| Summary: | IPV6 forced on in busybox | ||
|---|---|---|---|
| Product: | buildroot | Reporter: | Kenric Smith <ksmith> |
| Component: | Other | Assignee: | unassigned |
| Status: | RESOLVED FIXED | ||
| Severity: | trivial | CC: | buildroot |
| Priority: | P5 | ||
| Version: | 2015.11 | ||
| Target Milestone: | --- | ||
| Hardware: | Other | ||
| OS: | Linux | ||
| Host: | Target: | ||
| Build: | |||
| Attachments: | Patch to package makefile to remove forcing IPV6 | ||
(In reply to Kenric Smith from comment #0) That sounds like a bug in busybox that should be fixed instead of worked around then. What exactly goes wrong? Have you reported the bug to the busybox developers? Sorry for the confusion. Busybox is acting as it should, the issue is that Buildroot is telling Busybox to enable IPV6 no matter what. For Buildroot systems where IPV6 is not enabled (in the kernel), Buildroot shouldn't enable IPV6 in Busybox. Previously it appears that IPV6 was a Buildroot configuration option and that option was properly being passed into Busybox. When that option was removed, IPV6 was forced on in Busybox by the Buildroot configuration. The patch attached just allows Busybox to be configured or not for IPV6, in the menuconfig option of Busybox, instead of forcing it to be enabled in Busybox. (In reply to Kenric Smith from comment #0) Notice: I'm not against us removing the forced IPv6 (and just enable it in the default defconfig instead). We should probably also do it for largefile now that we always have that enabled, but I would like to understand what the problem is you are seeing. I did a quick test with qemu_arm_versatile_defconfig and the busybox defconfig tweaked to enable the ntpd daemon, and everything seems to work OK here: # date Thu Jan 1 00:00:50 UTC 1970 # ntpd -d -n -p pool.ntp.org ntpd: sending query to 77.243.43.213 ntpd: reply from 77.243.43.213: offset:+1457039097.162244 delay:0.036664 status:0x24 strat:2 refid:0xf00726c0 rootdelay:0.003418 reach:0x01 ntpd: sending query to 77.243.43.213 ntpd: reply from 77.243.43.213: offset:+1457039097.161038 delay:0.032410 status:0x24 strat:2 refid:0xf00726c0 rootdelay:0.003418 reach:0x03 ntpd: setting time to 2016-03-03 21:06:05.090957 (offset +1457039097.161038s) ntpd: sending query to 77.243.43.213 ntpd: reply from 77.243.43.213: offset:+0.001327 delay:0.033226 status:0x24 strat:2 refid:0xf00726c0 rootdelay:0.003418 reach:0x07 Try ntp.vt.edu. When I try it, I get: [root ~]# nslookup ntp.vt.edu Server: ******** Address 1: ******** Name: ntp.vt.edu Address 1: 2001:468:c80:2101:0:14b:a768:154f ntp-1.ipv6.vt.edu Address 2: 198.82.247.108 ntp-3.vt.edu Address 3: 198.82.247.51 ntp-1.vt.edu Address 4: 198.82.247.131 ntp-4.vt.edu Address 5: 198.82.247.71 ntp-2.vt.edu [root ~]# ntpd -w -p ntp.vt.edu ntpd: socket: Address family not supported by protocol I should have added, with the patch enabled, I get this behavior: [root ~]# nslookup ntp.vt.edu Server: ******* Address 1: ******* Name: ntp.vt.edu Address 1: (null) Address 2: 198.82.247.51 ntp-1.vt.edu Address 3: 198.82.247.131 ntp-4.vt.edu Address 4: 198.82.247.108 ntp-3.vt.edu Address 5: 198.82.247.71 ntp-2.vt.edu [root ~]# ntpd -w -d -p ntp.vt.edu ntpd: sending query to 198.82.247.131 ntpd: reply from 198.82.247.131: offset:+21623.503688 delay:0.048829 status:0x24 strat:2 refid:0xa4f752c6 rootdelay:0.000381 reach:0x01 (In reply to Kenric Smith from comment #4) So I tried again, this time with IPv6 support disabled in the kernel but enabled in the toolchain (external codesourcery/glibc): buildroot login: root # ifconfig eth0 Link encap:Ethernet HWaddr 52:54:00:12:34:56 inet addr:10.0.2.15 Bcast:10.0.2.255 Mask:255.255.255.0 UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:2 errors:0 dropped:0 overruns:0 frame:0 TX packets:2 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:1000 RX bytes:1152 (1.1 KiB) TX bytes:684 (684.0 B) lo Link encap:Local Loopback inet addr:127.0.0.1 Mask:255.0.0.0 UP LOOPBACK RUNNING MTU:65536 Metric:1 RX packets:0 errors:0 dropped:0 overruns:0 frame:0 TX packets:0 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:0 RX bytes:0 (0.0 B) TX bytes:0 (0.0 B) # date Thu Jan 1 00:00:13 UTC 1970 # ntpd -w -p ntp.vt.edu ntpd: reply from 198.82.247.51: offset:+1457040206.437536 delay:0.123510 status:0x24 strat:2 refid:0xa4f752c6 rootdelay:0.000351 reach:0x01 ntpd: reply from 198.82.247.51: offset:+1457040206.437670 delay:0.109456 status:0x24 strat:2 refid:0xa4f752c6 rootdelay:0.000351 reach:0x03 ntpd: reply from 198.82.247.51: offset:+1457040206.438839 delay:0.109959 status:0x24 strat:2 refid:0xa4f752c6 rootdelay:0.000351 reach:0x07 ntpd: reply from 198.82.247.51: offset:+1457040206.437485 delay:0.107279 status:0x24 strat:2 refid:0xa4f752c6 rootdelay:0.000351 reach:0x0f random: nonblocking pool is initialized So that works. If I instead try ntp-1.ipv6.vt.edu (which only has an AAAA record) it indeed fails: # ntpd -w -p ntp-1.ipv6.vt.edu ntpd: socket: Address family not supported by protocol Which makes sense as it cannot create an IPv6 socket: socket(PF_INET6, SOCK_DGRAM, IPPROTO_IP) = -1 EAFNOSUPPORT (Address family not supported by protocol) write(2, "ntpd: socket: Address family not"..., 55ntpd: socket: Address family not supported by protocol ) = 55 Looking at the ntpd code, it ends up calling xhost2sockaddr() on the peer name, which ends up doing a getnameinfo() with AF_UNSPEC, and using the first result without checking that it can create a socket - So if your C library prefers IPv6 over IPv4 this explains it. The function does contain a workaround though: If you enable ENABLE_FEATURE_PREFER_IPV4_ADDRESS then it will use the first AF_INET result if available. What C library do you use? uClibc-ng and the default config file for it. Confirmed setting ENABLE_FEATURE_PREFER_IPV4_ADDRESS also corrects the behavior. I personally still think forcing IPV6 in Busybox isn't great, but I also understand if you don't want to change the configuration. Feel free to close the issue as resolved, if you want to change the forced IPV6 in Busybox is up to you. (it results in a 20% larger executable for me) Disregard 20% comment. It was incorrect, it is only slightly larger. Fixed in git by: commit 7a5be2a042f3c1d55c5599ad819333a2150955ff Author: Peter Korsgaard <peter@korsgaard.com> Date: Fri Mar 4 16:47:38 2016 +0100 busybox: tweak IPv6/largefile handling Fixes #8736 When IPv6 and largefile options were removed from Buildroot, the code to force these options in busybox were still left in. There's no strong reason to forcefully enable these options (only to disable options if the system cannot support it like we do for nommu), so instead enable the options in our default defconfig, allowing people to override this if they use a custom config. While we're at it, enable the prefer-ipv4 option so network applets like ntpd doesn't fail when dual stacked hosts are resolved from a system without IPv6 support enabled in the kernel. Signed-off-by: Peter Korsgaard <peter@korsgaard.com> |
Created attachment 6371 [details] Patch to package makefile to remove forcing IPV6 It appears from the commit logs that the IPV6 config option was removed for buildroot, but the remnants of the removal are forcing FEATURE_IPV6 in busybox. The attached patch would remove setting that feature in busybox. Without this patch, the ntpd applet in busybox doesn't function properly, attempting to contact IPV6 hosts.