[-- Attachment #1: Type: text/plain, Size: 1054 bytes --] connect() to the IPv6 loopback address can fail with EACCES on Linux if IPv6 is disabled on the lo interface, and causes getaddrinfo to fail without returning IPv4 addresses. It should be treated as if IPv6 is disabled. echo 1 >/proc/sys/net/ipv6/conf/lo/disable_ipv6 struct addrinfo hints, *res = NULL; hints.ai_family = PF_UNSPEC; hints.ai_flags = AI_ADDRCONFIG; getaddrinfo("192.168.1.1", "80", &hints, &res); strace: ======start======= socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, IPPROTO_UDP) = 14 connect(14, {sa_family=AF_INET, sin_port=htons(65535), sin_addr=inet_addr("127.0.0.1")}, 16) = 0 close(14) = 0 socket(AF_INET6, SOCK_DGRAM|SOCK_CLOEXEC, IPPROTO_UDP) = 14 connect(14, {sa_family=AF_INET6, sin6_port=htons(65535), inet_pton(AF_INET6, "::1", &sin6_addr), sin6_flowinfo=htonl(0), sin6_scope_id=0}, 28) = -1 EACCES (Permission denied) close(14) = 0 writev(2, [{iov_base="[warn] getaddrinfo: Permission denied\n", iov_len=38}, {iov_base=NULL, iov_len=0}], 2) = 38 ======end========= [-- Attachment #2: musl-getaddrinfo-ipv6-eacces.patch --] [-- Type: text/x-patch, Size: 304 bytes --] --- musl-1.1.24/src/network/getaddrinfo.c 2019-10-13 14:58:27.000000000 -0700 +++ musl-1.1.24/src/network/getaddrinfo.c 2019-12-04 14:52:11.003784091 -0800 @@ -76,6 +76,7 @@ case EHOSTUNREACH: case ENETDOWN: case ENETUNREACH: + case EACCES: break; default: return EAI_SYSTEM;
On Wed, Dec 04, 2019 at 06:44:29PM -0800, Bob Richmond wrote: > connect() to the IPv6 loopback address can fail with EACCES on Linux > if IPv6 is disabled on the lo interface, and causes getaddrinfo to > fail without returning IPv4 addresses. It should be treated as if > IPv6 is disabled. > > echo 1 >/proc/sys/net/ipv6/conf/lo/disable_ipv6 > > struct addrinfo hints, *res = NULL; > hints.ai_family = PF_UNSPEC; > hints.ai_flags = AI_ADDRCONFIG; > getaddrinfo("192.168.1.1", "80", &hints, &res); > > strace: > ======start======= > socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, IPPROTO_UDP) = 14 > connect(14, {sa_family=AF_INET, sin_port=htons(65535), > sin_addr=inet_addr("127.0.0.1")}, 16) = 0 > close(14) = 0 > > socket(AF_INET6, SOCK_DGRAM|SOCK_CLOEXEC, IPPROTO_UDP) = 14 > > connect(14, {sa_family=AF_INET6, sin6_port=htons(65535), > inet_pton(AF_INET6, "::1", &sin6_addr), sin6_flowinfo=htonl(0), > sin6_scope_id=0}, 28) = -1 EACCES (Permission denied) > close(14) = 0 > writev(2, [{iov_base="[warn] getaddrinfo: Permission denied\n", > iov_len=38}, {iov_base=NULL, iov_len=0}], 2) = 38 > ======end========= > --- musl-1.1.24/src/network/getaddrinfo.c 2019-10-13 14:58:27.000000000 -0700 > +++ musl-1.1.24/src/network/getaddrinfo.c 2019-12-04 14:52:11.003784091 -0800 > @@ -76,6 +76,7 @@ > case EHOSTUNREACH: > case ENETDOWN: > case ENETUNREACH: > + case EACCES: > break; > default: > return EAI_SYSTEM; This patch was overlooked at the time, and another user just stopped by #musl to ask why it wasn't applied. I'm going to go ahead and apply it now. Sorry for the long delay! Rich
On Thu, Apr 29, 2021 at 08:13:03PM -0400, Rich Felker wrote:
> On Wed, Dec 04, 2019 at 06:44:29PM -0800, Bob Richmond wrote:
> > connect() to the IPv6 loopback address can fail with EACCES on Linux
> > if IPv6 is disabled on the lo interface, and causes getaddrinfo to
> > fail without returning IPv4 addresses. It should be treated as if
> > IPv6 is disabled.
> >
> > echo 1 >/proc/sys/net/ipv6/conf/lo/disable_ipv6
> >
> > struct addrinfo hints, *res = NULL;
> > hints.ai_family = PF_UNSPEC;
> > hints.ai_flags = AI_ADDRCONFIG;
> > getaddrinfo("192.168.1.1", "80", &hints, &res);
> >
> > strace:
> > ======start=======
> > socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, IPPROTO_UDP) = 14
> > connect(14, {sa_family=AF_INET, sin_port=htons(65535),
> > sin_addr=inet_addr("127.0.0.1")}, 16) = 0
> > close(14) = 0
> >
> > socket(AF_INET6, SOCK_DGRAM|SOCK_CLOEXEC, IPPROTO_UDP) = 14
> >
> > connect(14, {sa_family=AF_INET6, sin6_port=htons(65535),
> > inet_pton(AF_INET6, "::1", &sin6_addr), sin6_flowinfo=htonl(0),
> > sin6_scope_id=0}, 28) = -1 EACCES (Permission denied)
> > close(14) = 0
> > writev(2, [{iov_base="[warn] getaddrinfo: Permission denied\n",
> > iov_len=38}, {iov_base=NULL, iov_len=0}], 2) = 38
> > ======end=========
>
> > --- musl-1.1.24/src/network/getaddrinfo.c 2019-10-13 14:58:27.000000000 -0700
> > +++ musl-1.1.24/src/network/getaddrinfo.c 2019-12-04 14:52:11.003784091 -0800
> > @@ -76,6 +76,7 @@
> > case EHOSTUNREACH:
> > case ENETDOWN:
> > case ENETUNREACH:
> > + case EACCES:
> > break;
> > default:
> > return EAI_SYSTEM;
>
> This patch was overlooked at the time, and another user just stopped
> by #musl to ask why it wasn't applied. I'm going to go ahead and apply
> it now. Sorry for the long delay!
It's been raised that this is NOT a result of
echo 1 >/proc/sys/net/ipv6/conf/lo/disable_ipv6
but rather appears to be fib6 policy setup by OpenWRT for some reason,
whereby the kernel (net/ipv6/fib6_rules.c: fib6_rule_action)
synthesizes error codes for routing policy reasons. This is probably
wrong for the kernel to do -- especially their re-appropriation of
EINVAL for FR_ACT_BLACKHOLE when POSIX already specifies it for
"The address_len argument is not a valid length for the address
family; or invalid address family in the sockaddr structure."
So in light of this mess, the patch may be correct, despite the
problem being misattributed, but it should probably also handle the
EINVAL case. Also it's not 100% clear whether we should interpret this
as "no IPv6" or ignore it as an access control policy rather than
reflection of IPv6 existing. If there are any other ways the kernel
can return EACCES or EINVAL here, we would not want to misinterpret
that in a way that breaks IPv6.
Someone should probably also ping OpenWRT about why they're using this
arcane mechanism to block IPv6 to localhost.
Rich
On Fri, Apr 30, 2021 at 08:38:04AM -0400, Rich Felker wrote: > Someone should probably also ping OpenWRT about why they're using this > arcane mechanism to block IPv6 to localhost. at least I can see: https://git.openwrt.org/?p=openwrt/svn-archive/archive.git;a=blob;f=target/linux/generic/patches-3.19/670-ipv6-allow-rejecting-with-source-address-failed-policy.patch;h=f32458df30ad466d4e3ac8224cbec1bd074b43ec;hb=35d90ba52069c96afd1a74600b91499e5feed0e0 I was last refreshed on Tue Mar 30 22:01:27 2021 +0100 and says: "RFC6204 L-14 requires rejecting traffic from invalid addresses with ICMPv6 Destination Unreachable, Code 5 (Source address failed ingress/ egress policy) on the LAN side, so add an appropriate rule for that." But that is just guessing... bye, Bastian
On Fri, Apr 30, 2021 at 04:40:33PM +0000, Bastian Bittorf wrote:
> On Fri, Apr 30, 2021 at 08:38:04AM -0400, Rich Felker wrote:
> > Someone should probably also ping OpenWRT about why they're using this
> > arcane mechanism to block IPv6 to localhost.
>
> at least I can see:
> https://git.openwrt.org/?p=openwrt/svn-archive/archive.git;a=blob;f=target/linux/generic/patches-3.19/670-ipv6-allow-rejecting-with-source-address-failed-policy.patch;h=f32458df30ad466d4e3ac8224cbec1bd074b43ec;hb=35d90ba52069c96afd1a74600b91499e5feed0e0
>
> I was last refreshed on Tue Mar 30 22:01:27 2021 +0100
> and says:
>
> "RFC6204 L-14 requires rejecting traffic from invalid addresses with
> ICMPv6 Destination Unreachable, Code 5 (Source address failed ingress/
> egress policy) on the LAN side, so add an appropriate rule for that."
>
> But that is just guessing...
Presumably that would be for traffic originating from another host
OpenWRT is forwarding for (blocking it from spoofing ::1 on the wire?)
not for traffic originating on the OpenWRT box itself (where ::1
should work). I'm worried that if we just "fix" this issue on the musl
side, OpenWRT is just wrongly going to conclude there's no IPv6
coonnectivity rather than that they have a configuration error
breaking it..
Rich
On Fri, Apr 30, 2021 at 8:38 AM Rich Felker <dalias@libc.org> wrote:
> ....
> It's been raised that this is NOT a result of
>
> echo 1 >/proc/sys/net/ipv6/conf/lo/disable_ipv6
>
> but rather appears to be fib6 policy setup by OpenWRT for some reason,
> whereby the kernel (net/ipv6/fib6_rules.c: fib6_rule_action)
> synthesizes error codes for routing policy reasons. This is probably
> wrong for the kernel to do -- especially their re-appropriation of
> EINVAL for FR_ACT_BLACKHOLE when POSIX already specifies it for
>
> "The address_len argument is not a valid length for the address
> family; or invalid address family in the sockaddr structure."
>
> So in light of this mess, the patch may be correct, despite the
> problem being misattributed, but it should probably also handle the
> EINVAL case. Also it's not 100% clear whether we should interpret this
> as "no IPv6" or ignore it as an access control policy rather than
> reflection of IPv6 existing. If there are any other ways the kernel
> can return EACCES or EINVAL here, we would not want to misinterpret
> that in a way that breaks IPv6.
>
> Someone should probably also ping OpenWRT about why they're using this
> arcane mechanism to block IPv6 to localhost.
The kernel has been doing that stupid thing for ages. They have no
interest in fixing it. (I brought it up on one of the kernel mailing
lists).
It gets worse with components like SELinux. They hijack error codes
there, too. You can waste hours trying to track down an EACCESS on a
web server only to find out the kernel hijacked the return code in
SELinux.
God forbid they actually provide a selinux_errno to check for SELinux errors...
Jeff
Rich Felker <dalias@libc.org> writes: > So in light of this mess, the patch may be correct, despite the > problem being misattributed, but it should probably also handle the > EINVAL case. Also it's not 100% clear whether we should interpret this > as "no IPv6" or ignore it as an access control policy rather than > reflection of IPv6 existing. If there are any other ways the kernel > can return EACCES or EINVAL here, we would not want to misinterpret > that in a way that breaks IPv6. I've sent an updated patch to the list. I did submit a change to OpenWrt to not create these rules on devices where IPv6 is disabled: https://patchwork.ozlabs.org/project/openwrt/patch/20210430143037.6763-1-julian@cipht.net/ but it seems to me that the patch to musl still makes sense, in as much as these return codes from Linux make any sense at all. <#secure method=pgpmime mode=sign> -- Julian Squires
On Fri, Apr 30, 2021 at 12:59:39PM -0400, Jeffrey Walton wrote:
> God forbid they actually provide a selinux_errno to check for SELinux errors...
>
> Jeff
Well, that would be difficult. Although the concept of "nicer" errors
has been floated in the past, and having some kind of parametrization
for errno would be helpful (e.g. if ENOENT is returned, actually saying
which file could not be found would be helpful. Because it's not always
obvious). But right now, errno is the only error handling mechanism
established in the ABI, and it is transported by having the system call
return a value between -1 and -4096 (though I'm not sure if that lower
bound is general or just AMD64). Having a second errno would require
either establishing a new system call to read it out, or modifying the
ABI to allow for the information to be transported. There are many
hurdles in the way of the latter (can't use return value, can't use
registers, can only use memory on an opt-in basis, but then you can also
just add another system call directly), so it's going to be the former.
Then the question arrises whether the abstraction is even correct.
Technically, SELinux is just a plug-in security module, and a given
Linux kernel may have many of those. Shall each get their own errno?
Where does it end?
So yeah, it's not as simple as"just add another variable".
Ciao,
Markus
On Fri, Apr 30, 2021 at 08:49:17PM +0200, Markus Wichmann wrote:
> On Fri, Apr 30, 2021 at 12:59:39PM -0400, Jeffrey Walton wrote:
> > God forbid they actually provide a selinux_errno to check for SELinux errors...
> >
> > Jeff
>
> Well, that would be difficult. Although the concept of "nicer" errors
> has been floated in the past, and having some kind of parametrization
> for errno would be helpful (e.g. if ENOENT is returned, actually saying
> which file could not be found would be helpful. Because it's not always
> obvious). But right now, errno is the only error handling mechanism
> established in the ABI, and it is transported by having the system call
> return a value between -1 and -4096 (though I'm not sure if that lower
> bound is general or just AMD64). Having a second errno would require
> either establishing a new system call to read it out, or modifying the
> ABI to allow for the information to be transported. There are many
> hurdles in the way of the latter (can't use return value, can't use
> registers, can only use memory on an opt-in basis, but then you can also
> just add another system call directly), so it's going to be the former.
>
> Then the question arrises whether the abstraction is even correct.
> Technically, SELinux is just a plug-in security module, and a given
> Linux kernel may have many of those. Shall each get their own errno?
> Where does it end?
>
> So yeah, it's not as simple as"just add another variable".
Yes, the underlying issue is just SELinux, LSMs, firewall/routing
policy, seccomp, etc. producing wrong error codes (or letting the user
configure it to produce wrong error codes), rather than hard-coding
the cause. For example in the case here, nobody needs to know EACCES.
The important piece of information to the application is that the
connection is not routeable, and this is ENETUNREACH or EHOSTUNREACH.
Routing is always a matter of some sort of policy, so "this was
blocked by fib6 [vs iptables vs just not having a route, etc.]" is not
meaningful to the application.
If there were a secondary way to give more detailed error information,
it might be nice to give it, but in the absence of that, the right
solution is just "don't replace well-specified general error codes the
application knows to handle with overly specific ones it can't (and
that might even mean something else to it, in the case of EINVAL).
Rich