflags |= NLM_F_CREATE;
1353 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1357 new_fa->fa_info = fi;
1358 new_fa->fa_tos = tos;
1359 new_fa->fa_type = cfg->fc_type;
1360 new_fa->fa_state = 0;
1361 new_fa->fa_slen = slen;
1362 new_fa->tb_id = tb->tb_id;
1363 new_fa->fa_default = -1;
1364 new_fa->offload = 0;
1366 new_fa->offload_failed = 0;
1368 /* Insert new entry to the list. */
1369 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1371 goto out_free_new_fa;
1373 /* The alias was already inserted, so the node must exist. */
1374 l = l ? l : fib_find_node(t, &tp, key);
1375 if (WARN_ON_ONCE(!l))
1376 goto out_free_new_fa;
1378 if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1380 enum fib_event_type fib_event;
1382 fib_event = FIB_EVENT_ENTRY_REPLACE;
1383 err = call_fib_entry_notifiers(net, fib_event, key, plen,
1386 goto out_remove_new_fa;
1390 tb->tb_num_default++;
1392 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1393 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1394 &cfg->fc_nlinfo, nlflags);
1399 fib_remove_alias(t, tp, l, new_fa);
1401 kmem_cache_free(fn_alias_kmem, new_fa);
1403 fib_release_info(fi);
1408 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1410 t_key prefix = n->key;
1412 return (key ^ prefix) & (prefix | -prefix);
1415 bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1416 const struct flowi4 *flp)
1418 if (nhc->nhc_flags & RTNH_F_DEAD)
1421 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1422 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1423 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1426 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1427 if (flp->flowi4_oif &&
1428 flp->flowi4_oif != nhc->nhc_oif)
1435 /* should be called with rcu_read_lock */
1436 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1437 struct fib_result *res, int fib_flags)
1439 struct trie *t = (struct trie *) tb->tb_data;
1440 #ifdef CONFIG_IP_FIB_TRIE_STATS
1441 struct trie_use_stats __percpu *stats = t->stats;
1443 const t_key key = ntohl(flp->daddr);
1444 struct key_vector *n, *pn;
1445 struct fib_alias *fa;
1446 unsigned long index;
1452 n = get_child_rcu(pn, cindex);
1454 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1458 #ifdef CONFIG_IP_FIB_TRIE_STATS
1459 this_cpu_inc(stats->gets);
1462 /* Step 1: Travel to the longest prefix match in the trie */
1464 index = get_cindex(key, n);
1466 /* This bit of code is a bit tricky but it combines multiple
1467 * checks into a single check. The prefix consists of the
1468 * prefix plus zeros for the "bits" in the prefix. The index
1469 * is the difference between the key and this value. From
1470 * this we can actually derive several pieces of data.
1471 * if (index >= (1ul << bits))
1472 * we have a mismatch in skip bits and failed
1474 * we know the value is cindex
1476 * This check is safe even if bits == KEYLENGTH due to the
1477 * fact that we can only allocate a node with 32 bits if a
1478 * long is greater than 32 bits.
1480 if (index >= (1ul << n->bits))
1483 /* we have found a leaf. Prefixes have already been compared */
1487 /* only record pn and cindex if we are going to be chopping
1488 * bits later. Otherwise we are just wasting cycles.
1490 if (n->slen > n->pos) {
1495 n = get_child_rcu(n, index);
1500 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1502 /* record the pointer where our next node pointer is stored */
1503 struct key_vector __rcu **cptr = n->tnode;
1505 /* This test verifies that none of the bits that differ
1506 * between the key and the prefix exist in the region of
1507 * the lsb and higher in the prefix.
1509 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1512 /* exit out and process leaf */
1513 if (unlikely(IS_LEAF(n)))
1516 /* Don't bother recording parent info. Since we are in
1517 * prefix match mode we will have to come back to wherever
1518 * we started this traversal anyway
1521 while ((n = rcu_dereference(*cptr)) == NULL) {
1523 #ifdef CONFIG_IP_FIB_TRIE_STATS
1525 this_cpu_inc(stats->null_node_hit);
1527 /* If we are at cindex 0 there are no more bits for
1528 * us to strip at this level so we must ascend back
1529 * up one level to see if there are any more bits to
1530 * be stripped there.
1533 t_key pkey = pn->key;
1535 /* If we don't have a parent then there is
1536 * nothing for us to do as we do not have any
1537 * further nodes to parse.
1540 trace_fib_table_lookup(tb->tb_id, flp,
1544 #ifdef CONFIG_IP_FIB_TRIE_STATS
1545 this_cpu_inc(stats->backtrack);
1547 /* Get Child's index */
1548 pn = node_parent_rcu(pn);
1549 cindex = get_index(pkey, pn);
1552 /* strip the least significant bit from the cindex */
1553 cindex &= cindex - 1;
1555 /* grab pointer for next child node */
1556 cptr = &pn->tnode[cindex];
1561 /* this line carries forward the xor from earlier in the function */
1562 index = key ^ n->key;
1564 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1565 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1566 struct fib_info *fi = fa->fa_info;
1567 struct fib_nh_common *nhc;
1570 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1571 if (index >= (1ul << fa->fa_slen))
1574 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1578 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1580 fib_alias_accessed(fa);
1581 err = fib_props[fa->fa_type].error;
1582 if (unlikely(err < 0)) {
1584 #ifdef CONFIG_IP_FIB_TRIE_STATS
1585 this_cpu_inc(stats->semantic_match_passed);
1587 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1590 if (fi->fib_flags & RTNH_F_DEAD)
1593 if (unlikely(fi->nh)) {
1594 if (nexthop_is_blackhole(fi->nh)) {
1595 err = fib_props[RTN_BLACKHOLE].error;
1599 nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1606 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1607 nhc = fib_info_nhc(fi, nhsel);
1609 if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1612 if (!(fib_flags & FIB_LOOKUP_NOREF))
1613 refcount_inc(&fi->fib_clntref);
1615 res->prefix = htonl(n->key);
1616 res->prefixlen = KEYLENGTH - fa->fa_slen;
1617 res->nh_sel = nhsel;
1619 res->type = fa->fa_type;
1620 res->scope = fi->fib_scope;
1623 res->fa_head = &n->leaf;
1624 #ifdef CONFIG_IP_FIB_TRIE_STATS
1625 this_cpu_inc(stats->semantic_match_passed);
1627 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1633 #ifdef CONFIG_IP_FIB_TRIE_STATS
1634 this_cpu_inc(stats->semantic_match_miss);
1638 EXPORT_SYMBOL_GPL(fib_table_lookup);
1640 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1641 struct key_vector *l, struct fib_alias *old)
1643 /* record the location of the previous list_info entry */
1644 struct hlist_node **pprev = old->fa_list.pprev;
1645 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1647 /* remove the fib_alias from the list */
1648 hlist_del_rcu(&old->fa_list);
1650 /* if we emptied the list this leaf will be freed and we can sort
1651 * out parent suffix lengths as a part of trie_rebalance
1653 if (hlist_empty(&l->leaf)) {
1654 if (tp->slen == l->slen)
1655 node_pull_suffix(tp, tp->pos);
1656 put_child_root(tp, l->key, NULL);
1658 trie_rebalance(t, tp);
1662 /* only access fa if it is pointing at the last valid hlist_node */
1666 /* update the trie with the latest suffix length */
1667 l->slen = fa->fa_slen;
1668 node_pull_suffix(tp, fa->fa_slen);
1671 static void fib_notify_alias_delete(struct net *net, u32 key,
1672 struct hlist_head *fah,
1673 struct fib_alias *fa_to_delete,
1674 struct netlink_ext_ack *extack)
1676 struct fib_alias *fa_next, *fa_to_notify;
1677 u32 tb_id = fa_to_delete->tb_id;
1678 u8 slen = fa_to_delete->fa_slen;
1679 enum fib_event_type fib_event;
1681 /* Do not notify if we do not care about the route. */
1682 if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1685 /* Determine if the route should be replaced by the next route in the
1688 fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1689 struct fib_alias, fa_list);
1690 if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1691 fib_event = FIB_EVENT_ENTRY_REPLACE;
1692 fa_to_notify = fa_next;
1694 fib_event = FIB_EVENT_ENTRY_DEL;
1695 fa_to_notify = fa_to_delete;
1697 call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1698 fa_to_notify, extack);
1701 /* Caller must hold RTNL. */
1702 int fib_table_delete(struct net *net, struct fib_table *tb,
1703 struct fib_config *cfg, struct netlink_ext_ack *extack)
1705 struct trie *t = (struct trie *) tb->tb_data;
1706 struct fib_alias *fa, *fa_to_delete;
1707 struct key_vector *l, *tp;
1708 u8 plen = cfg->fc_dst_len;
1709 u8 slen = KEYLENGTH - plen;
1710 u8 tos = cfg->fc_tos;
1713 key = ntohl(cfg->fc_dst);
1715 if (!fib_valid_key_len(key, plen, extack))
1718 l = fib_find_node(t, &tp, key);
1722 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id, false);
1726 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1728 fa_to_delete = NULL;
1729 hlist_for_each_entry_from(fa, fa_list) {
1730 struct fib_info *fi = fa->fa_info;
1732 if ((fa->fa_slen != slen) ||
1733 (fa->tb_id != tb->tb_id) ||
1734 (fa->fa_tos != tos))
1737 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1738 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1739 fa->fa_info->fib_scope == cfg->fc_scope) &&
1740 (!cfg->fc_prefsrc ||
1741 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1742 (!cfg->fc_protocol ||
1743 fi->fib_protocol == cfg->fc_protocol) &&
1744 fib_nh_match(net, cfg, fi, extack) == 0 &&
1745 fib_metrics_match(cfg, fi)) {
1754 fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1755 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1756 &cfg->fc_nlinfo, 0);
1759 tb->tb_num_default--;
1761 fib_remove_alias(t, tp, l, fa_to_delete);
1763 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1764 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1766 fib_release_info(fa_to_delete->fa_info);
1767 alias_free_mem_rcu(fa_to_delete);
1771 /* Scan for the next leaf starting at the provided key value */
1772 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1774 struct key_vector *pn, *n = *tn;
1775 unsigned long cindex;
1777 /* this loop is meant to try and find the key in the trie */
1779 /* record parent and next child index */
1781 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1783 if (cindex >> pn->bits)
1786 /* descend into the next child */
1787 n = get_child_rcu(pn, cindex++);
1791 /* guarantee forward progress on the keys */
1792 if (IS_LEAF(n) && (n->key >= key))
1794 } while (IS_TNODE(n));
1796 /* this loop will search for the next leaf with a greater key */
1797 while (!IS_TRIE(pn)) {
1798 /* if we exhausted the parent node we will need to climb */
1799 if (cindex >= (1ul << pn->bits)) {
1800 t_key pkey = pn->key;
1802 pn = node_parent_rcu(pn);
1803 cindex = get_index(pkey, pn) + 1;
1807 /* grab the next available node */
1808 n = get_child_rcu(pn, cindex++);
1812 /* no need to compare keys since we bumped the index */
1816 /* Rescan start scanning in new node */
1822 return NULL; /* Root of trie */
1824 /* if we are at the limit for keys just return NULL for the tnode */
1829 static void fib_trie_free(struct fib_table *tb)
1831 struct trie *t = (struct trie *)tb->tb_data;
1832 struct key_vector *pn = t->kv;
1833 unsigned long cindex = 1;
1834 struct hlist_node *tmp;
1835 struct fib_alias *fa;
1837 /* walk trie in reverse order and free everything */
1839 struct key_vector *n;
1842 t_key pkey = pn->key;
1848 pn = node_parent(pn);
1850 /* drop emptied tnode */
1851 put_child_root(pn, n->key, NULL);
1854 cindex = get_index(pkey, pn);
1859 /* grab the next available node */
1860 n = get_child(pn, cindex);
1865 /* record pn and cindex for leaf walking */
1867 cindex = 1ul << n->bits;
1872 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1873 hlist_del_rcu(&fa->fa_list);
1874 alias_free_mem_rcu(fa);
1877 put_child_root(pn, n->key, NULL);
1881 #ifdef CONFIG_IP_FIB_TRIE_STATS
1882 free_percpu(t->stats);
1887 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1889 struct trie *ot = (struct trie *)oldtb->tb_data;
1890 struct key_vector *l, *tp = ot->kv;
1891 struct fib_table *local_tb;
1892 struct fib_alias *fa;
1896 if (oldtb->tb_data == oldtb->__data)
1899 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1903 lt = (struct trie *)local_tb->tb_data;
1905 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1906 struct key_vector *local_l = NULL, *local_tp;
1908 hlist_for_each_entry(fa, &l->leaf, fa_list) {
1909 struct fib_alias *new_fa;
1911 if (local_tb->tb_id != fa->tb_id)
1914 /* clone fa for new local table */
1915 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1919 memcpy(new_fa, fa, sizeof(*fa));
1921 /* insert clone into table */
1923 local_l = fib_find_node(lt, &local_tp, l->key);
1925 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1927 kmem_cache_free(fn_alias_kmem, new_fa);
1932 /* stop loop if key wrapped back to 0 */
1940 fib_trie_free(local_tb);
1945 /* Caller must hold RTNL */
1946 void fib_table_flush_external(struct fib_table *tb)
1948 struct trie *t = (struct trie *)tb->tb_data;
1949 struct key_vector *pn = t->kv;
1950 unsigned long cindex = 1;
1951 struct hlist_node *tmp;
1952 struct fib_alias *fa;
1954 /* walk trie in reverse order */
1956 unsigned char slen = 0;
1957 struct key_vector *n;
1960 t_key pkey = pn->key;
1962 /* cannot resize the trie vector */
1966 /* update the suffix to address pulled leaves */
1967 if (pn->slen > pn->pos)
1970 /* resize completed node */
1972 cindex = get_index(pkey, pn);
1977 /* grab the next available node */
1978 n = get_child(pn, cindex);
1983 /* record pn and cindex for leaf walking */
1985 cindex = 1ul << n->bits;
1990 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1991 /* if alias was cloned to local then we just
1992 * need to remove the local copy from main
1994 if (tb->tb_id != fa->tb_id) {
1995 hlist_del_rcu(&fa->fa_list);
1996 alias_free_mem_rcu(fa);
2000 /* record local slen */
2004 /* update leaf slen */
2007 if (hlist_empty(&n->leaf)) {
2008 put_child_root(pn, n->key, NULL);
2014 /* Caller must hold RTNL. */
2015 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
2017 struct trie *t = (struct trie *)tb->tb_data;
2018 struct key_vector *pn = t->kv;
2019 unsigned long cindex = 1;
2020 struct hlist_node *tmp;
2021 struct fib_alias *fa;
2024 /* walk trie in reverse order */
2026 unsigned char slen = 0;
2027 struct key_vector *n;
2030 t_key pkey = pn->key;
2032 /* cannot resize the trie vector */
2036 /* update the suffix to address pulled leaves */
2037 if (pn->slen > pn->pos)
2040 /* resize completed node */
2042 cindex = get_index(pkey, pn);
2047 /* grab the next available node */
2048 n = get_child(pn, cindex);
2053 /* record pn and cindex for leaf walking */
2055 cindex = 1ul << n->bits;
2060 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2061 struct fib_info *fi = fa->fa_info;
2063 if (!fi || tb->tb_id != fa->tb_id ||
2064 (!(fi->fib_flags & RTNH_F_DEAD) &&
2065 !fib_props[fa->fa_type].error)) {
2070 /* Do not flush error routes if network namespace is
2071 * not being dismantled
2073 if (!flush_all && fib_props[fa->fa_type].error) {
2078 fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2080 hlist_del_rcu(&fa->fa_list);
2081 fib_release_info(fa->fa_info);
2082 alias_free_mem_rcu(fa);
2086 /* update leaf slen */
2089 if (hlist_empty(&n->leaf)) {
2090 put_child_root(pn, n->key, NULL);
2095 pr_debug("trie_flush found=%d\n", found);
2099 /* derived from fib_trie_free */
2100 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2101 struct nl_info *info)
2103 struct trie *t = (struct trie *)tb->tb_data;
2104 struct key_vector *pn = t->kv;
2105 unsigned long cindex = 1;
2106 struct fib_alias *fa;
2109 struct key_vector *n;
2112 t_key pkey = pn->key;
2117 pn = node_parent(pn);
2118 cindex = get_index(pkey, pn);
2122 /* grab the next available node */
2123 n = get_child(pn, cindex);
2128 /* record pn and cindex for leaf walking */
2130 cindex = 1ul << n->bits;
2135 hlist_for_each_entry(fa, &n->leaf, fa_list) {
2136 struct fib_info *fi = fa->fa_info;
2138 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2141 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2142 KEYLENGTH - fa->fa_slen, tb->tb_id,
2143 info, NLM_F_REPLACE);
2148 void fib_info_notify_update(struct net *net, struct nl_info *info)
2152 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2153 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2154 struct fib_table *tb;
2156 hlist_for_each_entry_rcu(tb, head, tb_hlist,
2157 lockdep_rtnl_is_held())
2158 __fib_info_notify_update(net, tb, info);
2162 static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2163 struct notifier_block *nb,
2164 struct netlink_ext_ack *extack)
2166 struct fib_alias *fa;
2170 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2171 struct fib_info *fi = fa->fa_info;
2176 /* local and main table can share the same trie,
2177 * so don't notify twice for the same entry.
2179 if (tb->tb_id != fa->tb_id)
2182 if (fa->fa_slen == last_slen)
2185 last_slen = fa->fa_slen;
2186 err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2187 l->key, KEYLENGTH - fa->fa_slen,
2195 static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2196 struct netlink_ext_ack *extack)
2198 struct trie *t = (struct trie *)tb->tb_data;
2199 struct key_vector *l, *tp = t->kv;
2203 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2204 err = fib_leaf_notify(l, tb, nb, extack);
2209 /* stop in case of wrap around */
2216 int fib_notify(struct net *net, struct notifier_block *nb,
2217 struct netlink_ext_ack *extack)
2222 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2223 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2224 struct fib_table *tb;
2226 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2227 err = fib_table_notify(tb, nb, extack);
2235 static void __trie_free_rcu(struct rcu_head *head)
2237 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2238 #ifdef CONFIG_IP_FIB_TRIE_STATS
2239 struct trie *t = (struct trie *)tb->tb_data;
2241 if (tb->tb_data == tb->__data)
2242 free_percpu(t->stats);
2243 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2247 void fib_free_table(struct fib_table *tb)
2249 call_rcu(&tb->rcu, __trie_free_rcu);
2252 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2253 struct sk_buff *skb, struct netlink_callback *cb,
2254 struct fib_dump_filter *filter)
2256 unsigned int flags = NLM_F_MULTI;
2257 __be32 xkey = htonl(l->key);
2258 int i, s_i, i_fa, s_fa, err;
2259 struct fib_alias *fa;
2261 if (filter->filter_set ||
2262 !filter->dump_exceptions || !filter->dump_routes)
2263 flags |= NLM_F_DUMP_FILTERED;
2269 /* rcu_read_lock is hold by caller */
2270 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2271 struct fib_info *fi = fa->fa_info;
2278 if (tb->tb_id != fa->tb_id)
2281 if (filter->filter_set) {
2282 if (filter->rt_type && fa->fa_type != filter->rt_type)
2285 if ((filter->protocol &&
2286 fi->fib_protocol != filter->protocol))
2290 !fib_info_nh_uses_dev(fi, filter->dev))
2294 if (filter->dump_routes) {
2296 struct fib_rt_info fri;
2299 fri.tb_id = tb->tb_id;
2301 fri.dst_len = KEYLENGTH - fa->fa_slen;
2302 fri.tos = fa->fa_tos;
2303 fri.type = fa->fa_type;
2304 fri.offload = READ_ONCE(fa->offload);
2305 fri.trap = READ_ONCE(fa->trap);
2306 fri.offload_failed = READ_ONCE(fa->offload_failed);
2307 err = fib_dump_info(skb,
2308 NETLINK_CB(cb->skb).portid,
2310 RTM_NEWROUTE, &fri, flags);
2318 if (filter->dump_exceptions) {
2319 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2320 &i_fa, s_fa, flags);
2338 /* rcu_read_lock needs to be hold by caller from readside */
2339 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2340 struct netlink_callback *cb, struct fib_dump_filter *filter)
2342 struct trie *t = (struct trie *)tb->tb_data;
2343 struct key_vector *l, *tp = t->kv;
2344 /* Dump starting at last key.
2345 * Note: 0.0.0.0/0 (ie default) is first key.
2347 int count = cb->args[2];
2348 t_key key = cb->args[3];
2350 /* First time here, count and key are both always 0. Count > 0
2351 * and key == 0 means the dump has wrapped around and we are done.
2356 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2359 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2362 cb->args[2] = count;
2369 memset(&cb->args[4], 0,
2370 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2372 /* stop loop if key wrapped back to 0 */
2378 cb->args[2] = count;
2383 void __init fib_trie_init(void)
2385 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2386 sizeof(struct fib_alias),
2387 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2389 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2391 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2394 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2396 struct fib_table *tb;
2398 size_t sz = sizeof(*tb);
2401 sz += sizeof(struct trie);
2403 tb = kzalloc(sz, GFP_KERNEL);
2408 tb->tb_num_default = 0;
2409 tb->tb_data = (alias ? alias->__data : tb->__data);
2414 t = (struct trie *) tb->tb_data;
2415 t->kv[0].pos = KEYLENGTH;
2416 t->kv[0].slen = KEYLENGTH;
2417 #ifdef CONFIG_IP_FIB_TRIE_STATS
2418 t->stats = alloc_percpu(struct trie_use_stats);
2428 #ifdef CONFIG_PROC_FS
2429 /* Depth first Trie walk iterator */
2430 struct fib_trie_iter {
2431 struct seq_net_private p;
2432 struct fib_table *tb;
2433 struct key_vector *tnode;
2438 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2440 unsigned long cindex = iter->index;
2441 struct key_vector *pn = iter->tnode;
2444 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2445 iter->tnode, iter->index, iter->depth);
2447 while (!IS_TRIE(pn)) {
2448 while (cindex < child_length(pn)) {
2449 struct key_vector *n = get_child_rcu(pn, cindex++);
2456 iter->index = cindex;
2458 /* push down one level */
2467 /* Current node exhausted, pop back up */
2469 pn = node_parent_rcu(pn);
2470 cindex = get_index(pkey, pn) + 1;
2474 /* record root node so further searches know we are done */
2481 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2484 struct key_vector *n, *pn;
2490 n = rcu_dereference(pn->tnode[0]);
2507 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2509 struct key_vector *n;
2510 struct fib_trie_iter iter;
2512 memset(s, 0, sizeof(*s));
2515 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2517 struct fib_alias *fa;
2520 s->totdepth += iter.depth;
2521 if (iter.depth > s->maxdepth)
2522 s->maxdepth = iter.depth;
2524 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2528 if (n->bits < MAX_STAT_DEPTH)
2529 s->nodesizes[n->bits]++;
2530 s->nullpointers += tn_info(n)->empty_children;
2537 * This outputs /proc/net/fib_triestats
2539 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2541 unsigned int i, max, pointers, bytes, avdepth;
2544 avdepth = stat->totdepth*100 / stat->leaves;
2548 seq_printf(seq, "\tAver depth: %u.%02d\n",
2549 avdepth / 100, avdepth % 100);
2550 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2552 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2553 bytes = LEAF_SIZE * stat->leaves;
2555 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2556 bytes += sizeof(struct fib_alias) * stat->prefixes;
2558 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2559 bytes += TNODE_SIZE(0) * stat->tnodes;
2561 max = MAX_STAT_DEPTH;
2562 while (max > 0 && stat->nodesizes[max-1] == 0)
2566 for (i = 1; i < max; i++)
2567 if (stat->nodesizes[i] != 0) {
2568 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2569 pointers += (1<<i) * stat->nodesizes[i];
2571 seq_putc(seq, '\n');
2572 seq_printf(seq, "\tPointers: %u\n", pointers);
2574 bytes += sizeof(struct key_vector *) * pointers;
2575 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2576 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2579 #ifdef CONFIG_IP_FIB_TRIE_STATS
2580 static void trie_show_usage(struct seq_file *seq,
2581 const struct trie_use_stats __percpu *stats)
2583 struct trie_use_stats s = { 0 };
2586 /* loop through all of the CPUs and gather up the stats */
2587 for_each_possible_cpu(cpu) {
2588 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2590 s.gets += pcpu->gets;
2591 s.backtrack += pcpu->backtrack;
2592 s.semantic_match_passed += pcpu->semantic_match_passed;
2593 s.semantic_match_miss += pcpu->semantic_match_miss;
2594 s.null_node_hit += pcpu->null_node_hit;
2595 s.resize_node_skipped += pcpu->resize_node_skipped;
2598 seq_printf(seq, "\nCounters:\n---------\n");
2599 seq_printf(seq, "gets = %u\n", s.gets);
2600 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2601 seq_printf(seq, "semantic match passed = %u\n",
2602 s.semantic_match_passed);
2603 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2604 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2605 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2607 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2609 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2611 if (tb->tb_id == RT_TABLE_LOCAL)
2612 seq_puts(seq, "Local:\n");
2613 else if (tb->tb_id == RT_TABLE_MAIN)
2614 seq_puts(seq, "Main:\n");
2616 seq_printf(seq, "Id %d:\n", tb->tb_id);
2620 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2622 struct net *net = (struct net *)seq->private;
2626 "Basic info: size of leaf:"
2627 " %zd bytes, size of tnode: %zd bytes.\n",
2628 LEAF_SIZE, TNODE_SIZE(0));
2631 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2632 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2633 struct fib_table *tb;
2635 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2636 struct trie *t = (struct trie *) tb->tb_data;
2637 struct trie_stat stat;
2642 fib_table_print(seq, tb);
2644 trie_collect_stats(t, &stat);
2645 trie_show_stats(seq, &stat);
2646 #ifdef CONFIG_IP_FIB_TRIE_STATS
2647 trie_show_usage(seq, t->stats);
2657 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2659 struct fib_trie_iter *iter = seq->private;
2660 struct net *net = seq_file_net(seq);
2664 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2665 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2666 struct fib_table *tb;
2668 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2669 struct key_vector *n;
2671 for (n = fib_trie_get_first(iter,
2672 (struct trie *) tb->tb_data);
2673 n; n = fib_trie_get_next(iter))
2684 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2688 return fib_trie_get_idx(seq, *pos);
2691 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2693 struct fib_trie_iter *iter = seq->private;
2694 struct net *net = seq_file_net(seq);
2695 struct fib_table *tb = iter->tb;
2696 struct hlist_node *tb_node;
2698 struct key_vector *n;
2701 /* next node in same table */
2702 n = fib_trie_get_next(iter);
2706 /* walk rest of this hash chain */
2707 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2708 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2709 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2710 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2715 /* new hash chain */
2716 while (++h < FIB_TABLE_HASHSZ) {
2717 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2718 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2719 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2731 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2737 static void seq_indent(struct seq_file *seq, int n)
2743 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2746 case RT_SCOPE_UNIVERSE: return "universe";
2747 case RT_SCOPE_SITE: return "site";
2748 case RT_SCOPE_LINK: return "link";
2749 case RT_SCOPE_HOST: return "host";
2750 case RT_SCOPE_NOWHERE: return "nowhere";
2752 snprintf(buf, len, "scope=%d", s);
2757 static const char *const rtn_type_names[__RTN_MAX] = {
2758 [RTN_UNSPEC] = "UNSPEC",
2759 [RTN_UNICAST] = "UNICAST",
2760 [RTN_LOCAL] = "LOCAL",
2761 [RTN_BROADCAST] = "BROADCAST",
2762 [RTN_ANYCAST] = "ANYCAST",
2763 [RTN_MULTICAST] = "MULTICAST",
2764 [RTN_BLACKHOLE] = "BLACKHOLE",
2765 [RTN_UNREACHABLE] = "UNREACHABLE",
2766 [RTN_PROHIBIT] = "PROHIBIT",
2767 [RTN_THROW] = "THROW",
2769 [RTN_XRESOLVE] = "XRESOLVE",
2772 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2774 if (t < __RTN_MAX && rtn_type_names[t])
2775 return rtn_type_names[t];
2776 snprintf(buf, len, "type %u", t);
2780 /* Pretty print the trie */
2781 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2783 const struct fib_trie_iter *iter = seq->private;
2784 struct key_vector *n = v;
2786 if (IS_TRIE(node_parent_rcu(n)))
2787 fib_table_print(seq, iter->tb);
2790 __be32 prf = htonl(n->key);
2792 seq_indent(seq, iter->depth-1);
2793 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2794 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2795 tn_info(n)->full_children,
2796 tn_info(n)->empty_children);
2798 __be32 val = htonl(n->key);
2799 struct fib_alias *fa;
2801 seq_indent(seq, iter->depth);
2802 seq_printf(seq, " |-- %pI4\n", &val);
2804 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2805 char buf1[32], buf2[32];
2807 seq_indent(seq, iter->depth + 1);
2808 seq_printf(seq, " /%zu %s %s",
2809 KEYLENGTH - fa->fa_slen,
2810 rtn_scope(buf1, sizeof(buf1),
2811 fa->fa_info->fib_scope),
2812 rtn_type(buf2, sizeof(buf2),
2815 seq_printf(seq, " tos=%d", fa->fa_tos);
2816 seq_putc(seq, '\n');
2823 static const struct seq_operations fib_trie_seq_ops = {
2824 .start = fib_trie_seq_start,
2825 .next = fib_trie_seq_next,
2826 .stop = fib_trie_seq_stop,
2827 .show = fib_trie_seq_show,
2830 struct fib_route_iter {
2831 struct seq_net_private p;
2832 struct fib_table *main_tb;
2833 struct key_vector *tnode;
2838 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2841 struct key_vector *l, **tp = &iter->tnode;
2844 /* use cached location of previously found key */
2845 if (iter->pos > 0 && pos >= iter->pos) {
2854 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2859 /* handle unlikely case of a key wrap */
2865 iter->key = l->key; /* remember it */
2867 iter->pos = 0; /* forget it */
2872 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2875 struct fib_route_iter *iter = seq->private;
2876 struct fib_table *tb;
2881 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2886 t = (struct trie *)tb->tb_data;
2887 iter->tnode = t->kv;
2890 return fib_route_get_idx(iter, *pos);
2893 iter->key = KEY_MAX;
2895 return SEQ_START_TOKEN;
2898 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2900 struct fib_route_iter *iter = seq->private;
2901 struct key_vector *l = NULL;
2902 t_key key = iter->key + 1;
2906 /* only allow key of 0 for start of sequence */
2907 if ((v == SEQ_START_TOKEN) || key)
2908 l = leaf_walk_rcu(&iter->tnode, key);
2920 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2926 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2928 unsigned int flags = 0;
2930 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2933 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2935 if (nhc->nhc_gw.ipv4)
2936 flags |= RTF_GATEWAY;
2938 if (mask == htonl(0xFFFFFFFF))
2945 * This outputs /proc/net/route.
2946 * The format of the file is not supposed to be changed
2947 * and needs to be same as fib_hash output to avoid breaking
2950 static int fib_route_seq_show(struct seq_file *seq, void *v)
2952 struct fib_route_iter *iter = seq->private;
2953 struct fib_table *tb = iter->main_tb;
2954 struct fib_alias *fa;
2955 struct key_vector *l = v;
2958 if (v == SEQ_START_TOKEN) {
2959 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2960 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2965 prefix = htonl(l->key);
2967 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2968 struct fib_info *fi = fa->fa_info;
2969 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2970 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2972 if ((fa->fa_type == RTN_BROADCAST) ||
2973 (fa->fa_type == RTN_MULTICAST))
2976 if (fa->tb_id != tb->tb_id)
2979 seq_setwidth(seq, 127);
2982 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2985 if (nhc->nhc_gw_family == AF_INET)
2986 gw = nhc->nhc_gw.ipv4;
2989 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2990 "%d\t%08X\t%d\t%u\t%u",
2991 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2992 prefix, gw, flags, 0, 0,
2996 fi->fib_advmss + 40 : 0),
3001 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
3002 "%d\t%08X\t%d\t%u\t%u",
3003 prefix, 0, flags, 0, 0, 0,
3012 static const struct seq_operations fib_route_seq_ops = {
3013 .start = fib_route_seq_start,
3014 .next = fib_route_seq_next,
3015 .stop = fib_route_seq_stop,
3016 .show = fib_route_seq_show,
3019 int __net_init fib_proc_init(struct net *net)
3021 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
3022 sizeof(struct fib_trie_iter)))
3025 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
3026 fib_triestat_seq_show, NULL))
3029 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
3030 sizeof(struct fib_route_iter)))
3036 remove_proc_entry("fib_triestat", net->proc_net);
3038 remove_proc_entry("fib_trie", net->proc_net);
3043 void __net_exit fib_proc_exit(struct net *net)
3045 remove_proc_entry("fib_trie", net->proc_net);
3046 remove_proc_entry("fib_triestat", net->proc_net);
3047 remove_proc_entry("route", net->proc_net);
3050 #endif /* CONFIG_PROC_FS */