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/rcupdate_wait.h>
56 #include <linux/skbuff.h>
57 #include <linux/netlink.h>
58 #include <linux/init.h>
59 #include <linux/list.h>
60 #include <linux/slab.h>
61 #include <linux/export.h>
62 #include <linux/vmalloc.h>
63 #include <linux/notifier.h>
64 #include <net/net_namespace.h>
65 #include <net/inet_dscp.h>
67 #include <net/protocol.h>
68 #include <net/route.h>
71 #include <net/ip_fib.h>
72 #include <net/fib_notifier.h>
73 #include <trace/events/fib.h>
74 #include "fib_lookup.h"
76 static int call_fib_entry_notifier(struct notifier_block *nb,
77 enum fib_event_type event_type, u32 dst,
78 int dst_len, struct fib_alias *fa,
79 struct netlink_ext_ack *extack)
81 struct fib_entry_notifier_info info = {
82 .info.extack = extack,
90 return call_fib4_notifier(nb, event_type, &info.info);
93 static int call_fib_entry_notifiers(struct net *net,
94 enum fib_event_type event_type, u32 dst,
95 int dst_len, struct fib_alias *fa,
96 struct netlink_ext_ack *extack)
98 struct fib_entry_notifier_info info = {
99 .info.extack = extack,
107 return call_fib4_notifiers(net, event_type, &info.info);
110 #define MAX_STAT_DEPTH 32
112 #define KEYLENGTH (8*sizeof(t_key))
113 #define KEY_MAX ((t_key)~0)
115 typedef unsigned int t_key;
117 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
118 #define IS_TNODE(n) ((n)->bits)
119 #define IS_LEAF(n) (!(n)->bits)
123 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
124 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
127 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
128 struct hlist_head leaf;
129 /* This array is valid if (pos | bits) > 0 (TNODE) */
130 DECLARE_FLEX_ARRAY(struct key_vector __rcu *, tnode);
136 t_key empty_children; /* KEYLENGTH bits needed */
137 t_key full_children; /* KEYLENGTH bits needed */
138 struct key_vector __rcu *parent;
139 struct key_vector kv[1];
140 #define tn_bits kv[0].bits
143 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
144 #define LEAF_SIZE TNODE_SIZE(1)
146 #ifdef CONFIG_IP_FIB_TRIE_STATS
147 struct trie_use_stats {
149 unsigned int backtrack;
150 unsigned int semantic_match_passed;
151 unsigned int semantic_match_miss;
152 unsigned int null_node_hit;
153 unsigned int resize_node_skipped;
158 unsigned int totdepth;
159 unsigned int maxdepth;
162 unsigned int nullpointers;
163 unsigned int prefixes;
164 unsigned int nodesizes[MAX_STAT_DEPTH];
168 struct key_vector kv[1];
169 #ifdef CONFIG_IP_FIB_TRIE_STATS
170 struct trie_use_stats __percpu *stats;
174 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
175 static unsigned int tnode_free_size;
178 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
179 * especially useful before resizing the root node with PREEMPT_NONE configs;
180 * the value was obtained experimentally, aiming to avoid visible slowdown.
182 unsigned int sysctl_fib_sync_mem = 512 * 1024;
183 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
184 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
186 static struct kmem_cache *fn_alias_kmem __ro_after_init;
187 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
189 static inline struct tnode *tn_info(struct key_vector *kv)
191 return container_of(kv, struct tnode, kv[0]);
194 /* caller must hold RTNL */
195 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
196 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
198 /* caller must hold RCU read lock or RTNL */
199 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
200 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
202 /* wrapper for rcu_assign_pointer */
203 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
206 rcu_assign_pointer(tn_info(n)->parent, tp);
209 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
211 /* This provides us with the number of children in this node, in the case of a
212 * leaf this will return 0 meaning none of the children are accessible.
214 static inline unsigned long child_length(const struct key_vector *tn)
216 return (1ul << tn->bits) & ~(1ul);
219 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
221 static inline unsigned long get_index(t_key key, struct key_vector *kv)
223 unsigned long index = key ^ kv->key;
225 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
228 return index >> kv->pos;
231 /* To understand this stuff, an understanding of keys and all their bits is
232 * necessary. Every node in the trie has a key associated with it, but not
233 * all of the bits in that key are significant.
235 * Consider a node 'n' and its parent 'tp'.
237 * If n is a leaf, every bit in its key is significant. Its presence is
238 * necessitated by path compression, since during a tree traversal (when
239 * searching for a leaf - unless we are doing an insertion) we will completely
240 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
241 * a potentially successful search, that we have indeed been walking the
244 * Note that we can never "miss" the correct key in the tree if present by
245 * following the wrong path. Path compression ensures that segments of the key
246 * that are the same for all keys with a given prefix are skipped, but the
247 * skipped part *is* identical for each node in the subtrie below the skipped
248 * bit! trie_insert() in this implementation takes care of that.
250 * if n is an internal node - a 'tnode' here, the various parts of its key
251 * have many different meanings.
254 * _________________________________________________________________
255 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
256 * -----------------------------------------------------------------
257 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
259 * _________________________________________________________________
260 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
261 * -----------------------------------------------------------------
262 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
269 * First, let's just ignore the bits that come before the parent tp, that is
270 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
271 * point we do not use them for anything.
273 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
274 * index into the parent's child array. That is, they will be used to find
275 * 'n' among tp's children.
277 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
280 * All the bits we have seen so far are significant to the node n. The rest
281 * of the bits are really not needed or indeed known in n->key.
283 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
284 * n's child array, and will of course be different for each child.
286 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
290 static const int halve_threshold = 25;
291 static const int inflate_threshold = 50;
292 static const int halve_threshold_root = 15;
293 static const int inflate_threshold_root = 30;
295 static void __alias_free_mem(struct rcu_head *head)
297 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
298 kmem_cache_free(fn_alias_kmem, fa);
301 static inline void alias_free_mem_rcu(struct fib_alias *fa)
303 call_rcu(&fa->rcu, __alias_free_mem);
306 #define TNODE_VMALLOC_MAX \
307 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
309 static void __node_free_rcu(struct rcu_head *head)
311 struct tnode *n = container_of(head, struct tnode, rcu);
314 kmem_cache_free(trie_leaf_kmem, n);
319 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
321 static struct tnode *tnode_alloc(int bits)
325 /* verify bits is within bounds */
326 if (bits > TNODE_VMALLOC_MAX)
329 /* determine size and verify it is non-zero and didn't overflow */
330 size = TNODE_SIZE(1ul << bits);
332 if (size <= PAGE_SIZE)
333 return kzalloc(size, GFP_KERNEL);
335 return vzalloc(size);
338 static inline void empty_child_inc(struct key_vector *n)
340 tn_info(n)->empty_children++;
342 if (!tn_info(n)->empty_children)
343 tn_info(n)->full_children++;
346 static inline void empty_child_dec(struct key_vector *n)
348 if (!tn_info(n)->empty_children)
349 tn_info(n)->full_children--;
351 tn_info(n)->empty_children--;
354 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
356 struct key_vector *l;
359 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
363 /* initialize key vector */
368 l->slen = fa->fa_slen;
370 /* link leaf to fib alias */
371 INIT_HLIST_HEAD(&l->leaf);
372 hlist_add_head(&fa->fa_list, &l->leaf);
377 static struct key_vector *tnode_new(t_key key, int pos, int bits)
379 unsigned int shift = pos + bits;
380 struct key_vector *tn;
383 /* verify bits and pos their msb bits clear and values are valid */
384 BUG_ON(!bits || (shift > KEYLENGTH));
386 tnode = tnode_alloc(bits);
390 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
391 sizeof(struct key_vector *) << bits);
393 if (bits == KEYLENGTH)
394 tnode->full_children = 1;
396 tnode->empty_children = 1ul << bits;
399 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
407 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
408 * and no bits are skipped. See discussion in dyntree paper p. 6
410 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
412 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
415 /* Add a child at position i overwriting the old value.
416 * Update the value of full_children and empty_children.
418 static void put_child(struct key_vector *tn, unsigned long i,
419 struct key_vector *n)
421 struct key_vector *chi = get_child(tn, i);
424 BUG_ON(i >= child_length(tn));
426 /* update emptyChildren, overflow into fullChildren */
432 /* update fullChildren */
433 wasfull = tnode_full(tn, chi);
434 isfull = tnode_full(tn, n);
436 if (wasfull && !isfull)
437 tn_info(tn)->full_children--;
438 else if (!wasfull && isfull)
439 tn_info(tn)->full_children++;
441 if (n && (tn->slen < n->slen))
444 rcu_assign_pointer(tn->tnode[i], n);
447 static void update_children(struct key_vector *tn)
451 /* update all of the child parent pointers */
452 for (i = child_length(tn); i;) {
453 struct key_vector *inode = get_child(tn, --i);
458 /* Either update the children of a tnode that
459 * already belongs to us or update the child
460 * to point to ourselves.
462 if (node_parent(inode) == tn)
463 update_children(inode);
465 node_set_parent(inode, tn);
469 static inline void put_child_root(struct key_vector *tp, t_key key,
470 struct key_vector *n)
473 rcu_assign_pointer(tp->tnode[0], n);
475 put_child(tp, get_index(key, tp), n);
478 static inline void tnode_free_init(struct key_vector *tn)
480 tn_info(tn)->rcu.next = NULL;
483 static inline void tnode_free_append(struct key_vector *tn,
484 struct key_vector *n)
486 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
487 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
490 static void tnode_free(struct key_vector *tn)
492 struct callback_head *head = &tn_info(tn)->rcu;
496 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
499 tn = container_of(head, struct tnode, rcu)->kv;
502 if (tnode_free_size >= READ_ONCE(sysctl_fib_sync_mem)) {
508 static struct key_vector *replace(struct trie *t,
509 struct key_vector *oldtnode,
510 struct key_vector *tn)
512 struct key_vector *tp = node_parent(oldtnode);
515 /* setup the parent pointer out of and back into this node */
516 NODE_INIT_PARENT(tn, tp);
517 put_child_root(tp, tn->key, tn);
519 /* update all of the child parent pointers */
522 /* all pointers should be clean so we are done */
523 tnode_free(oldtnode);
525 /* resize children now that oldtnode is freed */
526 for (i = child_length(tn); i;) {
527 struct key_vector *inode = get_child(tn, --i);
529 /* resize child node */
530 if (tnode_full(tn, inode))
531 tn = resize(t, inode);
537 static struct key_vector *inflate(struct trie *t,
538 struct key_vector *oldtnode)
540 struct key_vector *tn;
544 pr_debug("In inflate\n");
546 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
550 /* prepare oldtnode to be freed */
551 tnode_free_init(oldtnode);
553 /* Assemble all of the pointers in our cluster, in this case that
554 * represents all of the pointers out of our allocated nodes that
555 * point to existing tnodes and the links between our allocated
558 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
559 struct key_vector *inode = get_child(oldtnode, --i);
560 struct key_vector *node0, *node1;
567 /* A leaf or an internal node with skipped bits */
568 if (!tnode_full(oldtnode, inode)) {
569 put_child(tn, get_index(inode->key, tn), inode);
573 /* drop the node in the old tnode free list */
574 tnode_free_append(oldtnode, inode);
576 /* An internal node with two children */
577 if (inode->bits == 1) {
578 put_child(tn, 2 * i + 1, get_child(inode, 1));
579 put_child(tn, 2 * i, get_child(inode, 0));
583 /* We will replace this node 'inode' with two new
584 * ones, 'node0' and 'node1', each with half of the
585 * original children. The two new nodes will have
586 * a position one bit further down the key and this
587 * means that the "significant" part of their keys
588 * (see the discussion near the top of this file)
589 * will differ by one bit, which will be "0" in
590 * node0's key and "1" in node1's key. Since we are
591 * moving the key position by one step, the bit that
592 * we are moving away from - the bit at position
593 * (tn->pos) - is the one that will differ between
594 * node0 and node1. So... we synthesize that bit in the
597 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
600 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
602 tnode_free_append(tn, node1);
605 tnode_free_append(tn, node0);
607 /* populate child pointers in new nodes */
608 for (k = child_length(inode), j = k / 2; j;) {
609 put_child(node1, --j, get_child(inode, --k));
610 put_child(node0, j, get_child(inode, j));
611 put_child(node1, --j, get_child(inode, --k));
612 put_child(node0, j, get_child(inode, j));
615 /* link new nodes to parent */
616 NODE_INIT_PARENT(node1, tn);
617 NODE_INIT_PARENT(node0, tn);
619 /* link parent to nodes */
620 put_child(tn, 2 * i + 1, node1);
621 put_child(tn, 2 * i, node0);
624 /* setup the parent pointers into and out of this node */
625 return replace(t, oldtnode, tn);
627 /* all pointers should be clean so we are done */
633 static struct key_vector *halve(struct trie *t,
634 struct key_vector *oldtnode)
636 struct key_vector *tn;
639 pr_debug("In halve\n");
641 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
645 /* prepare oldtnode to be freed */
646 tnode_free_init(oldtnode);
648 /* Assemble all of the pointers in our cluster, in this case that
649 * represents all of the pointers out of our allocated nodes that
650 * point to existing tnodes and the links between our allocated
653 for (i = child_length(oldtnode); i;) {
654 struct key_vector *node1 = get_child(oldtnode, --i);
655 struct key_vector *node0 = get_child(oldtnode, --i);
656 struct key_vector *inode;
658 /* At least one of the children is empty */
659 if (!node1 || !node0) {
660 put_child(tn, i / 2, node1 ? : node0);
664 /* Two nonempty children */
665 inode = tnode_new(node0->key, oldtnode->pos, 1);
668 tnode_free_append(tn, inode);
670 /* initialize pointers out of node */
671 put_child(inode, 1, node1);
672 put_child(inode, 0, node0);
673 NODE_INIT_PARENT(inode, tn);
675 /* link parent to node */
676 put_child(tn, i / 2, inode);
679 /* setup the parent pointers into and out of this node */
680 return replace(t, oldtnode, tn);
682 /* all pointers should be clean so we are done */
688 static struct key_vector *collapse(struct trie *t,
689 struct key_vector *oldtnode)
691 struct key_vector *n, *tp;
694 /* scan the tnode looking for that one child that might still exist */
695 for (n = NULL, i = child_length(oldtnode); !n && i;)
696 n = get_child(oldtnode, --i);
698 /* compress one level */
699 tp = node_parent(oldtnode);
700 put_child_root(tp, oldtnode->key, n);
701 node_set_parent(n, tp);
709 static unsigned char update_suffix(struct key_vector *tn)
711 unsigned char slen = tn->pos;
712 unsigned long stride, i;
713 unsigned char slen_max;
715 /* only vector 0 can have a suffix length greater than or equal to
716 * tn->pos + tn->bits, the second highest node will have a suffix
717 * length at most of tn->pos + tn->bits - 1
719 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
721 /* search though the list of children looking for nodes that might
722 * have a suffix greater than the one we currently have. This is
723 * why we start with a stride of 2 since a stride of 1 would
724 * represent the nodes with suffix length equal to tn->pos
726 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
727 struct key_vector *n = get_child(tn, i);
729 if (!n || (n->slen <= slen))
732 /* update stride and slen based on new value */
733 stride <<= (n->slen - slen);
737 /* stop searching if we have hit the maximum possible value */
738 if (slen >= slen_max)
747 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
748 * the Helsinki University of Technology and Matti Tikkanen of Nokia
749 * Telecommunications, page 6:
750 * "A node is doubled if the ratio of non-empty children to all
751 * children in the *doubled* node is at least 'high'."
753 * 'high' in this instance is the variable 'inflate_threshold'. It
754 * is expressed as a percentage, so we multiply it with
755 * child_length() and instead of multiplying by 2 (since the
756 * child array will be doubled by inflate()) and multiplying
757 * the left-hand side by 100 (to handle the percentage thing) we
758 * multiply the left-hand side by 50.
760 * The left-hand side may look a bit weird: child_length(tn)
761 * - tn->empty_children is of course the number of non-null children
762 * in the current node. tn->full_children is the number of "full"
763 * children, that is non-null tnodes with a skip value of 0.
764 * All of those will be doubled in the resulting inflated tnode, so
765 * we just count them one extra time here.
767 * A clearer way to write this would be:
769 * to_be_doubled = tn->full_children;
770 * not_to_be_doubled = child_length(tn) - tn->empty_children -
773 * new_child_length = child_length(tn) * 2;
775 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
777 * if (new_fill_factor >= inflate_threshold)
779 * ...and so on, tho it would mess up the while () loop.
782 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
786 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
787 * inflate_threshold * new_child_length
789 * expand not_to_be_doubled and to_be_doubled, and shorten:
790 * 100 * (child_length(tn) - tn->empty_children +
791 * tn->full_children) >= inflate_threshold * new_child_length
793 * expand new_child_length:
794 * 100 * (child_length(tn) - tn->empty_children +
795 * tn->full_children) >=
796 * inflate_threshold * child_length(tn) * 2
799 * 50 * (tn->full_children + child_length(tn) -
800 * tn->empty_children) >= inflate_threshold *
804 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
806 unsigned long used = child_length(tn);
807 unsigned long threshold = used;
809 /* Keep root node larger */
810 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
811 used -= tn_info(tn)->empty_children;
812 used += tn_info(tn)->full_children;
814 /* if bits == KEYLENGTH then pos = 0, and will fail below */
816 return (used > 1) && tn->pos && ((50 * used) >= threshold);
819 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
821 unsigned long used = child_length(tn);
822 unsigned long threshold = used;
824 /* Keep root node larger */
825 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
826 used -= tn_info(tn)->empty_children;
828 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
830 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
833 static inline bool should_collapse(struct key_vector *tn)
835 unsigned long used = child_length(tn);
837 used -= tn_info(tn)->empty_children;
839 /* account for bits == KEYLENGTH case */
840 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
843 /* One child or none, time to drop us from the trie */
848 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
850 #ifdef CONFIG_IP_FIB_TRIE_STATS
851 struct trie_use_stats __percpu *stats = t->stats;
853 struct key_vector *tp = node_parent(tn);
854 unsigned long cindex = get_index(tn->key, tp);
855 int max_work = MAX_WORK;
857 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
858 tn, inflate_threshold, halve_threshold);
860 /* track the tnode via the pointer from the parent instead of
861 * doing it ourselves. This way we can let RCU fully do its
862 * thing without us interfering
864 BUG_ON(tn != get_child(tp, cindex));
866 /* Double as long as the resulting node has a number of
867 * nonempty nodes that are above the threshold.
869 while (should_inflate(tp, tn) && max_work) {
872 #ifdef CONFIG_IP_FIB_TRIE_STATS
873 this_cpu_inc(stats->resize_node_skipped);
879 tn = get_child(tp, cindex);
882 /* update parent in case inflate failed */
883 tp = node_parent(tn);
885 /* Return if at least one inflate is run */
886 if (max_work != MAX_WORK)
889 /* Halve as long as the number of empty children in this
890 * node is above threshold.
892 while (should_halve(tp, tn) && max_work) {
895 #ifdef CONFIG_IP_FIB_TRIE_STATS
896 this_cpu_inc(stats->resize_node_skipped);
902 tn = get_child(tp, cindex);
905 /* Only one child remains */
906 if (should_collapse(tn))
907 return collapse(t, tn);
909 /* update parent in case halve failed */
910 return node_parent(tn);
913 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
915 unsigned char node_slen = tn->slen;
917 while ((node_slen > tn->pos) && (node_slen > slen)) {
918 slen = update_suffix(tn);
919 if (node_slen == slen)
922 tn = node_parent(tn);
923 node_slen = tn->slen;
927 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
929 while (tn->slen < slen) {
931 tn = node_parent(tn);
935 /* rcu_read_lock needs to be hold by caller from readside */
936 static struct key_vector *fib_find_node(struct trie *t,
937 struct key_vector **tp, u32 key)
939 struct key_vector *pn, *n = t->kv;
940 unsigned long index = 0;
944 n = get_child_rcu(n, index);
949 index = get_cindex(key, n);
951 /* This bit of code is a bit tricky but it combines multiple
952 * checks into a single check. The prefix consists of the
953 * prefix plus zeros for the bits in the cindex. The index
954 * is the difference between the key and this value. From
955 * this we can actually derive several pieces of data.
956 * if (index >= (1ul << bits))
957 * we have a mismatch in skip bits and failed
959 * we know the value is cindex
961 * This check is safe even if bits == KEYLENGTH due to the
962 * fact that we can only allocate a node with 32 bits if a
963 * long is greater than 32 bits.
965 if (index >= (1ul << n->bits)) {
970 /* keep searching until we find a perfect match leaf or NULL */
971 } while (IS_TNODE(n));
978 /* Return the first fib alias matching DSCP with
979 * priority less than or equal to PRIO.
980 * If 'find_first' is set, return the first matching
981 * fib alias, regardless of DSCP and priority.
983 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
984 dscp_t dscp, u32 prio, u32 tb_id,
987 struct fib_alias *fa;
992 hlist_for_each_entry(fa, fah, fa_list) {
993 /* Avoid Sparse warning when using dscp_t in inequalities */
994 u8 __fa_dscp = inet_dscp_to_dsfield(fa->fa_dscp);
995 u8 __dscp = inet_dscp_to_dsfield(dscp);
997 if (fa->fa_slen < slen)
999 if (fa->fa_slen != slen)
1001 if (fa->tb_id > tb_id)
1003 if (fa->tb_id != tb_id)
1007 if (__fa_dscp > __dscp)
1009 if (fa->fa_info->fib_priority >= prio || __fa_dscp < __dscp)
1016 static struct fib_alias *
1017 fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1019 u8 slen = KEYLENGTH - fri->dst_len;
1020 struct key_vector *l, *tp;
1021 struct fib_table *tb;
1022 struct fib_alias *fa;
1025 tb = fib_get_table(net, fri->tb_id);
1029 t = (struct trie *)tb->tb_data;
1030 l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1034 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1035 if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1036 fa->fa_dscp == fri->dscp && fa->fa_info == fri->fi &&
1037 fa->fa_type == fri->type)
1044 void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1046 u8 fib_notify_on_flag_change;
1047 struct fib_alias *fa_match;
1048 struct sk_buff *skb;
1053 fa_match = fib_find_matching_alias(net, fri);
1057 /* These are paired with the WRITE_ONCE() happening in this function.
1058 * The reason is that we are only protected by RCU at this point.
1060 if (READ_ONCE(fa_match->offload) == fri->offload &&
1061 READ_ONCE(fa_match->trap) == fri->trap &&
1062 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1065 WRITE_ONCE(fa_match->offload, fri->offload);
1066 WRITE_ONCE(fa_match->trap, fri->trap);
1068 fib_notify_on_flag_change = READ_ONCE(net->ipv4.sysctl_fib_notify_on_flag_change);
1070 /* 2 means send notifications only if offload_failed was changed. */
1071 if (fib_notify_on_flag_change == 2 &&
1072 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1075 WRITE_ONCE(fa_match->offload_failed, fri->offload_failed);
1077 if (!fib_notify_on_flag_change)
1080 skb = nlmsg_new(fib_nlmsg_size(fa_match->fa_info), GFP_ATOMIC);
1086 err = fib_dump_info(skb, 0, 0, RTM_NEWROUTE, fri, 0);
1088 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
1089 WARN_ON(err == -EMSGSIZE);
1094 rtnl_notify(skb, net, 0, RTNLGRP_IPV4_ROUTE, NULL, GFP_ATOMIC);
1098 rtnl_set_sk_err(net, RTNLGRP_IPV4_ROUTE, err);
1102 EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1104 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1106 while (!IS_TRIE(tn))
1110 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1111 struct fib_alias *new, t_key key)
1113 struct key_vector *n, *l;
1115 l = leaf_new(key, new);
1119 /* retrieve child from parent node */
1120 n = get_child(tp, get_index(key, tp));
1122 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1124 * Add a new tnode here
1125 * first tnode need some special handling
1126 * leaves us in position for handling as case 3
1129 struct key_vector *tn;
1131 tn = tnode_new(key, __fls(key ^ n->key), 1);
1135 /* initialize routes out of node */
1136 NODE_INIT_PARENT(tn, tp);
1137 put_child(tn, get_index(key, tn) ^ 1, n);
1139 /* start adding routes into the node */
1140 put_child_root(tp, key, tn);
1141 node_set_parent(n, tn);
1143 /* parent now has a NULL spot where the leaf can go */
1147 /* Case 3: n is NULL, and will just insert a new leaf */
1148 node_push_suffix(tp, new->fa_slen);
1149 NODE_INIT_PARENT(l, tp);
1150 put_child_root(tp, key, l);
1151 trie_rebalance(t, tp);
1160 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1161 struct key_vector *l, struct fib_alias *new,
1162 struct fib_alias *fa, t_key key)
1165 return fib_insert_node(t, tp, new, key);
1168 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1170 struct fib_alias *last;
1172 hlist_for_each_entry(last, &l->leaf, fa_list) {
1173 if (new->fa_slen < last->fa_slen)
1175 if ((new->fa_slen == last->fa_slen) &&
1176 (new->tb_id > last->tb_id))
1182 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1184 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1187 /* if we added to the tail node then we need to update slen */
1188 if (l->slen < new->fa_slen) {
1189 l->slen = new->fa_slen;
1190 node_push_suffix(tp, new->fa_slen);
1196 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1198 if (plen > KEYLENGTH) {
1199 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1203 if ((plen < KEYLENGTH) && (key << plen)) {
1204 NL_SET_ERR_MSG(extack,
1205 "Invalid prefix for given prefix length");
1212 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1213 struct key_vector *l, struct fib_alias *old);
1215 /* Caller must hold RTNL. */
1216 int fib_table_insert(struct net *net, struct fib_table *tb,
1217 struct fib_config *cfg, struct netlink_ext_ack *extack)
1219 struct trie *t = (struct trie *)tb->tb_data;
1220 struct fib_alias *fa, *new_fa;
1221 struct key_vector *l, *tp;
1222 u16 nlflags = NLM_F_EXCL;
1223 struct fib_info *fi;
1224 u8 plen = cfg->fc_dst_len;
1225 u8 slen = KEYLENGTH - plen;
1230 key = ntohl(cfg->fc_dst);
1232 if (!fib_valid_key_len(key, plen, extack))
1235 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1237 fi = fib_create_info(cfg, extack);
1243 dscp = cfg->fc_dscp;
1244 l = fib_find_node(t, &tp, key);
1245 fa = l ? fib_find_alias(&l->leaf, slen, dscp, fi->fib_priority,
1246 tb->tb_id, false) : NULL;
1248 /* Now fa, if non-NULL, points to the first fib alias
1249 * with the same keys [prefix,dscp,priority], if such key already
1250 * exists or to the node before which we will insert new one.
1252 * If fa is NULL, we will need to allocate a new one and
1253 * insert to the tail of the section matching the suffix length
1257 if (fa && fa->fa_dscp == dscp &&
1258 fa->fa_info->fib_priority == fi->fib_priority) {
1259 struct fib_alias *fa_first, *fa_match;
1262 if (cfg->fc_nlflags & NLM_F_EXCL)
1265 nlflags &= ~NLM_F_EXCL;
1268 * 1. Find exact match for type, scope, fib_info to avoid
1270 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1274 hlist_for_each_entry_from(fa, fa_list) {
1275 if ((fa->fa_slen != slen) ||
1276 (fa->tb_id != tb->tb_id) ||
1277 (fa->fa_dscp != dscp))
1279 if (fa->fa_info->fib_priority != fi->fib_priority)
1281 if (fa->fa_type == cfg->fc_type &&
1282 fa->fa_info == fi) {
1288 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1289 struct fib_info *fi_drop;
1292 nlflags |= NLM_F_REPLACE;
1300 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1304 fi_drop = fa->fa_info;
1305 new_fa->fa_dscp = fa->fa_dscp;
1306 new_fa->fa_info = fi;
1307 new_fa->fa_type = cfg->fc_type;
1308 state = fa->fa_state;
1309 new_fa->fa_state = state & ~FA_S_ACCESSED;
1310 new_fa->fa_slen = fa->fa_slen;
1311 new_fa->tb_id = tb->tb_id;
1312 new_fa->fa_default = -1;
1313 new_fa->offload = 0;
1315 new_fa->offload_failed = 0;
1317 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1319 if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1320 tb->tb_id, true) == new_fa) {
1321 enum fib_event_type fib_event;
1323 fib_event = FIB_EVENT_ENTRY_REPLACE;
1324 err = call_fib_entry_notifiers(net, fib_event,
1328 hlist_replace_rcu(&new_fa->fa_list,
1330 goto out_free_new_fa;
1334 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1335 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1337 alias_free_mem_rcu(fa);
1339 fib_release_info(fi_drop);
1340 if (state & FA_S_ACCESSED)
1341 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1345 /* Error if we find a perfect match which
1346 * uses the same scope, type, and nexthop
1352 if (cfg->fc_nlflags & NLM_F_APPEND)
1353 nlflags |= NLM_F_APPEND;
1358 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1361 nlflags |= NLM_F_CREATE;
1363 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1367 new_fa->fa_info = fi;
1368 new_fa->fa_dscp = dscp;
1369 new_fa->fa_type = cfg->fc_type;
1370 new_fa->fa_state = 0;
1371 new_fa->fa_slen = slen;
1372 new_fa->tb_id = tb->tb_id;
1373 new_fa->fa_default = -1;
1374 new_fa->offload = 0;
1376 new_fa->offload_failed = 0;
1378 /* Insert new entry to the list. */
1379 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1381 goto out_free_new_fa;
1383 /* The alias was already inserted, so the node must exist. */
1384 l = l ? l : fib_find_node(t, &tp, key);
1385 if (WARN_ON_ONCE(!l)) {
1387 goto out_free_new_fa;
1390 if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1392 enum fib_event_type fib_event;
1394 fib_event = FIB_EVENT_ENTRY_REPLACE;
1395 err = call_fib_entry_notifiers(net, fib_event, key, plen,
1398 goto out_remove_new_fa;
1402 tb->tb_num_default++;
1404 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1405 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1406 &cfg->fc_nlinfo, nlflags);
1411 fib_remove_alias(t, tp, l, new_fa);
1413 kmem_cache_free(fn_alias_kmem, new_fa);
1415 fib_release_info(fi);
1420 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1422 t_key prefix = n->key;
1424 return (key ^ prefix) & (prefix | -prefix);
1427 bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1428 const struct flowi4 *flp)
1430 if (nhc->nhc_flags & RTNH_F_DEAD)
1433 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1434 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1435 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1438 if (flp->flowi4_oif && flp->flowi4_oif != nhc->nhc_oif)
1444 /* should be called with rcu_read_lock */
1445 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1446 struct fib_result *res, int fib_flags)
1448 struct trie *t = (struct trie *) tb->tb_data;
1449 #ifdef CONFIG_IP_FIB_TRIE_STATS
1450 struct trie_use_stats __percpu *stats = t->stats;
1452 const t_key key = ntohl(flp->daddr);
1453 struct key_vector *n, *pn;
1454 struct fib_alias *fa;
1455 unsigned long index;
1461 n = get_child_rcu(pn, cindex);
1463 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1467 #ifdef CONFIG_IP_FIB_TRIE_STATS
1468 this_cpu_inc(stats->gets);
1471 /* Step 1: Travel to the longest prefix match in the trie */
1473 index = get_cindex(key, n);
1475 /* This bit of code is a bit tricky but it combines multiple
1476 * checks into a single check. The prefix consists of the
1477 * prefix plus zeros for the "bits" in the prefix. The index
1478 * is the difference between the key and this value. From
1479 * this we can actually derive several pieces of data.
1480 * if (index >= (1ul << bits))
1481 * we have a mismatch in skip bits and failed
1483 * we know the value is cindex
1485 * This check is safe even if bits == KEYLENGTH due to the
1486 * fact that we can only allocate a node with 32 bits if a
1487 * long is greater than 32 bits.
1489 if (index >= (1ul << n->bits))
1492 /* we have found a leaf. Prefixes have already been compared */
1496 /* only record pn and cindex if we are going to be chopping
1497 * bits later. Otherwise we are just wasting cycles.
1499 if (n->slen > n->pos) {
1504 n = get_child_rcu(n, index);
1509 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1511 /* record the pointer where our next node pointer is stored */
1512 struct key_vector __rcu **cptr = n->tnode;
1514 /* This test verifies that none of the bits that differ
1515 * between the key and the prefix exist in the region of
1516 * the lsb and higher in the prefix.
1518 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1521 /* exit out and process leaf */
1522 if (unlikely(IS_LEAF(n)))
1525 /* Don't bother recording parent info. Since we are in
1526 * prefix match mode we will have to come back to wherever
1527 * we started this traversal anyway
1530 while ((n = rcu_dereference(*cptr)) == NULL) {
1532 #ifdef CONFIG_IP_FIB_TRIE_STATS
1534 this_cpu_inc(stats->null_node_hit);
1536 /* If we are at cindex 0 there are no more bits for
1537 * us to strip at this level so we must ascend back
1538 * up one level to see if there are any more bits to
1539 * be stripped there.
1542 t_key pkey = pn->key;
1544 /* If we don't have a parent then there is
1545 * nothing for us to do as we do not have any
1546 * further nodes to parse.
1549 trace_fib_table_lookup(tb->tb_id, flp,
1553 #ifdef CONFIG_IP_FIB_TRIE_STATS
1554 this_cpu_inc(stats->backtrack);
1556 /* Get Child's index */
1557 pn = node_parent_rcu(pn);
1558 cindex = get_index(pkey, pn);
1561 /* strip the least significant bit from the cindex */
1562 cindex &= cindex - 1;
1564 /* grab pointer for next child node */
1565 cptr = &pn->tnode[cindex];
1570 /* this line carries forward the xor from earlier in the function */
1571 index = key ^ n->key;
1573 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1574 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1575 struct fib_info *fi = fa->fa_info;
1576 struct fib_nh_common *nhc;
1579 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1580 if (index >= (1ul << fa->fa_slen))
1584 inet_dscp_to_dsfield(fa->fa_dscp) != flp->flowi4_tos)
1586 /* Paired with WRITE_ONCE() in fib_release_info() */
1587 if (READ_ONCE(fi->fib_dead))
1589 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1591 fib_alias_accessed(fa);
1592 err = fib_props[fa->fa_type].error;
1593 if (unlikely(err < 0)) {
1595 #ifdef CONFIG_IP_FIB_TRIE_STATS
1596 this_cpu_inc(stats->semantic_match_passed);
1598 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1601 if (fi->fib_flags & RTNH_F_DEAD)
1604 if (unlikely(fi->nh)) {
1605 if (nexthop_is_blackhole(fi->nh)) {
1606 err = fib_props[RTN_BLACKHOLE].error;
1610 nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1617 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1618 nhc = fib_info_nhc(fi, nhsel);
1620 if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1623 if (!(fib_flags & FIB_LOOKUP_NOREF))
1624 refcount_inc(&fi->fib_clntref);
1626 res->prefix = htonl(n->key);
1627 res->prefixlen = KEYLENGTH - fa->fa_slen;
1628 res->nh_sel = nhsel;
1630 res->type = fa->fa_type;
1631 res->scope = fi->fib_scope;
1634 res->fa_head = &n->leaf;
1635 #ifdef CONFIG_IP_FIB_TRIE_STATS
1636 this_cpu_inc(stats->semantic_match_passed);
1638 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1644 #ifdef CONFIG_IP_FIB_TRIE_STATS
1645 this_cpu_inc(stats->semantic_match_miss);
1649 EXPORT_SYMBOL_GPL(fib_table_lookup);
1651 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1652 struct key_vector *l, struct fib_alias *old)
1654 /* record the location of the previous list_info entry */
1655 struct hlist_node **pprev = old->fa_list.pprev;
1656 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1658 /* remove the fib_alias from the list */
1659 hlist_del_rcu(&old->fa_list);
1661 /* if we emptied the list this leaf will be freed and we can sort
1662 * out parent suffix lengths as a part of trie_rebalance
1664 if (hlist_empty(&l->leaf)) {
1665 if (tp->slen == l->slen)
1666 node_pull_suffix(tp, tp->pos);
1667 put_child_root(tp, l->key, NULL);
1669 trie_rebalance(t, tp);
1673 /* only access fa if it is pointing at the last valid hlist_node */
1677 /* update the trie with the latest suffix length */
1678 l->slen = fa->fa_slen;
1679 node_pull_suffix(tp, fa->fa_slen);
1682 static void fib_notify_alias_delete(struct net *net, u32 key,
1683 struct hlist_head *fah,
1684 struct fib_alias *fa_to_delete,
1685 struct netlink_ext_ack *extack)
1687 struct fib_alias *fa_next, *fa_to_notify;
1688 u32 tb_id = fa_to_delete->tb_id;
1689 u8 slen = fa_to_delete->fa_slen;
1690 enum fib_event_type fib_event;
1692 /* Do not notify if we do not care about the route. */
1693 if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1696 /* Determine if the route should be replaced by the next route in the
1699 fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1700 struct fib_alias, fa_list);
1701 if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1702 fib_event = FIB_EVENT_ENTRY_REPLACE;
1703 fa_to_notify = fa_next;
1705 fib_event = FIB_EVENT_ENTRY_DEL;
1706 fa_to_notify = fa_to_delete;
1708 call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1709 fa_to_notify, extack);
1712 /* Caller must hold RTNL. */
1713 int fib_table_delete(struct net *net, struct fib_table *tb,
1714 struct fib_config *cfg, struct netlink_ext_ack *extack)
1716 struct trie *t = (struct trie *) tb->tb_data;
1717 struct fib_alias *fa, *fa_to_delete;
1718 struct key_vector *l, *tp;
1719 u8 plen = cfg->fc_dst_len;
1720 u8 slen = KEYLENGTH - plen;
1724 key = ntohl(cfg->fc_dst);
1726 if (!fib_valid_key_len(key, plen, extack))
1729 l = fib_find_node(t, &tp, key);
1733 dscp = cfg->fc_dscp;
1734 fa = fib_find_alias(&l->leaf, slen, dscp, 0, tb->tb_id, false);
1738 pr_debug("Deleting %08x/%d dsfield=0x%02x t=%p\n", key, plen,
1739 inet_dscp_to_dsfield(dscp), t);
1741 fa_to_delete = NULL;
1742 hlist_for_each_entry_from(fa, fa_list) {
1743 struct fib_info *fi = fa->fa_info;
1745 if ((fa->fa_slen != slen) ||
1746 (fa->tb_id != tb->tb_id) ||
1747 (fa->fa_dscp != dscp))
1750 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1751 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1752 fa->fa_info->fib_scope == cfg->fc_scope) &&
1753 (!cfg->fc_prefsrc ||
1754 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1755 (!cfg->fc_protocol ||
1756 fi->fib_protocol == cfg->fc_protocol) &&
1757 fib_nh_match(net, cfg, fi, extack) == 0 &&
1758 fib_metrics_match(cfg, fi)) {
1767 fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1768 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1769 &cfg->fc_nlinfo, 0);
1772 tb->tb_num_default--;
1774 fib_remove_alias(t, tp, l, fa_to_delete);
1776 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1777 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1779 fib_release_info(fa_to_delete->fa_info);
1780 alias_free_mem_rcu(fa_to_delete);
1784 /* Scan for the next leaf starting at the provided key value */
1785 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1787 struct key_vector *pn, *n = *tn;
1788 unsigned long cindex;
1790 /* this loop is meant to try and find the key in the trie */
1792 /* record parent and next child index */
1794 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1796 if (cindex >> pn->bits)
1799 /* descend into the next child */
1800 n = get_child_rcu(pn, cindex++);
1804 /* guarantee forward progress on the keys */
1805 if (IS_LEAF(n) && (n->key >= key))
1807 } while (IS_TNODE(n));
1809 /* this loop will search for the next leaf with a greater key */
1810 while (!IS_TRIE(pn)) {
1811 /* if we exhausted the parent node we will need to climb */
1812 if (cindex >= (1ul << pn->bits)) {
1813 t_key pkey = pn->key;
1815 pn = node_parent_rcu(pn);
1816 cindex = get_index(pkey, pn) + 1;
1820 /* grab the next available node */
1821 n = get_child_rcu(pn, cindex++);
1825 /* no need to compare keys since we bumped the index */
1829 /* Rescan start scanning in new node */
1835 return NULL; /* Root of trie */
1837 /* if we are at the limit for keys just return NULL for the tnode */
1842 static void fib_trie_free(struct fib_table *tb)
1844 struct trie *t = (struct trie *)tb->tb_data;
1845 struct key_vector *pn = t->kv;
1846 unsigned long cindex = 1;
1847 struct hlist_node *tmp;
1848 struct fib_alias *fa;
1850 /* walk trie in reverse order and free everything */
1852 struct key_vector *n;
1855 t_key pkey = pn->key;
1861 pn = node_parent(pn);
1863 /* drop emptied tnode */
1864 put_child_root(pn, n->key, NULL);
1867 cindex = get_index(pkey, pn);
1872 /* grab the next available node */
1873 n = get_child(pn, cindex);
1878 /* record pn and cindex for leaf walking */
1880 cindex = 1ul << n->bits;
1885 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1886 hlist_del_rcu(&fa->fa_list);
1887 alias_free_mem_rcu(fa);
1890 put_child_root(pn, n->key, NULL);
1894 #ifdef CONFIG_IP_FIB_TRIE_STATS
1895 free_percpu(t->stats);
1900 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1902 struct trie *ot = (struct trie *)oldtb->tb_data;
1903 struct key_vector *l, *tp = ot->kv;
1904 struct fib_table *local_tb;
1905 struct fib_alias *fa;
1909 if (oldtb->tb_data == oldtb->__data)
1912 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1916 lt = (struct trie *)local_tb->tb_data;
1918 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1919 struct key_vector *local_l = NULL, *local_tp;
1921 hlist_for_each_entry(fa, &l->leaf, fa_list) {
1922 struct fib_alias *new_fa;
1924 if (local_tb->tb_id != fa->tb_id)
1927 /* clone fa for new local table */
1928 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1932 memcpy(new_fa, fa, sizeof(*fa));
1934 /* insert clone into table */
1936 local_l = fib_find_node(lt, &local_tp, l->key);
1938 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1940 kmem_cache_free(fn_alias_kmem, new_fa);
1945 /* stop loop if key wrapped back to 0 */
1953 fib_trie_free(local_tb);
1958 /* Caller must hold RTNL */
1959 void fib_table_flush_external(struct fib_table *tb)
1961 struct trie *t = (struct trie *)tb->tb_data;
1962 struct key_vector *pn = t->kv;
1963 unsigned long cindex = 1;
1964 struct hlist_node *tmp;
1965 struct fib_alias *fa;
1967 /* walk trie in reverse order */
1969 unsigned char slen = 0;
1970 struct key_vector *n;
1973 t_key pkey = pn->key;
1975 /* cannot resize the trie vector */
1979 /* update the suffix to address pulled leaves */
1980 if (pn->slen > pn->pos)
1983 /* resize completed node */
1985 cindex = get_index(pkey, pn);
1990 /* grab the next available node */
1991 n = get_child(pn, cindex);
1996 /* record pn and cindex for leaf walking */
1998 cindex = 1ul << n->bits;
2003 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2004 /* if alias was cloned to local then we just
2005 * need to remove the local copy from main
2007 if (tb->tb_id != fa->tb_id) {
2008 hlist_del_rcu(&fa->fa_list);
2009 alias_free_mem_rcu(fa);
2013 /* record local slen */
2017 /* update leaf slen */
2020 if (hlist_empty(&n->leaf)) {
2021 put_child_root(pn, n->key, NULL);
2027 /* Caller must hold RTNL. */
2028 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
2030 struct trie *t = (struct trie *)tb->tb_data;
2031 struct nl_info info = { .nl_net = net };
2032 struct key_vector *pn = t->kv;
2033 unsigned long cindex = 1;
2034 struct hlist_node *tmp;
2035 struct fib_alias *fa;
2038 /* walk trie in reverse order */
2040 unsigned char slen = 0;
2041 struct key_vector *n;
2044 t_key pkey = pn->key;
2046 /* cannot resize the trie vector */
2050 /* update the suffix to address pulled leaves */
2051 if (pn->slen > pn->pos)
2054 /* resize completed node */
2056 cindex = get_index(pkey, pn);
2061 /* grab the next available node */
2062 n = get_child(pn, cindex);
2067 /* record pn and cindex for leaf walking */
2069 cindex = 1ul << n->bits;
2074 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2075 struct fib_info *fi = fa->fa_info;
2077 if (!fi || tb->tb_id != fa->tb_id ||
2078 (!(fi->fib_flags & RTNH_F_DEAD) &&
2079 !fib_props[fa->fa_type].error)) {
2084 /* Do not flush error routes if network namespace is
2085 * not being dismantled
2087 if (!flush_all && fib_props[fa->fa_type].error) {
2092 fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2094 if (fi->pfsrc_removed)
2095 rtmsg_fib(RTM_DELROUTE, htonl(n->key), fa,
2096 KEYLENGTH - fa->fa_slen, tb->tb_id, &info, 0);
2097 hlist_del_rcu(&fa->fa_list);
2098 fib_release_info(fa->fa_info);
2099 alias_free_mem_rcu(fa);
2103 /* update leaf slen */
2106 if (hlist_empty(&n->leaf)) {
2107 put_child_root(pn, n->key, NULL);
2112 pr_debug("trie_flush found=%d\n", found);
2116 /* derived from fib_trie_free */
2117 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2118 struct nl_info *info)
2120 struct trie *t = (struct trie *)tb->tb_data;
2121 struct key_vector *pn = t->kv;
2122 unsigned long cindex = 1;
2123 struct fib_alias *fa;
2126 struct key_vector *n;
2129 t_key pkey = pn->key;
2134 pn = node_parent(pn);
2135 cindex = get_index(pkey, pn);
2139 /* grab the next available node */
2140 n = get_child(pn, cindex);
2145 /* record pn and cindex for leaf walking */
2147 cindex = 1ul << n->bits;
2152 hlist_for_each_entry(fa, &n->leaf, fa_list) {
2153 struct fib_info *fi = fa->fa_info;
2155 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2158 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2159 KEYLENGTH - fa->fa_slen, tb->tb_id,
2160 info, NLM_F_REPLACE);
2165 void fib_info_notify_update(struct net *net, struct nl_info *info)
2169 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2170 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2171 struct fib_table *tb;
2173 hlist_for_each_entry_rcu(tb, head, tb_hlist,
2174 lockdep_rtnl_is_held())
2175 __fib_info_notify_update(net, tb, info);
2179 static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2180 struct notifier_block *nb,
2181 struct netlink_ext_ack *extack)
2183 struct fib_alias *fa;
2187 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2188 struct fib_info *fi = fa->fa_info;
2193 /* local and main table can share the same trie,
2194 * so don't notify twice for the same entry.
2196 if (tb->tb_id != fa->tb_id)
2199 if (fa->fa_slen == last_slen)
2202 last_slen = fa->fa_slen;
2203 err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2204 l->key, KEYLENGTH - fa->fa_slen,
2212 static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2213 struct netlink_ext_ack *extack)
2215 struct trie *t = (struct trie *)tb->tb_data;
2216 struct key_vector *l, *tp = t->kv;
2220 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2221 err = fib_leaf_notify(l, tb, nb, extack);
2226 /* stop in case of wrap around */
2233 int fib_notify(struct net *net, struct notifier_block *nb,
2234 struct netlink_ext_ack *extack)
2239 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2240 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2241 struct fib_table *tb;
2243 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2244 err = fib_table_notify(tb, nb, extack);
2252 static void __trie_free_rcu(struct rcu_head *head)
2254 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2255 #ifdef CONFIG_IP_FIB_TRIE_STATS
2256 struct trie *t = (struct trie *)tb->tb_data;
2258 if (tb->tb_data == tb->__data)
2259 free_percpu(t->stats);
2260 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2264 void fib_free_table(struct fib_table *tb)
2266 call_rcu(&tb->rcu, __trie_free_rcu);
2269 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2270 struct sk_buff *skb, struct netlink_callback *cb,
2271 struct fib_dump_filter *filter)
2273 unsigned int flags = NLM_F_MULTI;
2274 __be32 xkey = htonl(l->key);
2275 int i, s_i, i_fa, s_fa, err;
2276 struct fib_alias *fa;
2278 if (filter->filter_set ||
2279 !filter->dump_exceptions || !filter->dump_routes)
2280 flags |= NLM_F_DUMP_FILTERED;
2286 /* rcu_read_lock is hold by caller */
2287 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2288 struct fib_info *fi = fa->fa_info;
2295 if (tb->tb_id != fa->tb_id)
2298 if (filter->filter_set) {
2299 if (filter->rt_type && fa->fa_type != filter->rt_type)
2302 if ((filter->protocol &&
2303 fi->fib_protocol != filter->protocol))
2307 !fib_info_nh_uses_dev(fi, filter->dev))
2311 if (filter->dump_routes) {
2313 struct fib_rt_info fri;
2316 fri.tb_id = tb->tb_id;
2318 fri.dst_len = KEYLENGTH - fa->fa_slen;
2319 fri.dscp = fa->fa_dscp;
2320 fri.type = fa->fa_type;
2321 fri.offload = READ_ONCE(fa->offload);
2322 fri.trap = READ_ONCE(fa->trap);
2323 fri.offload_failed = READ_ONCE(fa->offload_failed);
2324 err = fib_dump_info(skb,
2325 NETLINK_CB(cb->skb).portid,
2327 RTM_NEWROUTE, &fri, flags);
2335 if (filter->dump_exceptions) {
2336 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2337 &i_fa, s_fa, flags);
2355 /* rcu_read_lock needs to be hold by caller from readside */
2356 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2357 struct netlink_callback *cb, struct fib_dump_filter *filter)
2359 struct trie *t = (struct trie *)tb->tb_data;
2360 struct key_vector *l, *tp = t->kv;
2361 /* Dump starting at last key.
2362 * Note: 0.0.0.0/0 (ie default) is first key.
2364 int count = cb->args[2];
2365 t_key key = cb->args[3];
2367 /* First time here, count and key are both always 0. Count > 0
2368 * and key == 0 means the dump has wrapped around and we are done.
2373 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2376 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2379 cb->args[2] = count;
2386 memset(&cb->args[4], 0,
2387 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2389 /* stop loop if key wrapped back to 0 */
2395 cb->args[2] = count;
2400 void __init fib_trie_init(void)
2402 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2403 sizeof(struct fib_alias),
2404 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2406 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2408 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2411 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2413 struct fib_table *tb;
2415 size_t sz = sizeof(*tb);
2418 sz += sizeof(struct trie);
2420 tb = kzalloc(sz, GFP_KERNEL);
2425 tb->tb_num_default = 0;
2426 tb->tb_data = (alias ? alias->__data : tb->__data);
2431 t = (struct trie *) tb->tb_data;
2432 t->kv[0].pos = KEYLENGTH;
2433 t->kv[0].slen = KEYLENGTH;
2434 #ifdef CONFIG_IP_FIB_TRIE_STATS
2435 t->stats = alloc_percpu(struct trie_use_stats);
2445 #ifdef CONFIG_PROC_FS
2446 /* Depth first Trie walk iterator */
2447 struct fib_trie_iter {
2448 struct seq_net_private p;
2449 struct fib_table *tb;
2450 struct key_vector *tnode;
2455 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2457 unsigned long cindex = iter->index;
2458 struct key_vector *pn = iter->tnode;
2461 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2462 iter->tnode, iter->index, iter->depth);
2464 while (!IS_TRIE(pn)) {
2465 while (cindex < child_length(pn)) {
2466 struct key_vector *n = get_child_rcu(pn, cindex++);
2473 iter->index = cindex;
2475 /* push down one level */
2484 /* Current node exhausted, pop back up */
2486 pn = node_parent_rcu(pn);
2487 cindex = get_index(pkey, pn) + 1;
2491 /* record root node so further searches know we are done */
2498 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2501 struct key_vector *n, *pn;
2507 n = rcu_dereference(pn->tnode[0]);
2524 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2526 struct key_vector *n;
2527 struct fib_trie_iter iter;
2529 memset(s, 0, sizeof(*s));
2532 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2534 struct fib_alias *fa;
2537 s->totdepth += iter.depth;
2538 if (iter.depth > s->maxdepth)
2539 s->maxdepth = iter.depth;
2541 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2545 if (n->bits < MAX_STAT_DEPTH)
2546 s->nodesizes[n->bits]++;
2547 s->nullpointers += tn_info(n)->empty_children;
2554 * This outputs /proc/net/fib_triestats
2556 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2558 unsigned int i, max, pointers, bytes, avdepth;
2561 avdepth = stat->totdepth*100 / stat->leaves;
2565 seq_printf(seq, "\tAver depth: %u.%02d\n",
2566 avdepth / 100, avdepth % 100);
2567 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2569 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2570 bytes = LEAF_SIZE * stat->leaves;
2572 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2573 bytes += sizeof(struct fib_alias) * stat->prefixes;
2575 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2576 bytes += TNODE_SIZE(0) * stat->tnodes;
2578 max = MAX_STAT_DEPTH;
2579 while (max > 0 && stat->nodesizes[max-1] == 0)
2583 for (i = 1; i < max; i++)
2584 if (stat->nodesizes[i] != 0) {
2585 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2586 pointers += (1<<i) * stat->nodesizes[i];
2588 seq_putc(seq, '\n');
2589 seq_printf(seq, "\tPointers: %u\n", pointers);
2591 bytes += sizeof(struct key_vector *) * pointers;
2592 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2593 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2596 #ifdef CONFIG_IP_FIB_TRIE_STATS
2597 static void trie_show_usage(struct seq_file *seq,
2598 const struct trie_use_stats __percpu *stats)
2600 struct trie_use_stats s = { 0 };
2603 /* loop through all of the CPUs and gather up the stats */
2604 for_each_possible_cpu(cpu) {
2605 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2607 s.gets += pcpu->gets;
2608 s.backtrack += pcpu->backtrack;
2609 s.semantic_match_passed += pcpu->semantic_match_passed;
2610 s.semantic_match_miss += pcpu->semantic_match_miss;
2611 s.null_node_hit += pcpu->null_node_hit;
2612 s.resize_node_skipped += pcpu->resize_node_skipped;
2615 seq_printf(seq, "\nCounters:\n---------\n");
2616 seq_printf(seq, "gets = %u\n", s.gets);
2617 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2618 seq_printf(seq, "semantic match passed = %u\n",
2619 s.semantic_match_passed);
2620 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2621 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2622 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2624 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2626 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2628 if (tb->tb_id == RT_TABLE_LOCAL)
2629 seq_puts(seq, "Local:\n");
2630 else if (tb->tb_id == RT_TABLE_MAIN)
2631 seq_puts(seq, "Main:\n");
2633 seq_printf(seq, "Id %d:\n", tb->tb_id);
2637 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2639 struct net *net = seq->private;
2643 "Basic info: size of leaf:"
2644 " %zd bytes, size of tnode: %zd bytes.\n",
2645 LEAF_SIZE, TNODE_SIZE(0));
2648 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2649 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2650 struct fib_table *tb;
2652 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2653 struct trie *t = (struct trie *) tb->tb_data;
2654 struct trie_stat stat;
2659 fib_table_print(seq, tb);
2661 trie_collect_stats(t, &stat);
2662 trie_show_stats(seq, &stat);
2663 #ifdef CONFIG_IP_FIB_TRIE_STATS
2664 trie_show_usage(seq, t->stats);
2674 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2676 struct fib_trie_iter *iter = seq->private;
2677 struct net *net = seq_file_net(seq);
2681 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2682 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2683 struct fib_table *tb;
2685 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2686 struct key_vector *n;
2688 for (n = fib_trie_get_first(iter,
2689 (struct trie *) tb->tb_data);
2690 n; n = fib_trie_get_next(iter))
2701 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2705 return fib_trie_get_idx(seq, *pos);
2708 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2710 struct fib_trie_iter *iter = seq->private;
2711 struct net *net = seq_file_net(seq);
2712 struct fib_table *tb = iter->tb;
2713 struct hlist_node *tb_node;
2715 struct key_vector *n;
2718 /* next node in same table */
2719 n = fib_trie_get_next(iter);
2723 /* walk rest of this hash chain */
2724 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2725 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2726 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2727 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2732 /* new hash chain */
2733 while (++h < FIB_TABLE_HASHSZ) {
2734 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2735 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2736 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2748 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2754 static void seq_indent(struct seq_file *seq, int n)
2760 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2763 case RT_SCOPE_UNIVERSE: return "universe";
2764 case RT_SCOPE_SITE: return "site";
2765 case RT_SCOPE_LINK: return "link";
2766 case RT_SCOPE_HOST: return "host";
2767 case RT_SCOPE_NOWHERE: return "nowhere";
2769 snprintf(buf, len, "scope=%d", s);
2774 static const char *const rtn_type_names[__RTN_MAX] = {
2775 [RTN_UNSPEC] = "UNSPEC",
2776 [RTN_UNICAST] = "UNICAST",
2777 [RTN_LOCAL] = "LOCAL",
2778 [RTN_BROADCAST] = "BROADCAST",
2779 [RTN_ANYCAST] = "ANYCAST",
2780 [RTN_MULTICAST] = "MULTICAST",
2781 [RTN_BLACKHOLE] = "BLACKHOLE",
2782 [RTN_UNREACHABLE] = "UNREACHABLE",
2783 [RTN_PROHIBIT] = "PROHIBIT",
2784 [RTN_THROW] = "THROW",
2786 [RTN_XRESOLVE] = "XRESOLVE",
2789 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2791 if (t < __RTN_MAX && rtn_type_names[t])
2792 return rtn_type_names[t];
2793 snprintf(buf, len, "type %u", t);
2797 /* Pretty print the trie */
2798 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2800 const struct fib_trie_iter *iter = seq->private;
2801 struct key_vector *n = v;
2803 if (IS_TRIE(node_parent_rcu(n)))
2804 fib_table_print(seq, iter->tb);
2807 __be32 prf = htonl(n->key);
2809 seq_indent(seq, iter->depth-1);
2810 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2811 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2812 tn_info(n)->full_children,
2813 tn_info(n)->empty_children);
2815 __be32 val = htonl(n->key);
2816 struct fib_alias *fa;
2818 seq_indent(seq, iter->depth);
2819 seq_printf(seq, " |-- %pI4\n", &val);
2821 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2822 char buf1[32], buf2[32];
2824 seq_indent(seq, iter->depth + 1);
2825 seq_printf(seq, " /%zu %s %s",
2826 KEYLENGTH - fa->fa_slen,
2827 rtn_scope(buf1, sizeof(buf1),
2828 fa->fa_info->fib_scope),
2829 rtn_type(buf2, sizeof(buf2),
2832 seq_printf(seq, " tos=%d",
2833 inet_dscp_to_dsfield(fa->fa_dscp));
2834 seq_putc(seq, '\n');
2841 static const struct seq_operations fib_trie_seq_ops = {
2842 .start = fib_trie_seq_start,
2843 .next = fib_trie_seq_next,
2844 .stop = fib_trie_seq_stop,
2845 .show = fib_trie_seq_show,
2848 struct fib_route_iter {
2849 struct seq_net_private p;
2850 struct fib_table *main_tb;
2851 struct key_vector *tnode;
2856 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2859 struct key_vector *l, **tp = &iter->tnode;
2862 /* use cached location of previously found key */
2863 if (iter->pos > 0 && pos >= iter->pos) {
2872 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2877 /* handle unlikely case of a key wrap */
2883 iter->key = l->key; /* remember it */
2885 iter->pos = 0; /* forget it */
2890 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2893 struct fib_route_iter *iter = seq->private;
2894 struct fib_table *tb;
2899 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2904 t = (struct trie *)tb->tb_data;
2905 iter->tnode = t->kv;
2908 return fib_route_get_idx(iter, *pos);
2911 iter->key = KEY_MAX;
2913 return SEQ_START_TOKEN;
2916 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2918 struct fib_route_iter *iter = seq->private;
2919 struct key_vector *l = NULL;
2920 t_key key = iter->key + 1;
2924 /* only allow key of 0 for start of sequence */
2925 if ((v == SEQ_START_TOKEN) || key)
2926 l = leaf_walk_rcu(&iter->tnode, key);
2938 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2944 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2946 unsigned int flags = 0;
2948 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2951 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2953 if (nhc->nhc_gw.ipv4)
2954 flags |= RTF_GATEWAY;
2956 if (mask == htonl(0xFFFFFFFF))
2963 * This outputs /proc/net/route.
2964 * The format of the file is not supposed to be changed
2965 * and needs to be same as fib_hash output to avoid breaking
2968 static int fib_route_seq_show(struct seq_file *seq, void *v)
2970 struct fib_route_iter *iter = seq->private;
2971 struct fib_table *tb = iter->main_tb;
2972 struct fib_alias *fa;
2973 struct key_vector *l = v;
2976 if (v == SEQ_START_TOKEN) {
2977 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2978 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2983 prefix = htonl(l->key);
2985 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2986 struct fib_info *fi = fa->fa_info;
2987 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2988 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2990 if ((fa->fa_type == RTN_BROADCAST) ||
2991 (fa->fa_type == RTN_MULTICAST))
2994 if (fa->tb_id != tb->tb_id)
2997 seq_setwidth(seq, 127);
3000 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
3003 if (nhc->nhc_gw_family == AF_INET)
3004 gw = nhc->nhc_gw.ipv4;
3007 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
3008 "%d\t%08X\t%d\t%u\t%u",
3009 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
3010 prefix, gw, flags, 0, 0,
3014 fi->fib_advmss + 40 : 0),
3019 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
3020 "%d\t%08X\t%d\t%u\t%u",
3021 prefix, 0, flags, 0, 0, 0,
3030 static const struct seq_operations fib_route_seq_ops = {
3031 .start = fib_route_seq_start,
3032 .next = fib_route_seq_next,
3033 .stop = fib_route_seq_stop,
3034 .show = fib_route_seq_show,
3037 int __net_init fib_proc_init(struct net *net)
3039 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
3040 sizeof(struct fib_trie_iter)))
3043 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
3044 fib_triestat_seq_show, NULL))
3047 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
3048 sizeof(struct fib_route_iter)))
3054 remove_proc_entry("fib_triestat", net->proc_net);
3056 remove_proc_entry("fib_trie", net->proc_net);
3061 void __net_exit fib_proc_exit(struct net *net)
3063 remove_proc_entry("fib_trie", net->proc_net);
3064 remove_proc_entry("fib_triestat", net->proc_net);
3065 remove_proc_entry("route", net->proc_net);
3068 #endif /* CONFIG_PROC_FS */