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 * https://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 #include <linux/cache.h>
39 #include <linux/uaccess.h>
40 #include <linux/bitops.h>
41 #include <linux/types.h>
42 #include <linux/kernel.h>
43 #include <linux/mm.h>
44 #include <linux/string.h>
45 #include <linux/socket.h>
46 #include <linux/sockios.h>
47 #include <linux/errno.h>
48 #include <linux/in.h>
49 #include <linux/inet.h>
50 #include <linux/inetdevice.h>
51 #include <linux/netdevice.h>
52 #include <linux/if_arp.h>
53 #include <linux/proc_fs.h>
54 #include <linux/rcupdate.h>
55 #include <linux/rcupdate_wait.h>
56 #include <linux/skbuff.h>
57 #include <linux/netlink.h>
58 #include <linux/init.h>
59 #include <linux/list.h>
60 #include <linux/slab.h>
61 #include <linux/export.h>
62 #include <linux/vmalloc.h>
63 #include <linux/notifier.h>
64 #include <net/net_namespace.h>
65 #include <net/inet_dscp.h>
66 #include <net/ip.h>
67 #include <net/protocol.h>
68 #include <net/route.h>
69 #include <net/tcp.h>
70 #include <net/sock.h>
71 #include <net/ip_fib.h>
72 #include <net/fib_notifier.h>
73 #include <trace/events/fib.h>
74 #include "fib_lookup.h"
75
call_fib_entry_notifier(struct notifier_block * nb,enum fib_event_type event_type,u32 dst,int dst_len,struct fib_alias * fa,struct netlink_ext_ack * extack)76 static int call_fib_entry_notifier(struct notifier_block *nb,
77 enum fib_event_type event_type, u32 dst,
78 int dst_len, struct fib_alias *fa,
79 struct netlink_ext_ack *extack)
80 {
81 struct fib_entry_notifier_info info = {
82 .info.extack = extack,
83 .dst = dst,
84 .dst_len = dst_len,
85 .fi = fa->fa_info,
86 .dscp = fa->fa_dscp,
87 .type = fa->fa_type,
88 .tb_id = fa->tb_id,
89 };
90 return call_fib4_notifier(nb, event_type, &info.info);
91 }
92
call_fib_entry_notifiers(struct net * net,enum fib_event_type event_type,u32 dst,int dst_len,struct fib_alias * fa,struct netlink_ext_ack * extack)93 static int call_fib_entry_notifiers(struct net *net,
94 enum fib_event_type event_type, u32 dst,
95 int dst_len, struct fib_alias *fa,
96 struct netlink_ext_ack *extack)
97 {
98 struct fib_entry_notifier_info info = {
99 .info.extack = extack,
100 .dst = dst,
101 .dst_len = dst_len,
102 .fi = fa->fa_info,
103 .dscp = fa->fa_dscp,
104 .type = fa->fa_type,
105 .tb_id = fa->tb_id,
106 };
107 return call_fib4_notifiers(net, event_type, &info.info);
108 }
109
110 #define MAX_STAT_DEPTH 32
111
112 #define KEYLENGTH (8*sizeof(t_key))
113 #define KEY_MAX ((t_key)~0)
114
115 typedef unsigned int t_key;
116
117 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
118 #define IS_TNODE(n) ((n)->bits)
119 #define IS_LEAF(n) (!(n)->bits)
120
121 struct key_vector {
122 t_key key;
123 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
124 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
125 unsigned char slen;
126 union {
127 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
128 struct hlist_head leaf;
129 /* This array is valid if (pos | bits) > 0 (TNODE) */
130 DECLARE_FLEX_ARRAY(struct key_vector __rcu *, tnode);
131 };
132 };
133
134 struct tnode {
135 struct rcu_head rcu;
136 t_key empty_children; /* KEYLENGTH bits needed */
137 t_key full_children; /* KEYLENGTH bits needed */
138 struct key_vector __rcu *parent;
139 struct key_vector kv[1];
140 #define tn_bits kv[0].bits
141 };
142
143 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
144 #define LEAF_SIZE TNODE_SIZE(1)
145
146 #ifdef CONFIG_IP_FIB_TRIE_STATS
147 struct trie_use_stats {
148 unsigned int gets;
149 unsigned int backtrack;
150 unsigned int semantic_match_passed;
151 unsigned int semantic_match_miss;
152 unsigned int null_node_hit;
153 unsigned int resize_node_skipped;
154 };
155 #endif
156
157 struct trie_stat {
158 unsigned int totdepth;
159 unsigned int maxdepth;
160 unsigned int tnodes;
161 unsigned int leaves;
162 unsigned int nullpointers;
163 unsigned int prefixes;
164 unsigned int nodesizes[MAX_STAT_DEPTH];
165 };
166
167 struct trie {
168 struct key_vector kv[1];
169 #ifdef CONFIG_IP_FIB_TRIE_STATS
170 struct trie_use_stats __percpu *stats;
171 #endif
172 };
173
174 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
175 static unsigned int tnode_free_size;
176
177 /*
178 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
179 * especially useful before resizing the root node with PREEMPT_NONE configs;
180 * the value was obtained experimentally, aiming to avoid visible slowdown.
181 */
182 unsigned int sysctl_fib_sync_mem = 512 * 1024;
183 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
184 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
185
186 static struct kmem_cache *fn_alias_kmem __ro_after_init;
187 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
188
tn_info(struct key_vector * kv)189 static inline struct tnode *tn_info(struct key_vector *kv)
190 {
191 return container_of(kv, struct tnode, kv[0]);
192 }
193
194 /* caller must hold RTNL */
195 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
196 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
197
198 /* caller must hold RCU read lock or RTNL */
199 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
200 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
201
202 /* wrapper for rcu_assign_pointer */
node_set_parent(struct key_vector * n,struct key_vector * tp)203 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
204 {
205 if (n)
206 rcu_assign_pointer(tn_info(n)->parent, tp);
207 }
208
209 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
210
211 /* This provides us with the number of children in this node, in the case of a
212 * leaf this will return 0 meaning none of the children are accessible.
213 */
child_length(const struct key_vector * tn)214 static inline unsigned long child_length(const struct key_vector *tn)
215 {
216 return (1ul << tn->bits) & ~(1ul);
217 }
218
219 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
220
get_index(t_key key,struct key_vector * kv)221 static inline unsigned long get_index(t_key key, struct key_vector *kv)
222 {
223 unsigned long index = key ^ kv->key;
224
225 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
226 return 0;
227
228 return index >> kv->pos;
229 }
230
231 /* To understand this stuff, an understanding of keys and all their bits is
232 * necessary. Every node in the trie has a key associated with it, but not
233 * all of the bits in that key are significant.
234 *
235 * Consider a node 'n' and its parent 'tp'.
236 *
237 * If n is a leaf, every bit in its key is significant. Its presence is
238 * necessitated by path compression, since during a tree traversal (when
239 * searching for a leaf - unless we are doing an insertion) we will completely
240 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
241 * a potentially successful search, that we have indeed been walking the
242 * correct key path.
243 *
244 * Note that we can never "miss" the correct key in the tree if present by
245 * following the wrong path. Path compression ensures that segments of the key
246 * that are the same for all keys with a given prefix are skipped, but the
247 * skipped part *is* identical for each node in the subtrie below the skipped
248 * bit! trie_insert() in this implementation takes care of that.
249 *
250 * if n is an internal node - a 'tnode' here, the various parts of its key
251 * have many different meanings.
252 *
253 * Example:
254 * _________________________________________________________________
255 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
256 * -----------------------------------------------------------------
257 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
258 *
259 * _________________________________________________________________
260 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
261 * -----------------------------------------------------------------
262 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
263 *
264 * tp->pos = 22
265 * tp->bits = 3
266 * n->pos = 13
267 * n->bits = 4
268 *
269 * First, let's just ignore the bits that come before the parent tp, that is
270 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
271 * point we do not use them for anything.
272 *
273 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
274 * index into the parent's child array. That is, they will be used to find
275 * 'n' among tp's children.
276 *
277 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
278 * for the node n.
279 *
280 * All the bits we have seen so far are significant to the node n. The rest
281 * of the bits are really not needed or indeed known in n->key.
282 *
283 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
284 * n's child array, and will of course be different for each child.
285 *
286 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
287 * at this point.
288 */
289
290 static const int halve_threshold = 25;
291 static const int inflate_threshold = 50;
292 static const int halve_threshold_root = 15;
293 static const int inflate_threshold_root = 30;
294
alias_free_mem_rcu(struct fib_alias * fa)295 static inline void alias_free_mem_rcu(struct fib_alias *fa)
296 {
297 kfree_rcu(fa, rcu);
298 }
299
300 #define TNODE_VMALLOC_MAX \
301 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
302
__node_free_rcu(struct rcu_head * head)303 static void __node_free_rcu(struct rcu_head *head)
304 {
305 struct tnode *n = container_of(head, struct tnode, rcu);
306
307 if (!n->tn_bits)
308 kmem_cache_free(trie_leaf_kmem, n);
309 else
310 kvfree(n);
311 }
312
313 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
314
tnode_alloc(int bits)315 static struct tnode *tnode_alloc(int bits)
316 {
317 size_t size;
318
319 /* verify bits is within bounds */
320 if (bits > TNODE_VMALLOC_MAX)
321 return NULL;
322
323 /* determine size and verify it is non-zero and didn't overflow */
324 size = TNODE_SIZE(1ul << bits);
325
326 if (size <= PAGE_SIZE)
327 return kzalloc(size, GFP_KERNEL);
328 else
329 return vzalloc(size);
330 }
331
empty_child_inc(struct key_vector * n)332 static inline void empty_child_inc(struct key_vector *n)
333 {
334 tn_info(n)->empty_children++;
335
336 if (!tn_info(n)->empty_children)
337 tn_info(n)->full_children++;
338 }
339
empty_child_dec(struct key_vector * n)340 static inline void empty_child_dec(struct key_vector *n)
341 {
342 if (!tn_info(n)->empty_children)
343 tn_info(n)->full_children--;
344
345 tn_info(n)->empty_children--;
346 }
347
leaf_new(t_key key,struct fib_alias * fa)348 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
349 {
350 struct key_vector *l;
351 struct tnode *kv;
352
353 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
354 if (!kv)
355 return NULL;
356
357 /* initialize key vector */
358 l = kv->kv;
359 l->key = key;
360 l->pos = 0;
361 l->bits = 0;
362 l->slen = fa->fa_slen;
363
364 /* link leaf to fib alias */
365 INIT_HLIST_HEAD(&l->leaf);
366 hlist_add_head(&fa->fa_list, &l->leaf);
367
368 return l;
369 }
370
tnode_new(t_key key,int pos,int bits)371 static struct key_vector *tnode_new(t_key key, int pos, int bits)
372 {
373 unsigned int shift = pos + bits;
374 struct key_vector *tn;
375 struct tnode *tnode;
376
377 /* verify bits and pos their msb bits clear and values are valid */
378 BUG_ON(!bits || (shift > KEYLENGTH));
379
380 tnode = tnode_alloc(bits);
381 if (!tnode)
382 return NULL;
383
384 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
385 sizeof(struct key_vector *) << bits);
386
387 if (bits == KEYLENGTH)
388 tnode->full_children = 1;
389 else
390 tnode->empty_children = 1ul << bits;
391
392 tn = tnode->kv;
393 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
394 tn->pos = pos;
395 tn->bits = bits;
396 tn->slen = pos;
397
398 return tn;
399 }
400
401 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
402 * and no bits are skipped. See discussion in dyntree paper p. 6
403 */
tnode_full(struct key_vector * tn,struct key_vector * n)404 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
405 {
406 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
407 }
408
409 /* Add a child at position i overwriting the old value.
410 * Update the value of full_children and empty_children.
411 */
put_child(struct key_vector * tn,unsigned long i,struct key_vector * n)412 static void put_child(struct key_vector *tn, unsigned long i,
413 struct key_vector *n)
414 {
415 struct key_vector *chi = get_child(tn, i);
416 int isfull, wasfull;
417
418 BUG_ON(i >= child_length(tn));
419
420 /* update emptyChildren, overflow into fullChildren */
421 if (!n && chi)
422 empty_child_inc(tn);
423 if (n && !chi)
424 empty_child_dec(tn);
425
426 /* update fullChildren */
427 wasfull = tnode_full(tn, chi);
428 isfull = tnode_full(tn, n);
429
430 if (wasfull && !isfull)
431 tn_info(tn)->full_children--;
432 else if (!wasfull && isfull)
433 tn_info(tn)->full_children++;
434
435 if (n && (tn->slen < n->slen))
436 tn->slen = n->slen;
437
438 rcu_assign_pointer(tn->tnode[i], n);
439 }
440
update_children(struct key_vector * tn)441 static void update_children(struct key_vector *tn)
442 {
443 unsigned long i;
444
445 /* update all of the child parent pointers */
446 for (i = child_length(tn); i;) {
447 struct key_vector *inode = get_child(tn, --i);
448
449 if (!inode)
450 continue;
451
452 /* Either update the children of a tnode that
453 * already belongs to us or update the child
454 * to point to ourselves.
455 */
456 if (node_parent(inode) == tn)
457 update_children(inode);
458 else
459 node_set_parent(inode, tn);
460 }
461 }
462
put_child_root(struct key_vector * tp,t_key key,struct key_vector * n)463 static inline void put_child_root(struct key_vector *tp, t_key key,
464 struct key_vector *n)
465 {
466 if (IS_TRIE(tp))
467 rcu_assign_pointer(tp->tnode[0], n);
468 else
469 put_child(tp, get_index(key, tp), n);
470 }
471
tnode_free_init(struct key_vector * tn)472 static inline void tnode_free_init(struct key_vector *tn)
473 {
474 tn_info(tn)->rcu.next = NULL;
475 }
476
tnode_free_append(struct key_vector * tn,struct key_vector * n)477 static inline void tnode_free_append(struct key_vector *tn,
478 struct key_vector *n)
479 {
480 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
481 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
482 }
483
tnode_free(struct key_vector * tn)484 static void tnode_free(struct key_vector *tn)
485 {
486 struct callback_head *head = &tn_info(tn)->rcu;
487
488 while (head) {
489 head = head->next;
490 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
491 node_free(tn);
492
493 tn = container_of(head, struct tnode, rcu)->kv;
494 }
495
496 if (tnode_free_size >= READ_ONCE(sysctl_fib_sync_mem)) {
497 tnode_free_size = 0;
498 synchronize_net();
499 }
500 }
501
replace(struct trie * t,struct key_vector * oldtnode,struct key_vector * tn)502 static struct key_vector *replace(struct trie *t,
503 struct key_vector *oldtnode,
504 struct key_vector *tn)
505 {
506 struct key_vector *tp = node_parent(oldtnode);
507 unsigned long i;
508
509 /* setup the parent pointer out of and back into this node */
510 NODE_INIT_PARENT(tn, tp);
511 put_child_root(tp, tn->key, tn);
512
513 /* update all of the child parent pointers */
514 update_children(tn);
515
516 /* all pointers should be clean so we are done */
517 tnode_free(oldtnode);
518
519 /* resize children now that oldtnode is freed */
520 for (i = child_length(tn); i;) {
521 struct key_vector *inode = get_child(tn, --i);
522
523 /* resize child node */
524 if (tnode_full(tn, inode))
525 tn = resize(t, inode);
526 }
527
528 return tp;
529 }
530
inflate(struct trie * t,struct key_vector * oldtnode)531 static struct key_vector *inflate(struct trie *t,
532 struct key_vector *oldtnode)
533 {
534 struct key_vector *tn;
535 unsigned long i;
536 t_key m;
537
538 pr_debug("In inflate\n");
539
540 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
541 if (!tn)
542 goto notnode;
543
544 /* prepare oldtnode to be freed */
545 tnode_free_init(oldtnode);
546
547 /* Assemble all of the pointers in our cluster, in this case that
548 * represents all of the pointers out of our allocated nodes that
549 * point to existing tnodes and the links between our allocated
550 * nodes.
551 */
552 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
553 struct key_vector *inode = get_child(oldtnode, --i);
554 struct key_vector *node0, *node1;
555 unsigned long j, k;
556
557 /* An empty child */
558 if (!inode)
559 continue;
560
561 /* A leaf or an internal node with skipped bits */
562 if (!tnode_full(oldtnode, inode)) {
563 put_child(tn, get_index(inode->key, tn), inode);
564 continue;
565 }
566
567 /* drop the node in the old tnode free list */
568 tnode_free_append(oldtnode, inode);
569
570 /* An internal node with two children */
571 if (inode->bits == 1) {
572 put_child(tn, 2 * i + 1, get_child(inode, 1));
573 put_child(tn, 2 * i, get_child(inode, 0));
574 continue;
575 }
576
577 /* We will replace this node 'inode' with two new
578 * ones, 'node0' and 'node1', each with half of the
579 * original children. The two new nodes will have
580 * a position one bit further down the key and this
581 * means that the "significant" part of their keys
582 * (see the discussion near the top of this file)
583 * will differ by one bit, which will be "0" in
584 * node0's key and "1" in node1's key. Since we are
585 * moving the key position by one step, the bit that
586 * we are moving away from - the bit at position
587 * (tn->pos) - is the one that will differ between
588 * node0 and node1. So... we synthesize that bit in the
589 * two new keys.
590 */
591 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
592 if (!node1)
593 goto nomem;
594 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
595
596 tnode_free_append(tn, node1);
597 if (!node0)
598 goto nomem;
599 tnode_free_append(tn, node0);
600
601 /* populate child pointers in new nodes */
602 for (k = child_length(inode), j = k / 2; j;) {
603 put_child(node1, --j, get_child(inode, --k));
604 put_child(node0, j, get_child(inode, j));
605 put_child(node1, --j, get_child(inode, --k));
606 put_child(node0, j, get_child(inode, j));
607 }
608
609 /* link new nodes to parent */
610 NODE_INIT_PARENT(node1, tn);
611 NODE_INIT_PARENT(node0, tn);
612
613 /* link parent to nodes */
614 put_child(tn, 2 * i + 1, node1);
615 put_child(tn, 2 * i, node0);
616 }
617
618 /* setup the parent pointers into and out of this node */
619 return replace(t, oldtnode, tn);
620 nomem:
621 /* all pointers should be clean so we are done */
622 tnode_free(tn);
623 notnode:
624 return NULL;
625 }
626
halve(struct trie * t,struct key_vector * oldtnode)627 static struct key_vector *halve(struct trie *t,
628 struct key_vector *oldtnode)
629 {
630 struct key_vector *tn;
631 unsigned long i;
632
633 pr_debug("In halve\n");
634
635 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
636 if (!tn)
637 goto notnode;
638
639 /* prepare oldtnode to be freed */
640 tnode_free_init(oldtnode);
641
642 /* Assemble all of the pointers in our cluster, in this case that
643 * represents all of the pointers out of our allocated nodes that
644 * point to existing tnodes and the links between our allocated
645 * nodes.
646 */
647 for (i = child_length(oldtnode); i;) {
648 struct key_vector *node1 = get_child(oldtnode, --i);
649 struct key_vector *node0 = get_child(oldtnode, --i);
650 struct key_vector *inode;
651
652 /* At least one of the children is empty */
653 if (!node1 || !node0) {
654 put_child(tn, i / 2, node1 ? : node0);
655 continue;
656 }
657
658 /* Two nonempty children */
659 inode = tnode_new(node0->key, oldtnode->pos, 1);
660 if (!inode)
661 goto nomem;
662 tnode_free_append(tn, inode);
663
664 /* initialize pointers out of node */
665 put_child(inode, 1, node1);
666 put_child(inode, 0, node0);
667 NODE_INIT_PARENT(inode, tn);
668
669 /* link parent to node */
670 put_child(tn, i / 2, inode);
671 }
672
673 /* setup the parent pointers into and out of this node */
674 return replace(t, oldtnode, tn);
675 nomem:
676 /* all pointers should be clean so we are done */
677 tnode_free(tn);
678 notnode:
679 return NULL;
680 }
681
collapse(struct trie * t,struct key_vector * oldtnode)682 static struct key_vector *collapse(struct trie *t,
683 struct key_vector *oldtnode)
684 {
685 struct key_vector *n, *tp;
686 unsigned long i;
687
688 /* scan the tnode looking for that one child that might still exist */
689 for (n = NULL, i = child_length(oldtnode); !n && i;)
690 n = get_child(oldtnode, --i);
691
692 /* compress one level */
693 tp = node_parent(oldtnode);
694 put_child_root(tp, oldtnode->key, n);
695 node_set_parent(n, tp);
696
697 /* drop dead node */
698 node_free(oldtnode);
699
700 return tp;
701 }
702
update_suffix(struct key_vector * tn)703 static unsigned char update_suffix(struct key_vector *tn)
704 {
705 unsigned char slen = tn->pos;
706 unsigned long stride, i;
707 unsigned char slen_max;
708
709 /* only vector 0 can have a suffix length greater than or equal to
710 * tn->pos + tn->bits, the second highest node will have a suffix
711 * length at most of tn->pos + tn->bits - 1
712 */
713 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
714
715 /* search though the list of children looking for nodes that might
716 * have a suffix greater than the one we currently have. This is
717 * why we start with a stride of 2 since a stride of 1 would
718 * represent the nodes with suffix length equal to tn->pos
719 */
720 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
721 struct key_vector *n = get_child(tn, i);
722
723 if (!n || (n->slen <= slen))
724 continue;
725
726 /* update stride and slen based on new value */
727 stride <<= (n->slen - slen);
728 slen = n->slen;
729 i &= ~(stride - 1);
730
731 /* stop searching if we have hit the maximum possible value */
732 if (slen >= slen_max)
733 break;
734 }
735
736 tn->slen = slen;
737
738 return slen;
739 }
740
741 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
742 * the Helsinki University of Technology and Matti Tikkanen of Nokia
743 * Telecommunications, page 6:
744 * "A node is doubled if the ratio of non-empty children to all
745 * children in the *doubled* node is at least 'high'."
746 *
747 * 'high' in this instance is the variable 'inflate_threshold'. It
748 * is expressed as a percentage, so we multiply it with
749 * child_length() and instead of multiplying by 2 (since the
750 * child array will be doubled by inflate()) and multiplying
751 * the left-hand side by 100 (to handle the percentage thing) we
752 * multiply the left-hand side by 50.
753 *
754 * The left-hand side may look a bit weird: child_length(tn)
755 * - tn->empty_children is of course the number of non-null children
756 * in the current node. tn->full_children is the number of "full"
757 * children, that is non-null tnodes with a skip value of 0.
758 * All of those will be doubled in the resulting inflated tnode, so
759 * we just count them one extra time here.
760 *
761 * A clearer way to write this would be:
762 *
763 * to_be_doubled = tn->full_children;
764 * not_to_be_doubled = child_length(tn) - tn->empty_children -
765 * tn->full_children;
766 *
767 * new_child_length = child_length(tn) * 2;
768 *
769 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
770 * new_child_length;
771 * if (new_fill_factor >= inflate_threshold)
772 *
773 * ...and so on, tho it would mess up the while () loop.
774 *
775 * anyway,
776 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
777 * inflate_threshold
778 *
779 * avoid a division:
780 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
781 * inflate_threshold * new_child_length
782 *
783 * expand not_to_be_doubled and to_be_doubled, and shorten:
784 * 100 * (child_length(tn) - tn->empty_children +
785 * tn->full_children) >= inflate_threshold * new_child_length
786 *
787 * expand new_child_length:
788 * 100 * (child_length(tn) - tn->empty_children +
789 * tn->full_children) >=
790 * inflate_threshold * child_length(tn) * 2
791 *
792 * shorten again:
793 * 50 * (tn->full_children + child_length(tn) -
794 * tn->empty_children) >= inflate_threshold *
795 * child_length(tn)
796 *
797 */
should_inflate(struct key_vector * tp,struct key_vector * tn)798 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
799 {
800 unsigned long used = child_length(tn);
801 unsigned long threshold = used;
802
803 /* Keep root node larger */
804 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
805 used -= tn_info(tn)->empty_children;
806 used += tn_info(tn)->full_children;
807
808 /* if bits == KEYLENGTH then pos = 0, and will fail below */
809
810 return (used > 1) && tn->pos && ((50 * used) >= threshold);
811 }
812
should_halve(struct key_vector * tp,struct key_vector * tn)813 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
814 {
815 unsigned long used = child_length(tn);
816 unsigned long threshold = used;
817
818 /* Keep root node larger */
819 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
820 used -= tn_info(tn)->empty_children;
821
822 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
823
824 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
825 }
826
should_collapse(struct key_vector * tn)827 static inline bool should_collapse(struct key_vector *tn)
828 {
829 unsigned long used = child_length(tn);
830
831 used -= tn_info(tn)->empty_children;
832
833 /* account for bits == KEYLENGTH case */
834 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
835 used -= KEY_MAX;
836
837 /* One child or none, time to drop us from the trie */
838 return used < 2;
839 }
840
841 #define MAX_WORK 10
resize(struct trie * t,struct key_vector * tn)842 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
843 {
844 #ifdef CONFIG_IP_FIB_TRIE_STATS
845 struct trie_use_stats __percpu *stats = t->stats;
846 #endif
847 struct key_vector *tp = node_parent(tn);
848 unsigned long cindex = get_index(tn->key, tp);
849 int max_work = MAX_WORK;
850
851 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
852 tn, inflate_threshold, halve_threshold);
853
854 /* track the tnode via the pointer from the parent instead of
855 * doing it ourselves. This way we can let RCU fully do its
856 * thing without us interfering
857 */
858 BUG_ON(tn != get_child(tp, cindex));
859
860 /* Double as long as the resulting node has a number of
861 * nonempty nodes that are above the threshold.
862 */
863 while (should_inflate(tp, tn) && max_work) {
864 tp = inflate(t, tn);
865 if (!tp) {
866 #ifdef CONFIG_IP_FIB_TRIE_STATS
867 this_cpu_inc(stats->resize_node_skipped);
868 #endif
869 break;
870 }
871
872 max_work--;
873 tn = get_child(tp, cindex);
874 }
875
876 /* update parent in case inflate failed */
877 tp = node_parent(tn);
878
879 /* Return if at least one inflate is run */
880 if (max_work != MAX_WORK)
881 return tp;
882
883 /* Halve as long as the number of empty children in this
884 * node is above threshold.
885 */
886 while (should_halve(tp, tn) && max_work) {
887 tp = halve(t, tn);
888 if (!tp) {
889 #ifdef CONFIG_IP_FIB_TRIE_STATS
890 this_cpu_inc(stats->resize_node_skipped);
891 #endif
892 break;
893 }
894
895 max_work--;
896 tn = get_child(tp, cindex);
897 }
898
899 /* Only one child remains */
900 if (should_collapse(tn))
901 return collapse(t, tn);
902
903 /* update parent in case halve failed */
904 return node_parent(tn);
905 }
906
node_pull_suffix(struct key_vector * tn,unsigned char slen)907 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
908 {
909 unsigned char node_slen = tn->slen;
910
911 while ((node_slen > tn->pos) && (node_slen > slen)) {
912 slen = update_suffix(tn);
913 if (node_slen == slen)
914 break;
915
916 tn = node_parent(tn);
917 node_slen = tn->slen;
918 }
919 }
920
node_push_suffix(struct key_vector * tn,unsigned char slen)921 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
922 {
923 while (tn->slen < slen) {
924 tn->slen = slen;
925 tn = node_parent(tn);
926 }
927 }
928
929 /* rcu_read_lock needs to be hold by caller from readside */
fib_find_node(struct trie * t,struct key_vector ** tp,u32 key)930 static struct key_vector *fib_find_node(struct trie *t,
931 struct key_vector **tp, u32 key)
932 {
933 struct key_vector *pn, *n = t->kv;
934 unsigned long index = 0;
935
936 do {
937 pn = n;
938 n = get_child_rcu(n, index);
939
940 if (!n)
941 break;
942
943 index = get_cindex(key, n);
944
945 /* This bit of code is a bit tricky but it combines multiple
946 * checks into a single check. The prefix consists of the
947 * prefix plus zeros for the bits in the cindex. The index
948 * is the difference between the key and this value. From
949 * this we can actually derive several pieces of data.
950 * if (index >= (1ul << bits))
951 * we have a mismatch in skip bits and failed
952 * else
953 * we know the value is cindex
954 *
955 * This check is safe even if bits == KEYLENGTH due to the
956 * fact that we can only allocate a node with 32 bits if a
957 * long is greater than 32 bits.
958 */
959 if (index >= (1ul << n->bits)) {
960 n = NULL;
961 break;
962 }
963
964 /* keep searching until we find a perfect match leaf or NULL */
965 } while (IS_TNODE(n));
966
967 *tp = pn;
968
969 return n;
970 }
971
972 /* Return the first fib alias matching DSCP with
973 * priority less than or equal to PRIO.
974 * If 'find_first' is set, return the first matching
975 * fib alias, regardless of DSCP and priority.
976 */
fib_find_alias(struct hlist_head * fah,u8 slen,dscp_t dscp,u32 prio,u32 tb_id,bool find_first)977 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
978 dscp_t dscp, u32 prio, u32 tb_id,
979 bool find_first)
980 {
981 struct fib_alias *fa;
982
983 if (!fah)
984 return NULL;
985
986 hlist_for_each_entry(fa, fah, fa_list) {
987 /* Avoid Sparse warning when using dscp_t in inequalities */
988 u8 __fa_dscp = inet_dscp_to_dsfield(fa->fa_dscp);
989 u8 __dscp = inet_dscp_to_dsfield(dscp);
990
991 if (fa->fa_slen < slen)
992 continue;
993 if (fa->fa_slen != slen)
994 break;
995 if (fa->tb_id > tb_id)
996 continue;
997 if (fa->tb_id != tb_id)
998 break;
999 if (find_first)
1000 return fa;
1001 if (__fa_dscp > __dscp)
1002 continue;
1003 if (fa->fa_info->fib_priority >= prio || __fa_dscp < __dscp)
1004 return fa;
1005 }
1006
1007 return NULL;
1008 }
1009
1010 static struct fib_alias *
fib_find_matching_alias(struct net * net,const struct fib_rt_info * fri)1011 fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1012 {
1013 u8 slen = KEYLENGTH - fri->dst_len;
1014 struct key_vector *l, *tp;
1015 struct fib_table *tb;
1016 struct fib_alias *fa;
1017 struct trie *t;
1018
1019 tb = fib_get_table(net, fri->tb_id);
1020 if (!tb)
1021 return NULL;
1022
1023 t = (struct trie *)tb->tb_data;
1024 l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1025 if (!l)
1026 return NULL;
1027
1028 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1029 if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1030 fa->fa_dscp == fri->dscp && fa->fa_info == fri->fi &&
1031 fa->fa_type == fri->type)
1032 return fa;
1033 }
1034
1035 return NULL;
1036 }
1037
fib_alias_hw_flags_set(struct net * net,const struct fib_rt_info * fri)1038 void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1039 {
1040 u8 fib_notify_on_flag_change;
1041 struct fib_alias *fa_match;
1042 struct sk_buff *skb;
1043 int err;
1044
1045 rcu_read_lock();
1046
1047 fa_match = fib_find_matching_alias(net, fri);
1048 if (!fa_match)
1049 goto out;
1050
1051 /* These are paired with the WRITE_ONCE() happening in this function.
1052 * The reason is that we are only protected by RCU at this point.
1053 */
1054 if (READ_ONCE(fa_match->offload) == fri->offload &&
1055 READ_ONCE(fa_match->trap) == fri->trap &&
1056 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1057 goto out;
1058
1059 WRITE_ONCE(fa_match->offload, fri->offload);
1060 WRITE_ONCE(fa_match->trap, fri->trap);
1061
1062 fib_notify_on_flag_change = READ_ONCE(net->ipv4.sysctl_fib_notify_on_flag_change);
1063
1064 /* 2 means send notifications only if offload_failed was changed. */
1065 if (fib_notify_on_flag_change == 2 &&
1066 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1067 goto out;
1068
1069 WRITE_ONCE(fa_match->offload_failed, fri->offload_failed);
1070
1071 if (!fib_notify_on_flag_change)
1072 goto out;
1073
1074 skb = nlmsg_new(fib_nlmsg_size(fa_match->fa_info), GFP_ATOMIC);
1075 if (!skb) {
1076 err = -ENOBUFS;
1077 goto errout;
1078 }
1079
1080 err = fib_dump_info(skb, 0, 0, RTM_NEWROUTE, fri, 0);
1081 if (err < 0) {
1082 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
1083 WARN_ON(err == -EMSGSIZE);
1084 kfree_skb(skb);
1085 goto errout;
1086 }
1087
1088 rtnl_notify(skb, net, 0, RTNLGRP_IPV4_ROUTE, NULL, GFP_ATOMIC);
1089 goto out;
1090
1091 errout:
1092 rtnl_set_sk_err(net, RTNLGRP_IPV4_ROUTE, err);
1093 out:
1094 rcu_read_unlock();
1095 }
1096 EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1097
trie_rebalance(struct trie * t,struct key_vector * tn)1098 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1099 {
1100 while (!IS_TRIE(tn))
1101 tn = resize(t, tn);
1102 }
1103
fib_insert_node(struct trie * t,struct key_vector * tp,struct fib_alias * new,t_key key)1104 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1105 struct fib_alias *new, t_key key)
1106 {
1107 struct key_vector *n, *l;
1108
1109 l = leaf_new(key, new);
1110 if (!l)
1111 goto noleaf;
1112
1113 /* retrieve child from parent node */
1114 n = get_child(tp, get_index(key, tp));
1115
1116 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1117 *
1118 * Add a new tnode here
1119 * first tnode need some special handling
1120 * leaves us in position for handling as case 3
1121 */
1122 if (n) {
1123 struct key_vector *tn;
1124
1125 tn = tnode_new(key, __fls(key ^ n->key), 1);
1126 if (!tn)
1127 goto notnode;
1128
1129 /* initialize routes out of node */
1130 NODE_INIT_PARENT(tn, tp);
1131 put_child(tn, get_index(key, tn) ^ 1, n);
1132
1133 /* start adding routes into the node */
1134 put_child_root(tp, key, tn);
1135 node_set_parent(n, tn);
1136
1137 /* parent now has a NULL spot where the leaf can go */
1138 tp = tn;
1139 }
1140
1141 /* Case 3: n is NULL, and will just insert a new leaf */
1142 node_push_suffix(tp, new->fa_slen);
1143 NODE_INIT_PARENT(l, tp);
1144 put_child_root(tp, key, l);
1145 trie_rebalance(t, tp);
1146
1147 return 0;
1148 notnode:
1149 node_free(l);
1150 noleaf:
1151 return -ENOMEM;
1152 }
1153
fib_insert_alias(struct trie * t,struct key_vector * tp,struct key_vector * l,struct fib_alias * new,struct fib_alias * fa,t_key key)1154 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1155 struct key_vector *l, struct fib_alias *new,
1156 struct fib_alias *fa, t_key key)
1157 {
1158 if (!l)
1159 return fib_insert_node(t, tp, new, key);
1160
1161 if (fa) {
1162 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1163 } else {
1164 struct fib_alias *last;
1165
1166 hlist_for_each_entry(last, &l->leaf, fa_list) {
1167 if (new->fa_slen < last->fa_slen)
1168 break;
1169 if ((new->fa_slen == last->fa_slen) &&
1170 (new->tb_id > last->tb_id))
1171 break;
1172 fa = last;
1173 }
1174
1175 if (fa)
1176 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1177 else
1178 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1179 }
1180
1181 /* if we added to the tail node then we need to update slen */
1182 if (l->slen < new->fa_slen) {
1183 l->slen = new->fa_slen;
1184 node_push_suffix(tp, new->fa_slen);
1185 }
1186
1187 return 0;
1188 }
1189
1190 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1191 struct key_vector *l, struct fib_alias *old);
1192
1193 /* Caller must hold RTNL. */
fib_table_insert(struct net * net,struct fib_table * tb,struct fib_config * cfg,struct netlink_ext_ack * extack)1194 int fib_table_insert(struct net *net, struct fib_table *tb,
1195 struct fib_config *cfg, struct netlink_ext_ack *extack)
1196 {
1197 struct trie *t = (struct trie *)tb->tb_data;
1198 struct fib_alias *fa, *new_fa;
1199 struct key_vector *l, *tp;
1200 u16 nlflags = NLM_F_EXCL;
1201 struct fib_info *fi;
1202 u8 plen = cfg->fc_dst_len;
1203 u8 slen = KEYLENGTH - plen;
1204 dscp_t dscp;
1205 u32 key;
1206 int err;
1207
1208 key = ntohl(cfg->fc_dst);
1209
1210 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1211
1212 fi = fib_create_info(cfg, extack);
1213 if (IS_ERR(fi)) {
1214 err = PTR_ERR(fi);
1215 goto err;
1216 }
1217
1218 dscp = cfg->fc_dscp;
1219 l = fib_find_node(t, &tp, key);
1220 fa = l ? fib_find_alias(&l->leaf, slen, dscp, fi->fib_priority,
1221 tb->tb_id, false) : NULL;
1222
1223 /* Now fa, if non-NULL, points to the first fib alias
1224 * with the same keys [prefix,dscp,priority], if such key already
1225 * exists or to the node before which we will insert new one.
1226 *
1227 * If fa is NULL, we will need to allocate a new one and
1228 * insert to the tail of the section matching the suffix length
1229 * of the new alias.
1230 */
1231
1232 if (fa && fa->fa_dscp == dscp &&
1233 fa->fa_info->fib_priority == fi->fib_priority) {
1234 struct fib_alias *fa_first, *fa_match;
1235
1236 err = -EEXIST;
1237 if (cfg->fc_nlflags & NLM_F_EXCL)
1238 goto out;
1239
1240 nlflags &= ~NLM_F_EXCL;
1241
1242 /* We have 2 goals:
1243 * 1. Find exact match for type, scope, fib_info to avoid
1244 * duplicate routes
1245 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1246 */
1247 fa_match = NULL;
1248 fa_first = fa;
1249 hlist_for_each_entry_from(fa, fa_list) {
1250 if ((fa->fa_slen != slen) ||
1251 (fa->tb_id != tb->tb_id) ||
1252 (fa->fa_dscp != dscp))
1253 break;
1254 if (fa->fa_info->fib_priority != fi->fib_priority)
1255 break;
1256 if (fa->fa_type == cfg->fc_type &&
1257 fa->fa_info == fi) {
1258 fa_match = fa;
1259 break;
1260 }
1261 }
1262
1263 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1264 struct fib_info *fi_drop;
1265 u8 state;
1266
1267 nlflags |= NLM_F_REPLACE;
1268 fa = fa_first;
1269 if (fa_match) {
1270 if (fa == fa_match)
1271 err = 0;
1272 goto out;
1273 }
1274 err = -ENOBUFS;
1275 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1276 if (!new_fa)
1277 goto out;
1278
1279 fi_drop = fa->fa_info;
1280 new_fa->fa_dscp = fa->fa_dscp;
1281 new_fa->fa_info = fi;
1282 new_fa->fa_type = cfg->fc_type;
1283 state = fa->fa_state;
1284 new_fa->fa_state = state & ~FA_S_ACCESSED;
1285 new_fa->fa_slen = fa->fa_slen;
1286 new_fa->tb_id = tb->tb_id;
1287 new_fa->fa_default = -1;
1288 new_fa->offload = 0;
1289 new_fa->trap = 0;
1290 new_fa->offload_failed = 0;
1291
1292 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1293
1294 if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1295 tb->tb_id, true) == new_fa) {
1296 enum fib_event_type fib_event;
1297
1298 fib_event = FIB_EVENT_ENTRY_REPLACE;
1299 err = call_fib_entry_notifiers(net, fib_event,
1300 key, plen,
1301 new_fa, extack);
1302 if (err) {
1303 hlist_replace_rcu(&new_fa->fa_list,
1304 &fa->fa_list);
1305 goto out_free_new_fa;
1306 }
1307 }
1308
1309 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1310 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1311
1312 alias_free_mem_rcu(fa);
1313
1314 fib_release_info(fi_drop);
1315 if (state & FA_S_ACCESSED)
1316 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1317
1318 goto succeeded;
1319 }
1320 /* Error if we find a perfect match which
1321 * uses the same scope, type, and nexthop
1322 * information.
1323 */
1324 if (fa_match)
1325 goto out;
1326
1327 if (cfg->fc_nlflags & NLM_F_APPEND)
1328 nlflags |= NLM_F_APPEND;
1329 else
1330 fa = fa_first;
1331 }
1332 err = -ENOENT;
1333 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1334 goto out;
1335
1336 nlflags |= NLM_F_CREATE;
1337 err = -ENOBUFS;
1338 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1339 if (!new_fa)
1340 goto out;
1341
1342 new_fa->fa_info = fi;
1343 new_fa->fa_dscp = dscp;
1344 new_fa->fa_type = cfg->fc_type;
1345 new_fa->fa_state = 0;
1346 new_fa->fa_slen = slen;
1347 new_fa->tb_id = tb->tb_id;
1348 new_fa->fa_default = -1;
1349 new_fa->offload = 0;
1350 new_fa->trap = 0;
1351 new_fa->offload_failed = 0;
1352
1353 /* Insert new entry to the list. */
1354 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1355 if (err)
1356 goto out_free_new_fa;
1357
1358 /* The alias was already inserted, so the node must exist. */
1359 l = l ? l : fib_find_node(t, &tp, key);
1360 if (WARN_ON_ONCE(!l)) {
1361 err = -ENOENT;
1362 goto out_free_new_fa;
1363 }
1364
1365 if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1366 new_fa) {
1367 enum fib_event_type fib_event;
1368
1369 fib_event = FIB_EVENT_ENTRY_REPLACE;
1370 err = call_fib_entry_notifiers(net, fib_event, key, plen,
1371 new_fa, extack);
1372 if (err)
1373 goto out_remove_new_fa;
1374 }
1375
1376 if (!plen)
1377 tb->tb_num_default++;
1378
1379 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1380 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1381 &cfg->fc_nlinfo, nlflags);
1382 succeeded:
1383 return 0;
1384
1385 out_remove_new_fa:
1386 fib_remove_alias(t, tp, l, new_fa);
1387 out_free_new_fa:
1388 kmem_cache_free(fn_alias_kmem, new_fa);
1389 out:
1390 fib_release_info(fi);
1391 err:
1392 return err;
1393 }
1394
prefix_mismatch(t_key key,struct key_vector * n)1395 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1396 {
1397 t_key prefix = n->key;
1398
1399 return (key ^ prefix) & (prefix | -prefix);
1400 }
1401
fib_lookup_good_nhc(const struct fib_nh_common * nhc,int fib_flags,const struct flowi4 * flp)1402 bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1403 const struct flowi4 *flp)
1404 {
1405 if (nhc->nhc_flags & RTNH_F_DEAD)
1406 return false;
1407
1408 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1409 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1410 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1411 return false;
1412
1413 if (flp->flowi4_oif && flp->flowi4_oif != nhc->nhc_oif)
1414 return false;
1415
1416 return true;
1417 }
1418
1419 /* should be called with rcu_read_lock */
fib_table_lookup(struct fib_table * tb,const struct flowi4 * flp,struct fib_result * res,int fib_flags)1420 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1421 struct fib_result *res, int fib_flags)
1422 {
1423 struct trie *t = (struct trie *) tb->tb_data;
1424 #ifdef CONFIG_IP_FIB_TRIE_STATS
1425 struct trie_use_stats __percpu *stats = t->stats;
1426 #endif
1427 const t_key key = ntohl(flp->daddr);
1428 struct key_vector *n, *pn;
1429 struct fib_alias *fa;
1430 unsigned long index;
1431 t_key cindex;
1432
1433 pn = t->kv;
1434 cindex = 0;
1435
1436 n = get_child_rcu(pn, cindex);
1437 if (!n) {
1438 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1439 return -EAGAIN;
1440 }
1441
1442 #ifdef CONFIG_IP_FIB_TRIE_STATS
1443 this_cpu_inc(stats->gets);
1444 #endif
1445
1446 /* Step 1: Travel to the longest prefix match in the trie */
1447 for (;;) {
1448 index = get_cindex(key, n);
1449
1450 /* This bit of code is a bit tricky but it combines multiple
1451 * checks into a single check. The prefix consists of the
1452 * prefix plus zeros for the "bits" in the prefix. The index
1453 * is the difference between the key and this value. From
1454 * this we can actually derive several pieces of data.
1455 * if (index >= (1ul << bits))
1456 * we have a mismatch in skip bits and failed
1457 * else
1458 * we know the value is cindex
1459 *
1460 * This check is safe even if bits == KEYLENGTH due to the
1461 * fact that we can only allocate a node with 32 bits if a
1462 * long is greater than 32 bits.
1463 */
1464 if (index >= (1ul << n->bits))
1465 break;
1466
1467 /* we have found a leaf. Prefixes have already been compared */
1468 if (IS_LEAF(n))
1469 goto found;
1470
1471 /* only record pn and cindex if we are going to be chopping
1472 * bits later. Otherwise we are just wasting cycles.
1473 */
1474 if (n->slen > n->pos) {
1475 pn = n;
1476 cindex = index;
1477 }
1478
1479 n = get_child_rcu(n, index);
1480 if (unlikely(!n))
1481 goto backtrace;
1482 }
1483
1484 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1485 for (;;) {
1486 /* record the pointer where our next node pointer is stored */
1487 struct key_vector __rcu **cptr = n->tnode;
1488
1489 /* This test verifies that none of the bits that differ
1490 * between the key and the prefix exist in the region of
1491 * the lsb and higher in the prefix.
1492 */
1493 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1494 goto backtrace;
1495
1496 /* exit out and process leaf */
1497 if (unlikely(IS_LEAF(n)))
1498 break;
1499
1500 /* Don't bother recording parent info. Since we are in
1501 * prefix match mode we will have to come back to wherever
1502 * we started this traversal anyway
1503 */
1504
1505 while ((n = rcu_dereference(*cptr)) == NULL) {
1506 backtrace:
1507 #ifdef CONFIG_IP_FIB_TRIE_STATS
1508 if (!n)
1509 this_cpu_inc(stats->null_node_hit);
1510 #endif
1511 /* If we are at cindex 0 there are no more bits for
1512 * us to strip at this level so we must ascend back
1513 * up one level to see if there are any more bits to
1514 * be stripped there.
1515 */
1516 while (!cindex) {
1517 t_key pkey = pn->key;
1518
1519 /* If we don't have a parent then there is
1520 * nothing for us to do as we do not have any
1521 * further nodes to parse.
1522 */
1523 if (IS_TRIE(pn)) {
1524 trace_fib_table_lookup(tb->tb_id, flp,
1525 NULL, -EAGAIN);
1526 return -EAGAIN;
1527 }
1528 #ifdef CONFIG_IP_FIB_TRIE_STATS
1529 this_cpu_inc(stats->backtrack);
1530 #endif
1531 /* Get Child's index */
1532 pn = node_parent_rcu(pn);
1533 cindex = get_index(pkey, pn);
1534 }
1535
1536 /* strip the least significant bit from the cindex */
1537 cindex &= cindex - 1;
1538
1539 /* grab pointer for next child node */
1540 cptr = &pn->tnode[cindex];
1541 }
1542 }
1543
1544 found:
1545 /* this line carries forward the xor from earlier in the function */
1546 index = key ^ n->key;
1547
1548 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1549 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1550 struct fib_info *fi = fa->fa_info;
1551 struct fib_nh_common *nhc;
1552 int nhsel, err;
1553
1554 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1555 if (index >= (1ul << fa->fa_slen))
1556 continue;
1557 }
1558 if (fa->fa_dscp && !fib_dscp_masked_match(fa->fa_dscp, flp))
1559 continue;
1560 /* Paired with WRITE_ONCE() in fib_release_info() */
1561 if (READ_ONCE(fi->fib_dead))
1562 continue;
1563 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1564 continue;
1565 fib_alias_accessed(fa);
1566 err = fib_props[fa->fa_type].error;
1567 if (unlikely(err < 0)) {
1568 out_reject:
1569 #ifdef CONFIG_IP_FIB_TRIE_STATS
1570 this_cpu_inc(stats->semantic_match_passed);
1571 #endif
1572 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1573 return err;
1574 }
1575 if (fi->fib_flags & RTNH_F_DEAD)
1576 continue;
1577
1578 if (unlikely(fi->nh)) {
1579 if (nexthop_is_blackhole(fi->nh)) {
1580 err = fib_props[RTN_BLACKHOLE].error;
1581 goto out_reject;
1582 }
1583
1584 nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1585 &nhsel);
1586 if (nhc)
1587 goto set_result;
1588 goto miss;
1589 }
1590
1591 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1592 nhc = fib_info_nhc(fi, nhsel);
1593
1594 if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1595 continue;
1596 set_result:
1597 if (!(fib_flags & FIB_LOOKUP_NOREF))
1598 refcount_inc(&fi->fib_clntref);
1599
1600 res->prefix = htonl(n->key);
1601 res->prefixlen = KEYLENGTH - fa->fa_slen;
1602 res->nh_sel = nhsel;
1603 res->nhc = nhc;
1604 res->type = fa->fa_type;
1605 res->scope = fi->fib_scope;
1606 res->dscp = fa->fa_dscp;
1607 res->fi = fi;
1608 res->table = tb;
1609 res->fa_head = &n->leaf;
1610 #ifdef CONFIG_IP_FIB_TRIE_STATS
1611 this_cpu_inc(stats->semantic_match_passed);
1612 #endif
1613 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1614
1615 return err;
1616 }
1617 }
1618 miss:
1619 #ifdef CONFIG_IP_FIB_TRIE_STATS
1620 this_cpu_inc(stats->semantic_match_miss);
1621 #endif
1622 goto backtrace;
1623 }
1624 EXPORT_SYMBOL_GPL(fib_table_lookup);
1625
fib_remove_alias(struct trie * t,struct key_vector * tp,struct key_vector * l,struct fib_alias * old)1626 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1627 struct key_vector *l, struct fib_alias *old)
1628 {
1629 /* record the location of the previous list_info entry */
1630 struct hlist_node **pprev = old->fa_list.pprev;
1631 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1632
1633 /* remove the fib_alias from the list */
1634 hlist_del_rcu(&old->fa_list);
1635
1636 /* if we emptied the list this leaf will be freed and we can sort
1637 * out parent suffix lengths as a part of trie_rebalance
1638 */
1639 if (hlist_empty(&l->leaf)) {
1640 if (tp->slen == l->slen)
1641 node_pull_suffix(tp, tp->pos);
1642 put_child_root(tp, l->key, NULL);
1643 node_free(l);
1644 trie_rebalance(t, tp);
1645 return;
1646 }
1647
1648 /* only access fa if it is pointing at the last valid hlist_node */
1649 if (*pprev)
1650 return;
1651
1652 /* update the trie with the latest suffix length */
1653 l->slen = fa->fa_slen;
1654 node_pull_suffix(tp, fa->fa_slen);
1655 }
1656
fib_notify_alias_delete(struct net * net,u32 key,struct hlist_head * fah,struct fib_alias * fa_to_delete,struct netlink_ext_ack * extack)1657 static void fib_notify_alias_delete(struct net *net, u32 key,
1658 struct hlist_head *fah,
1659 struct fib_alias *fa_to_delete,
1660 struct netlink_ext_ack *extack)
1661 {
1662 struct fib_alias *fa_next, *fa_to_notify;
1663 u32 tb_id = fa_to_delete->tb_id;
1664 u8 slen = fa_to_delete->fa_slen;
1665 enum fib_event_type fib_event;
1666
1667 /* Do not notify if we do not care about the route. */
1668 if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1669 return;
1670
1671 /* Determine if the route should be replaced by the next route in the
1672 * list.
1673 */
1674 fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1675 struct fib_alias, fa_list);
1676 if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1677 fib_event = FIB_EVENT_ENTRY_REPLACE;
1678 fa_to_notify = fa_next;
1679 } else {
1680 fib_event = FIB_EVENT_ENTRY_DEL;
1681 fa_to_notify = fa_to_delete;
1682 }
1683 call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1684 fa_to_notify, extack);
1685 }
1686
1687 /* Caller must hold RTNL. */
fib_table_delete(struct net * net,struct fib_table * tb,struct fib_config * cfg,struct netlink_ext_ack * extack)1688 int fib_table_delete(struct net *net, struct fib_table *tb,
1689 struct fib_config *cfg, struct netlink_ext_ack *extack)
1690 {
1691 struct trie *t = (struct trie *) tb->tb_data;
1692 struct fib_alias *fa, *fa_to_delete;
1693 struct key_vector *l, *tp;
1694 u8 plen = cfg->fc_dst_len;
1695 u8 slen = KEYLENGTH - plen;
1696 dscp_t dscp;
1697 u32 key;
1698
1699 key = ntohl(cfg->fc_dst);
1700
1701 l = fib_find_node(t, &tp, key);
1702 if (!l)
1703 return -ESRCH;
1704
1705 dscp = cfg->fc_dscp;
1706 fa = fib_find_alias(&l->leaf, slen, dscp, 0, tb->tb_id, false);
1707 if (!fa)
1708 return -ESRCH;
1709
1710 pr_debug("Deleting %08x/%d dsfield=0x%02x t=%p\n", key, plen,
1711 inet_dscp_to_dsfield(dscp), t);
1712
1713 fa_to_delete = NULL;
1714 hlist_for_each_entry_from(fa, fa_list) {
1715 struct fib_info *fi = fa->fa_info;
1716
1717 if ((fa->fa_slen != slen) ||
1718 (fa->tb_id != tb->tb_id) ||
1719 (fa->fa_dscp != dscp))
1720 break;
1721
1722 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1723 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1724 fa->fa_info->fib_scope == cfg->fc_scope) &&
1725 (!cfg->fc_prefsrc ||
1726 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1727 (!cfg->fc_protocol ||
1728 fi->fib_protocol == cfg->fc_protocol) &&
1729 fib_nh_match(net, cfg, fi, extack) == 0 &&
1730 fib_metrics_match(cfg, fi)) {
1731 fa_to_delete = fa;
1732 break;
1733 }
1734 }
1735
1736 if (!fa_to_delete)
1737 return -ESRCH;
1738
1739 fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1740 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1741 &cfg->fc_nlinfo, 0);
1742
1743 if (!plen)
1744 tb->tb_num_default--;
1745
1746 fib_remove_alias(t, tp, l, fa_to_delete);
1747
1748 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1749 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1750
1751 fib_release_info(fa_to_delete->fa_info);
1752 alias_free_mem_rcu(fa_to_delete);
1753 return 0;
1754 }
1755
1756 /* Scan for the next leaf starting at the provided key value */
leaf_walk_rcu(struct key_vector ** tn,t_key key)1757 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1758 {
1759 struct key_vector *pn, *n = *tn;
1760 unsigned long cindex;
1761
1762 /* this loop is meant to try and find the key in the trie */
1763 do {
1764 /* record parent and next child index */
1765 pn = n;
1766 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1767
1768 if (cindex >> pn->bits)
1769 break;
1770
1771 /* descend into the next child */
1772 n = get_child_rcu(pn, cindex++);
1773 if (!n)
1774 break;
1775
1776 /* guarantee forward progress on the keys */
1777 if (IS_LEAF(n) && (n->key >= key))
1778 goto found;
1779 } while (IS_TNODE(n));
1780
1781 /* this loop will search for the next leaf with a greater key */
1782 while (!IS_TRIE(pn)) {
1783 /* if we exhausted the parent node we will need to climb */
1784 if (cindex >= (1ul << pn->bits)) {
1785 t_key pkey = pn->key;
1786
1787 pn = node_parent_rcu(pn);
1788 cindex = get_index(pkey, pn) + 1;
1789 continue;
1790 }
1791
1792 /* grab the next available node */
1793 n = get_child_rcu(pn, cindex++);
1794 if (!n)
1795 continue;
1796
1797 /* no need to compare keys since we bumped the index */
1798 if (IS_LEAF(n))
1799 goto found;
1800
1801 /* Rescan start scanning in new node */
1802 pn = n;
1803 cindex = 0;
1804 }
1805
1806 *tn = pn;
1807 return NULL; /* Root of trie */
1808 found:
1809 /* if we are at the limit for keys just return NULL for the tnode */
1810 *tn = pn;
1811 return n;
1812 }
1813
fib_trie_free(struct fib_table * tb)1814 static void fib_trie_free(struct fib_table *tb)
1815 {
1816 struct trie *t = (struct trie *)tb->tb_data;
1817 struct key_vector *pn = t->kv;
1818 unsigned long cindex = 1;
1819 struct hlist_node *tmp;
1820 struct fib_alias *fa;
1821
1822 /* walk trie in reverse order and free everything */
1823 for (;;) {
1824 struct key_vector *n;
1825
1826 if (!(cindex--)) {
1827 t_key pkey = pn->key;
1828
1829 if (IS_TRIE(pn))
1830 break;
1831
1832 n = pn;
1833 pn = node_parent(pn);
1834
1835 /* drop emptied tnode */
1836 put_child_root(pn, n->key, NULL);
1837 node_free(n);
1838
1839 cindex = get_index(pkey, pn);
1840
1841 continue;
1842 }
1843
1844 /* grab the next available node */
1845 n = get_child(pn, cindex);
1846 if (!n)
1847 continue;
1848
1849 if (IS_TNODE(n)) {
1850 /* record pn and cindex for leaf walking */
1851 pn = n;
1852 cindex = 1ul << n->bits;
1853
1854 continue;
1855 }
1856
1857 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1858 hlist_del_rcu(&fa->fa_list);
1859 alias_free_mem_rcu(fa);
1860 }
1861
1862 put_child_root(pn, n->key, NULL);
1863 node_free(n);
1864 }
1865
1866 #ifdef CONFIG_IP_FIB_TRIE_STATS
1867 free_percpu(t->stats);
1868 #endif
1869 kfree(tb);
1870 }
1871
fib_trie_unmerge(struct fib_table * oldtb)1872 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1873 {
1874 struct trie *ot = (struct trie *)oldtb->tb_data;
1875 struct key_vector *l, *tp = ot->kv;
1876 struct fib_table *local_tb;
1877 struct fib_alias *fa;
1878 struct trie *lt;
1879 t_key key = 0;
1880
1881 if (oldtb->tb_data == oldtb->__data)
1882 return oldtb;
1883
1884 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1885 if (!local_tb)
1886 return NULL;
1887
1888 lt = (struct trie *)local_tb->tb_data;
1889
1890 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1891 struct key_vector *local_l = NULL, *local_tp;
1892
1893 hlist_for_each_entry(fa, &l->leaf, fa_list) {
1894 struct fib_alias *new_fa;
1895
1896 if (local_tb->tb_id != fa->tb_id)
1897 continue;
1898
1899 /* clone fa for new local table */
1900 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1901 if (!new_fa)
1902 goto out;
1903
1904 memcpy(new_fa, fa, sizeof(*fa));
1905
1906 /* insert clone into table */
1907 if (!local_l)
1908 local_l = fib_find_node(lt, &local_tp, l->key);
1909
1910 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1911 NULL, l->key)) {
1912 kmem_cache_free(fn_alias_kmem, new_fa);
1913 goto out;
1914 }
1915 }
1916
1917 /* stop loop if key wrapped back to 0 */
1918 key = l->key + 1;
1919 if (key < l->key)
1920 break;
1921 }
1922
1923 return local_tb;
1924 out:
1925 fib_trie_free(local_tb);
1926
1927 return NULL;
1928 }
1929
1930 /* Caller must hold RTNL */
fib_table_flush_external(struct fib_table * tb)1931 void fib_table_flush_external(struct fib_table *tb)
1932 {
1933 struct trie *t = (struct trie *)tb->tb_data;
1934 struct key_vector *pn = t->kv;
1935 unsigned long cindex = 1;
1936 struct hlist_node *tmp;
1937 struct fib_alias *fa;
1938
1939 /* walk trie in reverse order */
1940 for (;;) {
1941 unsigned char slen = 0;
1942 struct key_vector *n;
1943
1944 if (!(cindex--)) {
1945 t_key pkey = pn->key;
1946
1947 /* cannot resize the trie vector */
1948 if (IS_TRIE(pn))
1949 break;
1950
1951 /* update the suffix to address pulled leaves */
1952 if (pn->slen > pn->pos)
1953 update_suffix(pn);
1954
1955 /* resize completed node */
1956 pn = resize(t, pn);
1957 cindex = get_index(pkey, pn);
1958
1959 continue;
1960 }
1961
1962 /* grab the next available node */
1963 n = get_child(pn, cindex);
1964 if (!n)
1965 continue;
1966
1967 if (IS_TNODE(n)) {
1968 /* record pn and cindex for leaf walking */
1969 pn = n;
1970 cindex = 1ul << n->bits;
1971
1972 continue;
1973 }
1974
1975 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1976 /* if alias was cloned to local then we just
1977 * need to remove the local copy from main
1978 */
1979 if (tb->tb_id != fa->tb_id) {
1980 hlist_del_rcu(&fa->fa_list);
1981 alias_free_mem_rcu(fa);
1982 continue;
1983 }
1984
1985 /* record local slen */
1986 slen = fa->fa_slen;
1987 }
1988
1989 /* update leaf slen */
1990 n->slen = slen;
1991
1992 if (hlist_empty(&n->leaf)) {
1993 put_child_root(pn, n->key, NULL);
1994 node_free(n);
1995 }
1996 }
1997 }
1998
1999 /* Caller must hold RTNL. */
fib_table_flush(struct net * net,struct fib_table * tb,bool flush_all)2000 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
2001 {
2002 struct trie *t = (struct trie *)tb->tb_data;
2003 struct nl_info info = { .nl_net = net };
2004 struct key_vector *pn = t->kv;
2005 unsigned long cindex = 1;
2006 struct hlist_node *tmp;
2007 struct fib_alias *fa;
2008 int found = 0;
2009
2010 /* walk trie in reverse order */
2011 for (;;) {
2012 unsigned char slen = 0;
2013 struct key_vector *n;
2014
2015 if (!(cindex--)) {
2016 t_key pkey = pn->key;
2017
2018 /* cannot resize the trie vector */
2019 if (IS_TRIE(pn))
2020 break;
2021
2022 /* update the suffix to address pulled leaves */
2023 if (pn->slen > pn->pos)
2024 update_suffix(pn);
2025
2026 /* resize completed node */
2027 pn = resize(t, pn);
2028 cindex = get_index(pkey, pn);
2029
2030 continue;
2031 }
2032
2033 /* grab the next available node */
2034 n = get_child(pn, cindex);
2035 if (!n)
2036 continue;
2037
2038 if (IS_TNODE(n)) {
2039 /* record pn and cindex for leaf walking */
2040 pn = n;
2041 cindex = 1ul << n->bits;
2042
2043 continue;
2044 }
2045
2046 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2047 struct fib_info *fi = fa->fa_info;
2048
2049 if (!fi || tb->tb_id != fa->tb_id ||
2050 (!(fi->fib_flags & RTNH_F_DEAD) &&
2051 !fib_props[fa->fa_type].error)) {
2052 slen = fa->fa_slen;
2053 continue;
2054 }
2055
2056 /* Do not flush error routes if network namespace is
2057 * not being dismantled
2058 */
2059 if (!flush_all && fib_props[fa->fa_type].error) {
2060 slen = fa->fa_slen;
2061 continue;
2062 }
2063
2064 fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2065 NULL);
2066 if (fi->pfsrc_removed)
2067 rtmsg_fib(RTM_DELROUTE, htonl(n->key), fa,
2068 KEYLENGTH - fa->fa_slen, tb->tb_id, &info, 0);
2069 hlist_del_rcu(&fa->fa_list);
2070 fib_release_info(fa->fa_info);
2071 alias_free_mem_rcu(fa);
2072 found++;
2073 }
2074
2075 /* update leaf slen */
2076 n->slen = slen;
2077
2078 if (hlist_empty(&n->leaf)) {
2079 put_child_root(pn, n->key, NULL);
2080 node_free(n);
2081 }
2082 }
2083
2084 pr_debug("trie_flush found=%d\n", found);
2085 return found;
2086 }
2087
2088 /* derived from fib_trie_free */
__fib_info_notify_update(struct net * net,struct fib_table * tb,struct nl_info * info)2089 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2090 struct nl_info *info)
2091 {
2092 struct trie *t = (struct trie *)tb->tb_data;
2093 struct key_vector *pn = t->kv;
2094 unsigned long cindex = 1;
2095 struct fib_alias *fa;
2096
2097 for (;;) {
2098 struct key_vector *n;
2099
2100 if (!(cindex--)) {
2101 t_key pkey = pn->key;
2102
2103 if (IS_TRIE(pn))
2104 break;
2105
2106 pn = node_parent(pn);
2107 cindex = get_index(pkey, pn);
2108 continue;
2109 }
2110
2111 /* grab the next available node */
2112 n = get_child(pn, cindex);
2113 if (!n)
2114 continue;
2115
2116 if (IS_TNODE(n)) {
2117 /* record pn and cindex for leaf walking */
2118 pn = n;
2119 cindex = 1ul << n->bits;
2120
2121 continue;
2122 }
2123
2124 hlist_for_each_entry(fa, &n->leaf, fa_list) {
2125 struct fib_info *fi = fa->fa_info;
2126
2127 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2128 continue;
2129
2130 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2131 KEYLENGTH - fa->fa_slen, tb->tb_id,
2132 info, NLM_F_REPLACE);
2133 }
2134 }
2135 }
2136
fib_info_notify_update(struct net * net,struct nl_info * info)2137 void fib_info_notify_update(struct net *net, struct nl_info *info)
2138 {
2139 unsigned int h;
2140
2141 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2142 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2143 struct fib_table *tb;
2144
2145 hlist_for_each_entry_rcu(tb, head, tb_hlist,
2146 lockdep_rtnl_is_held())
2147 __fib_info_notify_update(net, tb, info);
2148 }
2149 }
2150
fib_leaf_notify(struct key_vector * l,struct fib_table * tb,struct notifier_block * nb,struct netlink_ext_ack * extack)2151 static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2152 struct notifier_block *nb,
2153 struct netlink_ext_ack *extack)
2154 {
2155 struct fib_alias *fa;
2156 int last_slen = -1;
2157 int err;
2158
2159 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2160 struct fib_info *fi = fa->fa_info;
2161
2162 if (!fi)
2163 continue;
2164
2165 /* local and main table can share the same trie,
2166 * so don't notify twice for the same entry.
2167 */
2168 if (tb->tb_id != fa->tb_id)
2169 continue;
2170
2171 if (fa->fa_slen == last_slen)
2172 continue;
2173
2174 last_slen = fa->fa_slen;
2175 err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2176 l->key, KEYLENGTH - fa->fa_slen,
2177 fa, extack);
2178 if (err)
2179 return err;
2180 }
2181 return 0;
2182 }
2183
fib_table_notify(struct fib_table * tb,struct notifier_block * nb,struct netlink_ext_ack * extack)2184 static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2185 struct netlink_ext_ack *extack)
2186 {
2187 struct trie *t = (struct trie *)tb->tb_data;
2188 struct key_vector *l, *tp = t->kv;
2189 t_key key = 0;
2190 int err;
2191
2192 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2193 err = fib_leaf_notify(l, tb, nb, extack);
2194 if (err)
2195 return err;
2196
2197 key = l->key + 1;
2198 /* stop in case of wrap around */
2199 if (key < l->key)
2200 break;
2201 }
2202 return 0;
2203 }
2204
fib_notify(struct net * net,struct notifier_block * nb,struct netlink_ext_ack * extack)2205 int fib_notify(struct net *net, struct notifier_block *nb,
2206 struct netlink_ext_ack *extack)
2207 {
2208 unsigned int h;
2209 int err;
2210
2211 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2212 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2213 struct fib_table *tb;
2214
2215 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2216 err = fib_table_notify(tb, nb, extack);
2217 if (err)
2218 return err;
2219 }
2220 }
2221 return 0;
2222 }
2223
__trie_free_rcu(struct rcu_head * head)2224 static void __trie_free_rcu(struct rcu_head *head)
2225 {
2226 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2227 #ifdef CONFIG_IP_FIB_TRIE_STATS
2228 struct trie *t = (struct trie *)tb->tb_data;
2229
2230 if (tb->tb_data == tb->__data)
2231 free_percpu(t->stats);
2232 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2233 kfree(tb);
2234 }
2235
fib_free_table(struct fib_table * tb)2236 void fib_free_table(struct fib_table *tb)
2237 {
2238 call_rcu(&tb->rcu, __trie_free_rcu);
2239 }
2240
fn_trie_dump_leaf(struct key_vector * l,struct fib_table * tb,struct sk_buff * skb,struct netlink_callback * cb,struct fib_dump_filter * filter)2241 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2242 struct sk_buff *skb, struct netlink_callback *cb,
2243 struct fib_dump_filter *filter)
2244 {
2245 unsigned int flags = NLM_F_MULTI;
2246 __be32 xkey = htonl(l->key);
2247 int i, s_i, i_fa, s_fa, err;
2248 struct fib_alias *fa;
2249
2250 if (filter->filter_set ||
2251 !filter->dump_exceptions || !filter->dump_routes)
2252 flags |= NLM_F_DUMP_FILTERED;
2253
2254 s_i = cb->args[4];
2255 s_fa = cb->args[5];
2256 i = 0;
2257
2258 /* rcu_read_lock is hold by caller */
2259 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2260 struct fib_info *fi = fa->fa_info;
2261
2262 if (i < s_i)
2263 goto next;
2264
2265 i_fa = 0;
2266
2267 if (tb->tb_id != fa->tb_id)
2268 goto next;
2269
2270 if (filter->filter_set) {
2271 if (filter->rt_type && fa->fa_type != filter->rt_type)
2272 goto next;
2273
2274 if ((filter->protocol &&
2275 fi->fib_protocol != filter->protocol))
2276 goto next;
2277
2278 if (filter->dev &&
2279 !fib_info_nh_uses_dev(fi, filter->dev))
2280 goto next;
2281 }
2282
2283 if (filter->dump_routes) {
2284 if (!s_fa) {
2285 struct fib_rt_info fri;
2286
2287 fri.fi = fi;
2288 fri.tb_id = tb->tb_id;
2289 fri.dst = xkey;
2290 fri.dst_len = KEYLENGTH - fa->fa_slen;
2291 fri.dscp = fa->fa_dscp;
2292 fri.type = fa->fa_type;
2293 fri.offload = READ_ONCE(fa->offload);
2294 fri.trap = READ_ONCE(fa->trap);
2295 fri.offload_failed = READ_ONCE(fa->offload_failed);
2296 err = fib_dump_info(skb,
2297 NETLINK_CB(cb->skb).portid,
2298 cb->nlh->nlmsg_seq,
2299 RTM_NEWROUTE, &fri, flags);
2300 if (err < 0)
2301 goto stop;
2302 }
2303
2304 i_fa++;
2305 }
2306
2307 if (filter->dump_exceptions) {
2308 err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2309 &i_fa, s_fa, flags);
2310 if (err < 0)
2311 goto stop;
2312 }
2313
2314 next:
2315 i++;
2316 }
2317
2318 cb->args[4] = i;
2319 return skb->len;
2320
2321 stop:
2322 cb->args[4] = i;
2323 cb->args[5] = i_fa;
2324 return err;
2325 }
2326
2327 /* rcu_read_lock needs to be hold by caller from readside */
fib_table_dump(struct fib_table * tb,struct sk_buff * skb,struct netlink_callback * cb,struct fib_dump_filter * filter)2328 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2329 struct netlink_callback *cb, struct fib_dump_filter *filter)
2330 {
2331 struct trie *t = (struct trie *)tb->tb_data;
2332 struct key_vector *l, *tp = t->kv;
2333 /* Dump starting at last key.
2334 * Note: 0.0.0.0/0 (ie default) is first key.
2335 */
2336 int count = cb->args[2];
2337 t_key key = cb->args[3];
2338
2339 /* First time here, count and key are both always 0. Count > 0
2340 * and key == 0 means the dump has wrapped around and we are done.
2341 */
2342 if (count && !key)
2343 return 0;
2344
2345 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2346 int err;
2347
2348 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2349 if (err < 0) {
2350 cb->args[3] = key;
2351 cb->args[2] = count;
2352 return err;
2353 }
2354
2355 ++count;
2356 key = l->key + 1;
2357
2358 memset(&cb->args[4], 0,
2359 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2360
2361 /* stop loop if key wrapped back to 0 */
2362 if (key < l->key)
2363 break;
2364 }
2365
2366 cb->args[3] = key;
2367 cb->args[2] = count;
2368
2369 return 0;
2370 }
2371
fib_trie_init(void)2372 void __init fib_trie_init(void)
2373 {
2374 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2375 sizeof(struct fib_alias),
2376 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2377
2378 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2379 LEAF_SIZE,
2380 0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2381 }
2382
fib_trie_table(u32 id,struct fib_table * alias)2383 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2384 {
2385 struct fib_table *tb;
2386 struct trie *t;
2387 size_t sz = sizeof(*tb);
2388
2389 if (!alias)
2390 sz += sizeof(struct trie);
2391
2392 tb = kzalloc(sz, GFP_KERNEL);
2393 if (!tb)
2394 return NULL;
2395
2396 tb->tb_id = id;
2397 tb->tb_num_default = 0;
2398 tb->tb_data = (alias ? alias->__data : tb->__data);
2399
2400 if (alias)
2401 return tb;
2402
2403 t = (struct trie *) tb->tb_data;
2404 t->kv[0].pos = KEYLENGTH;
2405 t->kv[0].slen = KEYLENGTH;
2406 #ifdef CONFIG_IP_FIB_TRIE_STATS
2407 t->stats = alloc_percpu(struct trie_use_stats);
2408 if (!t->stats) {
2409 kfree(tb);
2410 tb = NULL;
2411 }
2412 #endif
2413
2414 return tb;
2415 }
2416
2417 #ifdef CONFIG_PROC_FS
2418 /* Depth first Trie walk iterator */
2419 struct fib_trie_iter {
2420 struct seq_net_private p;
2421 struct fib_table *tb;
2422 struct key_vector *tnode;
2423 unsigned int index;
2424 unsigned int depth;
2425 };
2426
fib_trie_get_next(struct fib_trie_iter * iter)2427 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2428 {
2429 unsigned long cindex = iter->index;
2430 struct key_vector *pn = iter->tnode;
2431 t_key pkey;
2432
2433 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2434 iter->tnode, iter->index, iter->depth);
2435
2436 while (!IS_TRIE(pn)) {
2437 while (cindex < child_length(pn)) {
2438 struct key_vector *n = get_child_rcu(pn, cindex++);
2439
2440 if (!n)
2441 continue;
2442
2443 if (IS_LEAF(n)) {
2444 iter->tnode = pn;
2445 iter->index = cindex;
2446 } else {
2447 /* push down one level */
2448 iter->tnode = n;
2449 iter->index = 0;
2450 ++iter->depth;
2451 }
2452
2453 return n;
2454 }
2455
2456 /* Current node exhausted, pop back up */
2457 pkey = pn->key;
2458 pn = node_parent_rcu(pn);
2459 cindex = get_index(pkey, pn) + 1;
2460 --iter->depth;
2461 }
2462
2463 /* record root node so further searches know we are done */
2464 iter->tnode = pn;
2465 iter->index = 0;
2466
2467 return NULL;
2468 }
2469
fib_trie_get_first(struct fib_trie_iter * iter,struct trie * t)2470 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2471 struct trie *t)
2472 {
2473 struct key_vector *n, *pn;
2474
2475 if (!t)
2476 return NULL;
2477
2478 pn = t->kv;
2479 n = rcu_dereference(pn->tnode[0]);
2480 if (!n)
2481 return NULL;
2482
2483 if (IS_TNODE(n)) {
2484 iter->tnode = n;
2485 iter->index = 0;
2486 iter->depth = 1;
2487 } else {
2488 iter->tnode = pn;
2489 iter->index = 0;
2490 iter->depth = 0;
2491 }
2492
2493 return n;
2494 }
2495
trie_collect_stats(struct trie * t,struct trie_stat * s)2496 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2497 {
2498 struct key_vector *n;
2499 struct fib_trie_iter iter;
2500
2501 memset(s, 0, sizeof(*s));
2502
2503 rcu_read_lock();
2504 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2505 if (IS_LEAF(n)) {
2506 struct fib_alias *fa;
2507
2508 s->leaves++;
2509 s->totdepth += iter.depth;
2510 if (iter.depth > s->maxdepth)
2511 s->maxdepth = iter.depth;
2512
2513 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2514 ++s->prefixes;
2515 } else {
2516 s->tnodes++;
2517 if (n->bits < MAX_STAT_DEPTH)
2518 s->nodesizes[n->bits]++;
2519 s->nullpointers += tn_info(n)->empty_children;
2520 }
2521 }
2522 rcu_read_unlock();
2523 }
2524
2525 /*
2526 * This outputs /proc/net/fib_triestats
2527 */
trie_show_stats(struct seq_file * seq,struct trie_stat * stat)2528 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2529 {
2530 unsigned int i, max, pointers, bytes, avdepth;
2531
2532 if (stat->leaves)
2533 avdepth = stat->totdepth*100 / stat->leaves;
2534 else
2535 avdepth = 0;
2536
2537 seq_printf(seq, "\tAver depth: %u.%02d\n",
2538 avdepth / 100, avdepth % 100);
2539 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2540
2541 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2542 bytes = LEAF_SIZE * stat->leaves;
2543
2544 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2545 bytes += sizeof(struct fib_alias) * stat->prefixes;
2546
2547 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2548 bytes += TNODE_SIZE(0) * stat->tnodes;
2549
2550 max = MAX_STAT_DEPTH;
2551 while (max > 0 && stat->nodesizes[max-1] == 0)
2552 max--;
2553
2554 pointers = 0;
2555 for (i = 1; i < max; i++)
2556 if (stat->nodesizes[i] != 0) {
2557 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2558 pointers += (1<<i) * stat->nodesizes[i];
2559 }
2560 seq_putc(seq, '\n');
2561 seq_printf(seq, "\tPointers: %u\n", pointers);
2562
2563 bytes += sizeof(struct key_vector *) * pointers;
2564 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2565 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2566 }
2567
2568 #ifdef CONFIG_IP_FIB_TRIE_STATS
trie_show_usage(struct seq_file * seq,const struct trie_use_stats __percpu * stats)2569 static void trie_show_usage(struct seq_file *seq,
2570 const struct trie_use_stats __percpu *stats)
2571 {
2572 struct trie_use_stats s = { 0 };
2573 int cpu;
2574
2575 /* loop through all of the CPUs and gather up the stats */
2576 for_each_possible_cpu(cpu) {
2577 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2578
2579 s.gets += pcpu->gets;
2580 s.backtrack += pcpu->backtrack;
2581 s.semantic_match_passed += pcpu->semantic_match_passed;
2582 s.semantic_match_miss += pcpu->semantic_match_miss;
2583 s.null_node_hit += pcpu->null_node_hit;
2584 s.resize_node_skipped += pcpu->resize_node_skipped;
2585 }
2586
2587 seq_printf(seq, "\nCounters:\n---------\n");
2588 seq_printf(seq, "gets = %u\n", s.gets);
2589 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2590 seq_printf(seq, "semantic match passed = %u\n",
2591 s.semantic_match_passed);
2592 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2593 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2594 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2595 }
2596 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2597
fib_table_print(struct seq_file * seq,struct fib_table * tb)2598 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2599 {
2600 if (tb->tb_id == RT_TABLE_LOCAL)
2601 seq_puts(seq, "Local:\n");
2602 else if (tb->tb_id == RT_TABLE_MAIN)
2603 seq_puts(seq, "Main:\n");
2604 else
2605 seq_printf(seq, "Id %d:\n", tb->tb_id);
2606 }
2607
2608
fib_triestat_seq_show(struct seq_file * seq,void * v)2609 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2610 {
2611 struct net *net = seq->private;
2612 unsigned int h;
2613
2614 seq_printf(seq,
2615 "Basic info: size of leaf:"
2616 " %zd bytes, size of tnode: %zd bytes.\n",
2617 LEAF_SIZE, TNODE_SIZE(0));
2618
2619 rcu_read_lock();
2620 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2621 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2622 struct fib_table *tb;
2623
2624 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2625 struct trie *t = (struct trie *) tb->tb_data;
2626 struct trie_stat stat;
2627
2628 if (!t)
2629 continue;
2630
2631 fib_table_print(seq, tb);
2632
2633 trie_collect_stats(t, &stat);
2634 trie_show_stats(seq, &stat);
2635 #ifdef CONFIG_IP_FIB_TRIE_STATS
2636 trie_show_usage(seq, t->stats);
2637 #endif
2638 }
2639 cond_resched_rcu();
2640 }
2641 rcu_read_unlock();
2642
2643 return 0;
2644 }
2645
fib_trie_get_idx(struct seq_file * seq,loff_t pos)2646 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2647 {
2648 struct fib_trie_iter *iter = seq->private;
2649 struct net *net = seq_file_net(seq);
2650 loff_t idx = 0;
2651 unsigned int h;
2652
2653 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2654 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2655 struct fib_table *tb;
2656
2657 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2658 struct key_vector *n;
2659
2660 for (n = fib_trie_get_first(iter,
2661 (struct trie *) tb->tb_data);
2662 n; n = fib_trie_get_next(iter))
2663 if (pos == idx++) {
2664 iter->tb = tb;
2665 return n;
2666 }
2667 }
2668 }
2669
2670 return NULL;
2671 }
2672
fib_trie_seq_start(struct seq_file * seq,loff_t * pos)2673 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2674 __acquires(RCU)
2675 {
2676 rcu_read_lock();
2677 return fib_trie_get_idx(seq, *pos);
2678 }
2679
fib_trie_seq_next(struct seq_file * seq,void * v,loff_t * pos)2680 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2681 {
2682 struct fib_trie_iter *iter = seq->private;
2683 struct net *net = seq_file_net(seq);
2684 struct fib_table *tb = iter->tb;
2685 struct hlist_node *tb_node;
2686 unsigned int h;
2687 struct key_vector *n;
2688
2689 ++*pos;
2690 /* next node in same table */
2691 n = fib_trie_get_next(iter);
2692 if (n)
2693 return n;
2694
2695 /* walk rest of this hash chain */
2696 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2697 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2698 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2699 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2700 if (n)
2701 goto found;
2702 }
2703
2704 /* new hash chain */
2705 while (++h < FIB_TABLE_HASHSZ) {
2706 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2707 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2708 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2709 if (n)
2710 goto found;
2711 }
2712 }
2713 return NULL;
2714
2715 found:
2716 iter->tb = tb;
2717 return n;
2718 }
2719
fib_trie_seq_stop(struct seq_file * seq,void * v)2720 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2721 __releases(RCU)
2722 {
2723 rcu_read_unlock();
2724 }
2725
seq_indent(struct seq_file * seq,int n)2726 static void seq_indent(struct seq_file *seq, int n)
2727 {
2728 while (n-- > 0)
2729 seq_puts(seq, " ");
2730 }
2731
rtn_scope(char * buf,size_t len,enum rt_scope_t s)2732 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2733 {
2734 switch (s) {
2735 case RT_SCOPE_UNIVERSE: return "universe";
2736 case RT_SCOPE_SITE: return "site";
2737 case RT_SCOPE_LINK: return "link";
2738 case RT_SCOPE_HOST: return "host";
2739 case RT_SCOPE_NOWHERE: return "nowhere";
2740 default:
2741 snprintf(buf, len, "scope=%d", s);
2742 return buf;
2743 }
2744 }
2745
2746 static const char *const rtn_type_names[__RTN_MAX] = {
2747 [RTN_UNSPEC] = "UNSPEC",
2748 [RTN_UNICAST] = "UNICAST",
2749 [RTN_LOCAL] = "LOCAL",
2750 [RTN_BROADCAST] = "BROADCAST",
2751 [RTN_ANYCAST] = "ANYCAST",
2752 [RTN_MULTICAST] = "MULTICAST",
2753 [RTN_BLACKHOLE] = "BLACKHOLE",
2754 [RTN_UNREACHABLE] = "UNREACHABLE",
2755 [RTN_PROHIBIT] = "PROHIBIT",
2756 [RTN_THROW] = "THROW",
2757 [RTN_NAT] = "NAT",
2758 [RTN_XRESOLVE] = "XRESOLVE",
2759 };
2760
rtn_type(char * buf,size_t len,unsigned int t)2761 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2762 {
2763 if (t < __RTN_MAX && rtn_type_names[t])
2764 return rtn_type_names[t];
2765 snprintf(buf, len, "type %u", t);
2766 return buf;
2767 }
2768
2769 /* Pretty print the trie */
fib_trie_seq_show(struct seq_file * seq,void * v)2770 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2771 {
2772 const struct fib_trie_iter *iter = seq->private;
2773 struct key_vector *n = v;
2774
2775 if (IS_TRIE(node_parent_rcu(n)))
2776 fib_table_print(seq, iter->tb);
2777
2778 if (IS_TNODE(n)) {
2779 __be32 prf = htonl(n->key);
2780
2781 seq_indent(seq, iter->depth-1);
2782 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2783 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2784 tn_info(n)->full_children,
2785 tn_info(n)->empty_children);
2786 } else {
2787 __be32 val = htonl(n->key);
2788 struct fib_alias *fa;
2789
2790 seq_indent(seq, iter->depth);
2791 seq_printf(seq, " |-- %pI4\n", &val);
2792
2793 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2794 char buf1[32], buf2[32];
2795
2796 seq_indent(seq, iter->depth + 1);
2797 seq_printf(seq, " /%zu %s %s",
2798 KEYLENGTH - fa->fa_slen,
2799 rtn_scope(buf1, sizeof(buf1),
2800 fa->fa_info->fib_scope),
2801 rtn_type(buf2, sizeof(buf2),
2802 fa->fa_type));
2803 if (fa->fa_dscp)
2804 seq_printf(seq, " tos=%d",
2805 inet_dscp_to_dsfield(fa->fa_dscp));
2806 seq_putc(seq, '\n');
2807 }
2808 }
2809
2810 return 0;
2811 }
2812
2813 static const struct seq_operations fib_trie_seq_ops = {
2814 .start = fib_trie_seq_start,
2815 .next = fib_trie_seq_next,
2816 .stop = fib_trie_seq_stop,
2817 .show = fib_trie_seq_show,
2818 };
2819
2820 struct fib_route_iter {
2821 struct seq_net_private p;
2822 struct fib_table *main_tb;
2823 struct key_vector *tnode;
2824 loff_t pos;
2825 t_key key;
2826 };
2827
fib_route_get_idx(struct fib_route_iter * iter,loff_t pos)2828 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2829 loff_t pos)
2830 {
2831 struct key_vector *l, **tp = &iter->tnode;
2832 t_key key;
2833
2834 /* use cached location of previously found key */
2835 if (iter->pos > 0 && pos >= iter->pos) {
2836 key = iter->key;
2837 } else {
2838 iter->pos = 1;
2839 key = 0;
2840 }
2841
2842 pos -= iter->pos;
2843
2844 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2845 key = l->key + 1;
2846 iter->pos++;
2847 l = NULL;
2848
2849 /* handle unlikely case of a key wrap */
2850 if (!key)
2851 break;
2852 }
2853
2854 if (l)
2855 iter->key = l->key; /* remember it */
2856 else
2857 iter->pos = 0; /* forget it */
2858
2859 return l;
2860 }
2861
fib_route_seq_start(struct seq_file * seq,loff_t * pos)2862 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2863 __acquires(RCU)
2864 {
2865 struct fib_route_iter *iter = seq->private;
2866 struct fib_table *tb;
2867 struct trie *t;
2868
2869 rcu_read_lock();
2870
2871 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2872 if (!tb)
2873 return NULL;
2874
2875 iter->main_tb = tb;
2876 t = (struct trie *)tb->tb_data;
2877 iter->tnode = t->kv;
2878
2879 if (*pos != 0)
2880 return fib_route_get_idx(iter, *pos);
2881
2882 iter->pos = 0;
2883 iter->key = KEY_MAX;
2884
2885 return SEQ_START_TOKEN;
2886 }
2887
fib_route_seq_next(struct seq_file * seq,void * v,loff_t * pos)2888 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2889 {
2890 struct fib_route_iter *iter = seq->private;
2891 struct key_vector *l = NULL;
2892 t_key key = iter->key + 1;
2893
2894 ++*pos;
2895
2896 /* only allow key of 0 for start of sequence */
2897 if ((v == SEQ_START_TOKEN) || key)
2898 l = leaf_walk_rcu(&iter->tnode, key);
2899
2900 if (l) {
2901 iter->key = l->key;
2902 iter->pos++;
2903 } else {
2904 iter->pos = 0;
2905 }
2906
2907 return l;
2908 }
2909
fib_route_seq_stop(struct seq_file * seq,void * v)2910 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2911 __releases(RCU)
2912 {
2913 rcu_read_unlock();
2914 }
2915
fib_flag_trans(int type,__be32 mask,struct fib_info * fi)2916 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2917 {
2918 unsigned int flags = 0;
2919
2920 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2921 flags = RTF_REJECT;
2922 if (fi) {
2923 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2924
2925 if (nhc->nhc_gw.ipv4)
2926 flags |= RTF_GATEWAY;
2927 }
2928 if (mask == htonl(0xFFFFFFFF))
2929 flags |= RTF_HOST;
2930 flags |= RTF_UP;
2931 return flags;
2932 }
2933
2934 /*
2935 * This outputs /proc/net/route.
2936 * The format of the file is not supposed to be changed
2937 * and needs to be same as fib_hash output to avoid breaking
2938 * legacy utilities
2939 */
fib_route_seq_show(struct seq_file * seq,void * v)2940 static int fib_route_seq_show(struct seq_file *seq, void *v)
2941 {
2942 struct fib_route_iter *iter = seq->private;
2943 struct fib_table *tb = iter->main_tb;
2944 struct fib_alias *fa;
2945 struct key_vector *l = v;
2946 __be32 prefix;
2947
2948 if (v == SEQ_START_TOKEN) {
2949 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2950 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2951 "\tWindow\tIRTT");
2952 return 0;
2953 }
2954
2955 prefix = htonl(l->key);
2956
2957 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2958 struct fib_info *fi = fa->fa_info;
2959 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2960 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2961
2962 if ((fa->fa_type == RTN_BROADCAST) ||
2963 (fa->fa_type == RTN_MULTICAST))
2964 continue;
2965
2966 if (fa->tb_id != tb->tb_id)
2967 continue;
2968
2969 seq_setwidth(seq, 127);
2970
2971 if (fi) {
2972 struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2973 __be32 gw = 0;
2974
2975 if (nhc->nhc_gw_family == AF_INET)
2976 gw = nhc->nhc_gw.ipv4;
2977
2978 seq_printf(seq,
2979 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2980 "%u\t%08X\t%d\t%u\t%u",
2981 nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2982 prefix, gw, flags, 0, 0,
2983 fi->fib_priority,
2984 mask,
2985 (fi->fib_advmss ?
2986 fi->fib_advmss + 40 : 0),
2987 fi->fib_window,
2988 fi->fib_rtt >> 3);
2989 } else {
2990 seq_printf(seq,
2991 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2992 "%u\t%08X\t%d\t%u\t%u",
2993 prefix, 0, flags, 0, 0, 0,
2994 mask, 0, 0, 0);
2995 }
2996 seq_pad(seq, '\n');
2997 }
2998
2999 return 0;
3000 }
3001
3002 static const struct seq_operations fib_route_seq_ops = {
3003 .start = fib_route_seq_start,
3004 .next = fib_route_seq_next,
3005 .stop = fib_route_seq_stop,
3006 .show = fib_route_seq_show,
3007 };
3008
fib_proc_init(struct net * net)3009 int __net_init fib_proc_init(struct net *net)
3010 {
3011 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
3012 sizeof(struct fib_trie_iter)))
3013 goto out1;
3014
3015 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
3016 fib_triestat_seq_show, NULL))
3017 goto out2;
3018
3019 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
3020 sizeof(struct fib_route_iter)))
3021 goto out3;
3022
3023 return 0;
3024
3025 out3:
3026 remove_proc_entry("fib_triestat", net->proc_net);
3027 out2:
3028 remove_proc_entry("fib_trie", net->proc_net);
3029 out1:
3030 return -ENOMEM;
3031 }
3032
fib_proc_exit(struct net * net)3033 void __net_exit fib_proc_exit(struct net *net)
3034 {
3035 remove_proc_entry("fib_trie", net->proc_net);
3036 remove_proc_entry("fib_triestat", net->proc_net);
3037 remove_proc_entry("route", net->proc_net);
3038 }
3039
3040 #endif /* CONFIG_PROC_FS */
3041