1/*-
2 * Copyright 1994, 1995 Massachusetts Institute of Technology
3 *
4 * Permission to use, copy, modify, and distribute this software and
5 * its documentation for any purpose and without fee is hereby
6 * granted, provided that both the above copyright notice and this
7 * permission notice appear in all copies, that both the above
8 * copyright notice and this permission notice appear in all
9 * supporting documentation, and that the name of M.I.T. not be used
10 * in advertising or publicity pertaining to distribution of the
11 * software without specific, written prior permission. M.I.T. makes
12 * no representations about the suitability of this software for any
13 * purpose. It is provided "as is" without express or implied
14 * warranty.
15 *
16 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS
17 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
18 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
19 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
20 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
23 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
24 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
25 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
26 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30/*
31 * This code does two things necessary for the enhanced TCP metrics to
32 * function in a useful manner:
33 * 1) It marks all non-host routes as `cloning', thus ensuring that
34 * every actual reference to such a route actually gets turned
35 * into a reference to a host route to the specific destination
36 * requested.
37 * 2) When such routes lose all their references, it arranges for them
38 * to be deleted in some random collection of circumstances, so that
39 * a large quantity of stale routing data is not kept in kernel memory
40 * indefinitely. See in_rtqtimo() below for the exact mechanism.
41 */
42
43#include <sys/cdefs.h>
| 1/*-
2 * Copyright 1994, 1995 Massachusetts Institute of Technology
3 *
4 * Permission to use, copy, modify, and distribute this software and
5 * its documentation for any purpose and without fee is hereby
6 * granted, provided that both the above copyright notice and this
7 * permission notice appear in all copies, that both the above
8 * copyright notice and this permission notice appear in all
9 * supporting documentation, and that the name of M.I.T. not be used
10 * in advertising or publicity pertaining to distribution of the
11 * software without specific, written prior permission. M.I.T. makes
12 * no representations about the suitability of this software for any
13 * purpose. It is provided "as is" without express or implied
14 * warranty.
15 *
16 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS
17 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
18 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
19 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
20 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
23 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
24 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
25 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
26 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30/*
31 * This code does two things necessary for the enhanced TCP metrics to
32 * function in a useful manner:
33 * 1) It marks all non-host routes as `cloning', thus ensuring that
34 * every actual reference to such a route actually gets turned
35 * into a reference to a host route to the specific destination
36 * requested.
37 * 2) When such routes lose all their references, it arranges for them
38 * to be deleted in some random collection of circumstances, so that
39 * a large quantity of stale routing data is not kept in kernel memory
40 * indefinitely. See in_rtqtimo() below for the exact mechanism.
41 */
42
43#include <sys/cdefs.h>
|
44__FBSDID("$FreeBSD: head/sys/netinet/in_rmx.c 191816 2009-05-05 10:56:12Z zec $");
| 44__FBSDID("$FreeBSD: head/sys/netinet/in_rmx.c 193232 2009-06-01 15:49:42Z bz $");
|
45
46#include "opt_route.h"
47
48#include <sys/param.h>
49#include <sys/systm.h>
50#include <sys/kernel.h>
51#include <sys/sysctl.h>
52#include <sys/socket.h>
53#include <sys/mbuf.h>
54#include <sys/syslog.h>
55#include <sys/callout.h>
56#include <sys/vimage.h>
57
58#include <net/if.h>
59#include <net/route.h>
60#include <net/vnet.h>
61
62#include <netinet/in.h>
63#include <netinet/in_var.h>
64#include <netinet/ip_var.h>
65#include <netinet/vinet.h>
66
67extern int in_inithead(void **head, int off);
68
69#define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */
70
71/*
72 * Do what we need to do when inserting a route.
73 */
74static struct radix_node *
75in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
76 struct radix_node *treenodes)
77{
78 struct rtentry *rt = (struct rtentry *)treenodes;
79 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
80
81 RADIX_NODE_HEAD_WLOCK_ASSERT(head);
82 /*
83 * A little bit of help for both IP output and input:
84 * For host routes, we make sure that RTF_BROADCAST
85 * is set for anything that looks like a broadcast address.
86 * This way, we can avoid an expensive call to in_broadcast()
87 * in ip_output() most of the time (because the route passed
88 * to ip_output() is almost always a host route).
89 *
90 * We also do the same for local addresses, with the thought
91 * that this might one day be used to speed up ip_input().
92 *
93 * We also mark routes to multicast addresses as such, because
94 * it's easy to do and might be useful (but this is much more
95 * dubious since it's so easy to inspect the address).
96 */
97 if (rt->rt_flags & RTF_HOST) {
98 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
99 rt->rt_flags |= RTF_BROADCAST;
100 } else if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr ==
101 sin->sin_addr.s_addr) {
102 rt->rt_flags |= RTF_LOCAL;
103 }
104 }
105 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
106 rt->rt_flags |= RTF_MULTICAST;
107
108 if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp)
109 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
110
111 return (rn_addroute(v_arg, n_arg, head, treenodes));
112}
113
114/*
115 * This code is the inverse of in_clsroute: on first reference, if we
116 * were managing the route, stop doing so and set the expiration timer
117 * back off again.
118 */
119static struct radix_node *
120in_matroute(void *v_arg, struct radix_node_head *head)
121{
122 struct radix_node *rn = rn_match(v_arg, head);
123 struct rtentry *rt = (struct rtentry *)rn;
124
125 /*XXX locking? */
126 if (rt && rt->rt_refcnt == 0) { /* this is first reference */
127 if (rt->rt_flags & RTPRF_OURS) {
128 rt->rt_flags &= ~RTPRF_OURS;
129 rt->rt_rmx.rmx_expire = 0;
130 }
131 }
132 return rn;
133}
134
135#ifdef VIMAGE_GLOBALS
136static int rtq_reallyold;
137static int rtq_minreallyold;
138static int rtq_toomany;
139#endif
140
141SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_RTEXPIRE, rtexpire,
142 CTLFLAG_RW, rtq_reallyold, 0,
143 "Default expiration time on dynamically learned routes");
144
145SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_RTMINEXPIRE,
146 rtminexpire, CTLFLAG_RW, rtq_minreallyold, 0,
147 "Minimum time to attempt to hold onto dynamically learned routes");
148
149SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_RTMAXCACHE,
150 rtmaxcache, CTLFLAG_RW, rtq_toomany, 0,
151 "Upper limit on dynamically learned routes");
152
153/*
154 * On last reference drop, mark the route as belong to us so that it can be
155 * timed out.
156 */
157static void
158in_clsroute(struct radix_node *rn, struct radix_node_head *head)
159{
160 INIT_VNET_INET(curvnet);
161 struct rtentry *rt = (struct rtentry *)rn;
162
163 RT_LOCK_ASSERT(rt);
164
165 if (!(rt->rt_flags & RTF_UP))
166 return; /* prophylactic measures */
167
168 if (rt->rt_flags & RTPRF_OURS)
169 return;
170
171 if (!(rt->rt_flags & RTF_DYNAMIC))
172 return;
173
174 /*
175 * If rtq_reallyold is 0, just delete the route without
176 * waiting for a timeout cycle to kill it.
177 */
178 if (V_rtq_reallyold != 0) {
179 rt->rt_flags |= RTPRF_OURS;
180 rt->rt_rmx.rmx_expire = time_uptime + V_rtq_reallyold;
181 } else {
182 rtexpunge(rt);
183 }
184}
185
186struct rtqk_arg {
187 struct radix_node_head *rnh;
188 int draining;
189 int killed;
190 int found;
191 int updating;
192 time_t nextstop;
193};
194
195/*
196 * Get rid of old routes. When draining, this deletes everything, even when
197 * the timeout is not expired yet. When updating, this makes sure that
198 * nothing has a timeout longer than the current value of rtq_reallyold.
199 */
200static int
201in_rtqkill(struct radix_node *rn, void *rock)
202{
203 INIT_VNET_INET(curvnet);
204 struct rtqk_arg *ap = rock;
205 struct rtentry *rt = (struct rtentry *)rn;
206 int err;
207
208 RADIX_NODE_HEAD_WLOCK_ASSERT(ap->rnh);
209
210 if (rt->rt_flags & RTPRF_OURS) {
211 ap->found++;
212
213 if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) {
214 if (rt->rt_refcnt > 0)
215 panic("rtqkill route really not free");
216
217 err = in_rtrequest(RTM_DELETE,
218 (struct sockaddr *)rt_key(rt),
219 rt->rt_gateway, rt_mask(rt),
220 rt->rt_flags | RTF_RNH_LOCKED, 0,
221 rt->rt_fibnum);
222 if (err) {
223 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
224 } else {
225 ap->killed++;
226 }
227 } else {
228 if (ap->updating &&
229 (rt->rt_rmx.rmx_expire - time_uptime >
230 V_rtq_reallyold)) {
231 rt->rt_rmx.rmx_expire =
232 time_uptime + V_rtq_reallyold;
233 }
234 ap->nextstop = lmin(ap->nextstop,
235 rt->rt_rmx.rmx_expire);
236 }
237 }
238
239 return 0;
240}
241
242#define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */
243#ifdef VIMAGE_GLOBALS
244static int rtq_timeout;
245static struct callout rtq_timer;
246#endif
247
248static void in_rtqtimo_one(void *rock);
249
250static void
251in_rtqtimo(void *rock)
252{
253 CURVNET_SET((struct vnet *) rock);
| 45
46#include "opt_route.h"
47
48#include <sys/param.h>
49#include <sys/systm.h>
50#include <sys/kernel.h>
51#include <sys/sysctl.h>
52#include <sys/socket.h>
53#include <sys/mbuf.h>
54#include <sys/syslog.h>
55#include <sys/callout.h>
56#include <sys/vimage.h>
57
58#include <net/if.h>
59#include <net/route.h>
60#include <net/vnet.h>
61
62#include <netinet/in.h>
63#include <netinet/in_var.h>
64#include <netinet/ip_var.h>
65#include <netinet/vinet.h>
66
67extern int in_inithead(void **head, int off);
68
69#define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */
70
71/*
72 * Do what we need to do when inserting a route.
73 */
74static struct radix_node *
75in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
76 struct radix_node *treenodes)
77{
78 struct rtentry *rt = (struct rtentry *)treenodes;
79 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
80
81 RADIX_NODE_HEAD_WLOCK_ASSERT(head);
82 /*
83 * A little bit of help for both IP output and input:
84 * For host routes, we make sure that RTF_BROADCAST
85 * is set for anything that looks like a broadcast address.
86 * This way, we can avoid an expensive call to in_broadcast()
87 * in ip_output() most of the time (because the route passed
88 * to ip_output() is almost always a host route).
89 *
90 * We also do the same for local addresses, with the thought
91 * that this might one day be used to speed up ip_input().
92 *
93 * We also mark routes to multicast addresses as such, because
94 * it's easy to do and might be useful (but this is much more
95 * dubious since it's so easy to inspect the address).
96 */
97 if (rt->rt_flags & RTF_HOST) {
98 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
99 rt->rt_flags |= RTF_BROADCAST;
100 } else if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr ==
101 sin->sin_addr.s_addr) {
102 rt->rt_flags |= RTF_LOCAL;
103 }
104 }
105 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
106 rt->rt_flags |= RTF_MULTICAST;
107
108 if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp)
109 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
110
111 return (rn_addroute(v_arg, n_arg, head, treenodes));
112}
113
114/*
115 * This code is the inverse of in_clsroute: on first reference, if we
116 * were managing the route, stop doing so and set the expiration timer
117 * back off again.
118 */
119static struct radix_node *
120in_matroute(void *v_arg, struct radix_node_head *head)
121{
122 struct radix_node *rn = rn_match(v_arg, head);
123 struct rtentry *rt = (struct rtentry *)rn;
124
125 /*XXX locking? */
126 if (rt && rt->rt_refcnt == 0) { /* this is first reference */
127 if (rt->rt_flags & RTPRF_OURS) {
128 rt->rt_flags &= ~RTPRF_OURS;
129 rt->rt_rmx.rmx_expire = 0;
130 }
131 }
132 return rn;
133}
134
135#ifdef VIMAGE_GLOBALS
136static int rtq_reallyold;
137static int rtq_minreallyold;
138static int rtq_toomany;
139#endif
140
141SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_RTEXPIRE, rtexpire,
142 CTLFLAG_RW, rtq_reallyold, 0,
143 "Default expiration time on dynamically learned routes");
144
145SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_RTMINEXPIRE,
146 rtminexpire, CTLFLAG_RW, rtq_minreallyold, 0,
147 "Minimum time to attempt to hold onto dynamically learned routes");
148
149SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_RTMAXCACHE,
150 rtmaxcache, CTLFLAG_RW, rtq_toomany, 0,
151 "Upper limit on dynamically learned routes");
152
153/*
154 * On last reference drop, mark the route as belong to us so that it can be
155 * timed out.
156 */
157static void
158in_clsroute(struct radix_node *rn, struct radix_node_head *head)
159{
160 INIT_VNET_INET(curvnet);
161 struct rtentry *rt = (struct rtentry *)rn;
162
163 RT_LOCK_ASSERT(rt);
164
165 if (!(rt->rt_flags & RTF_UP))
166 return; /* prophylactic measures */
167
168 if (rt->rt_flags & RTPRF_OURS)
169 return;
170
171 if (!(rt->rt_flags & RTF_DYNAMIC))
172 return;
173
174 /*
175 * If rtq_reallyold is 0, just delete the route without
176 * waiting for a timeout cycle to kill it.
177 */
178 if (V_rtq_reallyold != 0) {
179 rt->rt_flags |= RTPRF_OURS;
180 rt->rt_rmx.rmx_expire = time_uptime + V_rtq_reallyold;
181 } else {
182 rtexpunge(rt);
183 }
184}
185
186struct rtqk_arg {
187 struct radix_node_head *rnh;
188 int draining;
189 int killed;
190 int found;
191 int updating;
192 time_t nextstop;
193};
194
195/*
196 * Get rid of old routes. When draining, this deletes everything, even when
197 * the timeout is not expired yet. When updating, this makes sure that
198 * nothing has a timeout longer than the current value of rtq_reallyold.
199 */
200static int
201in_rtqkill(struct radix_node *rn, void *rock)
202{
203 INIT_VNET_INET(curvnet);
204 struct rtqk_arg *ap = rock;
205 struct rtentry *rt = (struct rtentry *)rn;
206 int err;
207
208 RADIX_NODE_HEAD_WLOCK_ASSERT(ap->rnh);
209
210 if (rt->rt_flags & RTPRF_OURS) {
211 ap->found++;
212
213 if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) {
214 if (rt->rt_refcnt > 0)
215 panic("rtqkill route really not free");
216
217 err = in_rtrequest(RTM_DELETE,
218 (struct sockaddr *)rt_key(rt),
219 rt->rt_gateway, rt_mask(rt),
220 rt->rt_flags | RTF_RNH_LOCKED, 0,
221 rt->rt_fibnum);
222 if (err) {
223 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
224 } else {
225 ap->killed++;
226 }
227 } else {
228 if (ap->updating &&
229 (rt->rt_rmx.rmx_expire - time_uptime >
230 V_rtq_reallyold)) {
231 rt->rt_rmx.rmx_expire =
232 time_uptime + V_rtq_reallyold;
233 }
234 ap->nextstop = lmin(ap->nextstop,
235 rt->rt_rmx.rmx_expire);
236 }
237 }
238
239 return 0;
240}
241
242#define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */
243#ifdef VIMAGE_GLOBALS
244static int rtq_timeout;
245static struct callout rtq_timer;
246#endif
247
248static void in_rtqtimo_one(void *rock);
249
250static void
251in_rtqtimo(void *rock)
252{
253 CURVNET_SET((struct vnet *) rock);
|
254 INIT_VNET_NET(curvnet);
| |
255 INIT_VNET_INET(curvnet);
256 int fibnum;
257 void *newrock;
258 struct timeval atv;
259
260 for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
| 254 INIT_VNET_INET(curvnet);
255 int fibnum;
256 void *newrock;
257 struct timeval atv;
258
259 for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
|
261 if ((newrock = V_rt_tables[fibnum][AF_INET]) != NULL)
| 260 newrock = rt_tables_get_rnh(fibnum, AF_INET);
261 if (newrock != NULL)
|
262 in_rtqtimo_one(newrock);
263 }
264 atv.tv_usec = 0;
265 atv.tv_sec = V_rtq_timeout;
266 callout_reset(&V_rtq_timer, tvtohz(&atv), in_rtqtimo, rock);
267 CURVNET_RESTORE();
268}
269
270static void
271in_rtqtimo_one(void *rock)
272{
273 INIT_VNET_INET(curvnet);
274 struct radix_node_head *rnh = rock;
275 struct rtqk_arg arg;
276 static time_t last_adjusted_timeout = 0;
277
278 arg.found = arg.killed = 0;
279 arg.rnh = rnh;
280 arg.nextstop = time_uptime + V_rtq_timeout;
281 arg.draining = arg.updating = 0;
282 RADIX_NODE_HEAD_LOCK(rnh);
283 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
284 RADIX_NODE_HEAD_UNLOCK(rnh);
285
286 /*
287 * Attempt to be somewhat dynamic about this:
288 * If there are ``too many'' routes sitting around taking up space,
289 * then crank down the timeout, and see if we can't make some more
290 * go away. However, we make sure that we will never adjust more
291 * than once in rtq_timeout seconds, to keep from cranking down too
292 * hard.
293 */
294 if ((arg.found - arg.killed > V_rtq_toomany) &&
295 (time_uptime - last_adjusted_timeout >= V_rtq_timeout) &&
296 V_rtq_reallyold > V_rtq_minreallyold) {
297 V_rtq_reallyold = 2 * V_rtq_reallyold / 3;
298 if (V_rtq_reallyold < V_rtq_minreallyold) {
299 V_rtq_reallyold = V_rtq_minreallyold;
300 }
301
302 last_adjusted_timeout = time_uptime;
303#ifdef DIAGNOSTIC
304 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
305 V_rtq_reallyold);
306#endif
307 arg.found = arg.killed = 0;
308 arg.updating = 1;
309 RADIX_NODE_HEAD_LOCK(rnh);
310 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
311 RADIX_NODE_HEAD_UNLOCK(rnh);
312 }
313
314}
315
316void
317in_rtqdrain(void)
318{
319 VNET_ITERATOR_DECL(vnet_iter);
320 struct radix_node_head *rnh;
321 struct rtqk_arg arg;
322 int fibnum;
323
324 VNET_LIST_RLOCK();
325 VNET_FOREACH(vnet_iter) {
326 CURVNET_SET(vnet_iter);
| 262 in_rtqtimo_one(newrock);
263 }
264 atv.tv_usec = 0;
265 atv.tv_sec = V_rtq_timeout;
266 callout_reset(&V_rtq_timer, tvtohz(&atv), in_rtqtimo, rock);
267 CURVNET_RESTORE();
268}
269
270static void
271in_rtqtimo_one(void *rock)
272{
273 INIT_VNET_INET(curvnet);
274 struct radix_node_head *rnh = rock;
275 struct rtqk_arg arg;
276 static time_t last_adjusted_timeout = 0;
277
278 arg.found = arg.killed = 0;
279 arg.rnh = rnh;
280 arg.nextstop = time_uptime + V_rtq_timeout;
281 arg.draining = arg.updating = 0;
282 RADIX_NODE_HEAD_LOCK(rnh);
283 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
284 RADIX_NODE_HEAD_UNLOCK(rnh);
285
286 /*
287 * Attempt to be somewhat dynamic about this:
288 * If there are ``too many'' routes sitting around taking up space,
289 * then crank down the timeout, and see if we can't make some more
290 * go away. However, we make sure that we will never adjust more
291 * than once in rtq_timeout seconds, to keep from cranking down too
292 * hard.
293 */
294 if ((arg.found - arg.killed > V_rtq_toomany) &&
295 (time_uptime - last_adjusted_timeout >= V_rtq_timeout) &&
296 V_rtq_reallyold > V_rtq_minreallyold) {
297 V_rtq_reallyold = 2 * V_rtq_reallyold / 3;
298 if (V_rtq_reallyold < V_rtq_minreallyold) {
299 V_rtq_reallyold = V_rtq_minreallyold;
300 }
301
302 last_adjusted_timeout = time_uptime;
303#ifdef DIAGNOSTIC
304 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
305 V_rtq_reallyold);
306#endif
307 arg.found = arg.killed = 0;
308 arg.updating = 1;
309 RADIX_NODE_HEAD_LOCK(rnh);
310 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
311 RADIX_NODE_HEAD_UNLOCK(rnh);
312 }
313
314}
315
316void
317in_rtqdrain(void)
318{
319 VNET_ITERATOR_DECL(vnet_iter);
320 struct radix_node_head *rnh;
321 struct rtqk_arg arg;
322 int fibnum;
323
324 VNET_LIST_RLOCK();
325 VNET_FOREACH(vnet_iter) {
326 CURVNET_SET(vnet_iter);
|
327 INIT_VNET_NET(vnet_iter);
| |
328
329 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
| 327
328 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
|
330 rnh = V_rt_tables[fibnum][AF_INET];
| 329 rnh = rt_tables_get_rnh(fibnum, AF_INET);
|
331 arg.found = arg.killed = 0;
332 arg.rnh = rnh;
333 arg.nextstop = 0;
334 arg.draining = 1;
335 arg.updating = 0;
336 RADIX_NODE_HEAD_LOCK(rnh);
337 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
338 RADIX_NODE_HEAD_UNLOCK(rnh);
339 }
340 CURVNET_RESTORE();
341 }
342 VNET_LIST_RUNLOCK();
343}
344
345static int _in_rt_was_here;
346/*
347 * Initialize our routing tree.
348 */
349int
350in_inithead(void **head, int off)
351{
352 INIT_VNET_INET(curvnet);
353 struct radix_node_head *rnh;
354
355 /* XXX MRT
356 * This can be called from vfs_export.c too in which case 'off'
357 * will be 0. We know the correct value so just use that and
358 * return directly if it was 0.
359 * This is a hack that replaces an even worse hack on a bad hack
360 * on a bad design. After RELENG_7 this should be fixed but that
361 * will change the ABI, so for now do it this way.
362 */
363 if (!rn_inithead(head, 32))
364 return 0;
365
366 if (off == 0) /* XXX MRT see above */
367 return 1; /* only do the rest for a real routing table */
368
369 V_rtq_reallyold = 60*60; /* one hour is "really old" */
370 V_rtq_minreallyold = 10; /* never automatically crank down to less */
371 V_rtq_toomany = 128; /* 128 cached routes is "too many" */
372 V_rtq_timeout = RTQ_TIMEOUT;
373
374 rnh = *head;
375 rnh->rnh_addaddr = in_addroute;
376 rnh->rnh_matchaddr = in_matroute;
377 rnh->rnh_close = in_clsroute;
378 if (_in_rt_was_here == 0 ) {
379 callout_init(&V_rtq_timer, CALLOUT_MPSAFE);
380 callout_reset(&V_rtq_timer, 1, in_rtqtimo, curvnet);
381 _in_rt_was_here = 1;
382 }
383 return 1;
384}
385
386/*
387 * This zaps old routes when the interface goes down or interface
388 * address is deleted. In the latter case, it deletes static routes
389 * that point to this address. If we don't do this, we may end up
390 * using the old address in the future. The ones we always want to
391 * get rid of are things like ARP entries, since the user might down
392 * the interface, walk over to a completely different network, and
393 * plug back in.
394 */
395struct in_ifadown_arg {
396 struct ifaddr *ifa;
397 int del;
398};
399
400static int
401in_ifadownkill(struct radix_node *rn, void *xap)
402{
403 struct in_ifadown_arg *ap = xap;
404 struct rtentry *rt = (struct rtentry *)rn;
405
406 RT_LOCK(rt);
407 if (rt->rt_ifa == ap->ifa &&
408 (ap->del || !(rt->rt_flags & RTF_STATIC))) {
409 /*
410 * We need to disable the automatic prune that happens
411 * in this case in rtrequest() because it will blow
412 * away the pointers that rn_walktree() needs in order
413 * continue our descent. We will end up deleting all
414 * the routes that rtrequest() would have in any case,
415 * so that behavior is not needed there.
416 */
417 rtexpunge(rt);
418 }
419 RT_UNLOCK(rt);
420 return 0;
421}
422
423int
424in_ifadown(struct ifaddr *ifa, int delete)
425{
| 330 arg.found = arg.killed = 0;
331 arg.rnh = rnh;
332 arg.nextstop = 0;
333 arg.draining = 1;
334 arg.updating = 0;
335 RADIX_NODE_HEAD_LOCK(rnh);
336 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
337 RADIX_NODE_HEAD_UNLOCK(rnh);
338 }
339 CURVNET_RESTORE();
340 }
341 VNET_LIST_RUNLOCK();
342}
343
344static int _in_rt_was_here;
345/*
346 * Initialize our routing tree.
347 */
348int
349in_inithead(void **head, int off)
350{
351 INIT_VNET_INET(curvnet);
352 struct radix_node_head *rnh;
353
354 /* XXX MRT
355 * This can be called from vfs_export.c too in which case 'off'
356 * will be 0. We know the correct value so just use that and
357 * return directly if it was 0.
358 * This is a hack that replaces an even worse hack on a bad hack
359 * on a bad design. After RELENG_7 this should be fixed but that
360 * will change the ABI, so for now do it this way.
361 */
362 if (!rn_inithead(head, 32))
363 return 0;
364
365 if (off == 0) /* XXX MRT see above */
366 return 1; /* only do the rest for a real routing table */
367
368 V_rtq_reallyold = 60*60; /* one hour is "really old" */
369 V_rtq_minreallyold = 10; /* never automatically crank down to less */
370 V_rtq_toomany = 128; /* 128 cached routes is "too many" */
371 V_rtq_timeout = RTQ_TIMEOUT;
372
373 rnh = *head;
374 rnh->rnh_addaddr = in_addroute;
375 rnh->rnh_matchaddr = in_matroute;
376 rnh->rnh_close = in_clsroute;
377 if (_in_rt_was_here == 0 ) {
378 callout_init(&V_rtq_timer, CALLOUT_MPSAFE);
379 callout_reset(&V_rtq_timer, 1, in_rtqtimo, curvnet);
380 _in_rt_was_here = 1;
381 }
382 return 1;
383}
384
385/*
386 * This zaps old routes when the interface goes down or interface
387 * address is deleted. In the latter case, it deletes static routes
388 * that point to this address. If we don't do this, we may end up
389 * using the old address in the future. The ones we always want to
390 * get rid of are things like ARP entries, since the user might down
391 * the interface, walk over to a completely different network, and
392 * plug back in.
393 */
394struct in_ifadown_arg {
395 struct ifaddr *ifa;
396 int del;
397};
398
399static int
400in_ifadownkill(struct radix_node *rn, void *xap)
401{
402 struct in_ifadown_arg *ap = xap;
403 struct rtentry *rt = (struct rtentry *)rn;
404
405 RT_LOCK(rt);
406 if (rt->rt_ifa == ap->ifa &&
407 (ap->del || !(rt->rt_flags & RTF_STATIC))) {
408 /*
409 * We need to disable the automatic prune that happens
410 * in this case in rtrequest() because it will blow
411 * away the pointers that rn_walktree() needs in order
412 * continue our descent. We will end up deleting all
413 * the routes that rtrequest() would have in any case,
414 * so that behavior is not needed there.
415 */
416 rtexpunge(rt);
417 }
418 RT_UNLOCK(rt);
419 return 0;
420}
421
422int
423in_ifadown(struct ifaddr *ifa, int delete)
424{
|
426 INIT_VNET_NET(curvnet);
| |
427 struct in_ifadown_arg arg;
428 struct radix_node_head *rnh;
429 int fibnum;
430
431 if (ifa->ifa_addr->sa_family != AF_INET)
432 return 1;
433
434 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
| 425 struct in_ifadown_arg arg;
426 struct radix_node_head *rnh;
427 int fibnum;
428
429 if (ifa->ifa_addr->sa_family != AF_INET)
430 return 1;
431
432 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
|
435 rnh = V_rt_tables[fibnum][AF_INET];
| 433 rnh = rt_tables_get_rnh(fibnum, AF_INET);
|
436 arg.ifa = ifa;
437 arg.del = delete;
438 RADIX_NODE_HEAD_LOCK(rnh);
439 rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
440 RADIX_NODE_HEAD_UNLOCK(rnh);
441 ifa->ifa_flags &= ~IFA_ROUTE; /* XXXlocking? */
442 }
443 return 0;
444}
445
446/*
447 * inet versions of rt functions. These have fib extensions and
448 * for now will just reference the _fib variants.
449 * eventually this order will be reversed,
450 */
451void
452in_rtalloc_ign(struct route *ro, u_long ignflags, u_int fibnum)
453{
454 rtalloc_ign_fib(ro, ignflags, fibnum);
455}
456
457int
458in_rtrequest( int req,
459 struct sockaddr *dst,
460 struct sockaddr *gateway,
461 struct sockaddr *netmask,
462 int flags,
463 struct rtentry **ret_nrt,
464 u_int fibnum)
465{
466 return (rtrequest_fib(req, dst, gateway, netmask,
467 flags, ret_nrt, fibnum));
468}
469
470struct rtentry *
471in_rtalloc1(struct sockaddr *dst, int report, u_long ignflags, u_int fibnum)
472{
473 return (rtalloc1_fib(dst, report, ignflags, fibnum));
474}
475
476void
477in_rtredirect(struct sockaddr *dst,
478 struct sockaddr *gateway,
479 struct sockaddr *netmask,
480 int flags,
481 struct sockaddr *src,
482 u_int fibnum)
483{
484 rtredirect_fib(dst, gateway, netmask, flags, src, fibnum);
485}
486
487void
488in_rtalloc(struct route *ro, u_int fibnum)
489{
490 rtalloc_ign_fib(ro, 0UL, fibnum);
491}
492
493#if 0
494int in_rt_getifa(struct rt_addrinfo *, u_int fibnum);
495int in_rtioctl(u_long, caddr_t, u_int);
496int in_rtrequest1(int, struct rt_addrinfo *, struct rtentry **, u_int);
497#endif
498
499
| 434 arg.ifa = ifa;
435 arg.del = delete;
436 RADIX_NODE_HEAD_LOCK(rnh);
437 rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
438 RADIX_NODE_HEAD_UNLOCK(rnh);
439 ifa->ifa_flags &= ~IFA_ROUTE; /* XXXlocking? */
440 }
441 return 0;
442}
443
444/*
445 * inet versions of rt functions. These have fib extensions and
446 * for now will just reference the _fib variants.
447 * eventually this order will be reversed,
448 */
449void
450in_rtalloc_ign(struct route *ro, u_long ignflags, u_int fibnum)
451{
452 rtalloc_ign_fib(ro, ignflags, fibnum);
453}
454
455int
456in_rtrequest( int req,
457 struct sockaddr *dst,
458 struct sockaddr *gateway,
459 struct sockaddr *netmask,
460 int flags,
461 struct rtentry **ret_nrt,
462 u_int fibnum)
463{
464 return (rtrequest_fib(req, dst, gateway, netmask,
465 flags, ret_nrt, fibnum));
466}
467
468struct rtentry *
469in_rtalloc1(struct sockaddr *dst, int report, u_long ignflags, u_int fibnum)
470{
471 return (rtalloc1_fib(dst, report, ignflags, fibnum));
472}
473
474void
475in_rtredirect(struct sockaddr *dst,
476 struct sockaddr *gateway,
477 struct sockaddr *netmask,
478 int flags,
479 struct sockaddr *src,
480 u_int fibnum)
481{
482 rtredirect_fib(dst, gateway, netmask, flags, src, fibnum);
483}
484
485void
486in_rtalloc(struct route *ro, u_int fibnum)
487{
488 rtalloc_ign_fib(ro, 0UL, fibnum);
489}
490
491#if 0
492int in_rt_getifa(struct rt_addrinfo *, u_int fibnum);
493int in_rtioctl(u_long, caddr_t, u_int);
494int in_rtrequest1(int, struct rt_addrinfo *, struct rtentry **, u_int);
495#endif
496
497
|