Re: [Acl-devel] "Malformed access ACL" Issues with 2.2.53, 2.2.52 is fine

[Fabio Coatti]

On Monday 15 October 2018 21:21:24 CEST Andreas Grünbacher wrote:
> Am So., 14. Okt. 2018 um 14:38 Uhr schrieb Fabio Coatti
> 
> <address@hidden>:
> > Hi all, I'm having an issue with 2.2.53 version of acl.
> > Basically, with latest version acls it is not possible to set acl, I get
> > this> 
> > error back:
> > > setfacl -m g:cova:rwx 53
> > 
> > setfacl: 53: Malformed access ACL `group:cova:rwx,mask::rwx,other::r--':
> > Missing or wrong entry at entry 1
> 
> The output of getfacl (just other::r--) is already invalid, but
> getfacl relies on that the kernel gives it. I also cannot reproduce
> locally.
> 

Thanks for the answer and sorry for my late reply. So, let's see:

> What's the output of "getfattr -n system.posix_acl_access -e hex",
> i.e., the raw acl value?

raw value for 52 is the following:
getfattr -n system.posix_acl_access -e hex 52
# file: 52
system.posix_acl_access=0x0200000001000600ffffffff04000400ffffffff08000700e803000010000700ffffffff20000400ffffffff

Indeed, the 53 gives me a weird answer:
getfattr -n system.posix_acl_access -e hex 53
53: system.posix_acl_access: No such attribute

I've only recompiled acl pacakge, no other changes whatsoever. 

> What patches does gentoo have applied, and if any, can you reproduce
> with a stock getfacl?

As far I can see, no patches are applied by ebuild. However, this remark made 
me to verify if any of the compilation/linking flags can have an impact and 
indeed LTO was the culprit.
I'm using quite a few of "optimizations" (mostly to experiment a bit) and my 
usual list is like this, basically most of my system is compiled with it:

-march=native -O3 -fgraphite-identity -floop-nest-optimize -ftree-loop-
distribution -flto=4 -fuse-linker-plugin -pipe -fpie -fpic -fstack-protector-
strong

If i remove -flto=4 the get/set facl command behaves properly, version 2.2.53.

Maybe some change between 52 and 53 caused the LTO side effect. Now I have a 
working 53 version; however if you find worthwile to investigate this change 
and you need more data, just let me know. 

Right now my setup is 
gcc --version
gcc (Gentoo 8.2.0-r3 p1.3) 8.2.0
ld --version
GNU gold (Gentoo 2.31.1 p3 2.31.1) 1.16

(I switched to gold after detecting the issue, so maybe bfd/gold have the same 
behaviour here)
Also kernel version seems not related.

> Is this on a weird filesystem?

Nope, or at least not only. I tried this on ext4, btrfs and tmpfs.

> 
> Thanks,
> Andreas

Many thanks for the answer!
--
Fabio