Skip to content

Finit v4.1
Compare
Choose a tag to compare
@troglobit troglobit released this 06 Jun 15:35
· 954 commits to master since this release
4.1
c8d6ad6

Bug fix release. Also disables handlers for SIGINT and SIGPWR, a new set of sys conditions are instead generated which can be used to trigger external programs.

Changes

  • Change behavior on SIGUSR1 to be compatible with sysvinit and systemd. Previously SIGUSR1 caused Finit to halt, like BusyBox init. This had "interesting" side effects on Debian systems when coexisting with sysvinit (upgrading/reinstalling causes scripts to kill -USR1 1)
  • Change how contrib/debian/install.sh sets up a Grub boot entry for finit. We now modify the $SUPPORTED_INITS variable in 10_linux
  • Disable default kernel ctrl-alt-delete handler and let Finit instead catch SIGINT from kernel to be able to perform a proper reboot. There is no default command for this, you need to set up a task that triggers on <sys/key/ctrlaltdel> to issue initctl reboot
  • Added keventd to provide <sys/pwr/ac> condition to Finit. keventd is currently only responsible for monitoring /sys/class/power_supply for changes to active AC mains power online status. Enable keventd with configure --with-keventd
  • For handling power fail events (from UPS and similar) a process may send SIGPWR to PID 1. Finit no longer redirects this to SIGUSR1 (poweroff). There is no default command for this, you need to set up a task that triggers on <sys/pwr/fail> to issue initctl poweroff
  • Built-in Finit getty is now a standalone program
  • Default termios for TTYs now enable IUTF8 on input
  • If /bin/login is not found, Finit now tries sulogin before it falls back to an unauthenticated /bin/sh
  • Dropped (broken) support for multiple consoles. Finit now follows the default console selected by the kernel, /dev/console
  • Dropped signal handlers for SIGSTOP/TSTP and SIGCONT
  • Added support for \n, in addition to \r, in "Please press Enter" prompt before starting getty
  • Finit no longer parses /proc/cmdline for its options. Instead all options are by default now read from argv[], like a normal program, this is also what the kernel does by default. Please note, this may not work if your systems boots with an initramfs (ymmv), for such cases, see configure --enable-kernel-cmdline
  • The following plugins are now possible to disable (for containers): rtc.so, urandom.so, you may also want to disable hotplug.so. They are all enabled by default, as in Finit 4.0, but may be moved to external tools or entries in finit.conf in Finit 5.0
  • Added support for reading PRETTY_NAME from /etc/os-release to use as heading in progress output, unless --with-heading=GREET is used.
  • Added manual pages for finit(8), initctl(8), and finit.conf(5)

Fixes

  • Stricter interface name validation in netlink plugin, modeled after the kernel. Suggested by Coverity Scan

  • Fix problem of re-registering a service as a task. Previously, if a fundamental change, like type, was made to an active service/run/task it did not take. Only possible workaround was to remove from config

  • initctl: drop warning when removing a non-existing usr condition

  • initctl: drop confusing errno 0 when timing out waiting for reply

  • Ensure services in plugins and from Finit main belong to a cgroup

  • Ensure init top-level cgroup remains a leaf group

  • Fix tty parse error for detecting use of external getty

  • Fix default name: and :ID for tty's, e.g. ttyS0 now gives tty:S0 as expected. This was default for built-in getty already

  • Fix max username (32 chars) in bundled Finit getty

  • The contrib/*/install.sh scripts failed to run from tarball

  • Finit no longer forcibly mounts; /dev, /proc, or /sys, instead it checks first if they are already mounted (devtmpfs or container)

  • Fix /etc/fstab parser to properly check for 'ro' to not remount the root filesystem at boot. The wrong field was read, so a root mounted by an initramfs, or by lxc for a container, had their root remounted

  • Fix SIGCHLD handler, waitpd() may be interrupted by a signal

  • Reset starting flag of services being stopped. When a service is started and then stopped before it has created its pid file, it could be left forever in the stopping state, unless we reset the starting flag.

  • Fix #170: detect loss of default route when interfaces go down. This emulates the missing kernel netlink message to remove the condition net/default/route to allow stopping dependent services

  • Fix #171: restore automatic mount of /dev/shm, /dev/pts, /run and /tmp, unless mounted already by /etc/fstab. This is what most desktop systems expect PID 1 to do. Here we also make sure to mount /run/lock as a tmpfs as well, with write perms for regular users, this prevents regular users from filling up /run and causing DoS.

  • Fix #173: netlink plugin runs out of socket buffer space;

      finit[1]: nl_callback():recv(): No buffer space available

    Fixed by adding support for resync with kernel on ENOBUFS. See netlink(7) for details. As a spin-off the plugin now supports any number of interfaces and routes on a system. On resync, the following message is now logged, as a warning:

      finit[1]: nl_callback():busy system, resynchronizing with kernel.

  • Fix #174: loss of log messages using combo of prio and facility, e.g., logit(LOG_CONSOLE | LOG_NOTICE, ...), by Jacques de Laval, Westermo

  • Fix #175: ensure Finit does not acquire a controlling TTY when checking if a device is a TTY before starting a getty. This fixes an old bug where Ctrl-C after logout from a shell could cause PID 1 to get SIGINT, which in turn could lead to a system reboot

Assets 8
Loading
0 Join discussion