Distributions patches

[xosevp]

Debian: https://salsa.debian.org/debian/net-tools/-/tree/debian/sid/debian/patches
( git clone https://salsa.debian.org/debian/net-tools.git net-tools-debian )
@NightTsarina @utkarsh2102 @carnil

• Add_missing_headers.patch
• Bug_549397-fix-decoding-of-MII-vendor-ids.pach
• Bug_900962-man-de-typos.patch
• man-fr_FR-route.8.patch
• translations.patch

SuSE: https://build.opensuse.org/package/show/openSUSE%3AFactory/net-tools
@lnussel @dirkmueller @stanislav-brabec

• 0001-Add-ether-wake-binary.patch (we recommend busybox version)
• 0002-Do-not-warn-about-interface-socket-not-binded.patch
• 0004-By-default-do-not-fopen-anything-in-netrom_gr.patch (historic patch)
• 0006-Allow-interface-stacking.patch
• 0007-Introduce-T-notrim-option-in-netstat.patch

PLD: https://git.pld-linux.org/?p=packages/net-tools.git;a=tree
( git clone git://git.pld-linux.org/packages/net-tools.git net-tools-pld )
@arekm

• net-tools-interface.patch
• net-tools-make_config_h.patch
• net-tools-man.patch
• net-tools-netstat-netlink-diag.patch
• net-tools-opt.patch
• net-tools-tr.patch (I am not sure this is correct)

Fedora: https://src.fedoraproject.org/rpms/net-tools/tree/rawhide
( git clone https://src.fedoraproject.org/rpms/net-tools.git net-tools-fedora )
@mruprich

• ether-wake.*
• iptunnel.*
• mii-diag.*
• *.patch ( for net-tools-2.0 )

Pending bugs on SourceForge: https://sf.net/p/net-tools/bugs/

[ecki]

thanks for the list. Will check it. did you also check fedora/redhat or has that ship sailed? :)

[xosevp]

thanks for the list. Will check it. did you also check fedora/redhat or has that ship sailed? :)

Added, but it is based on net-tools-2.0

[ecki]

Bug_549397-fix-decoding-of-MII-vendor-ids.patch
Bug_900962-man-de-typos.patch
man-fr_FR-route.8.patch

These are upstream commited now.

[stanislav-brabec]

The SUSE 0006-Allow-interface-stacking.patch was originally introduced in 2006 as net-tools-1.60-interface_stack.patch originates from Fedora:

Add patch from fedora to avoid stack smashing https://bugzilla.redhat.com/show_bug.cgi?id=142461

This patch was actually broken by 7a8f42f and needs porting. I am working on an updated version.

[stanislav-brabec]

0001-Allow-interface-stacking.patch

[ecki]

0001-Allow-interface-stacking.patch

Thanks that Looks good (I am not quite sure if the subject is helpful - not working since the fix, what’s the difference in behavior? How is 142461 related to this?

I noticed some utf8 Junk, maybe that should be asciified as well.

[stanislav-brabec]

Thanks that Looks good (I am not quite sure if the subject is helpful - not working since the fix, what’s the difference in behavior?

How is 142461 related to this?

My mistake. It should be a reference to SUSE Bugzilla. It says that the patch comes from Fedora, but does not refer any Fedora bug.

The SUSE changelog from 2006 mentions that it provides a fix:

Wed Jan 11 15:10:08 CET 2006 - mmj@suse.de

- Add patch from fedora to avoid stack smashing [#142461]

Its name was net-tools-1.60-interface_stack.patch, but later it was renamed to 0006-Allow-interface-stacking.patch.

The SUSE bug is not public, so I extract the bug description here:

net-tools-1.60-543.8: ifconfig can fail to list aliases due to get_name() induced stack corruption

The net-tools interface library routine if_readlist_proc() can overwrite its own stack space for large interface byte counts. This has been seen at customer sites as "ifconfig -a" not returning alias information under certain circumstances. What we found during debug is that the if_readconf() routine was not being called in if_readlist() even though if_readlist_proc() was not indicating any errors.

Code review uncovered that the get_name() routine called by if_readlist_proc() did not have any bounds checking, and would overwrite the stack causing if_readlist_proc() to return a bogus value for err if the combination of interface name and receive byte count exceeded the size allocated to the name buffer (16 bytes):

        while (isdigit(*p))
            *name++ = *p++;

The following /proc/net/dev interface line was sufficiently long enough to cause the local err variable of if_readlist_proc() to be overwritten:

ether2:1333731384 1014828...

The get_name() routine needs to be called with an additional size argument indicating the size of the passed in buffer, and the parsing loops modified to stop when the end of the buffer is reached.

I just ported the patch on top of recent security fix 7a8f42f.

Surprisingly, the current Fedora does not have this patch.

I noticed some utf8 Junk, maybe that should be asciified as well.

The non-ASCII character in the code comes from 7a8f42f:

/* Safe version — guarantees at most IFNAMSIZ‑1 bytes are copied

The dash character is an UNICODE character EM DASH.

[stanislav-brabec]

If you need more info about other SUSE patches, I can dig into SUSE archives. Changes since September 2006 are publicly available at https://build.opensuse.org/package/show/openSUSE:Factory/net-tools

Changes between October 1997 and September 2006 are still archived at SUSE.

[mruprich]

Surprisingly, the current Fedora does not have this patch.

Actually the patch is in Fedora but it might be slightly different. It is a part of a bigger patch that was added a long time ago - https://src.fedoraproject.org/rpms/net-tools/blob/rawhide/f/net-tools-cycle.patch . AFAICT this patch also has the solution to 7a8f42f that's why I closed the bug in Fedora.

Just a quick question about the purpose of this issue - so you are trying to bring patches scattered across different distributions and bring them back here to the Upstream of net-tools? So is this some kind of an attempt to revive net-tools? Sorry for asking like this but I've been maintaining net-tools in Fedora for 9-something years and I kind of had the impression that there is no Upstream to speak of...

Regards,
Michal Ruprich

[stanislav-brabec]

Just a quick question about the purpose of this issue - so you are trying to bring patches scattered across different distributions and bring them back here to the Upstream of net-tools? So is this some kind of an attempt to revive net-tools?

I did not open this issue. I just seen it and I like the idea.

Regarding SUSE (openSUSE Tumbleweed, openSUSE Leap 16, SUSE Linux Enterprise 16 and other recent products):

I became a maintainer of net-tools few months ago, as there was a need to fix CVE-2025-46836. net-tools are deprecated in favor of the more recent tools. It was even removed from the default installation. We do not plan to revive net-tools. But there are still many utilities that depend on it, and we do not want ti force people to migrate yet. So it will remain in the maintenance mode for many years. Getting all patches together and use an unpatched upstream code as a common code base across all distributions will save our time. Less bugs, no need for porting of fixes.

Sorry for asking like this but I've been maintaining net-tools in Fedora for 9-something years and I kind of had the impression that there is no Upstream to speak of...

I see that CVE-2025-46836 was fixed here, so this repository looks promising.

[ecki]

If you ask me, then net-tools remains deprecated and essentially dead (would require major rewrite to get rid of proc usage), but there are very occasional maintenance releases (in fact I still plan to release the main branch soon and it would be good if the known fixes get merged here)

[ecki]

Code review uncovered that the get_name() routine called by if_readlist_proc() did not have any bounds checking

Ah thanks for digging this up. Is this patch “only”
To make the code nicer or do you have proof the latest branch version has a regression or is unfixed?

The allocation in getline looks better but not sure it’s complete as the length seems to be not referenced as well. Need to run some tests, so that’s why I ask for the purpose of the patch - maybe a condition I can use as test case like the extra stacked name or such?