On Thu, 8 Jun 2017, Al Viro wrote:
On Mon, Jun 05, 2017 at 04:49:36AM +0100, Al Viro wrote:
Grrr... OK, I'll put together a patch fixing that idiocy. As it is, rw mounts
of ufs on-disk ->i_blocks not updated and, AFAICS, can lead to disk space leaks
on inode deletion.
Turns out it's even worse than that, and some of the breakage in there is my fault.
I'll fix what I can, but for now I would suggest
* making UFS_FS_WRITE depend on BROKEN; it had always been experimental,
but since 2010 it had been actively buggering filesystems (aforementioned Jan's
bug + a bug of mine in 2015 series). I'll fix that stuff, but I'd expect
it to take a month or so, including serious testing. I have set the things up
for such testing (which has promptly caught all kinds of fun), but we are at -rc4
now, so it'll probably be the next cycle fodder, with interesting backporting
afterwards.
* for ->s_maxbytes, let's go with tree-imposed limit. Proper way of
dealing with ->i_blocks overflows is -ENOSPC when attempt to grab a block would've
caused such an overflow.
* If Evgeniy can resurface and take part in that fun, I'd be happy to help.
If not, well... guess I'll take the thing over until somebody willing to adopt it
comes along.
I am willing to test. I just turned on UFS_FS_WRITE for the very first
time running 4.12-rc4 and was able to copy a file of more than 2GB from
one r/o FreeBSD subpartition to another r/w FreeBSD subpartition.
So it is already looking pretty good.
I did notice an ->i_blocks bug where it was set to zero on the output file
instead of the actual block count. This showed up when I poped over to FreeBSD 11.0 and
did an fsck on the r/w subpartition.
