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>
51 #define DEBUG_MOD_NAME nhop
52 #define DEBUG_MAX_LEVEL LOG_DEBUG
53 #include <net/route/route_debug.h>
54 _DECLARE_DEBUG(LOG_INFO);
57 * This file contains data structures management logic for the nexthop ("nhop")
60 * Nexthops in the original sense are the objects containing all the necessary
61 * information to forward the packet to the selected destination.
62 * In particular, nexthop is defined by a combination of
63 * ifp, ifa, aifp, mtu, gw addr(if set), nh_type, nh_family, mask of rt_flags and
66 * All nexthops are stored in the resizable hash table.
67 * Additionally, each nexthop gets assigned its unique index (nexthop index)
68 * so userland programs can interact with the nexthops easier. Index allocation
69 * is backed by the bitmask array.
72 MALLOC_DEFINE(M_NHOP, "nhops", "nexthops data");
74 /* Hash management functions */
77 nhops_init_rib(struct rib_head *rh)
79 struct nh_control *ctl;
81 uint32_t num_buckets, num_items;
84 ctl = malloc(sizeof(struct nh_control), M_NHOP, M_WAITOK | M_ZERO);
87 * Allocate nexthop hash. Start with 16 items by default (128 bytes).
88 * This will be enough for most of the cases.
91 alloc_size = CHT_SLIST_GET_RESIZE_SIZE(num_buckets);
92 ptr = malloc(alloc_size, M_NHOP, M_WAITOK | M_ZERO);
93 CHT_SLIST_INIT(&ctl->nh_head, ptr, num_buckets);
96 * Allocate nexthop index bitmask.
98 num_items = 128 * 8; /* 128 bytes */
99 ptr = malloc(bitmask_get_size(num_items), M_NHOP, M_WAITOK | M_ZERO);
100 bitmask_init(&ctl->nh_idx_head, ptr, num_items);
102 NHOPS_LOCK_INIT(ctl);
104 rh->nh_control = ctl;
107 FIB_CTL_LOG(LOG_DEBUG2, ctl, "nhops init: ctl %p rh %p", ctl, rh);
113 destroy_ctl(struct nh_control *ctl)
116 NHOPS_LOCK_DESTROY(ctl);
117 free(ctl->nh_head.ptr, M_NHOP);
118 free(ctl->nh_idx_head.idx, M_NHOP);
126 * Epoch callback indicating ctl is safe to destroy
129 destroy_ctl_epoch(epoch_context_t ctx)
131 struct nh_control *ctl;
133 ctl = __containerof(ctx, struct nh_control, ctl_epoch_ctx);
139 nhops_destroy_rib(struct rib_head *rh)
141 struct nh_control *ctl;
142 struct nhop_priv *nh_priv;
144 ctl = rh->nh_control;
147 * All routes should have been deleted in rt_table_destroy().
148 * However, TCP stack or other consumers may store referenced
149 * nexthop pointers. When these references go to zero,
150 * nhop_free() will try to unlink these records from the
151 * datastructures, most likely leading to panic.
153 * Avoid that by explicitly marking all of the remaining
154 * nexthops as unlinked by removing a reference from a special
155 * counter. Please see nhop_free() comments for more
160 CHT_SLIST_FOREACH(&ctl->nh_head, nhops, nh_priv) {
161 FIB_RH_LOG(LOG_DEBUG3, rh, "marking nhop %u unlinked", nh_priv->nh_idx);
162 refcount_release(&nh_priv->nh_linked);
163 } CHT_SLIST_FOREACH_END;
165 nhgrp_ctl_unlink_all(ctl);
170 * Postpone destruction till the end of current epoch
171 * so nhop_free() can safely use nh_control pointer.
173 epoch_call(net_epoch_preempt, destroy_ctl_epoch,
174 &ctl->ctl_epoch_ctx);
178 * Nexhop hash calculation:
180 * Nexthops distribution:
181 * 2 "mandatory" nexthops per interface ("interface route", "loopback").
182 * For direct peering: 1 nexthop for the peering router per ifp/af.
183 * For Ix-like peering: tens to hundreds nexthops of neghbors per ifp/af.
184 * IGP control plane & broadcast segment: tens of nexthops per ifp/af.
186 * Each fib/af combination has its own hash table.
187 * With that in mind, hash nexthops by the combination of the interface
190 * To optimize hash calculation, ignore lower bits of ifnet pointer,
191 * as they give very little entropy.
192 * Similarly, use lower 4 bytes of IPv6 address to distinguish between the
203 djb_hash(const unsigned char *h, const int len)
205 unsigned int result = 0;
208 for (i = 0; i < len; i++)
209 result = 33 * result ^ h[i];
215 hash_priv(const struct nhop_priv *priv)
217 struct nhop_object *nh = priv->nh;
218 struct _hash_data key = {
219 .ifentropy = (uint16_t)((((uintptr_t)nh->nh_ifp) >> 6) & 0xFFFF),
220 .family = nh->gw_sa.sa_family,
221 .nh_type = priv->nh_type & 0xFF,
222 .gw_addr = (nh->gw_sa.sa_family == AF_INET6) ?
223 nh->gw6_sa.sin6_addr.s6_addr32[3] :
224 nh->gw4_sa.sin_addr.s_addr
227 return (uint32_t)(djb_hash((const unsigned char *)&key, sizeof(key)));
231 * Checks if hash needs resizing and performs this resize if necessary
235 consider_resize(struct nh_control *ctl, uint32_t new_nh_buckets, uint32_t new_idx_items)
237 void *nh_ptr, *nh_idx_ptr;
242 if (new_nh_buckets != 0) {
243 alloc_size = CHT_SLIST_GET_RESIZE_SIZE(new_nh_buckets);
244 nh_ptr = malloc(alloc_size, M_NHOP, M_NOWAIT | M_ZERO);
248 if (new_idx_items != 0) {
249 alloc_size = bitmask_get_size(new_idx_items);
250 nh_idx_ptr = malloc(alloc_size, M_NHOP, M_NOWAIT | M_ZERO);
253 if (nh_ptr == NULL && nh_idx_ptr == NULL) {
254 /* Either resize is not required or allocations have failed. */
258 FIB_CTL_LOG(LOG_DEBUG, ctl,
259 "going to resize: nh:[ptr:%p sz:%u] idx:[ptr:%p sz:%u]",
260 nh_ptr, new_nh_buckets, nh_idx_ptr, new_idx_items);
265 if (nh_ptr != NULL) {
266 CHT_SLIST_RESIZE(&ctl->nh_head, nhops, nh_ptr, new_nh_buckets);
268 if (nh_idx_ptr != NULL) {
269 if (bitmask_copy(&ctl->nh_idx_head, nh_idx_ptr, new_idx_items) == 0)
270 bitmask_swap(&ctl->nh_idx_head, nh_idx_ptr, new_idx_items, &old_idx_ptr);
275 free(nh_ptr, M_NHOP);
276 if (old_idx_ptr != NULL)
277 free(old_idx_ptr, M_NHOP);
281 * Links nextop @nh_priv to the nexhop hash table and allocates
283 * Returns allocated index or 0 on failure.
286 link_nhop(struct nh_control *ctl, struct nhop_priv *nh_priv)
289 uint32_t num_buckets_new, num_items_new;
291 KASSERT((nh_priv->nh_idx == 0), ("nhop index is already allocated"));
295 * Check if we need to resize hash and index.
296 * The following 2 functions returns either new size or 0
297 * if resize is not required.
299 num_buckets_new = CHT_SLIST_GET_RESIZE_BUCKETS(&ctl->nh_head);
300 num_items_new = bitmask_get_resize_items(&ctl->nh_idx_head);
302 if (bitmask_alloc_idx(&ctl->nh_idx_head, &idx) != 0) {
304 FIB_CTL_LOG(LOG_INFO, ctl, "Unable to allocate nhop index");
305 RTSTAT_INC(rts_nh_idx_alloc_failure);
306 consider_resize(ctl, num_buckets_new, num_items_new);
310 nh_priv->nh_idx = idx;
311 nh_priv->nh_control = ctl;
312 nh_priv->nh_finalized = 1;
314 CHT_SLIST_INSERT_HEAD(&ctl->nh_head, nhops, nh_priv);
318 FIB_RH_LOG(LOG_DEBUG2, ctl->ctl_rh,
319 "Linked nhop priv %p to %d, hash %u, ctl %p",
320 nh_priv, idx, hash_priv(nh_priv), ctl);
321 consider_resize(ctl, num_buckets_new, num_items_new);
327 * Unlinks nexthop specified by @nh_priv data from the hash.
329 * Returns found nexthop or NULL.
332 unlink_nhop(struct nh_control *ctl, struct nhop_priv *nh_priv_del)
334 struct nhop_priv *priv_ret;
336 uint32_t num_buckets_new, num_items_new;
341 CHT_SLIST_REMOVE(&ctl->nh_head, nhops, nh_priv_del, priv_ret);
343 if (priv_ret != NULL) {
344 idx = priv_ret->nh_idx;
345 priv_ret->nh_idx = 0;
347 KASSERT((idx != 0), ("bogus nhop index 0"));
348 if ((bitmask_free_idx(&ctl->nh_idx_head, idx)) != 0) {
349 FIB_CTL_LOG(LOG_DEBUG, ctl,
350 "Unable to remove index %d from fib %u af %d",
351 idx, ctl->ctl_rh->rib_fibnum, ctl->ctl_rh->rib_family);
355 /* Check if hash or index needs to be resized */
356 num_buckets_new = CHT_SLIST_GET_RESIZE_BUCKETS(&ctl->nh_head);
357 num_items_new = bitmask_get_resize_items(&ctl->nh_idx_head);
360 FIB_CTL_LOG(LOG_INFO, ctl, "Unable to unlink nhop priv %p from hash, hash %u ctl %p",
361 nh_priv_del, hash_priv(nh_priv_del), ctl);
363 if (priv_ret == NULL) {
364 FIB_CTL_LOG(LOG_INFO, ctl,
365 "Unable to unlink nhop priv %p from hash, hash %u ctl %p",
366 nh_priv_del, hash_priv(nh_priv_del), ctl);
368 FIB_CTL_LOG(LOG_DEBUG2, ctl, "Unlinked nhop %p priv idx %d",
372 consider_resize(ctl, num_buckets_new, num_items_new);
378 * Searches for the nexthop by data specifcied in @nh_priv.
379 * Returns referenced nexthop or NULL.
382 find_nhop(struct nh_control *ctl, const struct nhop_priv *nh_priv)
384 struct nhop_priv *nh_priv_ret;
387 CHT_SLIST_FIND_BYOBJ(&ctl->nh_head, nhops, nh_priv, nh_priv_ret);
388 if (nh_priv_ret != NULL) {
389 if (refcount_acquire_if_not_zero(&nh_priv_ret->nh_refcnt) == 0){
390 /* refcount was 0 -> nhop is being deleted */
396 return (nh_priv_ret);