[Top] [All Lists]

Re: [REVIEW][PATCH 13/15] userns: Add basic quota support

To: Jan Kara <jack@xxxxxxx>
Subject: Re: [REVIEW][PATCH 13/15] userns: Add basic quota support
From: ebiederm@xxxxxxxxxxxx (Eric W. Biederman)
Date: Mon, 27 Aug 2012 08:54:21 -0700
Cc: linux-kernel@xxxxxxxxxxxxxxx, netdev@xxxxxxxxxxxxxxx, linux-fsdevel@xxxxxxxxxxxxxxx, "Serge E. Hallyn" <serge@xxxxxxxxxx>, David Miller <davem@xxxxxxxxxxxxx>, Steven Whitehouse <swhiteho@xxxxxxxxxx>, cluster-devel@xxxxxxxxxx, Mark Fasheh <mfasheh@xxxxxxxx>, Joel Becker <jlbec@xxxxxxxxxxxx>, ocfs2-devel@xxxxxxxxxxxxxx, Ben Myers <bpm@xxxxxxx>, Alex Elder <elder@xxxxxxxxxx>, xfs@xxxxxxxxxxx, Dmitry Monakhov <dmonakhov@xxxxxxxxxx>
In-reply-to: <20120827085034.GA8998@xxxxxxxxxxxxx> (Jan Kara's message of "Mon, 27 Aug 2012 10:50:34 +0200")
References: <87lih2h6i4.fsf@xxxxxxxxxxxx> <87harqecvk.fsf@xxxxxxxxxxxx> <20120827085034.GA8998@xxxxxxxxxxxxx>
User-agent: Gnus/5.13 (Gnus v5.13) Emacs/24.1 (gnu/linux)
Jan Kara <jack@xxxxxxx> writes:

>   Hello,
> On Sat 25-08-12 17:05:35, Eric W. Biederman wrote:
>> Two helper are added dqgetusr and dqgetgrp to allow the quota
>> infrastructure to be called with a kuid and a kgid respectively.  This
>> creates type safe variants of dqget and leads to shorter more
>> comprehensible code.
>   It would look more comprehensible to me to have functions like:
> kuid2qown() and kgid2qown() and then call dqget(sb, kuid2qown(attr->uid))
> (see below for qown_t change proposal). The code then at the first look
> explains what is going on... Hmm?
>> Place the USRQUOTA and GRPQUOTA defines into enum quota_type.  This
>> brings with it the ability for the compiler to check that switch
>> statements handle every quota type, and the ability to mark which
>> values store the type of a quota entry.
>   OK, makes sense.
>> Add the data type qown_t a union of kuid_t and kgid_t. qown_t is a
>> replacement for the implicit union of uid and gid stored in an
>> unsigned int that is was used in the quota data structures.  Making
>> the data type explicit allows the kuid_t and kgid_t type safety to
>> propogate more thoroughly through the code, revealing more places
>> where uid/gid conversions need be made.
>   Hum, when we already do this, wouldn't it make more sense to embed quota
> type in qown_t? Because with the union thing you have no meaningful way of
> accessing that type without having quota type anyway. So having that in a
> single structure makes a lot of sense, plus it makes prototypes shorter...
> And you have to call make_qown() anyway...

So I think there was a reason for having the type separate but I'm not
seeing that reason off the top of my head.

I think it was filesystems like gfs2 doing weird things.  But I tell
you want I will play with this and if I can't reproduce find a reason
for putting them in one structure I will because there are definitely
good reasons for doing that.

>> Allong with the data type qown_t comes the helper functions
>> qown_eq, from_qown, from_qown_munged, qown_valid, and make_qown.
>> Update struct dquot dq_id to be a qown_t.
>> Update the signature of dqget, quota_send_warning, dquot_get_dqblk,
>> and dquot_set_dqblk to use enum quota_type and qown_t.
>> Make minimal changes to gfs2, ocfs2, and xfs to deal with the change
>> in quota structures and signatures.  The ocfs2 changes are larger than
>> most because of the extensive tracing throughout the ocfs2 quota code
>> that prints out dq_id.
>   Otherwise the changes look OK to me, although I didn't check them in
> detail yet (as above suggestions will change the code anyway).
>                                                               Honza



<Prev in Thread] Current Thread [Next in Thread>