RFE: please propagate documented error codes from /usr/bin/swapoff in swapoff failed due to OOM

[poettering]

IIUC the kernel swapoff() syscall returns ENOMEM if there's not enough RAM to place the swapped out data into. It would be great if the swapoff binary would propagate that error code in some documented way through its exit status, so that systemd and similar systems can react to that condition in some way.

[karelzak]

Implemented. The magic exit status you look for is "2" (for ENOMEM). The rest is (copy & past from man page):

       0      success
       1      system has insufficient memory to stop swapping (OOM)
       2      swapoff syscall failed for another reason
       4      non-swapoff syscall system error (out of memory, ...)
       16     usage or syntax error
       32     all swapoff failed on --all
       64     some swapoff succeeded on --all

 The command swapoff --all returns 0 (all succeeded), 32 (all failed), or 64 (some failed, some succeeded).

We need also clean up swapon(8) to keep it consistent, but tomorrow ;-)

[poettering]

What did the tool return previously on failure? always 1? if so, for compat reasons, wouldn't it make sense to change the table above, and map OOM to 2 and "failed for another reason" to 1? That way, systemd could check for error code 2, and on new swapoff it's a clear indication for OOM, and on old OOM it would never happen. But with your current work we couldn't distuingish the case "old swapoff" from "new swapoff and oom", if you follow what i mean?

[karelzak]

The old version returns 255 on all swapoff(2) errors and 1 for non-swaoff() and for EPERM errors. Maybe avoid 1 at all would be the best thing. It means use 0, 2, 4, 8, 16, 32, and 64.

[poettering]

Yes, that would make sense!

For the systemd case we are primarily interested in having a recognizable error for the OOM case that wasn't previously used otherwise.

[karelzak]

Released in util-linux v2.36.