On 2017-09-17, at 10:27 AM, larkost <[email protected]> wrote: > As to #2, you are going to have to be a lot clearer about what you are > trying to do. Apple’s recommendation for handling startup dependencies is to > have each tool monitor its dependencies itself, and handle the cases where it > needs to wait for something. My general approach has been to write a Python > script in front of each one that has dependencies that waits for them then > uses something like `execv` to hand over the PID to the real process.
This is the sort of thing that a "all systems should ship secure and no user can make it insecure by accident" person will hate. Basically, consider that everything started has hidden, unstated dependencies. For example: /tmp will be ready for use. The system clock has been set and is accurate. Hostname is valid. Etc. In deep history, the idea was that /etc/rc started by doing cleanup -- recovering the password file if it was edited, saving editor temp states, wiping /tmp back to clean AFTER both of those, etc. Included here would be things like mounting drives that the system needed (good ol /etc/fstab, which is empty, and heck, I have attempted to use it and not been able to get partitions to mount properly), There was a point in /etc/rc where it started to run daemon programs (that did a fork/exec), or run other programs with an "&". That is the point where things start up for multiuser. Everything after that point has hidden assumptions of basically, "Everything that /etc/rc would do to prepare the system has been done". So, what I'm looking for is basically a way (putting it into writing like this actually makes it easy to state) to add a dependency tag to everything apple ships that says "A site-local startup script has run", and add in a program (bash, python, executable, whatever) that gets a chance to run before anything else gets going. ===== Launchd used to be flexible at boot. Going from 10.7 to 10.9 gave me trouble. It turns out that Launchd as pid 0 will refuse to boot the system unless several paths exist on the root partition as directories with specific ownerships/permissions. You are not allowed to even have them as symbolic links to a second partition on the boot drive. Something I learned early on doing system admin on unix systems (hey, I learned these lessons in V7 and S3, I can use "unix") was that you wanted your root partition to be as little-write as possible. Neither tmp nor swap go on root. The idea of per-user temp and cache directories to prevent people from interfering with each other is great, but why does that have to be on a partition that doesn't have 20 GB of wasted temp space? Or why do I have to waste that much space? Being able to say "Fine, here's a directory for launchd to be happy, now let me mount a fake that supplies a symbolic link before anything else runs and starts writing into the wrong place"? That's kinda my first goal. /var, /var/vm, /var/tmp, /var/folders, and a few others are forced to be root partitions. --- Entertaining minecraft videos http://YouTube.com/keybounce
_______________________________________________ MacOSX-talk mailing list [email protected] http://www.omnigroup.com/mailman/listinfo/macosx-talk
