From mboxrd@z Thu Jan 1 00:00:00 1970 From: Neil Brown Subject: Re: simple question: where is internal bitmap stored? Date: Wed, 13 May 2009 20:42:34 +1000 Message-ID: <18954.42010.122591.899035@notabene.brown> References: <4A0A86B8.5030905@eyal.emu.id.au> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: message from Eyal Lebedinsky on Wednesday May 13 Sender: linux-raid-owner@vger.kernel.org To: Eyal Lebedinsky Cc: linux-raid@vger.kernel.org List-Id: linux-raid.ids On Wednesday May 13, eyal@eyal.emu.id.au wrote: > If I have a live raid and want to enable bitmap then I do > mdadm --grow /dev/md0 --bitmap=internal > Does this bitmap use space in the already present header? If so then > does this mean that the headers have enough reserve for this? Why > not always use the bitmap then? 1/ Yes. 2/ Yes. 3/ because there are other reasons for not wanting a bitmap. Using a write-intent bitmap can cause a reduction in throughput. It is conceivable that having a bitmap should be the default, but it should definitely be options. In fact, ever array does has a default bitmap. However that bitmap contains exactly one bit, and it cover the whole array. It is the clean/active flag. Defaulting to a few more bits is possibly a good idea... > > If it carves extra space from the device which already has data > (e.g. a fs) then surely this will be problematic. Yes. It surely would be. NeilBrown