On 11/25/24 12:15 PM, Daniel P. Berrangé wrote:
On Mon, Nov 25, 2024 at 11:56:31AM -0500, Laine Stump wrote:
On 11/25/24 5:44 AM, Daniel P. Berrangé wrote:
On Fri, Nov 22, 2024 at 04:16:38PM -0500, Laine Stump wrote:
If the layer of a FirewallCmd is "raw", then the first arg is the name
of an arbitrary binary to exec, and the rest are the arguments to that
binary.
raw layer doesn't support auto-rollback command creation (any rollback
needs to be added manually with virFirewallAddRollbackCmd()), and also
raw layer isn't supported by the iptables backend (it would have been
straightforward to add, but the iptables backend doesn't need it, and
I didn't want to take the chance of causing a regression in that
code for no good reason).
I guess the obvious question to ask is why you chose to define
a "raw" layer, as opposed to defining a "tc" layer ? Being
more targetted about the anticipated usage feels better IMHO.
I thought about that, but while layer is used to figure out the binary name
for the iptables backend (because the different layers use ebtables,
iptables, or ip6tables), in the case of the nftables backend all of the
different layers use "nft" as the binary, and the layer indicates changes in
a few of the arguments to that command (and really both your suggestion and
mine are technically messed up, since the layer in the case of this
checksum-fix filter should really be "ipv4").
Maybe we just shouldn't be pretending this is a firewall command at
all ?
Even with iptables, this really isn't anything to do with traffic
filtering.
Well, if you're going to be pedantic and say that the only things that
are a part of the "firewall" are those bits that control whether or not
the traffic passes, and *not* the bits that modify packets on their way
through, then none of the rules that setup NAT should be a part of the
firewall either.
iptables just happened to be a convenient place to put
the logic in the kernel at the time.
> > 'tc' is the new "convenient" place to put the logic today. How about
putting a virNetDevFixDHCPChecksum() in virnetdev.h/c ?
and just invoking this API after we've setup nftables rules ?
That's kind of where I started out with this (having the tc command run
not even as a part of networkAddFirewallRules(), but at the same level),
but this required adding a call to the "FixChecksum()' function at each
and every place we called networkAddFirewallRules() (along with extra
bits to recover from errors). Then I moved the call to FixChecksum()
inside of networkAddFirewallRules, and that was obviously better, but
made it more obvious that this was just one more external command that
needed to be executed, just like all the nft commands, so handling it as
a special case (rather than as just one more in the list of commands to
run) complicated the code for no good reason.
As to whether or not the tc command to fix the dhcp checksum belongs in
the virFirewall object - a virFirewall is just a list of commands to be
executed to setup (or tear down) packet filtering and packet
modification for an interface. One type of packet modification is to
implement NAT, and another type of modification is to fix incorrect
checksums. Arguably the 2nd item shouldn't need to even exist, but here
we are :-P. And if modifying packets for NAT can be considered a part of
the firewall, then IMO modifying packet checksums is just as "firewally"
(newly invented word of the day - I adjectived a noun!).
Also I like having the tc command be a part of the virFirewall object
because that means it automatically takes advantage of the
"pseudo-atomic" nature of virFirewall, as well as its rollback
functionality:
1) error cleanup is simpler - this is just one more rollback command in
the virFirewall that gets execute if there is a failure at any point
rather than extra code around a one-off virCommand execution that is
really just duplicating the code that's already in virFirewallApply(), and
2) the command to remove the tc filter becomes a part of the fwRemoval
object saved in the network status, making it much simpler to reliably
remove the filter when virtnetworkd is restarted (and we
remove/reinstall all the firewall rules) or the network is shutdown.
Otherwise we could lose track of whether or not the tc filter was in
place or exactly what the filter was and end up doing the wrong thing,
e.g. if the backend setting changed from iptables to nftables (or vice
versa) during a virtnetworkd restart, or if the tc command to add the
filter changed from one release to the next leaving us attempting to
remove the old filter with a command that wouldn't actually remove it,
and/or attempting (and failing) to add a tc filter when one is already
in place. This is actually something that happened when I tried adding
the checksum filter by itself with no direct connection to the
virFirewall, and part of what led me to make it another virFirewallCmd
on the virFirewall's list.
