Re: [GIT PULL] namespace updates for v3.17-rc1

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Am 21.08.2014 06:53, schrieb Eric W. Biederman:
> The bugs fixed are security issues, so if we have to break a small
> number of userspace applications we will.  Anything that we can
> reasonably do to avoid regressions will be done.
> 
> Could you please look at my user-namespace.git#for-next branch I have a
> fix for at least one regresion causing issue in there.  I think it may
> fix your issues but I am not fully certain more comments below.

I'll run this on my LXC testbed today.

>>         /*
>>          * We can't immediately set the MS_RDONLY flag when mounting filesystems
>>          * because (in at least some kernel versions) this will propagate back
>>          * to the original mount in the host OS, turning it readonly too. Thus
>>          * we mount the filesystem in read-write mode initially, and then do a
>>          * separate read-only bind mount on top of that.
>>          */
>>         bindOverReadonly = !!(mnt_mflags & MS_RDONLY);
>>
>>         VIR_DEBUG("Mount %s on %s type=%s flags=%x",
>>                   mnt_src, mnt->dst, mnt->type, mnt_mflags & ~MS_RDONLY);
>>         if (mount(mnt_src, mnt->dst, mnt->type, mnt_mflags &
>> ~MS_RDONLY, NULL) < 0) {
>>
>> ^^^^ Here it fails for sysfs because with user namespaces we bind the
>> existing /sys into the container
>> and would have to read out all existing mount flags from the current /sys mount.
>> Otherwise mount() fails with EPERM.
>> On my test system /sys is mounted with
>> "rw,nosuid,nodev,noexec,relatime" and libvirt
>> misses the realtime...
> 
> Not specifying any atime flags to mount should be safe as that asks for
> the default atime flags which for remount I have made the default atime
> flags the existing atime flags.  So I am scratching my head a little on
> this one.

Okay, let me find out why exactly libvirt gets a EPERM here.
Maybe there are more odds hidden.

>>
>>             virReportSystemError(errno,
>>                                  _("Failed to mount %s on %s type %s flags=%x"),
>>                                  mnt_src, mnt->dst, NULLSTR(mnt->type),
>>                                  mnt_mflags & ~MS_RDONLY);
>>             goto cleanup;
>>         }
>>
>>         if (bindOverReadonly &&
>>             mount(mnt_src, mnt->dst, NULL,
>>                   MS_BIND|MS_REMOUNT|MS_RDONLY, NULL) < 0) {
>>
>> ^^^ Here it fails because now we'd have to specify all flags as used
>> for the first
>> mount. For the procfs case MS_NOSUID|MS_NOEXEC|MS_NODEV.
>> See lxcBasicMounts[].
>> In this case the fix is easy, add mnt_mflags to the mount flags.
> 
> That has always been a bug in general because remount has always
> required specifying the complete set of mount flags you want to have.
> 
> That fact that flags such as nosuid are now properly locked so you can
> not change them if you are not the global root user just makes this
> obvious.
> 
> Andy Lutermorski has observed that statvfs will return the mount flags
> making reading them simple.

Thanks for the clarification, I'll create a fix for libvirt.

Thanks,
//richard
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html




[Index of Archives]     [Linux Ext4 Filesystem]     [Union Filesystem]     [Filesystem Testing]     [Ceph Users]     [Ecryptfs]     [AutoFS]     [Kernel Newbies]     [Share Photos]     [Security]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux Cachefs]     [Reiser Filesystem]     [Linux RAID]     [Samba]     [Device Mapper]     [CEPH Development]
  Powered by Linux