Skip to content

Releases

Gola Bootlace 0.5.2-rc.3

Pre-release

Choose a tag to compare

@ghazichaieb ghazichaieb released this 31 Oct 12:06

A pre-release for trying -offer-delay on the PCs that never accept the offers: update -v 0.5.2-rc.3 in the container installs it (a plain update only follows full releases). It also contains everything of 0.5.2-rc.2.

  • -offer-delay (environment BOOTLACE_OFFER_DELAY): an experiment for PCs that never accept the offers (#51). Two PCs with a Realtek network chip (an ASUS P8Z77-V LX and an HP laptop) keep repeating their DHCP discover and never send a request, although both the router and Bootlace answer within the same millisecond with well-formed offers (a wire capture of the HP is in #51); an Intel chip (the ExpertBook) takes the same two offers at once. With a value such as -offer-delay 300ms, proxy mode holds back its offer to a discover so that the router's offer arrives first. Requests are answered at once and other PCs are not kept waiting. Off by default (0). Not yet tried on those PCs.

  • The ExpertBook's refusal of the shim is explained, and the notes say how to check a PC. The ASUS ExpertBook B3405CCA still answers "Invalid signature" to the shim signed with both Microsoft CAs. Its key database was read (2026-10-31): it holds only "Microsoft Windows Production PCA 2011", neither "Microsoft Corporation UEFI CA 2011" nor "Microsoft UEFI CA 2023", so no shim can pass on it. The compatibility notes and the Secure Boot troubleshooting table now say so, with the PowerShell lines that read a PC's database. No program change.

Gola Bootlace 0.5.2-rc.2

Pre-release

Choose a tag to compare

@ghazichaieb ghazichaieb released this 30 Oct 21:39

A pre-release for trying the shim on the ASUS ExpertBook B3405CCA: update -v 0.5.2-rc.2 in the container installs it (a plain update only follows full releases). It also contains everything of 0.5.2-rc.1.

  • The built-in shim is signed with both Microsoft CAs (2011 and 2023). A PC whose firmware no longer trusts Microsoft's 2011 third-party CA cannot start a shim that only the 2011 CA signed, and the shim in the iPXE v2.0.0 archive was that build. The ASUS ExpertBook B3405CCA refused the route, probably for this reason (its key database was not read). The iPXE project replaced its ipxe-16.1 shim release on 2026-05-27 with a dual-signed one; Bootlace now carries that file (1,048,424 bytes, SHA-256 5eecca27…fa6cf), while ipxe.efi stays the v2.0.0 one. tools/update-secure-boot-assets.sh takes the shim from a shim release with --shim-release <tag> <sha256> and records the release, the address and the CAs found in its signature (signedWith) in MANIFEST.json; the Server page shows them; a test fails when the shim does not carry both certificates. Not yet tried on the ExpertBook.

Gola Bootlace 0.5.2-rc.1

Pre-release

Choose a tag to compare

@ghazichaieb ghazichaieb released this 30 Oct 14:29

A pre-release for trying the fix on a real PC: update -v 0.5.2-rc.1 in the container installs it (a plain update only follows full releases).

  • The drivers of the library are loaded in Windows PE at all. The command file only loaded them if exist "%SystemDrive%\$WinPEDriver$\nul", and that folder test is false on Windows PE's drive, so the step silently skipped every driver (Windows Setup itself picks up a $WinPEDriver$ folder, which hid it for disk drivers in the earlier tests). Found on a real PC: the Intel network card had no driver and the same for /r ... drvload loop typed by hand worked at once, with a DHCP address. The test is gone (the folder is always in the image when the drivers are on), and a test keeps it out of the command file.

Gola Bootlace 0.5.1

Choose a tag to compare

@ghazichaieb ghazichaieb released this 30 Oct 12:20
  • The drivers of the library are loaded again after wpeinit, and what drvload says is kept. On a real PC (ASUS ExpertBook, Intel I219-V card) loading the drivers only before wpeinit left the card without a driver, while the same load run by hand afterwards gave it a driver and a DHCP address. The command file now loads them before and after wpeinit (loading a driver that is already loaded is harmless), writes what drvload says to X:\drvload.log, and shows the refusals on the failure screen. The failure screen's list of devices without a driver uses find, because findstr is not part of Windows PE.
  • A Windows set-up that cannot reach the share names the devices that lack a driver. When the set-up environment has no driver for the PC's network card it cannot connect, and the screen only showed an adapter list with no adapter in it. It now also lists the devices Windows has no driver for (with their PCI\VEN_ and DEV_ numbers, the way to look a driver up) and says where a driver package goes (drivers\all\<name>\ in the library). Found on an ASUS ExpertBook whose Intel I219 network chip (VEN_8086&DEV_57A1) is not in Windows 11's set-up image.
  • The Server page no longer reads like an error about the own-key way while another way is in use. The note says that the own-key way is not set up yet, that this only matters if you choose "Own key", and what the folder needs; the sentence for a folder that has neither the shim nor the file it asks for now says "are missing" and "Copy them".
  • The container's update command lists the changes when the installed version is a pre-release. The list is built by comparing versions, and sort -V ranks 0.5.0-rc.1 above 0.5.0, so an installation that ran a release candidate was told nothing at all ("installed: 0.5.0-rc.1 available: 0.5.0" and straight to the question). The comparison is now a small function (vcmp) in which a pre-release is lower than its release and a test build such as 0.3.0-9-gabc is higher than the release it follows; the host script uses it too. Tests: go test ./tools/proxmox runs the script's own text with sh and awk. The fix reaches a container with its next update (the archive carries the command); to have it at once, run on the Proxmox host curl -fsSL https://raw.githubusercontent.com/ghazichaieb/gola-bootlace/main/tools/proxmox/bootlace-lxc.sh | bash -s -- --print-update >/tmp/update.new && pct push <number> /tmp/update.new /usr/local/bin/update --perms 0755.

Gola Bootlace 0.5.0

Choose a tag to compare

@ghazichaieb ghazichaieb released this 29 Oct 20:53
  • Configured boot entries survive a server restart without waiting for inspection (#70, cache follow-up). Bounded initramfs facts are cached locally and used to rebuild compatible starts on the first request. Changed files, program versions or server addresses invalidate the cache; corrupt caches fall back safely, and explicit rescan discards them. Thanks to @omar-akermi (PR #75).

  • Secure Boot works out of the box: the iPXE project's signed shim and iPXE are built into the program (#10). The two x86-64 files of the iPXE project's release v2.0.0 (x86_64-sb/shimx64.efi, served as ipxe-shimx64.efi, and x86_64-sb/ipxe.efi; the combination that started the menu, Ubuntu Server, Ubuntu 25.04 and a Windows 11 installation on a real MSI PC with Secure Boot on) are embedded unchanged, with internal/secureboot/assets/MANIFEST.json (source, release, checksum of the release archive, size and checksum of each file). The program checks its own bytes against the manifest at start and serves nothing, with an ERROR line in the log and a message in the web interface, when they do not match. TFTP serves exactly these two names under sb/ in this mode. The ARM64 files of the same release are not included (ARM64 PCs are not started yet, and are never sent the x86-64 shim).

  • Secure Boot way and scope are settings (#10). secureBootMode is builtin, own (a folder with a distribution's shim and an iPXE signed with your own key, as before) or off; secureBootScope is all (every 64-bit UEFI PC, what -secure-boot-all did) or marked (only the PCs marked on the PCs page). They are in the settings file and so in backups. A new installation starts with builtin and all; a settings file from an older version has neither, which stays off and marked, so updating changes no behaviour, and the log, the Dashboard and the Server page say Secure Boot support is built in: turn it on until a way is chosen. New options -secure-boot builtin|own|off and -secure-boot-scope all|marked (BOOTLACE_SECURE_BOOT, BOOTLACE_SECURE_BOOT_SCOPE); -secure-boot-dir still selects the own-key way and -secure-boot-all the scope all; options win over the saved settings and are shown as such. Without -secure-boot-dir, the own-key way uses the folder secure-boot next to the settings file.

  • The Server page chooses the way (#10). The Secure Boot block has a control for the way (Built in, Own key, Off) and one for the scope, a plain explanation of each way, the served files with size and checksum and the iPXE release they come from, what is missing for the own key, and a question before switching to the own key (every PC needs the key enrolled) or to every PC. A change counts from the next DHCP answer with no restart. API: GET /api/secureboot now has mode (builtin, own, off; it was "", pcs, all), scope, modeSource/scopeSource, modeLocked/scopeLocked, unset, builtin and own; new PUT /api/secureboot ({"mode": ..., "scope": ...}; 400 for an unknown value, 409 for what an option decides, for damaged built-in files and for the own key without its shim). The start-up log says which way is on and why.

  • tools/update-secure-boot-assets.sh <release> <sha256> renews the built-in files. It downloads the iPXE project's release archive, stops unless its SHA-256 is the one given (or --trust-printed-hash is used, which only prints and records the checksum), takes out the two files, rewrites the manifest and prints which checksums changed. docs/secure-boot.md says when to do it and why: PCs refuse builds that a newer iPXE or shim release revoked (SBAT, dbx).

  • Sources and licences of the built-in files. NOTICE and third_party/SOURCES (and so the third-party-sources archive of every release) carry the iPXE v2.0.0 source, the iPXE project's shim source (tag ipxe-16.1) and its gnu-efi, with the licence terms read from them; the exact commit the shim was built from is not recorded by the project, which NOTICE says. The shim in the file carries Microsoft's 2011 UEFI CA signature only.

  • The Dashboard no longer says that Secure Boot is not supported, and the PCs page's button text no longer says that a key must always be enrolled.

  • The Secure Boot route says what it is doing (#74). Found by testing 0.5.0-rc.1 on a real MSI PC with Secure Boot on, in proxy-DHCP mode: with -secure-boot-all and the iPXE project's signed shim and ipxe.efi the menu came up and Ubuntu Server and Ubuntu 25.04 booted, but a missing option or file only showed as Security Violation on the PC. The log now says at the start Secure Boot route: ON ... or OFF ... set BOOTLACE_SECURE_BOOT_DIR, see docs/secure-boot.md, and warns when the folder has the shim but not the file the shim asks for next (ipxe.efi for a shim whose name contains ipxe, grubx64.efi otherwise); the start is not refused for that, because the folder is read at every request and the file can be copied in while the server runs. The Server page has a Secure Boot route block (on or off, the folder, every expected file marked present or missing, all PCs or only marked ones, and a sentence about what is missing), with the same facts in GET /api/secureboot and in /api/status.

  • A PC that only asked for a boot file is labelled (#73). Its row on the PCs page says Asked for a boot file; Secure Boot may be refusing iPXE. with what to do (turn the route on, or press "Secure Boot route"), it has a History entry for each ask, and it can be forgotten. The page's self-refresh no longer stops while the search box or the page-size drop-down has the focus (it paused for anything inside the page, so a PC that appeared in the meantime stayed unlisted until the page was clicked elsewhere or reloaded). The server side was checked end to end with the real server, DHCP answer and admin handler and fake packets (discover from 0.0.0.0, request, architecture codes 7, 9 and 16, route on and off, restart with a saved list, full list): the PC is listed in all of them, so the missing row seen on the real PC was not reproduced; see the issue.

  • Secure Boot documentation. docs/secure-boot.md now starts with the iPXE project's signed shim as the recommended route, with copy-paste steps for a Proxmox container (download, checksum, folder, two lines in /etc/default/bootlace), per-PC button versus -secure-boot-all, the log lines of a working start, and a troubleshooting table; the real-PC results (menu, Ubuntu Server and Ubuntu 25.04 on an MSI PC, proxy DHCP) are in its results tables and in the compatibility page. Windows Setup through the route has not been tried on a real PC.

Gola Bootlace 0.5.0-rc.1

Pre-release

Choose a tag to compare

@ghazichaieb ghazichaieb released this 27 Oct 10:02

A pre-release: it exists so that the Secure Boot route can be tried on a real server and real PCs. Nothing changes unless the new option is set. To try the route without enrolling any key, put the iPXE project's signed ipxe-shimx64.efi and ipxe.efi (release file ipxeboot.tar.gz, folder x86_64-sb, the shim file shimx64.efi renamed) in a folder and start with -secure-boot-dir <folder> -secure-boot-shim ipxe-shimx64.efi; the steps are in docs/secure-boot.md ("Way 2").

  • Secure Boot for other Linux ISOs: a prototype, off by default (#10). New option -secure-boot-dir <folder> (BOOTLACE_SECURE_BOOT_DIR; -secure-boot-shim names the boot file, -secure-boot-all sends every 64-bit UEFI PC): the DHCP answer for a PC marked on the PCs page ("Secure Boot route") names a Microsoft-signed shim under sb/, TFTP serves the folder's files by name (the shim's second stage, MokManager), and the menu entries that start a Linux kernel get an iPXE shim line so that the ISO's own shim checks the kernel. tools/ipxe/build-sb.sh builds iPXE with SECUREBOOT=1 and its own SBAT line and signs it with a key given by path (nothing is signed or shipped by the project). Measured in a Hyper-V lab with Secure Boot on: the menu, and Ubuntu, Linux Mint, Debian and Fedora to their desktops or installer, openSUSE and Ubuntu Server to their kernels; an unsigned kernel (CachyOS) and Windows through wimboot do not start. The shim needs the signing certificate enrolled once per PC (8 screens), unless the iPXE project's own signed shim and binaries are used instead, which needed no enrolment in the lab. Full table, limits and risks in docs/secure-boot.md. Not tested on a real PC. tools/lab.ps1 and tools/pxe_vm.ps1 can start Secure Boot VMs (any template, an extra disk).

  • A correction in the Secure Boot page. wimboot is signed by Microsoft (UEFI CA 2011); only iPXE is unsigned.

  • Changing management pages while data loads is safe (#69). Late page responses and failures are discarded after navigation, and abandoned pages cannot restart polling. Background uploads continue normally. Thanks to @omar-akermi (#71).

Gola Bootlace 0.4.2

Choose a tag to compare

@ghazichaieb ghazichaieb released this 26 Oct 20:05
  • The container's update command renews itself. A container's update command was only rewritten when the host script ran again, so a container made before the "what is new" list existed (0.4.0 or earlier scripts) kept an old copy that asked "Update now?" without listing the changes. The Linux release archives now carry the current update command, and after a successful update it replaces the installed one (the archive has just passed its checksum; an archive without one, or one that does not parse, leaves the old copy). For a container that already has the old copy, run the host script once with CTID=<number> (see docs/proxmox.md); after that it stays current. When the changelog cannot be read, update now says so instead of printing nothing.

Gola Bootlace 0.4.1

Choose a tag to compare

@ghazichaieb ghazichaieb released this 26 Oct 16:28

Please update if you use boot passwords: with 0.4.0 an ISO protected by a boot password stopped with a kernel panic after the right password was typed (fixed below). Also: compressed disk images (.zst, .xz) with a hostile dictionary or window size are now refused before decoding.

  • PC cleanup tests no longer depend on clock precision (#34). Their fixtures are kept away from the age cutoff, avoiding the intermittent 19-versus-20 failure on Windows. Thanks to @omar-akermi (#66).
  • Zstandard image unpacking has a bounded window (#34). Frames requiring more than 128 MiB of history are refused before decoding, and extraction uses one decoder worker. Ordinary images still stream to disk; this is a window limit, not a limit on unpacked image size. Thanks to @omar-akermi (#67).
  • XZ image unpacking has the same limit as Zstandard (#34). The XZ library raises its own dictionary limit to whatever the file declares, so a small hostile .xz could ask for up to 4 GiB before the free-space check helps. The first block's declared dictionary is now read first and a file above 128 MiB is refused with a message that says how to repack it (follow-up to #67). Only the first block is checked; the documentation says so.
  • Manage the entries of an ISO's boot menu (#35). The ISOs page can hide and rename compatible entries, add arguments to one entry, and select which entry starts when the ISO is chosen (including automatic starts and PC rules). The usual start remains available. Choices survive restarts, backups and library renames; a missing choice falls back to the usual start. Thanks to @omar-akermi (#68).
  • Extra arguments go in front of the kernel's -- (#35). A boot-menu line of an ISO can contain the word -- (the Linux Mint entries do): the kernel stops reading its parameters there and hands the rest to init, so the arguments an ISO, an entry or a PC adds (and the debug words) were ignored by the kernel on such a line. They are now put in front of the --; a line without it, and one with --- (installers), is unchanged. Found by starting a UEFI VM with maxcpus=1 added to a Mint entry.
  • An ISO with a boot password starts again (#24). The menu fetched the password check's answer as an iPXE image and left it there; iPXE hands every image it holds to a Linux kernel as an initrd, so the kernel failed to unpack it and stopped with "Unable to mount root fs" (Linux Mint, started in a UEFI VM after the right password). The image is freed right after the check.

What was verified, and what was not

Seen working in real runs: the per-ISO boot entries in a real UEFI iPXE VM and in a real browser (rename, hide, arguments, a non-usual default chosen by hand, by the timeout and by a PC rule, the boot password, Linux Mint and Debian live); the order of arguments in the guest's own /proc/cmdline; the boot-password panic and its fix (a panic before, the desktop after); the menu scripts of 0.4.0 and this version are identical for 12 views (UEFI and BIOS) when no entry settings are used. go vet and the whole test suite pass.

Not verified: a BIOS (non-UEFI) walk-through of the entry settings (script comparison and unit tests only); an installer run to the end; the backup restore by clicking the page's buttons (the same API calls were used). Known small follow-ups are in #69 and #70.

Contributors: thank you @omar-akermi for the per-ISO boot entries (#68), the Zstandard decoder limits (#67) and the stable cleanup test (#66).

Gola Bootlace 0.4.0

Choose a tag to compare

@ghazichaieb ghazichaieb released this 25 Oct 15:50
  • Answer files are given to fewer hosts (#60). The address of an answer file has no password (an installer cannot log in), so a file is now given only to a MAC address that asked for the menu in the last two hours, and only twelve files after each time that PC asked for the menu; then the address answers "not found" until the PC asks again (a refusal says why in the log, the requester sees nothing more). The same network address as the menu request is NOT required by default: in a real run the installer of Ubuntu Server was given another address than iPXE had (the router leases by client identifier), so that condition refused the PC it was meant for; it is a switch ("Stricter") on the Unattended page for networks where a PC keeps its address. Saving, listing and attaching a file now warn when a line holds a password in plain text for its kind (Debian preseed, Ubuntu autoinstall, kickstart, AutoYaST); nothing is refused. This does not stop a host of the local network that asks the menu with the MAC address it wants and then fetches: the documentation says so. Verified with Ubuntu Server 26.04.1 in a UEFI VM (see the issue); the gate, the counters and the warnings are covered by unit tests.
  • Windows answer files per PC or per ISO (#17). A new answer file kind, Windows autounattend.xml, attached to an ISO or to one PC like the other kinds (same opt-in, confirmation, masked passwords, plain-password warning and limits as for the web address). The Windows set-up environment has no web client (no curl, getmac, wmic, powershell, certutil or findstr), so the file travels over the share: the ISO's folder shows _bootlace\<MAC>\autounattend.xml for a PC that has a file, made for that PC ({{hostname}}, {{computername}} cut to 15 characters, {{name}}, {{mac}}, {{ip}}, {{serial}}; words are XML-escaped), and bootlace.cmd reads its own MAC addresses from ipconfig /all, copies the file and starts Setup with /unattend:. An autounattend.xml in the ISO's own folder keeps working when nothing is attached (an attached file wins). A "native" (Secure Boot) start counts as asking for the menu. Verified in a real Windows 11 26H2 Setup in Hyper-V: unattended install onto an empty disk, computer name SHOP-PC-7 taken from the PC's name, account made, first-logon command ran. Windows opens the folder that holds a file before copying it, so the share also shows the two folders on the way (found by the first real run, which failed with "path not found").
  • Answer file text is escaped for XML kinds. Values put into AutoYaST and Windows files (the PC's name, serial ...) are XML-escaped.
  • Windows Setup options (#18). Menu page > Tuning: the five Windows 11 check ticks were tried in a real 26H2 Setup (VM without TPM, 2 GB, 40 GB disk): without the TPM tick Setup stops with "This PC can't run Windows 11", with it Setup goes on; the memory and storage ticks changed nothing visible there and the Secure Boot and processor ticks could not be tried (see the compatibility page, which also lists the runs). New Local account at the first start tick: Windows 11 26H2's first start has no way past its network page (OOBE\BYPASSNRO and the BypassNRO value no longer do anything: tried); HideWirelessSetupInOOBE in an answer file is what removes the page, and the first start then asks for a local account (verified offline). With the tick on, Bootlace makes that small answer file when no Windows answer file is attached (a PC set to "No answer file" is left alone). New Drivers tick: driver packages in <library>/drivers/all/ (every Windows ISO) or drivers/<ISO file name without .iso>/ are added to the boot.wim copy the server hands out (the ISO itself is never changed), loaded with drvload before Setup starts and kept at \$WinPEDriver$ so that Setup also takes them into the new Windows; the iPXE start now loads that copy too. Verified: a virtio-scsi disk that Setup did not see ("We couldn't find any drives") is seen with the drivers on, and the package is in the new Windows' driver store afterwards. Not seen: the installed Windows booting from that disk (the Proxmox test VM's QEMU exits at 75 % of the install, probably the old CPU). setup.exe /installdrivers does not exist in the set-up image's Setup and is not used. Without wimlib-imagex the drivers are ignored with a log line instead of breaking the start.
  • Answer files with control characters in an XML kind are refused. A stray vertical tab or form feed in an AutoYaST or Windows file (it came from a backslash escape in a script during the tests) made Windows Setup stop with "an internal error while loading or searching for an unattend answer file". The page now says which line holds it.
  • A dual-stack web port (#29, stage 1). The web port listens on IPv4 and IPv6. This showed a real bug: on Windows a link-local peer comes as 169.254.x.x%<interface> (and fe80::1%eth0 on Linux), net.ParseIP refuses it, and the access check refused the PC as "unknown"; the zone is now dropped (access.Host).
  • The Proxmox update shows what is new. update in the container (and update --check), and the script that updates a container from the host, now list a short line for each changelog entry between the installed version and the offered one, not just the latest release: a server three releases behind sees all three, security notes are marked with !, a test build counts as newer than the release it came from, and nothing is shown for an older version or when the changelog cannot be read. (A script change: it takes effect when the host script runs, no release is needed.)
  • Debian and Kali netinst installers can install over the network (#59). The CD flavour of the Debian installer can only continue with the CD it was started from, so a netinst ISO started over the network stopped at "Detect and mount installation media". Bootlace now gives it a CD: an archive made for each ISO adds the network card drivers, iscsistart and a start-up script, all taken from the ISO's own packages for the kernel the ISO starts (nothing names a Debian version, so a new kernel needs no change here), and the script logs in to the ISO's iSCSI CD-ROM before the installer's first step. The installer's own CD detection then mounts it like a disc, and the answer files of the Unattended page (preseed) are read as they are for any installer. It needs the iSCSI server (on by default); without it the menu entry now says that the installer cannot install. Run to an installed, booting system in Proxmox VMs: Debian 13.7 (by hand, UEFI), Debian 13.7 (also in BIOS) and Debian 12.15 with a preseed file, Kali 2026.2 (kernel 7.1) with a preseed file. The graphical installer was run with 3 GB and 1 GB; with 768 MB it switches to its low-memory (text) mode and goes on; with 512 MB it did not get past its start.
  • The preseed template no longer uses the reserved user name admin (#63). The Debian installer refuses it ("reserved for use by the system") and an unattended install stopped there. The template says localadmin.
  • Debian preseed installs get the PC's hostname before networking (#64). When a preseed is attached, the kernel line carries the same sanitized PC name (or MAC-based fallback) as the answer file's hostname word, including an override of DHCP-provided names. Explicit ISO or PC hostname arguments are preserved. Network answers inside a network-fetched preseed arrive too late to configure the initial network. Real run on a Debian 13.7 netinst UEFI VM: a PC named "Front Desk 7" installed as front-desk-7, an unnamed one as pc-<last 6 hex of its MAC>, and without the change the installed name was debian. Thanks to @omar-akermi (#65).

What was verified, and what was not

Seen working in real runs: Ubuntu Server autoinstall with the new answer-file limits (#60); a Windows 11 26H2 Setup started from a per-PC answer file in Hyper-V (#17); the Windows 11 check ticks in a Setup on a VM without TPM (only the TPM tick changed anything there); a real driver (virtio-scsi) making Setup see its disk (#18); Debian 13.7 netinst installs taking the PC's name, with an explicit name kept and the old behaviour (debian) reproduced without the change (#64, #65); the Proxmox update summary.

Not verified: the Windows tick boxes and driver folders on real PC hardware (please try them before relying on them); the Secure Boot "native" Windows start with an attached Windows answer file; a non-English Windows with answer files; groups of PCs; kickstart and AutoYaST installs to the end; the installed Windows booting from the virtio-scsi disk (the Proxmox test host's older processor stopped the install about three quarters through, the same ISO finishes on Hyper-V). Not built, by decision: NFS export (#30), imaging (#32), a catalog of network installers (#21); IPv6 is only stage 1 (the web port), see #29.

Contributors: thank you @omar-akermi for the Debian preseed hostname fix (#65, fixes #64).

Gola Bootlace 0.3.1

Choose a tag to compare

@ghazichaieb ghazichaieb released this 25 Oct 09:02
  • A separate debug shell switch (#28). Verbose start-up no longer opens a shell: Proxmox's proxdebug (and the shells of the other systems) now belong to a new Debug shell switch next to Verbose start-up and Serial console, on the ISOs page and in a PC's Kernel options. It adds the word that stops the start-up at a shell: proxdebug (Proxmox VE), rd.break=pre-mount (Fedora and the other dracut systems), break=premount (Debian live, Ubuntu); systems without such a word ignore it. Each word was checked on a real UEFI virtual machine started through Bootlace (Proxmox VE 9.1, Debian 13 live, Fedora 44 live, Ubuntu 22.04). Settings made before keep working; a PC or ISO with only this switch is kept like one with the others.
  • Next page and Previous page keep their direction. A page reached with a page button now highlights that same button, so pressing Enter again keeps going the same way: forwards lands on Next page, backwards on Previous page (before, going back landed on Next page of the earlier page). Where the button does not exist (page 1 reached backwards, the last page reached forwards) the other one is highlighted. It applies to the UEFI picture menu, the BIOS text menu and the pages of Other ways to start an ISO; the page links carry the direction (focus=prev / focus=next), and an older address without it behaves as before. Checked on real UEFI and BIOS iPXE through four and seven pages, forwards and backwards.
  • Configured PCs keep their history when the list is full (#61). A PC with an administrator-assigned name, MAC rule or non-empty kernel/answer options is no longer evicted by new MAC addresses. The 200-PC cap stays: if every slot is protected, new PCs can still boot but are not recorded until a slot is available. Automatic age-based cleanup and explicit Forget actions are unchanged. Thanks to @omar-akermi (#62).
  • The DHCP listener can no longer grow the list of PCs without limit (#61). The menu path was bounded and protected the configured PCs (#62), but a PC that only asked for a boot file was added by a second piece of code with no bound at all, so a flood of DHCP requests with made-up MAC addresses still grew the list (and pcs.json) without end. Both paths now go through one function that applies the same bound and the same protection.
  • Verbose start-up removes silent splash settings (#28). Valued splash options such as Proxmox's splash=silent are removed and other kernel log levels are replaced by loglevel=7. The same correction applies to per-PC verbose options. Thanks to @omar-akermi (#58).

Contributors: thank you @omar-akermi for the protection of configured PCs in the history list (#62, with a follow-up here for the DHCP path) and the verbose start-up fix (#58).