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>
64 #include <net/inet_dscp.h>
66 #include <net/protocol.h>
67 #include <net/route.h>
70 #include <net/ip_fib.h>
71 #include <net/fib_notifier.h>
72 #include <trace/events/fib.h>
73 #include "fib_lookup.h"
75 static int call_fib_entry_notifier(struct notifier_block *nb,
76 enum fib_event_type event_type, u32 dst,
77 int dst_len, struct fib_alias *fa,
78 struct netlink_ext_ack *extack)
80 struct fib_entry_notifier_info info = {
81 .info.extack = extack,
89 return call_fib4_notifier(nb, event_type, &info.info);
92 static int call_fib_entry_notifiers(struct net *net,
93 enum fib_event_type event_type, u32 dst,
94 int dst_len, struct fib_alias *fa,
95 struct netlink_ext_ack *extack)
97 struct fib_entry_notifier_info info = {
98 .info.extack = extack,
106 return call_fib4_notifiers(net, event_type, &info.info);
109 #define MAX_STAT_DEPTH 32
111 #define KEYLENGTH (8*sizeof(t_key))
112 #define KEY_MAX ((t_key)~0)
114 typedef unsigned int t_key;
116 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
117 #define IS_TNODE(n) ((n)->bits)
118 #define IS_LEAF(n) (!(n)->bits)
122 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
123 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
126 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
127 struct hlist_head leaf;
128 /* This array is valid if (pos | bits) > 0 (TNODE) */
129 DECLARE_FLEX_ARRAY(struct key_vector __rcu *, tnode);
135 t_key empty_children; /* KEYLENGTH bits needed */
136 t_key full_children; /* KEYLENGTH bits needed */
137 struct key_vector __rcu *parent;
138 struct key_vector kv[1];
139 #define tn_bits kv[0].bits
142 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
143 #define LEAF_SIZE TNODE_SIZE(1)
145 #ifdef CONFIG_IP_FIB_TRIE_STATS
146 struct trie_use_stats {
148 unsigned int backtrack;
149 unsigned int semantic_match_passed;
150 unsigned int semantic_match_miss;
151 unsigned int null_node_hit;
152 unsigned int resize_node_skipped;
157 unsigned int totdepth;
158 unsigned int maxdepth;
161 unsigned int nullpointers;
162 unsigned int prefixes;
163 unsigned int nodesizes[MAX_STAT_DEPTH];
167 struct key_vector kv[1];
168 #ifdef CONFIG_IP_FIB_TRIE_STATS
169 struct trie_use_stats __percpu *stats;
173 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
174 static unsigned int tnode_free_size;
177 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
178 * especially useful before resizing the root node with PREEMPT_NONE configs;
179 * the value was obtained experimentally, aiming to avoid visible slowdown.
181 unsigned int sysctl_fib_sync_mem = 512 * 1024;
182 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
183 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
185 static struct kmem_cache *fn_alias_kmem __ro_after_init;
186 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
188 static inline struct tnode *tn_info(struct key_vector *kv)
190 return container_of(kv, struct tnode, kv[0]);
193 /* caller must hold RTNL */
194 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
195 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
197 /* caller must hold RCU read lock or RTNL */
198 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
199 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
201 /* wrapper for rcu_assign_pointer */
202 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
205 rcu_assign_pointer(tn_info(n)->parent, tp);
208 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
210 /* This provides us with the number of children in this node, in the case of a
211 * leaf this will return 0 meaning none of the children are accessible.
213 static inline unsigned long child_length(const struct key_vector *tn)
215 return (1ul << tn->bits) & ~(1ul);
218 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
220 static inline unsigned long get_index(t_key key, struct key_vector *kv)
222 unsigned long index = key ^ kv->key;
224 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
227 return index >> kv->pos;
230 /* To understand this stuff, an understanding of keys and all their bits is
231 * necessary. Every node in the trie has a key associated with it, but not
232 * all of the bits in that key are significant.
234 * Consider a node 'n' and its parent 'tp'.
236 * If n is a leaf, every bit in its key is significant. Its presence is
237 * necessitated by path compression, since during a tree traversal (when
238 * searching for a leaf - unless we are doing an insertion) we will completely
239 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
240 * a potentially successful search, that we have indeed been walking the
243 * Note that we can never "miss" the correct key in the tree if present by
244 * following the wrong path. Path compression ensures that segments of the key
245 * that are the same for all keys with a given prefix are skipped, but the
246 * skipped part *is* identical for each node in the subtrie below the skipped
247 * bit! trie_insert() in this implementation takes care of that.
249 * if n is an internal node - a 'tnode' here, the various parts of its key
250 * have many different meanings.
253 * _________________________________________________________________
254 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
255 * -----------------------------------------------------------------
256 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
258 * _________________________________________________________________
259 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
260 * -----------------------------------------------------------------
261 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
268 * First, let's just ignore the bits that come before the parent tp, that is
269 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
270 * point we do not use them for anything.
272 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
273 * index into the parent's child array. That is, they will be used to find
274 * 'n' among tp's children.
276 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
279 * All the bits we have seen so far are significant to the node n. The rest
280 * of the bits are really not needed or indeed known in n->key.
282 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
283 * n's child array, and will of course be different for each child.
285 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
289 static const int halve_threshold = 25;
290 static const int inflate_threshold = 50;
291 static const int halve_threshold_root = 15;
292 static const int inflate_threshold_root = 30;
294 static void __alias_free_mem(struct rcu_head *head)
296 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
297 kmem_cache_free(fn_alias_kmem, fa);
300 static inline void alias_free_mem_rcu(struct fib_alias *fa)
302 call_rcu(&fa->rcu, __alias_free_mem);
305 #define TNODE_VMALLOC_MAX \
306 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
308 static void __node_free_rcu(struct rcu_head *head)
310 struct tnode *n = container_of(head, struct tnode, rcu);
313 kmem_cache_free(trie_leaf_kmem, n);
318 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
320 static struct tnode *tnode_alloc(int bits)
324 /* verify bits is within bounds */
325 if (bits > TNODE_VMALLOC_MAX)
328 /* determine size and verify it is non-zero and didn't overflow */
329 size = TNODE_SIZE(1ul << bits);
331 if (size <= PAGE_SIZE)
332 return kzalloc(size, GFP_KERNEL);
334 return vzalloc(size);
337 static inline void empty_child_inc(struct key_vector *n)
339 tn_info(n)->empty_children++;
341 if (!tn_info(n)->empty_children)
342 tn_info(n)->full_children++;
345 static inline void empty_child_dec(struct key_vector *n)
347 if (!tn_info(n)->empty_children)
348 tn_info(n)->full_children--;
350 tn_info(n)->empty_children--;
353 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
355 struct key_vector *l;
358 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
362 /* initialize key vector */
367 l->slen = fa->fa_slen;
369 /* link leaf to fib alias */
370 INIT_HLIST_HEAD(&l->leaf);
371 hlist_add_head(&fa->fa_list, &l->leaf);
376 static struct key_vector *tnode_new(t_key key, int pos, int bits)
378 unsigned int shift = pos + bits;
379 struct key_vector *tn;
382 /* verify bits and pos their msb bits clear and values are valid */
383 BUG_ON(!bits || (shift > KEYLENGTH));
385 tnode = tnode_alloc(bits);
389 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
390 sizeof(struct key_vector *) << bits);
392 if (bits == KEYLENGTH)
393 tnode->full_children = 1;
395 tnode->empty_children = 1ul << bits;
398 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
406 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
407 * and no bits are skipped. See discussion in dyntree paper p. 6
409 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
411 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
414 /* Add a child at position i overwriting the old value.
415 * Update the value of full_children and empty_children.
417 static void put_child(struct key_vector *tn, unsigned long i,
418 struct key_vector *n)
420 struct key_vector *chi = get_child(tn, i);
423 BUG_ON(i >= child_length(tn));
425 /* update emptyChildren, overflow into fullChildren */
431 /* update fullChildren */
432 wasfull = tnode_full(tn, chi);
433 isfull = tnode_full(tn, n);
435 if (wasfull && !isfull)
436 tn_info(tn)->full_children--;
437 else if (!wasfull && isfull)
438 tn_info(tn)->full_children++;
440 if (n && (tn->slen < n->slen))
443 rcu_assign_pointer(tn->tnode[i], n);
446 static void update_children(struct key_vector *tn)
450 /* update all of the child parent pointers */
451 for (i = child_length(tn); i;) {
452 struct key_vector *inode = get_child(tn, --i);
457 /* Either update the children of a tnode that
458 * already belongs to us or update the child
459 * to point to ourselves.
461 if (node_parent(inode) == tn)
462 update_children(inode);
464 node_set_parent(inode, tn);
468 static inline void put_child_root(struct key_vector *tp, t_key key,
469 struct key_vector *n)
472 rcu_assign_pointer(tp->tnode[0], n);
474 put_child(tp, get_index(key, tp), n);
477 static inline void tnode_free_init(struct key_vector *tn)
479 tn_info(tn)->rcu.next = NULL;
482 static inline void tnode_free_append(struct key_vector *tn,
483 struct key_vector *n)
485 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
486 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
489 static void tnode_free(struct key_vector *tn)
491 struct callback_head *head = &tn_info(tn)->rcu;
495 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
498 tn = container_of(head, struct tnode, rcu)->kv;
501 if (tnode_free_size >= READ_ONCE(sysctl_fib_sync_mem)) {
507 static struct key_vector *replace(struct trie *t,
508 struct key_vector *oldtnode,
509 struct key_vector *tn)
511 struct key_vector *tp = node_parent(oldtnode);
514 /* setup the parent pointer out of and back into this node */
515 NODE_INIT_PARENT(tn, tp);
516 put_child_root(tp, tn->key, tn);
518 /* update all of the child parent pointers */
521 /* all pointers should be clean so we are done */
522 tnode_free(oldtnode);
524 /* resize children now that oldtnode is freed */
525 for (i = child_length(tn); i;) {
526 struct key_vector *inode = get_child(tn, --i);
528 /* resize child node */
529 if (tnode_full(tn, inode))
530 tn = resize(t, inode);
536 static struct key_vector *inflate(struct trie *t,
537 struct key_vector *oldtnode)
539 struct key_vector *tn;
543 pr_debug("In inflate\n");
545 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
549 /* prepare oldtnode to be freed */
550 tnode_free_init(oldtnode);
552 /* Assemble all of the pointers in our cluster, in this case that
553 * represents all of the pointers out of our allocated nodes that
554 * point to existing tnodes and the links between our allocated
557 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
558 struct key_vector *inode = get_child(oldtnode, --i);
559 struct key_vector *node0, *node1;
566 /* A leaf or an internal node with skipped bits */
567 if (!tnode_full(oldtnode, inode)) {
568 put_child(tn, get_index(inode->key, tn), inode);
572 /* drop the node in the old tnode free list */
573 tnode_free_append(oldtnode, inode);
575 /* An internal node with two children */
576 if (inode->bits == 1) {
577 put_child(tn, 2 * i + 1, get_child(inode, 1));
578 put_child(tn, 2 * i, get_child(inode, 0));
582 /* We will replace this node 'inode' with two new
583 * ones, 'node0' and 'node1', each with half of the
584 * original children. The two new nodes will have
585 * a position one bit further down the key and this
586 * means that the "significant" part of their keys
587 * (see the discussion near the top of this file)
588 * will differ by one bit, which will be "0" in
589 * node0's key and "1" in node1's key. Since we are
590 * moving the key position by one step, the bit that
591 * we are moving away from - the bit at position
592 * (tn->pos) - is the one that will differ between
593 * node0 and node1. So... we synthesize that bit in the
596 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
599 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
601 tnode_free_append(tn, node1);
604 tnode_free_append(tn, node0);
606 /* populate child pointers in new nodes */
607 for (k = child_length(inode), j = k / 2; j;) {
608 put_child(node1, --j, get_child(inode, --k));
609 put_child(node0, j, get_child(inode, j));
610 put_child(node1, --j, get_child(inode, --k));
611 put_child(node0, j, get_child(inode, j));
614 /* link new nodes to parent */
615 NODE_INIT_PARENT(node1, tn);
616 NODE_INIT_PARENT(node0, tn);
618 /* link parent to nodes */
619 put_child(tn, 2 * i + 1, node1);
620 put_child(tn, 2 * i, node0);
623 /* setup the parent pointers into and out of this node */
624 return replace(t, oldtnode, tn);
626 /* all pointers should be clean so we are done */
632 static struct key_vector *halve(struct trie *t,
633 struct key_vector *oldtnode)
635 struct key_vector *tn;
638 pr_debug("In halve\n");
640 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
644 /* prepare oldtnode to be freed */
645 tnode_free_init(oldtnode);
647 /* Assemble all of the pointers in our cluster, in this case that
648 * represents all of the pointers out of our allocated nodes that
649 * point to existing tnodes and the links between our allocated
652 for (i = child_length(oldtnode); i;) {
653 struct key_vector *node1 = get_child(oldtnode, --i);
654 struct key_vector *node0 = get_child(oldtnode, --i);
655 struct key_vector *inode;
657 /* At least one of the children is empty */
658 if (!node1 || !node0) {
659 put_child(tn, i / 2, node1 ? : node0);
663 /* Two nonempty children */
664 inode = tnode_new(node0->key, oldtnode->pos, 1);
667 tnode_free_append(tn, inode);
669 /* initialize pointers out of node */
670 put_child(inode, 1, node1);
671 put_child(inode, 0, node0);
672 NODE_INIT_PARENT(inode, tn);
674 /* link parent to node */
675 put_child(tn, i / 2, inode);
678 /* setup the parent pointers into and out of this node */
679 return replace(t, oldtnode, tn);
681 /* all pointers should be clean so we are done */
687 static struct key_vector *collapse(struct trie *t,
688 struct key_vector *oldtnode)
690 struct key_vector *n, *tp;
693 /* scan the tnode looking for that one child that might still exist */
694 for (n = NULL, i = child_length(oldtnode); !n && i;)
695 n = get_child(oldtnode, --i);
697 /* compress one level */
698 tp = node_parent(oldtnode);
699 put_child_root(tp, oldtnode->key, n);
700 node_set_parent(n, tp);
708 static unsigned char update_suffix(struct key_vector *tn)
710 unsigned char slen = tn->pos;
711 unsigned long stride, i;
712 unsigned char slen_max;
714 /* only vector 0 can have a suffix length greater than or equal to
715 * tn->pos + tn->bits, the second highest node will have a suffix
716 * length at most of tn->pos + tn->bits - 1
718 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
720 /* search though the list of children looking for nodes that might
721 * have a suffix greater than the one we currently have. This is
722 * why we start with a stride of 2 since a stride of 1 would
723 * represent the nodes with suffix length equal to tn->pos
725 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
726 struct key_vector *n = get_child(tn, i);
728 if (!n || (n->slen <= slen))
731 /* update stride and slen based on new value */
732 stride <<= (n->slen - slen);
736 /* stop searching if we have hit the maximum possible value */
737 if (slen >= slen_max)
746 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
747 * the Helsinki University of Technology and Matti Tikkanen of Nokia
748 * Telecommunications, page 6:
749 * "A node is doubled if the ratio of non-empty children to all
750 * children in the *doubled* node is at least 'high'."
752 * 'high' in this instance is the variable 'inflate_threshold'. It
753 * is expressed as a percentage, so we multiply it with
754 * child_length() and instead of multiplying by 2 (since the
755 * child array will be doubled by inflate()) and multiplying
756 * the left-hand side by 100 (to handle the percentage thing) we
757 * multiply the left-hand side by 50.
759 * The left-hand side may look a bit weird: child_length(tn)
760 * - tn->empty_children is of course the number of non-null children
761 * in the current node. tn->full_children is the number of "full"
762 * children, that is non-null tnodes with a skip value of 0.
763 * All of those will be doubled in the resulting inflated tnode, so
764 * we just count them one extra time here.
766 * A clearer way to write this would be:
768 * to_be_doubled = tn->full_children;
769 * not_to_be_doubled = child_length(tn) - tn->empty_children -
772 * new_child_length = child_length(tn) * 2;
774 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
776 * if (new_fill_factor >= inflate_threshold)
778 * ...and so on, tho it would mess up the while () loop.
781 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
785 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
786 * inflate_threshold * new_child_length
788 * expand not_to_be_doubled and to_be_doubled, and shorten:
789 * 100 * (child_length(tn) - tn->empty_children +
790 * tn->full_children) >= inflate_threshold * new_child_length
792 * expand new_child_length:
793 * 100 * (child_length(tn) - tn->empty_children +
794 * tn->full_children) >=
795 * inflate_threshold * child_length(tn) * 2
798 * 50 * (tn->full_children + child_length(tn) -
799 * tn->empty_children) >= inflate_threshold *
803 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
805 unsigned long used = child_length(tn);
806 unsigned long threshold = used;
808 /* Keep root node larger */
809 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
810 used -= tn_info(tn)->empty_children;
811 used += tn_info(tn)->full_children;
813 /* if bits == KEYLENGTH then pos = 0, and will fail below */
815 return (used > 1) && tn->pos && ((50 * used) >= threshold);
818 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
820 unsigned long used = child_length(tn);
821 unsigned long threshold = used;
823 /* Keep root node larger */
824 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
825 used -= tn_info(tn)->empty_children;
827 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
829 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
832 static inline bool should_collapse(struct key_vector *tn)
834 unsigned long used = child_length(tn);
836 used -= tn_info(tn)->empty_children;
838 /* account for bits == KEYLENGTH case */
839 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
842 /* One child or none, time to drop us from the trie */
847 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
849 #ifdef CONFIG_IP_FIB_TRIE_STATS
850 struct trie_use_stats __percpu *stats = t->stats;
852 struct key_vector *tp = node_parent(tn);
853 unsigned long cindex = get_index(tn->key, tp);
854 int max_work = MAX_WORK;
856 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
857 tn, inflate_threshold, halve_threshold);
859 /* track the tnode via the pointer from the parent instead of
860 * doing it ourselves. This way we can let RCU fully do its
861 * thing without us interfering
863 BUG_ON(tn != get_child(tp, cindex));
865 /* Double as long as the resulting node has a number of
866 * nonempty nodes that are above the threshold.
868 while (should_inflate(tp, tn) && max_work) {
871 #ifdef CONFIG_IP_FIB_TRIE_STATS
872 this_cpu_inc(stats->resize_node_skipped);
878 tn = get_child(tp, cindex);
881 /* update parent in case inflate failed */
882 tp = node_parent(tn);
884 /* Return if at least one inflate is run */
885 if (max_work != MAX_WORK)
888 /* Halve as long as the number of empty children in this
889 * node is above threshold.
891 while (should_halve(tp, tn) && max_work) {
894 #ifdef CONFIG_IP_FIB_TRIE_STATS
895 this_cpu_inc(stats->resize_node_skipped);
901 tn = get_child(tp, cindex);
904 /* Only one child remains */
905 if (should_collapse(tn))
906 return collapse(t, tn);
908 /* update parent in case halve failed */
909 return node_parent(tn);
912 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
914 unsigned char node_slen = tn->slen;
916 while ((node_slen > tn->pos) && (node_slen > slen)) {
917 slen = update_suffix(tn);
918 if (node_slen == slen)
921 tn = node_parent(tn);
922 node_slen = tn->slen;
926 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
928 while (tn->slen < slen) {
930 tn = node_parent(tn);
934 /* rcu_read_lock needs to be hold by caller from readside */
935 static struct key_vector *fib_find_node(struct trie *t,
936 struct key_vector **tp, u32 key)
938 struct key_vector *pn, *n = t->kv;
939 unsigned long index = 0;
943 n = get_child_rcu(n, index);
948 index = get_cindex(key, n);
950 /* This bit of code is a bit tricky but it combines multiple
951 * checks into a single check. The prefix consists of the
952 * prefix plus zeros for the bits in the cindex. The index
953 * is the difference between the key and this value. From
954 * this we can actually derive several pieces of data.
955 * if (index >= (1ul << bits))
956 * we have a mismatch in skip bits and failed
958 * we know the value is cindex
960 * This check is safe even if bits == KEYLENGTH due to the
961 * fact that we can only allocate a node with 32 bits if a
962 * long is greater than 32 bits.
964 if (index >= (1ul << n->bits)) {
969 /* keep searching until we find a perfect match leaf or NULL */
970 } while (IS_TNODE(n));
977 /* Return the first fib alias matching DSCP with
978 * priority less than or equal to PRIO.
979 * If 'find_first' is set, return the first matching
980 * fib alias, regardless of DSCP and priority.
982 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
983 dscp_t dscp, u32 prio, u32 tb_id,
986 struct fib_alias *fa;
991 hlist_for_each_entry(fa, fah, fa_list) {
992 /* Avoid Sparse warning when using dscp_t in inequalities */
993 u8 __fa_dscp = inet_dscp_to_dsfield(fa->fa_dscp);
994 u8 __dscp = inet_dscp_to_dsfield(dscp);
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_dscp > __dscp)
1008 if (fa->fa_info->fib_priority >= prio || __fa_dscp < __dscp)
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_dscp == fri->dscp && 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 u8 fib_notify_on_flag_change;
1046 struct fib_alias *fa_match;
1047 struct sk_buff *skb;
1052 fa_match = fib_find_matching_alias(net, fri);
1056 /* These are paired with the WRITE_ONCE() happening in this function.
1057 * The reason is that we are only protected by RCU at this point.
1059 if (READ_ONCE(fa_match->offload) == fri->offload &&
1060 READ_ONCE(fa_match->trap) == fri->trap &&
1061 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1064 WRITE_ONCE(fa_match->offload, fri->offload);
1065 WRITE_ONCE(fa_match->trap, fri->trap);
1067 fib_notify_on_flag_change = READ_ONCE(net->ipv4.sysctl_fib_notify_on_flag_change);
1069 /* 2 means send notifications only if offload_failed was changed. */
1070 if (fib_notify_on_flag_change == 2 &&
1071 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1074 WRITE_ONCE(fa_match->offload_failed, fri->offload_failed);
1076 if (!fib_notify_on_flag_change)
1079 skb = nlmsg_new(fib_nlmsg_size(fa_match->fa_info), GFP_ATOMIC);
1085 err = fib_dump_info(skb, 0, 0, RTM_NEWROUTE, fri, 0);
1087 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
1088 WARN_ON(err == -EMSGSIZE);
1093 rtnl_notify(skb, net, 0, RTNLGRP_IPV4_ROUTE, NULL, GFP_ATOMIC);
1097 rtnl_set_sk_err(net, RTNLGRP_IPV4_ROUTE, err);
1101 EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1103 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1105 while (!IS_TRIE(tn))
1109 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1110 struct fib_alias *new, t_key key)
1112 struct key_vector *n, *l;
1114 l = leaf_new(key, new);
1118 /* retrieve child from parent node */
1119 n = get_child(tp, get_index(key, tp));
1121 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1123 * Add a new tnode here
1124 * first tnode need some special handling
1125 * leaves us in position for handling as case 3
1128 struct key_vector *tn;
1130 tn = tnode_new(key, __fls(key ^ n->key), 1);
1134 /* initialize routes out of node */
1135 NODE_INIT_PARENT(tn, tp);
1136 put_child(tn, get_index(key, tn) ^ 1, n);
1138 /* start adding routes into the node */
1139 put_child_root(tp, key, tn);
1140 node_set_parent(n, tn);
1142 /* parent now has a NULL spot where the leaf can go */
1146 /* Case 3: n is NULL, and will just insert a new leaf */
1147 node_push_suffix(tp, new->fa_slen);
1148 NODE_INIT_PARENT(l, tp);
1149 put_child_root(tp, key, l);
1150 trie_rebalance(t, tp);
1159 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1160 struct key_vector *l, struct fib_alias *new,
1161 struct fib_alias *fa, t_key key)
1164 return fib_insert_node(t, tp, new, key);
1167 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1169 struct fib_alias *last;
1171 hlist_for_each_entry(last, &l->leaf, fa_list) {
1172 if (new->fa_slen < last->fa_slen)
1174 if ((new->fa_slen == last->fa_slen) &&
1175 (new->tb_id > last->tb_id))
1181 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1183 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1186 /* if we added to the tail node then we need to update slen */
1187 if (l->slen < new->fa_slen) {
1188 l->slen = new->fa_slen;
1189 node_push_suffix(tp, new->fa_slen);
1195 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1197 if (plen > KEYLENGTH) {
1198 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1202 if ((plen < KEYLENGTH) && (key << plen)) {
1203 NL_SET_ERR_MSG(extack,
1204 "Invalid prefix for given prefix length");
1211 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1212 struct key_vector *l, struct fib_alias *old);
1214 /* Caller must hold RTNL. */
1215 int fib_table_insert(struct net *net, struct fib_table *tb,
1216 struct fib_config *cfg, struct netlink_ext_ack *extack)
1218 struct trie *t = (struct trie *)tb->tb_data;
1219 struct fib_alias *fa, *new_fa;
1220 struct key_vector *l, *tp;
1221 u16 nlflags = NLM_F_EXCL;
1222 struct fib_info *fi;
1223 u8 plen = cfg->fc_dst_len;
1224 u8 slen = KEYLENGTH - plen;
1229 key = ntohl(cfg->fc_dst);
1231 if (!fib_valid_key_len(key, plen, extack))
1234 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1236 fi = fib_create_info(cfg, extack);
1242 dscp = cfg->fc_dscp;
1243 l = fib_find_node(t, &tp, key);
1244 fa = l ? fib_find_alias(&l->leaf, slen, dscp, fi->fib_priority,
1245 tb->tb_id, false) : NULL;
1247 /* Now fa, if non-NULL, points to the first fib alias
1248 * with the same keys [prefix,dscp,priority], if such key already
1249 * exists or to the node before which we will insert new one.
1251 * If fa is NULL, we will need to allocate a new one and
1252 * insert to the tail of the section matching the suffix length
1256 if (fa && fa->fa_dscp == dscp &&
1257 fa->fa_info->fib_priority == fi->fib_priority) {
1258 struct fib_alias *fa_first, *fa_match;
1261 if (cfg->fc_nlflags & NLM_F_EXCL)
1264 nlflags &= ~NLM_F_EXCL;
1267 * 1. Find exact match for type, scope, fib_info to avoid
1269 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1273 hlist_for_each_entry_from(fa, fa_list) {
1274 if ((fa->fa_slen != slen) ||
1275 (fa->tb_id != tb->tb_id) ||
1276 (fa->fa_dscp != dscp))
1278 if (fa->fa_info->fib_priority != fi->fib_priority)
1280 if (fa->fa_type == cfg->fc_type &&
1281 fa->fa_info == fi) {
1287 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1288 struct fib_info *fi_drop;
1291 nlflags |= NLM_F_REPLACE;
1299 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1303 fi_drop = fa->fa_info;
1304 new_fa->fa_dscp = fa->fa_dscp;
1305 new_fa->fa_info = fi;
1306 new_fa->fa_type = cfg->fc_type;
1307 state = fa->fa_state;
1308 new_fa->fa_state = state & ~FA_S_ACCESSED;
1309 new_fa->fa_slen = fa->fa_slen;
1310 new_fa->tb_id = tb->tb_id;
1311 new_fa->fa_default = -1;
1312 new_fa->offload = 0;
1314 new_fa->offload_failed = 0;
1316 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1318 if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1319 tb->tb_id, true) == new_fa) {
1320 enum fib_event_type fib_event;
1322 fib_event = FIB_EVENT_ENTRY_REPLACE;
1323 err = call_fib_entry_notifiers(net, fib_event,
1327 hlist_replace_rcu(&new_fa->fa_list,
1329 goto out_free_new_fa;
1333 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1334 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1336 alias_free_mem_rcu(fa);
1338 fib_release_info(fi_drop);
1339 if (state & FA_S_ACCESSED)
1340 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1344 /* Error if we find a perfect match which
1345 * uses the same scope, type, and nexthop
1351 if (cfg->fc_nlflags & NLM_F_APPEND)
1352 nlflags |= NLM_F_APPEND;
1357 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1360 nlflags |= NLM_F_CREATE;
1362 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1366 new_fa->fa_info = fi;
1367 new_fa->fa_dscp = dscp;
1368 new_fa->fa_type = cfg->fc_type;
1369 new_fa->fa_state = 0;
1370 new_fa->fa_slen = slen;
1371 new_fa->tb_id = tb->tb_id;
1372 new_fa->fa_default = -1;
1373 new_fa->offload = 0;
1375 new_fa->offload_failed = 0;
1377 /* Insert new entry to the list. */
1378 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1380 goto out_free_new_fa;
1382 /* The alias was already inserted, so the node must exist. */
1383 l = l ? l : fib_find_node(t, &tp, key);
1384 if (WARN_ON_ONCE(!l)) {
1386 goto out_free_new_fa;
1389 if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1391 enum fib_event_type fib_event;
1393 fib_event = FIB_EVENT_ENTRY_REPLACE;
1394 err = call_fib_entry_notifiers(net, fib_event, key, plen,
1397 goto out_remove_new_fa;
1401 tb->tb_num_default++;
1403 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1404 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1405 &cfg->fc_nlinfo, nlflags);
1410 fib_remove_alias(t, tp, l, new_fa);
1412 kmem_cache_free(fn_alias_kmem, new_fa);
1414 fib_release_info(fi);
1419 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1421 t_key prefix = n->key;
1423 return (key ^ prefix) & (prefix | -prefix);
1426 bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1427 const struct flowi4 *flp)
1429 if (nhc->nhc_flags & RTNH_F_DEAD)
1432 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1433 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1434 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1437 if (flp->flowi4_oif && flp->flowi4_oif != nhc->nhc_oif)
1443 /* should be called with rcu_read_lock */
1444 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1445 struct fib_result *res, int fib_flags)
1447 struct trie *t = (struct trie *) tb->tb_data;
1448 #ifdef CONFIG_IP_FIB_TRIE_STATS
1449 struct trie_use_stats __percpu *stats = t->stats;
1451 const t_key key = ntohl(flp->daddr);
1452 struct key_vector *n, *pn;
1453 struct fib_alias *fa;
1454 unsigned long index;
1460 n = get_child_rcu(pn, cindex);
1462 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1466 #ifdef CONFIG_IP_FIB_TRIE_STATS
1467 this_cpu_inc(stats->gets);
1470 /* Step 1: Travel to the longest prefix match in the trie */
1472 index = get_cindex(key, n);
1474 /* This bit of code is a bit tricky but it combines multiple
1475 * checks into a single check. The prefix consists of the
1476 * prefix plus zeros for the "bits" in the prefix. The index
1477 * is the difference between the key and this value. From
1478 * this we can actually derive several pieces of data.
1479 * if (index >= (1ul << bits))
1480 * we have a mismatch in skip bits and failed
1482 * we know the value is cindex
1484 * This check is safe even if bits == KEYLENGTH due to the
1485 * fact that we can only allocate a node with 32 bits if a
1486 * long is greater than 32 bits.
1488 if (index >= (1ul << n->bits))
1491 /* we have found a leaf. Prefixes have already been compared */
1495 /* only record pn and cindex if we are going to be chopping
1496 * bits later. Otherwise we are just wasting cycles.
1498 if (n->slen > n->pos) {
1503 n = get_child_rcu(n, index);
1508 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1510 /* record the pointer where our next node pointer is stored */
1511 struct key_vector __rcu **cptr = n->tnode;
1513 /* This test verifies that none of the bits that differ
1514 * between the key and the prefix exist in the region of
1515 * the lsb and higher in the prefix.
1517 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1520 /* exit out and process leaf */
1521 if (unlikely(IS_LEAF(n)))
1524 /* Don't bother recording parent info. Since we are in
1525 * prefix match mode we will have to come back to wherever
1526 * we started this traversal anyway
1529 while ((n = rcu_dereference(*cptr)) == NULL) {
1531 #ifdef CONFIG_IP_FIB_TRIE_STATS
1533 this_cpu_inc(stats->null_node_hit);
1535 /* If we are at cindex 0 there are no more bits for
1536 * us to strip at this level so we must ascend back
1537 * up one level to see if there are any more bits to
1538 * be stripped there.
1541 t_key pkey = pn->key;
1543 /* If we don't have a parent then there is
1544 * nothing for us to do as we do not have any
1545 * further nodes to parse.
1548 trace_fib_table_lookup(tb->tb_id, flp,
1552 #ifdef CONFIG_IP_FIB_TRIE_STATS
1553 this_cpu_inc(stats->backtrack);
1555 /* Get Child's index */
1556 pn = node_parent_rcu(pn);
1557 cindex = get_index(pkey, pn);
1560 /* strip the least significant bit from the cindex */
1561 cindex &= cindex - 1;
1563 /* grab pointer for next child node */
1564 cptr = &pn->tnode[cindex];
1569 /* this line carries forward the xor from earlier in the function */
1570 index = key ^ n->key;
1572 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1573 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1574 struct fib_info *fi = fa->fa_info;
1575 struct fib_nh_common *nhc;
1578 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1579 if (index >= (1ul << fa->fa_slen))
1583 inet_dscp_to_dsfield(fa->fa_dscp) != flp->flowi4_tos)
1585 /* Paired with WRITE_ONCE() in fib_release_info() */
1586 if (READ_ONCE(fi->fib_dead))
1588 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1590 fib_alias_accessed(fa);
1591 err = fib_props[fa->fa_type].error;
1592 if (unlikely(err < 0)) {
1594 #ifdef CONFIG_IP_FIB_TRIE_STATS
1595 this_cpu_inc(stats->semantic_match_passed);
1597 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1600 if (fi->fib_flags & RTNH_F_DEAD)
1603 if (unlikely(fi->nh)) {
1604 if (nexthop_is_blackhole(fi->nh)) {
1605 err = fib_props[RTN_BLACKHOLE].error;
1609 nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1616 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1617 nhc = fib_info_nhc(fi, nhsel);
1619 if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1622 if (!(fib_flags & FIB_LOOKUP_NOREF))
1623 refcount_inc(&fi->fib_clntref);
1625 res->prefix = htonl(n->key);
1626 res->prefixlen = KEYLENGTH - fa->fa_slen;
1627 res->nh_sel = nhsel;
1629 res->type = fa->fa_type;
1630 res->scope = fi->fib_scope;
1633 res->fa_head = &n->leaf;
1634 #ifdef CONFIG_IP_FIB_TRIE_STATS
1635 this_cpu_inc(stats->semantic_match_passed);
1637 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1643 #ifdef CONFIG_IP_FIB_TRIE_STATS
1644 this_cpu_inc(stats->semantic_match_miss);
1648 EXPORT_SYMBOL_GPL(fib_table_lookup);
1650 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1651 struct key_vector *l, struct fib_alias *old)
1653 /* record the location of the previous list_info entry */
1654 struct hlist_node **pprev = old->fa_list.pprev;
1655 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1657 /* remove the fib_alias from the list */
1658 hlist_del_rcu(&old->fa_list);
1660 /* if we emptied the list this leaf will be freed and we can sort
1661 * out parent suffix lengths as a part of trie_rebalance
1663 if (hlist_empty(&l->leaf)) {
1664 if (tp->slen == l->slen)
1665 node_pull_suffix(tp, tp->pos);
1666 put_child_root(tp, l->key, NULL);
1668 trie_rebalance(t, tp);
1672 /* only access fa if it is pointing at the last valid hlist_node */
1676 /* update the trie with the latest suffix length */
1677 l->slen = fa->fa_slen;
1678 node_pull_suffix(tp, fa->fa_slen);
1681 static void fib_notify_alias_delete(struct net *net, u32 key,
1682 struct hlist_head *fah,
1683 struct fib_alias *fa_to_delete,
1684 struct netlink_ext_ack *extack)
1686 struct fib_alias *fa_next, *fa_to_notify;
1687 u32 tb_id = fa_to_delete->tb_id;
1688 u8 slen = fa_to_delete->fa_slen;
1689 enum fib_event_type fib_event;
1691 /* Do not notify if we do not care about the route. */
1692 if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1695 /* Determine if the route should be replaced by the next route in the
1698 fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1699 struct fib_alias, fa_list);
1700 if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1701 fib_event = FIB_EVENT_ENTRY_REPLACE;
1702 fa_to_notify = fa_next;
1704 fib_event = FIB_EVENT_ENTRY_DEL;
1705 fa_to_notify = fa_to_delete;
1707 call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1708 fa_to_notify, extack);
1711 /* Caller must hold RTNL. */
1712 int fib_table_delete(struct net *net, struct fib_table *tb,
1713 struct fib_config *cfg, struct netlink_ext_ack *extack)
1715 struct trie *t = (struct trie *) tb->tb_data;
1716 struct fib_alias *fa, *fa_to_delete;
1717 struct key_vector *l, *tp;
1718 u8 plen = cfg->fc_dst_len;
1719 u8 slen = KEYLENGTH - plen;
1723 key = ntohl(cfg->fc_dst);
1725 if (!fib_valid_key_len(key, plen, extack))
1728 l = fib_find_node(t, &tp, key);
1732 dscp = cfg->fc_dscp;
1733 fa = fib_find_alias(&l->leaf, slen, dscp, 0, tb->tb_id, false);
1737 pr_debug("Deleting %08x/%d dsfield=0x%02x t=%p\n", key, plen,
1738 inet_dscp_to_dsfield(dscp), t);
1740 fa_to_delete = NULL;
1741 hlist_for_each_entry_from(fa, fa_list) {
1742 struct fib_info *fi = fa->fa_info;
1744 if ((fa->fa_slen != slen) ||
1745 (fa->tb_id != tb->tb_id) ||
1746 (fa->fa_dscp != dscp))
1749 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1750 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1751 fa->fa_info->fib_scope == cfg->fc_scope) &&
1752 (!cfg->fc_prefsrc ||
1753 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1754 (!cfg->fc_protocol ||
1755 fi->fib_protocol == cfg->fc_protocol) &&
1756 fib_nh_match(net, cfg, fi, extack) == 0 &&
1757 fib_metrics_match(cfg, fi)) {
1766 fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1767 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1768 &cfg->fc_nlinfo, 0);
1771 tb->tb_num_default--;
1773 fib_remove_alias(t, tp, l, fa_to_delete);
1775 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1776 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1778 fib_release_info(fa_to_delete->fa_info);
1779 alias_free_mem_rcu(fa_to_delete);
1783 /* Scan for the next leaf starting at the provided key value */
1784 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1786 struct key_vector *pn, *n = *tn;
1787 unsigned long cindex;
1789 /* this loop is meant to try and find the key in the trie */
1791 /* record parent and next child index */
1793 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1795 if (cindex >> pn->bits)
1798 /* descend into the next child */
1799 n = get_child_rcu(pn, cindex++);
1803 /* guarantee forward progress on the keys */
1804 if (IS_LEAF(n) && (n->key >= key))
1806 } while (IS_TNODE(n));
1808 /* this loop will search for the next leaf with a greater key */
1809 while (!IS_TRIE(pn)) {
1810 /* if we exhausted the parent node we will need to climb */
1811 if (cindex >= (1ul << pn->bits)) {
1812 t_key pkey = pn->key;
1814 pn = node_parent_rcu(pn);
1815 cindex = get_index(pkey, pn) + 1;
1819 /* grab the next available node */
1820 n = get_child_rcu(pn, cindex++);
1824 /* no need to compare keys since we bumped the index */
1828 /* Rescan start scanning in new node */
1834 return NULL; /* Root of trie */
1836 /* if we are at the limit for keys just return NULL for the tnode */
1841 static void fib_trie_free(struct fib_table *tb)
1843 struct trie *t = (struct trie *)tb->tb_data;
1844 struct key_vector *pn = t->kv;
1845 unsigned long cindex = 1;
1846 struct hlist_node *tmp;
1847 struct fib_alias *fa;
1849 /* walk trie in reverse order and free everything */
1851 struct key_vector *n;
1854 t_key pkey = pn->key;
1860 pn = node_parent(pn);
1862 /* drop emptied tnode */
1863 put_child_root(pn, n->key, NULL);
1866 cindex = get_index(pkey, pn);
1871 /* grab the next available node */
1872 n = get_child(pn, cindex);
1877 /* record pn and cindex for leaf walking */
1879 cindex = 1ul << n->bits;
1884 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1885 hlist_del_rcu(&fa->fa_list);
1886 alias_free_mem_rcu(fa);
1889 put_child_root(pn, n->key, NULL);
1893 #ifdef CONFIG_IP_FIB_TRIE_STATS
1894 free_percpu(t->stats);
1899 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1901 struct trie *ot = (struct trie *)oldtb->tb_data;
1902 struct key_vector *l, *tp = ot->kv;
1903 struct fib_table *local_tb;
1904 struct fib_alias *fa;
1908 if (oldtb->tb_data == oldtb->__data)
1911 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1915 lt = (struct trie *)local_tb->tb_data;
1917 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1918 struct key_vector *local_l = NULL, *local_tp;
1920 hlist_for_each_entry(fa, &l->leaf, fa_list) {
1921 struct fib_alias *new_fa;
1923 if (local_tb->tb_id != fa->tb_id)
1926 /* clone fa for new local table */
1927 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1931 memcpy(new_fa, fa, sizeof(*fa));
1933 /* insert clone into table */
1935 local_l = fib_find_node(lt, &local_tp, l->key);
1937 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1939 kmem_cache_free(fn_alias_kmem, new_fa);
1944 /* stop loop if key wrapped back to 0 */
1952 fib_trie_free(local_tb);
1957 /* Caller must hold RTNL */
1958 void fib_table_flush_external(struct fib_table *tb)
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;
1966 /* walk trie in reverse order */
1968 unsigned char slen = 0;
1969 struct key_vector *n;
1972 t_key pkey = pn->key;
1974 /* cannot resize the trie vector */
1978 /* update the suffix to address pulled leaves */
1979 if (pn->slen > pn->pos)
1982 /* resize completed node */
1984 cindex = get_index(pkey, pn);
1989 /* grab the next available node */
1990 n = get_child(pn, cindex);
1995 /* record pn and cindex for leaf walking */
1997 cindex = 1ul << n->bits;
2002 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2003 /* if alias was cloned to local then we just
2004 * need to remove the local copy from main
2006 if (tb->tb_id != fa->tb_id) {
2007 hlist_del_rcu(&fa->fa_list);
2008 alias_free_mem_rcu(fa);
2012 /* record local slen */
2016 /* update leaf slen */
2019 if (hlist_empty(&n->leaf)) {
2020 put_child_root(pn, n->key, NULL);
2026 /* Caller must hold RTNL. */
2027 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
2029 struct trie *t = (struct trie *)tb->tb_data;
2030 struct key_vector *pn = t->kv;
2031 unsigned long cindex = 1;
2032 struct hlist_node *tmp;
2033 struct fib_alias *fa;
2036 /* walk trie in reverse order */
2038 unsigned char slen = 0;
2039 struct key_vector *n;
2042 t_key pkey = pn->key;
2044 /* cannot resize the trie vector */
2048 /* update the suffix to address pulled leaves */
2049 if (pn->slen > pn->pos)
2052 /* resize completed node */
2054 cindex = get_index(pkey, pn);
2059 /* grab the next available node */
2060 n = get_child(pn, cindex);
2065 /* record pn and cindex for leaf walking */
2067 cindex = 1ul << n->bits;
2072 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2073 struct fib_info *fi = fa->fa_info;
2075 if (!fi || tb->tb_id != fa->tb_id ||
2076 (!(fi->fib_flags & RTNH_F_DEAD) &&
2077 !fib_props[fa->fa_type].error)) {
2082 /* Do not flush error routes if network namespace is
2083 * not being dismantled
2085 if (!flush_all && fib_props[fa->fa_type].error) {
2090 fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2092 hlist_del_rcu(&fa->fa_list);
2093 fib_release_info(fa->fa_info);
2094 alias_free_mem_rcu(fa);
2098 /* update leaf slen */
2101 if (hlist_empty(&n->leaf)) {
2102 put_child_root(pn, n->key, NULL);
2107 pr_debug("trie_flush found=%d\n", found);
2111 /* derived from fib_trie_free */
2112 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2113 struct nl_info *info)
2115 struct trie *t = (struct trie *)tb->tb_data;
2116 struct key_vector *pn = t->kv;
2117 unsigned long cindex = 1;
2118 struct fib_alias *fa;
2121 struct key_vector *n;
2124 t_key pkey = pn->key;
2129 pn = node_parent(pn);
2130 cindex = get_index(pkey, pn);
2134 /* grab the next available node */
2135 n = get_child(pn, cindex);
2140 /* record pn and cindex for leaf walking */
2142 cindex = 1ul << n->bits;
2147 hlist_for_each_entry(fa, &n->leaf, fa_list) {
2148 struct fib_info *fi = fa->fa_info;
2150 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2153 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2154 KEYLENGTH - fa->fa_slen, tb->tb_id,
2155 info, NLM_F_REPLACE);
2160 void fib_info_notify_update(struct net *net, struct nl_info *info)
2164 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2165 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2166 struct fib_table *tb;
2168 hlist_for_each_entry_rcu(tb, head, tb_hlist,
2169 lockdep_rtnl_is_held())
2170 __fib_info_notify_update(net, tb, info);
2174 static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2175 struct notifier_block *nb,
2176 struct netlink_ext_ack *extack)
2178 struct fib_alias *fa;
2182 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2183 struct fib_info *fi = fa->fa_info;
2188 /* local and main table can share the same trie,
2189 * so don't notify twice for the same entry.
2191 if (tb->tb_id != fa->tb_id)
2194 if (fa->fa_slen == last_slen)
2197 last_slen = fa->fa_slen;
2198 err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2199 l->key, KEYLENGTH - fa->fa_slen,
2207 static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2208 struct netlink_ext_ack *extack)
2210 struct trie *t = (struct trie *)tb->tb_data;
2211 struct key_vector *l, *tp = t->kv;
2215 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2216 err = fib_leaf_notify(l, tb, nb, extack);
2221 /* stop in case of wrap around */
2228 int fib_notify(struct net *net, struct notifier_block *nb,
2229 struct netlink_ext_ack *extack)
2234 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2235 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2236 struct fib_table *tb;
2238 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2239 err = fib_table_notify(tb, nb, extack);
2247 static void __trie_free_rcu(struct rcu_head *head)
2249 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2250 #ifdef CONFIG_IP_FIB_TRIE_STATS
2251 struct trie *t = (struct trie *)tb->tb_data;
2253 if (tb->tb_data == tb->__data)
2254 free_percpu(t->stats);
2255 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2259 void fib_free_table(struct fib_table *tb)
2261 call_rcu(&tb->rcu, __trie_free_rcu);
2264 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2265 struct sk_buff *skb, struct netlink_callback *cb,
2266 struct fib_dump_filter *filter)
2268 unsigned int flags = NLM_F_MULTI;
2269 __be32 xkey = htonl(l->key);
2270 int i, s_i, i_fa, s_fa, err;
2271 struct fib_alias *fa;
2273 if (filter->filter_set ||
2274 !filter->dump_exceptions || !filter->dump_routes)
2275 flags |= NLM_F_DUMP_FILTERED;
2281 /* rcu_read_lock is hold by caller */
2282 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2283 struct fib_info *fi = fa->fa_info;
2290 if (tb->tb_id != fa->tb_id)
2293 if (filter->filter_set) {
2294 if (filter->rt_type && fa->fa_type != filter->rt_type)
2297 if ((filter->protocol &&
2298 fi->fib_protocol != filter->protocol))
2302 !fib_info_nh_uses_dev(fi, filter->dev))
2306 if (filter->dump_routes) {
2308 struct fib_rt_info fri;
2311 fri.tb_id = tb->tb_id;
2313 fri.dst_len = KEYLENGTH - fa->fa_slen;
2314 fri.dscp = fa->fa_dscp;
2315 fri.type = fa->fa_type;
2316 fri.offload = READ_ONCE(fa->offload);
2317 fri.trap = READ_ONCE(fa->trap);
2318 fri.offload_failed = READ_ONCE(fa->offload_failed);
2319 err = fib_dump_info(skb,
2320 NETLINK_CB(cb->skb).portid,
2322 RTM_NEWROUTE, &fri, flags);
2330 if (filter->dump_exceptions) {
2331 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2332 &i_fa, s_fa, flags);
2350 /* rcu_read_lock needs to be hold by caller from readside */
2351 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2352 struct netlink_callback *cb, struct fib_dump_filter *filter)
2354 struct trie *t = (struct trie *)tb->tb_data;
2355 struct key_vector *l, *tp = t->kv;
2356 /* Dump starting at last key.
2357 * Note: 0.0.0.0/0 (ie default) is first key.
2359 int count = cb->args[2];
2360 t_key key = cb->args[3];
2362 /* First time here, count and key are both always 0. Count > 0
2363 * and key == 0 means the dump has wrapped around and we are done.
2368 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2371 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2374 cb->args[2] = count;
2381 memset(&cb->args[4], 0,
2382 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2384 /* stop loop if key wrapped back to 0 */
2390 cb->args[2] = count;
2395 void __init fib_trie_init(void)
2397 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2398 sizeof(struct fib_alias),
2399 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2401 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2403 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2406 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2408 struct fib_table *tb;
2410 size_t sz = sizeof(*tb);
2413 sz += sizeof(struct trie);
2415 tb = kzalloc(sz, GFP_KERNEL);
2420 tb->tb_num_default = 0;
2421 tb->tb_data = (alias ? alias->__data : tb->__data);
2426 t = (struct trie *) tb->tb_data;
2427 t->kv[0].pos = KEYLENGTH;
2428 t->kv[0].slen = KEYLENGTH;
2429 #ifdef CONFIG_IP_FIB_TRIE_STATS
2430 t->stats = alloc_percpu(struct trie_use_stats);
2440 #ifdef CONFIG_PROC_FS
2441 /* Depth first Trie walk iterator */
2442 struct fib_trie_iter {
2443 struct seq_net_private p;
2444 struct fib_table *tb;
2445 struct key_vector *tnode;
2450 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2452 unsigned long cindex = iter->index;
2453 struct key_vector *pn = iter->tnode;
2456 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2457 iter->tnode, iter->index, iter->depth);
2459 while (!IS_TRIE(pn)) {
2460 while (cindex < child_length(pn)) {
2461 struct key_vector *n = get_child_rcu(pn, cindex++);
2468 iter->index = cindex;
2470 /* push down one level */
2479 /* Current node exhausted, pop back up */
2481 pn = node_parent_rcu(pn);
2482 cindex = get_index(pkey, pn) + 1;
2486 /* record root node so further searches know we are done */
2493 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2496 struct key_vector *n, *pn;
2502 n = rcu_dereference(pn->tnode[0]);
2519 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2521 struct key_vector *n;
2522 struct fib_trie_iter iter;
2524 memset(s, 0, sizeof(*s));
2527 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2529 struct fib_alias *fa;
2532 s->totdepth += iter.depth;
2533 if (iter.depth > s->maxdepth)
2534 s->maxdepth = iter.depth;
2536 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2540 if (n->bits < MAX_STAT_DEPTH)
2541 s->nodesizes[n->bits]++;
2542 s->nullpointers += tn_info(n)->empty_children;
2549 * This outputs /proc/net/fib_triestats
2551 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2553 unsigned int i, max, pointers, bytes, avdepth;
2556 avdepth = stat->totdepth*100 / stat->leaves;
2560 seq_printf(seq, "\tAver depth: %u.%02d\n",
2561 avdepth / 100, avdepth % 100);
2562 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2564 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2565 bytes = LEAF_SIZE * stat->leaves;
2567 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2568 bytes += sizeof(struct fib_alias) * stat->prefixes;
2570 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2571 bytes += TNODE_SIZE(0) * stat->tnodes;
2573 max = MAX_STAT_DEPTH;
2574 while (max > 0 && stat->nodesizes[max-1] == 0)
2578 for (i = 1; i < max; i++)
2579 if (stat->nodesizes[i] != 0) {
2580 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2581 pointers += (1<<i) * stat->nodesizes[i];
2583 seq_putc(seq, '\n');
2584 seq_printf(seq, "\tPointers: %u\n", pointers);
2586 bytes += sizeof(struct key_vector *) * pointers;
2587 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2588 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2591 #ifdef CONFIG_IP_FIB_TRIE_STATS
2592 static void trie_show_usage(struct seq_file *seq,
2593 const struct trie_use_stats __percpu *stats)
2595 struct trie_use_stats s = { 0 };
2598 /* loop through all of the CPUs and gather up the stats */
2599 for_each_possible_cpu(cpu) {
2600 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2602 s.gets += pcpu->gets;
2603 s.backtrack += pcpu->backtrack;
2604 s.semantic_match_passed += pcpu->semantic_match_passed;
2605 s.semantic_match_miss += pcpu->semantic_match_miss;
2606 s.null_node_hit += pcpu->null_node_hit;
2607 s.resize_node_skipped += pcpu->resize_node_skipped;
2610 seq_printf(seq, "\nCounters:\n---------\n");
2611 seq_printf(seq, "gets = %u\n", s.gets);
2612 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2613 seq_printf(seq, "semantic match passed = %u\n",
2614 s.semantic_match_passed);
2615 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2616 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2617 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2619 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2621 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2623 if (tb->tb_id == RT_TABLE_LOCAL)
2624 seq_puts(seq, "Local:\n");
2625 else if (tb->tb_id == RT_TABLE_MAIN)
2626 seq_puts(seq, "Main:\n");
2628 seq_printf(seq, "Id %d:\n", tb->tb_id);
2632 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2634 struct net *net = seq->private;
2638 "Basic info: size of leaf:"
2639 " %zd bytes, size of tnode: %zd bytes.\n",
2640 LEAF_SIZE, TNODE_SIZE(0));
2643 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2644 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2645 struct fib_table *tb;
2647 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2648 struct trie *t = (struct trie *) tb->tb_data;
2649 struct trie_stat stat;
2654 fib_table_print(seq, tb);
2656 trie_collect_stats(t, &stat);
2657 trie_show_stats(seq, &stat);
2658 #ifdef CONFIG_IP_FIB_TRIE_STATS
2659 trie_show_usage(seq, t->stats);
2669 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2671 struct fib_trie_iter *iter = seq->private;
2672 struct net *net = seq_file_net(seq);
2676 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2677 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2678 struct fib_table *tb;
2680 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2681 struct key_vector *n;
2683 for (n = fib_trie_get_first(iter,
2684 (struct trie *) tb->tb_data);
2685 n; n = fib_trie_get_next(iter))
2696 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2700 return fib_trie_get_idx(seq, *pos);
2703 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2705 struct fib_trie_iter *iter = seq->private;
2706 struct net *net = seq_file_net(seq);
2707 struct fib_table *tb = iter->tb;
2708 struct hlist_node *tb_node;
2710 struct key_vector *n;
2713 /* next node in same table */
2714 n = fib_trie_get_next(iter);
2718 /* walk rest of this hash chain */
2719 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2720 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2721 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2722 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2727 /* new hash chain */
2728 while (++h < FIB_TABLE_HASHSZ) {
2729 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2730 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2731 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2743 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2749 static void seq_indent(struct seq_file *seq, int n)
2755 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2758 case RT_SCOPE_UNIVERSE: return "universe";
2759 case RT_SCOPE_SITE: return "site";
2760 case RT_SCOPE_LINK: return "link";
2761 case RT_SCOPE_HOST: return "host";
2762 case RT_SCOPE_NOWHERE: return "nowhere";
2764 snprintf(buf, len, "scope=%d", s);
2769 static const char *const rtn_type_names[__RTN_MAX] = {
2770 [RTN_UNSPEC] = "UNSPEC",
2771 [RTN_UNICAST] = "UNICAST",
2772 [RTN_LOCAL] = "LOCAL",
2773 [RTN_BROADCAST] = "BROADCAST",
2774 [RTN_ANYCAST] = "ANYCAST",
2775 [RTN_MULTICAST] = "MULTICAST",
2776 [RTN_BLACKHOLE] = "BLACKHOLE",
2777 [RTN_UNREACHABLE] = "UNREACHABLE",
2778 [RTN_PROHIBIT] = "PROHIBIT",
2779 [RTN_THROW] = "THROW",
2781 [RTN_XRESOLVE] = "XRESOLVE",
2784 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2786 if (t < __RTN_MAX && rtn_type_names[t])
2787 return rtn_type_names[t];
2788 snprintf(buf, len, "type %u", t);
2792 /* Pretty print the trie */
2793 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2795 const struct fib_trie_iter *iter = seq->private;
2796 struct key_vector *n = v;
2798 if (IS_TRIE(node_parent_rcu(n)))
2799 fib_table_print(seq, iter->tb);
2802 __be32 prf = htonl(n->key);
2804 seq_indent(seq, iter->depth-1);
2805 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2806 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2807 tn_info(n)->full_children,
2808 tn_info(n)->empty_children);
2810 __be32 val = htonl(n->key);
2811 struct fib_alias *fa;
2813 seq_indent(seq, iter->depth);
2814 seq_printf(seq, " |-- %pI4\n", &val);
2816 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2817 char buf1[32], buf2[32];
2819 seq_indent(seq, iter->depth + 1);
2820 seq_printf(seq, " /%zu %s %s",
2821 KEYLENGTH - fa->fa_slen,
2822 rtn_scope(buf1, sizeof(buf1),
2823 fa->fa_info->fib_scope),
2824 rtn_type(buf2, sizeof(buf2),
2827 seq_printf(seq, " tos=%d",
2828 inet_dscp_to_dsfield(fa->fa_dscp));
2829 seq_putc(seq, '\n');
2836 static const struct seq_operations fib_trie_seq_ops = {
2837 .start = fib_trie_seq_start,
2838 .next = fib_trie_seq_next,
2839 .stop = fib_trie_seq_stop,
2840 .show = fib_trie_seq_show,
2843 struct fib_route_iter {
2844 struct seq_net_private p;
2845 struct fib_table *main_tb;
2846 struct key_vector *tnode;
2851 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2854 struct key_vector *l, **tp = &iter->tnode;
2857 /* use cached location of previously found key */
2858 if (iter->pos > 0 && pos >= iter->pos) {
2867 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2872 /* handle unlikely case of a key wrap */
2878 iter->key = l->key; /* remember it */
2880 iter->pos = 0; /* forget it */
2885 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2888 struct fib_route_iter *iter = seq->private;
2889 struct fib_table *tb;
2894 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2899 t = (struct trie *)tb->tb_data;
2900 iter->tnode = t->kv;
2903 return fib_route_get_idx(iter, *pos);
2906 iter->key = KEY_MAX;
2908 return SEQ_START_TOKEN;
2911 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2913 struct fib_route_iter *iter = seq->private;
2914 struct key_vector *l = NULL;
2915 t_key key = iter->key + 1;
2919 /* only allow key of 0 for start of sequence */
2920 if ((v == SEQ_START_TOKEN) || key)
2921 l = leaf_walk_rcu(&iter->tnode, key);
2933 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2939 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2941 unsigned int flags = 0;
2943 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2946 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2948 if (nhc->nhc_gw.ipv4)
2949 flags |= RTF_GATEWAY;
2951 if (mask == htonl(0xFFFFFFFF))
2958 * This outputs /proc/net/route.
2959 * The format of the file is not supposed to be changed
2960 * and needs to be same as fib_hash output to avoid breaking
2963 static int fib_route_seq_show(struct seq_file *seq, void *v)
2965 struct fib_route_iter *iter = seq->private;
2966 struct fib_table *tb = iter->main_tb;
2967 struct fib_alias *fa;
2968 struct key_vector *l = v;
2971 if (v == SEQ_START_TOKEN) {
2972 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2973 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2978 prefix = htonl(l->key);
2980 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2981 struct fib_info *fi = fa->fa_info;
2982 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2983 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2985 if ((fa->fa_type == RTN_BROADCAST) ||
2986 (fa->fa_type == RTN_MULTICAST))
2989 if (fa->tb_id != tb->tb_id)
2992 seq_setwidth(seq, 127);
2995 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2998 if (nhc->nhc_gw_family == AF_INET)
2999 gw = nhc->nhc_gw.ipv4;
3002 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
3003 "%d\t%08X\t%d\t%u\t%u",
3004 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
3005 prefix, gw, flags, 0, 0,
3009 fi->fib_advmss + 40 : 0),
3014 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
3015 "%d\t%08X\t%d\t%u\t%u",
3016 prefix, 0, flags, 0, 0, 0,
3025 static const struct seq_operations fib_route_seq_ops = {
3026 .start = fib_route_seq_start,
3027 .next = fib_route_seq_next,
3028 .stop = fib_route_seq_stop,
3029 .show = fib_route_seq_show,
3032 int __net_init fib_proc_init(struct net *net)
3034 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
3035 sizeof(struct fib_trie_iter)))
3038 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
3039 fib_triestat_seq_show, NULL))
3042 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
3043 sizeof(struct fib_route_iter)))
3049 remove_proc_entry("fib_triestat", net->proc_net);
3051 remove_proc_entry("fib_trie", net->proc_net);
3056 void __net_exit fib_proc_exit(struct net *net)
3058 remove_proc_entry("fib_trie", net->proc_net);
3059 remove_proc_entry("fib_triestat", net->proc_net);
3060 remove_proc_entry("route", net->proc_net);
3063 #endif /* CONFIG_PROC_FS */