diff mbox series

[net-next] net: macb: allow MTU change when the interface is up

Message ID 20240325152735.1708636-1-thomas.perrot@bootlin.com (mailing list archive)
State Changes Requested
Delegated to: Netdev Maintainers
Headers show
Series [net-next] net: macb: allow MTU change when the interface is up | expand

Checks

Context Check Description
netdev/series_format success Single patches do not need cover letters
netdev/tree_selection success Clearly marked for net-next
netdev/ynl success Generated files up to date; no warnings/errors; no diff in generated;
netdev/fixes_present success Fixes tag not required for -next series
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 965 this patch: 965
netdev/build_tools success No tools touched, skip
netdev/cc_maintainers success CCed 6 of 6 maintainers
netdev/build_clang success Errors and warnings before: 955 this patch: 955
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/deprecated_api success None detected
netdev/check_selftest success No net selftest shell script
netdev/verify_fixes success No Fixes tag
netdev/build_allmodconfig_warn success Errors and warnings before: 977 this patch: 977
netdev/checkpatch success total: 0 errors, 0 warnings, 0 checks, 21 lines checked
netdev/build_clang_rust success No Rust files in patch. Skipping build
netdev/kdoc success Errors and warnings before: 0 this patch: 0
netdev/source_inline success Was 0 now: 0
netdev/contest success net-next-2024-03-26--00-00 (tests: 943)

Commit Message

Thomas Perrot March 25, 2024, 3:27 p.m. UTC 
From: Thomas Perrot <thomas.perrot@bootlin.com>

The macb_change_mtu callback doesn't allow MTU changes when the
interface is up because the MTU is used to set the buffer size on
GEM.

So, to be able to change the MTU from userspace it is necessary to
explicitly down then to up the interface, which is problematic in
some cases.

Then we suggest to reset the interface when the interface is already
up to allow to change the MTU, regardless of if the interface is up
or not.

Signed-off-by: Thomas Perrot <thomas.perrot@bootlin.com>
---
 drivers/net/ethernet/cadence/macb_main.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

Comments

Simon Horman March 25, 2024, 8:54 p.m. UTC | #1 
On Mon, Mar 25, 2024 at 04:27:35PM +0100, thomas.perrot@bootlin.com wrote:
> From: Thomas Perrot <thomas.perrot@bootlin.com>
> 
> The macb_change_mtu callback doesn't allow MTU changes when the
> interface is up because the MTU is used to set the buffer size on
> GEM.
> 
> So, to be able to change the MTU from userspace it is necessary to
> explicitly down then to up the interface, which is problematic in
> some cases.
> 
> Then we suggest to reset the interface when the interface is already
> up to allow to change the MTU, regardless of if the interface is up
> or not.
> 
> Signed-off-by: Thomas Perrot <thomas.perrot@bootlin.com>

Hi Thomas,

I'm not sure that it is expected behaviour for an interface
to reset like this when a change of MTU is requested.
While conversely I think it is common (if not entirely desirable)
to prohibit changing the MTU when an interface is up.
What is the problem being addressed here?

> ---
>  drivers/net/ethernet/cadence/macb_main.c | 12 ++++++++++--
>  1 file changed, 10 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
> index 898debfd4db3..5c20b162c5da 100644
> --- a/drivers/net/ethernet/cadence/macb_main.c
> +++ b/drivers/net/ethernet/cadence/macb_main.c
> @@ -3019,11 +3019,19 @@ static int macb_close(struct net_device *dev)
>  
>  static int macb_change_mtu(struct net_device *dev, int new_mtu)
>  {
> -	if (netif_running(dev))
> -		return -EBUSY;
> +	int reset = 0;

nit: reset could be bool and set to true and false.

> +
> +	/* On GEM the buffer size is based on the MTU */
> +	if (netif_running(dev)) {
> +		macb_close(dev);
> +		reset = 1;
> +	}
>  
>  	dev->mtu = new_mtu;
>  
> +	if (reset)
> +		macb_open(dev);
> +
>  	return 0;
>  }
>  
> -- 
> 2.44.0
> 
>
Jakub Kicinski March 26, 2024, 1:56 a.m. UTC | #2 
On Mon, 25 Mar 2024 20:54:01 +0000 Simon Horman wrote:
> I'm not sure that it is expected behaviour for an interface
> to reset like this when a change of MTU is requested.
> While conversely I think it is common (if not entirely desirable)
> to prohibit changing the MTU when an interface is up.
> What is the problem being addressed here?

Right..

> >  	dev->mtu = new_mtu;
> >  
> > +	if (reset)
> > +		macb_open(dev);

.. imagine admin does this over SSH on a remote box and system 
is under memory pressure. Even ignoring the fact you're not checking
the return value, the result of changing MTU should be either having 
the requested MTU (success) or having the old MTU (failure).
Not "machine drops off the network" :(
Thomas Perrot March 26, 2024, 8:49 a.m. UTC | #3 
Hello,

On Mon, 2024-03-25 at 18:56 -0700, Jakub Kicinski wrote:
> On Mon, 25 Mar 2024 20:54:01 +0000 Simon Horman wrote:
> > I'm not sure that it is expected behaviour for an interface
> > to reset like this when a change of MTU is requested.
> > While conversely I think it is common (if not entirely desirable)
> > to prohibit changing the MTU when an interface is up.
> > What is the problem being addressed here?
> 

The problem being addressed here, is that NetworkManager isn't able to
apply the MTU value set in the connection configuration file because
the link is already up, then the change_mtu callback returns an error:

"NetworkManager[412]: <debug> [1709629970.1735] platform: (eth0) link:
setting mtu 1400                                                      
NetworkManager[412]: <trace> [1709629970.1737] platform-linux: delayed-
action: schedule wait-for-response-rtnl (seq 41, timeout in
0.199992796, response-type 0)                                         
NetworkManager[412]: <trace> [1709629970.1737] platform-linux: delayed-
action: schedule refresh-link (ifindex 2)                             
NetworkManager[412]: <trace> [1709629970.1738] platform-linux: delayed-
action: handle refresh-link (ifindex 2)                               
NetworkManager[412]: <debug> [1709629970.1738] platform-linux: do-
request-link: 2                                                       
NetworkManager[412]: <trace> [1709629970.1739] platform-linux: rtnl:
recvmsg: new message NLMSG_ERROR, flags 0, seq 41                     
NetworkManager[412]: <debug> [1709629970.1740] platform-linux: rtnl:
recvmsg: error message from kernel: Device or resource busy (-16) for
request 41"

Kind regards,
Thomas Perrot

> Right..
> 
> > >  	dev->mtu = new_mtu;
> > >  
> > > +	if (reset)
> > > +		macb_open(dev);
> 
> .. imagine admin does this over SSH on a remote box and system 
> is under memory pressure. Even ignoring the fact you're not checking
> the return value, the result of changing MTU should be either having 
> the requested MTU (success) or having the old MTU (failure).
> Not "machine drops off the network" :(
>
Simon Horman March 26, 2024, 9:30 a.m. UTC | #4 
On Tue, Mar 26, 2024 at 09:49:19AM +0100, Thomas Perrot wrote:
> Hello,
> 
> On Mon, 2024-03-25 at 18:56 -0700, Jakub Kicinski wrote:
> > On Mon, 25 Mar 2024 20:54:01 +0000 Simon Horman wrote:
> > > I'm not sure that it is expected behaviour for an interface
> > > to reset like this when a change of MTU is requested.
> > > While conversely I think it is common (if not entirely desirable)
> > > to prohibit changing the MTU when an interface is up.
> > > What is the problem being addressed here?
> > 
> 
> The problem being addressed here, is that NetworkManager isn't able to
> apply the MTU value set in the connection configuration file because
> the link is already up, then the change_mtu callback returns an error:
> 
> "NetworkManager[412]: <debug> [1709629970.1735] platform: (eth0) link:
> setting mtu 1400                                                      
> NetworkManager[412]: <trace> [1709629970.1737] platform-linux: delayed-
> action: schedule wait-for-response-rtnl (seq 41, timeout in
> 0.199992796, response-type 0)                                         
> NetworkManager[412]: <trace> [1709629970.1737] platform-linux: delayed-
> action: schedule refresh-link (ifindex 2)                             
> NetworkManager[412]: <trace> [1709629970.1738] platform-linux: delayed-
> action: handle refresh-link (ifindex 2)                               
> NetworkManager[412]: <debug> [1709629970.1738] platform-linux: do-
> request-link: 2                                                       
> NetworkManager[412]: <trace> [1709629970.1739] platform-linux: rtnl:
> recvmsg: new message NLMSG_ERROR, flags 0, seq 41                     
> NetworkManager[412]: <debug> [1709629970.1740] platform-linux: rtnl:
> recvmsg: error message from kernel: Device or resource busy (-16) for
> request 41"

Hi Thomas,

I understand this is not ideal. But from a driver level perspective, if the
HW doesn't support changing the MTU while the device is up, rejecting the
change MTU operation is actually the best practice (or least worst
depending on your perspective).

As Jakub pointed out elsewhere in this thread, with your proposed change,
there is a real chance the NIC could drop of the network entirely and not
come back. Whereas the expected outcome is for the NIC to remain on the
network with either the new or old MTU.
diff mbox series

Patch

diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
index 898debfd4db3..5c20b162c5da 100644
--- a/drivers/net/ethernet/cadence/macb_main.c
+++ b/drivers/net/ethernet/cadence/macb_main.c
@@ -3019,11 +3019,19 @@  static int macb_close(struct net_device *dev)
 
 static int macb_change_mtu(struct net_device *dev, int new_mtu)
 {
-	if (netif_running(dev))
-		return -EBUSY;
+	int reset = 0;
+
+	/* On GEM the buffer size is based on the MTU */
+	if (netif_running(dev)) {
+		macb_close(dev);
+		reset = 1;
+	}
 
 	dev->mtu = new_mtu;
 
+	if (reset)
+		macb_open(dev);
+
 	return 0;
 }