tap_user: DHCP fails after guest reboot

[LekKit]

The issue

• After rebooting a guest with an automatic DHCP configuration, networking isn't brought up immediately

Steps to reproduce

• Use Arch / Alpine image
• Enable DHCP via dhcpcd / systemd-networkd / NetworkManager
• Reboot
• Network is not immediately brought up, but comes back after a while

[LekKit]

Also affects Debian and OpenSUSE guests on initial boot, i.e. no working DHCP at all there.

[ahdiotllfer]

check out my repository https://github.com/ahdiotllfer/RVVM . it contains a fix for this problem and also a bunch of changes for my ios port

[ahdiotllfer]

or maybe not, at least this fixes the network problems I had with ios

[LekKit]

Oh pretty nice work actually. I do like the way you structured state handling code, I'd like to merge it as soon as you wish, it does not conflict with my working tree it seems.

Yeah, reinitializing TAP probably fixes most of the stuff. I'm gonna look into support for DHCP session resume later on, but this fix is still good as it drops any connections left from a previous boot. Maybe call it tap_reset() tho?

A few things come to my mind about device suspend/resume, which I plan on doing after snapshot support is in tree:

• Ideally make rvvm_state_read_u32() just return the read value (Or 0 on failure), then add some kind of rvvm_state_status() to return error status if any of the previous (de)serialization calls failes. This will likely simplify current device suspend code.
• Make some kind of named device section mechanism, i.e. rvvm_state_open_section("nvme_v0.7") instead of just the version number, then enumerate equally named sections in the same order when serializing/deserializing
• Pause device threads (or other background processing) in suspend() callback before serialization, resume in resume() callback. Allow passing a NULL state argument so no actual serialization happens, then callbacks return early, purely pausing/restarting device internals. This will be useful to stop all DMA transfers during guest RAM hotplug (in future), etc.
• The snapshot mechanism should probably dump the data to an internal rvvm_blk_dev_t handle (backed by a file, etc). I am working on a deduplication mechanism where multiple storage images merge their duplicated blocks (4kb by default), which will allow also merging multiple VM snapshots with themselves, and with the disk images too.

Also - is thread_create() identifier occupied on some platform? It's a good idea to rename then, but then what other identifiers are also already used?

[ahdiotllfer]

I don't know how to make a pull request since I pushed all the changes in one commit, and I'm afraid that among them there are changes that are useless for platforms other than iOS, and those that may break things that I haven't tested, so you'd better apply these changes selectively yourself

[ahdiotllfer]

I'm also working on adding a framebuffer to the ios port and just implemented snapshot saving for the framebuffer. these changes are not yet uploaded to github