Merge pull request #3502 from donaldsharp/socket_to_me_baby

BSD route socket refactoring/cleanup
This commit is contained in:
Russ White 2018-12-19 11:48:42 -05:00 committed by GitHub
commit c52e2ecf95
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -89,170 +89,6 @@ static int kernel_rtm_add_labels(struct mpls_label_stack *nh_label,
}
#endif
/* Interface between zebra message and rtm message. */
static int kernel_rtm_ipv4(int cmd, const struct prefix *p,
const struct nexthop_group *ng, uint32_t metric)
{
struct sockaddr_in *mask = NULL;
struct sockaddr_in sin_dest, sin_mask, sin_gate;
#ifdef __OpenBSD__
struct sockaddr_mpls smpls;
#endif
union sockunion *smplsp = NULL;
struct nexthop *nexthop;
int nexthop_num = 0;
ifindex_t ifindex = 0;
int gate = 0;
int error;
char prefix_buf[PREFIX_STRLEN];
enum blackhole_type bh_type = BLACKHOLE_UNSPEC;
if (IS_ZEBRA_DEBUG_RIB)
prefix2str(p, prefix_buf, sizeof(prefix_buf));
memset(&sin_dest, 0, sizeof(struct sockaddr_in));
sin_dest.sin_family = AF_INET;
#ifdef HAVE_STRUCT_SOCKADDR_IN_SIN_LEN
sin_dest.sin_len = sizeof(struct sockaddr_in);
#endif /* HAVE_STRUCT_SOCKADDR_IN_SIN_LEN */
sin_dest.sin_addr = p->u.prefix4;
memset(&sin_mask, 0, sizeof(struct sockaddr_in));
memset(&sin_gate, 0, sizeof(struct sockaddr_in));
sin_gate.sin_family = AF_INET;
#ifdef HAVE_STRUCT_SOCKADDR_IN_SIN_LEN
sin_gate.sin_len = sizeof(struct sockaddr_in);
#endif /* HAVE_STRUCT_SOCKADDR_IN_SIN_LEN */
/* Make gateway. */
for (ALL_NEXTHOPS_PTR(ng, nexthop)) {
if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_RECURSIVE))
continue;
gate = 0;
char gate_buf[INET_ADDRSTRLEN] = "NULL";
/*
* XXX We need to refrain from kernel operations in some cases,
* but this if statement seems overly cautious - what about
* other than ADD and DELETE?
*/
if ((cmd == RTM_ADD && NEXTHOP_IS_ACTIVE(nexthop->flags))
|| (cmd == RTM_DELETE)) {
if (nexthop->type == NEXTHOP_TYPE_IPV4
|| nexthop->type == NEXTHOP_TYPE_IPV4_IFINDEX) {
sin_gate.sin_addr = nexthop->gate.ipv4;
gate = 1;
}
if (nexthop->type == NEXTHOP_TYPE_IFINDEX
|| nexthop->type == NEXTHOP_TYPE_IPV4_IFINDEX)
ifindex = nexthop->ifindex;
if (nexthop->type == NEXTHOP_TYPE_BLACKHOLE) {
struct in_addr loopback;
loopback.s_addr = htonl(INADDR_LOOPBACK);
sin_gate.sin_addr = loopback;
bh_type = nexthop->bh_type;
gate = 1;
}
if (gate && p->prefixlen == 32)
mask = NULL;
else {
masklen2ip(p->prefixlen, &sin_mask.sin_addr);
sin_mask.sin_family = AF_INET;
#ifdef HAVE_STRUCT_SOCKADDR_IN_SIN_LEN
sin_mask.sin_len =
sin_masklen(sin_mask.sin_addr);
#endif /* HAVE_STRUCT_SOCKADDR_IN_SIN_LEN */
mask = &sin_mask;
}
#ifdef __OpenBSD__
if (nexthop->nh_label
&& !kernel_rtm_add_labels(nexthop->nh_label,
&smpls))
continue;
smplsp = (union sockunion *)&smpls;
#endif
error = rtm_write(cmd, (union sockunion *)&sin_dest,
(union sockunion *)mask,
gate ? (union sockunion *)&sin_gate
: NULL,
smplsp, ifindex, bh_type, metric);
if (IS_ZEBRA_DEBUG_KERNEL) {
if (!gate) {
zlog_debug(
"%s: %s: attention! gate not found for re",
__func__, prefix_buf);
} else
inet_ntop(AF_INET, &sin_gate.sin_addr,
gate_buf, INET_ADDRSTRLEN);
}
switch (error) {
/* We only flag nexthops as being in FIB if rtm_write()
* did its work. */
case ZEBRA_ERR_NOERROR:
nexthop_num++;
if (IS_ZEBRA_DEBUG_KERNEL)
zlog_debug(
"%s: %s: successfully did NH %s",
__func__, prefix_buf, gate_buf);
if (cmd == RTM_ADD)
SET_FLAG(nexthop->flags,
NEXTHOP_FLAG_FIB);
break;
/* The only valid case for this error is kernel's
* failure to install
* a multipath route, which is common for FreeBSD. This
* should be
* ignored silently, but logged as an error otherwise.
*/
case ZEBRA_ERR_RTEXIST:
if (cmd != RTM_ADD)
flog_err(
EC_LIB_SYSTEM_CALL,
"%s: rtm_write() returned %d for command %d",
__func__, error, cmd);
continue;
/* Note any unexpected status returns */
default:
flog_err(
EC_LIB_SYSTEM_CALL,
"%s: %s: rtm_write() unexpectedly returned %d for command %s",
__func__,
prefix2str(p, prefix_buf,
sizeof(prefix_buf)),
error,
lookup_msg(rtm_type_str, cmd, NULL));
break;
}
} /* if (cmd and flags make sense) */
else if (IS_ZEBRA_DEBUG_KERNEL)
zlog_debug("%s: odd command %s for flags %d", __func__,
lookup_msg(rtm_type_str, cmd, NULL),
nexthop->flags);
} /* for (ALL_NEXTHOPS(...))*/
/* If there was no useful nexthop, then complain. */
if (nexthop_num == 0) {
if (IS_ZEBRA_DEBUG_KERNEL)
zlog_debug("%s: No useful nexthops were found in RIB prefix %s",
__func__, prefix2str(p, prefix_buf,
sizeof(prefix_buf)));
return 1;
}
return 0; /*XXX*/
}
#ifdef SIN6_LEN
/* Calculate sin6_len value for netmask socket value. */
static int sin6_masklen(struct in6_addr mask)
@ -278,11 +114,11 @@ static int sin6_masklen(struct in6_addr mask)
#endif /* SIN6_LEN */
/* Interface between zebra message and rtm message. */
static int kernel_rtm_ipv6(int cmd, const struct prefix *p,
const struct nexthop_group *ng, uint32_t metric)
static int kernel_rtm(int cmd, const struct prefix *p,
const struct nexthop_group *ng, uint32_t metric)
{
struct sockaddr_in6 *mask;
struct sockaddr_in6 sin_dest, sin_mask, sin_gate;
union sockunion sin_dest, sin_mask, sin_gate;
#ifdef __OpenBSD__
struct sockaddr_mpls smpls;
#endif
@ -290,48 +126,81 @@ static int kernel_rtm_ipv6(int cmd, const struct prefix *p,
struct nexthop *nexthop;
int nexthop_num = 0;
ifindex_t ifindex = 0;
int gate = 0;
bool gate = false;
int error;
char prefix_buf[PREFIX_STRLEN];
enum blackhole_type bh_type = BLACKHOLE_UNSPEC;
memset(&sin_dest, 0, sizeof(struct sockaddr_in6));
sin_dest.sin6_family = AF_INET6;
#ifdef SIN6_LEN
sin_dest.sin6_len = sizeof(struct sockaddr_in6);
#endif /* SIN6_LEN */
sin_dest.sin6_addr = p->u.prefix6;
if (IS_ZEBRA_DEBUG_RIB)
prefix2str(p, prefix_buf, sizeof(prefix_buf));
memset(&sin_mask, 0, sizeof(struct sockaddr_in6));
/*
* We only have the ability to ADD or DELETE at this point
* in time.
*/
if (cmd != RTM_ADD && cmd != RTM_DELETE) {
if (IS_ZEBRA_DEBUG_KERNEL)
zlog_debug("%s: %s odd command %s for flags %d",
__func__, prefix_buf,
lookup_msg(rtm_type_str, cmd, NULL),
nexthop->flags);
return 0;
}
memset(&sin_gate, 0, sizeof(struct sockaddr_in6));
sin_gate.sin6_family = AF_INET6;
memset(&sin_dest, 0, sizeof(sin_dest));
memset(&sin_gate, 0, sizeof(sin_gate));
memset(&sin_mask, 0, sizeof(sin_mask));
switch (p->family) {
case AF_INET:
sin_dest.sin.sin_family = AF_INET;
#ifdef HAVE_STRUCT_SOCKADDR_IN_SIN_LEN
sin_gate.sin6_len = sizeof(struct sockaddr_in6);
sin_dest.sin.sin_len = sizeof(sin_dest);
sin_gate.sin.sin_len = sizeof(sin_gate);
#endif /* HAVE_STRUCT_SOCKADDR_IN_SIN_LEN */
sin_dest.sin.sin_addr = p->u.prefix4;
sin_gate.sin.sin_family = AF_INET;
break;
case AF_INET6:
sin_dest.sin6.sin6_family = AF_INET6;
#ifdef SIN6_LEN
sin_dest.sin6.sin6_len = sizeof(sin_dest);
#endif /* SIN6_LEN */
sin_dest.sin6.sin6_addr = p->u.prefix6;
sin_gate.sin6.sin6_family = AF_INET6;
#ifdef HAVE_STRUCT_SOCKADDR_IN_SIN_LEN
sin_gate.sin6.sin6_len = sizeof(sin_gate);
#endif /* HAVE_STRUCT_SOCKADDR_IN_SIN_LEN */
break;
}
/* Make gateway. */
for (ALL_NEXTHOPS_PTR(ng, nexthop)) {
if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_RECURSIVE))
/*
* We only want to use the actual good nexthops
*/
if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_RECURSIVE) ||
!CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_ACTIVE))
continue;
gate = 0;
smplsp = NULL;
gate = false;
char gate_buf[INET_ADDRSTRLEN] = "NULL";
if ((cmd == RTM_ADD && NEXTHOP_IS_ACTIVE(nexthop->flags))
|| (cmd == RTM_DELETE)) {
if (nexthop->type == NEXTHOP_TYPE_IPV6
|| nexthop->type == NEXTHOP_TYPE_IPV6_IFINDEX) {
sin_gate.sin6_addr = nexthop->gate.ipv6;
gate = 1;
}
if (nexthop->type == NEXTHOP_TYPE_IFINDEX
|| nexthop->type == NEXTHOP_TYPE_IPV6_IFINDEX)
ifindex = nexthop->ifindex;
if (nexthop->type == NEXTHOP_TYPE_BLACKHOLE)
bh_type = nexthop->bh_type;
}
/* Under kame set interface index to link local address. */
switch (nexthop->type) {
case NEXTHOP_TYPE_IPV4:
case NEXTHOP_TYPE_IPV4_IFINDEX:
sin_gate.sin.sin_addr = nexthop->gate.ipv4;
sin_gate.sin.sin_family = AF_INET;
ifindex = nexthop->ifindex;
gate = true;
break;
case NEXTHOP_TYPE_IPV6:
case NEXTHOP_TYPE_IPV6_IFINDEX:
sin_gate.sin6.sin6_addr = nexthop->gate.ipv6;
sin_gate.sin6.sin6_family = AF_INET6;
ifindex = nexthop->ifindex;
/* Under kame set interface index to link local address */
#ifdef KAME
#define SET_IN6_LINKLOCAL_IFINDEX(a, i) \
@ -340,19 +209,49 @@ static int kernel_rtm_ipv6(int cmd, const struct prefix *p,
(a).s6_addr[3] = (i)&0xff; \
} while (0)
if (gate && IN6_IS_ADDR_LINKLOCAL(&sin_gate.sin6_addr))
SET_IN6_LINKLOCAL_IFINDEX(sin_gate.sin6_addr, ifindex);
if (IN6_IS_ADDR_LINKLOCAL(&sin_gate.sin6.sin6_addr))
SET_IN6_LINKLOCAL_IFINDEX(
sin_gate.sin6.sin6_addr,
ifindex);
#endif /* KAME */
if (gate && p->prefixlen == 128)
mask = NULL;
else {
masklen2ip6(p->prefixlen, &sin_mask.sin6_addr);
sin_mask.sin6_family = AF_INET6;
gate = true;
break;
case NEXTHOP_TYPE_IFINDEX:
ifindex = nexthop->ifindex;
break;
case NEXTHOP_TYPE_BLACKHOLE:
bh_type = nexthop->bh_type;
switch (p->family) {
case AFI_IP: {
struct in_addr loopback;
loopback.s_addr = htonl(INADDR_LOOPBACK);
sin_gate.sin.sin_addr = loopback;
gate = true;
}
break;
case AFI_IP6:
break;
}
}
switch (p->family) {
case AF_INET:
masklen2ip(p->prefixlen, &sin_mask.sin.sin_addr);
sin_mask.sin.sin_family = AF_INET;
#ifdef HAVE_STRUCT_SOCKADDR_IN_SIN_LEN
sin_mask.sin.sin_len = sin_masklen(
sin_mask.sin.sin_addr);
#endif /* HAVE_STRUCT_SOCKADDR_IN_SIN_LEN */
break;
case AF_INET6:
masklen2ip6(p->prefixlen, &sin_mask.sin6.sin6_addr);
sin_mask.sin6.sin6_family = AF_INET6;
#ifdef SIN6_LEN
sin_mask.sin6_len = sin6_masklen(sin_mask.sin6_addr);
sin_mask.sin6.sin6_len = sin6_masklen(
sin_mask.sin6.sin6_addr);
#endif /* SIN6_LEN */
mask = &sin_mask;
break;
}
#ifdef __OpenBSD__
@ -361,41 +260,69 @@ static int kernel_rtm_ipv6(int cmd, const struct prefix *p,
continue;
smplsp = (union sockunion *)&smpls;
#endif
error = rtm_write(cmd, &sin_dest, &sin_mask,
gate ? &sin_gate : NULL, smplsp,
ifindex, bh_type, metric);
error = rtm_write(cmd, (union sockunion *)&sin_dest,
(union sockunion *)mask,
gate ? (union sockunion *)&sin_gate : NULL,
smplsp, ifindex, bh_type, metric);
if (IS_ZEBRA_DEBUG_KERNEL) {
if (!gate) {
zlog_debug("%s: %s: attention! gate not found for re",
__func__, prefix_buf);
} else
inet_ntop(p->family == AFI_IP ? AF_INET
: AF_INET6,
&sin_gate.sin.sin_addr,
gate_buf, INET_ADDRSTRLEN);
}
switch (error) {
/* We only flag nexthops as being in FIB if
* rtm_write() did its work. */
case ZEBRA_ERR_NOERROR:
nexthop_num++;
if (IS_ZEBRA_DEBUG_KERNEL)
zlog_debug("%s: %s: successfully did NH %s",
__func__, prefix_buf, gate_buf);
if (cmd == RTM_ADD)
SET_FLAG(nexthop->flags, NEXTHOP_FLAG_FIB);
break;
/* Update installed nexthop info on success */
if ((cmd == RTM_ADD) && (error == ZEBRA_ERR_NOERROR))
SET_FLAG(nexthop->flags, NEXTHOP_FLAG_FIB);
/* The only valid case for this error is
* kernel's failure to install a multipath
* route, which is common for FreeBSD. This
* should be ignored silently, but logged as an error
* otherwise.
*/
case ZEBRA_ERR_RTEXIST:
if (cmd != RTM_ADD)
flog_err(EC_LIB_SYSTEM_CALL,
"%s: rtm_write() returned %d for command %d",
__func__, error, cmd);
continue;
nexthop_num++;
}
/* Note any unexpected status returns */
default:
flog_err(EC_LIB_SYSTEM_CALL,
"%s: %s: rtm_write() unexpectedly returned %d for command %s",
__func__,
prefix2str(p, prefix_buf,
sizeof(prefix_buf)),
error, lookup_msg(rtm_type_str, cmd, NULL));
break;
}
} /* for (ALL_NEXTHOPS(...))*/
/* If there is no useful nexthop then return. */
/* If there was no useful nexthop, then complain. */
if (nexthop_num == 0) {
if (IS_ZEBRA_DEBUG_KERNEL)
zlog_debug("kernel_rtm_ipv6(): No useful nexthop.");
zlog_debug("%s: No useful nexthops were found in RIB prefix %s",
__func__, prefix2str(p, prefix_buf,
sizeof(prefix_buf)));
return 1;
}
return 0; /*XXX*/
}
static int kernel_rtm(int cmd, const struct prefix *p,
const struct nexthop_group *ng, uint32_t metric)
{
switch (PREFIX_FAMILY(p)) {
case AF_INET:
return kernel_rtm_ipv4(cmd, p, ng, metric);
case AF_INET6:
return kernel_rtm_ipv6(cmd, p, ng, metric);
}
return 0;
}
/*
* Update or delete a prefix from the kernel,
* using info from a dataplane context struct.