[CRIU] [PATCH 0/2] introduce --skip-mnt cli option

[Pavel Emelyanov]

On 03/31/2015 07:24 PM, Oleg Nesterov wrote:
> Hello.
> 
> To clarify: this depends on "[PATCH 0/5] parse_mountinfo: minor fixes
> and cleanups" series I sent before. But this is completely orthogonal
> to FSTYPE__AUTO feature we discuss in another thread, this series can
> be applied (or nacked ;) independently.
> 
> With this patch (plus FSTYPE__AUTO hacks) I can finally dump/restore
> 
> 	perl -e 'close STDIN; close STDOUT; close STDERR; sleep'
> 
> running with the unshared mnt ns on rhel-7 which has a lot of mounts
> by default.
> 
> "--enable-fs all" helps to dump xfs/configfs/more, but it obviously
> can't handle "autofs". Now I can pass --skip-mnt /proc/sys/fs/binfmt_misc
> to make it work.

Is this because there are some entries in binfmt_misc?

> And I am very sure that mysqld (the actual target I want to c/r) doesn't
> need this mountpoint, so I think I have the use-case ;)
> 
> Please review. Again, again, I mostly mean the feature itself.
> 
> 
> 
> As for implementation, see the changelog in 2/2. I think that we should
> filter out the mounts at "parse" time, but I am not sure you will like
> this.
> 
> Say, if I pass "--skip-mnt /sys" then mnt_build_tree() right after that
> correctly fails with "Mountpoint w/o parent" and this is what we want.
> So at least I think we should not postpone the filtering "too much".
> 
> TODO:
> 
> 	- support this option in RPC.
> 
> 	- perhaps --skip-fsname makes sense too. Or, better, we can
> 	  teach add_skip_mount() to check "fs=" or "mnt=" at the start
> 	  of optarg.
> 
> Oleg.
> 
> .
>