read() called twice for /proc files

read() called twice for /proc files

[Tim Hockin]

woops!  sent too early!

Is there a general solution to the case of read() being called minimum
twice for a file in /proc?  I have a small file in /proc, whose data takes
a fair time to generate.  My read() handler gets called once for the data,
and once so I can return 0 to terminate read().

This results in the actual read taking twice as long.  Perhaps I am missing
something...

What if the proc generic stuff used file->private_data as an EOF flag.  It
seems really bizarre that the read loop loops until return 0 or eof is
set.  We promptly throw away the EOF information that the read() handler
set.

Would it break anything?  Is there something I am not seeing in the larger
picture?

-- 
Tim Hockin
Systems Software Engineer
Sun Microsystems, Cobalt Server Appliances
thockin@sun.com

Re: read() called twice for /proc files

[Jonathan Lundell]

At 6:34 PM -0700 2001-09-24, Tim Hockin wrote:
>Is there a general solution to the case of read() being called minimum
>twice for a file in /proc?  I have a small file in /proc, whose data takes
>a fair time to generate.  My read() handler gets called once for the data,
>and once so I can return 0 to terminate read().

I'm guessing you mean that the user is calling read once (and not a 
second time to get the 0/eof return), and that proc_file_read() is 
calling your read_proc twice. If that's the case, you can just set 
the eof flag on the first call (on which you return the data), and 
you won't be called again unless the user reads again.

And of course, as David Miller suggests, you can check the offset 
first, before generating data.
-- 
/Jonathan Lundell.

read() called twice for /proc files

[Tim Hockin]

Is there a general solution to the case of read() being called minimum
twice for a file in /proc?  I have a small file in /proc, whose data takes
a fair time to generate.  My read() handler gets called once for the data,
and once so I can return 0 to terminate read().

This results in the actual read taking twice as long.  Perhaps I am missing
something...

-- 
Tim Hockin
Systems Software Engineer
Sun Microsystems, Cobalt Server Appliances
thockin@sun.com

Re: read() called twice for /proc files

[David S. Miller]

   From: Tim Hockin <thockin@sun.com>
   Date: Mon, 24 Sep 2001 18:29:49 -0700
   
   This results in the actual read taking twice as long.  Perhaps I am missing
   something...

EOF isn't known until you return the zero.  You can watch the loff_t
arg to the read operation to see if it is at the end, and if so just
avoid retouching the device or whatever and go straight to returning
0.

Franks a lot,
David S. Miller
davem@redhat.com