2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2020 Alexander V. Chernikov
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include "opt_route.h"
33 #include <sys/param.h>
34 #include <sys/systm.h>
36 #include <sys/rwlock.h>
37 #include <sys/malloc.h>
39 #include <sys/socket.h>
40 #include <sys/kernel.h>
43 #include <net/if_var.h>
44 #include <net/route.h>
45 #include <net/route/route_var.h>
46 #include <net/route/nhop_utils.h>
47 #include <net/route/nhop.h>
48 #include <net/route/nhop_var.h>
49 #include <net/route/shared.h>
53 * This file contains data structures management logic for the nexthop ("nhop")
56 * Nexthops in the original sense are the objects containing all the necessary
57 * information to forward the packet to the selected destination.
58 * In particular, nexthop is defined by a combination of
59 * ifp, ifa, aifp, mtu, gw addr(if set), nh_type, nh_family, mask of rt_flags and
62 * All nexthops are stored in the resizable hash table.
63 * Additionally, each nexthop gets assigned its unique index (nexthop index)
64 * so userland programs can interact with the nexthops easier. Index allocation
65 * is backed by the bitmask array.
68 static MALLOC_DEFINE(M_NHOP, "nhops", "nexthops data");
71 /* Hash management functions */
74 nhops_init_rib(struct rib_head *rh)
76 struct nh_control *ctl;
78 uint32_t num_buckets, num_items;
81 ctl = malloc(sizeof(struct nh_control), M_NHOP, M_WAITOK | M_ZERO);
84 * Allocate nexthop hash. Start with 16 items by default (128 bytes).
85 * This will be enough for most of the cases.
88 alloc_size = CHT_SLIST_GET_RESIZE_SIZE(num_buckets);
89 ptr = malloc(alloc_size, M_NHOP, M_WAITOK | M_ZERO);
90 CHT_SLIST_INIT(&ctl->nh_head, ptr, num_buckets);
93 * Allocate nexthop index bitmask.
95 num_items = 128 * 8; /* 128 bytes */
96 ptr = malloc(bitmask_get_size(num_items), M_NHOP, M_WAITOK | M_ZERO);
97 bitmask_init(&ctl->nh_idx_head, ptr, num_items);
101 rh->nh_control = ctl;
104 DPRINTF("NHOPS init for fib %u af %u: ctl %p rh %p", rh->rib_fibnum,
105 rh->rib_family, ctl, rh);
111 destroy_ctl(struct nh_control *ctl)
114 NHOPS_LOCK_DESTROY(ctl);
115 free(ctl->nh_head.ptr, M_NHOP);
116 free(ctl->nh_idx_head.idx, M_NHOP);
121 * Epoch callback indicating ctl is safe to destroy
124 destroy_ctl_epoch(epoch_context_t ctx)
126 struct nh_control *ctl;
128 ctl = __containerof(ctx, struct nh_control, ctl_epoch_ctx);
134 nhops_destroy_rib(struct rib_head *rh)
136 struct nh_control *ctl;
137 struct nhop_priv *nh_priv;
139 ctl = rh->nh_control;
142 * All routes should have been deleted in rt_table_destroy().
143 * However, TCP stack or other consumers may store referenced
144 * nexthop pointers. When these references go to zero,
145 * nhop_free() will try to unlink these records from the
146 * datastructures, most likely leading to panic.
148 * Avoid that by explicitly marking all of the remaining
149 * nexthops as unlinked by removing a reference from a special
150 * counter. Please see nhop_free() comments for more
155 CHT_SLIST_FOREACH(&ctl->nh_head, nhops, nh_priv) {
156 DPRINTF("Marking nhop %u unlinked", nh_priv->nh_idx);
157 refcount_release(&nh_priv->nh_linked);
158 } CHT_SLIST_FOREACH_END;
162 * Postpone destruction till the end of current epoch
163 * so nhop_free() can safely use nh_control pointer.
165 epoch_call(net_epoch_preempt, destroy_ctl_epoch,
166 &ctl->ctl_epoch_ctx);
170 * Nexhop hash calculation:
172 * Nexthops distribution:
173 * 2 "mandatory" nexthops per interface ("interface route", "loopback").
174 * For direct peering: 1 nexthop for the peering router per ifp/af.
175 * For Ix-like peering: tens to hundreds nexthops of neghbors per ifp/af.
176 * IGP control plane & broadcast segment: tens of nexthops per ifp/af.
178 * Each fib/af combination has its own hash table.
179 * With that in mind, hash nexthops by the combination of the interface
182 * To optimize hash calculation, ignore higher bytes of ifindex, as they
183 * give very little entropy.
184 * Similarly, use lower 4 bytes of IPv6 address to distinguish between the
195 djb_hash(const unsigned char *h, const int len)
197 unsigned int result = 0;
200 for (i = 0; i < len; i++)
201 result = 33 * result ^ h[i];
207 hash_priv(const struct nhop_priv *priv)
209 struct nhop_object *nh;
211 struct _hash_data key;
214 ifindex = nh->nh_ifp->if_index & 0xFFFF;
215 memset(&key, 0, sizeof(key));
217 key.ifindex = ifindex;
218 key.family = nh->gw_sa.sa_family;
219 key.nh_type = priv->nh_type & 0xFF;
220 if (nh->gw_sa.sa_family == AF_INET6)
221 memcpy(&key.gw_addr, &nh->gw6_sa.sin6_addr.s6_addr32[3], 4);
222 else if (nh->gw_sa.sa_family == AF_INET)
223 memcpy(&key.gw_addr, &nh->gw4_sa.sin_addr, 4);
225 return (uint32_t)(djb_hash((const unsigned char *)&key, sizeof(key)));
229 * Checks if hash needs resizing and performs this resize if necessary
233 consider_resize(struct nh_control *ctl, uint32_t new_nh_buckets, uint32_t new_idx_items)
235 void *nh_ptr, *nh_idx_ptr;
240 if (new_nh_buckets != 0) {
241 alloc_size = CHT_SLIST_GET_RESIZE_SIZE(new_nh_buckets);
242 nh_ptr = malloc(alloc_size, M_NHOP, M_NOWAIT | M_ZERO);
246 if (new_idx_items != 0) {
247 alloc_size = bitmask_get_size(new_idx_items);
248 nh_idx_ptr = malloc(alloc_size, M_NHOP, M_NOWAIT | M_ZERO);
251 if (nh_ptr == NULL && nh_idx_ptr == NULL) {
252 /* Either resize is not required or allocations have failed. */
256 DPRINTF("going to resize: nh:[ptr:%p sz:%u] idx:[ptr:%p sz:%u]", nh_ptr,
257 new_nh_buckets, nh_idx_ptr, new_idx_items);
262 if (nh_ptr != NULL) {
263 CHT_SLIST_RESIZE(&ctl->nh_head, nhops, nh_ptr, new_nh_buckets);
265 if (nh_idx_ptr != NULL) {
266 if (bitmask_copy(&ctl->nh_idx_head, nh_idx_ptr, new_idx_items) == 0)
267 bitmask_swap(&ctl->nh_idx_head, nh_idx_ptr, new_idx_items, &old_idx_ptr);
272 free(nh_ptr, M_NHOP);
273 if (old_idx_ptr != NULL)
274 free(old_idx_ptr, M_NHOP);
278 * Links nextop @nh_priv to the nexhop hash table and allocates
280 * Returns allocated index or 0 on failure.
283 link_nhop(struct nh_control *ctl, struct nhop_priv *nh_priv)
286 uint32_t num_buckets_new, num_items_new;
288 KASSERT((nh_priv->nh_idx == 0), ("nhop index is already allocated"));
292 * Check if we need to resize hash and index.
293 * The following 2 functions returns either new size or 0
294 * if resize is not required.
296 num_buckets_new = CHT_SLIST_GET_RESIZE_BUCKETS(&ctl->nh_head);
297 num_items_new = bitmask_get_resize_items(&ctl->nh_idx_head);
299 if (bitmask_alloc_idx(&ctl->nh_idx_head, &idx) != 0) {
301 DPRINTF("Unable to allocate nhop index");
302 RTSTAT_INC(rts_nh_idx_alloc_failure);
303 consider_resize(ctl, num_buckets_new, num_items_new);
307 nh_priv->nh_idx = idx;
308 nh_priv->nh_control = ctl;
310 CHT_SLIST_INSERT_HEAD(&ctl->nh_head, nhops, nh_priv);
314 DPRINTF("Linked nhop priv %p to %d, hash %u, ctl %p", nh_priv, idx,
315 hash_priv(nh_priv), ctl);
316 consider_resize(ctl, num_buckets_new, num_items_new);
322 * Unlinks nexthop specified by @nh_priv data from the hash.
324 * Returns found nexthop or NULL.
327 unlink_nhop(struct nh_control *ctl, struct nhop_priv *nh_priv_del)
329 struct nhop_priv *priv_ret;
331 uint32_t num_buckets_new, num_items_new;
336 CHT_SLIST_REMOVE_BYOBJ(&ctl->nh_head, nhops, nh_priv_del, priv_ret);
338 if (priv_ret != NULL) {
339 idx = priv_ret->nh_idx;
340 priv_ret->nh_idx = 0;
342 KASSERT((idx != 0), ("bogus nhop index 0"));
343 if ((bitmask_free_idx(&ctl->nh_idx_head, idx)) != 0) {
344 DPRINTF("Unable to remove index %d from fib %u af %d",
345 idx, ctl->ctl_rh->rib_fibnum,
346 ctl->ctl_rh->rib_family);
350 /* Check if hash or index needs to be resized */
351 num_buckets_new = CHT_SLIST_GET_RESIZE_BUCKETS(&ctl->nh_head);
352 num_items_new = bitmask_get_resize_items(&ctl->nh_idx_head);
356 if (priv_ret == NULL)
357 DPRINTF("Unable to unlink nhop priv %p from hash, hash %u ctl %p",
358 nh_priv_del, hash_priv(nh_priv_del), ctl);
360 DPRINTF("Unlinked nhop %p priv idx %d", priv_ret, idx);
362 consider_resize(ctl, num_buckets_new, num_items_new);
368 * Searches for the nexthop by data specifcied in @nh_priv.
369 * Returns referenced nexthop or NULL.
372 find_nhop(struct nh_control *ctl, const struct nhop_priv *nh_priv)
374 struct nhop_priv *nh_priv_ret;
377 CHT_SLIST_FIND_BYOBJ(&ctl->nh_head, nhops, nh_priv, nh_priv_ret);
378 if (nh_priv_ret != NULL) {
379 if (refcount_acquire_if_not_zero(&nh_priv_ret->nh_refcnt) == 0){
380 /* refcount was 0 -> nhop is being deleted */
386 return (nh_priv_ret);