Re: [PATCH v5 net-next 02/18] ionic: Add hardware init and device commands

[Shannon Nelson <snelson@pensando.io>]

On 8/27/19 12:50 PM, Andrew Lunn wrote:
> On Tue, Aug 27, 2019 at 10:39:20AM -0700, Shannon Nelson wrote:
>> On 8/26/19 7:26 PM, Andrew Lunn wrote:
>>> On Mon, Aug 26, 2019 at 02:33:23PM -0700, Shannon Nelson wrote:
>>>> +void ionic_debugfs_add_dev(struct ionic *ionic)
>>>> +{
>>>> +	struct dentry *dentry;
>>>> +
>>>> +	dentry = debugfs_create_dir(ionic_bus_info(ionic), ionic_dir);
>>>> +	if (IS_ERR_OR_NULL(dentry))
>>>> +		return;
>>>> +
>>>> +	ionic->dentry = dentry;
>>>> +}
>>> Hi Shannon
>>>
>>> There was recently a big patchset from GregKH which removed all error
>>> checking from drivers calling debugfs calls. I'm pretty sure you don't
>>> need this check here.
>> With this check I end up either with a valid dentry value or NULL in
>> ionic->dentry.  Without this check I possibly get an error value in
>> ionic->dentry, which can get used later as the parent dentry to try to make
>> a new debugfs node.
> Hi Shannon
>
> What you should find is that every debugfs function will have
> something like:
>
> 	if (IS_ERR(dentry))
> 	   return dentry;
> or
> 	if (IS_ERR(parent))
> 	   return parent;
>
> If you know of a API which is missing such protection, i'm sure GregKH
> would like to know. Especially since he just ripped all such
> protection in driver out. Meaning he just broken some drivers if such
> IS_ERR() calls are missing in the debugfs core.
>
Ah, here's the confusion: there's a start_creating() in the tracefs as 
well as in the debugfs, and the tracefs code doesn't have all of those 
checks.

sln