Re: read_proc issue

[Val Henson <val@nmt.edu>]

On Wed, Feb 27, 2002 at 11:33:18AM -0800, Randy.Dunlap wrote:
> On Tue, 26 Feb 2002, Laurent wrote:
<snippety snippety>
> | Problem is:
> | when I 'cat /proc/number' multiple times, instead of getting
> | 0
> | 1
> | 2
> | 3
> | ...
> |
> | I get
> | 0
> | 2
> | 4
> | 6
> | ...
<more snippage>
> app. calls read() [Key data: with offset=0], which calls sys_read(),
> which calls proc_file_read(), which returns N bytes, and then
> proc_file_read() and sys_read() and read() return to the app.
> 
> When the app. gets N bytes of actual data on its read(), it
> doesn't know whether it was at EOF or not, so it tries another
> read() [with offset=N], and sys_read() calls proc_file_read()
> again, both of which return 0, so the app. knows that there
> is no more data.

I've encountered this problem before, too.  What is the "One True Way"
to do this cleanly?  In other words, if you want to do a calculation
once every time someone runs "cat /proc/foo", what is the cleanest way
to do that?  The solution we came up with was to check the file offset
and only do the calculation if offset == 0, which seems pretty
hackish.

-VAL