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

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

Version: ~ [ linux-5.12-rc1 ] ~ [ linux-5.11.2 ] ~ [ linux-5.10.19 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.101 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.177 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.222 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.258 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.258 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-or-later
  2 /*
  3  *
  4  *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
  5  *     & Swedish University of Agricultural Sciences.
  6  *
  7  *   Jens Laas <jens.laas@data.slu.se> Swedish University of
  8  *     Agricultural Sciences.
  9  *
 10  *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
 11  *
 12  * This work is based on the LPC-trie which is originally described in:
 13  *
 14  * An experimental study of compression methods for dynamic tries
 15  * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
 16  * http://www.csc.kth.se/~snilsson/software/dyntrie2/
 17  *
 18  * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
 19  * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
 20  *
 21  * Code from fib_hash has been reused which includes the following header:
 22  *
 23  * INET         An implementation of the TCP/IP protocol suite for the LINUX
 24  *              operating system.  INET is implemented using the  BSD Socket
 25  *              interface as the means of communication with the user level.
 26  *
 27  *              IPv4 FIB: lookup engine and maintenance routines.
 28  *
 29  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 30  *
 31  * Substantial contributions to this work comes from:
 32  *
 33  *              David S. Miller, <davem@davemloft.net>
 34  *              Stephen Hemminger <shemminger@osdl.org>
 35  *              Paul E. McKenney <paulmck@us.ibm.com>
 36  *              Patrick McHardy <kaber@trash.net>
 37  */
 38 
 39 #define VERSION "0.409"
 40 
 41 #include <linux/cache.h>
 42 #include <linux/uaccess.h>
 43 #include <linux/bitops.h>
 44 #include <linux/types.h>
 45 #include <linux/kernel.h>
 46 #include <linux/mm.h>
 47 #include <linux/string.h>
 48 #include <linux/socket.h>
 49 #include <linux/sockios.h>
 50 #include <linux/errno.h>
 51 #include <linux/in.h>
 52 #include <linux/inet.h>
 53 #include <linux/inetdevice.h>
 54 #include <linux/netdevice.h>
 55 #include <linux/if_arp.h>
 56 #include <linux/proc_fs.h>
 57 #include <linux/rcupdate.h>
 58 #include <linux/skbuff.h>
 59 #include <linux/netlink.h>
 60 #include <linux/init.h>
 61 #include <linux/list.h>
 62 #include <linux/slab.h>
 63 #include <linux/export.h>
 64 #include <linux/vmalloc.h>
 65 #include <linux/notifier.h>
 66 #include <net/net_namespace.h>
 67 #include <net/ip.h>
 68 #include <net/protocol.h>
 69 #include <net/route.h>
 70 #include <net/tcp.h>
 71 #include <net/sock.h>
 72 #include <net/ip_fib.h>
 73 #include <net/fib_notifier.h>
 74 #include <trace/events/fib.h>
 75 #include "fib_lookup.h"
 76 
 77 static int call_fib_entry_notifier(struct notifier_block *nb,
 78                                    enum fib_event_type event_type, u32 dst,
 79                                    int dst_len, struct fib_alias *fa,
 80                                    struct netlink_ext_ack *extack)
 81 {
 82         struct fib_entry_notifier_info info = {
 83                 .info.extack = extack,
 84                 .dst = dst,
 85                 .dst_len = dst_len,
 86                 .fi = fa->fa_info,
 87                 .tos = fa->fa_tos,
 88                 .type = fa->fa_type,
 89                 .tb_id = fa->tb_id,
 90         };
 91         return call_fib4_notifier(nb, event_type, &info.info);
 92 }
 93 
 94 static int call_fib_entry_notifiers(struct net *net,
 95                                     enum fib_event_type event_type, u32 dst,
 96                                     int dst_len, struct fib_alias *fa,
 97                                     struct netlink_ext_ack *extack)
 98 {
 99         struct fib_entry_notifier_info info = {
100                 .info.extack = extack,
101                 .dst = dst,
102                 .dst_len = dst_len,
103                 .fi = fa->fa_info,
104                 .tos = fa->fa_tos,
105                 .type = fa->fa_type,
106                 .tb_id = fa->tb_id,
107         };
108         return call_fib4_notifiers(net, event_type, &info.info);
109 }
110 
111 #define MAX_STAT_DEPTH 32
112 
113 #define KEYLENGTH       (8*sizeof(t_key))
114 #define KEY_MAX         ((t_key)~0)
115 
116 typedef unsigned int t_key;
117 
118 #define IS_TRIE(n)      ((n)->pos >= KEYLENGTH)
119 #define IS_TNODE(n)     ((n)->bits)
120 #define IS_LEAF(n)      (!(n)->bits)
121 
122 struct key_vector {
123         t_key key;
124         unsigned char pos;              /* 2log(KEYLENGTH) bits needed */
125         unsigned char bits;             /* 2log(KEYLENGTH) bits needed */
126         unsigned char slen;
127         union {
128                 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
129                 struct hlist_head leaf;
130                 /* This array is valid if (pos | bits) > 0 (TNODE) */
131                 struct key_vector __rcu *tnode[0];
132         };
133 };
134 
135 struct tnode {
136         struct rcu_head rcu;
137         t_key empty_children;           /* KEYLENGTH bits needed */
138         t_key full_children;            /* KEYLENGTH bits needed */
139         struct key_vector __rcu *parent;
140         struct key_vector kv[1];
141 #define tn_bits kv[0].bits
142 };
143 
144 #define TNODE_SIZE(n)   offsetof(struct tnode, kv[0].tnode[n])
145 #define LEAF_SIZE       TNODE_SIZE(1)
146 
147 #ifdef CONFIG_IP_FIB_TRIE_STATS
148 struct trie_use_stats {
149         unsigned int gets;
150         unsigned int backtrack;
151         unsigned int semantic_match_passed;
152         unsigned int semantic_match_miss;
153         unsigned int null_node_hit;
154         unsigned int resize_node_skipped;
155 };
156 #endif
157 
158 struct trie_stat {
159         unsigned int totdepth;
160         unsigned int maxdepth;
161         unsigned int tnodes;
162         unsigned int leaves;
163         unsigned int nullpointers;
164         unsigned int prefixes;
165         unsigned int nodesizes[MAX_STAT_DEPTH];
166 };
167 
168 struct trie {
169         struct key_vector kv[1];
170 #ifdef CONFIG_IP_FIB_TRIE_STATS
171         struct trie_use_stats __percpu *stats;
172 #endif
173 };
174 
175 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
176 static unsigned int tnode_free_size;
177 
178 /*
179  * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
180  * especially useful before resizing the root node with PREEMPT_NONE configs;
181  * the value was obtained experimentally, aiming to avoid visible slowdown.
182  */
183 unsigned int sysctl_fib_sync_mem = 512 * 1024;
184 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
185 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
186 
187 static struct kmem_cache *fn_alias_kmem __ro_after_init;
188 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
189 
190 static inline struct tnode *tn_info(struct key_vector *kv)
191 {
192         return container_of(kv, struct tnode, kv[0]);
193 }
194 
195 /* caller must hold RTNL */
196 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
197 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
198 
199 /* caller must hold RCU read lock or RTNL */
200 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
201 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
202 
203 /* wrapper for rcu_assign_pointer */
204 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
205 {
206         if (n)
207                 rcu_assign_pointer(tn_info(n)->parent, tp);
208 }
209 
210 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
211 
212 /* This provides us with the number of children in this node, in the case of a
213  * leaf this will return 0 meaning none of the children are accessible.
214  */
215 static inline unsigned long child_length(const struct key_vector *tn)
216 {
217         return (1ul << tn->bits) & ~(1ul);
218 }
219 
220 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
221 
222 static inline unsigned long get_index(t_key key, struct key_vector *kv)
223 {
224         unsigned long index = key ^ kv->key;
225 
226         if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
227                 return 0;
228 
229         return index >> kv->pos;
230 }
231 
232 /* To understand this stuff, an understanding of keys and all their bits is
233  * necessary. Every node in the trie has a key associated with it, but not
234  * all of the bits in that key are significant.
235  *
236  * Consider a node 'n' and its parent 'tp'.
237  *
238  * If n is a leaf, every bit in its key is significant. Its presence is
239  * necessitated by path compression, since during a tree traversal (when
240  * searching for a leaf - unless we are doing an insertion) we will completely
241  * ignore all skipped bits we encounter. Thus we need to verify, at the end of
242  * a potentially successful search, that we have indeed been walking the
243  * correct key path.
244  *
245  * Note that we can never "miss" the correct key in the tree if present by
246  * following the wrong path. Path compression ensures that segments of the key
247  * that are the same for all keys with a given prefix are skipped, but the
248  * skipped part *is* identical for each node in the subtrie below the skipped
249  * bit! trie_insert() in this implementation takes care of that.
250  *
251  * if n is an internal node - a 'tnode' here, the various parts of its key
252  * have many different meanings.
253  *
254  * Example:
255  * _________________________________________________________________
256  * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
257  * -----------------------------------------------------------------
258  *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
259  *
260  * _________________________________________________________________
261  * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
262  * -----------------------------------------------------------------
263  *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
264  *
265  * tp->pos = 22
266  * tp->bits = 3
267  * n->pos = 13
268  * n->bits = 4
269  *
270  * First, let's just ignore the bits that come before the parent tp, that is
271  * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
272  * point we do not use them for anything.
273  *
274  * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
275  * index into the parent's child array. That is, they will be used to find
276  * 'n' among tp's children.
277  *
278  * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
279  * for the node n.
280  *
281  * All the bits we have seen so far are significant to the node n. The rest
282  * of the bits are really not needed or indeed known in n->key.
283  *
284  * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
285  * n's child array, and will of course be different for each child.
286  *
287  * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
288  * at this point.
289  */
290 
291 static const int halve_threshold = 25;
292 static const int inflate_threshold = 50;
293 static const int halve_threshold_root = 15;
294 static const int inflate_threshold_root = 30;
295 
296 static void __alias_free_mem(struct rcu_head *head)
297 {
298         struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
299         kmem_cache_free(fn_alias_kmem, fa);
300 }
301 
302 static inline void alias_free_mem_rcu(struct fib_alias *fa)
303 {
304         call_rcu(&fa->rcu, __alias_free_mem);
305 }
306 
307 #define TNODE_KMALLOC_MAX \
308         ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
309 #define TNODE_VMALLOC_MAX \
310         ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
311 
312 static void __node_free_rcu(struct rcu_head *head)
313 {
314         struct tnode *n = container_of(head, struct tnode, rcu);
315 
316         if (!n->tn_bits)
317                 kmem_cache_free(trie_leaf_kmem, n);
318         else
319                 kvfree(n);
320 }
321 
322 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
323 
324 static struct tnode *tnode_alloc(int bits)
325 {
326         size_t size;
327 
328         /* verify bits is within bounds */
329         if (bits > TNODE_VMALLOC_MAX)
330                 return NULL;
331 
332         /* determine size and verify it is non-zero and didn't overflow */
333         size = TNODE_SIZE(1ul << bits);
334 
335         if (size <= PAGE_SIZE)
336                 return kzalloc(size, GFP_KERNEL);
337         else
338                 return vzalloc(size);
339 }
340 
341 static inline void empty_child_inc(struct key_vector *n)
342 {
343         tn_info(n)->empty_children++;
344 
345         if (!tn_info(n)->empty_children)
346                 tn_info(n)->full_children++;
347 }
348 
349 static inline void empty_child_dec(struct key_vector *n)
350 {
351         if (!tn_info(n)->empty_children)
352                 tn_info(n)->full_children--;
353 
354         tn_info(n)->empty_children--;
355 }
356 
357 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
358 {
359         struct key_vector *l;
360         struct tnode *kv;
361 
362         kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
363         if (!kv)
364                 return NULL;
365 
366         /* initialize key vector */
367         l = kv->kv;
368         l->key = key;
369         l->pos = 0;
370         l->bits = 0;
371         l->slen = fa->fa_slen;
372 
373         /* link leaf to fib alias */
374         INIT_HLIST_HEAD(&l->leaf);
375         hlist_add_head(&fa->fa_list, &l->leaf);
376 
377         return l;
378 }
379 
380 static struct key_vector *tnode_new(t_key key, int pos, int bits)
381 {
382         unsigned int shift = pos + bits;
383         struct key_vector *tn;
384         struct tnode *tnode;
385 
386         /* verify bits and pos their msb bits clear and values are valid */
387         BUG_ON(!bits || (shift > KEYLENGTH));
388 
389         tnode = tnode_alloc(bits);
390         if (!tnode)
391                 return NULL;
392 
393         pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
394                  sizeof(struct key_vector *) << bits);
395 
396         if (bits == KEYLENGTH)
397                 tnode->full_children = 1;
398         else
399                 tnode->empty_children = 1ul << bits;
400 
401         tn = tnode->kv;
402         tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
403         tn->pos = pos;
404         tn->bits = bits;
405         tn->slen = pos;
406 
407         return tn;
408 }
409 
410 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
411  * and no bits are skipped. See discussion in dyntree paper p. 6
412  */
413 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
414 {
415         return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
416 }
417 
418 /* Add a child at position i overwriting the old value.
419  * Update the value of full_children and empty_children.
420  */
421 static void put_child(struct key_vector *tn, unsigned long i,
422                       struct key_vector *n)
423 {
424         struct key_vector *chi = get_child(tn, i);
425         int isfull, wasfull;
426 
427         BUG_ON(i >= child_length(tn));
428 
429         /* update emptyChildren, overflow into fullChildren */
430         if (!n && chi)
431                 empty_child_inc(tn);
432         if (n && !chi)
433                 empty_child_dec(tn);
434 
435         /* update fullChildren */
436         wasfull = tnode_full(tn, chi);
437         isfull = tnode_full(tn, n);
438 
439         if (wasfull && !isfull)
440                 tn_info(tn)->full_children--;
441         else if (!wasfull && isfull)
442                 tn_info(tn)->full_children++;
443 
444         if (n && (tn->slen < n->slen))
445                 tn->slen = n->slen;
446 
447         rcu_assign_pointer(tn->tnode[i], n);
448 }
449 
450 static void update_children(struct key_vector *tn)
451 {
452         unsigned long i;
453 
454         /* update all of the child parent pointers */
455         for (i = child_length(tn); i;) {
456                 struct key_vector *inode = get_child(tn, --i);
457 
458                 if (!inode)
459                         continue;
460 
461                 /* Either update the children of a tnode that
462                  * already belongs to us or update the child
463                  * to point to ourselves.
464                  */
465                 if (node_parent(inode) == tn)
466                         update_children(inode);
467                 else
468                         node_set_parent(inode, tn);
469         }
470 }
471 
472 static inline void put_child_root(struct key_vector *tp, t_key key,
473                                   struct key_vector *n)
474 {
475         if (IS_TRIE(tp))
476                 rcu_assign_pointer(tp->tnode[0], n);
477         else
478                 put_child(tp, get_index(key, tp), n);
479 }
480 
481 static inline void tnode_free_init(struct key_vector *tn)
482 {
483         tn_info(tn)->rcu.next = NULL;
484 }
485 
486 static inline void tnode_free_append(struct key_vector *tn,
487                                      struct key_vector *n)
488 {
489         tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
490         tn_info(tn)->rcu.next = &tn_info(n)->rcu;
491 }
492 
493 static void tnode_free(struct key_vector *tn)
494 {
495         struct callback_head *head = &tn_info(tn)->rcu;
496 
497         while (head) {
498                 head = head->next;
499                 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
500                 node_free(tn);
501 
502                 tn = container_of(head, struct tnode, rcu)->kv;
503         }
504 
505         if (tnode_free_size >= sysctl_fib_sync_mem) {
506                 tnode_free_size = 0;
507                 synchronize_rcu();
508         }
509 }
510 
511 static struct key_vector *replace(struct trie *t,
512                                   struct key_vector *oldtnode,
513                                   struct key_vector *tn)
514 {
515         struct key_vector *tp = node_parent(oldtnode);
516         unsigned long i;
517 
518         /* setup the parent pointer out of and back into this node */
519         NODE_INIT_PARENT(tn, tp);
520         put_child_root(tp, tn->key, tn);
521 
522         /* update all of the child parent pointers */
523         update_children(tn);
524 
525         /* all pointers should be clean so we are done */
526         tnode_free(oldtnode);
527 
528         /* resize children now that oldtnode is freed */
529         for (i = child_length(tn); i;) {
530                 struct key_vector *inode = get_child(tn, --i);
531 
532                 /* resize child node */
533                 if (tnode_full(tn, inode))
534                         tn = resize(t, inode);
535         }
536 
537         return tp;
538 }
539 
540 static struct key_vector *inflate(struct trie *t,
541                                   struct key_vector *oldtnode)
542 {
543         struct key_vector *tn;
544         unsigned long i;
545         t_key m;
546 
547         pr_debug("In inflate\n");
548 
549         tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
550         if (!tn)
551                 goto notnode;
552 
553         /* prepare oldtnode to be freed */
554         tnode_free_init(oldtnode);
555 
556         /* Assemble all of the pointers in our cluster, in this case that
557          * represents all of the pointers out of our allocated nodes that
558          * point to existing tnodes and the links between our allocated
559          * nodes.
560          */
561         for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
562                 struct key_vector *inode = get_child(oldtnode, --i);
563                 struct key_vector *node0, *node1;
564                 unsigned long j, k;
565 
566                 /* An empty child */
567                 if (!inode)
568                         continue;
569 
570                 /* A leaf or an internal node with skipped bits */
571                 if (!tnode_full(oldtnode, inode)) {
572                         put_child(tn, get_index(inode->key, tn), inode);
573                         continue;
574                 }
575 
576                 /* drop the node in the old tnode free list */
577                 tnode_free_append(oldtnode, inode);
578 
579                 /* An internal node with two children */
580                 if (inode->bits == 1) {
581                         put_child(tn, 2 * i + 1, get_child(inode, 1));
582                         put_child(tn, 2 * i, get_child(inode, 0));
583                         continue;
584                 }
585 
586                 /* We will replace this node 'inode' with two new
587                  * ones, 'node0' and 'node1', each with half of the
588                  * original children. The two new nodes will have
589                  * a position one bit further down the key and this
590                  * means that the "significant" part of their keys
591                  * (see the discussion near the top of this file)
592                  * will differ by one bit, which will be "" in
593                  * node0's key and "1" in node1's key. Since we are
594                  * moving the key position by one step, the bit that
595                  * we are moving away from - the bit at position
596                  * (tn->pos) - is the one that will differ between
597                  * node0 and node1. So... we synthesize that bit in the
598                  * two new keys.
599                  */
600                 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
601                 if (!node1)
602                         goto nomem;
603                 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
604 
605                 tnode_free_append(tn, node1);
606                 if (!node0)
607                         goto nomem;
608                 tnode_free_append(tn, node0);
609 
610                 /* populate child pointers in new nodes */
611                 for (k = child_length(inode), j = k / 2; j;) {
612                         put_child(node1, --j, get_child(inode, --k));
613                         put_child(node0, j, get_child(inode, j));
614                         put_child(node1, --j, get_child(inode, --k));
615                         put_child(node0, j, get_child(inode, j));
616                 }
617 
618                 /* link new nodes to parent */
619                 NODE_INIT_PARENT(node1, tn);
620                 NODE_INIT_PARENT(node0, tn);
621 
622                 /* link parent to nodes */
623                 put_child(tn, 2 * i + 1, node1);
624                 put_child(tn, 2 * i, node0);
625         }
626 
627         /* setup the parent pointers into and out of this node */
628         return replace(t, oldtnode, tn);
629 nomem:
630         /* all pointers should be clean so we are done */
631         tnode_free(tn);
632 notnode:
633         return NULL;
634 }
635 
636 static struct key_vector *halve(struct trie *t,
637                                 struct key_vector *oldtnode)
638 {
639         struct key_vector *tn;
640         unsigned long i;
641 
642         pr_debug("In halve\n");
643 
644         tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
645         if (!tn)
646                 goto notnode;
647 
648         /* prepare oldtnode to be freed */
649         tnode_free_init(oldtnode);
650 
651         /* Assemble all of the pointers in our cluster, in this case that
652          * represents all of the pointers out of our allocated nodes that
653          * point to existing tnodes and the links between our allocated
654          * nodes.
655          */
656         for (i = child_length(oldtnode); i;) {
657                 struct key_vector *node1 = get_child(oldtnode, --i);
658                 struct key_vector *node0 = get_child(oldtnode, --i);
659                 struct key_vector *inode;
660 
661                 /* At least one of the children is empty */
662                 if (!node1 || !node0) {
663                         put_child(tn, i / 2, node1 ? : node0);
664                         continue;
665                 }
666 
667                 /* Two nonempty children */
668                 inode = tnode_new(node0->key, oldtnode->pos, 1);
669                 if (!inode)
670                         goto nomem;
671                 tnode_free_append(tn, inode);
672 
673                 /* initialize pointers out of node */
674                 put_child(inode, 1, node1);
675                 put_child(inode, 0, node0);
676                 NODE_INIT_PARENT(inode, tn);
677 
678                 /* link parent to node */
679                 put_child(tn, i / 2, inode);
680         }
681 
682         /* setup the parent pointers into and out of this node */
683         return replace(t, oldtnode, tn);
684 nomem:
685         /* all pointers should be clean so we are done */
686         tnode_free(tn);
687 notnode:
688         return NULL;
689 }
690 
691 static struct key_vector *collapse(struct trie *t,
692                                    struct key_vector *oldtnode)
693 {
694         struct key_vector *n, *tp;
695         unsigned long i;
696 
697         /* scan the tnode looking for that one child that might still exist */
698         for (n = NULL, i = child_length(oldtnode); !n && i;)
699                 n = get_child(oldtnode, --i);
700 
701         /* compress one level */
702         tp = node_parent(oldtnode);
703         put_child_root(tp, oldtnode->key, n);
704         node_set_parent(n, tp);
705 
706         /* drop dead node */
707         node_free(oldtnode);
708 
709         return tp;
710 }
711 
712 static unsigned char update_suffix(struct key_vector *tn)
713 {
714         unsigned char slen = tn->pos;
715         unsigned long stride, i;
716         unsigned char slen_max;
717 
718         /* only vector 0 can have a suffix length greater than or equal to
719          * tn->pos + tn->bits, the second highest node will have a suffix
720          * length at most of tn->pos + tn->bits - 1
721          */
722         slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
723 
724         /* search though the list of children looking for nodes that might
725          * have a suffix greater than the one we currently have.  This is
726          * why we start with a stride of 2 since a stride of 1 would
727          * represent the nodes with suffix length equal to tn->pos
728          */
729         for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
730                 struct key_vector *n = get_child(tn, i);
731 
732                 if (!n || (n->slen <= slen))
733                         continue;
734 
735                 /* update stride and slen based on new value */
736                 stride <<= (n->slen - slen);
737                 slen = n->slen;
738                 i &= ~(stride - 1);
739 
740                 /* stop searching if we have hit the maximum possible value */
741                 if (slen >= slen_max)
742                         break;
743         }
744 
745         tn->slen = slen;
746 
747         return slen;
748 }
749 
750 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
751  * the Helsinki University of Technology and Matti Tikkanen of Nokia
752  * Telecommunications, page 6:
753  * "A node is doubled if the ratio of non-empty children to all
754  * children in the *doubled* node is at least 'high'."
755  *
756  * 'high' in this instance is the variable 'inflate_threshold'. It
757  * is expressed as a percentage, so we multiply it with
758  * child_length() and instead of multiplying by 2 (since the
759  * child array will be doubled by inflate()) and multiplying
760  * the left-hand side by 100 (to handle the percentage thing) we
761  * multiply the left-hand side by 50.
762  *
763  * The left-hand side may look a bit weird: child_length(tn)
764  * - tn->empty_children is of course the number of non-null children
765  * in the current node. tn->full_children is the number of "full"
766  * children, that is non-null tnodes with a skip value of 0.
767  * All of those will be doubled in the resulting inflated tnode, so
768  * we just count them one extra time here.
769  *
770  * A clearer way to write this would be:
771  *
772  * to_be_doubled = tn->full_children;
773  * not_to_be_doubled = child_length(tn) - tn->empty_children -
774  *     tn->full_children;
775  *
776  * new_child_length = child_length(tn) * 2;
777  *
778  * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
779  *      new_child_length;
780  * if (new_fill_factor >= inflate_threshold)
781  *
782  * ...and so on, tho it would mess up the while () loop.
783  *
784  * anyway,
785  * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
786  *      inflate_threshold
787  *
788  * avoid a division:
789  * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
790  *      inflate_threshold * new_child_length
791  *
792  * expand not_to_be_doubled and to_be_doubled, and shorten:
793  * 100 * (child_length(tn) - tn->empty_children +
794  *    tn->full_children) >= inflate_threshold * new_child_length
795  *
796  * expand new_child_length:
797  * 100 * (child_length(tn) - tn->empty_children +
798  *    tn->full_children) >=
799  *      inflate_threshold * child_length(tn) * 2
800  *
801  * shorten again:
802  * 50 * (tn->full_children + child_length(tn) -
803  *    tn->empty_children) >= inflate_threshold *
804  *    child_length(tn)
805  *
806  */
807 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
808 {
809         unsigned long used = child_length(tn);
810         unsigned long threshold = used;
811 
812         /* Keep root node larger */
813         threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
814         used -= tn_info(tn)->empty_children;
815         used += tn_info(tn)->full_children;
816 
817         /* if bits == KEYLENGTH then pos = 0, and will fail below */
818 
819         return (used > 1) && tn->pos && ((50 * used) >= threshold);
820 }
821 
822 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
823 {
824         unsigned long used = child_length(tn);
825         unsigned long threshold = used;
826 
827         /* Keep root node larger */
828         threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
829         used -= tn_info(tn)->empty_children;
830 
831         /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
832 
833         return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
834 }
835 
836 static inline bool should_collapse(struct key_vector *tn)
837 {
838         unsigned long used = child_length(tn);
839 
840         used -= tn_info(tn)->empty_children;
841 
842         /* account for bits == KEYLENGTH case */
843         if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
844                 used -= KEY_MAX;
845 
846         /* One child or none, time to drop us from the trie */
847         return used < 2;
848 }
849 
850 #define MAX_WORK 10
851 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
852 {
853 #ifdef CONFIG_IP_FIB_TRIE_STATS
854         struct trie_use_stats __percpu *stats = t->stats;
855 #endif
856         struct key_vector *tp = node_parent(tn);
857         unsigned long cindex = get_index(tn->key, tp);
858         int max_work = MAX_WORK;
859 
860         pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
861                  tn, inflate_threshold, halve_threshold);
862 
863         /* track the tnode via the pointer from the parent instead of
864          * doing it ourselves.  This way we can let RCU fully do its
865          * thing without us interfering
866          */
867         BUG_ON(tn != get_child(tp, cindex));
868 
869         /* Double as long as the resulting node has a number of
870          * nonempty nodes that are above the threshold.
871          */
872         while (should_inflate(tp, tn) && max_work) {
873                 tp = inflate(t, tn);
874                 if (!tp) {
875 #ifdef CONFIG_IP_FIB_TRIE_STATS
876                         this_cpu_inc(stats->resize_node_skipped);
877 #endif
878                         break;
879                 }
880 
881                 max_work--;
882                 tn = get_child(tp, cindex);
883         }
884 
885         /* update parent in case inflate failed */
886         tp = node_parent(tn);
887 
888         /* Return if at least one inflate is run */
889         if (max_work != MAX_WORK)
890                 return tp;
891 
892         /* Halve as long as the number of empty children in this
893          * node is above threshold.
894          */
895         while (should_halve(tp, tn) && max_work) {
896                 tp = halve(t, tn);
897                 if (!tp) {
898 #ifdef CONFIG_IP_FIB_TRIE_STATS
899                         this_cpu_inc(stats->resize_node_skipped);
900 #endif
901                         break;
902                 }
903 
904                 max_work--;
905                 tn = get_child(tp, cindex);
906         }
907 
908         /* Only one child remains */
909         if (should_collapse(tn))
910                 return collapse(t, tn);
911 
912         /* update parent in case halve failed */
913         return node_parent(tn);
914 }
915 
916 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
917 {
918         unsigned char node_slen = tn->slen;
919 
920         while ((node_slen > tn->pos) && (node_slen > slen)) {
921                 slen = update_suffix(tn);
922                 if (node_slen == slen)
923                         break;
924 
925                 tn = node_parent(tn);
926                 node_slen = tn->slen;
927         }
928 }
929 
930 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
931 {
932         while (tn->slen < slen) {
933                 tn->slen = slen;
934                 tn = node_parent(tn);
935         }
936 }
937 
938 /* rcu_read_lock needs to be hold by caller from readside */
939 static struct key_vector *fib_find_node(struct trie *t,
940                                         struct key_vector **tp, u32 key)
941 {
942         struct key_vector *pn, *n = t->kv;
943         unsigned long index = 0;
944 
945         do {
946                 pn = n;
947                 n = get_child_rcu(n, index);
948 
949                 if (!n)
950                         break;
951 
952                 index = get_cindex(key, n);
953 
954                 /* This bit of code is a bit tricky but it combines multiple
955                  * checks into a single check.  The prefix consists of the
956                  * prefix plus zeros for the bits in the cindex. The index
957                  * is the difference between the key and this value.  From
958                  * this we can actually derive several pieces of data.
959                  *   if (index >= (1ul << bits))
960                  *     we have a mismatch in skip bits and failed
961                  *   else
962                  *     we know the value is cindex
963                  *
964                  * This check is safe even if bits == KEYLENGTH due to the
965                  * fact that we can only allocate a node with 32 bits if a
966                  * long is greater than 32 bits.
967                  */
968                 if (index >= (1ul << n->bits)) {
969                         n = NULL;
970                         break;
971                 }
972 
973                 /* keep searching until we find a perfect match leaf or NULL */
974         } while (IS_TNODE(n));
975 
976         *tp = pn;
977 
978         return n;
979 }
980 
981 /* Return the first fib alias matching TOS with
982  * priority less than or equal to PRIO.
983  */
984 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
985                                         u8 tos, u32 prio, u32 tb_id)
986 {
987         struct fib_alias *fa;
988 
989         if (!fah)
990                 return NULL;
991 
992         hlist_for_each_entry(fa, fah, fa_list) {
993                 if (fa->fa_slen < slen)
994                         continue;
995                 if (fa->fa_slen != slen)
996                         break;
997                 if (fa->tb_id > tb_id)
998                         continue;
999                 if (fa->tb_id != tb_id)
1000                         break;
1001                 if (fa->fa_tos > tos)
1002                         continue;
1003                 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1004                         return fa;
1005         }
1006 
1007         return NULL;
1008 }
1009 
1010 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1011 {
1012         while (!IS_TRIE(tn))
1013                 tn = resize(t, tn);
1014 }
1015 
1016 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1017                            struct fib_alias *new, t_key key)
1018 {
1019         struct key_vector *n, *l;
1020 
1021         l = leaf_new(key, new);
1022         if (!l)
1023                 goto noleaf;
1024 
1025         /* retrieve child from parent node */
1026         n = get_child(tp, get_index(key, tp));
1027 
1028         /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1029          *
1030          *  Add a new tnode here
1031          *  first tnode need some special handling
1032          *  leaves us in position for handling as case 3
1033          */
1034         if (n) {
1035                 struct key_vector *tn;
1036 
1037                 tn = tnode_new(key, __fls(key ^ n->key), 1);
1038                 if (!tn)
1039                         goto notnode;
1040 
1041                 /* initialize routes out of node */
1042                 NODE_INIT_PARENT(tn, tp);
1043                 put_child(tn, get_index(key, tn) ^ 1, n);
1044 
1045                 /* start adding routes into the node */
1046                 put_child_root(tp, key, tn);
1047                 node_set_parent(n, tn);
1048 
1049                 /* parent now has a NULL spot where the leaf can go */
1050                 tp = tn;
1051         }
1052 
1053         /* Case 3: n is NULL, and will just insert a new leaf */
1054         node_push_suffix(tp, new->fa_slen);
1055         NODE_INIT_PARENT(l, tp);
1056         put_child_root(tp, key, l);
1057         trie_rebalance(t, tp);
1058 
1059         return 0;
1060 notnode:
1061         node_free(l);
1062 noleaf:
1063         return -ENOMEM;
1064 }
1065 
1066 /* fib notifier for ADD is sent before calling fib_insert_alias with
1067  * the expectation that the only possible failure ENOMEM
1068  */
1069 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1070                             struct key_vector *l, struct fib_alias *new,
1071                             struct fib_alias *fa, t_key key)
1072 {
1073         if (!l)
1074                 return fib_insert_node(t, tp, new, key);
1075 
1076         if (fa) {
1077                 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1078         } else {
1079                 struct fib_alias *last;
1080 
1081                 hlist_for_each_entry(last, &l->leaf, fa_list) {
1082                         if (new->fa_slen < last->fa_slen)
1083                                 break;
1084                         if ((new->fa_slen == last->fa_slen) &&
1085                             (new->tb_id > last->tb_id))
1086                                 break;
1087                         fa = last;
1088                 }
1089 
1090                 if (fa)
1091                         hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1092                 else
1093                         hlist_add_head_rcu(&new->fa_list, &l->leaf);
1094         }
1095 
1096         /* if we added to the tail node then we need to update slen */
1097         if (l->slen < new->fa_slen) {
1098                 l->slen = new->fa_slen;
1099                 node_push_suffix(tp, new->fa_slen);
1100         }
1101 
1102         return 0;
1103 }
1104 
1105 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1106 {
1107         if (plen > KEYLENGTH) {
1108                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1109                 return false;
1110         }
1111 
1112         if ((plen < KEYLENGTH) && (key << plen)) {
1113                 NL_SET_ERR_MSG(extack,
1114                                "Invalid prefix for given prefix length");
1115                 return false;
1116         }
1117 
1118         return true;
1119 }
1120 
1121 /* Caller must hold RTNL. */
1122 int fib_table_insert(struct net *net, struct fib_table *tb,
1123                      struct fib_config *cfg, struct netlink_ext_ack *extack)
1124 {
1125         enum fib_event_type event = FIB_EVENT_ENTRY_ADD;
1126         struct trie *t = (struct trie *)tb->tb_data;
1127         struct fib_alias *fa, *new_fa;
1128         struct key_vector *l, *tp;
1129         u16 nlflags = NLM_F_EXCL;
1130         struct fib_info *fi;
1131         u8 plen = cfg->fc_dst_len;
1132         u8 slen = KEYLENGTH - plen;
1133         u8 tos = cfg->fc_tos;
1134         u32 key;
1135         int err;
1136 
1137         key = ntohl(cfg->fc_dst);
1138 
1139         if (!fib_valid_key_len(key, plen, extack))
1140                 return -EINVAL;
1141 
1142         pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1143 
1144         fi = fib_create_info(cfg, extack);
1145         if (IS_ERR(fi)) {
1146                 err = PTR_ERR(fi);
1147                 goto err;
1148         }
1149 
1150         l = fib_find_node(t, &tp, key);
1151         fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1152                                 tb->tb_id) : NULL;
1153 
1154         /* Now fa, if non-NULL, points to the first fib alias
1155          * with the same keys [prefix,tos,priority], if such key already
1156          * exists or to the node before which we will insert new one.
1157          *
1158          * If fa is NULL, we will need to allocate a new one and
1159          * insert to the tail of the section matching the suffix length
1160          * of the new alias.
1161          */
1162 
1163         if (fa && fa->fa_tos == tos &&
1164             fa->fa_info->fib_priority == fi->fib_priority) {
1165                 struct fib_alias *fa_first, *fa_match;
1166 
1167                 err = -EEXIST;
1168                 if (cfg->fc_nlflags & NLM_F_EXCL)
1169                         goto out;
1170 
1171                 nlflags &= ~NLM_F_EXCL;
1172 
1173                 /* We have 2 goals:
1174                  * 1. Find exact match for type, scope, fib_info to avoid
1175                  * duplicate routes
1176                  * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1177                  */
1178                 fa_match = NULL;
1179                 fa_first = fa;
1180                 hlist_for_each_entry_from(fa, fa_list) {
1181                         if ((fa->fa_slen != slen) ||
1182                             (fa->tb_id != tb->tb_id) ||
1183                             (fa->fa_tos != tos))
1184                                 break;
1185                         if (fa->fa_info->fib_priority != fi->fib_priority)
1186                                 break;
1187                         if (fa->fa_type == cfg->fc_type &&
1188                             fa->fa_info == fi) {
1189                                 fa_match = fa;
1190                                 break;
1191                         }
1192                 }
1193 
1194                 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1195                         struct fib_info *fi_drop;
1196                         u8 state;
1197 
1198                         nlflags |= NLM_F_REPLACE;
1199                         fa = fa_first;
1200                         if (fa_match) {
1201                                 if (fa == fa_match)
1202                                         err = 0;
1203                                 goto out;
1204                         }
1205                         err = -ENOBUFS;
1206                         new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1207                         if (!new_fa)
1208                                 goto out;
1209 
1210                         fi_drop = fa->fa_info;
1211                         new_fa->fa_tos = fa->fa_tos;
1212                         new_fa->fa_info = fi;
1213                         new_fa->fa_type = cfg->fc_type;
1214                         state = fa->fa_state;
1215                         new_fa->fa_state = state & ~FA_S_ACCESSED;
1216                         new_fa->fa_slen = fa->fa_slen;
1217                         new_fa->tb_id = tb->tb_id;
1218                         new_fa->fa_default = -1;
1219 
1220                         err = call_fib_entry_notifiers(net,
1221                                                        FIB_EVENT_ENTRY_REPLACE,
1222                                                        key, plen, new_fa,
1223                                                        extack);
1224                         if (err)
1225                                 goto out_free_new_fa;
1226 
1227                         rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1228                                   tb->tb_id, &cfg->fc_nlinfo, nlflags);
1229 
1230                         hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1231 
1232                         alias_free_mem_rcu(fa);
1233 
1234                         fib_release_info(fi_drop);
1235                         if (state & FA_S_ACCESSED)
1236                                 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1237 
1238                         goto succeeded;
1239                 }
1240                 /* Error if we find a perfect match which
1241                  * uses the same scope, type, and nexthop
1242                  * information.
1243                  */
1244                 if (fa_match)
1245                         goto out;
1246 
1247                 if (cfg->fc_nlflags & NLM_F_APPEND) {
1248                         event = FIB_EVENT_ENTRY_APPEND;
1249                         nlflags |= NLM_F_APPEND;
1250                 } else {
1251                         fa = fa_first;
1252                 }
1253         }
1254         err = -ENOENT;
1255         if (!(cfg->fc_nlflags & NLM_F_CREATE))
1256                 goto out;
1257 
1258         nlflags |= NLM_F_CREATE;
1259         err = -ENOBUFS;
1260         new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1261         if (!new_fa)
1262                 goto out;
1263 
1264         new_fa->fa_info = fi;
1265         new_fa->fa_tos = tos;
1266         new_fa->fa_type = cfg->fc_type;
1267         new_fa->fa_state = 0;
1268         new_fa->fa_slen = slen;
1269         new_fa->tb_id = tb->tb_id;
1270         new_fa->fa_default = -1;
1271 
1272         err = call_fib_entry_notifiers(net, event, key, plen, new_fa, extack);
1273         if (err)
1274                 goto out_free_new_fa;
1275 
1276         /* Insert new entry to the list. */
1277         err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1278         if (err)
1279                 goto out_fib_notif;
1280 
1281         if (!plen)
1282                 tb->tb_num_default++;
1283 
1284         rt_cache_flush(cfg->fc_nlinfo.nl_net);
1285         rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1286                   &cfg->fc_nlinfo, nlflags);
1287 succeeded:
1288         return 0;
1289 
1290 out_fib_notif:
1291         /* notifier was sent that entry would be added to trie, but
1292          * the add failed and need to recover. Only failure for
1293          * fib_insert_alias is ENOMEM.
1294          */
1295         NL_SET_ERR_MSG(extack, "Failed to insert route into trie");
1296         call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key,
1297                                  plen, new_fa, NULL);
1298 out_free_new_fa:
1299         kmem_cache_free(fn_alias_kmem, new_fa);
1300 out:
1301         fib_release_info(fi);
1302 err:
1303         return err;
1304 }
1305 
1306 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1307 {
1308         t_key prefix = n->key;
1309 
1310         return (key ^ prefix) & (prefix | -prefix);
1311 }
1312 
1313 /* should be called with rcu_read_lock */
1314 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1315                      struct fib_result *res, int fib_flags)
1316 {
1317         struct trie *t = (struct trie *) tb->tb_data;
1318 #ifdef CONFIG_IP_FIB_TRIE_STATS
1319         struct trie_use_stats __percpu *stats = t->stats;
1320 #endif
1321         const t_key key = ntohl(flp->daddr);
1322         struct key_vector *n, *pn;
1323         struct fib_alias *fa;
1324         unsigned long index;
1325         t_key cindex;
1326 
1327         pn = t->kv;
1328         cindex = 0;
1329 
1330         n = get_child_rcu(pn, cindex);
1331         if (!n) {
1332                 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1333                 return -EAGAIN;
1334         }
1335 
1336 #ifdef CONFIG_IP_FIB_TRIE_STATS
1337         this_cpu_inc(stats->gets);
1338 #endif
1339 
1340         /* Step 1: Travel to the longest prefix match in the trie */
1341         for (;;) {
1342                 index = get_cindex(key, n);
1343 
1344                 /* This bit of code is a bit tricky but it combines multiple
1345                  * checks into a single check.  The prefix consists of the
1346                  * prefix plus zeros for the "bits" in the prefix. The index
1347                  * is the difference between the key and this value.  From
1348                  * this we can actually derive several pieces of data.
1349                  *   if (index >= (1ul << bits))
1350                  *     we have a mismatch in skip bits and failed
1351                  *   else
1352                  *     we know the value is cindex
1353                  *
1354                  * This check is safe even if bits == KEYLENGTH due to the
1355                  * fact that we can only allocate a node with 32 bits if a
1356                  * long is greater than 32 bits.
1357                  */
1358                 if (index >= (1ul << n->bits))
1359                         break;
1360 
1361                 /* we have found a leaf. Prefixes have already been compared */
1362                 if (IS_LEAF(n))
1363                         goto found;
1364 
1365                 /* only record pn and cindex if we are going to be chopping
1366                  * bits later.  Otherwise we are just wasting cycles.
1367                  */
1368                 if (n->slen > n->pos) {
1369                         pn = n;
1370                         cindex = index;
1371                 }
1372 
1373                 n = get_child_rcu(n, index);
1374                 if (unlikely(!n))
1375                         goto backtrace;
1376         }
1377 
1378         /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1379         for (;;) {
1380                 /* record the pointer where our next node pointer is stored */
1381                 struct key_vector __rcu **cptr = n->tnode;
1382 
1383                 /* This test verifies that none of the bits that differ
1384                  * between the key and the prefix exist in the region of
1385                  * the lsb and higher in the prefix.
1386                  */
1387                 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1388                         goto backtrace;
1389 
1390                 /* exit out and process leaf */
1391                 if (unlikely(IS_LEAF(n)))
1392                         break;
1393 
1394                 /* Don't bother recording parent info.  Since we are in
1395                  * prefix match mode we will have to come back to wherever
1396                  * we started this traversal anyway
1397                  */
1398 
1399                 while ((n = rcu_dereference(*cptr)) == NULL) {
1400 backtrace:
1401 #ifdef CONFIG_IP_FIB_TRIE_STATS
1402                         if (!n)
1403                                 this_cpu_inc(stats->null_node_hit);
1404 #endif
1405                         /* If we are at cindex 0 there are no more bits for
1406                          * us to strip at this level so we must ascend back
1407                          * up one level to see if there are any more bits to
1408                          * be stripped there.
1409                          */
1410                         while (!cindex) {
1411                                 t_key pkey = pn->key;
1412 
1413                                 /* If we don't have a parent then there is
1414                                  * nothing for us to do as we do not have any
1415                                  * further nodes to parse.
1416                                  */
1417                                 if (IS_TRIE(pn)) {
1418                                         trace_fib_table_lookup(tb->tb_id, flp,
1419                                                                NULL, -EAGAIN);
1420                                         return -EAGAIN;
1421                                 }
1422 #ifdef CONFIG_IP_FIB_TRIE_STATS
1423                                 this_cpu_inc(stats->backtrack);
1424 #endif
1425                                 /* Get Child's index */
1426                                 pn = node_parent_rcu(pn);
1427                                 cindex = get_index(pkey, pn);
1428                         }
1429 
1430                         /* strip the least significant bit from the cindex */
1431                         cindex &= cindex - 1;
1432 
1433                         /* grab pointer for next child node */
1434                         cptr = &pn->tnode[cindex];
1435                 }
1436         }
1437 
1438 found:
1439         /* this line carries forward the xor from earlier in the function */
1440         index = key ^ n->key;
1441 
1442         /* Step 3: Process the leaf, if that fails fall back to backtracing */
1443         hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1444                 struct fib_info *fi = fa->fa_info;
1445                 int nhsel, err;
1446 
1447                 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1448                         if (index >= (1ul << fa->fa_slen))
1449                                 continue;
1450                 }
1451                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1452                         continue;
1453                 if (fi->fib_dead)
1454                         continue;
1455                 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1456                         continue;
1457                 fib_alias_accessed(fa);
1458                 err = fib_props[fa->fa_type].error;
1459                 if (unlikely(err < 0)) {
1460 out_reject:
1461 #ifdef CONFIG_IP_FIB_TRIE_STATS
1462                         this_cpu_inc(stats->semantic_match_passed);
1463 #endif
1464                         trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1465                         return err;
1466                 }
1467                 if (fi->fib_flags & RTNH_F_DEAD)
1468                         continue;
1469 
1470                 if (unlikely(fi->nh && nexthop_is_blackhole(fi->nh))) {
1471                         err = fib_props[RTN_BLACKHOLE].error;
1472                         goto out_reject;
1473                 }
1474 
1475                 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1476                         struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
1477 
1478                         if (nhc->nhc_flags & RTNH_F_DEAD)
1479                                 continue;
1480                         if (ip_ignore_linkdown(nhc->nhc_dev) &&
1481                             nhc->nhc_flags & RTNH_F_LINKDOWN &&
1482                             !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1483                                 continue;
1484                         if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1485                                 if (flp->flowi4_oif &&
1486                                     flp->flowi4_oif != nhc->nhc_oif)
1487                                         continue;
1488                         }
1489 
1490                         if (!(fib_flags & FIB_LOOKUP_NOREF))
1491                                 refcount_inc(&fi->fib_clntref);
1492 
1493                         res->prefix = htonl(n->key);
1494                         res->prefixlen = KEYLENGTH - fa->fa_slen;
1495                         res->nh_sel = nhsel;
1496                         res->nhc = nhc;
1497                         res->type = fa->fa_type;
1498                         res->scope = fi->fib_scope;
1499                         res->fi = fi;
1500                         res->table = tb;
1501                         res->fa_head = &n->leaf;
1502 #ifdef CONFIG_IP_FIB_TRIE_STATS
1503                         this_cpu_inc(stats->semantic_match_passed);
1504 #endif
1505                         trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1506 
1507                         return err;
1508                 }
1509         }
1510 #ifdef CONFIG_IP_FIB_TRIE_STATS
1511         this_cpu_inc(stats->semantic_match_miss);
1512 #endif
1513         goto backtrace;
1514 }
1515 EXPORT_SYMBOL_GPL(fib_table_lookup);
1516 
1517 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1518                              struct key_vector *l, struct fib_alias *old)
1519 {
1520         /* record the location of the previous list_info entry */
1521         struct hlist_node **pprev = old->fa_list.pprev;
1522         struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1523 
1524         /* remove the fib_alias from the list */
1525         hlist_del_rcu(&old->fa_list);
1526 
1527         /* if we emptied the list this leaf will be freed and we can sort
1528          * out parent suffix lengths as a part of trie_rebalance
1529          */
1530         if (hlist_empty(&l->leaf)) {
1531                 if (tp->slen == l->slen)
1532                         node_pull_suffix(tp, tp->pos);
1533                 put_child_root(tp, l->key, NULL);
1534                 node_free(l);
1535                 trie_rebalance(t, tp);
1536                 return;
1537         }
1538 
1539         /* only access fa if it is pointing at the last valid hlist_node */
1540         if (*pprev)
1541                 return;
1542 
1543         /* update the trie with the latest suffix length */
1544         l->slen = fa->fa_slen;
1545         node_pull_suffix(tp, fa->fa_slen);
1546 }
1547 
1548 /* Caller must hold RTNL. */
1549 int fib_table_delete(struct net *net, struct fib_table *tb,
1550                      struct fib_config *cfg, struct netlink_ext_ack *extack)
1551 {
1552         struct trie *t = (struct trie *) tb->tb_data;
1553         struct fib_alias *fa, *fa_to_delete;
1554         struct key_vector *l, *tp;
1555         u8 plen = cfg->fc_dst_len;
1556         u8 slen = KEYLENGTH - plen;
1557         u8 tos = cfg->fc_tos;
1558         u32 key;
1559 
1560         key = ntohl(cfg->fc_dst);
1561 
1562         if (!fib_valid_key_len(key, plen, extack))
1563                 return -EINVAL;
1564 
1565         l = fib_find_node(t, &tp, key);
1566         if (!l)
1567                 return -ESRCH;
1568 
1569         fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1570         if (!fa)
1571                 return -ESRCH;
1572 
1573         pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1574 
1575         fa_to_delete = NULL;
1576         hlist_for_each_entry_from(fa, fa_list) {
1577                 struct fib_info *fi = fa->fa_info;
1578 
1579                 if ((fa->fa_slen != slen) ||
1580                     (fa->tb_id != tb->tb_id) ||
1581                     (fa->fa_tos != tos))
1582                         break;
1583 
1584                 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1585                     (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1586                      fa->fa_info->fib_scope == cfg->fc_scope) &&
1587                     (!cfg->fc_prefsrc ||
1588                      fi->fib_prefsrc == cfg->fc_prefsrc) &&
1589                     (!cfg->fc_protocol ||
1590                      fi->fib_protocol == cfg->fc_protocol) &&
1591                     fib_nh_match(cfg, fi, extack) == 0 &&
1592                     fib_metrics_match(cfg, fi)) {
1593                         fa_to_delete = fa;
1594                         break;
1595                 }
1596         }
1597 
1598         if (!fa_to_delete)
1599                 return -ESRCH;
1600 
1601         call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1602                                  fa_to_delete, extack);
1603         rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1604                   &cfg->fc_nlinfo, 0);
1605 
1606         if (!plen)
1607                 tb->tb_num_default--;
1608 
1609         fib_remove_alias(t, tp, l, fa_to_delete);
1610 
1611         if (fa_to_delete->fa_state & FA_S_ACCESSED)
1612                 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1613 
1614         fib_release_info(fa_to_delete->fa_info);
1615         alias_free_mem_rcu(fa_to_delete);
1616         return 0;
1617 }
1618 
1619 /* Scan for the next leaf starting at the provided key value */
1620 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1621 {
1622         struct key_vector *pn, *n = *tn;
1623         unsigned long cindex;
1624 
1625         /* this loop is meant to try and find the key in the trie */
1626         do {
1627                 /* record parent and next child index */
1628                 pn = n;
1629                 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1630 
1631                 if (cindex >> pn->bits)
1632                         break;
1633 
1634                 /* descend into the next child */
1635                 n = get_child_rcu(pn, cindex++);
1636                 if (!n)
1637                         break;
1638 
1639                 /* guarantee forward progress on the keys */
1640                 if (IS_LEAF(n) && (n->key >= key))
1641                         goto found;
1642         } while (IS_TNODE(n));
1643 
1644         /* this loop will search for the next leaf with a greater key */
1645         while (!IS_TRIE(pn)) {
1646                 /* if we exhausted the parent node we will need to climb */
1647                 if (cindex >= (1ul << pn->bits)) {
1648                         t_key pkey = pn->key;
1649 
1650                         pn = node_parent_rcu(pn);
1651                         cindex = get_index(pkey, pn) + 1;
1652                         continue;
1653                 }
1654 
1655                 /* grab the next available node */
1656                 n = get_child_rcu(pn, cindex++);
1657                 if (!n)
1658                         continue;
1659 
1660                 /* no need to compare keys since we bumped the index */
1661                 if (IS_LEAF(n))
1662                         goto found;
1663 
1664                 /* Rescan start scanning in new node */
1665                 pn = n;
1666                 cindex = 0;
1667         }
1668 
1669         *tn = pn;
1670         return NULL; /* Root of trie */
1671 found:
1672         /* if we are at the limit for keys just return NULL for the tnode */
1673         *tn = pn;
1674         return n;
1675 }
1676 
1677 static void fib_trie_free(struct fib_table *tb)
1678 {
1679         struct trie *t = (struct trie *)tb->tb_data;
1680         struct key_vector *pn = t->kv;
1681         unsigned long cindex = 1;
1682         struct hlist_node *tmp;
1683         struct fib_alias *fa;
1684 
1685         /* walk trie in reverse order and free everything */
1686         for (;;) {
1687                 struct key_vector *n;
1688 
1689                 if (!(cindex--)) {
1690                         t_key pkey = pn->key;
1691 
1692                         if (IS_TRIE(pn))
1693                                 break;
1694 
1695                         n = pn;
1696                         pn = node_parent(pn);
1697 
1698                         /* drop emptied tnode */
1699                         put_child_root(pn, n->key, NULL);
1700                         node_free(n);
1701 
1702                         cindex = get_index(pkey, pn);
1703 
1704                         continue;
1705                 }
1706 
1707                 /* grab the next available node */
1708                 n = get_child(pn, cindex);
1709                 if (!n)
1710                         continue;
1711 
1712                 if (IS_TNODE(n)) {
1713                         /* record pn and cindex for leaf walking */
1714                         pn = n;
1715                         cindex = 1ul << n->bits;
1716 
1717                         continue;
1718                 }
1719 
1720                 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1721                         hlist_del_rcu(&fa->fa_list);
1722                         alias_free_mem_rcu(fa);
1723                 }
1724 
1725                 put_child_root(pn, n->key, NULL);
1726                 node_free(n);
1727         }
1728 
1729 #ifdef CONFIG_IP_FIB_TRIE_STATS
1730         free_percpu(t->stats);
1731 #endif
1732         kfree(tb);
1733 }
1734 
1735 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1736 {
1737         struct trie *ot = (struct trie *)oldtb->tb_data;
1738         struct key_vector *l, *tp = ot->kv;
1739         struct fib_table *local_tb;
1740         struct fib_alias *fa;
1741         struct trie *lt;
1742         t_key key = 0;
1743 
1744         if (oldtb->tb_data == oldtb->__data)
1745                 return oldtb;
1746 
1747         local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1748         if (!local_tb)
1749                 return NULL;
1750 
1751         lt = (struct trie *)local_tb->tb_data;
1752 
1753         while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1754                 struct key_vector *local_l = NULL, *local_tp;
1755 
1756                 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1757                         struct fib_alias *new_fa;
1758 
1759                         if (local_tb->tb_id != fa->tb_id)
1760                                 continue;
1761 
1762                         /* clone fa for new local table */
1763                         new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1764                         if (!new_fa)
1765                                 goto out;
1766 
1767                         memcpy(new_fa, fa, sizeof(*fa));
1768 
1769                         /* insert clone into table */
1770                         if (!local_l)
1771                                 local_l = fib_find_node(lt, &local_tp, l->key);
1772 
1773                         if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1774                                              NULL, l->key)) {
1775                                 kmem_cache_free(fn_alias_kmem, new_fa);
1776                                 goto out;
1777                         }
1778                 }
1779 
1780                 /* stop loop if key wrapped back to 0 */
1781                 key = l->key + 1;
1782                 if (key < l->key)
1783                         break;
1784         }
1785 
1786         return local_tb;
1787 out:
1788         fib_trie_free(local_tb);
1789 
1790         return NULL;
1791 }
1792 
1793 /* Caller must hold RTNL */
1794 void fib_table_flush_external(struct fib_table *tb)
1795 {
1796         struct trie *t = (struct trie *)tb->tb_data;
1797         struct key_vector *pn = t->kv;
1798         unsigned long cindex = 1;
1799         struct hlist_node *tmp;
1800         struct fib_alias *fa;
1801 
1802         /* walk trie in reverse order */
1803         for (;;) {
1804                 unsigned char slen = 0;
1805                 struct key_vector *n;
1806 
1807                 if (!(cindex--)) {
1808                         t_key pkey = pn->key;
1809 
1810                         /* cannot resize the trie vector */
1811                         if (IS_TRIE(pn))
1812                                 break;
1813 
1814                         /* update the suffix to address pulled leaves */
1815                         if (pn->slen > pn->pos)
1816                                 update_suffix(pn);
1817 
1818                         /* resize completed node */
1819                         pn = resize(t, pn);
1820                         cindex = get_index(pkey, pn);
1821 
1822                         continue;
1823                 }
1824 
1825                 /* grab the next available node */
1826                 n = get_child(pn, cindex);
1827                 if (!n)
1828                         continue;
1829 
1830                 if (IS_TNODE(n)) {
1831                         /* record pn and cindex for leaf walking */
1832                         pn = n;
1833                         cindex = 1ul << n->bits;
1834 
1835                         continue;
1836                 }
1837 
1838                 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1839                         /* if alias was cloned to local then we just
1840                          * need to remove the local copy from main
1841                          */
1842                         if (tb->tb_id != fa->tb_id) {
1843                                 hlist_del_rcu(&fa->fa_list);
1844                                 alias_free_mem_rcu(fa);
1845                                 continue;
1846                         }
1847 
1848                         /* record local slen */
1849                         slen = fa->fa_slen;
1850                 }
1851 
1852                 /* update leaf slen */
1853                 n->slen = slen;
1854 
1855                 if (hlist_empty(&n->leaf)) {
1856                         put_child_root(pn, n->key, NULL);
1857                         node_free(n);
1858                 }
1859         }
1860 }
1861 
1862 /* Caller must hold RTNL. */
1863 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1864 {
1865         struct trie *t = (struct trie *)tb->tb_data;
1866         struct key_vector *pn = t->kv;
1867         unsigned long cindex = 1;
1868         struct hlist_node *tmp;
1869         struct fib_alias *fa;
1870         int found = 0;
1871 
1872         /* walk trie in reverse order */
1873         for (;;) {
1874                 unsigned char slen = 0;
1875                 struct key_vector *n;
1876 
1877                 if (!(cindex--)) {
1878                         t_key pkey = pn->key;
1879 
1880                         /* cannot resize the trie vector */
1881                         if (IS_TRIE(pn))
1882                                 break;
1883 
1884                         /* update the suffix to address pulled leaves */
1885                         if (pn->slen > pn->pos)
1886                                 update_suffix(pn);
1887 
1888                         /* resize completed node */
1889                         pn = resize(t, pn);
1890                         cindex = get_index(pkey, pn);
1891 
1892                         continue;
1893                 }
1894 
1895                 /* grab the next available node */
1896                 n = get_child(pn, cindex);
1897                 if (!n)
1898                         continue;
1899 
1900                 if (IS_TNODE(n)) {
1901                         /* record pn and cindex for leaf walking */
1902                         pn = n;
1903                         cindex = 1ul << n->bits;
1904 
1905                         continue;
1906                 }
1907 
1908                 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1909                         struct fib_info *fi = fa->fa_info;
1910 
1911                         if (!fi || tb->tb_id != fa->tb_id ||
1912                             (!(fi->fib_flags & RTNH_F_DEAD) &&
1913                              !fib_props[fa->fa_type].error)) {
1914                                 slen = fa->fa_slen;
1915                                 continue;
1916                         }
1917 
1918                         /* Do not flush error routes if network namespace is
1919                          * not being dismantled
1920                          */
1921                         if (!flush_all && fib_props[fa->fa_type].error) {
1922                                 slen = fa->fa_slen;
1923                                 continue;
1924                         }
1925 
1926                         call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1927                                                  n->key,
1928                                                  KEYLENGTH - fa->fa_slen, fa,
1929                                                  NULL);
1930                         hlist_del_rcu(&fa->fa_list);
1931                         fib_release_info(fa->fa_info);
1932                         alias_free_mem_rcu(fa);
1933                         found++;
1934                 }
1935 
1936                 /* update leaf slen */
1937                 n->slen = slen;
1938 
1939                 if (hlist_empty(&n->leaf)) {
1940                         put_child_root(pn, n->key, NULL);
1941                         node_free(n);
1942                 }
1943         }
1944 
1945         pr_debug("trie_flush found=%d\n", found);
1946         return found;
1947 }
1948 
1949 /* derived from fib_trie_free */
1950 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
1951                                      struct nl_info *info)
1952 {
1953         struct trie *t = (struct trie *)tb->tb_data;
1954         struct key_vector *pn = t->kv;
1955         unsigned long cindex = 1;
1956         struct fib_alias *fa;
1957 
1958         for (;;) {
1959                 struct key_vector *n;
1960 
1961                 if (!(cindex--)) {
1962                         t_key pkey = pn->key;
1963 
1964                         if (IS_TRIE(pn))
1965                                 break;
1966 
1967                         pn = node_parent(pn);
1968                         cindex = get_index(pkey, pn);
1969                         continue;
1970                 }
1971 
1972                 /* grab the next available node */
1973                 n = get_child(pn, cindex);
1974                 if (!n)
1975                         continue;
1976 
1977                 if (IS_TNODE(n)) {
1978                         /* record pn and cindex for leaf walking */
1979                         pn = n;
1980                         cindex = 1ul << n->bits;
1981 
1982                         continue;
1983                 }
1984 
1985                 hlist_for_each_entry(fa, &n->leaf, fa_list) {
1986                         struct fib_info *fi = fa->fa_info;
1987 
1988                         if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
1989                                 continue;
1990 
1991                         rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
1992                                   KEYLENGTH - fa->fa_slen, tb->tb_id,
1993                                   info, NLM_F_REPLACE);
1994 
1995                         /* call_fib_entry_notifiers will be removed when
1996                          * in-kernel notifier is implemented and supported
1997                          * for nexthop objects
1998                          */
1999                         call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
2000                                                  n->key,
2001                                                  KEYLENGTH - fa->fa_slen, fa,
2002                                                  NULL);
2003                 }
2004         }
2005 }
2006 
2007 void fib_info_notify_update(struct net *net, struct nl_info *info)
2008 {
2009         unsigned int h;
2010 
2011         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2012                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2013                 struct fib_table *tb;
2014 
2015                 hlist_for_each_entry_rcu(tb, head, tb_hlist)
2016                         __fib_info_notify_update(net, tb, info);
2017         }
2018 }
2019 
2020 static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2021                            struct notifier_block *nb,
2022                            struct netlink_ext_ack *extack)
2023 {
2024         struct fib_alias *fa;
2025         int err;
2026 
2027         hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2028                 struct fib_info *fi = fa->fa_info;
2029 
2030                 if (!fi)
2031                         continue;
2032 
2033                 /* local and main table can share the same trie,
2034                  * so don't notify twice for the same entry.
2035                  */
2036                 if (tb->tb_id != fa->tb_id)
2037                         continue;
2038 
2039                 err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_ADD, l->key,
2040                                               KEYLENGTH - fa->fa_slen,
2041                                               fa, extack);
2042                 if (err)
2043                         return err;
2044         }
2045         return 0;
2046 }
2047 
2048 static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2049                             struct netlink_ext_ack *extack)
2050 {
2051         struct trie *t = (struct trie *)tb->tb_data;
2052         struct key_vector *l, *tp = t->kv;
2053         t_key key = 0;
2054         int err;
2055 
2056         while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2057                 err = fib_leaf_notify(l, tb, nb, extack);
2058                 if (err)
2059                         return err;
2060 
2061                 key = l->key + 1;
2062                 /* stop in case of wrap around */
2063                 if (key < l->key)
2064                         break;
2065         }
2066         return 0;
2067 }
2068 
2069 int fib_notify(struct net *net, struct notifier_block *nb,
2070                struct netlink_ext_ack *extack)
2071 {
2072         unsigned int h;
2073         int err;
2074 
2075         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2076                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2077                 struct fib_table *tb;
2078 
2079                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2080                         err = fib_table_notify(tb, nb, extack);
2081                         if (err)
2082                                 return err;
2083                 }
2084         }
2085         return 0;
2086 }
2087 
2088 static void __trie_free_rcu(struct rcu_head *head)
2089 {
2090         struct fib_table *tb = container_of(head, struct fib_table, rcu);
2091 #ifdef CONFIG_IP_FIB_TRIE_STATS
2092         struct trie *t = (struct trie *)tb->tb_data;
2093 
2094         if (tb->tb_data == tb->__data)
2095                 free_percpu(t->stats);
2096 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2097         kfree(tb);
2098 }
2099 
2100 void fib_free_table(struct fib_table *tb)
2101 {
2102         call_rcu(&tb->rcu, __trie_free_rcu);
2103 }
2104 
2105 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2106                              struct sk_buff *skb, struct netlink_callback *cb,
2107                              struct fib_dump_filter *filter)
2108 {
2109         unsigned int flags = NLM_F_MULTI;
2110         __be32 xkey = htonl(l->key);
2111         int i, s_i, i_fa, s_fa, err;
2112         struct fib_alias *fa;
2113 
2114         if (filter->filter_set ||
2115             !filter->dump_exceptions || !filter->dump_routes)
2116                 flags |= NLM_F_DUMP_FILTERED;
2117 
2118         s_i = cb->args[4];
2119         s_fa = cb->args[5];
2120         i = 0;
2121 
2122         /* rcu_read_lock is hold by caller */
2123         hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2124                 struct fib_info *fi = fa->fa_info;
2125 
2126                 if (i < s_i)
2127                         goto next;
2128 
2129                 i_fa = 0;
2130 
2131                 if (tb->tb_id != fa->tb_id)
2132                         goto next;
2133 
2134                 if (filter->filter_set) {
2135                         if (filter->rt_type && fa->fa_type != filter->rt_type)
2136                                 goto next;
2137 
2138                         if ((filter->protocol &&
2139                              fi->fib_protocol != filter->protocol))
2140                                 goto next;
2141 
2142                         if (filter->dev &&
2143                             !fib_info_nh_uses_dev(fi, filter->dev))
2144                                 goto next;
2145                 }
2146 
2147                 if (filter->dump_routes) {
2148                         if (!s_fa) {
2149                                 err = fib_dump_info(skb,
2150                                                     NETLINK_CB(cb->skb).portid,
2151                                                     cb->nlh->nlmsg_seq,
2152                                                     RTM_NEWROUTE,
2153                                                     tb->tb_id, fa->fa_type,
2154                                                     xkey,
2155                                                     KEYLENGTH - fa->fa_slen,
2156                                                     fa->fa_tos, fi, flags);
2157                                 if (err < 0)
2158                                         goto stop;
2159                         }
2160 
2161                         i_fa++;
2162                 }
2163 
2164                 if (filter->dump_exceptions) {
2165                         err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2166                                                  &i_fa, s_fa, flags);
2167                         if (err < 0)
2168                                 goto stop;
2169                 }
2170 
2171 next:
2172                 i++;
2173         }
2174 
2175         cb->args[4] = i;
2176         return skb->len;
2177 
2178 stop:
2179         cb->args[4] = i;
2180         cb->args[5] = i_fa;
2181         return err;
2182 }
2183 
2184 /* rcu_read_lock needs to be hold by caller from readside */
2185 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2186                    struct netlink_callback *cb, struct fib_dump_filter *filter)
2187 {
2188         struct trie *t = (struct trie *)tb->tb_data;
2189         struct key_vector *l, *tp = t->kv;
2190         /* Dump starting at last key.
2191          * Note: 0.0.0.0/0 (ie default) is first key.
2192          */
2193         int count = cb->args[2];
2194         t_key key = cb->args[3];
2195 
2196         /* First time here, count and key are both always 0. Count > 0
2197          * and key == 0 means the dump has wrapped around and we are done.
2198          */
2199         if (count && !key)
2200                 return skb->len;
2201 
2202         while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2203                 int err;
2204 
2205                 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2206                 if (err < 0) {
2207                         cb->args[3] = key;
2208                         cb->args[2] = count;
2209                         return err;
2210                 }
2211 
2212                 ++count;
2213                 key = l->key + 1;
2214 
2215                 memset(&cb->args[4], 0,
2216                        sizeof(cb->args) - 4*sizeof(cb->args[0]));
2217 
2218                 /* stop loop if key wrapped back to 0 */
2219                 if (key < l->key)
2220                         break;
2221         }
2222 
2223         cb->args[3] = key;
2224         cb->args[2] = count;
2225 
2226         return skb->len;
2227 }
2228 
2229 void __init fib_trie_init(void)
2230 {
2231         fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2232                                           sizeof(struct fib_alias),
2233                                           0, SLAB_PANIC, NULL);
2234 
2235         trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2236                                            LEAF_SIZE,
2237                                            0, SLAB_PANIC, NULL);
2238 }
2239 
2240 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2241 {
2242         struct fib_table *tb;
2243         struct trie *t;
2244         size_t sz = sizeof(*tb);
2245 
2246         if (!alias)
2247                 sz += sizeof(struct trie);
2248 
2249         tb = kzalloc(sz, GFP_KERNEL);
2250         if (!tb)
2251                 return NULL;
2252 
2253         tb->tb_id = id;
2254         tb->tb_num_default = 0;
2255         tb->tb_data = (alias ? alias->__data : tb->__data);
2256 
2257         if (alias)
2258                 return tb;
2259 
2260         t = (struct trie *) tb->tb_data;
2261         t->kv[0].pos = KEYLENGTH;
2262         t->kv[0].slen = KEYLENGTH;
2263 #ifdef CONFIG_IP_FIB_TRIE_STATS
2264         t->stats = alloc_percpu(struct trie_use_stats);
2265         if (!t->stats) {
2266                 kfree(tb);
2267                 tb = NULL;
2268         }
2269 #endif
2270 
2271         return tb;
2272 }
2273 
2274 #ifdef CONFIG_PROC_FS
2275 /* Depth first Trie walk iterator */
2276 struct fib_trie_iter {
2277         struct seq_net_private p;
2278         struct fib_table *tb;
2279         struct key_vector *tnode;
2280         unsigned int index;
2281         unsigned int depth;
2282 };
2283 
2284 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2285 {
2286         unsigned long cindex = iter->index;
2287         struct key_vector *pn = iter->tnode;
2288         t_key pkey;
2289 
2290         pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2291                  iter->tnode, iter->index, iter->depth);
2292 
2293         while (!IS_TRIE(pn)) {
2294                 while (cindex < child_length(pn)) {
2295                         struct key_vector *n = get_child_rcu(pn, cindex++);
2296 
2297                         if (!n)
2298                                 continue;
2299 
2300                         if (IS_LEAF(n)) {
2301                                 iter->tnode = pn;
2302                                 iter->index = cindex;
2303                         } else {
2304                                 /* push down one level */
2305                                 iter->tnode = n;
2306                                 iter->index = 0;
2307                                 ++iter->depth;
2308                         }
2309 
2310                         return n;
2311                 }
2312 
2313                 /* Current node exhausted, pop back up */
2314                 pkey = pn->key;
2315                 pn = node_parent_rcu(pn);
2316                 cindex = get_index(pkey, pn) + 1;
2317                 --iter->depth;
2318         }
2319 
2320         /* record root node so further searches know we are done */
2321         iter->tnode = pn;
2322         iter->index = 0;
2323 
2324         return NULL;
2325 }
2326 
2327 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2328                                              struct trie *t)
2329 {
2330         struct key_vector *n, *pn;
2331 
2332         if (!t)
2333                 return NULL;
2334 
2335         pn = t->kv;
2336         n = rcu_dereference(pn->tnode[0]);
2337         if (!n)
2338                 return NULL;
2339 
2340         if (IS_TNODE(n)) {
2341                 iter->tnode = n;
2342                 iter->index = 0;
2343                 iter->depth = 1;
2344         } else {
2345                 iter->tnode = pn;
2346                 iter->index = 0;
2347                 iter->depth = 0;
2348         }
2349 
2350         return n;
2351 }
2352 
2353 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2354 {
2355         struct key_vector *n;
2356         struct fib_trie_iter iter;
2357 
2358         memset(s, 0, sizeof(*s));
2359 
2360         rcu_read_lock();
2361         for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2362                 if (IS_LEAF(n)) {
2363                         struct fib_alias *fa;
2364 
2365                         s->leaves++;
2366                         s->totdepth += iter.depth;
2367                         if (iter.depth > s->maxdepth)
2368                                 s->maxdepth = iter.depth;
2369 
2370                         hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2371                                 ++s->prefixes;
2372                 } else {
2373                         s->tnodes++;
2374                         if (n->bits < MAX_STAT_DEPTH)
2375                                 s->nodesizes[n->bits]++;
2376                         s->nullpointers += tn_info(n)->empty_children;
2377                 }
2378         }
2379         rcu_read_unlock();
2380 }
2381 
2382 /*
2383  *      This outputs /proc/net/fib_triestats
2384  */
2385 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2386 {
2387         unsigned int i, max, pointers, bytes, avdepth;
2388 
2389         if (stat->leaves)
2390                 avdepth = stat->totdepth*100 / stat->leaves;
2391         else
2392                 avdepth = 0;
2393 
2394         seq_printf(seq, "\tAver depth:     %u.%02d\n",
2395                    avdepth / 100, avdepth % 100);
2396         seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2397 
2398         seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2399         bytes = LEAF_SIZE * stat->leaves;
2400 
2401         seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2402         bytes += sizeof(struct fib_alias) * stat->prefixes;
2403 
2404         seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2405         bytes += TNODE_SIZE(0) * stat->tnodes;
2406 
2407         max = MAX_STAT_DEPTH;
2408         while (max > 0 && stat->nodesizes[max-1] == 0)
2409                 max--;
2410 
2411         pointers = 0;
2412         for (i = 1; i < max; i++)
2413                 if (stat->nodesizes[i] != 0) {
2414                         seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2415                         pointers += (1<<i) * stat->nodesizes[i];
2416                 }
2417         seq_putc(seq, '\n');
2418         seq_printf(seq, "\tPointers: %u\n", pointers);
2419 
2420         bytes += sizeof(struct key_vector *) * pointers;
2421         seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2422         seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2423 }
2424 
2425 #ifdef CONFIG_IP_FIB_TRIE_STATS
2426 static void trie_show_usage(struct seq_file *seq,
2427                             const struct trie_use_stats __percpu *stats)
2428 {
2429         struct trie_use_stats s = { 0 };
2430         int cpu;
2431 
2432         /* loop through all of the CPUs and gather up the stats */
2433         for_each_possible_cpu(cpu) {
2434                 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2435 
2436                 s.gets += pcpu->gets;
2437                 s.backtrack += pcpu->backtrack;
2438                 s.semantic_match_passed += pcpu->semantic_match_passed;
2439                 s.semantic_match_miss += pcpu->semantic_match_miss;
2440                 s.null_node_hit += pcpu->null_node_hit;
2441                 s.resize_node_skipped += pcpu->resize_node_skipped;
2442         }
2443 
2444         seq_printf(seq, "\nCounters:\n---------\n");
2445         seq_printf(seq, "gets = %u\n", s.gets);
2446         seq_printf(seq, "backtracks = %u\n", s.backtrack);
2447         seq_printf(seq, "semantic match passed = %u\n",
2448                    s.semantic_match_passed);
2449         seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2450         seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2451         seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2452 }
2453 #endif /*  CONFIG_IP_FIB_TRIE_STATS */
2454 
2455 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2456 {
2457         if (tb->tb_id == RT_TABLE_LOCAL)
2458                 seq_puts(seq, "Local:\n");
2459         else if (tb->tb_id == RT_TABLE_MAIN)
2460                 seq_puts(seq, "Main:\n");
2461         else
2462                 seq_printf(seq, "Id %d:\n", tb->tb_id);
2463 }
2464 
2465 
2466 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2467 {
2468         struct net *net = (struct net *)seq->private;
2469         unsigned int h;
2470 
2471         seq_printf(seq,
2472                    "Basic info: size of leaf:"
2473                    " %zd bytes, size of tnode: %zd bytes.\n",
2474                    LEAF_SIZE, TNODE_SIZE(0));
2475 
2476         rcu_read_lock();
2477         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2478                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2479                 struct fib_table *tb;
2480 
2481                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2482                         struct trie *t = (struct trie *) tb->tb_data;
2483                         struct trie_stat stat;
2484 
2485                         if (!t)
2486                                 continue;
2487 
2488                         fib_table_print(seq, tb);
2489 
2490                         trie_collect_stats(t, &stat);
2491                         trie_show_stats(seq, &stat);
2492 #ifdef CONFIG_IP_FIB_TRIE_STATS
2493                         trie_show_usage(seq, t->stats);
2494 #endif
2495                 }
2496                 cond_resched_rcu();
2497         }
2498         rcu_read_unlock();
2499 
2500         return 0;
2501 }
2502 
2503 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2504 {
2505         struct fib_trie_iter *iter = seq->private;
2506         struct net *net = seq_file_net(seq);
2507         loff_t idx = 0;
2508         unsigned int h;
2509 
2510         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2511                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2512                 struct fib_table *tb;
2513 
2514                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2515                         struct key_vector *n;
2516 
2517                         for (n = fib_trie_get_first(iter,
2518                                                     (struct trie *) tb->tb_data);
2519                              n; n = fib_trie_get_next(iter))
2520                                 if (pos == idx++) {
2521                                         iter->tb = tb;
2522                                         return n;
2523                                 }
2524                 }
2525         }
2526 
2527         return NULL;
2528 }
2529 
2530 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2531         __acquires(RCU)
2532 {
2533         rcu_read_lock();
2534         return fib_trie_get_idx(seq, *pos);
2535 }
2536 
2537 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2538 {
2539         struct fib_trie_iter *iter = seq->private;
2540         struct net *net = seq_file_net(seq);
2541         struct fib_table *tb = iter->tb;
2542         struct hlist_node *tb_node;
2543         unsigned int h;
2544         struct key_vector *n;
2545 
2546         ++*pos;
2547         /* next node in same table */
2548         n = fib_trie_get_next(iter);
2549         if (n)
2550                 return n;
2551 
2552         /* walk rest of this hash chain */
2553         h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2554         while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2555                 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2556                 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2557                 if (n)
2558                         goto found;
2559         }
2560 
2561         /* new hash chain */
2562         while (++h < FIB_TABLE_HASHSZ) {
2563                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2564                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2565                         n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2566                         if (n)
2567                                 goto found;
2568                 }
2569         }
2570         return NULL;
2571 
2572 found:
2573         iter->tb = tb;
2574         return n;
2575 }
2576 
2577 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2578         __releases(RCU)
2579 {
2580         rcu_read_unlock();
2581 }
2582 
2583 static void seq_indent(struct seq_file *seq, int n)
2584 {
2585         while (n-- > 0)
2586                 seq_puts(seq, "   ");
2587 }
2588 
2589 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2590 {
2591         switch (s) {
2592         case RT_SCOPE_UNIVERSE: return "universe";
2593         case RT_SCOPE_SITE:     return "site";
2594         case RT_SCOPE_LINK:     return "link";
2595         case RT_SCOPE_HOST:     return "host";
2596         case RT_SCOPE_NOWHERE:  return "nowhere";
2597         default:
2598                 snprintf(buf, len, "scope=%d", s);
2599                 return buf;
2600         }
2601 }
2602 
2603 static const char *const rtn_type_names[__RTN_MAX] = {
2604         [RTN_UNSPEC] = "UNSPEC",
2605         [RTN_UNICAST] = "UNICAST",
2606         [RTN_LOCAL] = "LOCAL",
2607         [RTN_BROADCAST] = "BROADCAST",
2608         [RTN_ANYCAST] = "ANYCAST",
2609         [RTN_MULTICAST] = "MULTICAST",
2610         [RTN_BLACKHOLE] = "BLACKHOLE",
2611         [RTN_UNREACHABLE] = "UNREACHABLE",
2612         [RTN_PROHIBIT] = "PROHIBIT",
2613         [RTN_THROW] = "THROW",
2614         [RTN_NAT] = "NAT",
2615         [RTN_XRESOLVE] = "XRESOLVE",
2616 };
2617 
2618 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2619 {
2620         if (t < __RTN_MAX && rtn_type_names[t])
2621                 return rtn_type_names[t];
2622         snprintf(buf, len, "type %u", t);
2623         return buf;
2624 }
2625 
2626 /* Pretty print the trie */
2627 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2628 {
2629         const struct fib_trie_iter *iter = seq->private;
2630         struct key_vector *n = v;
2631 
2632         if (IS_TRIE(node_parent_rcu(n)))
2633                 fib_table_print(seq, iter->tb);
2634 
2635         if (IS_TNODE(n)) {
2636                 __be32 prf = htonl(n->key);
2637 
2638                 seq_indent(seq, iter->depth-1);
2639                 seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2640                            &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2641                            tn_info(n)->full_children,
2642                            tn_info(n)->empty_children);
2643         } else {
2644                 __be32 val = htonl(n->key);
2645                 struct fib_alias *fa;
2646 
2647                 seq_indent(seq, iter->depth);
2648                 seq_printf(seq, "  |-- %pI4\n", &val);
2649 
2650                 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2651                         char buf1[32], buf2[32];
2652 
2653                         seq_indent(seq, iter->depth + 1);
2654                         seq_printf(seq, "  /%zu %s %s",
2655                                    KEYLENGTH - fa->fa_slen,
2656                                    rtn_scope(buf1, sizeof(buf1),
2657                                              fa->fa_info->fib_scope),
2658                                    rtn_type(buf2, sizeof(buf2),
2659                                             fa->fa_type));
2660                         if (fa->fa_tos)
2661                                 seq_printf(seq, " tos=%d", fa->fa_tos);
2662                         seq_putc(seq, '\n');
2663                 }
2664         }
2665 
2666         return 0;
2667 }
2668 
2669 static const struct seq_operations fib_trie_seq_ops = {
2670         .start  = fib_trie_seq_start,
2671         .next   = fib_trie_seq_next,
2672         .stop   = fib_trie_seq_stop,
2673         .show   = fib_trie_seq_show,
2674 };
2675 
2676 struct fib_route_iter {
2677         struct seq_net_private p;
2678         struct fib_table *main_tb;
2679         struct key_vector *tnode;
2680         loff_t  pos;
2681         t_key   key;
2682 };
2683 
2684 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2685                                             loff_t pos)
2686 {
2687         struct key_vector *l, **tp = &iter->tnode;
2688         t_key key;
2689 
2690         /* use cached location of previously found key */
2691         if (iter->pos > 0 && pos >= iter->pos) {
2692                 key = iter->key;
2693         } else {
2694                 iter->pos = 1;
2695                 key = 0;
2696         }
2697 
2698         pos -= iter->pos;
2699 
2700         while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2701                 key = l->key + 1;
2702                 iter->pos++;
2703                 l = NULL;
2704 
2705                 /* handle unlikely case of a key wrap */
2706                 if (!key)
2707                         break;
2708         }
2709 
2710         if (l)
2711                 iter->key = l->key;     /* remember it */
2712         else
2713                 iter->pos = 0;          /* forget it */
2714 
2715         return l;
2716 }
2717 
2718 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2719         __acquires(RCU)
2720 {
2721         struct fib_route_iter *iter = seq->private;
2722         struct fib_table *tb;
2723         struct trie *t;
2724 
2725         rcu_read_lock();
2726 
2727         tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2728         if (!tb)
2729                 return NULL;
2730 
2731         iter->main_tb = tb;
2732         t = (struct trie *)tb->tb_data;
2733         iter->tnode = t->kv;
2734 
2735         if (*pos != 0)
2736                 return fib_route_get_idx(iter, *pos);
2737 
2738         iter->pos = 0;
2739         iter->key = KEY_MAX;
2740 
2741         return SEQ_START_TOKEN;
2742 }
2743 
2744 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2745 {
2746         struct fib_route_iter *iter = seq->private;
2747         struct key_vector *l = NULL;
2748         t_key key = iter->key + 1;
2749 
2750         ++*pos;
2751 
2752         /* only allow key of 0 for start of sequence */
2753         if ((v == SEQ_START_TOKEN) || key)
2754                 l = leaf_walk_rcu(&iter->tnode, key);
2755 
2756         if (l) {
2757                 iter->key = l->key;
2758                 iter->pos++;
2759         } else {
2760                 iter->pos = 0;
2761         }
2762 
2763         return l;
2764 }
2765 
2766 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2767         __releases(RCU)
2768 {
2769         rcu_read_unlock();
2770 }
2771 
2772 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2773 {
2774         unsigned int flags = 0;
2775 
2776         if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2777                 flags = RTF_REJECT;
2778         if (fi) {
2779                 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2780 
2781                 if (nhc->nhc_gw.ipv4)
2782                         flags |= RTF_GATEWAY;
2783         }
2784         if (mask == htonl(0xFFFFFFFF))
2785                 flags |= RTF_HOST;
2786         flags |= RTF_UP;
2787         return flags;
2788 }
2789 
2790 /*
2791  *      This outputs /proc/net/route.
2792  *      The format of the file is not supposed to be changed
2793  *      and needs to be same as fib_hash output to avoid breaking
2794  *      legacy utilities
2795  */
2796 static int fib_route_seq_show(struct seq_file *seq, void *v)
2797 {
2798         struct fib_route_iter *iter = seq->private;
2799         struct fib_table *tb = iter->main_tb;
2800         struct fib_alias *fa;
2801         struct key_vector *l = v;
2802         __be32 prefix;
2803 
2804         if (v == SEQ_START_TOKEN) {
2805                 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2806                            "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2807                            "\tWindow\tIRTT");
2808                 return 0;
2809         }
2810 
2811         prefix = htonl(l->key);
2812 
2813         hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2814                 struct fib_info *fi = fa->fa_info;
2815                 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2816                 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2817 
2818                 if ((fa->fa_type == RTN_BROADCAST) ||
2819                     (fa->fa_type == RTN_MULTICAST))
2820                         continue;
2821 
2822                 if (fa->tb_id != tb->tb_id)
2823                         continue;
2824 
2825                 seq_setwidth(seq, 127);
2826 
2827                 if (fi) {
2828                         struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2829                         __be32 gw = 0;
2830 
2831                         if (nhc->nhc_gw_family == AF_INET)
2832                                 gw = nhc->nhc_gw.ipv4;
2833 
2834                         seq_printf(seq,
2835                                    "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2836                                    "%d\t%08X\t%d\t%u\t%u",
2837                                    nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2838                                    prefix, gw, flags, 0, 0,
2839                                    fi->fib_priority,
2840                                    mask,
2841                                    (fi->fib_advmss ?
2842                                     fi->fib_advmss + 40 : 0),
2843                                    fi->fib_window,
2844                                    fi->fib_rtt >> 3);
2845                 } else {
2846                         seq_printf(seq,
2847                                    "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2848                                    "%d\t%08X\t%d\t%u\t%u",
2849                                    prefix, 0, flags, 0, 0, 0,
2850                                    mask, 0, 0, 0);
2851                 }
2852                 seq_pad(seq, '\n');
2853         }
2854 
2855         return 0;
2856 }
2857 
2858 static const struct seq_operations fib_route_seq_ops = {
2859         .start  = fib_route_seq_start,
2860         .next   = fib_route_seq_next,
2861         .stop   = fib_route_seq_stop,
2862         .show   = fib_route_seq_show,
2863 };
2864 
2865 int __net_init fib_proc_init(struct net *net)
2866 {
2867         if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2868                         sizeof(struct fib_trie_iter)))
2869                 goto out1;
2870 
2871         if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
2872                         fib_triestat_seq_show, NULL))
2873                 goto out2;
2874 
2875         if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
2876                         sizeof(struct fib_route_iter)))
2877                 goto out3;
2878 
2879         return 0;
2880 
2881 out3:
2882         remove_proc_entry("fib_triestat", net->proc_net);
2883 out2:
2884         remove_proc_entry("fib_trie", net->proc_net);
2885 out1:
2886         return -ENOMEM;
2887 }
2888 
2889 void __net_exit fib_proc_exit(struct net *net)
2890 {
2891         remove_proc_entry("fib_trie", net->proc_net);
2892         remove_proc_entry("fib_triestat", net->proc_net);
2893         remove_proc_entry("route", net->proc_net);
2894 }
2895 
2896 #endif /* CONFIG_PROC_FS */
2897 

~ [ 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