Thomas Hood wrote:

> Unfortunately, your method #3 conflicts with methods #0 through #2,
> which exhaust the range of possible values that may be returned
> in *start.  Any value greater than buffer is regarded as being
> "within the buffer".


I guess I don't understand the conflict.

There is case #0 where start is NULL.  In this case start is computed in 
proc_file_read as page + *ppos so (unsigned long)start < PROC_BLOCK_SIZE 
is NOT true and so start is used as before.

In case #1 (unsigned long)start < PROC_BLOCK_SIZE so it will grab the 
data from page and use start as the length to copy as it did before.

And finally cases #2 and #3 are the same:  use start as an explicit data 
address.  It doesn't matter whether this points into page or if it is 
space provided by read_proc (which is why I suggested not even 
mentioning a "case #3").


Did you get a chance to read the patch?  I'll attach it again just in 
case.  Or is there a chance that start >= PROC_BLOCK_SIZE (but start < 
page) in case #1?  If that is true I am wondering how it could possibly 
be correct since start will be used as a length which is greater than 
the size of the page.

-todd


> 
> Introducing method #1 was a bad idea because this hack made it
> impossible cleanly to implement what you suggest.
> 
> --
> Thomas Hood
> 
> On Mon, 2002-03-25 at 13:18, Todd Inglett wrote:
> 
>>How about applying my trivial patch and then adding this to your nice 
>>comment?
>>
>>3) Set *start = an address outside the buffer.
>>    Put the data of the requested offset at *start.
>>    Return the number of bytes of data placed there.
>>    If this number is greater than zero and you
>>    didn't signal eof and the reader is prepared to
>>    take more data you will be called again with the
>>    requested offset advanced by the number ob tyes
>>    absorbed.
>>
>>The code should still work with the other cases now that the hack is 
>>fixed.  Of course, rather than add 3), it would be better to re-word 2) 
>>(e.g. "Set *start = address of the buffer which may or may not be in the 
>>given buffer.).
>>
>>There are cases where the data is available and need not be copied.  My 
>>code got simpler when I got rid of the need to copy my data around.
>>