Re: [RFC] globmatch() helper function

[Tejun Heo <tj@kernel.org>]

Peter Zijlstra wrote:
> On Thu, 2008-12-18 at 01:33 +0900, Tejun Heo wrote:
>> Steven Rostedt wrote:
>>> On Thu, 2008-12-18 at 01:22 +0900, Tejun Heo wrote:
>>>> Hello,
>>>>
>>>> George Spelvin wrote:
>>>>> Do people think that would be, on balance, better?  It would be plenty
>>>>> good enough for the blacklist application.
>>>> Just pass a depth parameter and trigger WARN_ON() and return -EINVAL
>>>> when it exceeds ten.  It's a five minute change and should be enough
>>>> for kernel usages.
>>> If this is ever expected to be used by userspace, I would not include
>>> the WARN_ON. If this is a generic function, then I'll include in in
>>> ftrace as well, and that takes userspace input. The last thing I want is
>>> a DoS because of printk's to the serial console because some userspace
>>> app is constantly writing bad patterns to this file.
>> Well, then, how about printk_ratelimit()?  Having one too many
>> asterisk will be a very rare occasion and when it happens it's
>> something which can easily escape attention, so I think some form of
>> whining is in order.
> 
> having the write fail with -EINVAL seems suitably whiney..

Yeah, I suppose so but that makes return value handling weird.  -EINVAL
for error, 0 for mis-match and 1 for match.  Three-way returns are
usually very sucky to handle and error-prone, so I was thinking about
returning -EINVAL in the internal function and printing out the caller
address and pattern on ratelimit or once.

Thanks.

-- 
tejun