On 12/24/11 11:34, Roland Smith wrote:
On Sat, Dec 24, 2011 at 09:57:38AM +1000, Da Rock wrote:
FreeBSD be default already does buffering in the VFS layer (unless you turn
that off). I don't think that adding more buffering would help. It might even
make matters worse. If data is buffered and not immediately written to the USB
stick, it will show no activity. This might even give the user a false
impression it is finished...
That there is exactly the problem. Any way to prevent that though?
Yes. Using the '-o sync' option with mount. To the best of my understanding
that means that a write action will be executed immediately and that write(2)
will not return until it is finished.
<snip>
It is a heck of a problem. Distributed filesystems like Coda
[http://en.wikipedia.org/wiki/Coda_%28file_system%29] might offer a partial
solution. But it doesn't seem like they're being widely used.
The only sane way to handle this is for the application to get an error from
the next write reporting that the filesystem has disappeared. Which it should
then report to the user because that's the person that pulled the plug, so to
speak.
Man, What a mess! The real solution is to keep the light flashing until
all the data written to disk.
The abovementioned -o sync option will do that for you.
The price is that the write syscall will block until it is finished. This
might render the application performing it unresponsive during that time.
Bingo! Thats exactly the issue. Taking a look at FUSE I can see there
are calls that FUSE_FSYNC which I would say is the closest to what is
needed. The question is are they implemented in say fuse_msdosfs (I may
have just barfed that, but you get the idea) and maybe some others? That
way the filesystem is treated as normal, the app writes and can return
happily, and FUSE takes the responsibility to ensure the data continues
to write until finished instead of holding it in buffer for an
indeterminate period leaving the user thinking alls well (ergo, the
light continues flashing).
Failing that, how does one ensure data is synced manually? Is there a
call that could be run periodically at a shorter, more regular intervals?
Either that or something that just tells the disk to keep flashing its
little led while the buffer is not empty :)
All the mayor players in this drama, hal, policykit and dbus are maintained by
gnome@. In practice that _might_ mean that no single person cares enough to
care and feed them.
Ahh. Now that may explain some things. But by your meaning are you
talking about the software development itself or the developers? LOL
What I mean here is that there are no dedicated maintainers for the FreeBSD
ports. When that is the case, that maintainer address of a port is usually
that of a relevant mailing list. That way problems with such a port at least
get the attention of people with relevant interests.
As for the original developers, who knows? IMNSHO their solutions look overly
complex,
Indeed. At least from a BSD perspective, anyway. Linux is whole other
animal... family, xenotype even :) Too many chiefs and not enough
indians over there I reckon.
In my mind reading your comments I was imagining some starving,
malnourished developers trying to build these complex software designs
surrounded by some Doc Brown-like inventions in the dungeons of Novell
or something... :) Hell, it could even be the dungeons of Microsoft,
with Gates uttering "Now I'll get you all, and your little dog too.." :D
And I'm kind of allergic to desktop
environments. Over the years I've tried several window managers, but I keep
coming back to good old FVWM. :-)
Me too- can't get away from icewm.
_______________________________________________
[email protected] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "[email protected]"
