On Wed, Mar 30, 2022 at 08:13:00PM +0200, наб wrote: > Which take an user, group, and mode, respectively, and set them on the > target after mounting > > This is vaguely similar to tmpfs(5)'s [ug]id= and mode= options, > but we POSIX-parse the user- and group names > > Oft requested in systemd/zram-generator, since a common use-case > is to use it to create /tmp or an equivalent directory that needs > to be a=rwx,o+t (or a user's private temp that needs to be owned > by them) ‒ this is impossible without terrible hacks, cf. > https://github.com/systemd/zram-generator/issues/150, > https://github.com/systemd/zram-generator/issues/146, &c. > > This started off as a Set{User,Group,Mode}= systemd mount unit, > but was poetterung into libmount options: > https://github.com/systemd/systemd/pull/22889 I see how usable this feature could be, but it also increases complexity of the mount(8) command. > But I'm not sure how to proceed. I've stuffed the parsing and chowning > stage into mnt_context_finalize_mount() for ease-of-PoC, but should: > (a) the post-syscall error handling in mnt_context_get_mount_excode() > be extended to recognise MNT_ERR_MOUNTOPT? Yes, see MNT_ERR_NAMESPACE, it the same thing. We have it in the pre-syscall and in post-syscall sections. > (b) the parsing/chowning stages be split (parsing in pre-mount prep, > chowning in post-mount)? with a new MNT_ERR_ flag potentially? Yes, this is probably the most robust way (don't call mount syscall if X-mount.* is invalid). I guess you can add tgt_owner, tgt_group and tgt_mode to libmnt_context to keep parsed results, fill the fields in mnt_context_prepare_mount() (or so) and use it mnt_context_do_mount() (before mnt_context_switch_ns()). Don't forget to reset it in mnt_reset_context(). Karel -- Karel Zak <kzak@xxxxxxxxxx> http://karelzak.blogspot.com
