On Wed, Oct 11, 2023 at 4:07 PM Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > > On Wed, 11 Oct 2023 at 14:07, Amir Goldstein <amir73il@xxxxxxxxx> wrote: > > > > On Wed, Oct 11, 2023 at 1:18 PM Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > > > > > > On Wed, 11 Oct 2023 at 10:45, Amir Goldstein <amir73il@xxxxxxxxx> wrote: > > > > > > > > > > > We could add new keys: > > > > lowerdir_first=,lowerdir_next=,lowerdatadir_next= > > > > as explicit variants for the "[^:]",":","::" prefix detection > > > > and those don't need to be unescaped. > > > > > > Good idea. I'd merge "lowerdir_first" and "lowerdir_next" into > > > "lowerdir_one" or whatever for simplicity. I'd also consider dropping > > > the prefix detection, since it has only been in mainline for one > > > cycle. > > > > > > > OK. > > > > Christian, > > > > Do you know any userspace that already uses your new append prefixes? > > Do we have any good reason to support "lowerdir_first" > > so a lower dir stack could be reset before creating the sb? > > If that is a requirement, I suggest extending fsconfig(2) to allow > resetting an option. > FWIW, I see that Christian has also implemented reset of lowerdir stack with an empty string. > > > > > > > > > > > > Anyway, let's focus on what you would like best. > > > > > > If you prefer to just fix the regression, it is doable. > > > > > > If you prefer the upperdirfd, workdirfd, lowerdirfd API, I think we can > > > > > > find a volunteer to write it up. > > Can't the existing option names be overloaded if a separate cmd > (FSCONFIG_SET_PATH or FSCONFIG_SET_PATH_EMPTY) is used in fsconfig()? > > > > > BTW, it looks like we also don't display the user passed lowerdir > > > > parameter as is in the case of escaped characters in lowerdirs. > > > > Admittedly, that is another change of behavior from the new mount > > > > api param parsers. > > > > > > And it's a bug (regardless of being a regression or not) since commas > > > and whitespace must be escaped on this interface, and colon too for > > > being a separator of lower layers. > > > > OK. I think it should be easy to fix this bug. > > I can look into it. > > Thanks. > > > > More fun: upperdir and workdir use seq_show_option() which escapes > > > commas and whitespace, so any escaped characters during mount will end > > > up being double escaped. > > > FYI, I tried to reproduce this bug, but could not. The reason (I think) is that the escaping provided by seq_show_option() is not the dumb "prepend \", but it's the "convert to \octal". It turns out that I also needed to store the lowerdir before unescaping, so that a path like 'lower\:2' will be presented correctly (with \) in mount options. Thanks, Amir.
