d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Make net.fibs growable.

Browse Source 
Allow to dynamically grow the amount of fibs in each vnet.

This change alters current behavior. Currently, if one defines
 ROUTETABLES > 1 in the kernel config, each vnet will be created
 with the number of fibs defined in the kernel config.
 After this commit vnets will be created with fibs=1.

Dynamic net.fibs is not compatible with net.add_addr_allfibs.
 The plan is to deprecate the latter and make
 net.add_addr_allfibs=0 default behaviour.

Reviewed by:	glebius
Relnotes:	yes
Differential Revision:	https://reviews.freebsd.org/D26062
This commit is contained in:
Alexander V. Chernikov 2020-08-21 21:34:52 +00:00
parent 19337211f8
commit f5247a232a
Notes: svn2git 2020-12-20 02:59:44 +00:00 
svn path=/head/; revision=364465
5 changed files with 334 additions and 181 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
sys/conf/files
View File

@ -4102,6 +4102,7 @@ net/route/nhop_utils.c standard
net/route/route_ctl.c standard
net/route/route_ddb.c optional ddb
net/route/route_helpers.c standard
net/route/route_tables.c standard
net/route/route_temporal.c standard
net/rss_config.c optional inet rss | inet6 rss
net/rtsock.c standard

178
sys/net/route.c
View File

@ -74,29 +74,6 @@
#include <netinet/in.h>
#include <netinet/ip_mroute.h>
#include <vm/uma.h>
#define RT_MAXFIBS UINT16_MAX
/* Kernel config default option. */
#ifdef ROUTETABLES
#if ROUTETABLES <= 0
#error "ROUTETABLES defined too low"
#endif
#if ROUTETABLES > RT_MAXFIBS
#error "ROUTETABLES defined too big"
#endif
#define RT_NUMFIBS ROUTETABLES
#endif /* ROUTETABLES */
/* Initialize to default if not otherwise set. */
#ifndef RT_NUMFIBS
#define RT_NUMFIBS 1
#endif
/* This is read-only.. */
u_int rt_numfibs = RT_NUMFIBS;
SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, "");
/*
* By default add routes to all fibs for new interfaces.
* Once this is set to 0 then only allocate routes on interface
@ -118,10 +95,6 @@ VNET_PCPUSTAT_SYSINIT(rtstat);
VNET_PCPUSTAT_SYSUNINIT(rtstat);
#endif
VNET_DEFINE(struct rib_head *, rt_tables);
#define V_rt_tables VNET(rt_tables)
EVENTHANDLER_LIST_DEFINE(rt_addrmsg);
static int rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *,
@ -129,63 +102,6 @@ static int rt_ifdelroute(const struct rtentry *rt, const struct nhop_object *,
static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info,
int flags);
/*
* handler for net.my_fibnum
*/
static int
sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
{
int fibnum;
int error;
fibnum = curthread->td_proc->p_fibnum;
error = sysctl_handle_int(oidp, &fibnum, 0, req);
return (error);
}
SYSCTL_PROC(_net, OID_AUTO, my_fibnum,
CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
&sysctl_my_fibnum, "I",
"default FIB of caller");
static __inline struct rib_head **
rt_tables_get_rnh_ptr(int table, int fam)
{
struct rib_head **rnh;
KASSERT(table >= 0 && table < rt_numfibs,
("%s: table out of bounds (0 <= %d < %d)", __func__, table,
rt_numfibs));
KASSERT(fam >= 0 && fam < (AF_MAX + 1),
("%s: fam out of bounds (0 <= %d < %d)", __func__, fam, AF_MAX+1));
/* rnh is [fib=0][af=0]. */
rnh = (struct rib_head **)V_rt_tables;
/* Get the offset to the requested table and fam. */
rnh += table * (AF_MAX+1) + fam;
return (rnh);
}
struct rib_head *
rt_tables_get_rnh(int table, int fam)
{
return (*rt_tables_get_rnh_ptr(table, fam));
}
u_int
rt_tables_get_gen(int table, int fam)
{
struct rib_head *rnh;
rnh = *rt_tables_get_rnh_ptr(table, fam);
KASSERT(rnh != NULL, ("%s: NULL rib_head pointer table %d fam %d",
__func__, table, fam));
return (rnh->rnh_gen);
}
/*
* route initialization must occur before ip6_init2(), which happenas at
* SI_ORDER_MIDDLE.
@ -194,89 +110,10 @@ static void
route_init(void)
{
/* whack the tunable ints into line. */
if (rt_numfibs > RT_MAXFIBS)
rt_numfibs = RT_MAXFIBS;
if (rt_numfibs == 0)
rt_numfibs = 1;
nhops_init();
}
SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, NULL);
static void
vnet_route_init(const void *unused __unused)
{
struct domain *dom;
struct rib_head **rnh;
int table;
int fam;
V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) *
sizeof(struct rib_head *), M_RTABLE, M_WAITOK|M_ZERO);
vnet_rtzone_init();
for (dom = domains; dom; dom = dom->dom_next) {
if (dom->dom_rtattach == NULL)
continue;
for (table = 0; table < rt_numfibs; table++) {
fam = dom->dom_family;
if (table != 0 && fam != AF_INET6 && fam != AF_INET)
break;
rnh = rt_tables_get_rnh_ptr(table, fam);
if (rnh == NULL)
panic("%s: rnh NULL", __func__);
*rnh = dom->dom_rtattach(table);
}
}
}
VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
vnet_route_init, 0);
#ifdef VIMAGE
static void
vnet_route_uninit(const void *unused __unused)
{
int table;
int fam;
struct domain *dom;
struct rib_head **rnh;
for (dom = domains; dom; dom = dom->dom_next) {
if (dom->dom_rtdetach == NULL)
continue;
for (table = 0; table < rt_numfibs; table++) {
fam = dom->dom_family;
if (table != 0 && fam != AF_INET6 && fam != AF_INET)
break;
rnh = rt_tables_get_rnh_ptr(table, fam);
if (rnh == NULL)
panic("%s: rnh NULL", __func__);
dom->dom_rtdetach(*rnh);
}
}
/*
* dom_rtdetach calls rt_table_destroy(), which
* schedules deletion for all rtentries, nexthops and control
* structures. Wait for the destruction callbacks to fire.
* Note that this should result in freeing all rtentries, but
* nexthops deletions will be scheduled for the next epoch run
* and will be completed after vnet teardown.
*/
epoch_drain_callbacks(net_epoch_preempt);
free(V_rt_tables, M_RTABLE);
vnet_rtzone_destroy();
}
VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
vnet_route_uninit, 0);
#endif
struct rib_head *
rt_table_init(int offset, int family, u_int fibnum)
{
@ -347,21 +184,6 @@ rt_table_destroy(struct rib_head *rh)
free(rh, M_RTABLE);
}
#ifndef _SYS_SYSPROTO_H_
struct setfib_args {
int fibnum;
};
#endif
int
sys_setfib(struct thread *td, struct setfib_args *uap)
{
if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs)
return EINVAL;
td->td_proc->p_fibnum = uap->fibnum;
return (0);
}
/*
* Adds a temporal redirect entry to the routing table.
* @fibnum: fib number

7
sys/net/route.h
View File

@ -115,7 +115,10 @@ typedef volatile u_int rt_gen_t; /* tree generation (for adds) */
#define RT_DEFAULT_FIB 0 /* Explicitly mark fib=0 restricted cases */
#define RT_ALL_FIBS -1 /* Announce event for every fib */
#ifdef _KERNEL
extern u_int rt_numfibs; /* number of usable routing tables */
VNET_DECLARE(uint32_t, _rt_numfibs); /* number of existing route tables */
#define V_rt_numfibs VNET(_rt_numfibs)
/* temporary compat arg */
#define rt_numfibs V_rt_numfibs
VNET_DECLARE(u_int, rt_add_addr_allfibs); /* Announce interfaces to all fibs */
#define V_rt_add_addr_allfibs VNET(rt_add_addr_allfibs)
#endif
@ -379,7 +382,7 @@ void rt_newmaddrmsg(int, struct ifmultiaddr *);
void rt_maskedcopy(struct sockaddr *, struct sockaddr *, struct sockaddr *);
struct rib_head *rt_table_init(int, int, u_int);
void rt_table_destroy(struct rib_head *);
u_int rt_tables_get_gen(int table, int fam);
u_int rt_tables_get_gen(uint32_t table, sa_family_t family);
int rtsock_addrmsg(int, struct ifaddr *, int);
int rtsock_routemsg(int, struct rtentry *, struct ifnet *ifp, int, int);

326
sys/net/route/route_tables.c Normal file
View File

@ -0,0 +1,326 @@
/*-
* SPDX-License-Identifier: BSD-3-Clause
*
* Copyright (c) 1980, 1986, 1991, 1993
* The Regents of the University of California. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
/************************************************************************
* Note: In this file a 'fib' is a "forwarding information base" *
* Which is the new name for an in kernel routing (next hop) table. *
***********************************************************************/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_route.h"
#include <sys/param.h>
#include <sys/socket.h>
#include <sys/systm.h>
#include <sys/malloc.h>
#include <sys/jail.h>
#include <sys/sysctl.h>
#include <sys/syslog.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/sx.h>
#include <sys/domain.h>
#include <sys/sysproto.h>
#include <net/vnet.h>
#include <net/route.h>
#include <net/route/route_var.h>
/* Kernel config default option. */
#ifdef ROUTETABLES
#if ROUTETABLES <= 0
#error "ROUTETABLES defined too low"
#endif
#if ROUTETABLES > RT_MAXFIBS
#error "ROUTETABLES defined too big"
#endif
#define RT_NUMFIBS ROUTETABLES
#endif /* ROUTETABLES */
/* Initialize to default if not otherwise set. */
#ifndef RT_NUMFIBS
#define RT_NUMFIBS 1
#endif
static void grow_rtables(uint32_t num_fibs);
VNET_DEFINE_STATIC(struct sx, rtables_lock);
#define V_rtables_lock VNET(rtables_lock)
#define RTABLES_LOCK() sx_xlock(&V_rtables_lock)
#define RTABLES_UNLOCK() sx_xunlock(&V_rtables_lock)
#define RTABLES_LOCK_INIT() sx_init(&V_rtables_lock, "rtables lock")
#define RTABLES_LOCK_ASSERT() sx_assert(&V_rtables_lock, SA_LOCKED)
VNET_DEFINE_STATIC(struct rib_head **, rt_tables);
#define V_rt_tables VNET(rt_tables)
VNET_DEFINE(uint32_t, _rt_numfibs) = RT_NUMFIBS;
/*
* Handler for net.my_fibnum.
* Returns current fib of the process.
*/
static int
sysctl_my_fibnum(SYSCTL_HANDLER_ARGS)
{
int fibnum;
int error;
fibnum = curthread->td_proc->p_fibnum;
error = sysctl_handle_int(oidp, &fibnum, 0, req);
return (error);
}
SYSCTL_PROC(_net, OID_AUTO, my_fibnum,
CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
&sysctl_my_fibnum, "I",
"default FIB of caller");
static uint32_t
normalize_num_rtables(uint32_t num_rtables)
{
if (num_rtables > RT_MAXFIBS)
num_rtables = RT_MAXFIBS;
else if (num_rtables == 0)
num_rtables = 1;
return (num_rtables);
}
/*
* Sets the number of fibs in the current vnet.
* Function does not allow shrinking number of rtables.
*/
static int
sysctl_fibs(SYSCTL_HANDLER_ARGS)
{
uint32_t new_fibs;
int error;
RTABLES_LOCK();
new_fibs = V_rt_numfibs;
error = sysctl_handle_32(oidp, &new_fibs, 0, req);
if (error == 0) {
new_fibs = normalize_num_rtables(new_fibs);
if (new_fibs < V_rt_numfibs)
error = ENOTCAPABLE;
if (new_fibs > V_rt_numfibs)
grow_rtables(new_fibs);
}
RTABLES_UNLOCK();
return (error);
}
SYSCTL_PROC(_net, OID_AUTO, fibs,
CTLFLAG_VNET | CTLTYPE_U32 | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, NULL, 0,
&sysctl_fibs, "IU",
"set number of fibs");
/*
* Sets fib of a current process.
*/
int
sys_setfib(struct thread *td, struct setfib_args *uap)
{
int error = 0;
CURVNET_SET(TD_TO_VNET(td));
if (uap->fibnum >= 0 && uap->fibnum < V_rt_numfibs)
td->td_proc->p_fibnum = uap->fibnum;
else
error = EINVAL;
CURVNET_RESTORE();
return (error);
}
/*
* Grows up the number of routing tables in the current fib.
* Function creates new index array for all rtables and allocates
* remaining routing tables.
*/
static void
grow_rtables(uint32_t num_tables)
{
struct domain *dom;
struct rib_head **prnh;
struct rib_head **new_rt_tables, **old_rt_tables;
int family;
RTABLES_LOCK_ASSERT();
KASSERT(num_tables >= V_rt_numfibs, ("num_tables(%u) < rt_numfibs(%u)\n",
num_tables, V_rt_numfibs));
new_rt_tables = mallocarray(num_tables * (AF_MAX + 1), sizeof(void *),
M_RTABLE, M_WAITOK | M_ZERO);
/*
* Current rt_tables layout:
* fib0[af0, af1, af2, .., AF_MAX]fib1[af0, af1, af2, .., Af_MAX]..
* this allows to copy existing tables data by using memcpy()
*/
if (V_rt_tables != NULL)
memcpy(new_rt_tables, V_rt_tables,
V_rt_numfibs * (AF_MAX + 1) * sizeof(void *));
/* Populate the remainders */
for (dom = domains; dom; dom = dom->dom_next) {
if (dom->dom_rtattach == NULL)
continue;
family = dom->dom_family;
for (int i = 0; i < num_tables; i++) {
prnh = &new_rt_tables[i * (AF_MAX + 1) + family];
if (*prnh != NULL)
continue;
*prnh = dom->dom_rtattach(i);
if (*prnh == NULL)
log(LOG_ERR, "unable to create routing tables for domain %d\n",
dom->dom_family);
}
}
/*
* Update rtables pointer.
* Ensure all writes to new_rt_tables has been completed before
* switching pointer.
*/
atomic_thread_fence_rel();
old_rt_tables = V_rt_tables;
V_rt_tables = new_rt_tables;
/* Wait till all cpus see new pointers */
atomic_thread_fence_rel();
epoch_wait_preempt(net_epoch_preempt);
/* Finally, set number of fibs to a new value */
V_rt_numfibs = num_tables;
if (old_rt_tables != NULL)
free(old_rt_tables, M_RTABLE);
}
static void
vnet_rtables_init(const void *unused __unused)
{
int num_rtables_base;
if (IS_DEFAULT_VNET(curvnet)) {
num_rtables_base = RT_NUMFIBS;
TUNABLE_INT_FETCH("net.fibs", &num_rtables_base);
V_rt_numfibs = normalize_num_rtables(num_rtables_base);
} else
V_rt_numfibs = 1;
vnet_rtzone_init();
RTABLES_LOCK_INIT();
RTABLES_LOCK();
grow_rtables(V_rt_numfibs);
RTABLES_UNLOCK();
}
VNET_SYSINIT(vnet_rtables_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
vnet_rtables_init, 0);
#ifdef VIMAGE
static void
rtables_destroy(const void *unused __unused)
{
struct rib_head *rnh;
struct domain *dom;
int family;
RTABLES_LOCK();
for (dom = domains; dom; dom = dom->dom_next) {
if (dom->dom_rtdetach == NULL)
continue;
family = dom->dom_family;
for (int i = 0; i < V_rt_numfibs; i++) {
rnh = rt_tables_get_rnh(i, family);
dom->dom_rtdetach(rnh);
}
}
RTABLES_UNLOCK();
/*
* dom_rtdetach calls rt_table_destroy(), which
* schedules deletion for all rtentries, nexthops and control
* structures. Wait for the destruction callbacks to fire.
* Note that this should result in freeing all rtentries, but
* nexthops deletions will be scheduled for the next epoch run
* and will be completed after vnet teardown.
*/
epoch_drain_callbacks(net_epoch_preempt);
free(V_rt_tables, M_RTABLE);
vnet_rtzone_destroy();
}
VNET_SYSUNINIT(rtables_destroy, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
rtables_destroy, 0);
#endif
static inline struct rib_head *
rt_tables_get_rnh_ptr(uint32_t table, sa_family_t family)
{
struct rib_head **prnh;
KASSERT(table < V_rt_numfibs,
("%s: table out of bounds (%d < %d)", __func__, table,
V_rt_numfibs));
KASSERT(family < (AF_MAX + 1),
("%s: fam out of bounds (%d < %d)", __func__, family, AF_MAX + 1));
/* rnh is [fib=0][af=0]. */
prnh = V_rt_tables;
/* Get the offset to the requested table and fam. */
prnh += table * (AF_MAX + 1) + family;
return (*prnh);
}
struct rib_head *
rt_tables_get_rnh(uint32_t table, sa_family_t family)
{
return (rt_tables_get_rnh_ptr(table, family));
}
u_int
rt_tables_get_gen(uint32_t table, sa_family_t family)
{
struct rib_head *rnh;
rnh = rt_tables_get_rnh_ptr(table, family);
KASSERT(rnh != NULL, ("%s: NULL rib_head pointer table %d family %d",
__func__, table, family));
return (rnh->rnh_gen);
}

3
sys/net/route/route_var.h
View File

@ -79,6 +79,7 @@ struct rib_head {
/* Constants */
#define RIB_MAX_RETRIES 3
#define RT_MAXFIBS UINT16_MAX
/* Macro for verifying fields in af-specific 'struct route' structures */
#define CHK_STRUCT_FIELD_GENERIC(_s1, _f1, _s2, _f2) \
@ -104,7 +105,7 @@ CHK_STRUCT_ROUTE_FIELDS(_ro_new); \
_Static_assert(__offsetof(struct route, ro_dst) == __offsetof(_ro_new, _dst_new),\
"ro_dst and " #_dst_new " are at different offset")
struct rib_head *rt_tables_get_rnh(int fib, int family);
struct rib_head *rt_tables_get_rnh(uint32_t table, sa_family_t family);
void rt_mpath_init_rnh(struct rib_head *rnh);
int rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum);
void rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt);