squashfuse using fuse-t

[udance4ever]

hi! I can compile squashfuse no problem with macfuse installed.

I am wondering if it is possible to compile squashfuse using fuse-t which doesn't depend on kernel extensions?

https://github.com/macos-fuse-t/fuse-t

if I have fuse-t installed (but not macfuse), it doesn't get past configure as it cannot find FUSE.

[vasi]

Hi, yes this is possible, I was just looking into it! It's a bit complicated, so I'll answer in two parts.

Easy version

• Install the FUSE-T 1.0.39e pre-release
• Download a tarball of squashfuse, eg: the most recent version 0.5.2. Unpack the tarball.
• Run LDFLAGS=-Wl,-rpath,/usr/local/lib ./configure --with-fuse-soname=fuse-t, then make -j8

Explanations of complications

1. There's a serious bug in FUSE-T that prevents squashfuse's readdir method from working. A fix is in progress! The squashfuse master branch has a workaround (at the cost of slightly reduced efficiency), and FUSE-T has the bug fixed in the pre-release above.
2. FUSE-T uses a different library name. That's why you need the soname flag. We should probably make squashfuse look for this specifically.
3. FUSE-T's library uses rpath weirdly. That's why you need to pass LDFLAGS manually, so the resulting binaries can find the FUSE-T library. I should probably file a bug against FUSE-T for this.
4. Building from git needs automake. It's traditional that a release tarball includes the configure script, but git itself doesn't. As the README explains, if you want to build from git you'll need to get autoconf/automake/libtool from eg: homebrew, and then run ./autogen.sh.
5. FUSE-T and MacFUSE don't provide a way of switching between them. If you have both of them installed, you may have to put extra effort in to make configure specifically choose FUSE-T. I found the following worked: CPPFLAGS='-DFUSE_USE_VERSION=26' LDFLAGS='-Wl,-rpath,/usr/local/lib' ./configure --with-fuse-soname=fuse-t --with-fuse-include=/usr/local/include/fuse --with-fuse-lib=/usr/local/lib. Perhaps we could add a new configure flag to deal with this situation, though I'm not sure how common it is.

[udance4ever]

thank you so much for the detailed explanations! helps me understand what’s going on :)

I gave this a shot:

user@host foo % squashfuse file.ext.squashfs  file.ext
fuse: mount failed with errro: -1

is there a log I can look at to see why it throws this error?

I don’t know if it’s also a function of being on macOS Sequoia beta - I already have macfuse installed in macOS Sonoma so I figure this is a beta, why not muck with it :)

[vasi]

Hmm, I don't have a Sequoia install to test with unfortunately. Is it possible that file.ext does not exist, or is a file rather than a directory? You should invoke squashfuse like squashfuse somefile.squashfs somedir.

If that's not it, you can try running with the -d option to turn on debugging. But I'm not sure it will show anything interesting that early in the mount process.

[udance4ever]

hey good news - I got it to work! 🙌🏼

I ran it again with -d and for some reason instead of throwing an error, the process waited while I gave Terminal permission to access files (paraphrasing, I think that's what it was asking for - I've seen the dialog box before and this is a new install)

after mounting once, I dismounted and tried it again without the -d option and it just works like you would expect.

This is really super not to have to go into Startup Security Utility and not have to permit kernel extensions! 🙏🏼

so glad you got this working. this is definitely the way forward as I wouldn't be surprised if macOS just decides to give kernel extensions the boot any release now!

[vasi]

Thanks for testing! I wish I understood better what made it work in the end—do you believe it was the permission dialog?

[udance4ever]

yeah- I feel ya - always nicer when the world is more predictable! :)

I do think the permission dialog made a difference but I can't explain why running it the first time without the debug option outright failed (-1) and then when I ran it again with the debug flag, all the debug spewed out (looked normal) and the permission dialog box showed up and it just worked!