On Thu, Jul 22, 2021 at 09:33:59AM -0700, Daeho Jeong wrote:
> On Thu, Jul 22, 2021 at 9:26 AM Eric Biggers <[email protected]> wrote:
> >
> > On Thu, Jul 22, 2021 at 09:17:54AM -0700, Eric Biggers wrote:
> > > > I think it matches with FIEMAP_EXTENT_UNWRITTEN. Otherwise, we should
> > > > shorten the last extent like below.
> > > >
> > > >         fe_logical=0    fe_physical=16384  length=4096
> > > >         fe_logical=4096 fe_physical=81920  length=4096
> > > >         fe_logical=8192 fe_physical=147456 length=4096
> > > >
> > > >
> > > >   Unwritten extent - the extent is allocated but its data has not been
> > > >   initialized.  This indicates the extent's data will be all zero if 
> > > > read
> > > >   through the filesystem but the contents are undefined if read 
> > > > directly from
> > > >   the device.
> > >
> > > Well, as I said before, using UNWRITTEN isn't appropriate because it 
> > > indicates
> > > that the data is all zeroes, which in this case it is not.  Similarly, 
> > > reporting
> > > a hole isn't appropriate because it also indicates that the data is all 
> > > zeroes
> > > and also that it has no space allocated on-disk at all.
> > >
> > > I think we should just over-report the physical length of the last extent 
> > > in the
> > > cluster, which is what btrfs does...
> > >
> >
> > Also keep in mind that as far as fiemap is concerned, when 
> > FIEMAP_EXTENT_ENCODED
> > is set (indicating that the data is compressed or encrypted), then reading 
> > the
> > data from disk will have "undefined results"; see
> > Documentation/filesystems/fiemap.rst.  As such, when someone decides to do 
> > so
> > anyway (which is necessary for encryption testing), they really need to know
> > *exactly* what they're doing.  So I think it's less bad to bend the rules on
> > extents where FIEMAP_EXTENT_ENCODED is already set.
> >
> > In contrast, your suggestion would incorreectly report some parts of the 
> > file as
> > standard extents (or holes) without FIEMAP_EXTENT_ENCODED, so it would be
> > expected that the standard meaning would apply to those parts.
> >
> > - Eric
> 
> I also think it is okay with the FIEMAP_EXTENT_ENCODED flag.
> It is actually all zero in a view of the filesystem internal and still
> undefined if read directly from the device.
> If we remove this extent, it might be confusing to understand the layout of 
> it.

FIEMAP_EXTENT_ENCODED | FIEMAP_EXTENT_UNWRITTEN is contradictory, so that
doesn't seem like a good option.  And no, the range is *not* zero when read from
the file by userspace, which is what UNWRITTEN is supposed to indicate.

Compressed data is terminated by an end-of-stream marker, so it is possible to
decompress even if extra data is appended to it.  So that's another reason why I
feel that my suggestion is not as bad as the other options.

> Plus, I think we need to remove the last extent, when we return back
> the block to the free space pool to filesystem using releasing
> reserved space ioctl.

This seems to be a filesystem implementation detail.  Again, FS_IOC_FIEMAP is
just about returning the actual extent mapping, not about other filesystem
implementation details.

- Eric


_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Reply via email to