Repository navigation
A NIC goes to the net zone with its own address, Wake-on-LAN off and an MTU of 1500 - #306
Merged
Merged
Conversation
The net zone holds CAP_NET_ADMIN in its namespace and can give its NIC any name the kernel takes: a leading dash, a quote, an escape, bytes that are not UTF-8. When the zone ends the NIC comes back to zone 0 under that name, and the next start handed it on as it was. A name that is not UTF-8 failed every later start, since kryptikd looked up the lossy copy, and kept the machine offline until a reboot. Any other reached the fresh zone's script: a quote ended the name in the nft uplink set and could add kryptik0 to it, and a leading dash reached dhcpcd as an option. kryptikd now reads each physical interface's name as the kernel holds it, with its index, and renames one that is not plain, a lower-case letter then lower-case letters and digits, to the kernel's nic%d before it moves. netzone-init.sh names the device to ip with dev, so a plain name that is also an ip keyword is still read as a name.
…tart making kryptik0
… zone left on it The net zone can add netdevs to its radio, an AP, mesh or monitor one, or delete the one it has, and cfg80211 returns the wiphy to zone 0 with whatever is on it. A second netdev failed every later start: the first carry moved the wiphy with both, and the second found no such interface. A radio with none was never carried again, so Wi-Fi was gone until a reboot. Before it lists what to carry, kryptikd now gives a radio with no netdev a station named nic<N> (nl80211 NEW_INTERFACE), deletes all but the first netdev of a radio with several (DEL_INTERFACE), and lists each radio once. The zones suite leaves the net zone's radio with a monitor netdev beside the station, then with none, restarts the zone each time, and sees the next one get the radio with one station and join the access point again.
… since which to keep was the last zone's choice
…an MTU of 1500 What the last net zone set on its NIC outlived it: the address it chose, which would follow the machine from network to network after the restart meant to evict that zone; Wake-on-LAN, which lets anyone on the network wake the machine; and the MTU. Before a NIC is carried, kryptikd now sets its address back to the one the hardware came with (ETHTOOL_GPERMADDR), its MTU to 1500, and Wake-on-LAN off (ETHTOOL_SWOL), for the NICs "*" finds and for a NIC named in the zone file. The zones suite's rename round also gives the NIC another address and an MTU of 1400 inside the net zone, sees both come back to zone 0, and sees the next net zone get the NIC with its own address and 1500.
An altname sits in the same name table as names and moves with the NIC, so a net zone that gave its NIC the altname kryptik0 stopped every later start making the bridge, as #304's plain names otherwise prevent. The alias is free text that ip link prints. reset_nic now reads both with RTM_GETLINK, removes the altnames with RTM_DELLINKPROP and drops the alias with the address and MTU. mac_of was test-only, which the release build refused.
…e reset transact_on read into one 8 KiB buffer and refused a longer reply, and a NIC may hold altnames until their list nears 64 KiB. A net zone that gave its NIC seventy long altnames beside kryptik0 made the listing fail, so kryptik0 stayed and the next start could not make the bridge. Each read is now sized to its message (MSG_PEEK | MSG_TRUNC) under a 128 KiB cap, del_altnames removes them 64 to a request, as the list's length is 16 bits, and a listing that fails all the same still removes kryptik0 by name.
…cannot find by name through zone 0's sysfs
…ik0 as an altname, as only then can a zone
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Builds on #305 (radio-one-station), which builds on #304; merge those first. This PR's own change is its last three commits, ab1e852, 9166a00 and dd0ce8b.
What was wrong. linux-distro-fe found that what the last net zone set on its NIC outlived it, until a reboot:
kryptik0stopped every later start from making the bridge, which A NIC the net zone renamed is renamed nic<N> before the next net zone takes it #304's plain names otherwise prevent. The alias is free text thatip linkprints.Plain names (A NIC the net zone renamed is renamed nic<N> before the next net zone takes it #304) and a radio's shape (A radio leaves zone 0 with one station, whatever netdevs the last net zone left on it #305) reset none of these. linux-distro-7d asked for the reset before 1.0.
What changed.
compartments/kryptikd/src/netzone.rs:reset_nicruns in zone 0 before a NIC is carried, both for every NICnic = "*"finds and for a NIC named in the zone file. It sets the address back to the one the hardware came with when they differ, sets the MTU of an Ethernet-type NIC to 1500 when it differs, removes any altnames, drops the alias, and turns Wake-on-LAN off when it is on. Each change and each failure is logged; a failure does not stop the start.compartments/kryptikd/src/netlink.rs:perm_mac_ofusesETHTOOL_GPERMADDR, which readsdev->perm_addr, the addressregister_netdevicekeeps for a NIC whose driver gave one.wol_offusesETHTOOL_GWOL, thenETHTOOL_SWOLwithwolopts0. A driver with no Wake-on-LAN answers EOPNOTSUPP, which is not an error here.set_linksendsRTM_NEWLINKwithIFLA_ADDRESS,IFLA_MTUand an emptyIFLA_IFALIAS, by index.names_leftreads the altnames (IFLA_PROP_LIST, nestedIFLA_ALT_IFNAME) and whether an alias is set, with oneRTM_GETLINK;del_altnamesremoves the altnames withRTM_DELLINKPROP.mac_ofwas test-only, which the release build refused at ab1e852; it is built always now.transact_onsizes each read to its message (MSG_PEEK | MSG_TRUNC) under a 128 KiB cap instead of one 8 KiB buffer: a NIC may hold altnames until their list nears 64 KiB, and a listing that failed would have leftkryptik0in place (fe's read).del_altnamesremoves them 64 to a request, since a property list's length is 16 bits, and if the listing fails all the same,reset_nicstill removeskryptik0by name.build/guest-tests/zones-check.sh: A NIC the net zone renamed is renamed nic<N> before the next net zone takes it #304's rename round now also gives the NIC the address 02:00:5e:00:53:01 and an MTU of 1400 inside the net zone, seventy altnames of a hundred bytes and thenkryptik0, and an alias, before the rename. The new checkuplink-state-resetsees the address and MTU come back to zone 0 under the hostile name, then requires the next net zone'snic<N>to read its own address (as recorded inside the zone first) and 1500, with no altname or alias left. On ab1e852 the altname alone keeps that start from comingREADY.tools/image/zones-test.sh: lists the verdict.docs/design/net-zone.md: what is set back and why, and what is not.What is not covered. EEE, offloads, rings and coalescing are not set back; they stay with the ethtool-netlink limit on the owner's list. Nor are a radio's wake triggers (nl80211 WoWLAN), which act only in suspend. The VM's virtio-net has no Wake-on-LAN, so no suite shows it being turned off; the check says so in an INFO line. Testing on physical hardware is planned for October 2026.
How the run proves it. CI compiles the new netlink calls and runs the unit suites. In the Distro run's zones suite,
uplink-state-resetfails on #305's code: the next zone'snic0keeps 02:00:5e:00:53:01 and MTU 1400. With this change it reads 52:54:00:12:34:56 and 1500.