Re: [patchset 0/4] osd: Stop usage of blk_rq_append_bio

[Boaz Harrosh <bharrosh@panasas.com>]

On 05/13/2009 05:47 PM, James Bottomley wrote:
> On Wed, 2009-05-13 at 17:36 +0300, Boaz Harrosh wrote:
>> On 05/13/2009 05:28 PM, Boaz Harrosh wrote:
>>> On 05/12/2009 02:25 PM, Jens Axboe wrote:
>>>> On Thu, May 07 2009, Boaz Harrosh wrote:
>>>>> Osd library needs to submit pre-allocated bios, form several sources.
>>>>> osdblk exofs and pNFS-layout driver all have prepared bios for IO submission.
>>>>> On top of that the osd library needs to append additional segments to the
>>>>> IO memory, for get/set attributes and more.
>>>>>
>>>>> All these are done today by use of a temporary hack - blk_rq_append_bio.
>>>>> This is bad on few accounts.
>>>>> 1. blk_rq_append_bio was not meant to be exported and is very specific to its users.
>>>>> 2. blk_rq_append_bio does not support chained bios.
>>>>> 3. blk_rq_append_bio does not bounce the bio and therefore current osd implementation
>>>>>    has a bug.
>>>>>
>>>>> The proposed solution adds two new fixtures to the block layer, and a corresponding
>>>>> fixing patch to osd. These are:
>>>>>
>>>>> [PATCH 1/4] allow blk_rq_map_kern to append to requests
>>>>> [PATCH 2/4] libosd: Use new blk_rq_map_kern
>>>>>
>>>>>   This is originally a James patch and it's used, to let blk_rq_map_kern append it's buffer
>>>>>   to existing bio, and there for is able to be called multiple times in a loop, to append
>>>>>   multiple segments. This API can also be useful for scsi/block targets that have segment
>>>>>   information in some other memory structure (like scatterlist) and wants to set it into
>>>>>   a request. Until such time that they have a proper support for mapping scatterlists directly.
>>>>>   (Since above called on long lists might not be good for performance)
>>>>>
>>>>>   Here in osd it makes tons of sense, and should be considered for inclusion.
>>>>>   (The patches are based on linus-tip but should patch on block tree)
>>>>>
>>>>> [RFC 3/4] New blk_make_request(), takes bio, returns a request
>>>>> [RFC 4/4] libosd: Use of new blk_make_request
>>>>>   
>>>>>   Here I propose a new block API, that will support proper delegation of a bio
>>>>>   to a full request. Please read inside the patch descriptions for details.
>>>>>   After this patch both osd and block layer will have the proper support for osdblk
>>>>>   driver as well as future needs.
>>>>>   These patches also eliminate the last use of blk_rq_append_bio which can be now un-exported.
>>>>>
>>>>>   These two patches conflic with Tejun's branch and are based on linus-tip. Upon positive review
>>>>>   I will serialize them with Tejun and submit them properly. But first they must be agreed upon.
>>>>>   Jens, I specially need your opinion on this?
>>>> Looks sane to me. Can you resubmit against 'for-2.6.31' of the block git
>>>> repo?
>>>>
>>> Thanks Jens.
>>>
>>> I have done the rebase and ran some tests, however I was unable to test these patches
>>> as is, because there are some inter tree fallouts.
>>>
>>> Jens, James, Stephan, I please need your help
>>>
>>> The situation is like that.
>>> - Both block/for-next and scsi/master are based on an old osd upstream-point (v2.6.30-rc3--ce8a7424)
>>> - Linus tip has important OSD patches that went in via scsi-rc-fixes which changed Wire format
> 
> So just pull them into Linus head and build on that ... as long as you
> explain what the base was, I can rebase scsi-misc (or run a post merge
> tree) to cope.  It needs rebasing anyway to redo the mvsas patches.
> 
>>> - If I try and merge block/for-next ontop of plain linus/master I get a merge conflict
>>> - If I try merge scsi/master block/for-next I get build errors / conflicts
> 
> This is the problem of the renames ... I think we need a block postmerge
> tree to fix this up, but that probably needs sorting out first.
> 
>>> So there is no sane tree point that I can test on.
>>>
>>> It would be nice if both Jens block/for-next and scsi-misc/master could be rebased on Linus rc5++
>>> and resolve these conflicts. (And scsi-misc conflicts with Tejun's cleanups be put in a second stage
>>> tree)
>>>
>>> Should I send the patches as is half tested? Or wait for things to settle after I tested them
>>> with all changes included?
>>>
>>> I have cut a new osd/linux-next branch which is based, not on linus, but on v2.6.30-rc3--ce8a7424
>>> the base point for block/for-next and scsi-misc/master. So in next it should all come together
>>> well, and I will try to clone tomorrow's next and test on top of that.
>>>
>> This will not work I have one patch [3/4] New blk_make_request(), takes bio, returns a request
>> which will conflict with block/for-next if I rebase it on v2.6.30-rc3--ce8a7424.
>>
>> Should I cut osd/linux-next on top of block/for-next ?
> 
> What you really want is on the combination of the necessary trees.  If
> it's only block, then sure ... if it's block and SCSI, that's postmerge
> territory.
> 
> James
> 
> 

Thank you for your reply.

I did more-less what you said rebased block/for-next on linus-tip and
fixed the merge with scsi-misc as per Stephan advise. Plus my patches
last. Test ran well.

I see that you rebased by now, though I suspect the fc's blk_end_request
call will fail to build if merged with block tree.

I'm also seeing some recent changes to block git so I suspect that Jens is
in the middle of rebasing too. (I hope)

As far as OSD, I managed to separate the two block-based changes to osd
from the rest of the changes scheduled for 2.6.31 in such a way that they do not
conflict and can merge either way (block first or scsi-misc first).

I'll repost all these patches. Sunday hopfuly after Jens rebases.

Thanks
Boaz