~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/ipv4/fib_trie.c

Version: ~ [ linux-5.19-rc3 ] ~ [ linux-5.18.5 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.48 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.123 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.199 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.248 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.284 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.319 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *   This program is free software; you can redistribute it and/or
  3  *   modify it under the terms of the GNU General Public License
  4  *   as published by the Free Software Foundation; either version
  5  *   2 of the License, or (at your option) any later version.
  6  *
  7  *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
  8  *     & Swedish University of Agricultural Sciences.
  9  *
 10  *   Jens Laas <jens.laas@data.slu.se> Swedish University of
 11  *     Agricultural Sciences.
 12  *
 13  *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
 14  *
 15  * This work is based on the LPC-trie which is originally described in:
 16  *
 17  * An experimental study of compression methods for dynamic tries
 18  * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
 19  * http://www.csc.kth.se/~snilsson/software/dyntrie2/
 20  *
 21  *
 22  * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
 23  * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
 24  *
 25  *
 26  * Code from fib_hash has been reused which includes the following header:
 27  *
 28  *
 29  * INET         An implementation of the TCP/IP protocol suite for the LINUX
 30  *              operating system.  INET is implemented using the  BSD Socket
 31  *              interface as the means of communication with the user level.
 32  *
 33  *              IPv4 FIB: lookup engine and maintenance routines.
 34  *
 35  *
 36  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 37  *
 38  *              This program is free software; you can redistribute it and/or
 39  *              modify it under the terms of the GNU General Public License
 40  *              as published by the Free Software Foundation; either version
 41  *              2 of the License, or (at your option) any later version.
 42  *
 43  * Substantial contributions to this work comes from:
 44  *
 45  *              David S. Miller, <davem@davemloft.net>
 46  *              Stephen Hemminger <shemminger@osdl.org>
 47  *              Paul E. McKenney <paulmck@us.ibm.com>
 48  *              Patrick McHardy <kaber@trash.net>
 49  */
 50 
 51 #define VERSION "0.409"
 52 
 53 #include <asm/uaccess.h>
 54 #include <linux/bitops.h>
 55 #include <linux/types.h>
 56 #include <linux/kernel.h>
 57 #include <linux/mm.h>
 58 #include <linux/string.h>
 59 #include <linux/socket.h>
 60 #include <linux/sockios.h>
 61 #include <linux/errno.h>
 62 #include <linux/in.h>
 63 #include <linux/inet.h>
 64 #include <linux/inetdevice.h>
 65 #include <linux/netdevice.h>
 66 #include <linux/if_arp.h>
 67 #include <linux/proc_fs.h>
 68 #include <linux/rcupdate.h>
 69 #include <linux/skbuff.h>
 70 #include <linux/netlink.h>
 71 #include <linux/init.h>
 72 #include <linux/list.h>
 73 #include <linux/slab.h>
 74 #include <linux/export.h>
 75 #include <linux/vmalloc.h>
 76 #include <net/net_namespace.h>
 77 #include <net/ip.h>
 78 #include <net/protocol.h>
 79 #include <net/route.h>
 80 #include <net/tcp.h>
 81 #include <net/sock.h>
 82 #include <net/ip_fib.h>
 83 #include <net/switchdev.h>
 84 #include <trace/events/fib.h>
 85 #include "fib_lookup.h"
 86 
 87 #define MAX_STAT_DEPTH 32
 88 
 89 #define KEYLENGTH       (8*sizeof(t_key))
 90 #define KEY_MAX         ((t_key)~0)
 91 
 92 typedef unsigned int t_key;
 93 
 94 #define IS_TRIE(n)      ((n)->pos >= KEYLENGTH)
 95 #define IS_TNODE(n)     ((n)->bits)
 96 #define IS_LEAF(n)      (!(n)->bits)
 97 
 98 struct key_vector {
 99         t_key key;
100         unsigned char pos;              /* 2log(KEYLENGTH) bits needed */
101         unsigned char bits;             /* 2log(KEYLENGTH) bits needed */
102         unsigned char slen;
103         union {
104                 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
105                 struct hlist_head leaf;
106                 /* This array is valid if (pos | bits) > 0 (TNODE) */
107                 struct key_vector __rcu *tnode[0];
108         };
109 };
110 
111 struct tnode {
112         struct rcu_head rcu;
113         t_key empty_children;           /* KEYLENGTH bits needed */
114         t_key full_children;            /* KEYLENGTH bits needed */
115         struct key_vector __rcu *parent;
116         struct key_vector kv[1];
117 #define tn_bits kv[0].bits
118 };
119 
120 #define TNODE_SIZE(n)   offsetof(struct tnode, kv[0].tnode[n])
121 #define LEAF_SIZE       TNODE_SIZE(1)
122 
123 #ifdef CONFIG_IP_FIB_TRIE_STATS
124 struct trie_use_stats {
125         unsigned int gets;
126         unsigned int backtrack;
127         unsigned int semantic_match_passed;
128         unsigned int semantic_match_miss;
129         unsigned int null_node_hit;
130         unsigned int resize_node_skipped;
131 };
132 #endif
133 
134 struct trie_stat {
135         unsigned int totdepth;
136         unsigned int maxdepth;
137         unsigned int tnodes;
138         unsigned int leaves;
139         unsigned int nullpointers;
140         unsigned int prefixes;
141         unsigned int nodesizes[MAX_STAT_DEPTH];
142 };
143 
144 struct trie {
145         struct key_vector kv[1];
146 #ifdef CONFIG_IP_FIB_TRIE_STATS
147         struct trie_use_stats __percpu *stats;
148 #endif
149 };
150 
151 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
152 static size_t tnode_free_size;
153 
154 /*
155  * synchronize_rcu after call_rcu for that many pages; it should be especially
156  * useful before resizing the root node with PREEMPT_NONE configs; the value was
157  * obtained experimentally, aiming to avoid visible slowdown.
158  */
159 static const int sync_pages = 128;
160 
161 static struct kmem_cache *fn_alias_kmem __read_mostly;
162 static struct kmem_cache *trie_leaf_kmem __read_mostly;
163 
164 static inline struct tnode *tn_info(struct key_vector *kv)
165 {
166         return container_of(kv, struct tnode, kv[0]);
167 }
168 
169 /* caller must hold RTNL */
170 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
171 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
172 
173 /* caller must hold RCU read lock or RTNL */
174 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
175 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
176 
177 /* wrapper for rcu_assign_pointer */
178 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
179 {
180         if (n)
181                 rcu_assign_pointer(tn_info(n)->parent, tp);
182 }
183 
184 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
185 
186 /* This provides us with the number of children in this node, in the case of a
187  * leaf this will return 0 meaning none of the children are accessible.
188  */
189 static inline unsigned long child_length(const struct key_vector *tn)
190 {
191         return (1ul << tn->bits) & ~(1ul);
192 }
193 
194 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
195 
196 static inline unsigned long get_index(t_key key, struct key_vector *kv)
197 {
198         unsigned long index = key ^ kv->key;
199 
200         if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
201                 return 0;
202 
203         return index >> kv->pos;
204 }
205 
206 /* To understand this stuff, an understanding of keys and all their bits is
207  * necessary. Every node in the trie has a key associated with it, but not
208  * all of the bits in that key are significant.
209  *
210  * Consider a node 'n' and its parent 'tp'.
211  *
212  * If n is a leaf, every bit in its key is significant. Its presence is
213  * necessitated by path compression, since during a tree traversal (when
214  * searching for a leaf - unless we are doing an insertion) we will completely
215  * ignore all skipped bits we encounter. Thus we need to verify, at the end of
216  * a potentially successful search, that we have indeed been walking the
217  * correct key path.
218  *
219  * Note that we can never "miss" the correct key in the tree if present by
220  * following the wrong path. Path compression ensures that segments of the key
221  * that are the same for all keys with a given prefix are skipped, but the
222  * skipped part *is* identical for each node in the subtrie below the skipped
223  * bit! trie_insert() in this implementation takes care of that.
224  *
225  * if n is an internal node - a 'tnode' here, the various parts of its key
226  * have many different meanings.
227  *
228  * Example:
229  * _________________________________________________________________
230  * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
231  * -----------------------------------------------------------------
232  *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
233  *
234  * _________________________________________________________________
235  * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
236  * -----------------------------------------------------------------
237  *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
238  *
239  * tp->pos = 22
240  * tp->bits = 3
241  * n->pos = 13
242  * n->bits = 4
243  *
244  * First, let's just ignore the bits that come before the parent tp, that is
245  * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
246  * point we do not use them for anything.
247  *
248  * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
249  * index into the parent's child array. That is, they will be used to find
250  * 'n' among tp's children.
251  *
252  * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
253  * for the node n.
254  *
255  * All the bits we have seen so far are significant to the node n. The rest
256  * of the bits are really not needed or indeed known in n->key.
257  *
258  * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
259  * n's child array, and will of course be different for each child.
260  *
261  * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
262  * at this point.
263  */
264 
265 static const int halve_threshold = 25;
266 static const int inflate_threshold = 50;
267 static const int halve_threshold_root = 15;
268 static const int inflate_threshold_root = 30;
269 
270 static void __alias_free_mem(struct rcu_head *head)
271 {
272         struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
273         kmem_cache_free(fn_alias_kmem, fa);
274 }
275 
276 static inline void alias_free_mem_rcu(struct fib_alias *fa)
277 {
278         call_rcu(&fa->rcu, __alias_free_mem);
279 }
280 
281 #define TNODE_KMALLOC_MAX \
282         ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
283 #define TNODE_VMALLOC_MAX \
284         ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
285 
286 static void __node_free_rcu(struct rcu_head *head)
287 {
288         struct tnode *n = container_of(head, struct tnode, rcu);
289 
290         if (!n->tn_bits)
291                 kmem_cache_free(trie_leaf_kmem, n);
292         else
293                 kvfree(n);
294 }
295 
296 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
297 
298 static struct tnode *tnode_alloc(int bits)
299 {
300         size_t size;
301 
302         /* verify bits is within bounds */
303         if (bits > TNODE_VMALLOC_MAX)
304                 return NULL;
305 
306         /* determine size and verify it is non-zero and didn't overflow */
307         size = TNODE_SIZE(1ul << bits);
308 
309         if (size <= PAGE_SIZE)
310                 return kzalloc(size, GFP_KERNEL);
311         else
312                 return vzalloc(size);
313 }
314 
315 static inline void empty_child_inc(struct key_vector *n)
316 {
317         ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
318 }
319 
320 static inline void empty_child_dec(struct key_vector *n)
321 {
322         tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
323 }
324 
325 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
326 {
327         struct key_vector *l;
328         struct tnode *kv;
329 
330         kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
331         if (!kv)
332                 return NULL;
333 
334         /* initialize key vector */
335         l = kv->kv;
336         l->key = key;
337         l->pos = 0;
338         l->bits = 0;
339         l->slen = fa->fa_slen;
340 
341         /* link leaf to fib alias */
342         INIT_HLIST_HEAD(&l->leaf);
343         hlist_add_head(&fa->fa_list, &l->leaf);
344 
345         return l;
346 }
347 
348 static struct key_vector *tnode_new(t_key key, int pos, int bits)
349 {
350         unsigned int shift = pos + bits;
351         struct key_vector *tn;
352         struct tnode *tnode;
353 
354         /* verify bits and pos their msb bits clear and values are valid */
355         BUG_ON(!bits || (shift > KEYLENGTH));
356 
357         tnode = tnode_alloc(bits);
358         if (!tnode)
359                 return NULL;
360 
361         pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
362                  sizeof(struct key_vector *) << bits);
363 
364         if (bits == KEYLENGTH)
365                 tnode->full_children = 1;
366         else
367                 tnode->empty_children = 1ul << bits;
368 
369         tn = tnode->kv;
370         tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
371         tn->pos = pos;
372         tn->bits = bits;
373         tn->slen = pos;
374 
375         return tn;
376 }
377 
378 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
379  * and no bits are skipped. See discussion in dyntree paper p. 6
380  */
381 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
382 {
383         return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
384 }
385 
386 /* Add a child at position i overwriting the old value.
387  * Update the value of full_children and empty_children.
388  */
389 static void put_child(struct key_vector *tn, unsigned long i,
390                       struct key_vector *n)
391 {
392         struct key_vector *chi = get_child(tn, i);
393         int isfull, wasfull;
394 
395         BUG_ON(i >= child_length(tn));
396 
397         /* update emptyChildren, overflow into fullChildren */
398         if (!n && chi)
399                 empty_child_inc(tn);
400         if (n && !chi)
401                 empty_child_dec(tn);
402 
403         /* update fullChildren */
404         wasfull = tnode_full(tn, chi);
405         isfull = tnode_full(tn, n);
406 
407         if (wasfull && !isfull)
408                 tn_info(tn)->full_children--;
409         else if (!wasfull && isfull)
410                 tn_info(tn)->full_children++;
411 
412         if (n && (tn->slen < n->slen))
413                 tn->slen = n->slen;
414 
415         rcu_assign_pointer(tn->tnode[i], n);
416 }
417 
418 static void update_children(struct key_vector *tn)
419 {
420         unsigned long i;
421 
422         /* update all of the child parent pointers */
423         for (i = child_length(tn); i;) {
424                 struct key_vector *inode = get_child(tn, --i);
425 
426                 if (!inode)
427                         continue;
428 
429                 /* Either update the children of a tnode that
430                  * already belongs to us or update the child
431                  * to point to ourselves.
432                  */
433                 if (node_parent(inode) == tn)
434                         update_children(inode);
435                 else
436                         node_set_parent(inode, tn);
437         }
438 }
439 
440 static inline void put_child_root(struct key_vector *tp, t_key key,
441                                   struct key_vector *n)
442 {
443         if (IS_TRIE(tp))
444                 rcu_assign_pointer(tp->tnode[0], n);
445         else
446                 put_child(tp, get_index(key, tp), n);
447 }
448 
449 static inline void tnode_free_init(struct key_vector *tn)
450 {
451         tn_info(tn)->rcu.next = NULL;
452 }
453 
454 static inline void tnode_free_append(struct key_vector *tn,
455                                      struct key_vector *n)
456 {
457         tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
458         tn_info(tn)->rcu.next = &tn_info(n)->rcu;
459 }
460 
461 static void tnode_free(struct key_vector *tn)
462 {
463         struct callback_head *head = &tn_info(tn)->rcu;
464 
465         while (head) {
466                 head = head->next;
467                 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
468                 node_free(tn);
469 
470                 tn = container_of(head, struct tnode, rcu)->kv;
471         }
472 
473         if (tnode_free_size >= PAGE_SIZE * sync_pages) {
474                 tnode_free_size = 0;
475                 synchronize_rcu();
476         }
477 }
478 
479 static struct key_vector *replace(struct trie *t,
480                                   struct key_vector *oldtnode,
481                                   struct key_vector *tn)
482 {
483         struct key_vector *tp = node_parent(oldtnode);
484         unsigned long i;
485 
486         /* setup the parent pointer out of and back into this node */
487         NODE_INIT_PARENT(tn, tp);
488         put_child_root(tp, tn->key, tn);
489 
490         /* update all of the child parent pointers */
491         update_children(tn);
492 
493         /* all pointers should be clean so we are done */
494         tnode_free(oldtnode);
495 
496         /* resize children now that oldtnode is freed */
497         for (i = child_length(tn); i;) {
498                 struct key_vector *inode = get_child(tn, --i);
499 
500                 /* resize child node */
501                 if (tnode_full(tn, inode))
502                         tn = resize(t, inode);
503         }
504 
505         return tp;
506 }
507 
508 static struct key_vector *inflate(struct trie *t,
509                                   struct key_vector *oldtnode)
510 {
511         struct key_vector *tn;
512         unsigned long i;
513         t_key m;
514 
515         pr_debug("In inflate\n");
516 
517         tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
518         if (!tn)
519                 goto notnode;
520 
521         /* prepare oldtnode to be freed */
522         tnode_free_init(oldtnode);
523 
524         /* Assemble all of the pointers in our cluster, in this case that
525          * represents all of the pointers out of our allocated nodes that
526          * point to existing tnodes and the links between our allocated
527          * nodes.
528          */
529         for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
530                 struct key_vector *inode = get_child(oldtnode, --i);
531                 struct key_vector *node0, *node1;
532                 unsigned long j, k;
533 
534                 /* An empty child */
535                 if (!inode)
536                         continue;
537 
538                 /* A leaf or an internal node with skipped bits */
539                 if (!tnode_full(oldtnode, inode)) {
540                         put_child(tn, get_index(inode->key, tn), inode);
541                         continue;
542                 }
543 
544                 /* drop the node in the old tnode free list */
545                 tnode_free_append(oldtnode, inode);
546 
547                 /* An internal node with two children */
548                 if (inode->bits == 1) {
549                         put_child(tn, 2 * i + 1, get_child(inode, 1));
550                         put_child(tn, 2 * i, get_child(inode, 0));
551                         continue;
552                 }
553 
554                 /* We will replace this node 'inode' with two new
555                  * ones, 'node0' and 'node1', each with half of the
556                  * original children. The two new nodes will have
557                  * a position one bit further down the key and this
558                  * means that the "significant" part of their keys
559                  * (see the discussion near the top of this file)
560                  * will differ by one bit, which will be "" in
561                  * node0's key and "1" in node1's key. Since we are
562                  * moving the key position by one step, the bit that
563                  * we are moving away from - the bit at position
564                  * (tn->pos) - is the one that will differ between
565                  * node0 and node1. So... we synthesize that bit in the
566                  * two new keys.
567                  */
568                 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
569                 if (!node1)
570                         goto nomem;
571                 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
572 
573                 tnode_free_append(tn, node1);
574                 if (!node0)
575                         goto nomem;
576                 tnode_free_append(tn, node0);
577 
578                 /* populate child pointers in new nodes */
579                 for (k = child_length(inode), j = k / 2; j;) {
580                         put_child(node1, --j, get_child(inode, --k));
581                         put_child(node0, j, get_child(inode, j));
582                         put_child(node1, --j, get_child(inode, --k));
583                         put_child(node0, j, get_child(inode, j));
584                 }
585 
586                 /* link new nodes to parent */
587                 NODE_INIT_PARENT(node1, tn);
588                 NODE_INIT_PARENT(node0, tn);
589 
590                 /* link parent to nodes */
591                 put_child(tn, 2 * i + 1, node1);
592                 put_child(tn, 2 * i, node0);
593         }
594 
595         /* setup the parent pointers into and out of this node */
596         return replace(t, oldtnode, tn);
597 nomem:
598         /* all pointers should be clean so we are done */
599         tnode_free(tn);
600 notnode:
601         return NULL;
602 }
603 
604 static struct key_vector *halve(struct trie *t,
605                                 struct key_vector *oldtnode)
606 {
607         struct key_vector *tn;
608         unsigned long i;
609 
610         pr_debug("In halve\n");
611 
612         tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
613         if (!tn)
614                 goto notnode;
615 
616         /* prepare oldtnode to be freed */
617         tnode_free_init(oldtnode);
618 
619         /* Assemble all of the pointers in our cluster, in this case that
620          * represents all of the pointers out of our allocated nodes that
621          * point to existing tnodes and the links between our allocated
622          * nodes.
623          */
624         for (i = child_length(oldtnode); i;) {
625                 struct key_vector *node1 = get_child(oldtnode, --i);
626                 struct key_vector *node0 = get_child(oldtnode, --i);
627                 struct key_vector *inode;
628 
629                 /* At least one of the children is empty */
630                 if (!node1 || !node0) {
631                         put_child(tn, i / 2, node1 ? : node0);
632                         continue;
633                 }
634 
635                 /* Two nonempty children */
636                 inode = tnode_new(node0->key, oldtnode->pos, 1);
637                 if (!inode)
638                         goto nomem;
639                 tnode_free_append(tn, inode);
640 
641                 /* initialize pointers out of node */
642                 put_child(inode, 1, node1);
643                 put_child(inode, 0, node0);
644                 NODE_INIT_PARENT(inode, tn);
645 
646                 /* link parent to node */
647                 put_child(tn, i / 2, inode);
648         }
649 
650         /* setup the parent pointers into and out of this node */
651         return replace(t, oldtnode, tn);
652 nomem:
653         /* all pointers should be clean so we are done */
654         tnode_free(tn);
655 notnode:
656         return NULL;
657 }
658 
659 static struct key_vector *collapse(struct trie *t,
660                                    struct key_vector *oldtnode)
661 {
662         struct key_vector *n, *tp;
663         unsigned long i;
664 
665         /* scan the tnode looking for that one child that might still exist */
666         for (n = NULL, i = child_length(oldtnode); !n && i;)
667                 n = get_child(oldtnode, --i);
668 
669         /* compress one level */
670         tp = node_parent(oldtnode);
671         put_child_root(tp, oldtnode->key, n);
672         node_set_parent(n, tp);
673 
674         /* drop dead node */
675         node_free(oldtnode);
676 
677         return tp;
678 }
679 
680 static unsigned char update_suffix(struct key_vector *tn)
681 {
682         unsigned char slen = tn->pos;
683         unsigned long stride, i;
684         unsigned char slen_max;
685 
686         /* only vector 0 can have a suffix length greater than or equal to
687          * tn->pos + tn->bits, the second highest node will have a suffix
688          * length at most of tn->pos + tn->bits - 1
689          */
690         slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
691 
692         /* search though the list of children looking for nodes that might
693          * have a suffix greater than the one we currently have.  This is
694          * why we start with a stride of 2 since a stride of 1 would
695          * represent the nodes with suffix length equal to tn->pos
696          */
697         for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
698                 struct key_vector *n = get_child(tn, i);
699 
700                 if (!n || (n->slen <= slen))
701                         continue;
702 
703                 /* update stride and slen based on new value */
704                 stride <<= (n->slen - slen);
705                 slen = n->slen;
706                 i &= ~(stride - 1);
707 
708                 /* stop searching if we have hit the maximum possible value */
709                 if (slen >= slen_max)
710                         break;
711         }
712 
713         tn->slen = slen;
714 
715         return slen;
716 }
717 
718 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
719  * the Helsinki University of Technology and Matti Tikkanen of Nokia
720  * Telecommunications, page 6:
721  * "A node is doubled if the ratio of non-empty children to all
722  * children in the *doubled* node is at least 'high'."
723  *
724  * 'high' in this instance is the variable 'inflate_threshold'. It
725  * is expressed as a percentage, so we multiply it with
726  * child_length() and instead of multiplying by 2 (since the
727  * child array will be doubled by inflate()) and multiplying
728  * the left-hand side by 100 (to handle the percentage thing) we
729  * multiply the left-hand side by 50.
730  *
731  * The left-hand side may look a bit weird: child_length(tn)
732  * - tn->empty_children is of course the number of non-null children
733  * in the current node. tn->full_children is the number of "full"
734  * children, that is non-null tnodes with a skip value of 0.
735  * All of those will be doubled in the resulting inflated tnode, so
736  * we just count them one extra time here.
737  *
738  * A clearer way to write this would be:
739  *
740  * to_be_doubled = tn->full_children;
741  * not_to_be_doubled = child_length(tn) - tn->empty_children -
742  *     tn->full_children;
743  *
744  * new_child_length = child_length(tn) * 2;
745  *
746  * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
747  *      new_child_length;
748  * if (new_fill_factor >= inflate_threshold)
749  *
750  * ...and so on, tho it would mess up the while () loop.
751  *
752  * anyway,
753  * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
754  *      inflate_threshold
755  *
756  * avoid a division:
757  * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
758  *      inflate_threshold * new_child_length
759  *
760  * expand not_to_be_doubled and to_be_doubled, and shorten:
761  * 100 * (child_length(tn) - tn->empty_children +
762  *    tn->full_children) >= inflate_threshold * new_child_length
763  *
764  * expand new_child_length:
765  * 100 * (child_length(tn) - tn->empty_children +
766  *    tn->full_children) >=
767  *      inflate_threshold * child_length(tn) * 2
768  *
769  * shorten again:
770  * 50 * (tn->full_children + child_length(tn) -
771  *    tn->empty_children) >= inflate_threshold *
772  *    child_length(tn)
773  *
774  */
775 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
776 {
777         unsigned long used = child_length(tn);
778         unsigned long threshold = used;
779 
780         /* Keep root node larger */
781         threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
782         used -= tn_info(tn)->empty_children;
783         used += tn_info(tn)->full_children;
784 
785         /* if bits == KEYLENGTH then pos = 0, and will fail below */
786 
787         return (used > 1) && tn->pos && ((50 * used) >= threshold);
788 }
789 
790 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
791 {
792         unsigned long used = child_length(tn);
793         unsigned long threshold = used;
794 
795         /* Keep root node larger */
796         threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
797         used -= tn_info(tn)->empty_children;
798 
799         /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
800 
801         return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
802 }
803 
804 static inline bool should_collapse(struct key_vector *tn)
805 {
806         unsigned long used = child_length(tn);
807 
808         used -= tn_info(tn)->empty_children;
809 
810         /* account for bits == KEYLENGTH case */
811         if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
812                 used -= KEY_MAX;
813 
814         /* One child or none, time to drop us from the trie */
815         return used < 2;
816 }
817 
818 #define MAX_WORK 10
819 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
820 {
821 #ifdef CONFIG_IP_FIB_TRIE_STATS
822         struct trie_use_stats __percpu *stats = t->stats;
823 #endif
824         struct key_vector *tp = node_parent(tn);
825         unsigned long cindex = get_index(tn->key, tp);
826         int max_work = MAX_WORK;
827 
828         pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
829                  tn, inflate_threshold, halve_threshold);
830 
831         /* track the tnode via the pointer from the parent instead of
832          * doing it ourselves.  This way we can let RCU fully do its
833          * thing without us interfering
834          */
835         BUG_ON(tn != get_child(tp, cindex));
836 
837         /* Double as long as the resulting node has a number of
838          * nonempty nodes that are above the threshold.
839          */
840         while (should_inflate(tp, tn) && max_work) {
841                 tp = inflate(t, tn);
842                 if (!tp) {
843 #ifdef CONFIG_IP_FIB_TRIE_STATS
844                         this_cpu_inc(stats->resize_node_skipped);
845 #endif
846                         break;
847                 }
848 
849                 max_work--;
850                 tn = get_child(tp, cindex);
851         }
852 
853         /* update parent in case inflate failed */
854         tp = node_parent(tn);
855 
856         /* Return if at least one inflate is run */
857         if (max_work != MAX_WORK)
858                 return tp;
859 
860         /* Halve as long as the number of empty children in this
861          * node is above threshold.
862          */
863         while (should_halve(tp, tn) && max_work) {
864                 tp = halve(t, tn);
865                 if (!tp) {
866 #ifdef CONFIG_IP_FIB_TRIE_STATS
867                         this_cpu_inc(stats->resize_node_skipped);
868 #endif
869                         break;
870                 }
871 
872                 max_work--;
873                 tn = get_child(tp, cindex);
874         }
875 
876         /* Only one child remains */
877         if (should_collapse(tn))
878                 return collapse(t, tn);
879 
880         /* update parent in case halve failed */
881         return node_parent(tn);
882 }
883 
884 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
885 {
886         unsigned char node_slen = tn->slen;
887 
888         while ((node_slen > tn->pos) && (node_slen > slen)) {
889                 slen = update_suffix(tn);
890                 if (node_slen == slen)
891                         break;
892 
893                 tn = node_parent(tn);
894                 node_slen = tn->slen;
895         }
896 }
897 
898 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
899 {
900         while (tn->slen < slen) {
901                 tn->slen = slen;
902                 tn = node_parent(tn);
903         }
904 }
905 
906 /* rcu_read_lock needs to be hold by caller from readside */
907 static struct key_vector *fib_find_node(struct trie *t,
908                                         struct key_vector **tp, u32 key)
909 {
910         struct key_vector *pn, *n = t->kv;
911         unsigned long index = 0;
912 
913         do {
914                 pn = n;
915                 n = get_child_rcu(n, index);
916 
917                 if (!n)
918                         break;
919 
920                 index = get_cindex(key, n);
921 
922                 /* This bit of code is a bit tricky but it combines multiple
923                  * checks into a single check.  The prefix consists of the
924                  * prefix plus zeros for the bits in the cindex. The index
925                  * is the difference between the key and this value.  From
926                  * this we can actually derive several pieces of data.
927                  *   if (index >= (1ul << bits))
928                  *     we have a mismatch in skip bits and failed
929                  *   else
930                  *     we know the value is cindex
931                  *
932                  * This check is safe even if bits == KEYLENGTH due to the
933                  * fact that we can only allocate a node with 32 bits if a
934                  * long is greater than 32 bits.
935                  */
936                 if (index >= (1ul << n->bits)) {
937                         n = NULL;
938                         break;
939                 }
940 
941                 /* keep searching until we find a perfect match leaf or NULL */
942         } while (IS_TNODE(n));
943 
944         *tp = pn;
945 
946         return n;
947 }
948 
949 /* Return the first fib alias matching TOS with
950  * priority less than or equal to PRIO.
951  */
952 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
953                                         u8 tos, u32 prio, u32 tb_id)
954 {
955         struct fib_alias *fa;
956 
957         if (!fah)
958                 return NULL;
959 
960         hlist_for_each_entry(fa, fah, fa_list) {
961                 if (fa->fa_slen < slen)
962                         continue;
963                 if (fa->fa_slen != slen)
964                         break;
965                 if (fa->tb_id > tb_id)
966                         continue;
967                 if (fa->tb_id != tb_id)
968                         break;
969                 if (fa->fa_tos > tos)
970                         continue;
971                 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
972                         return fa;
973         }
974 
975         return NULL;
976 }
977 
978 static void trie_rebalance(struct trie *t, struct key_vector *tn)
979 {
980         while (!IS_TRIE(tn))
981                 tn = resize(t, tn);
982 }
983 
984 static int fib_insert_node(struct trie *t, struct key_vector *tp,
985                            struct fib_alias *new, t_key key)
986 {
987         struct key_vector *n, *l;
988 
989         l = leaf_new(key, new);
990         if (!l)
991                 goto noleaf;
992 
993         /* retrieve child from parent node */
994         n = get_child(tp, get_index(key, tp));
995 
996         /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
997          *
998          *  Add a new tnode here
999          *  first tnode need some special handling
1000          *  leaves us in position for handling as case 3
1001          */
1002         if (n) {
1003                 struct key_vector *tn;
1004 
1005                 tn = tnode_new(key, __fls(key ^ n->key), 1);
1006                 if (!tn)
1007                         goto notnode;
1008 
1009                 /* initialize routes out of node */
1010                 NODE_INIT_PARENT(tn, tp);
1011                 put_child(tn, get_index(key, tn) ^ 1, n);
1012 
1013                 /* start adding routes into the node */
1014                 put_child_root(tp, key, tn);
1015                 node_set_parent(n, tn);
1016 
1017                 /* parent now has a NULL spot where the leaf can go */
1018                 tp = tn;
1019         }
1020 
1021         /* Case 3: n is NULL, and will just insert a new leaf */
1022         node_push_suffix(tp, new->fa_slen);
1023         NODE_INIT_PARENT(l, tp);
1024         put_child_root(tp, key, l);
1025         trie_rebalance(t, tp);
1026 
1027         return 0;
1028 notnode:
1029         node_free(l);
1030 noleaf:
1031         return -ENOMEM;
1032 }
1033 
1034 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1035                             struct key_vector *l, struct fib_alias *new,
1036                             struct fib_alias *fa, t_key key)
1037 {
1038         if (!l)
1039                 return fib_insert_node(t, tp, new, key);
1040 
1041         if (fa) {
1042                 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1043         } else {
1044                 struct fib_alias *last;
1045 
1046                 hlist_for_each_entry(last, &l->leaf, fa_list) {
1047                         if (new->fa_slen < last->fa_slen)
1048                                 break;
1049                         if ((new->fa_slen == last->fa_slen) &&
1050                             (new->tb_id > last->tb_id))
1051                                 break;
1052                         fa = last;
1053                 }
1054 
1055                 if (fa)
1056                         hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1057                 else
1058                         hlist_add_head_rcu(&new->fa_list, &l->leaf);
1059         }
1060 
1061         /* if we added to the tail node then we need to update slen */
1062         if (l->slen < new->fa_slen) {
1063                 l->slen = new->fa_slen;
1064                 node_push_suffix(tp, new->fa_slen);
1065         }
1066 
1067         return 0;
1068 }
1069 
1070 /* Caller must hold RTNL. */
1071 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1072 {
1073         struct trie *t = (struct trie *)tb->tb_data;
1074         struct fib_alias *fa, *new_fa;
1075         struct key_vector *l, *tp;
1076         unsigned int nlflags = 0;
1077         struct fib_info *fi;
1078         u8 plen = cfg->fc_dst_len;
1079         u8 slen = KEYLENGTH - plen;
1080         u8 tos = cfg->fc_tos;
1081         u32 key;
1082         int err;
1083 
1084         if (plen > KEYLENGTH)
1085                 return -EINVAL;
1086 
1087         key = ntohl(cfg->fc_dst);
1088 
1089         pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1090 
1091         if ((plen < KEYLENGTH) && (key << plen))
1092                 return -EINVAL;
1093 
1094         fi = fib_create_info(cfg);
1095         if (IS_ERR(fi)) {
1096                 err = PTR_ERR(fi);
1097                 goto err;
1098         }
1099 
1100         l = fib_find_node(t, &tp, key);
1101         fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1102                                 tb->tb_id) : NULL;
1103 
1104         /* Now fa, if non-NULL, points to the first fib alias
1105          * with the same keys [prefix,tos,priority], if such key already
1106          * exists or to the node before which we will insert new one.
1107          *
1108          * If fa is NULL, we will need to allocate a new one and
1109          * insert to the tail of the section matching the suffix length
1110          * of the new alias.
1111          */
1112 
1113         if (fa && fa->fa_tos == tos &&
1114             fa->fa_info->fib_priority == fi->fib_priority) {
1115                 struct fib_alias *fa_first, *fa_match;
1116 
1117                 err = -EEXIST;
1118                 if (cfg->fc_nlflags & NLM_F_EXCL)
1119                         goto out;
1120 
1121                 /* We have 2 goals:
1122                  * 1. Find exact match for type, scope, fib_info to avoid
1123                  * duplicate routes
1124                  * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1125                  */
1126                 fa_match = NULL;
1127                 fa_first = fa;
1128                 hlist_for_each_entry_from(fa, fa_list) {
1129                         if ((fa->fa_slen != slen) ||
1130                             (fa->tb_id != tb->tb_id) ||
1131                             (fa->fa_tos != tos))
1132                                 break;
1133                         if (fa->fa_info->fib_priority != fi->fib_priority)
1134                                 break;
1135                         if (fa->fa_type == cfg->fc_type &&
1136                             fa->fa_info == fi) {
1137                                 fa_match = fa;
1138                                 break;
1139                         }
1140                 }
1141 
1142                 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1143                         struct fib_info *fi_drop;
1144                         u8 state;
1145 
1146                         fa = fa_first;
1147                         if (fa_match) {
1148                                 if (fa == fa_match)
1149                                         err = 0;
1150                                 goto out;
1151                         }
1152                         err = -ENOBUFS;
1153                         new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1154                         if (!new_fa)
1155                                 goto out;
1156 
1157                         fi_drop = fa->fa_info;
1158                         new_fa->fa_tos = fa->fa_tos;
1159                         new_fa->fa_info = fi;
1160                         new_fa->fa_type = cfg->fc_type;
1161                         state = fa->fa_state;
1162                         new_fa->fa_state = state & ~FA_S_ACCESSED;
1163                         new_fa->fa_slen = fa->fa_slen;
1164                         new_fa->tb_id = tb->tb_id;
1165                         new_fa->fa_default = -1;
1166 
1167                         err = switchdev_fib_ipv4_add(key, plen, fi,
1168                                                      new_fa->fa_tos,
1169                                                      cfg->fc_type,
1170                                                      cfg->fc_nlflags,
1171                                                      tb->tb_id);
1172                         if (err) {
1173                                 switchdev_fib_ipv4_abort(fi);
1174                                 kmem_cache_free(fn_alias_kmem, new_fa);
1175                                 goto out;
1176                         }
1177 
1178                         hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1179 
1180                         alias_free_mem_rcu(fa);
1181 
1182                         fib_release_info(fi_drop);
1183                         if (state & FA_S_ACCESSED)
1184                                 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1185                         rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1186                                 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1187 
1188                         goto succeeded;
1189                 }
1190                 /* Error if we find a perfect match which
1191                  * uses the same scope, type, and nexthop
1192                  * information.
1193                  */
1194                 if (fa_match)
1195                         goto out;
1196 
1197                 if (cfg->fc_nlflags & NLM_F_APPEND)
1198                         nlflags = NLM_F_APPEND;
1199                 else
1200                         fa = fa_first;
1201         }
1202         err = -ENOENT;
1203         if (!(cfg->fc_nlflags & NLM_F_CREATE))
1204                 goto out;
1205 
1206         err = -ENOBUFS;
1207         new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1208         if (!new_fa)
1209                 goto out;
1210 
1211         new_fa->fa_info = fi;
1212         new_fa->fa_tos = tos;
1213         new_fa->fa_type = cfg->fc_type;
1214         new_fa->fa_state = 0;
1215         new_fa->fa_slen = slen;
1216         new_fa->tb_id = tb->tb_id;
1217         new_fa->fa_default = -1;
1218 
1219         /* (Optionally) offload fib entry to switch hardware. */
1220         err = switchdev_fib_ipv4_add(key, plen, fi, tos, cfg->fc_type,
1221                                      cfg->fc_nlflags, tb->tb_id);
1222         if (err) {
1223                 switchdev_fib_ipv4_abort(fi);
1224                 goto out_free_new_fa;
1225         }
1226 
1227         /* Insert new entry to the list. */
1228         err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1229         if (err)
1230                 goto out_sw_fib_del;
1231 
1232         if (!plen)
1233                 tb->tb_num_default++;
1234 
1235         rt_cache_flush(cfg->fc_nlinfo.nl_net);
1236         rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1237                   &cfg->fc_nlinfo, nlflags);
1238 succeeded:
1239         return 0;
1240 
1241 out_sw_fib_del:
1242         switchdev_fib_ipv4_del(key, plen, fi, tos, cfg->fc_type, tb->tb_id);
1243 out_free_new_fa:
1244         kmem_cache_free(fn_alias_kmem, new_fa);
1245 out:
1246         fib_release_info(fi);
1247 err:
1248         return err;
1249 }
1250 
1251 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1252 {
1253         t_key prefix = n->key;
1254 
1255         return (key ^ prefix) & (prefix | -prefix);
1256 }
1257 
1258 /* should be called with rcu_read_lock */
1259 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1260                      struct fib_result *res, int fib_flags)
1261 {
1262         struct trie *t = (struct trie *) tb->tb_data;
1263 #ifdef CONFIG_IP_FIB_TRIE_STATS
1264         struct trie_use_stats __percpu *stats = t->stats;
1265 #endif
1266         const t_key key = ntohl(flp->daddr);
1267         struct key_vector *n, *pn;
1268         struct fib_alias *fa;
1269         unsigned long index;
1270         t_key cindex;
1271 
1272         trace_fib_table_lookup(tb->tb_id, flp);
1273 
1274         pn = t->kv;
1275         cindex = 0;
1276 
1277         n = get_child_rcu(pn, cindex);
1278         if (!n)
1279                 return -EAGAIN;
1280 
1281 #ifdef CONFIG_IP_FIB_TRIE_STATS
1282         this_cpu_inc(stats->gets);
1283 #endif
1284 
1285         /* Step 1: Travel to the longest prefix match in the trie */
1286         for (;;) {
1287                 index = get_cindex(key, n);
1288 
1289                 /* This bit of code is a bit tricky but it combines multiple
1290                  * checks into a single check.  The prefix consists of the
1291                  * prefix plus zeros for the "bits" in the prefix. The index
1292                  * is the difference between the key and this value.  From
1293                  * this we can actually derive several pieces of data.
1294                  *   if (index >= (1ul << bits))
1295                  *     we have a mismatch in skip bits and failed
1296                  *   else
1297                  *     we know the value is cindex
1298                  *
1299                  * This check is safe even if bits == KEYLENGTH due to the
1300                  * fact that we can only allocate a node with 32 bits if a
1301                  * long is greater than 32 bits.
1302                  */
1303                 if (index >= (1ul << n->bits))
1304                         break;
1305 
1306                 /* we have found a leaf. Prefixes have already been compared */
1307                 if (IS_LEAF(n))
1308                         goto found;
1309 
1310                 /* only record pn and cindex if we are going to be chopping
1311                  * bits later.  Otherwise we are just wasting cycles.
1312                  */
1313                 if (n->slen > n->pos) {
1314                         pn = n;
1315                         cindex = index;
1316                 }
1317 
1318                 n = get_child_rcu(n, index);
1319                 if (unlikely(!n))
1320                         goto backtrace;
1321         }
1322 
1323         /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1324         for (;;) {
1325                 /* record the pointer where our next node pointer is stored */
1326                 struct key_vector __rcu **cptr = n->tnode;
1327 
1328                 /* This test verifies that none of the bits that differ
1329                  * between the key and the prefix exist in the region of
1330                  * the lsb and higher in the prefix.
1331                  */
1332                 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1333                         goto backtrace;
1334 
1335                 /* exit out and process leaf */
1336                 if (unlikely(IS_LEAF(n)))
1337                         break;
1338 
1339                 /* Don't bother recording parent info.  Since we are in
1340                  * prefix match mode we will have to come back to wherever
1341                  * we started this traversal anyway
1342                  */
1343 
1344                 while ((n = rcu_dereference(*cptr)) == NULL) {
1345 backtrace:
1346 #ifdef CONFIG_IP_FIB_TRIE_STATS
1347                         if (!n)
1348                                 this_cpu_inc(stats->null_node_hit);
1349 #endif
1350                         /* If we are at cindex 0 there are no more bits for
1351                          * us to strip at this level so we must ascend back
1352                          * up one level to see if there are any more bits to
1353                          * be stripped there.
1354                          */
1355                         while (!cindex) {
1356                                 t_key pkey = pn->key;
1357 
1358                                 /* If we don't have a parent then there is
1359                                  * nothing for us to do as we do not have any
1360                                  * further nodes to parse.
1361                                  */
1362                                 if (IS_TRIE(pn))
1363                                         return -EAGAIN;
1364 #ifdef CONFIG_IP_FIB_TRIE_STATS
1365                                 this_cpu_inc(stats->backtrack);
1366 #endif
1367                                 /* Get Child's index */
1368                                 pn = node_parent_rcu(pn);
1369                                 cindex = get_index(pkey, pn);
1370                         }
1371 
1372                         /* strip the least significant bit from the cindex */
1373                         cindex &= cindex - 1;
1374 
1375                         /* grab pointer for next child node */
1376                         cptr = &pn->tnode[cindex];
1377                 }
1378         }
1379 
1380 found:
1381         /* this line carries forward the xor from earlier in the function */
1382         index = key ^ n->key;
1383 
1384         /* Step 3: Process the leaf, if that fails fall back to backtracing */
1385         hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1386                 struct fib_info *fi = fa->fa_info;
1387                 int nhsel, err;
1388 
1389                 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1390                         if (index >= (1ul << fa->fa_slen))
1391                                 continue;
1392                 }
1393                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1394                         continue;
1395                 if (fi->fib_dead)
1396                         continue;
1397                 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1398                         continue;
1399                 fib_alias_accessed(fa);
1400                 err = fib_props[fa->fa_type].error;
1401                 if (unlikely(err < 0)) {
1402 #ifdef CONFIG_IP_FIB_TRIE_STATS
1403                         this_cpu_inc(stats->semantic_match_passed);
1404 #endif
1405                         return err;
1406                 }
1407                 if (fi->fib_flags & RTNH_F_DEAD)
1408                         continue;
1409                 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1410                         const struct fib_nh *nh = &fi->fib_nh[nhsel];
1411                         struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1412 
1413                         if (nh->nh_flags & RTNH_F_DEAD)
1414                                 continue;
1415                         if (in_dev &&
1416                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1417                             nh->nh_flags & RTNH_F_LINKDOWN &&
1418                             !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1419                                 continue;
1420                         if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1421                                 if (flp->flowi4_oif &&
1422                                     flp->flowi4_oif != nh->nh_oif)
1423                                         continue;
1424                         }
1425 
1426                         if (!(fib_flags & FIB_LOOKUP_NOREF))
1427                                 atomic_inc(&fi->fib_clntref);
1428 
1429                         res->prefixlen = KEYLENGTH - fa->fa_slen;
1430                         res->nh_sel = nhsel;
1431                         res->type = fa->fa_type;
1432                         res->scope = fi->fib_scope;
1433                         res->fi = fi;
1434                         res->table = tb;
1435                         res->fa_head = &n->leaf;
1436 #ifdef CONFIG_IP_FIB_TRIE_STATS
1437                         this_cpu_inc(stats->semantic_match_passed);
1438 #endif
1439                         trace_fib_table_lookup_nh(nh);
1440 
1441                         return err;
1442                 }
1443         }
1444 #ifdef CONFIG_IP_FIB_TRIE_STATS
1445         this_cpu_inc(stats->semantic_match_miss);
1446 #endif
1447         goto backtrace;
1448 }
1449 EXPORT_SYMBOL_GPL(fib_table_lookup);
1450 
1451 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1452                              struct key_vector *l, struct fib_alias *old)
1453 {
1454         /* record the location of the previous list_info entry */
1455         struct hlist_node **pprev = old->fa_list.pprev;
1456         struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1457 
1458         /* remove the fib_alias from the list */
1459         hlist_del_rcu(&old->fa_list);
1460 
1461         /* if we emptied the list this leaf will be freed and we can sort
1462          * out parent suffix lengths as a part of trie_rebalance
1463          */
1464         if (hlist_empty(&l->leaf)) {
1465                 if (tp->slen == l->slen)
1466                         node_pull_suffix(tp, tp->pos);
1467                 put_child_root(tp, l->key, NULL);
1468                 node_free(l);
1469                 trie_rebalance(t, tp);
1470                 return;
1471         }
1472 
1473         /* only access fa if it is pointing at the last valid hlist_node */
1474         if (*pprev)
1475                 return;
1476 
1477         /* update the trie with the latest suffix length */
1478         l->slen = fa->fa_slen;
1479         node_pull_suffix(tp, fa->fa_slen);
1480 }
1481 
1482 /* Caller must hold RTNL. */
1483 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1484 {
1485         struct trie *t = (struct trie *) tb->tb_data;
1486         struct fib_alias *fa, *fa_to_delete;
1487         struct key_vector *l, *tp;
1488         u8 plen = cfg->fc_dst_len;
1489         u8 slen = KEYLENGTH - plen;
1490         u8 tos = cfg->fc_tos;
1491         u32 key;
1492 
1493         if (plen > KEYLENGTH)
1494                 return -EINVAL;
1495 
1496         key = ntohl(cfg->fc_dst);
1497 
1498         if ((plen < KEYLENGTH) && (key << plen))
1499                 return -EINVAL;
1500 
1501         l = fib_find_node(t, &tp, key);
1502         if (!l)
1503                 return -ESRCH;
1504 
1505         fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1506         if (!fa)
1507                 return -ESRCH;
1508 
1509         pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1510 
1511         fa_to_delete = NULL;
1512         hlist_for_each_entry_from(fa, fa_list) {
1513                 struct fib_info *fi = fa->fa_info;
1514 
1515                 if ((fa->fa_slen != slen) ||
1516                     (fa->tb_id != tb->tb_id) ||
1517                     (fa->fa_tos != tos))
1518                         break;
1519 
1520                 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1521                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1522                      fa->fa_info->fib_scope == cfg->fc_scope) &&
1523                     (!cfg->fc_prefsrc ||
1524                      fi->fib_prefsrc == cfg->fc_prefsrc) &&
1525                     (!cfg->fc_protocol ||
1526                      fi->fib_protocol == cfg->fc_protocol) &&
1527                     fib_nh_match(cfg, fi) == 0) {
1528                         fa_to_delete = fa;
1529                         break;
1530                 }
1531         }
1532 
1533         if (!fa_to_delete)
1534                 return -ESRCH;
1535 
1536         switchdev_fib_ipv4_del(key, plen, fa_to_delete->fa_info, tos,
1537                                cfg->fc_type, tb->tb_id);
1538 
1539         rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1540                   &cfg->fc_nlinfo, 0);
1541 
1542         if (!plen)
1543                 tb->tb_num_default--;
1544 
1545         fib_remove_alias(t, tp, l, fa_to_delete);
1546 
1547         if (fa_to_delete->fa_state & FA_S_ACCESSED)
1548                 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1549 
1550         fib_release_info(fa_to_delete->fa_info);
1551         alias_free_mem_rcu(fa_to_delete);
1552         return 0;
1553 }
1554 
1555 /* Scan for the next leaf starting at the provided key value */
1556 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1557 {
1558         struct key_vector *pn, *n = *tn;
1559         unsigned long cindex;
1560 
1561         /* this loop is meant to try and find the key in the trie */
1562         do {
1563                 /* record parent and next child index */
1564                 pn = n;
1565                 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1566 
1567                 if (cindex >> pn->bits)
1568                         break;
1569 
1570                 /* descend into the next child */
1571                 n = get_child_rcu(pn, cindex++);
1572                 if (!n)
1573                         break;
1574 
1575                 /* guarantee forward progress on the keys */
1576                 if (IS_LEAF(n) && (n->key >= key))
1577                         goto found;
1578         } while (IS_TNODE(n));
1579 
1580         /* this loop will search for the next leaf with a greater key */
1581         while (!IS_TRIE(pn)) {
1582                 /* if we exhausted the parent node we will need to climb */
1583                 if (cindex >= (1ul << pn->bits)) {
1584                         t_key pkey = pn->key;
1585 
1586                         pn = node_parent_rcu(pn);
1587                         cindex = get_index(pkey, pn) + 1;
1588                         continue;
1589                 }
1590 
1591                 /* grab the next available node */
1592                 n = get_child_rcu(pn, cindex++);
1593                 if (!n)
1594                         continue;
1595 
1596                 /* no need to compare keys since we bumped the index */
1597                 if (IS_LEAF(n))
1598                         goto found;
1599 
1600                 /* Rescan start scanning in new node */
1601                 pn = n;
1602                 cindex = 0;
1603         }
1604 
1605         *tn = pn;
1606         return NULL; /* Root of trie */
1607 found:
1608         /* if we are at the limit for keys just return NULL for the tnode */
1609         *tn = pn;
1610         return n;
1611 }
1612 
1613 static void fib_trie_free(struct fib_table *tb)
1614 {
1615         struct trie *t = (struct trie *)tb->tb_data;
1616         struct key_vector *pn = t->kv;
1617         unsigned long cindex = 1;
1618         struct hlist_node *tmp;
1619         struct fib_alias *fa;
1620 
1621         /* walk trie in reverse order and free everything */
1622         for (;;) {
1623                 struct key_vector *n;
1624 
1625                 if (!(cindex--)) {
1626                         t_key pkey = pn->key;
1627 
1628                         if (IS_TRIE(pn))
1629                                 break;
1630 
1631                         n = pn;
1632                         pn = node_parent(pn);
1633 
1634                         /* drop emptied tnode */
1635                         put_child_root(pn, n->key, NULL);
1636                         node_free(n);
1637 
1638                         cindex = get_index(pkey, pn);
1639 
1640                         continue;
1641                 }
1642 
1643                 /* grab the next available node */
1644                 n = get_child(pn, cindex);
1645                 if (!n)
1646                         continue;
1647 
1648                 if (IS_TNODE(n)) {
1649                         /* record pn and cindex for leaf walking */
1650                         pn = n;
1651                         cindex = 1ul << n->bits;
1652 
1653                         continue;
1654                 }
1655 
1656                 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1657                         hlist_del_rcu(&fa->fa_list);
1658                         alias_free_mem_rcu(fa);
1659                 }
1660 
1661                 put_child_root(pn, n->key, NULL);
1662                 node_free(n);
1663         }
1664 
1665 #ifdef CONFIG_IP_FIB_TRIE_STATS
1666         free_percpu(t->stats);
1667 #endif
1668         kfree(tb);
1669 }
1670 
1671 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1672 {
1673         struct trie *ot = (struct trie *)oldtb->tb_data;
1674         struct key_vector *l, *tp = ot->kv;
1675         struct fib_table *local_tb;
1676         struct fib_alias *fa;
1677         struct trie *lt;
1678         t_key key = 0;
1679 
1680         if (oldtb->tb_data == oldtb->__data)
1681                 return oldtb;
1682 
1683         local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1684         if (!local_tb)
1685                 return NULL;
1686 
1687         lt = (struct trie *)local_tb->tb_data;
1688 
1689         while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1690                 struct key_vector *local_l = NULL, *local_tp;
1691 
1692                 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1693                         struct fib_alias *new_fa;
1694 
1695                         if (local_tb->tb_id != fa->tb_id)
1696                                 continue;
1697 
1698                         /* clone fa for new local table */
1699                         new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1700                         if (!new_fa)
1701                                 goto out;
1702 
1703                         memcpy(new_fa, fa, sizeof(*fa));
1704 
1705                         /* insert clone into table */
1706                         if (!local_l)
1707                                 local_l = fib_find_node(lt, &local_tp, l->key);
1708 
1709                         if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1710                                              NULL, l->key)) {
1711                                 kmem_cache_free(fn_alias_kmem, new_fa);
1712                                 goto out;
1713                         }
1714                 }
1715 
1716                 /* stop loop if key wrapped back to 0 */
1717                 key = l->key + 1;
1718                 if (key < l->key)
1719                         break;
1720         }
1721 
1722         return local_tb;
1723 out:
1724         fib_trie_free(local_tb);
1725 
1726         return NULL;
1727 }
1728 
1729 /* Caller must hold RTNL */
1730 void fib_table_flush_external(struct fib_table *tb)
1731 {
1732         struct trie *t = (struct trie *)tb->tb_data;
1733         struct key_vector *pn = t->kv;
1734         unsigned long cindex = 1;
1735         struct hlist_node *tmp;
1736         struct fib_alias *fa;
1737 
1738         /* walk trie in reverse order */
1739         for (;;) {
1740                 unsigned char slen = 0;
1741                 struct key_vector *n;
1742 
1743                 if (!(cindex--)) {
1744                         t_key pkey = pn->key;
1745 
1746                         /* cannot resize the trie vector */
1747                         if (IS_TRIE(pn))
1748                                 break;
1749 
1750                         /* update the suffix to address pulled leaves */
1751                         if (pn->slen > pn->pos)
1752                                 update_suffix(pn);
1753 
1754                         /* resize completed node */
1755                         pn = resize(t, pn);
1756                         cindex = get_index(pkey, pn);
1757 
1758                         continue;
1759                 }
1760 
1761                 /* grab the next available node */
1762                 n = get_child(pn, cindex);
1763                 if (!n)
1764                         continue;
1765 
1766                 if (IS_TNODE(n)) {
1767                         /* record pn and cindex for leaf walking */
1768                         pn = n;
1769                         cindex = 1ul << n->bits;
1770 
1771                         continue;
1772                 }
1773 
1774                 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1775                         struct fib_info *fi = fa->fa_info;
1776 
1777                         /* if alias was cloned to local then we just
1778                          * need to remove the local copy from main
1779                          */
1780                         if (tb->tb_id != fa->tb_id) {
1781                                 hlist_del_rcu(&fa->fa_list);
1782                                 alias_free_mem_rcu(fa);
1783                                 continue;
1784                         }
1785 
1786                         /* record local slen */
1787                         slen = fa->fa_slen;
1788 
1789                         if (!fi || !(fi->fib_flags & RTNH_F_OFFLOAD))
1790                                 continue;
1791 
1792                         switchdev_fib_ipv4_del(n->key, KEYLENGTH - fa->fa_slen,
1793                                                fi, fa->fa_tos, fa->fa_type,
1794                                                tb->tb_id);
1795                 }
1796 
1797                 /* update leaf slen */
1798                 n->slen = slen;
1799 
1800                 if (hlist_empty(&n->leaf)) {
1801                         put_child_root(pn, n->key, NULL);
1802                         node_free(n);
1803                 }
1804         }
1805 }
1806 
1807 /* Caller must hold RTNL. */
1808 int fib_table_flush(struct fib_table *tb)
1809 {
1810         struct trie *t = (struct trie *)tb->tb_data;
1811         struct key_vector *pn = t->kv;
1812         unsigned long cindex = 1;
1813         struct hlist_node *tmp;
1814         struct fib_alias *fa;
1815         int found = 0;
1816 
1817         /* walk trie in reverse order */
1818         for (;;) {
1819                 unsigned char slen = 0;
1820                 struct key_vector *n;
1821 
1822                 if (!(cindex--)) {
1823                         t_key pkey = pn->key;
1824 
1825                         /* cannot resize the trie vector */
1826                         if (IS_TRIE(pn))
1827                                 break;
1828 
1829                         /* update the suffix to address pulled leaves */
1830                         if (pn->slen > pn->pos)
1831                                 update_suffix(pn);
1832 
1833                         /* resize completed node */
1834                         pn = resize(t, pn);
1835                         cindex = get_index(pkey, pn);
1836 
1837                         continue;
1838                 }
1839 
1840                 /* grab the next available node */
1841                 n = get_child(pn, cindex);
1842                 if (!n)
1843                         continue;
1844 
1845                 if (IS_TNODE(n)) {
1846                         /* record pn and cindex for leaf walking */
1847                         pn = n;
1848                         cindex = 1ul << n->bits;
1849 
1850                         continue;
1851                 }
1852 
1853                 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1854                         struct fib_info *fi = fa->fa_info;
1855 
1856                         if (!fi || !(fi->fib_flags & RTNH_F_DEAD)) {
1857                                 slen = fa->fa_slen;
1858                                 continue;
1859                         }
1860 
1861                         switchdev_fib_ipv4_del(n->key, KEYLENGTH - fa->fa_slen,
1862                                                fi, fa->fa_tos, fa->fa_type,
1863                                                tb->tb_id);
1864                         hlist_del_rcu(&fa->fa_list);
1865                         fib_release_info(fa->fa_info);
1866                         alias_free_mem_rcu(fa);
1867                         found++;
1868                 }
1869 
1870                 /* update leaf slen */
1871                 n->slen = slen;
1872 
1873                 if (hlist_empty(&n->leaf)) {
1874                         put_child_root(pn, n->key, NULL);
1875                         node_free(n);
1876                 }
1877         }
1878 
1879         pr_debug("trie_flush found=%d\n", found);
1880         return found;
1881 }
1882 
1883 static void __trie_free_rcu(struct rcu_head *head)
1884 {
1885         struct fib_table *tb = container_of(head, struct fib_table, rcu);
1886 #ifdef CONFIG_IP_FIB_TRIE_STATS
1887         struct trie *t = (struct trie *)tb->tb_data;
1888 
1889         if (tb->tb_data == tb->__data)
1890                 free_percpu(t->stats);
1891 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1892         kfree(tb);
1893 }
1894 
1895 void fib_free_table(struct fib_table *tb)
1896 {
1897         call_rcu(&tb->rcu, __trie_free_rcu);
1898 }
1899 
1900 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
1901                              struct sk_buff *skb, struct netlink_callback *cb)
1902 {
1903         __be32 xkey = htonl(l->key);
1904         struct fib_alias *fa;
1905         int i, s_i;
1906 
1907         s_i = cb->args[4];
1908         i = 0;
1909 
1910         /* rcu_read_lock is hold by caller */
1911         hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1912                 if (i < s_i) {
1913                         i++;
1914                         continue;
1915                 }
1916 
1917                 if (tb->tb_id != fa->tb_id) {
1918                         i++;
1919                         continue;
1920                 }
1921 
1922                 if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
1923                                   cb->nlh->nlmsg_seq,
1924                                   RTM_NEWROUTE,
1925                                   tb->tb_id,
1926                                   fa->fa_type,
1927                                   xkey,
1928                                   KEYLENGTH - fa->fa_slen,
1929                                   fa->fa_tos,
1930                                   fa->fa_info, NLM_F_MULTI) < 0) {
1931                         cb->args[4] = i;
1932                         return -1;
1933                 }
1934                 i++;
1935         }
1936 
1937         cb->args[4] = i;
1938         return skb->len;
1939 }
1940 
1941 /* rcu_read_lock needs to be hold by caller from readside */
1942 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1943                    struct netlink_callback *cb)
1944 {
1945         struct trie *t = (struct trie *)tb->tb_data;
1946         struct key_vector *l, *tp = t->kv;
1947         /* Dump starting at last key.
1948          * Note: 0.0.0.0/0 (ie default) is first key.
1949          */
1950         int count = cb->args[2];
1951         t_key key = cb->args[3];
1952 
1953         while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1954                 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1955                         cb->args[3] = key;
1956                         cb->args[2] = count;
1957                         return -1;
1958                 }
1959 
1960                 ++count;
1961                 key = l->key + 1;
1962 
1963                 memset(&cb->args[4], 0,
1964                        sizeof(cb->args) - 4*sizeof(cb->args[0]));
1965 
1966                 /* stop loop if key wrapped back to 0 */
1967                 if (key < l->key)
1968                         break;
1969         }
1970 
1971         cb->args[3] = key;
1972         cb->args[2] = count;
1973 
1974         return skb->len;
1975 }
1976 
1977 void __init fib_trie_init(void)
1978 {
1979         fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1980                                           sizeof(struct fib_alias),
1981                                           0, SLAB_PANIC, NULL);
1982 
1983         trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1984                                            LEAF_SIZE,
1985                                            0, SLAB_PANIC, NULL);
1986 }
1987 
1988 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
1989 {
1990         struct fib_table *tb;
1991         struct trie *t;
1992         size_t sz = sizeof(*tb);
1993 
1994         if (!alias)
1995                 sz += sizeof(struct trie);
1996 
1997         tb = kzalloc(sz, GFP_KERNEL);
1998         if (!tb)
1999                 return NULL;
2000 
2001         tb->tb_id = id;
2002         tb->tb_num_default = 0;
2003         tb->tb_data = (alias ? alias->__data : tb->__data);
2004 
2005         if (alias)
2006                 return tb;
2007 
2008         t = (struct trie *) tb->tb_data;
2009         t->kv[0].pos = KEYLENGTH;
2010         t->kv[0].slen = KEYLENGTH;
2011 #ifdef CONFIG_IP_FIB_TRIE_STATS
2012         t->stats = alloc_percpu(struct trie_use_stats);
2013         if (!t->stats) {
2014                 kfree(tb);
2015                 tb = NULL;
2016         }
2017 #endif
2018 
2019         return tb;
2020 }
2021 
2022 #ifdef CONFIG_PROC_FS
2023 /* Depth first Trie walk iterator */
2024 struct fib_trie_iter {
2025         struct seq_net_private p;
2026         struct fib_table *tb;
2027         struct key_vector *tnode;
2028         unsigned int index;
2029         unsigned int depth;
2030 };
2031 
2032 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2033 {
2034         unsigned long cindex = iter->index;
2035         struct key_vector *pn = iter->tnode;
2036         t_key pkey;
2037 
2038         pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2039                  iter->tnode, iter->index, iter->depth);
2040 
2041         while (!IS_TRIE(pn)) {
2042                 while (cindex < child_length(pn)) {
2043                         struct key_vector *n = get_child_rcu(pn, cindex++);
2044 
2045                         if (!n)
2046                                 continue;
2047 
2048                         if (IS_LEAF(n)) {
2049                                 iter->tnode = pn;
2050                                 iter->index = cindex;
2051                         } else {
2052                                 /* push down one level */
2053                                 iter->tnode = n;
2054                                 iter->index = 0;
2055                                 ++iter->depth;
2056                         }
2057 
2058                         return n;
2059                 }
2060 
2061                 /* Current node exhausted, pop back up */
2062                 pkey = pn->key;
2063                 pn = node_parent_rcu(pn);
2064                 cindex = get_index(pkey, pn) + 1;
2065                 --iter->depth;
2066         }
2067 
2068         /* record root node so further searches know we are done */
2069         iter->tnode = pn;
2070         iter->index = 0;
2071 
2072         return NULL;
2073 }
2074 
2075 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2076                                              struct trie *t)
2077 {
2078         struct key_vector *n, *pn;
2079 
2080         if (!t)
2081                 return NULL;
2082 
2083         pn = t->kv;
2084         n = rcu_dereference(pn->tnode[0]);
2085         if (!n)
2086                 return NULL;
2087 
2088         if (IS_TNODE(n)) {
2089                 iter->tnode = n;
2090                 iter->index = 0;
2091                 iter->depth = 1;
2092         } else {
2093                 iter->tnode = pn;
2094                 iter->index = 0;
2095                 iter->depth = 0;
2096         }
2097 
2098         return n;
2099 }
2100 
2101 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2102 {
2103         struct key_vector *n;
2104         struct fib_trie_iter iter;
2105 
2106         memset(s, 0, sizeof(*s));
2107 
2108         rcu_read_lock();
2109         for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2110                 if (IS_LEAF(n)) {
2111                         struct fib_alias *fa;
2112 
2113                         s->leaves++;
2114                         s->totdepth += iter.depth;
2115                         if (iter.depth > s->maxdepth)
2116                                 s->maxdepth = iter.depth;
2117 
2118                         hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2119                                 ++s->prefixes;
2120                 } else {
2121                         s->tnodes++;
2122                         if (n->bits < MAX_STAT_DEPTH)
2123                                 s->nodesizes[n->bits]++;
2124                         s->nullpointers += tn_info(n)->empty_children;
2125                 }
2126         }
2127         rcu_read_unlock();
2128 }
2129 
2130 /*
2131  *      This outputs /proc/net/fib_triestats
2132  */
2133 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2134 {
2135         unsigned int i, max, pointers, bytes, avdepth;
2136 
2137         if (stat->leaves)
2138                 avdepth = stat->totdepth*100 / stat->leaves;
2139         else
2140                 avdepth = 0;
2141 
2142         seq_printf(seq, "\tAver depth:     %u.%02d\n",
2143                    avdepth / 100, avdepth % 100);
2144         seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2145 
2146         seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2147         bytes = LEAF_SIZE * stat->leaves;
2148 
2149         seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2150         bytes += sizeof(struct fib_alias) * stat->prefixes;
2151 
2152         seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2153         bytes += TNODE_SIZE(0) * stat->tnodes;
2154 
2155         max = MAX_STAT_DEPTH;
2156         while (max > 0 && stat->nodesizes[max-1] == 0)
2157                 max--;
2158 
2159         pointers = 0;
2160         for (i = 1; i < max; i++)
2161                 if (stat->nodesizes[i] != 0) {
2162                         seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2163                         pointers += (1<<i) * stat->nodesizes[i];
2164                 }
2165         seq_putc(seq, '\n');
2166         seq_printf(seq, "\tPointers: %u\n", pointers);
2167 
2168         bytes += sizeof(struct key_vector *) * pointers;
2169         seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2170         seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2171 }
2172 
2173 #ifdef CONFIG_IP_FIB_TRIE_STATS
2174 static void trie_show_usage(struct seq_file *seq,
2175                             const struct trie_use_stats __percpu *stats)
2176 {
2177         struct trie_use_stats s = { 0 };
2178         int cpu;
2179 
2180         /* loop through all of the CPUs and gather up the stats */
2181         for_each_possible_cpu(cpu) {
2182                 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2183 
2184                 s.gets += pcpu->gets;
2185                 s.backtrack += pcpu->backtrack;
2186                 s.semantic_match_passed += pcpu->semantic_match_passed;
2187                 s.semantic_match_miss += pcpu->semantic_match_miss;
2188                 s.null_node_hit += pcpu->null_node_hit;
2189                 s.resize_node_skipped += pcpu->resize_node_skipped;
2190         }
2191 
2192         seq_printf(seq, "\nCounters:\n---------\n");
2193         seq_printf(seq, "gets = %u\n", s.gets);
2194         seq_printf(seq, "backtracks = %u\n", s.backtrack);
2195         seq_printf(seq, "semantic match passed = %u\n",
2196                    s.semantic_match_passed);
2197         seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2198         seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2199         seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2200 }
2201 #endif /*  CONFIG_IP_FIB_TRIE_STATS */
2202 
2203 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2204 {
2205         if (tb->tb_id == RT_TABLE_LOCAL)
2206                 seq_puts(seq, "Local:\n");
2207         else if (tb->tb_id == RT_TABLE_MAIN)
2208                 seq_puts(seq, "Main:\n");
2209         else
2210                 seq_printf(seq, "Id %d:\n", tb->tb_id);
2211 }
2212 
2213 
2214 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2215 {
2216         struct net *net = (struct net *)seq->private;
2217         unsigned int h;
2218 
2219         seq_printf(seq,
2220                    "Basic info: size of leaf:"
2221                    " %Zd bytes, size of tnode: %Zd bytes.\n",
2222                    LEAF_SIZE, TNODE_SIZE(0));
2223 
2224         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2225                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2226                 struct fib_table *tb;
2227 
2228                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2229                         struct trie *t = (struct trie *) tb->tb_data;
2230                         struct trie_stat stat;
2231 
2232                         if (!t)
2233                                 continue;
2234 
2235                         fib_table_print(seq, tb);
2236 
2237                         trie_collect_stats(t, &stat);
2238                         trie_show_stats(seq, &stat);
2239 #ifdef CONFIG_IP_FIB_TRIE_STATS
2240                         trie_show_usage(seq, t->stats);
2241 #endif
2242                 }
2243         }
2244 
2245         return 0;
2246 }
2247 
2248 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2249 {
2250         return single_open_net(inode, file, fib_triestat_seq_show);
2251 }
2252 
2253 static const struct file_operations fib_triestat_fops = {
2254         .owner  = THIS_MODULE,
2255         .open   = fib_triestat_seq_open,
2256         .read   = seq_read,
2257         .llseek = seq_lseek,
2258         .release = single_release_net,
2259 };
2260 
2261 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2262 {
2263         struct fib_trie_iter *iter = seq->private;
2264         struct net *net = seq_file_net(seq);
2265         loff_t idx = 0;
2266         unsigned int h;
2267 
2268         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2269                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2270                 struct fib_table *tb;
2271 
2272                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2273                         struct key_vector *n;
2274 
2275                         for (n = fib_trie_get_first(iter,
2276                                                     (struct trie *) tb->tb_data);
2277                              n; n = fib_trie_get_next(iter))
2278                                 if (pos == idx++) {
2279                                         iter->tb = tb;
2280                                         return n;
2281                                 }
2282                 }
2283         }
2284 
2285         return NULL;
2286 }
2287 
2288 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2289         __acquires(RCU)
2290 {
2291         rcu_read_lock();
2292         return fib_trie_get_idx(seq, *pos);
2293 }
2294 
2295 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2296 {
2297         struct fib_trie_iter *iter = seq->private;
2298         struct net *net = seq_file_net(seq);
2299         struct fib_table *tb = iter->tb;
2300         struct hlist_node *tb_node;
2301         unsigned int h;
2302         struct key_vector *n;
2303 
2304         ++*pos;
2305         /* next node in same table */
2306         n = fib_trie_get_next(iter);
2307         if (n)
2308                 return n;
2309 
2310         /* walk rest of this hash chain */
2311         h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2312         while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2313                 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2314                 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2315                 if (n)
2316                         goto found;
2317         }
2318 
2319         /* new hash chain */
2320         while (++h < FIB_TABLE_HASHSZ) {
2321                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2322                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2323                         n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2324                         if (n)
2325                                 goto found;
2326                 }
2327         }
2328         return NULL;
2329 
2330 found:
2331         iter->tb = tb;
2332         return n;
2333 }
2334 
2335 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2336         __releases(RCU)
2337 {
2338         rcu_read_unlock();
2339 }
2340 
2341 static void seq_indent(struct seq_file *seq, int n)
2342 {
2343         while (n-- > 0)
2344                 seq_puts(seq, "   ");
2345 }
2346 
2347 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2348 {
2349         switch (s) {
2350         case RT_SCOPE_UNIVERSE: return "universe";
2351         case RT_SCOPE_SITE:     return "site";
2352         case RT_SCOPE_LINK:     return "link";
2353         case RT_SCOPE_HOST:     return "host";
2354         case RT_SCOPE_NOWHERE:  return "nowhere";
2355         default:
2356                 snprintf(buf, len, "scope=%d", s);
2357                 return buf;
2358         }
2359 }
2360 
2361 static const char *const rtn_type_names[__RTN_MAX] = {
2362         [RTN_UNSPEC] = "UNSPEC",
2363         [RTN_UNICAST] = "UNICAST",
2364         [RTN_LOCAL] = "LOCAL",
2365         [RTN_BROADCAST] = "BROADCAST",
2366         [RTN_ANYCAST] = "ANYCAST",
2367         [RTN_MULTICAST] = "MULTICAST",
2368         [RTN_BLACKHOLE] = "BLACKHOLE",
2369         [RTN_UNREACHABLE] = "UNREACHABLE",
2370         [RTN_PROHIBIT] = "PROHIBIT",
2371         [RTN_THROW] = "THROW",
2372         [RTN_NAT] = "NAT",
2373         [RTN_XRESOLVE] = "XRESOLVE",
2374 };
2375 
2376 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2377 {
2378         if (t < __RTN_MAX && rtn_type_names[t])
2379                 return rtn_type_names[t];
2380         snprintf(buf, len, "type %u", t);
2381         return buf;
2382 }
2383 
2384 /* Pretty print the trie */
2385 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2386 {
2387         const struct fib_trie_iter *iter = seq->private;
2388         struct key_vector *n = v;
2389 
2390         if (IS_TRIE(node_parent_rcu(n)))
2391                 fib_table_print(seq, iter->tb);
2392 
2393         if (IS_TNODE(n)) {
2394                 __be32 prf = htonl(n->key);
2395 
2396                 seq_indent(seq, iter->depth-1);
2397                 seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2398                            &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2399                            tn_info(n)->full_children,
2400                            tn_info(n)->empty_children);
2401         } else {
2402                 __be32 val = htonl(n->key);
2403                 struct fib_alias *fa;
2404 
2405                 seq_indent(seq, iter->depth);
2406                 seq_printf(seq, "  |-- %pI4\n", &val);
2407 
2408                 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2409                         char buf1[32], buf2[32];
2410 
2411                         seq_indent(seq, iter->depth + 1);
2412                         seq_printf(seq, "  /%zu %s %s",
2413                                    KEYLENGTH - fa->fa_slen,
2414                                    rtn_scope(buf1, sizeof(buf1),
2415                                              fa->fa_info->fib_scope),
2416                                    rtn_type(buf2, sizeof(buf2),
2417                                             fa->fa_type));
2418                         if (fa->fa_tos)
2419                                 seq_printf(seq, " tos=%d", fa->fa_tos);
2420                         seq_putc(seq, '\n');
2421                 }
2422         }
2423 
2424         return 0;
2425 }
2426 
2427 static const struct seq_operations fib_trie_seq_ops = {
2428         .start  = fib_trie_seq_start,
2429         .next   = fib_trie_seq_next,
2430         .stop   = fib_trie_seq_stop,
2431         .show   = fib_trie_seq_show,
2432 };
2433 
2434 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2435 {
2436         return seq_open_net(inode, file, &fib_trie_seq_ops,
2437                             sizeof(struct fib_trie_iter));
2438 }
2439 
2440 static const struct file_operations fib_trie_fops = {
2441         .owner  = THIS_MODULE,
2442         .open   = fib_trie_seq_open,
2443         .read   = seq_read,
2444         .llseek = seq_lseek,
2445         .release = seq_release_net,
2446 };
2447 
2448 struct fib_route_iter {
2449         struct seq_net_private p;
2450         struct fib_table *main_tb;
2451         struct key_vector *tnode;
2452         loff_t  pos;
2453         t_key   key;
2454 };
2455 
2456 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2457                                             loff_t pos)
2458 {
2459         struct key_vector *l, **tp = &iter->tnode;
2460         t_key key;
2461 
2462         /* use cached location of previously found key */
2463         if (iter->pos > 0 && pos >= iter->pos) {
2464                 key = iter->key;
2465         } else {
2466                 iter->pos = 1;
2467                 key = 0;
2468         }
2469 
2470         pos -= iter->pos;
2471 
2472         while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2473                 key = l->key + 1;
2474                 iter->pos++;
2475                 l = NULL;
2476 
2477                 /* handle unlikely case of a key wrap */
2478                 if (!key)
2479                         break;
2480         }
2481 
2482         if (l)
2483                 iter->key = l->key;     /* remember it */
2484         else
2485                 iter->pos = 0;          /* forget it */
2486 
2487         return l;
2488 }
2489 
2490 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2491         __acquires(RCU)
2492 {
2493         struct fib_route_iter *iter = seq->private;
2494         struct fib_table *tb;
2495         struct trie *t;
2496 
2497         rcu_read_lock();
2498 
2499         tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2500         if (!tb)
2501                 return NULL;
2502 
2503         iter->main_tb = tb;
2504         t = (struct trie *)tb->tb_data;
2505         iter->tnode = t->kv;
2506 
2507         if (*pos != 0)
2508                 return fib_route_get_idx(iter, *pos);
2509 
2510         iter->pos = 0;
2511         iter->key = KEY_MAX;
2512 
2513         return SEQ_START_TOKEN;
2514 }
2515 
2516 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2517 {
2518         struct fib_route_iter *iter = seq->private;
2519         struct key_vector *l = NULL;
2520         t_key key = iter->key + 1;
2521 
2522         ++*pos;
2523 
2524         /* only allow key of 0 for start of sequence */
2525         if ((v == SEQ_START_TOKEN) || key)
2526                 l = leaf_walk_rcu(&iter->tnode, key);
2527 
2528         if (l) {
2529                 iter->key = l->key;
2530                 iter->pos++;
2531         } else {
2532                 iter->pos = 0;
2533         }
2534 
2535         return l;
2536 }
2537 
2538 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2539         __releases(RCU)
2540 {
2541         rcu_read_unlock();
2542 }
2543 
2544 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2545 {
2546         unsigned int flags = 0;
2547 
2548         if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2549                 flags = RTF_REJECT;
2550         if (fi && fi->fib_nh->nh_gw)
2551                 flags |= RTF_GATEWAY;
2552         if (mask == htonl(0xFFFFFFFF))
2553                 flags |= RTF_HOST;
2554         flags |= RTF_UP;
2555         return flags;
2556 }
2557 
2558 /*
2559  *      This outputs /proc/net/route.
2560  *      The format of the file is not supposed to be changed
2561  *      and needs to be same as fib_hash output to avoid breaking
2562  *      legacy utilities
2563  */
2564 static int fib_route_seq_show(struct seq_file *seq, void *v)
2565 {
2566         struct fib_route_iter *iter = seq->private;
2567         struct fib_table *tb = iter->main_tb;
2568         struct fib_alias *fa;
2569         struct key_vector *l = v;
2570         __be32 prefix;
2571 
2572         if (v == SEQ_START_TOKEN) {
2573                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2574                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2575                            "\tWindow\tIRTT");
2576                 return 0;
2577         }
2578 
2579         prefix = htonl(l->key);
2580 
2581         hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2582                 const struct fib_info *fi = fa->fa_info;
2583                 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2584                 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2585 
2586                 if ((fa->fa_type == RTN_BROADCAST) ||
2587                     (fa->fa_type == RTN_MULTICAST))
2588                         continue;
2589 
2590                 if (fa->tb_id != tb->tb_id)
2591                         continue;
2592 
2593                 seq_setwidth(seq, 127);
2594 
2595                 if (fi)
2596                         seq_printf(seq,
2597                                    "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2598                                    "%d\t%08X\t%d\t%u\t%u",
2599                                    fi->fib_dev ? fi->fib_dev->name : "*",
2600                                    prefix,
2601                                    fi->fib_nh->nh_gw, flags, 0, 0,
2602                                    fi->fib_priority,
2603                                    mask,
2604                                    (fi->fib_advmss ?
2605                                     fi->fib_advmss + 40 : 0),
2606                                    fi->fib_window,
2607                                    fi->fib_rtt >> 3);
2608                 else
2609                         seq_printf(seq,
2610                                    "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2611                                    "%d\t%08X\t%d\t%u\t%u",
2612                                    prefix, 0, flags, 0, 0, 0,
2613                                    mask, 0, 0, 0);
2614 
2615                 seq_pad(seq, '\n');
2616         }
2617 
2618         return 0;
2619 }
2620 
2621 static const struct seq_operations fib_route_seq_ops = {
2622         .start  = fib_route_seq_start,
2623         .next   = fib_route_seq_next,
2624         .stop   = fib_route_seq_stop,
2625         .show   = fib_route_seq_show,
2626 };
2627 
2628 static int fib_route_seq_open(struct inode *inode, struct file *file)
2629 {
2630         return seq_open_net(inode, file, &fib_route_seq_ops,
2631                             sizeof(struct fib_route_iter));
2632 }
2633 
2634 static const struct file_operations fib_route_fops = {
2635         .owner  = THIS_MODULE,
2636         .open   = fib_route_seq_open,
2637         .read   = seq_read,
2638         .llseek = seq_lseek,
2639         .release = seq_release_net,
2640 };
2641 
2642 int __net_init fib_proc_init(struct net *net)
2643 {
2644         if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2645                 goto out1;
2646 
2647         if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2648                          &fib_triestat_fops))
2649                 goto out2;
2650 
2651         if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2652                 goto out3;
2653 
2654         return 0;
2655 
2656 out3:
2657         remove_proc_entry("fib_triestat", net->proc_net);
2658 out2:
2659         remove_proc_entry("fib_trie", net->proc_net);
2660 out1:
2661         return -ENOMEM;
2662 }
2663 
2664 void __net_exit fib_proc_exit(struct net *net)
2665 {
2666         remove_proc_entry("fib_trie", net->proc_net);
2667         remove_proc_entry("fib_triestat", net->proc_net);
2668         remove_proc_entry("route", net->proc_net);
2669 }
2670 
2671 #endif /* CONFIG_PROC_FS */
2672 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp