Re: ntfs ubc_uiomove error

[Antti Kantee <pooka%NetBSD.org@localhost>]

On Tue, Sep 01, 2009 at 01:20:35PM +0100, Patrick Welche wrote:
> On Thu, Aug 27, 2009 at 01:20:11PM -0400, Christos Zoulas wrote:
> > On Aug 27,  4:21pm, dholland-current%netbsd.org@localhost (David Holland) 
> > wrote:
> > -- Subject: Re: ntfs ubc_uiomove error
> > 
> > | On Wed, Aug 26, 2009 at 10:22:24AM -0400, Christos Zoulas wrote:
> > |  > | I was _reading_ from ntfs vnd, and writing to /tmp (tmpfs or ffs, 
> > not sure
> > |  > | which now...) The surprise above was that cat worked where cp 
> > failed(?!)
> > |  > 
> > |  > Ah, ok... Now I understand :-) NTFS does not implement mmap, and cp 
> > uses
> > |  > mmap.
> > | 
> > | Shouldn't cp be adjusted so it'll fall back to read()?
> > 
> > Last I checked it failed differently, mmap(2) returned an invalid
> > pointer and the the application SEGV'd. Clearly this behavior has
> > now changed :-) I don't think that the error is being passed up
> > properly, still.
> 
> A quick look says that cp falls back to read(), and ntfs uses genfs' mmap
> which just returns zero, which means success, no? Rather than return
> not implemented...

An equally quick look tells ntfs supports mmap and therefore should
return success there.  I fixed the bug preventing genfs_compat_getpages()
from working and you should get better results with cp(1) now.