[fuse-devel] blocksize used for writing

[John Berthels <jjberthels@gm...>]

Hi there.

Thanks for fuse, it's a great piece of code. I'm using the perl
bindings to provide a fs wrapper around an HTTP/WebDAV server and got
to what I think it a workable system.

I currently do no buffering and translate each call to WRITE to an
HTTP PUT (with an appropriate content-range: header), which is simple
and works well.

Unfortunately, I seem to only be getting WRITEs in chunks of 4k -
which really hurts performance.

(I think) I'm returning a larger value in STATFS (and the same value
in GETATTR), but if even I have a process doing large writes, they are
still chunked down to 4k by the time my fuse code sees them:

strace dd if=large_file of=/mnt/tmp delete-me bs=1024000
...
read(0, "`0\30\f\6\203\301`0\30\f\6\203\301`0\30\f\6\203\301`0\30"...,
1024000) = 1024000
write(1, "`0\30\f\6\203\301`0\30\f\6\203\301`0\30\f\6\203\301`0\30"...,
1024000) = 1024000
read(0, "\313\355N\334\276\227\333\303\271=\231\333Oq{.\267\347"...,
1024000) = 1024000
write(1, "\313\355N\334\276\227\333\303\271=\231\333Oq{.\267\347"...,
1024000) = 1024000
...

but in my fuse code I see:

2007/07/26 09:34:02 WRITE: /delete-me [10117120, 4096]
2007/07/26 09:34:02 WRITE: /delete-me [10121216, 4096]
2007/07/26 09:34:02 WRITE: /delete-me [10125312, 4096]
2007/07/26 09:34:02 WRITE: /delete-me [10129408, 4096]

Can I ask if there is any other way to influence the blocksize which I
see on WRITE?

Or will I have to go the route of buffering to accomplish this?

jb

[fuse-devel] Problem with python, fuse and sqlite

[BrokenClock <brokenclock@fr...>]

Re: [fuse-devel] fuse works on command line but not from hotplug

[Leif Johnson <leif.t.johnson@gm...>]

On 7/24/07, Leif Johnson <leif.t.johnson@...> wrote:
>
>
>
> Okay.  I've been doing further testing, and hotplug can definitely run
> commands from that directory.  I haven't been able to make it return 127
> again, which means I probably had something out of place on that run.
> However it still isn't working.  The best I've been able to do is to get it
> to return 0 (indicating a successful mount), but the directory does not get
> mounted and it leaves a zombie process behind.
>
> leif
>
>
>
Hi again,

The platform I'm working on uses BusyBox v1.1.3.  It does not have a
/etc/mtab.  Note that /etc/mtab is NOT symlinked to /proc/mounts, /etc/mtab
does not exist.

fuse-2.6.5:
First, I tested using the '-d' option with fuse-2.6.5.  It worked!  Of
course, this isn't desirable as it hotplug waits for the fuse program to
finish before continuing.

fuse-2.7.0
Noting the lack of /etc/mtab, and the information from the thread about
fuse-2.7.0 on busybox, I made a crude hack to get 2.7.0 working by replacing
calls to /bin/mount with exit(0);

Note: I doubt my hack is the correct way to do it, as I don't seem to be
able to unmount fuse filesystems using it.

 It also worked using the -d option from hotplug.  Run normally from
hotplug, it returns successfully, and mount claims that the target directory
is mounted as a fuse filesystem.  However, if I do an ls of the directory, I
get the following error:
=============================================
mnt/tmpfs/media # mount
<snip>
fuse on /mnt/tmpfs/media/ipod type fuse
(rw,nosuid,nodev,user_id=0,group_id=0)
</snip>
/mnt/tmpfs/media # ls
ls: ./ipod: Transport endpoint is not connected
SD-card      USB          network      upnpBrowser
/mnt/tmpfs/media # ps | grep hello
 1505 root            Z < [hello]
===============================================
You can see that the hello program has become a zombie process.

Browsing the fuse-devel archives led me to believe that this means that the
fuse program has lost communication with the fuse kernel module.  Is this
correct?  Where should I start looking.

thanks,
leif