On Mon, Nov 11, 2024 at 02:12:16PM +0100, Miklos Szeredi wrote: > On Wed, 26 Jun 2024 at 14:23, Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > > > > On Tue, 25 Jun 2024 at 16:18, Josef Bacik <josef@xxxxxxxxxxxxxx> wrote: > > > > > But that means getting the buffer, and going back through it and replacing every > > > ',' with a '\0', because I'm sure as hell not going and changing all of our > > > ->show_options() callbacks to not put in a ','. > > > > > > Is this the direction we want to go? > > > > IMO yes. Having a clean interface is much more important than doing > > slightly less processing on the kernel side (which would likely be > > done anyway on the user side). > > So I went for an extended leave, and this interface was merged in the > meantime with much to be desired. > > The options are presented just the same as in /proc/self/mountinfo > (just the standard options left out). And that has all the same > problems: > > - options can't contain commas (this causes much headache for > overlayfs which has filenames in its options) > > - to allow the result to be consumed by fsconfig() for example > options need to be unescaped > > - mnt_opts is confusing, since these are *not* mount options, these > are super block options. > > This patchset was apparently hurried through without much thought and > review, and what review I did provide was ignored. So I'm > frustrated, but not sure what if anything can be done at this point, > since the interface went live in the last release and changing it > would probably break things... I understand your frustation but multiple people agreed that the interface as is is fine and Karel as main consumer agreed as well. So ultimately I didn't see a reason to delay the patchset. None of the issues you raised are really things that make the interface uncomsumable and Karel succeeded to port libmount to the new interfaces with success (minus the mnt_devname we're adding now that he requested) and was happy. If there's genuine behavioral problems that cause substatntial issues for userspace then I would request that you please add a new flag that changes escaping and parsing behavior for statmount().
