Add "." and ".." in directory listings

[DrDaveD]

This is a first attempt at

• fixes getdents does not list . and .. files on squashfuse-mounted filesystems #133

I'm sure it is incomplete but it does work to make ls -a show . and .. files.

@vasi Can you give me some hints on what else is needed? I didn't know what to fill in the inode and inode_number fields but it doesn't seem to make a difference for ls -a and ls -al. Also I would think that some of the other functions in dir.c that deal with offsets will need some changes but I don't really know what changes.

I came up with the idea of adding fake offset indexes from the corresponding squashfs kernel code.

[vasi]

Thanks! I'm on vacation now, so I'll only be able to look at this in a couple of weeks. Sorry for the delay.

[DrDaveD]

@vasi are you back?

[vasi]

Hi @DrDaveD , thanks for the ping. I just arrived home, I'll try to look at this soon. Just trying to unpack, figure out where my lost luggage is, etc!

[vasi]

Gonna do a quick pass now.

[vasi]

So the fundamental issue here is what we're trying to change the semantics of:

1. The squashfuse_ll/_hl FUSE programs, or
2. The squashfuse internal representation (and/or libsquashfuse)

What you've done here is a bit in the middle, yah? We're changing the meaning of sqfs_dir_entry->offset (and ->next_offset), but we're not changing the meaning of sqfs_traverse_next(). We should go either all-in on 1 or 2.

Advantages for each option:

1. Changing FUSE programs only

• Existing programs that use libsquashfuse keep working.
• No need to fix sqfs_dir_ff_offset() and friends to use our new definition of offset.
• Doesn't make non-UNIX drivers awkward, eg: if someone wanted to implement a Windows filesystem for squashfuse.

2. Changing libsquashfuse semantics

• We only have to implement ./.. once in the library code, not for each of low-level and high-level.

[vasi]

Minor nit: This probably looks a lot nicer using strncpy() (or similar) than doing character-by-character modification. Similarly below in traverse.c, using strcmp().

[vasi]

Also thank you so much for your patience, I'm sorry for the bad vacation timing.

[DrDaveD]

Ok, I didn't understand which files were considered part of libsquashfuse. It sounds there are more advantages to changing it only in the FUSE programs. Maybe the duplicate changes in hl/ll can be kept to a minimum via shared function(s).

It looks like both of the source files I changed are part of libsquashfuse. So if I avoided those, it would be the readdir functions in ll.c & hl.c that I should be modifying instead? I guess I would need to add an additional variable in the ll and hl structures to indicate whether it's time to return ., .., or call sqfs_dir_next.

[vasi]

I think you'd need to change sqfs_hl_op_readdir/sqfs_ll_op_readdir to:

• Translate between "UNIX offset" and "libsquashfuse offset". You'd want to translate UNIX -> libsquashfuse for the input parameter, and libsquashfuse -> UNIX for the offsets returned via fuse_fill_dir_t/sqfs_ll_add_direntry.
• When given an offset low enough, output the dotfiles before doing the real work.

Any common functions you build could go in fuseprivate.c, to minimize duplicate work.

Oh, and the last thing is that I have no idea whether we care about the inode values for the dot entries. We should try to test whether things like ls mnt/foo/.. works correctly with our change,

[DrDaveD]

Thanks, that seems quite clear.

I'm quite sure that .. and . as part of file paths already work, independent of what shows up in a directory listing, but I'll make sure they still work when I am finished.