fuse3?

[vaeth]

Are there any plans to support fuse3?
In the moment, it is not really a topic of course, but maybe on the long run some distributions might drop fuse in favour of the more current fuse3.
Experimentally, I had added "fuse3" in front of "fuse" in FUSE_NAMES in cmake/FindFUSE.cmake; compilation went smoothly, and it linked against libfuse3, but when trying to mount, the different API showed up:

fuse: warning: library too old, some operations may not not work
fuse: unknown option(s): `-o use_ino'
fuse failed.  …

I guess the warning is, because fuse3 switched to a "major"-"minor" version scheme (with "major"=3, currently). The removal of use_ino might of course indicate a more serious API change. I am afraid that rather different code paths are needed depending on whether fuse or fuse3 is in use.

[vaeth]

I had made the error vanish by removing

PUSHARG("-o");
PUSHARG("use_ino");

from the C code. However, the warning remains, and the mounted directory appears to be empty. So unfortunately, it is not only the option which has changed.

[blshkv]

would be good to have some progress here

[doj]

I've created #664 which works for me. Would be good if other people can test this branch.

[vaeth]

Would be good if other people can test this branch.

I tried it. Compilation was smoothly, and encfs seems to work fine without any options or with option --public.

However, encfs -f breaks if a relative path is specified as a first argument: The mount point is then not accessible. Absolute paths work also with option -f.

Therefore, a reasonable workaround is probably to not accept relative paths with option -f (as is apparently already the case without option -f).

[vaeth]

Since it essentially seems to work, I added an ebuild for gentoo to the mv overlay. Chances are good that there will arrive bug reports if there should be a major regression.

[agmartin2]

Hi,

I have been testing most of the changes in #664 on top of Debian encfs 1.9.5-3 (fuse2), trying to modify it to use fuse3. Results apparently work, but there are some strange things,

After creating a new encrypted volume and touch something inside it

$ mkdir -p .crypto/ target/
$ encfs `pwd`/.crypto `pwd`/target
$ touch target/00-HI-ENCRYPTED

ls -la yields some strange result

$ ls -al target/
total 8
drwxrwsr-x          2 myuser myuser 4096 dic  3 18:28 .
drwxrwsr-x          4 myuser myuser 4096 dic  3 18:28 ..
---------- 3651136296 root   saned     0 dic  1  4458884 00-HI-ENCRYPTED

After further mount I get

$ ls -al target/
total 69999410541872
drwxrwsr-x        2 myuser myuser           4096 dic  3 18:28 .
drwxrwsr-x        4 myuser myuser           4096 dic  3 18:29 ..
---------- 67103528 root   root   93983110258336 oct 13  4438351 00-HI-ENCRYPTED

This last changes slightly from mount to mount.

Using -f option with relative paths

$ encfs -f ./.crypto ./target

results in

d????????? ? ? ? ? ? target

be shown by ls -la in top dir. Using it with absolute paths results in something similar as above without -f.

Using original encfs package from Debian (still fuse2) I get the expected

$ ls -la target/
total 8
drwxrwsr-x 2 myuser myuser 4096 dic  3 17:43 .
drwxrwsr-x 4 myuser myuser 4096 dic  3 18:19 ..
-rw-rw-r-- 1 myuser myuser    0 dic  3 17:42 00-HI-ENCRYPTED

[agmartin2]

By the way, regarding the use_ino problem, I see that in fuse3 include/fuse_lowlevel.h file there are comments in the area of fuse_session and fuse_session_new_fn definitions about use_ino not being supported in low level session. Do not know if this may be related.

[blshkv]

@agmartin2 dont waste your time. encfs is dead. Use other alternatives like gocryptfs.