DragonFlyBSD

I've identified the issue. In devfs getattr vnop I'm doing:
vap->va_bytes = vap->va_size = sizeof(struct devfs_node);
and kern_lseek then does:
new_offset = offset + vattr.va_size;

I'm not quite sure how to solve this. Should the size be set to the physical
size if the device is a disk type? This would also result in fstat to return
the physical size for a disk.

Unless anyone considers it to always set va_size and st_size to the disk size
for a disk-type device in the next few hours or so, I'll do it this way.

Cheers,
Alex Hornung

oops; I meant "unless anyone considers it to be incorrect to always set
va_size..."

For reference:
http://www.opengroup.org/onlinepubs/9699919799/basedefs/sys_stat.h.html
So the standard does not specify what st_size has to be for devices.

Any opinion on this matter?

Cheers,
Alex Hornung

On Thu, 17 Dec 2009 14:08:49 +0000
"Alex Hornung \(via DragonFly issue tracker\)"
<bugs@crater.dragonflybsd.org> wrote:

Alex Hornung <ahornung@gmail.com> added the comment:

I've identified the issue. In devfs getattr vnop I'm doing:
vap->va_bytes = vap->va_size = sizeof(struct devfs_node);
and kern_lseek then does:
new_offset = offset + vattr.va_size;

I'm not quite sure how to solve this. Should the size be set to the
physical size if the device is a disk type? This would also result in
fstat to return the physical size for a disk.

The one place I know it is used is in ddrescue where it is
expected to return the amount of data that will be read by reading the
device until EOF, which for a DVD would be the amount of data written on
the media but for a hard disk would be the physical size of the disk.

Alex Hornung (via DragonFly issue tracker) wrote:

Alex Hornung <ahornung@gmail.com> added the comment:

For reference:
http://www.opengroup.org/onlinepubs/9699919799/basedefs/sys_stat.h.html
So the standard does not specify what st_size has to be for devices.

Any opinion on this matter?

If we have the size of the device in question (i.e. disk), it won't hurt to set it. Else maybe set 0 to avoid confusion? Alternatively seek to EOF could fail.

cheers
simon

Fixed in commit 2d0767552b6288b28a2ec45d633cfd93fa58ca64.
Steve, just fyi: you need to change your %ld formatter to %lld, as off_t is 64
bit wide.
I've tested this with several hard disks and partitions and a DVD. It seems to
work, but I'll wait for Steve's feedback before closing this issue.

Cheers,
Alex Hornung

On Thu, 17 Dec 2009 15:36:28 +0000
"Alex Hornung \(via DragonFly issue tracker\)"
<bugs@crater.dragonflybsd.org> wrote:

Fixed in commit 2d0767552b6288b28a2ec45d633cfd93fa58ca64.

Works like a charm - Thank you very much.

Steve, just fyi: you need to change your %ld formatter to %lld, as off_t
is 64 bit wide.

Yeah I know - it was just a quick lash up to find out why ddrescue
was misbehaving, I'll be deleting it now.

I've tested this with several hard disks and partitions and a DVD. It
seems to work, but I'll wait for Steve's feedback before closing this
issue.

Go ahead and close it.

:I've identified the issue. In devfs getattr vnop I'm doing:
:vap->va_bytes =3D vap->va_size =3D sizeof(struct devfs_node);
:and kern_lseek then does:
:new_offset =3D offset + vattr.va_size;
:
:I'm not quite sure how to solve this. Should the size be set to the physica=
:l=20
:size if the device is a disk type? This would also result in fstat to retur=
:n=20
:the physical size for a disk.
:
:Unless anyone considers it to always set va_size and st_size to the disk si=
:ze=20
:for a disk-type device in the next few hours or so, I'll do it this way.
:
:Cheers,
:Alex Hornung

BSDs in the past have not supported st_size on disk devices, but
    linux has.

I personally think that if we can easily support st_size for disk devices
    that we should.

-Matt

I don't think va_size should ever be set to the size of an internal
structure. That information is not meaningful to consumers / stat.
It should either be something meaningful, like the size of the
disk (partition/slice/etc) or 0.

For softlinks va_size is the length of the softlink string.

-Matt
                    Matthew Dillon 
                    <dillon@backplane.com>

Note that %lld is 128 bits on 64-bit machines. The best thing to do is
to cast to size_t or ssize_t:

"... %ju ...", ...  (size_t)blah
    "... %jd ...", ...  (ssize_t)blah

-Matt
                    Matthew Dillon 
                    <dillon@backplane.com>

Ugh. never mind. long long is 64 bits. But use size_t anyway.

-Matt
                    Matthew Dillon 
                    <dillon@backplane.com>