Module Name: src
Committed By: knakahara
Date: Wed Dec 7 08:33:02 UTC 2022
Modified Files:
src/sys/netinet: ip_encap.c ip_encap.h
Log Message:
Refactor ip_encap.[ch]
- remove encap_attach() which is no longer used
- remove USE_RADIX code in ip_encap.c, which is used for
encap_attach() only
- remove mask members in encaptab
To generate a diff of this commit:
cvs rdiff -u -r1.76 -r1.77 src/sys/netinet/ip_encap.c
cvs rdiff -u -r1.27 -r1.28 src/sys/netinet/ip_encap.h
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/sys/netinet/ip_encap.c
diff -u src/sys/netinet/ip_encap.c:1.76 src/sys/netinet/ip_encap.c:1.77
--- src/sys/netinet/ip_encap.c:1.76 Wed Dec 7 08:28:46 2022
+++ src/sys/netinet/ip_encap.c Wed Dec 7 08:33:02 2022
@@ -1,4 +1,4 @@
-/* $NetBSD: ip_encap.c,v 1.76 2022/12/07 08:28:46 knakahara Exp $ */
+/* $NetBSD: ip_encap.c,v 1.77 2022/12/07 08:33:02 knakahara Exp $ */
/* $KAME: ip_encap.c,v 1.73 2001/10/02 08:30:58 itojun Exp $ */
/*
@@ -57,18 +57,8 @@
*/
/* XXX is M_NETADDR correct? */
-/*
- * With USE_RADIX the code will use radix table for tunnel lookup, for
- * tunnels registered with encap_attach() with a addr/mask pair.
- * Faster on machines with thousands of tunnel registerations (= interfaces).
- *
- * The code assumes that radix table code can handle non-continuous netmask,
- * as it will pass radix table memory region with (src + dst) sockaddr pair.
- */
-#define USE_RADIX
-
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ip_encap.c,v 1.76 2022/12/07 08:28:46 knakahara Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ip_encap.c,v 1.77 2022/12/07 08:33:02 knakahara Exp $");
#ifdef _KERNEL_OPT
#include "opt_mrouting.h"
@@ -128,13 +118,6 @@ static struct encaptab *encap6_lookup(st
static int encap_add(struct encaptab *);
static int encap_remove(struct encaptab *);
static void encap_afcheck(int, const struct sockaddr *, const struct sockaddr *);
-#ifdef USE_RADIX
-static struct radix_node_head *encap_rnh(int);
-static int mask_matchlen(const struct sockaddr *);
-#else
-static int mask_match(const struct encaptab *, const struct sockaddr *,
- const struct sockaddr *);
-#endif
static void encap_key_init(struct encap_key *, const struct sockaddr *,
const struct sockaddr *);
static void encap_key_inc(struct encap_key *);
@@ -158,11 +141,6 @@ static struct {
struct lwp *busy;
} encap_whole __cacheline_aligned;
-#ifdef USE_RADIX
-struct radix_node_head *encap_head[2]; /* 0 for AF_INET, 1 for AF_INET6 */
-static bool encap_head_updating = false;
-#endif
-
static thmap_t *encap_map[2]; /* 0 for AF_INET, 1 for AF_INET6 */
static bool encap_initialized = false;
@@ -205,18 +183,6 @@ encap_init(void)
PSLIST_INIT(&encap_table);
#endif
-#ifdef USE_RADIX
- /*
- * initialize radix lookup table when the radix subsystem is inited.
- */
- rn_delayedinit((void *)&encap_head[0],
- sizeof(struct sockaddr_pack) << 3);
-#ifdef INET6
- rn_delayedinit((void *)&encap_head[1],
- sizeof(struct sockaddr_pack) << 3);
-#endif
-#endif
-
encap_map[0] = thmap_create(0, NULL, THMAP_NOCOPY);
#ifdef INET6
encap_map[1] = thmap_create(0, NULL, THMAP_NOCOPY);
@@ -233,10 +199,6 @@ encap4_lookup(struct mbuf *m, int off, i
struct encaptab *ep, *match;
int prio, matchprio;
int s;
-#ifdef USE_RADIX
- struct radix_node_head *rnh = encap_rnh(AF_INET);
- struct radix_node *rn;
-#endif
thmap_t *emap = encap_map[0];
struct encap_key key;
@@ -260,26 +222,6 @@ encap4_lookup(struct mbuf *m, int off, i
matchprio = 0;
s = pserialize_read_enter();
-#ifdef USE_RADIX
- if (encap_head_updating) {
- /*
- * Update in progress. Do nothing.
- */
- pserialize_read_exit(s);
- return NULL;
- }
-
- rn = rnh->rnh_matchaddr((void *)&pack, rnh);
- if (rn && (rn->rn_flags & RNF_ROOT) == 0) {
- struct encaptab *encapp = (struct encaptab *)rn;
-
- psref_acquire(match_psref, &encapp->psref,
- encaptab.elem_class);
- match = encapp;
- matchprio = mask_matchlen(match->srcmask) +
- mask_matchlen(match->dstmask);
- }
-#endif
encap_key_init(&key, sintosa(&pack.mine), sintosa(&pack.yours));
while ((ep = thmap_get(emap, &key, sizeof(key))) != NULL) {
@@ -335,42 +277,21 @@ encap4_lookup(struct mbuf *m, int off, i
psref_acquire(&elem_psref, &ep->psref,
encaptab.elem_class);
- if (ep->func) {
- pserialize_read_exit(s);
- /* ep->func is sleepable. e.g. rtalloc1 */
- prio = (*ep->func)(m, off, proto, ep->arg);
- s = pserialize_read_enter();
- } else {
-#ifdef USE_RADIX
- psref_release(&elem_psref, &ep->psref,
- encaptab.elem_class);
- continue;
-#else
- prio = mask_match(ep, (struct sockaddr *)&pack.mine,
- (struct sockaddr *)&pack.yours);
-#endif
- }
+ pserialize_read_exit(s);
+ /* ep->func is sleepable. e.g. rtalloc1 */
+ prio = (*ep->func)(m, off, proto, ep->arg);
+ s = pserialize_read_enter();
/*
* We prioritize the matches by using bit length of the
- * matches. mask_match() and user-supplied matching function
+ * matches. user-supplied matching function
* should return the bit length of the matches (for example,
* if both src/dst are matched for IPv4, 64 should be returned).
* 0 or negative return value means "it did not match".
*
- * The question is, since we have two "mask" portion, we
- * cannot really define total order between entries.
- * For example, which of these should be preferred?
- * mask_match() returns 48 (32 + 16) for both of them.
- * src=3ffe::/16, dst=3ffe:501::/32
- * src=3ffe:501::/32, dst=3ffe::/16
- *
* We need to loop through all the possible candidates
* to get the best match - the search takes O(n) for
* n attachments (i.e. interfaces).
- *
- * For radix-based lookup, I guess source takes precedence.
- * See rn_{refines,lexobetter} for the correct answer.
*/
if (prio <= 0) {
psref_release(&elem_psref, &ep->psref,
@@ -440,10 +361,6 @@ encap6_lookup(struct mbuf *m, int off, i
int prio, matchprio;
int s;
struct encaptab *ep, *match;
-#ifdef USE_RADIX
- struct radix_node_head *rnh = encap_rnh(AF_INET6);
- struct radix_node *rn;
-#endif
thmap_t *emap = encap_map[1];
struct encap_key key;
@@ -467,26 +384,6 @@ encap6_lookup(struct mbuf *m, int off, i
matchprio = 0;
s = pserialize_read_enter();
-#ifdef USE_RADIX
- if (encap_head_updating) {
- /*
- * Update in progress. Do nothing.
- */
- pserialize_read_exit(s);
- return NULL;
- }
-
- rn = rnh->rnh_matchaddr((void *)&pack, rnh);
- if (rn && (rn->rn_flags & RNF_ROOT) == 0) {
- struct encaptab *encapp = (struct encaptab *)rn;
-
- psref_acquire(match_psref, &encapp->psref,
- encaptab.elem_class);
- match = encapp;
- matchprio = mask_matchlen(match->srcmask) +
- mask_matchlen(match->dstmask);
- }
-#endif
encap_key_init(&key, sin6tosa(&pack.mine), sin6tosa(&pack.yours));
while ((ep = thmap_get(emap, &key, sizeof(key))) != NULL) {
@@ -542,21 +439,10 @@ encap6_lookup(struct mbuf *m, int off, i
psref_acquire(&elem_psref, &ep->psref,
encaptab.elem_class);
- if (ep->func) {
- pserialize_read_exit(s);
- /* ep->func is sleepable. e.g. rtalloc1 */
- prio = (*ep->func)(m, off, proto, ep->arg);
- s = pserialize_read_enter();
- } else {
-#ifdef USE_RADIX
- psref_release(&elem_psref, &ep->psref,
- encaptab.elem_class);
- continue;
-#else
- prio = mask_match(ep, (struct sockaddr *)&pack.mine,
- (struct sockaddr *)&pack.yours);
-#endif
- }
+ pserialize_read_exit(s);
+ /* ep->func is sleepable. e.g. rtalloc1 */
+ prio = (*ep->func)(m, off, proto, ep->arg);
+ s = pserialize_read_enter();
/* see encap4_lookup() for issues here */
if (prio <= 0) {
@@ -624,82 +510,24 @@ encap6_input(struct mbuf **mp, int *offp
}
#endif
-/*
- * XXX
- * The encaptab list and the rnh radix tree must be manipulated atomically.
- */
static int
encap_add(struct encaptab *ep)
{
-#ifdef USE_RADIX
- struct radix_node_head *rnh = encap_rnh(ep->af);
-#endif
KASSERT(encap_lock_held());
-#ifdef USE_RADIX
- if (!ep->func && rnh) {
- /* Disable access to the radix tree for reader. */
- encap_head_updating = true;
- /* Wait for all readers to drain. */
- pserialize_perform(encaptab.psz);
-
- if (!rnh->rnh_addaddr((void *)ep->addrpack,
- (void *)ep->maskpack, rnh, ep->nodes)) {
- encap_head_updating = false;
- return EEXIST;
- }
-
- /*
- * The ep added to the radix tree must be skipped while
- * encap[46]_lookup walks encaptab list. In other words,
- * encap_add() does not need to care whether the ep has
- * been added encaptab list or not yet.
- * So, we can re-enable access to the radix tree for now.
- */
- encap_head_updating = false;
- }
-#endif
PSLIST_WRITER_INSERT_HEAD(&encap_table, ep, chain);
return 0;
}
-/*
- * XXX
- * The encaptab list and the rnh radix tree must be manipulated atomically.
- */
static int
encap_remove(struct encaptab *ep)
{
-#ifdef USE_RADIX
- struct radix_node_head *rnh = encap_rnh(ep->af);
-#endif
int error = 0;
KASSERT(encap_lock_held());
-#ifdef USE_RADIX
- if (!ep->func && rnh) {
- /* Disable access to the radix tree for reader. */
- encap_head_updating = true;
- /* Wait for all readers to drain. */
- pserialize_perform(encaptab.psz);
-
- if (!rnh->rnh_deladdr((void *)ep->addrpack,
- (void *)ep->maskpack, rnh))
- error = ESRCH;
-
- /*
- * The ep added to the radix tree must be skipped while
- * encap[46]_lookup walks encaptab list. In other words,
- * encap_add() does not need to care whether the ep has
- * been added encaptab list or not yet.
- * So, we can re-enable access to the radix tree for now.
- */
- encap_head_updating = false;
- }
-#endif
PSLIST_WRITER_REMOVE(ep, chain);
return error;
@@ -717,141 +545,6 @@ encap_afcheck(int af, const struct socka
KASSERT(len != 0 && len == sp->sa_len && len == dp->sa_len);
}
-/*
- * sp (src ptr) is always my side, and dp (dst ptr) is always remote side.
- * length of mask (sm and dm) is assumed to be same as sp/dp.
- * Return value will be necessary as input (cookie) for encap_detach().
- */
-const struct encaptab *
-encap_attach(int af, int proto,
- const struct sockaddr *sp, const struct sockaddr *sm,
- const struct sockaddr *dp, const struct sockaddr *dm,
- const struct encapsw *esw, void *arg)
-{
- struct encaptab *ep;
- int error;
- int pss;
- size_t l;
- struct ip_pack4 *pack4;
-#ifdef INET6
- struct ip_pack6 *pack6;
-#endif
-#ifndef ENCAP_MPSAFE
- int s;
-
- s = splsoftnet();
-#endif
-
- ASSERT_SLEEPABLE();
-
- /* sanity check on args */
- encap_afcheck(af, sp, dp);
-
- /* check if anyone have already attached with exactly same config */
- pss = pserialize_read_enter();
- PSLIST_READER_FOREACH(ep, &encap_table, struct encaptab, chain) {
- if (ep->af != af)
- continue;
- if (ep->proto != proto)
- continue;
- if (ep->func)
- continue;
-
- KASSERT(ep->src != NULL);
- KASSERT(ep->dst != NULL);
- KASSERT(ep->srcmask != NULL);
- KASSERT(ep->dstmask != NULL);
-
- if (ep->src->sa_len != sp->sa_len ||
- memcmp(ep->src, sp, sp->sa_len) != 0 ||
- memcmp(ep->srcmask, sm, sp->sa_len) != 0)
- continue;
- if (ep->dst->sa_len != dp->sa_len ||
- memcmp(ep->dst, dp, dp->sa_len) != 0 ||
- memcmp(ep->dstmask, dm, dp->sa_len) != 0)
- continue;
-
- error = EEXIST;
- pserialize_read_exit(pss);
- goto fail;
- }
- pserialize_read_exit(pss);
-
- switch (af) {
- case AF_INET:
- l = sizeof(*pack4);
- break;
-#ifdef INET6
- case AF_INET6:
- l = sizeof(*pack6);
- break;
-#endif
- default:
- goto fail;
- }
-
- /* M_NETADDR ok? */
- ep = kmem_zalloc(sizeof(*ep), KM_SLEEP);
- ep->addrpack = kmem_zalloc(l, KM_SLEEP);
- ep->maskpack = kmem_zalloc(l, KM_SLEEP);
-
- ep->af = af;
- ep->proto = proto;
- ep->addrpack->sa_len = l & 0xff;
- ep->maskpack->sa_len = l & 0xff;
- switch (af) {
- case AF_INET:
- pack4 = (struct ip_pack4 *)ep->addrpack;
- ep->src = (struct sockaddr *)&pack4->mine;
- ep->dst = (struct sockaddr *)&pack4->yours;
- pack4 = (struct ip_pack4 *)ep->maskpack;
- ep->srcmask = (struct sockaddr *)&pack4->mine;
- ep->dstmask = (struct sockaddr *)&pack4->yours;
- break;
-#ifdef INET6
- case AF_INET6:
- pack6 = (struct ip_pack6 *)ep->addrpack;
- ep->src = (struct sockaddr *)&pack6->mine;
- ep->dst = (struct sockaddr *)&pack6->yours;
- pack6 = (struct ip_pack6 *)ep->maskpack;
- ep->srcmask = (struct sockaddr *)&pack6->mine;
- ep->dstmask = (struct sockaddr *)&pack6->yours;
- break;
-#endif
- }
-
- memcpy(ep->src, sp, sp->sa_len);
- memcpy(ep->srcmask, sm, sp->sa_len);
- memcpy(ep->dst, dp, dp->sa_len);
- memcpy(ep->dstmask, dm, dp->sa_len);
- ep->esw = esw;
- ep->arg = arg;
- psref_target_init(&ep->psref, encaptab.elem_class);
-
- error = encap_add(ep);
- if (error)
- goto gc;
-
- error = 0;
-#ifndef ENCAP_MPSAFE
- splx(s);
-#endif
- return ep;
-
-gc:
- if (ep->addrpack)
- kmem_free(ep->addrpack, l);
- if (ep->maskpack)
- kmem_free(ep->maskpack, l);
- if (ep)
- kmem_free(ep, sizeof(*ep));
-fail:
-#ifndef ENCAP_MPSAFE
- splx(s);
-#endif
- return NULL;
-}
-
const struct encaptab *
encap_attach_func(int af, int proto,
encap_priofunc_t *func,
@@ -1192,107 +885,11 @@ encap_detach(const struct encaptab *cook
pserialize_perform(encaptab.psz);
psref_target_destroy(&p->psref,
encaptab.elem_class);
- if (!ep->func) {
- kmem_free(p->addrpack, ep->addrpack->sa_len);
- kmem_free(p->maskpack, ep->maskpack->sa_len);
- }
kmem_free(p, sizeof(*p));
return 0;
}
-#ifdef USE_RADIX
-static struct radix_node_head *
-encap_rnh(int af)
-{
-
- switch (af) {
- case AF_INET:
- return encap_head[0];
-#ifdef INET6
- case AF_INET6:
- return encap_head[1];
-#endif
- default:
- return NULL;
- }
-}
-
-static int
-mask_matchlen(const struct sockaddr *sa)
-{
- const char *p, *ep;
- int l;
-
- p = (const char *)sa;
- ep = p + sa->sa_len;
- p += 2; /* sa_len + sa_family */
-
- l = 0;
- while (p < ep) {
- l += (*p ? 8 : 0); /* estimate */
- p++;
- }
- return l;
-}
-#endif
-
-#ifndef USE_RADIX
-static int
-mask_match(const struct encaptab *ep,
- const struct sockaddr *sp,
- const struct sockaddr *dp)
-{
- struct sockaddr_storage s;
- struct sockaddr_storage d;
- int i;
- const u_int8_t *p, *q;
- u_int8_t *r;
- int matchlen;
-
- KASSERTMSG(ep->func == NULL, "wrong encaptab passed to mask_match");
-
- if (sp->sa_len > sizeof(s) || dp->sa_len > sizeof(d))
- return 0;
- if (sp->sa_family != ep->af || dp->sa_family != ep->af)
- return 0;
- if (sp->sa_len != ep->src->sa_len || dp->sa_len != ep->dst->sa_len)
- return 0;
-
- matchlen = 0;
-
- p = (const u_int8_t *)sp;
- q = (const u_int8_t *)ep->srcmask;
- r = (u_int8_t *)&s;
- for (i = 0 ; i < sp->sa_len; i++) {
- r[i] = p[i] & q[i];
- /* XXX estimate */
- matchlen += (q[i] ? 8 : 0);
- }
-
- p = (const u_int8_t *)dp;
- q = (const u_int8_t *)ep->dstmask;
- r = (u_int8_t *)&d;
- for (i = 0 ; i < dp->sa_len; i++) {
- r[i] = p[i] & q[i];
- /* XXX rough estimate */
- matchlen += (q[i] ? 8 : 0);
- }
-
- /* need to overwrite len/family portion as we don't compare them */
- s.ss_len = sp->sa_len;
- s.ss_family = sp->sa_family;
- d.ss_len = dp->sa_len;
- d.ss_family = dp->sa_family;
-
- if (memcmp(&s, ep->src, ep->src->sa_len) == 0 &&
- memcmp(&d, ep->dst, ep->dst->sa_len) == 0) {
- return matchlen;
- } else
- return 0;
-}
-#endif
-
int
encap_lock_enter(void)
{
Index: src/sys/netinet/ip_encap.h
diff -u src/sys/netinet/ip_encap.h:1.27 src/sys/netinet/ip_encap.h:1.28
--- src/sys/netinet/ip_encap.h:1.27 Wed Dec 7 08:28:46 2022
+++ src/sys/netinet/ip_encap.h Wed Dec 7 08:33:02 2022
@@ -1,4 +1,4 @@
-/* $NetBSD: ip_encap.h,v 1.27 2022/12/07 08:28:46 knakahara Exp $ */
+/* $NetBSD: ip_encap.h,v 1.28 2022/12/07 08:33:02 knakahara Exp $ */
/* $KAME: ip_encap.h,v 1.7 2000/03/25 07:23:37 sumikawa Exp $ */
/*
@@ -35,10 +35,6 @@
#ifdef _KERNEL
-#ifndef RNF_NORMAL
-#include <net/radix.h>
-#endif
-
#include <sys/pslist.h>
#include <sys/psref.h>
@@ -87,16 +83,12 @@ struct encap_key {
};
struct encaptab {
- struct radix_node nodes[2];
struct pslist_entry chain;
int af;
int proto; /* -1: don't care, I'll check myself */
- struct sockaddr *addrpack; /* malloc'ed, for radix lookup */
- struct sockaddr *maskpack; /* ditto */
+ struct sockaddr *addrpack; /* malloc'ed, for lookup */
struct sockaddr *src; /* my addr */
- struct sockaddr *srcmask;
struct sockaddr *dst; /* remote addr */
- struct sockaddr *dstmask;
encap_priofunc_t *func;
const struct encapsw *esw;
void *arg;
@@ -107,7 +99,7 @@ struct encaptab {
#define IP_ENCAP_ADDR_ENABLE __BIT(0)
-/* to lookup a pair of address using radix tree */
+/* to lookup a pair of address using map */
struct sockaddr_pack {
u_int8_t sp_len;
u_int8_t sp_family; /* not really used */
@@ -130,9 +122,6 @@ void encapinit(void);
void encap_init(void);
void encap4_input(struct mbuf *, int, int);
int encap6_input(struct mbuf **, int *, int);
-const struct encaptab *encap_attach(int, int, const struct sockaddr *,
- const struct sockaddr *, const struct sockaddr *,
- const struct sockaddr *, const struct encapsw *, void *);
const struct encaptab *encap_attach_func(int, int,
encap_priofunc_t *,
const struct encapsw *, void *);
