On Thu, 19 Apr 2012, Karel Zak wrote: > On Thu, Apr 19, 2012 at 11:07:55AM +0200, Lukas Czerner wrote: > > Some mount flags can conflict with each other so they can not be > > handled together. Currently when conflicting flags are specified, > > some of them are silently ignored putting user in believe that > > they was handled correctly. > > Unfortunately, it's not so simple ;-) > > > - if (flags & MS_REMOUNT) > > + if (flags & MS_REMOUNT) { > > retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, > > data_page); > > - else if (flags & MS_BIND) > > + flags &= ~MS_REMOUNT; > > This is incorrect, the flags may also include many others flags. For > example MS_REMOUNT|MS_BIND|MS_RDONLY is valid (see do_remoun() code). ah, right we can specify more flags to be consumed by do_remount and others. Well, then we can just mask them all out like this: flags &= ~(MS_REMOUNT | MS_BIND ); > > And it's normal that for "mount -o remount" the mount command reads > flags from mtab/fstab so it includes for example MS_RELATIME, ... Yes, but MS_RELATIME is masked out from flags and it is used in mnt_flags which are handled by remount. That makes me wonder why we even have MS_RELATIME flag, since it is not used anywhere. > > > + } else if (flags & MS_BIND) { > > retval = do_loopback(&path, dev_name, flags & MS_REC); > > - else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) > > + flags &= ~MS_BIND; > > what about MS_REC ? Ok, we should put MS_REC into the mix since the do_remount does not use it. flags &= ~(MS_BIND | MS_REC); > > > + } else if (flags & (MS_SHARED | MS_PRIVATE | > > + MS_SLAVE | MS_UNBINDABLE)) { > > retval = do_change_type(&path, flags); > > - else if (flags & MS_MOVE) > > + flags &= ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE); > > what about MS_REC ? flags &= ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE | MS_REC); > > Note that do_change_type() already checks for unexpected flags and > returns -EINVAL if more flags are specified. I am not sure how this is related. Each of those function should probably do that but do_change_type() is actually the only one doing so. > > > + } else if (flags & MS_MOVE) { > > retval = do_move_mount(&path, dev_name); > > - else > > + flags &= ~MS_MOVE; > > + } else > > retval = do_new_mount(&path, type_page, flags, mnt_flags, > > dev_name, data_page); > > + > > + flags &= (MS_REMOUNT | MS_BIND | MS_SHARED | MS_PRIVATE | > > + MS_SLAVE | MS_UNBINDABLE | MS_MOVE); flags &= (MS_REMOUNT | MS_BIND | MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE | MS_MOVE | MS_REC); > > + > > + if (!retval && flags) > > + printk(KERN_WARNING "%s(%u): (%s -> %s) Conflicting mount flags" > > + " specified. These flags has been " > > + "ignored: %#.8lx\n", __func__, current->pid, > > + dev_name, dir_name, flags); But then, we'll might ignoring mnt_flags silently since they are masked out earlier. Well, maybe we could mas them out only do_remount() and do_new_mount(). > Karel Thanks! -Lukas -- 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