Re: [PATCH RESEND v4 6/8] vfs: Add get_vfsmount_sb() function to get vfsmount from a given sb.

[Al Viro <viro@ZenIV.linux.org.uk>]

On Fri, Jan 30, 2015 at 09:11:26AM +0800, Qu Wenruo wrote:
> For the mounted ro case, that's not a problem, since if one instance
> is mounted ro,
> other instances are also mounted ro, so that's not a problem.

Not really.

root@satch:~# cd /tmp/
root@satch:~# mkdir /tm/a
root@satch:~# mount --bind /tmp/ /tmp/a
root@satch:~# mount --bind -o remount,ro /tmp/a
root@satch:~# mkdir /tmp/b  
root@satch:~# mkdir /tmp/a/c
mkdir: cannot create directory '/tmp/a/c': Read-only file system
root@satch:~# stat /tmp/b /tmp/a/b
  File: '/tmp/b'
  Size: 4096            Blocks: 8          IO Block: 4096   directory
Device: 806h/2054d      Inode: 257537      Links: 2
Access: (0755/drwxr-xr-x)  Uid: (    0/    root)   Gid: (    0/    root)
Access: 2015-01-29 21:03:35.000000000 -0500
Modify: 2015-01-29 21:03:35.000000000 -0500
Change: 2015-01-29 21:03:35.000000000 -0500
 Birth: -
  File: '/tmp/a/b'
  Size: 4096            Blocks: 8          IO Block: 4096   directory
Device: 806h/2054d      Inode: 257537      Links: 2
Access: (0755/drwxr-xr-x)  Uid: (    0/    root)   Gid: (    0/    root)
Access: 2015-01-29 21:03:35.000000000 -0500
Modify: 2015-01-29 21:03:35.000000000 -0500
Change: 2015-01-29 21:03:35.000000000 -0500
 Birth: -
root@satch:~#

IOW, /tmp and /tmp/a bear the same filesystem (one from /dev/sda6), the former
is mounted r/w, the latter - r/o.

> >Assuming that your superblock is guaranteed to stay alive and usable for
> >whatever work you are trying to do, what's wrong with sb_want_write()?
> Did you mean change the function name and it's parameter to
> sb_want_write(sb) and sb_drop_write(sb).
> That looks much better.
> But I'm a little worried about just using sb_start_write() and
> s_readonly_remount/s_flags to do the protection,
> will it be enough?

Protection against what?  If superblock is r/w, it will stay r/w until you
do sb_drop_write(); if it's r/o, sb_want_write() will fail.  There might be
any number of vfsmounts over superblock; attempt to get write access via
vfsmount will succeed only of both the vfsmount and superblock are r/w -
mnt_want_write() does sb_want_write() and fails if that has failed.