Re: [PATCH 1/1] security: introduce fs caps

[KaiGai Kohei <kaigai@ak.jp.nec.com>]

Hi, Serge.

Serge E. Hallyn wrote:
> Quoting Bill O'Donnell (billodo@sgi.com):
>> 8102  execve("/sbin/setfcaps", ["setfcaps", "cap_net_raw=ep", "/bin/ping"], [/* 67 vars */]) = 0
     - snip -
>> 8102  capget(0x19980330, 0, {0, 0, 0})  = -1 EINVAL (Invalid argument)
> 
> I don't see why this capget is returning -EINVAL.  In fact I don't see
> why it happens at all - cap_inode_setxattr would check
> capable(CAP_SYS_ADMIN), but setxattr hasn't been called yet.  Looking at
> both libcap and setfcaps.c, I don't see where the capget comes from.
> 
> As for the -EINVAL, kernel/capability.c:sys_capget() returns -EINVAL if
> the _LINUX_CAPABILITY_VERSION is wrong - you have 0x19980330 which is
> correct - if pid < 0 - but you send in 0 - or if security_capget
> returns -EINVAL, which cap_capget (and dummy_capget) don't do.
> 
> Kaigai, do you have any ideas?

Bill said that he uses SLES10/ia64, so the version of libcap is different
from Fedora Core's one. 'libcap-1.92-499.4.src.rpm' is bandled.

Then, I found a strange code in libcap-1.92-499.4.src.rpm.

The setfcaps calls cap_from_text() which is defined in libcap to parse
the command line argument. It has the following function call chains:

cap_from_text()
   -> cap_init()
     -> _libcap_establish_api()

---- the definition of _libcap_establish_api() ----
void _libcap_establish_api(void)
{
     struct __user_cap_header_struct ch;
     struct __user_cap_data_struct cs;

     if (_libcap_kernel_version) { <-- _libcap_kernel_version is 0 initially.
         _cap_debug("already identified kernal api 0x%.8x",
                    _libcap_kernel_version);
         return;
     }

     memset(&ch, 0, sizeof(ch));
     memset(&cs, 0, sizeof(cs));

     (void) capget(&ch, &cs);  <-- (?)

     switch (ch.version) {

     case 0x19980330:
         _libcap_kernel_version = 0x19980330;
         _libcap_kernel_features = CAP_FEATURE_PROC;
         break;

     case 0x19990414:
         _libcap_kernel_version = 0x19990414;
         _libcap_kernel_features = CAP_FEATURE_PROC|CAP_FEATURE_FILE;
         break;

     default:
         _libcap_kernel_version = 0x00000000;
         _libcap_kernel_features = 0x00000000;
     }

     _cap_debug("version: %x, features: %x\n",
                _libcap_kernel_version, _libcap_kernel_features);
}
---------------------------------------------------

capget() is called from _libcap_establish_api() with full-zeroed
__user_cap_header_struct object at first time.
The result of this, sys_capget() in kernel will return -EINVAL.
(Why did strace say the first argument is 0x19980330?)

Probably, Bill didn't update libcap.so.

But I can't recommend Bill to update libcap immediately.
As Hawk Xu said, it may cause a serious problem on the distro
except Fedora Core 6. :(

I have to recommend to use 'fscaps-1.0-kg.i386.rpm' now.
It includes the implementation of interaction between application and xattr.
(Of couse, it's one of the features which should be provided by libcap.)

Thanks,
-- 
Open Source Software Promotion Center, NEC
KaiGai Kohei <kaigai@ak.jp.nec.com>