Re: [PATCH v6 04/16] nvme-core: introduce nvme_get_by_path()

From: Logan Gunthorpe <logang@deltatee.com>
To: Keith Busch <kbusch@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>,
	linux-kernel@vger.kernel.org, linux-nvme@lists.infradead.org,
	linux-block@vger.kernel.org, linux-fsdevel@vger.kernel.org,
	Christoph Hellwig <hch@lst.de>, Sagi Grimberg <sagi@grimberg.me>,
	Jens Axboe <axboe@fb.com>,
	Chaitanya Kulkarni <Chaitanya.Kulkarni@wdc.com>,
	Max Gurtovoy <maxg@mellanox.com>,
	Stephen Bates <sbates@raithlin.com>
Subject: Re: [PATCH v6 04/16] nvme-core: introduce nvme_get_by_path()
Date: Thu, 25 Jul 2019 14:37:05 -0600	[thread overview]
Message-ID: <3bb266ae-abf3-0146-5d93-e7a600453493@deltatee.com> (raw)
In-Reply-To: <20190725203118.GB7317@localhost.localdomain>

On 2019-07-25 2:31 p.m., Keith Busch wrote:
> On Thu, Jul 25, 2019 at 02:28:28PM -0600, Logan Gunthorpe wrote:
>>
>>
>> On 2019-07-25 1:58 p.m., Keith Busch wrote:
>>> On Thu, Jul 25, 2019 at 11:54:18AM -0600, Logan Gunthorpe wrote:
>>>>
>>>>
>>>> On 2019-07-25 11:50 a.m., Matthew Wilcox wrote:
>>>>> On Thu, Jul 25, 2019 at 11:23:23AM -0600, Logan Gunthorpe wrote:
>>>>>> nvme_get_by_path() is analagous to blkdev_get_by_path() except it
>>>>>> gets a struct nvme_ctrl from the path to its char dev (/dev/nvme0).
>>>>>>
>>>>>> The purpose of this function is to support NVMe-OF target passthru.
>>>>>
>>>>> I can't find anywhere that you use this in this patchset.
>>>>>
>>>>
>>>> Oh sorry, the commit message is out of date the function was actually
>>>> called nvme_ctrl_get_by_path() and it's used in Patch 10.
>>>
>>> Instead of by path, could we have configfs take something else, like
>>> the unique controller instance or serial number? I know that's different
>>> than how we handle blocks and files, but that way nvme core can lookup
>>> the cooresponding controller without adding new cdev dependencies.
>>
>> Well the previous version of the patchset just used the ctrl name
>> ("nvme1") and looped through all the controllers to find a match. But
>> this sucks because of the inconsistency and the fact that the name can
>> change if hardware changes and the number changes. Allowing the user to
>> make use of standard udev rules seems important to me.
> 
> Should we then create a new udev rule for persistent controller
> names? /dev/nvme1 may not be the same controller each time you refer
> to it.

Udev can only create symlinks from /dev/nvme0 to
/dev/nvme-persistent-name and users can do this as they need now. No
changes needed.

My point was if we use the ctrl name (nvme0) as a reference then the
kernel can't make use of these symlinks or anything udev does seeing
that name is internal to the kernel only.

If we use cdev_get_by_path()/nvme_ctrl_get_by_path() then this isn't a
problem as we can open a symlink to /dev/nvme0 without any issues.

Logan