Hi all, On 14 October 2017 at 01:20, Wojtek Porczyk <w...@invisiblethingslab.com> wrote:
> -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA256 > > On Sat, Oct 14, 2017 at 02:06:43AM +0200, Wojtek Porczyk wrote: > > On Sat, Oct 14, 2017 at 01:40:45AM +0200, Marek Marczykowski-Górecki > wrote: > > > On Sat, Oct 14, 2017 at 01:21:58AM +0200, Wojtek Porczyk wrote: > > > > On Sat, Oct 14, 2017 at 01:20:13AM +0200, Marek Marczykowski-Górecki > wrote: > > > > > You can't reload partition table while it is used (something from > there > > > > > is mounted). At least Linux doesn't support it. > > > > > > > > Just tested with loop device, fdisk, and ext4. Looks like it works. > Unless it > > > > works only with loop? > > > > > > No, it doesn't work, at least not directly: > > > > > (snip) > > > > The kernel still uses the old table. The new table will be used at > the next reboot or after you run partprobe(8) or kpartx(8). > > > > > > I wonder on what conditions partprobe would work. And how would mounted > > > filesystem react for it. > > > > I'm calling bullshit on that. Coredump follows: > > (...) > > > [root@qubes-dev tmp]# partprobe /dev/loop0 > > [root@qubes-dev tmp]# echo $? > > 0 > > (...) > > Look at strace of blockdev --rereadpt and partprobe. The difference is that > blockdev (and I assume fdisk also, since both are from util-linux) fires > ioctl(fd, BLKRRPART), but partprobe (from GNU parted) does something funny > to > /sys, which I didn't try to understand, but seems to work. > > My guess is that this is a missing feature in kernel, which parted works > around. > I am a Qubes user, and coincidentally, the original author of partprobe (and parted). I haven't looked at partprobe/parted since 2005. The code has changed a lot since then, but let me do my best... The BLKRRPART ioctl is no good because it can't accommodate busy block devices at all, i.e. resizing partition 1 when partition 2 on the same disk is mounted. Instead, Parted primarily uses the BLKPG family of ioctl to inform the kernel of partition table changes. (It also has "new" support for the device mapper -- but that's probably not relevant here.) You can read about the BLKPG ioctl in /usr/include/linux/blkpg.h. Since 2012, the linux kernel supports a new BLKPG feature to do online partition resizing, i.e. telling the kernel to modify a mounted partition. I think this is what is being used here. The relevant Parted code is in the function linux_disk_commit(), which calls _disk_sync_part_table() and _blkpg_resize_partition() inside http://git.savannah.gnu.org/cgit/parted.git/tree/libparted/arch/linux.c If the BLKPG ioctl fails, then partprobe/parted will throw an exception and tell you about it. Wojtek: what part of your shell transcript was unexpected? It looked like everything worked to me. Cheers, Andrew -- You received this message because you are subscribed to the Google Groups "qubes-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to qubes-devel+unsubscr...@googlegroups.com. To post to this group, send email to qubes-devel@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/qubes-devel/CAAXZBWL2p_84BdurZrYSG%2B_pZwv65qyUWRPKpG-p2OrP0zJ6uw%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
