support overmount

[bAndie91]

support overmount

[sloonz]

I have a (mostly untested) implementation in the at branch, would you mind testing it a bit on your end ?

[bAndie91]

hi sloonz, I also developed this feature a while ago - sorry i have not uploaded it earlier. see my branch @development/overmount. I and my family is using this version for 3 months to accelerate web browser (Vivaldi), ~/.config/vivaldi is overmounted on itself and sqlite journal files are redirected to memory fs. I have not noticed any glitch: no mem-leak, no hanging syscalls.

There are more changes on my branch, but the one in this scope is commit 12c12c3.
My approach operates not with openat() family functions, but (1) chdir to orig_fs and (2) access files by relative path. Because the programm does not chdir elsewhere, the cwd serves like orig_fd in your code. I'm not sure which is the better. Yours is more explicite. Mine saves a few calls in rewrite_opendir, rewrite_truncate, rewrite_statfs (poorly we dont have opendirat() in std lib).

I also removed the constraint which prohibits config file to be under orig_fs (recursion loop is avoided by the relative-path feature).

[bAndie91]

rewrite_chown():

fchownat() with flags=0 dereferences synlinks, howerver the replaced lchown() does not.
I suggest AT_SYMLINK_NOFOLLOW.

the same for rewrite_setxattr() and rewrite_getxattr() and rewrite_listxattr() and rewrite_removexattr():

suggest openat(..., O_RDONLY | O_NOFOLLOW) to operate on symlinks like lsetxattr() (and lgetxattr() and llistxattr() and lremovexattr()) did.

[sloonz]

Good catch on chown/chmod

For *xattr unfortunately there's no way to do that AFAIK. O_NOFOLLOW on openat does not do what you think, it just returns an error if the target is a symlink.

Your chdir approach can't work with FUSE 3 : libfuse chdir into / at the end of the initialization process.

I’m going to merge autocreate/at*/abmalloc into develop, do some more polishing today, and let it run on my machine for a few days (maybe weeks) before merging into master again.

[bAndie91]

hi @sloonz

1. afaik chmod is not affected as there is no lchmod

2. regarding to *xattr, IMO it should be disallowed on symlinks in order to preserve semantics. (eg. let's do O_NOFOLLOW and return error if it fails)

3. chdir-approach

   1. what do you think, does it worth to workaround fuse3 limitation, so do chdir on the first actual call to rewritefs instead to before libfuse initialization? rece condition may be avoided by saving cwd before libfuse and fchdir into it later.

   2. my chdir-approach indeed works only in foreground mode

4. note, auto_mkdir feature (239d802) is not thought through well by me; i committed it to the overmount branch only by lazyness; I'm not fully sure is it a good idea or is it implemented well.

thanks for consideration

[sloonz]

Merged in master