shutdown: several fixlets

[yuwata]

Hopefully fixes #28645.

[bluca]

uhm doesn't this undo the fix that I did before the release? Where stuff under /run was lost in the transition to initrd?

[yuwata]

IIUC, the previous issue is on initramfs -> host transition on boot. But this is for shutting down, and host -> initramfs transition.

[yuwata]

At least, v253 did not mount /run/ recursively. git diff v253..v254 src/shutdown/shutdown.c

@@ -162,17 +163,15 @@ static int parse_argv(int argc, char *argv[]) {
 }
 
 static int switch_root_initramfs(void) {
-        if (mount("/run/initramfs", "/run/initramfs", NULL, MS_BIND, NULL) < 0)
-                return log_error_errno(errno, "Failed to mount bind /run/initramfs on /run/initramfs: %m");
-
-        if (mount(NULL, "/run/initramfs", NULL, MS_PRIVATE, NULL) < 0)
-                return log_error_errno(errno, "Failed to make /run/initramfs private mount: %m");
-
-        /* switch_root with MS_BIND, because there might still be processes lurking around, which have open file descriptors.
-         * /run/initramfs/shutdown will take care of these.
-         * Also do not detach the old root, because /run/initramfs/shutdown needs to access it.
-         */
-        return switch_root("/run/initramfs", "/oldroot", false, MS_BIND);
+        /* Do not detach the old root, because /run/initramfs/shutdown needs to access it.
+         *
+         * Disable sync() during switch-root, we after all sync'ed here plenty, and a dumb sync (as opposed
+         * to the "smart" sync() we did here that looks at progress parameters) would defeat much of our
+         * efforts here. */
+        return switch_root(
+                        /* new_root= */ "/run/initramfs",
+                        /* old_root_after= */ "/oldroot",
+                        /* flags= */ SWITCH_ROOT_DONT_SYNC);
 }
 
 /* Read the following fields from /proc/meminfo:
@@ -341,6 +340,12 @@ int main(int argc, char *argv[]) {

[bluca]

Ok, let's give it a shot

[yuwata]

I'd like to wait for reporter feedback for a while, as I have not tested it.

[mrc0mmand]

Nope, still seeing the error (+ an extra debug message just to confirm I run the correct binary):

[yuwata]

@mrc0mmand Could you test again?

[mrc0mmand]

@mrc0mmand Could you test again?

The computer is still unhappy:

[yuwata]

@mrc0mmand Does v253 work fine?

[Freyon77]

@mrc0mmand Does v253 work fine?

Yep

[mrc0mmand]

@mrc0mmand Does v253 work fine?

It seems so, I don't see any errors even with some debug output sprinkled around:

[mrc0mmand]

And just a note: on v253 it seems to go through the pivot_root() branch, but it works even if I force it to use MS_MOVE instead:

[yuwata]

@mrc0mmand @Freyon77 Could you test again?

[mrc0mmand]

@mrc0mmand @Freyon77 Could you test again?

This looks promising:

[  OK  ] Reached target shutdown.target.
[  OK  ] Reached target final.target.
[  OK  ] Finished systemd-halt.service.
[  OK  ] Reached target halt.target.
[  118.668517] reboot: System halted

[yuwata]

Thank you!

[mrc0mmand]

@mrc0mmand Could you test again?

Still looks good!

[  OK  ] Reached target shutdown.target.
[  OK  ] Reached target final.target.
[  OK  ] Finished systemd-halt.service.
[  OK  ] Reached target halt.target.
[39099.296592] systemd-shutdown[1]: switch_root: about to pivot_root()
[39099.297278] systemd-shutdown[1]: switch_root: pivot_root() succeeded
[39099.394763] reboot: System halted

[  OK  ] Reached target shutdown.target.
[  OK  ] Reached target final.target.
[  OK  ] Finished systemd-halt.service.
[  OK  ] Reached target halt.target.
[  208.428841] systemd-shutdown[1]: Pivoting root file system failed, moving mounts instead: Operation not permitted
[  208.429875] systemd-shutdown[1]: switch_root: about to move '.' to '/'
[  208.430518] systemd-shutdown[1]: switch_root: about to chroot() into '.'
[  208.431182] systemd-shutdown[1]: switch_root: about to chdir() to '.'
[  208.431922] systemd-shutdown[1]: switch_root: all done!
[  208.458289] reboot: System halted

[Freyon77]

@mrc0mmand what about pivoting root file system failed?

[mrc0mmand]

@mrc0mmand what about pivoting root file system failed?

That's just me setting r to -1 to test the fallback branch, probably should've dropped that from the output.

[yuwata]

Thank you for testing so many times!

@bluca Could you take a look again?

[keszybz]

I queued up all three commits for v254.1.

[poettering]

So I am pretty sure the 2nd commit is not right. It's not load-bearing for the bugfix this PR is supposed to be, is it?

I prepped #28793 that reverts that part. Please take a look for an explanation.