Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 3 Sep 2018 14:28:04 +0200
From:   Michal Hocko <mhocko@kernel.org>
To:     Douglas Gilbert <dgilbert@interlog.com>
Cc:     Richard Weinberger <richard.weinberger@gmail.com>,
        drorl@infinidat.com, LKML <linux-kernel@vger.kernel.org>,
        linux-block@vger.kernel.org, linux-scsi@vger.kernel.org,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Christoph Hellwig <hch@infradead.org>,
        Jens Axboe <axboe@kernel.dk>
Subject: Re: Recent removal of bsg read/write support
Message-ID: <20180903122804.GA15074@dhcp22.suse.cz>
References: <CAOuWhkaGjX8Q_xghkFVTqzZkDrUv-sdH=VbS7hvnqFWrT+3eOQ@mail.gmail.com>
 <CAFLxGvxX8QW4wUJKommXVMpGBj7G_H3=SOwQGjGuyPZZrhCzKw@mail.gmail.com>
 <6b907fed-d83f-de75-bde4-11270a0b1b0b@interlog.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <6b907fed-d83f-de75-bde4-11270a0b1b0b@interlog.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sun 02-09-18 21:16:10, Douglas Gilbert wrote:
> On 2018-09-02 01:44 PM, Richard Weinberger wrote:
> > CC'ing relevant people. Otherwise your mail might get lost.
> > 
> > On Sun, Sep 2, 2018 at 1:37 PM Dror Levin <drorl@infinidat.com> wrote:
> > > 
> > > Note: I am not subscribed to LKML so please CC replies to this email.
> > > 
> > > Hi,
> > > 
> > > We have an internal tool that uses the bsg read/write interface to
> > > issue SCSI commands as part of a test suite for a storage device.
> > > 
> > > After recently reading on LWN that this interface is to be removed we
> > > tried porting our code to use sg instead. However, that raises new
> > > issues - mainly getting ENOMEM over iSCSI for unknown reasons.
> > > 
> > > Because of this we would like to continue using the bsg interface,
> > > even if some changes are required to meet security concerns.
> > > 
> > > Is there any chance for this removal to be reverted? I saw it was
> > > already included in 4.19-rc1.
> 
> Hi,
> Both bsg and sg are relatively thin shims over the same block layer
> pass-through calls. And neither driver will themselves generate ENOMEM
> unless the CPU is running low of memory.
> 
> In my experience, the main reason for unexpected ENOMEMs *** is from
> blk_rq_map_user_iov() in block/blk_map.c called from both drivers.
> That is a particular resource shortage rather than memory in general.
> I do notice the blk_rq_map_user_iov() is/was called with GFP_KERNEL
> in bsg and GFP_ATOMIC by sg. That suggests when you call write() on
> a sg device and get ENOMEM, then wait a little (depends on your app)
> and try again.

Well, what is the reason to use GFP_ATOMIC in the first place? I am not
familiar with the code so I might be easily wrong but sg_start_req which
calls blk_rq_map_user_iov resp. blk_rq_map_user with GFP_ATOMIC uses
mutex. It is a conditional usage so the sleeping context might depend
on the caller. But I guess it would be better to double check. It looks
suspicious to me.
-- 
Michal Hocko
SUSE Labs