Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Fri, 15 Jul 2022 16:55:57 +0800
From:   Matt Hsiao <matt.hsiao@hpe.com>
To:     Greg KH <gregkh@linuxfoundation.org>
Cc:     linux-kernel@vger.kernel.org, arnd@arndb.de, jerry.hoemann@hpe.com,
        scott.norton@hpe.com, camille.lu@hpe.com, geoffrey.ndu@hpe.com,
        gustavo.knuppe@hpe.com
Subject: Re: [PATCH v2 1/1] misc: hpilo: switch .{read,write} ops to
 .{read,write}_iter
Message-ID: <20220715085557.GC15061@blofly.os1.tw>
References: <20220713175452.4221-1-matt.hsiao@hpe.com>
 <20220713175452.4221-2-matt.hsiao@hpe.com>
 <Ys8OyLA35o/wr1jB@kroah.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Ys8OyLA35o/wr1jB@kroah.com>
User-Agent: Mutt/1.5.21 (2010-09-15)
Precedence: bulk

On Wed, Jul 13, 2022 at 08:28:24PM +0200, Greg KH wrote:
> On Thu, Jul 14, 2022 at 01:54:52AM +0800, matt.hsiao@hpe.com wrote:
> > From: Matt Hsiao <matt.hsiao@hpe.com>
> > 
> > Commit 4d03e3cc59828c82ee89 ("fs: don't allow kernel reads and writes
> > without iter ops") requested exclusive .{read,write}_iter ops for
> > kernel_{read,write}. To support dependent drivers to access hpilo by
> > kernel_{read,write}, switch .{read,write} ops to their iter variants.
> > 
> > Signed-off-by: Matt Hsiao <matt.hsiao@hpe.com>
> 
> So this fixes a bug?  What commit does this fix?

No, this is not a bug fix. Please see my explanation for your main question below. 

> 
> Should it go to stable branches?  If so, which ones?

No, it does not need to.

> 
> But my main question is I have no idea what the changelog means here.
> What is a "dependent driver"?  What does "exclusive" mean here?  What is
> a iter variant?

There is an out-of-box driver which is not in the upstream kernel yet
that uses kernel_{read,write} to access the hpilo driver for talking
to the iLO ASIC. Before commit 4d03e3cc59828c82ee89 ("fs: don't allow kernel
reads and writes without iter ops"), kernel_{read,write} would call the
.{read,write} file ops that hpilo already implemented, so there was no problem;
But after that commit, kernel_{read,write} would only allow the .{read,write}_iter
file ops, and disallowed the coexistence of .{read,write} file ops. Accessing
hpilo now fails since it does not have the .{read,write}_iter file ops. To make it
work, this patch implements the .{read,write}_iter file ops and removed the
.{read,write} ones.

> 
> 
> 
> > ---
> >  drivers/misc/hpilo.c | 31 ++++++++++++++++++-------------
> >  1 file changed, 18 insertions(+), 13 deletions(-)
> > 
> > diff --git a/drivers/misc/hpilo.c b/drivers/misc/hpilo.c
> > index 8d00df9243c4..5d431a56b7eb 100644
> > --- a/drivers/misc/hpilo.c
> > +++ b/drivers/misc/hpilo.c
> > @@ -23,6 +23,7 @@
> >  #include <linux/wait.h>
> >  #include <linux/poll.h>
> >  #include <linux/slab.h>
> > +#include <linux/uio.h>
> >  #include "hpilo.h"
> >  
> >  static struct class *ilo_class;
> > @@ -435,14 +436,14 @@ static void ilo_set_reset(struct ilo_hwinfo *hw)
> >  	}
> >  }
> >  
> > -static ssize_t ilo_read(struct file *fp, char __user *buf,
> > -			size_t len, loff_t *off)
> > +static ssize_t ilo_read_iter(struct kiocb *iocb, struct iov_iter *to)
> >  {
> > -	int err, found, cnt, pkt_id, pkt_len;
> > -	struct ccb_data *data = fp->private_data;
> > +	int err = 0, found, cnt, pkt_id, pkt_len;
> > +	struct ccb_data *data = iocb->ki_filp->private_data;
> >  	struct ccb *driver_ccb = &data->driver_ccb;
> >  	struct ilo_hwinfo *hw = data->ilo_hw;
> >  	void *pkt;
> > +	size_t len = iov_iter_count(to), copied;
> >  
> >  	if (is_channel_reset(driver_ccb)) {
> >  		/*
> > @@ -477,7 +478,9 @@ static ssize_t ilo_read(struct file *fp, char __user *buf,
> >  	if (pkt_len < len)
> >  		len = pkt_len;
> >  
> > -	err = copy_to_user(buf, pkt, len);
> > +	copied = copy_to_iter(pkt, len, to);
> > +	if (unlikely(copied != len))
> 
> Why unlikely?  If you can prove it is needed in benchmarks, great,
> otherwise never add likely/unlikely as they are almost always wrong and
> the compiler and cpu can do it better.

Will remove it in the next verion of patch.

> 
> 
> > +		err = -EFAULT;
> >  
> >  	/* return the received packet to the queue */
> >  	ilo_pkt_enqueue(hw, driver_ccb, RECVQ, pkt_id, desc_mem_sz(1));
> > @@ -485,14 +488,14 @@ static ssize_t ilo_read(struct file *fp, char __user *buf,
> >  	return err ? -EFAULT : len;
> >  }
> >  
> > -static ssize_t ilo_write(struct file *fp, const char __user *buf,
> > -			 size_t len, loff_t *off)
> > +static ssize_t ilo_write_iter(struct kiocb *iocb, struct iov_iter *from)
> >  {
> > -	int err, pkt_id, pkt_len;
> > -	struct ccb_data *data = fp->private_data;
> > +	int err = 0, pkt_id, pkt_len;
> > +	struct ccb_data *data = iocb->ki_filp->private_data;
> >  	struct ccb *driver_ccb = &data->driver_ccb;
> >  	struct ilo_hwinfo *hw = data->ilo_hw;
> >  	void *pkt;
> > +	size_t len = iov_iter_count(from), copied;
> >  
> >  	if (is_channel_reset(driver_ccb))
> >  		return -ENODEV;
> > @@ -506,9 +509,11 @@ static ssize_t ilo_write(struct file *fp, const char __user *buf,
> >  		len = pkt_len;
> >  
> >  	/* on failure, set the len to 0 to return empty packet to the device */
> > -	err = copy_from_user(pkt, buf, len);
> > -	if (err)
> > +	copied = copy_from_iter(pkt, len, from);
> > +	if (unlikely(copied != len)) {
> 
> Same here.

Will remove it in the next verion of patch.

> 
> thanks,
> 
> greg k-h