Summary: | [www/node] segfault when installing npm on 8.4, 9.3 on i386 | ||
---|---|---|---|
Product: | Ports & Packages | Reporter: | Marcin Cieślak <saper> |
Component: | Individual Port(s) | Assignee: | Po-Chuan Hsieh <sunpoet> |
Status: | Closed FIXED | ||
Severity: | Affects Many People | CC: | bdrewery, linpct |
Priority: | --- | ||
Version: | Latest | ||
Hardware: | i386 | ||
OS: | Any | ||
URL: | http://beefy1.isc.freebsd.org/bulk/84i386-default/latest/logs/errors/npm-2.6.0.log | ||
See Also: | https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=178881 |
Description
Marcin Cieślak
2015-03-06 11:39:12 UTC
So, managed to get a backtrace: (gdb) bt #0 0x0888d9ad in uv_interface_addresses (addresses=0xffffcb68, count=0xffffcb64) at ../deps/uv/src/unix/freebsd.c:380 #1 0x0882b2ed in GetInterfaceAddresses (args=@0xffffcba0) at ../src/node_os.cc:237 #2 0x083c138d in v8::internal::FunctionCallbackArguments::Call (this=0xffffcc30, f=0x882b2a0 <GetInterfaceAddresses>) at ../deps/v8/src/arguments.cc:33 #3 0x083ea953 in Builtin_HandleApiCall (args_length=2, args_object=0xffffccac, isolate=0x29018800) at ../deps/v8/src/builtins.cc:1144 (gdb) print *ent $2 = {ifa_next = 0x29c6c01c, ifa_name = 0x29c6c36e "\006\003\006", ifa_flags = 34883, ifa_addr = 0x29c6c150, ifa_netmask = 0x0, ifa_dstaddr = 0x0, ifa_data = 0x29c6c154} also "ifconfig -a" looks strange in this jail: # ifconfig -a : flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> inet 213.239.217.105 netmask 0xffffffe0 broadcast 213.239.217.127 inet 213.239.217.125 netmask 0xffffffe0 broadcast 213.239.217.127 ifconfig: ioctl(SIOCGIFINFO_IN6): Device not configured : flags=8049<UP,LOOPBACK,RUNNING,MULTICAST> inet 127.0.0.1 netmask 0xff000000 ifconfig: ioctl(SIOCGIFINFO_IN6): Device not configured I think data returned by getifaddrs(3) are bogus and this is bug 178881 *** This bug has been marked as a duplicate of bug 178881 *** Only on 10.1 getifaddrs(3) appears to work, I wonder why. (also i386) FreeBSD nodejs_10_1_i386-default 10.1-RELEASE-p6 FreeBSD 10.1-RELEASE-p6 i386 My kernel is r270025, 10.0-STABLE The best way to test this is to query "os.networkInterfaces()" in node
On 8.4 and 9.3 i386 os.networkInterfaces() crashes the process, on
10.1 it just gives incorrect results.
Here's output from 10.1 i386 jail (running on amd64 kernel):
> os.networkInterfaces()
{ /* .. */
lo0:
[ { address: '<unknown sa family>',
netmask: '255.255.255.224',
family: '<unknown>',
mac: '00:00:00:00:00:00',
internal: true },
{ address: '<unknown sa family>',
netmask: '255.255.255.224',
family: '<unknown>',
mac: '00:00:00:00:00:00',
internal: true },
{ address: '127.0.0.1',
netmask: '255.0.0.0',
family: 'IPv4',
mac: '00:00:00:00:00:00',
internal: true } ] }
correct output:
{ /* .. */
lo0:
[ { address: '::1',
netmask: 'ffff:ffff:ffff:ffff:ffff:ffff:ffff:ffff',
family: 'IPv6',
mac: '00:00:00:00:00:00',
scopeid: 0,
internal: true },
{ address: 'fe80::1',
netmask: 'ffff:ffff:ffff:ffff::',
family: 'IPv6',
mac: '00:00:00:00:00:00',
scopeid: 2,
internal: true },
{ address: '127.0.0.1',
netmask: '255.0.0.0',
family: 'IPv4',
mac: '00:00:00:00:00:00',
internal: true } ] }
Here is a quick workaround to get www/npm to build on amd64 host running i386 jail: diff -ruN /usr/ports/www/npm/Makefile npm/Makefile --- /usr/ports/www/npm/Makefile 2015-03-11 16:10:57.000000000 +0000 +++ npm/Makefile 2015-03-13 12:50:19.000000000 +0000 @@ -3,6 +3,7 @@ PORTNAME= npm PORTVERSION= 2.7.0 +PORTREVISION= 1 CATEGORIES= www MASTER_SITES= LOCAL/sunpoet @@ -79,4 +80,11 @@ @sudo ${INSTALL} -m 644 ${MS_WRKSRC}/${PORTNAME}-${MS_VERSION}.tar.xz ${DISTDIR}/ @cd ${.CURDIR}/ && ${MAKE_CMD} makesum -.include <bsd.port.mk> +.include <bsd.port.pre.mk> + +.if ${ARCH} == "i386" +# Workaround for kernel bug 178881 +EXTRA_PATCHES+= ${PATCHDIR}/bug178881-workaround +.endif + +.include <bsd.port.post.mk> diff -ruN /usr/ports/www/npm/files/bug178881-workaround npm/files/bug178881-workaround --- /usr/ports/www/npm/files/bug178881-workaround 1970-01-01 00:00:00.000000000 +0000 +++ npm/files/bug178881-workaround 2015-03-13 11:13:41.000000000 +0000 @@ -0,0 +1,20 @@ +--- lib/config/defaults.js.orig 2015-02-20 16:48:40.000000000 +0000 ++++ lib/config/defaults.js 2015-03-13 11:11:16.000000000 +0000 +@@ -325,16 +325,7 @@ + } + + function getLocalAddresses() { +- Object.keys(os.networkInterfaces()).map(function (nic) { +- return os.networkInterfaces()[nic].filter(function (addr) { +- return addr.family === "IPv4" +- }) +- .map(function (addr) { +- return addr.address +- }) +- }).reduce(function (curr, next) { +- return curr.concat(next) +- }, []).concat(undefined) ++ return [ '127.0.0.1', undefined ] + } + + exports.shorthands = it just cheats npm to think it only has 127.0.0.1 IPv4 address. of course the "i386" test is wrong, because we have a problem probably only on amd64 host/i386 guest combination. This may be a duplicate, but it is not fixed. I searched for open npm bugs and could not find this. A commit references this bug: Author: sunpoet Date: Sat May 16 17:02:38 UTC 2015 New revision: 386559 URL: https://svnweb.freebsd.org/changeset/ports/386559 Log: - Revert r386459 - Add workaround for i386 to deal with kernel bug 178881 [1] (PR/178881: getifaddrs(3) does not report IPv6 addresses properly in 32-bit compatibility mode) PR: 198350 [1] Submitted by: Marcin Cieslak <saper@saper.info> [1] Thanks to: adamw, bdrewery MFH: 2015Q2 Changes: head/www/npm/Makefile head/www/npm/files/ head/www/npm/files/extra-patch-bug-178881 Committed. Thanks! (In reply to commit-hook from comment #6) > - Add workaround for i386 to deal with kernel bug 178881 [1] > (PR/178881: getifaddrs(3) does not report IPv6 addresses properly in 32-bit > compatibility mode) This workaround disables getifaddr(3) functionality completely in npm. Actually it should work fine for i386 userland running on i386 kernels; only i386 userland on the amd64 kernel is affected. (i386 on i386 should not have this patch, only i386 on amd64 - that's why it's just a workaround, not a general solution). |