summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorDavid S. Miller <davem@davemloft.net>2021-05-17 23:47:58 +0300
committerDavid S. Miller <davem@davemloft.net>2021-05-17 23:47:58 +0300
commit58fee5fc83658aaacf60246aeab738946a9ba516 (patch)
tree23c6aecbf6b979243a90a6b025f1419eb7fbfaa7
parent77091933e453a258bbe9ff2aeb1c8d6fc1db7ef9 (diff)
parent6101ca0384e3778cf6ad4f938fbc094a0386ec01 (diff)
downloadlinux-58fee5fc83658aaacf60246aeab738946a9ba516.tar.xz
Merge branch 'ipv4-unicast'
Seth David Schoen says: ==================== Treat IPv4 lowest address as ordinary unicast address Treat the lowest address in a subnet (the address within the subnet which contains all 0 bits) as an ordinary unicast address instead of as a potential second broadcast address. For example, in subnet 192.168.17.24/29, which contains 8 addresses, make address 192.168.17.24 usable as a normal unicast address (while continuing to support 192.168.17.31 as a broadcast address). Since EVERY network number or subnet formerly had its host number 0 reserved, this patchset adds 1 more usable host address to every network and subnet (i.e., 2^(32-n)-1 instead of 2^(32-n)-2 addresses available for assignment on each IPv4 /n subnet). For small subnets, this is a significant gain; instead of 6 usable host addresses, a /29 would now contain 7, a 16% increase. The reserving of host number 0 for broadcast came about in RFC 1122 from 1989 (page 31, "IP addresses are not permitted to have the value 0 or -1 for any of the <Host-number>, <Network-number>, or <Subnet-number> fields (except in the special cases listed above)" and page 66, "There is a class of hosts [4.2BSD Unix and its derivatives, but not 4.3BSD] that use non-standard broadcast address forms, substituting 0 for -1. All hosts SHOULD recognize and accept any of these non-standard broadcast addresses as the destination address of an incoming datagram."). This has been repeated in subsequent RFCs, always with backwards-compatibility rationales. Network troubles (broadcast storms) ensued when some early hosts on a LAN treated the lowest address as unicast and others treated it as broadcast. Multiple 1989 changes to IP successfully prevented these. The key was adding the layering violation rule requiring hosts to ignore all IP datagrams with unicast destination addresses that were received in low-level (Ethernet) broadcasts. That change is still in effect, and this patchset does not alter it. All operating systems since 4.3BSD, including all the current BSD OSes, now use the standard IP broadcast address. 4.2BSD has been obsolete for more than 30 years, and all modern hosts ignore hardware broadcasts containing unicast IP addresses, so there is no modern likelihood of broadcast storms even when hosts disagree on the unicast vs. broadcast status of a given address. Tests with this patchset show that other Linux hosts on the local segment simply ignore a host numbered with the lowest address, both for incoming and outgoing packet purposes. They don't interoperate with it, but they also don't cause broadcast storms or any other malfunction. If patched, they have no trouble interoperating with a host at the lowest address. Unmodified "distant" hosts that are not on the same segment successfully interoperate, as long as the gateway on the local segment, and the local host itself using the lowest address, have this patch. (Distant hosts have no way of knowing whether a given address is the lowest address in a faraway network segment, so they treat it no differently than any other unicast address.) This means that each local site can change this behavior locally, resulting immediately in global interoperability with the newly usable lowest local address. Modern software and documentation continues to use the definition of the directed, or "net-directed", broadcast address as "a host ID of all one bits". The Internet no longer gets any benefit from having two different broadcast addresses usable on every Ethernet segment. I have not been able to find any documentation that suggests that users or software should ever intentionally use the all-zero form, or that justifies it other than as a historic Berkeleyism. RFCs 1112, 1812, and 3021 state that hosts and routers need to maintain compatibility with the old form -- but they give no rationale other than the past existence of the 4.2BSD behavior. We're happy to provide more historical details or information about behavior of other systems in this regard by e-mail or as future patches to kernel documentation files. ==================== Signed-off-by: David S. Miller <davem@davemloft.net>
-rw-r--r--net/ipv4/fib_frontend.c4
-rwxr-xr-xtools/testing/selftests/net/unicast_extensions.sh17
2 files changed, 10 insertions, 11 deletions
diff --git a/net/ipv4/fib_frontend.c b/net/ipv4/fib_frontend.c
index 84bb707bd88d..bfb345c88271 100644
--- a/net/ipv4/fib_frontend.c
+++ b/net/ipv4/fib_frontend.c
@@ -1122,10 +1122,8 @@ void fib_add_ifaddr(struct in_ifaddr *ifa)
prefix, ifa->ifa_prefixlen, prim,
ifa->ifa_rt_priority);
- /* Add network specific broadcasts, when it takes a sense */
+ /* Add the network broadcast address, when it makes sense */
if (ifa->ifa_prefixlen < 31) {
- fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
- prim, 0);
fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
32, prim, 0);
}
diff --git a/tools/testing/selftests/net/unicast_extensions.sh b/tools/testing/selftests/net/unicast_extensions.sh
index dbf0421986df..66354cdd5ce4 100755
--- a/tools/testing/selftests/net/unicast_extensions.sh
+++ b/tools/testing/selftests/net/unicast_extensions.sh
@@ -189,6 +189,15 @@ segmenttest 255.255.255.1 255.255.255.254 24 "assign and ping inside 255.255.255
route_test 240.5.6.7 240.5.6.1 255.1.2.1 255.1.2.3 24 "route between 240.5.6/24 and 255.1.2/24 (is allowed)"
route_test 0.200.6.7 0.200.38.1 245.99.101.1 245.99.200.111 16 "route between 0.200/16 and 245.99/16 (is allowed)"
#
+# Test support for lowest address ending in .0
+segmenttest 5.10.15.20 5.10.15.0 24 "assign and ping lowest address (/24)"
+#
+# Test support for lowest address not ending in .0
+segmenttest 192.168.101.192 192.168.101.193 26 "assign and ping lowest address (/26)"
+#
+# Routing using lowest address as a gateway/endpoint
+route_test 192.168.42.1 192.168.42.0 9.8.7.6 9.8.7.0 24 "routing using lowest address"
+#
# ==============================================
# ==== TESTS THAT CURRENTLY EXPECT FAILURE =====
# ==============================================
@@ -202,14 +211,6 @@ segmenttest 255.255.255.1 255.255.255.255 16 "assigning 255.255.255.255 (is forb
# Currently Linux does not allow this, so this should fail too
segmenttest 127.99.4.5 127.99.4.6 16 "assign and ping inside 127/8 (is forbidden)"
#
-# Test support for lowest address
-# Currently Linux does not allow this, so this should fail too
-segmenttest 5.10.15.20 5.10.15.0 24 "assign and ping lowest address (is forbidden)"
-#
-# Routing using lowest address as a gateway/endpoint
-# Currently Linux does not allow this, so this should fail too
-route_test 192.168.42.1 192.168.42.0 9.8.7.6 9.8.7.0 24 "routing using lowest address (is forbidden)"
-#
# Test support for unicast use of class D
# Currently Linux does not allow this, so this should fail too
segmenttest 225.1.2.3 225.1.2.200 24 "assign and ping class D address (is forbidden)"