Path setting corrupted (renamed issue)

[vazub]

finit/src/sig.c

Line 389 in e1fe870

run("mount -n -o remount,ro -t dummytype dummydev /", NULL);

Not a bug, but a curious implicit assumption, of using mount command from the util-linux package. When using other implementations, that do not support some of the options, e.g. from toybox or busybox, the command fails. Would it be reasonable to avoid running a dedicated command and call something from sys/mount.h instead?

[troglobit]

Barely remember this one, but it works for me with BusyBox.

[vazub]

Ok, I might have been wrong here indeed. My initial assumption was based on this issue I am facing:

I tracked it to the code part mentioned, and assumed it was the culprit. But having tested it again with 3 mount implementations (BusyBox, Toybox, util-linux), the issue persists for each of them, so apparently something else is at play here. Will have to debug more and look for more insights.

[vazub]

Alright, an update here. I found out that for some reason, at least with my build of finit for my custom non-FSH distro, the PATH is not getting parsed properly, so therefore initctl can not find commands it shells out to. This affects mount as in example above, as well as any service stanzas. A mitigation for this is to always use absolute paths, then it works fine. Still puzzled though as to why this happens, as based on the code finit is supposed to use either PATH returned by env, or _PATH_DEFPATH, or _PATH_STDPATH. My distro keeps everything in /bin, and all these env vars do contain it as part of specification. Not sure yet what causes this weird behavior. Seeing as no one raised this issue before, I tend to assume this is my special case?

[troglobit]

What is this sorcery? 🤔🤯 The checkenv() helper is real simple:

static inline int checkenv(const char *env)
{
        const char *val = getenv(env);

        if (!val || !val[0])
                return 0;

        return 1;
}

Maybe you could add some printf("%s: %s\n", env, val); to that function to see what you get?

[vazub]

Here is the output of logging env variables before and after calling checkenv/setenv 21753e2

And here is similar output when passing PATH=/bin to kernel cmdline in bootloader:

So looks like checkenv/setenv logic works correctly, by setting the PATH. However, that does not seem to be enough, as further along the road that flow does not work still.

[vazub]

More insights, after tracing PATH state change during main execution:

Mar  3 10:44:17 mimicOS finit[1]: PATH after conf_saverc = /bin:/usr/bin:/sbin:/usr/sbin

Mar  3 10:44:17 mimicOS finit[1]: iwatch_add1():adding new watcher for path /etc/finit.d
Mar  3 10:44:17 mimicOS finit[1]: iwatch_add1():adding new watcher for path /etc/finit.d/available/
Mar  3 10:44:17 mimicOS finit[1]: iwatch_add1():adding new watcher for path /etc/finit.d/enabled/
Mar  3 10:44:17 mimicOS finit[1]: iwatch_add1():adding new watcher for path /etc/finit.conf
Mar  3 10:44:17 mimicOS finit[1]: iwatch_add1():skipping /etc/default/: no such file or directory
Mar  3 10:44:17 mimicOS finit[1]: iwatch_add1():adding new watcher for path /etc/conf.d/
Mar  3 10:44:17 mimicOS finit[1]: conf_reload():Set time  daylight: 0  timezone: 0  tzname: UTC 
Mar  3 10:44:17 mimicOS finit[1]: parse_conf():*** Parsing /etc/finit.conf
Mar  3 10:44:17 mimicOS finit[1]: service_register():/etc/finit.conf: skipping initctl: No such file or directory
Mar  3 10:44:17 mimicOS last message buffered 1 times
Mar  3 10:44:17 mimicOS finit[1]: tty_parse_args():Registering built-in getty on TTY /dev/tty1 at 0 baud with term N/A
Mar  3 10:44:17 mimicOS finit[1]: service_register():Found existing svc for /pkg/finit/libexec/getty name tty id 1 type 8
Mar  3 10:44:17 mimicOS finit[1]: service_register():Service tty:1 runlevel 0x3e
Mar  3 10:44:17 mimicOS finit[1]: tty_parse_args():Registering built-in getty on TTY /dev/tty2 at 0 baud with term N/A
Mar  3 10:44:17 mimicOS finit[1]: service_register():Found existing svc for /pkg/finit/libexec/getty name tty id 2 type 8
Mar  3 10:44:17 mimicOS finit[1]: service_register():Service tty:2 runlevel 0x3c
Mar  3 10:44:17 mimicOS finit[1]: tty_parse_args():Registering built-in getty on TTY /dev/tty3 at 0 baud with term N/A
Mar  3 10:44:17 mimicOS finit[1]: service_register():Found existing svc for /pkg/finit/libexec/getty name tty id 3 type 8
Mar  3 10:44:17 mimicOS finit[1]: service_register():Service tty:3 runlevel 0x3c
Mar  3 10:44:17 mimicOS finit[1]: glob_append():conf_reload(): glob /pkg/finit/lib/system/*.conf ...
Mar  3 10:44:17 mimicOS finit[1]: glob_append():conf_reload(): glob /run/finit/system/*.conf ...
Mar  3 10:44:17 mimicOS finit[1]: glob_append():conf_reload(): glob /etc/finit.d/*.conf ...
Mar  3 10:44:17 mimicOS finit[1]: glob_append():conf_reload(): glob /etc/finit.d/enabled/*.conf ...
Mar  3 10:44:17 mimicOS finit[1]: parse_conf():*** Parsing /etc/finit.d/enabled/dhcpcd.conf
Mar  3 10:44:17 mimicOS finit[1]: service_register():/etc/finit.d/enabled/dhcpcd.conf: skipping dhcpcd: No such file or directory
Mar  3 10:44:17 mimicOS finit[1]: parse_conf():*** Parsing /etc/finit.d/enabled/syslogd.conf
Mar  3 10:44:17 mimicOS finit[1]: service_register():Creating new svc for /bin/syslogd name syslogd id  type 1
Mar  3 10:44:17 mimicOS finit[1]: service_register():Service syslogd runlevel 0x7fe
Mar  3 10:44:17 mimicOS finit[1]: parse_cmdline_args():Modified args for /bin/syslogd detected:  /bin/syslogd -Fss $SYSLOGD_OPTS
Mar  3 10:44:17 mimicOS finit[1]: parse_cgroup():syslogd: cgroup name:system leafname: cfg: delegate:no
Mar  3 10:44:17 mimicOS finit[1]: group_init():path /sys/fs/cgroup/init, leaf 1, cfg cpu.weight:100
Mar  3 10:44:17 mimicOS finit[1]: group_init():/sys/fs/cgroup/init <=> cpu.weight:100
Mar  3 10:44:17 mimicOS finit[1]: group_init():ptr: cpu.weight:100
Mar  3 10:44:17 mimicOS finit[1]: cgset():path /sys/fs/cgroup/init, ctrl cpu.weight:100, prop NIL
Mar  3 10:44:17 mimicOS finit[1]: cgset():/sys/fs/cgroup/init/cpu.weight <= 100
Mar  3 10:44:17 mimicOS finit[1]: iwatch_add1():skipping, /sys/fs/cgroup/init/cgroup.events already registered
Mar  3 10:44:17 mimicOS finit[1]: group_init():path /sys/fs/cgroup/system, leaf 0, cfg cpu.weight:9700
Mar  3 10:44:17 mimicOS finit[1]: group_init():/sys/fs/cgroup/system <=> cpu.weight:9700
Mar  3 10:44:17 mimicOS finit[1]: group_init():ptr: cpu.weight:9700
Mar  3 10:44:17 mimicOS finit[1]: cgset():path /sys/fs/cgroup/system, ctrl cpu.weight:9700, prop NIL
Mar  3 10:44:17 mimicOS finit[1]: cgset():/sys/fs/cgroup/system/cpu.weight <= 9700
Mar  3 10:44:17 mimicOS finit[1]: iwatch_add1():skipping, /sys/fs/cgroup/system/cgroup.events already registered
Mar  3 10:44:17 mimicOS finit[1]: group_init():path /sys/fs/cgroup/user, leaf 0, cfg cpu.weight:100
Mar  3 10:44:17 mimicOS finit[1]: group_init():/sys/fs/cgroup/user <=> cpu.weight:100
Mar  3 10:44:17 mimicOS finit[1]: group_init():ptr: cpu.weight:100
Mar  3 10:44:17 mimicOS finit[1]: cgset():path /sys/fs/cgroup/user, ctrl cpu.weight:100, prop NIL
Mar  3 10:44:17 mimicOS finit[1]: cgset():/sys/fs/cgroup/user/cpu.weight <= 100
Mar  3 10:44:17 mimicOS finit[1]: iwatch_add1():skipping, /sys/fs/cgroup/user/cgroup.events already registered
Mar  3 10:44:17 mimicOS finit[1]: PATH after conf_monitor = \x7f

Looks like the culprit hides inside conf_reload function. Probably related to some env reset, or some mistreatment of the path variable.

[vazub]

Ok, issue finally pinned down - the misuse of a global path var.
A patch with a minor code refactor incoming.

[troglobit]

Fixed in PR #481