[OpenAFS-devel] Linux atime nits

[Marc Dionne]

On Mon, Jan 18, 2010 at 1:12 PM, Marc Aurele La France <tsi@ualberta.ca> wr=
ote:
> On Mon, 18 Jan 2010, Derrick Brashear wrote:
>>
>> On Mon, Jan 18, 2010 at 12:26 PM, Marc Aurele La France <tsi@ualberta.ca=
>
>> wrote:
>>>
>>> On Sun, 17 Jan 2010, Simon Wilkinson wrote:
>>>>
>>>> On 17 Jan 2010, at 13:48, Marc Aurele La France wrote:
>>>>>
>>>>> Any chance of incorporating these into stable?
>
>>>> When we last discussed pullups for 1.4.12, we decided that these were
>>>> too high risk for that release (the changes appeared very late in the
>>>> release cycle). MS_NOATIME is sufficiently large that it doesn't appea=
r
>>>> to conflict with any inode flags, up to and including Linus's current
>>>> kernel, so there doesn't appear to be a risk associated with not takin=
g
>>>> them this time round.
>
>>> Only one more inode flag is needed to create a conflict. =A0Whether or =
not
>>> this occurs before 1.4.13 is released is a risk I gather you are willin=
g
>>> to
>>> take. =A0I think it's better to be pro-active.
>
>> The question is "will it happen in a kernel 1.4.12 also doesn't
>> support, making the issue somewhat less urgent?".
>
> Humm. =A0That depends on whether or not a inode flag conflict is the only
> reason it doesn't. =A0Incompatible changes to kernel APIs have died down =
quite
> a bit of late.

Things are indeed better lately, but I've yet to see a kernel that
required no changes at all.  I would also note that the latest inode
flag (S_PRIVATE) is 5'ish years old, so this is not something that
gets added very often.

One worry for 1.4 is that setting MS_NOATIME may result in subtle
changes in the VFS behaviour, which may change how/when our callback
functions get called.  In theory it shouldn't be an issue, but I guess
the feeling was that it deserves more testing before landing in 1.4 at
this stage of the release process.

Marc