Networking

Makar currently has a monolithic lwIP bring-up path over the active kernel netdev device. In the default QEMU test setup this is virtio-net connected to QEMU user networking (slirp).

The interface is exposed to userspace as eth0.

Virtual NICs

The kernel registers several PCI Ethernet backends and binds whichever QEMU device is present:

Driver	nettest selector	PCI ID
virtio-net legacy	./run.sh nettest virtio	1AF4:1000
RTL8139	./run.sh nettest rtl8139	10EC:8139
Intel E1000/E1000E family baseline	./run.sh nettest e1000	8086:100E, 8086:100F
AMD PCNet FAST	./run.sh nettest pcnet	1022:2000

Each backend is a small polled driver feeding the same netdev contract: send, rx_poll, and mac. Interrupt-driven RX/TX can come later without changing lwIP’s current attachment point.

For an interactive session, pass the same NIC name to the boot targets to attach that device with slirp user networking:

./run.sh iso boot [virtio|rtl8139|e1000|pcnet]    # default virtio
./run.sh hdd boot [virtio|rtl8139|e1000|pcnet]

Networking test section

The networking suites (netdev ARP, lwip_tcp, lwip_net_info) live in their own section, ktest_run_net(), dispatched by the test=nettest selector — they are not part of the default ./run.sh ktest gate, which stays NIC-agnostic. ./run.sh nettest <nic> builds the test ISO with test_mode test=nettest, attaches the chosen QEMU NIC, and asserts on the serial marker:

KTEST_NET_RESULT: PASS

The NIC is a positional argument (no environment variable needed); it defaults to virtio when omitted. The same section validates every backend by re-running it against a different NIC. RTL8139, E1000, and PCNet have each passed it against QEMU slirp with ARP and lwIP TCP coverage.

The section also covers ICMP echo (ping to the live gateway, asserted; to 1.1.1.1, informational), DNS resolution (a dotted-quad literal asserted, a real name informational), and wget (an HTTP GET against a slirp guestfwd HTTP fixture). Peer addresses are derived from the live netif, not hardcoded.

DNS, ping, and HTTP (wget / unzip)

The lwIP stack exposes a small client surface:

• DNS — net_lwip_resolve(host, ip, timeout) resolves a hostname (or dotted-quad) via the lwIP resolver and the slirp-provided DNS server.
• ICMP — the kernel can send echo requests over the raw API (used by the net test section).

• wget — fetch a plain-HTTP URL and save it to the VFS. Available both as an in-kernel shell command and as wget.elf (via the SYS_WGET syscall, number 255):

  wget http://host[:port]/path [outfile]      # outfile defaults to /tmp/<basename>
  

  Plain HTTP only — there is no TLS, so https:// URLs are rejected. To pull from an HTTPS origin (e.g. GitHub), terminate TLS on the host and serve the file over plain HTTP; the guest reaches a host server at the slirp gateway (10.0.2.2) over TCP NAT, which streams multi-MB transfers fine (unlike a QEMU guestfwd -cmd, which caps response bodies near 160 KiB).

• unzip — unzip <archive.zip> [destdir] extracts a ZIP archive onto the VFS. A from-scratch RFC-1951 inflate (fs/inflate.c) handles stored and deflate entries; fs/unzip.c walks the central directory, creates parent directories, and rejects unsafe (../absolute) names. Extract to a writable backend (/tmp, or a mounted ext2/FAT32 disk for persistence).

Example end-to-end (download + verify + extract), with a file served over plain HTTP from a host/LAN server:

wget http://10.0.0.111/main.zip /tmp/main.zip
cat /tmp/main.zip | md5            # compare against the published md5sum
unzip /tmp/main.zip /tmp/src
ls /tmp/src

maknetcfg

maknetcfg.elf is the current network configuration tool:

exec /apps/maknetcfg.elf
exec /apps/maknetcfg.elf release
exec /apps/maknetcfg.elf renew
exec /apps/maknetcfg.elf flush-dns
exec /apps/maknetcfg.elf help        # also -h / --help

maknetcfg help (or -h/--help) prints the command summary and exits 0; an unrecognised argument prints the same summary and exits 1.

Without arguments it prints an ipconfig-style summary:

• active Ethernet driver
• link state
• MAC address
• DHCP state
• IPv4 address
• subnet mask
• default gateway
• DNS server

release stops the lwIP DHCP client and deconfigures eth0: the IPv4 address, gateway, and DNS server are all cleared, so maknetcfg then reports DHCP state released with no assigned address (0.0.0.0) until a renew — matching the normal DHCP-release model.

renew clears the released state and restarts DHCP with a fresh DISCOVER, waiting briefly for a lease. If none arrives, Makar drops eth0 onto its static QEMU slirp fallback so the interface is usable again in the default virtualized setup:

IPv4: 10.0.2.15/24
Gateway: 10.0.2.2
DNS: 10.0.2.3

flush-dns clears lwIP’s DNS cache and pending resolver requests while keeping the configured DNS server.

Static configuration (Settings → Network)

By default eth0 is DHCP-driven. The Settings app (mxsettings.elf, the Network panel) can also assign a static IPv4 address via SYS_NET_CONFIG (net_cfg_t in makar_abi.h) → net_lwip_config():

• Static (dhcp = 0): under the net big-lock, stop the DHCP client (dhcp_release_and_stop + note DHCP off), then netif_set_addr() with the caller’s IP/netmask/gateway and dns_setserver() with the caller’s DNS — the same primitives as the slirp static fallback above, but with user-supplied values. Takes effect immediately on the live netif.
• DHCP (dhcp != 0): restart the DHCP client (equivalent to renew).

net_lwip_config() returns 0 on success, -1 if the stack isn’t ready. The panel reads the current state back through SYS_NET_INFO, so an applied static address shows up right away. There’s no persistence across reboots yet (the config is applied to the running interface only).

Current Limits

Networking is still early:

• there is no POSIX socket API yet
• ping, wget-lite, ftp, and telnet are not implemented yet
• only one active netdev is selected
• DHCP is best-effort with deterministic static fallback for QEMU slirp