lkml.org 
[lkml]   [2010]   [Nov]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH 4/4] fanotify: Expose the file changes to the user
From
On Fri, Nov 26, 2010 at 14:41, Tvrtko Ursulin <tvrtko.ursulin@sophos.com> wrote:
> On Friday 26 Nov 2010 11:21:18 Alexey Zaytsev wrote:
>> On Fri, Nov 26, 2010 at 13:11, Tvrtko Ursulin <tvrtko.ursulin@sophos.com>
> wrote:
>> > On Monday 22 Nov 2010 00:37:21 Alexey Zaytsev wrote:
>> >> +struct fanotify_opt_hdr {
>> >> +       __u8 type;
>> >> +       __u8 reserved;
>> >> +       __u16 len;
>> >> +       /* Payload goes here. */
>> >> +};
>> >> +
>> >> +#define FANOTIFY_METADATA_VERSION      3
>> >>
>> >>  struct fanotify_event_metadata {
>> >> -       __u16 event_len;
>> >> +       __u16 event_len; /* Including the options */
>> >>         __u8 vers;
>> >> -       __u8 reserved;
>> >> +       __u8 options_offset; /* Aka header length */
>> >>         __s32 fd;
>> >>         __aligned_u64 mask;
>> >>         __s32 pid;
>> >> +       /* Options go here. */
>> >>  };
>> >
>> > I am not 100% comfortable with having 16 bits length fields because I am
>> > just not sure there is a measurable performance difference versus just
>> > going with 32 bits.
>>
>> I'm not concerned so much with the performance, as with the storage.
>> If we are generating events for every access on a mount point, some
>> consumers might build a considerable backlog over a period of high
>> activity. Would be good if we could cut the event size by 1/3 for
>> free. And I don't see an event growing 64k even with the options. Do
>> you?
>
> I don't but maybe it is just lack of imagination.
>
> My bias is that I am mostly thinking about synchronous events where large
> backlog is not a realistic scenario. How realistic you think is this with
> async events?
>

Ok, you are probably right, we can't build a huge backlog, because
each event carries an open fd with it. So, given the unnoticeable
performance gain, it might be worth the leave the possibility to pass
events greater than 64k (with options) in the future, no matter how
unlikely this looks now. We still could save 32 bit by packing the
vers and the offset together. The common part of the events
(options_offset) should not grow over 256 byte, or it would start to
affect the performance for common events, and I hope that 256 versions
would also be enough, so we could even reserve 16 bit for further
extensions.

Eric?

>> > Also, options_offset is, if I understood it correctly, basically the
>> > lenght of fanotify_event_metadata. As such, isn't that field redundant
>> > since the lenght is implied from the protocol version?
>>
>> There are two problems there.
>>
>> 1) You lose backwards-compatibility. It's still an ABI breakage, even
>> if you tell the users about it.
>
> Assuming 2.6.37 release as starting point for ABI considerations?

Yep, that's why I'm asking to include the first patch into .37.

>
>> 2) You can't build a program to account for different fanotify versions:
>>         if (vers >= N) { use the cool stuff } else if {vers >= N-1}  {
>> still good }
>
> I don't get why not, but maybe I am just slow today. There will always be 1:1
> mapping from version to your options_offset field, no? How does then removing
> options_offset change anything?
>

You need to get the old version of the ABI definition somewhere. And
if you need to cast the events to different types, and use different
versions of FAN_OPTION* depending on the event version, the program
also becomes a bit complicated.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2010-11-26 13:15    [W:0.044 / U:2.276 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site