netdev - Re: [Possible regression?] ip route deletion behavior change

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 31 Oct 2019 14:41:49 -0600
From:   David Ahern <dsahern@...il.com>
To:     Hendrik Donner <hd@...cillation.de>,
        "David S. Miller" <davem@...emloft.net>, netdev@...r.kernel.org
Subject: Re: [Possible regression?] ip route deletion behavior change

On 10/31/19 9:44 AM, Hendrik Donner wrote:
> Hello,
> 
> analyzing a network issue on our embedded system product i found a change in behavior 
> regarding the removal of routing table entries when an IP address is removed.
> 
> On older kernel releases before commit 5a56a0b3a45dd0cc5b2f7bec6afd053a474ed9f5
> (simplified example):
> 
> Routing table:
> 
> # ip r
> default via 10.0.2.2 dev enp0s3 proto dhcp src 10.0.2.15 metric 1024
> 10.0.2.0/24 dev enp0s3 proto kernel scope link src 10.0.2.15
> 10.0.2.2 dev enp0s3 proto dhcp scope link src 10.0.2.15 metric 1024
> 10.20.0.0/14 via 10.0.2.2 dev enp0s3 src 10.20.40.100
> 
> The last route was manually added with ip r add.
> 
> Removing the IP 10.20.40.100 from enp0s3 also removes the last route:
> 
> # ip r
> default via 10.0.2.2 dev enp0s3 proto dhcp src 10.0.2.15 metric 1024
> 10.0.2.0/24 dev enp0s3 proto kernel scope link src 10.0.2.15
> 10.0.2.2 dev enp0s3 proto dhcp scope link src 10.0.2.15 metric 1024
> 
> After the mentioned commit - so since v4.10 - the route will no longer be removed. At 
> least for my team that's a surprising change in behavior because our system relies on
> the old behavior.
> 
> Reverting the commit restores the old behavior.
> 
> I'm aware that our use case is a bit odd, but according to the commit message commit 
> 5a56a0b3a45dd0cc5b2f7bec6afd053a474ed9f5 was meant to fix VRF related behavior while
> having the described (maybe unintended?) user visible side effect for non-VRF usage.
> 

devices not associated with a VRF table are implicitly tied to the
default == main table.

Can you test this change:

diff --git a/net/ipv4/fib_semantics.c b/net/ipv4/fib_semantics.c
index 0913a090b2bf..f1888c683426 100644
--- a/net/ipv4/fib_semantics.c
+++ b/net/ipv4/fib_semantics.c
@@ -1814,8 +1814,8 @@ int fib_sync_down_addr(struct net_device *dev,
__be32 local)
        int ret = 0;
        unsigned int hash = fib_laddr_hashfn(local);
        struct hlist_head *head = &fib_info_laddrhash[hash];
+       int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
        struct net *net = dev_net(dev);
-       int tb_id = l3mdev_fib_table(dev);
        struct fib_info *fi;

        if (!fib_info_laddrhash || local == 0)

[ As DaveM noted, you should cc maintainers and author(s) of suspected
regression patches ]