* [PATCH net] bridge: fix hello and hold timers starting/stopping
@ 2017-05-19 16:25 ` Ivan Vecera
0 siblings, 0 replies; 14+ messages in thread
From: Ivan Vecera @ 2017-05-19 16:25 UTC (permalink / raw)
To: netdev; +Cc: davem, sashok, stephen, bridge, lucien.xin, nikolay
Current bridge code incorrectly handles starting/stopping of hello and
hold timers during STP enable/disable.
1. Timers are stopped in br_stp_start() during NO_STP->USER_STP
transition. This is not correct as the timers are stopped in NO_STP
case.
2. Timers are started in br_stp_stop() during USER_STP->NO_STP transition.
This is not also correct as the timers should be stopped in NO_STP
state.
3. Timers are NOT stopped in br_stp_stop() during KERNEL_STP->NO_STP
transition. They should be stopped as they are running in KERNEL_STP
state and should not run in NO_STP case.
The patch is a follow-up for "bridge: start hello_timer when enabling
KERNEL_STP in br_stp_start" patch from Xin Long.
Cc: davem@davemloft.net
Cc: sashok@cumulusnetworks.com
Cc: stephen@networkplumber.org
Cc: bridge@lists.linux-foundation.org
Cc: lucien.xin@gmail.com
Cc: nikolay@cumulusnetworks.com
Signed-off-by: Ivan Vecera <cera@cera.cz>
---
net/bridge/br_stp_if.c | 15 +++++----------
1 file changed, 5 insertions(+), 10 deletions(-)
diff --git a/net/bridge/br_stp_if.c b/net/bridge/br_stp_if.c
index 0db8102995a5..f137ebf27755 100644
--- a/net/bridge/br_stp_if.c
+++ b/net/bridge/br_stp_if.c
@@ -150,7 +150,6 @@ static int br_stp_call_user(struct net_bridge *br, char *arg)
static void br_stp_start(struct net_bridge *br)
{
- struct net_bridge_port *p;
int err = -ENOENT;
if (net_eq(dev_net(br->dev), &init_net))
@@ -169,11 +168,6 @@ static void br_stp_start(struct net_bridge *br)
if (!err) {
br->stp_enabled = BR_USER_STP;
br_debug(br, "userspace STP started\n");
-
- /* Stop hello and hold timers */
- del_timer(&br->hello_timer);
- list_for_each_entry(p, &br->port_list, list)
- del_timer(&p->hold_timer);
} else {
br->stp_enabled = BR_KERNEL_STP;
br_debug(br, "using kernel STP\n");
@@ -197,13 +191,14 @@ static void br_stp_stop(struct net_bridge *br)
br_err(br, "failed to stop userspace STP (%d)\n", err);
/* To start timers on any ports left in blocking */
- mod_timer(&br->hello_timer, jiffies + br->hello_time);
- list_for_each_entry(p, &br->port_list, list)
- mod_timer(&p->hold_timer,
- round_jiffies(jiffies + BR_HOLD_TIME));
spin_lock_bh(&br->lock);
br_port_state_selection(br);
spin_unlock_bh(&br->lock);
+ } else {
+ /* BR_KERNEL_STP - stop hello and hold timers */
+ del_timer(&br->hello_timer);
+ list_for_each_entry(p, &br->port_list, list)
+ del_timer(&p->hold_timer);
}
br->stp_enabled = BR_NO_STP;
--
2.13.0
^ permalink raw reply related [flat|nested] 14+ messages in thread* Re: [Bridge] [PATCH net] bridge: fix hello and hold timers starting/stopping
2017-05-19 16:25 ` Ivan Vecera
@ 2017-05-19 16:38 ` Stephen Hemminger
-1 siblings, 0 replies; 14+ messages in thread
From: Stephen Hemminger @ 2017-05-19 16:38 UTC (permalink / raw)
To: Ivan Vecera; +Cc: lucien.xin, nikolay, netdev, bridge, davem
On Fri, 19 May 2017 18:25:43 +0200
Ivan Vecera <cera@cera.cz> wrote:
> Current bridge code incorrectly handles starting/stopping of hello and
> hold timers during STP enable/disable.
>
> 1. Timers are stopped in br_stp_start() during NO_STP->USER_STP
> transition. This is not correct as the timers are stopped in NO_STP
> case.
>
> 2. Timers are started in br_stp_stop() during USER_STP->NO_STP transition.
> This is not also correct as the timers should be stopped in NO_STP
> state.
>
> 3. Timers are NOT stopped in br_stp_stop() during KERNEL_STP->NO_STP
> transition. They should be stopped as they are running in KERNEL_STP
> state and should not run in NO_STP case.
>
> The patch is a follow-up for "bridge: start hello_timer when enabling
> KERNEL_STP in br_stp_start" patch from Xin Long.
>
> Cc: davem@davemloft.net
> Cc: sashok@cumulusnetworks.com
> Cc: stephen@networkplumber.org
> Cc: bridge@lists.linux-foundation.org
> Cc: lucien.xin@gmail.com
> Cc: nikolay@cumulusnetworks.com
> Signed-off-by: Ivan Vecera <cera@cera.cz>
Overall, this looks correct but the wording of commit message
is too terse.
It would be better to add a more complete description of the impact
of this from a user's point of view. I am concerned that this
might have other side effects.
For example, what is the sequence of commands to validated this.
What is the impact, should this go to stable?
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [PATCH net] bridge: fix hello and hold timers starting/stopping
@ 2017-05-19 16:38 ` Stephen Hemminger
0 siblings, 0 replies; 14+ messages in thread
From: Stephen Hemminger @ 2017-05-19 16:38 UTC (permalink / raw)
To: Ivan Vecera; +Cc: lucien.xin, nikolay, netdev, bridge, davem
On Fri, 19 May 2017 18:25:43 +0200
Ivan Vecera <cera@cera.cz> wrote:
> Current bridge code incorrectly handles starting/stopping of hello and
> hold timers during STP enable/disable.
>
> 1. Timers are stopped in br_stp_start() during NO_STP->USER_STP
> transition. This is not correct as the timers are stopped in NO_STP
> case.
>
> 2. Timers are started in br_stp_stop() during USER_STP->NO_STP transition.
> This is not also correct as the timers should be stopped in NO_STP
> state.
>
> 3. Timers are NOT stopped in br_stp_stop() during KERNEL_STP->NO_STP
> transition. They should be stopped as they are running in KERNEL_STP
> state and should not run in NO_STP case.
>
> The patch is a follow-up for "bridge: start hello_timer when enabling
> KERNEL_STP in br_stp_start" patch from Xin Long.
>
> Cc: davem@davemloft.net
> Cc: sashok@cumulusnetworks.com
> Cc: stephen@networkplumber.org
> Cc: bridge@lists.linux-foundation.org
> Cc: lucien.xin@gmail.com
> Cc: nikolay@cumulusnetworks.com
> Signed-off-by: Ivan Vecera <cera@cera.cz>
Overall, this looks correct but the wording of commit message
is too terse.
It would be better to add a more complete description of the impact
of this from a user's point of view. I am concerned that this
might have other side effects.
For example, what is the sequence of commands to validated this.
What is the impact, should this go to stable?
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [Bridge] [PATCH net] bridge: fix hello and hold timers starting/stopping
2017-05-19 16:25 ` Ivan Vecera
@ 2017-05-19 16:51 ` Xin Long
-1 siblings, 0 replies; 14+ messages in thread
From: Xin Long @ 2017-05-19 16:51 UTC (permalink / raw)
To: Ivan Vecera; +Cc: Nikolay Aleksandrov, network dev, bridge, davem
On Sat, May 20, 2017 at 12:25 AM, Ivan Vecera <cera@cera.cz> wrote:
> Current bridge code incorrectly handles starting/stopping of hello and
> hold timers during STP enable/disable.
>
> 1. Timers are stopped in br_stp_start() during NO_STP->USER_STP
> transition. This is not correct as the timers are stopped in NO_STP
> case.
>
> 2. Timers are started in br_stp_stop() during USER_STP->NO_STP transition.
> This is not also correct as the timers should be stopped in NO_STP
> state.
>
> 3. Timers are NOT stopped in br_stp_stop() during KERNEL_STP->NO_STP
> transition. They should be stopped as they are running in KERNEL_STP
> state and should not run in NO_STP case.
>
> The patch is a follow-up for "bridge: start hello_timer when enabling
> KERNEL_STP in br_stp_start" patch from Xin Long.
>
> Cc: davem@davemloft.net
> Cc: sashok@cumulusnetworks.com
> Cc: stephen@networkplumber.org
> Cc: bridge@lists.linux-foundation.org
> Cc: lucien.xin@gmail.com
> Cc: nikolay@cumulusnetworks.com
> Signed-off-by: Ivan Vecera <cera@cera.cz>
> ---
> net/bridge/br_stp_if.c | 15 +++++----------
> 1 file changed, 5 insertions(+), 10 deletions(-)
>
> diff --git a/net/bridge/br_stp_if.c b/net/bridge/br_stp_if.c
> index 0db8102995a5..f137ebf27755 100644
> --- a/net/bridge/br_stp_if.c
> +++ b/net/bridge/br_stp_if.c
> @@ -150,7 +150,6 @@ static int br_stp_call_user(struct net_bridge *br, char *arg)
>
> static void br_stp_start(struct net_bridge *br)
> {
> - struct net_bridge_port *p;
> int err = -ENOENT;
>
> if (net_eq(dev_net(br->dev), &init_net))
> @@ -169,11 +168,6 @@ static void br_stp_start(struct net_bridge *br)
> if (!err) {
> br->stp_enabled = BR_USER_STP;
> br_debug(br, "userspace STP started\n");
> -
> - /* Stop hello and hold timers */
> - del_timer(&br->hello_timer);
> - list_for_each_entry(p, &br->port_list, list)
> - del_timer(&p->hold_timer);
> } else {
> br->stp_enabled = BR_KERNEL_STP;
> br_debug(br, "using kernel STP\n");
> @@ -197,13 +191,14 @@ static void br_stp_stop(struct net_bridge *br)
> br_err(br, "failed to stop userspace STP (%d)\n", err);
>
> /* To start timers on any ports left in blocking */
> - mod_timer(&br->hello_timer, jiffies + br->hello_time);
> - list_for_each_entry(p, &br->port_list, list)
> - mod_timer(&p->hold_timer,
> - round_jiffies(jiffies + BR_HOLD_TIME));
> spin_lock_bh(&br->lock);
> br_port_state_selection(br);
> spin_unlock_bh(&br->lock);
> + } else {
> + /* BR_KERNEL_STP - stop hello and hold timers */
> + del_timer(&br->hello_timer);
> + list_for_each_entry(p, &br->port_list, list)
> + del_timer(&p->hold_timer);
I'm thinking, what if the timers are running when deleting them ?
del_timer may not be going to delete it, and still have to stop itself
next time when br->stp_enabled = BR_NO_STP.
So do you think it's better to do nothing here and just leave it to be
stopped by itself when checking br->stp_enabled in
br_hello_timer_expired ?
> }
>
> br->stp_enabled = BR_NO_STP;
> --
> 2.13.0
>
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [PATCH net] bridge: fix hello and hold timers starting/stopping
@ 2017-05-19 16:51 ` Xin Long
0 siblings, 0 replies; 14+ messages in thread
From: Xin Long @ 2017-05-19 16:51 UTC (permalink / raw)
To: Ivan Vecera
Cc: network dev, davem, sashok, Stephen Hemminger, bridge,
Nikolay Aleksandrov
On Sat, May 20, 2017 at 12:25 AM, Ivan Vecera <cera@cera.cz> wrote:
> Current bridge code incorrectly handles starting/stopping of hello and
> hold timers during STP enable/disable.
>
> 1. Timers are stopped in br_stp_start() during NO_STP->USER_STP
> transition. This is not correct as the timers are stopped in NO_STP
> case.
>
> 2. Timers are started in br_stp_stop() during USER_STP->NO_STP transition.
> This is not also correct as the timers should be stopped in NO_STP
> state.
>
> 3. Timers are NOT stopped in br_stp_stop() during KERNEL_STP->NO_STP
> transition. They should be stopped as they are running in KERNEL_STP
> state and should not run in NO_STP case.
>
> The patch is a follow-up for "bridge: start hello_timer when enabling
> KERNEL_STP in br_stp_start" patch from Xin Long.
>
> Cc: davem@davemloft.net
> Cc: sashok@cumulusnetworks.com
> Cc: stephen@networkplumber.org
> Cc: bridge@lists.linux-foundation.org
> Cc: lucien.xin@gmail.com
> Cc: nikolay@cumulusnetworks.com
> Signed-off-by: Ivan Vecera <cera@cera.cz>
> ---
> net/bridge/br_stp_if.c | 15 +++++----------
> 1 file changed, 5 insertions(+), 10 deletions(-)
>
> diff --git a/net/bridge/br_stp_if.c b/net/bridge/br_stp_if.c
> index 0db8102995a5..f137ebf27755 100644
> --- a/net/bridge/br_stp_if.c
> +++ b/net/bridge/br_stp_if.c
> @@ -150,7 +150,6 @@ static int br_stp_call_user(struct net_bridge *br, char *arg)
>
> static void br_stp_start(struct net_bridge *br)
> {
> - struct net_bridge_port *p;
> int err = -ENOENT;
>
> if (net_eq(dev_net(br->dev), &init_net))
> @@ -169,11 +168,6 @@ static void br_stp_start(struct net_bridge *br)
> if (!err) {
> br->stp_enabled = BR_USER_STP;
> br_debug(br, "userspace STP started\n");
> -
> - /* Stop hello and hold timers */
> - del_timer(&br->hello_timer);
> - list_for_each_entry(p, &br->port_list, list)
> - del_timer(&p->hold_timer);
> } else {
> br->stp_enabled = BR_KERNEL_STP;
> br_debug(br, "using kernel STP\n");
> @@ -197,13 +191,14 @@ static void br_stp_stop(struct net_bridge *br)
> br_err(br, "failed to stop userspace STP (%d)\n", err);
>
> /* To start timers on any ports left in blocking */
> - mod_timer(&br->hello_timer, jiffies + br->hello_time);
> - list_for_each_entry(p, &br->port_list, list)
> - mod_timer(&p->hold_timer,
> - round_jiffies(jiffies + BR_HOLD_TIME));
> spin_lock_bh(&br->lock);
> br_port_state_selection(br);
> spin_unlock_bh(&br->lock);
> + } else {
> + /* BR_KERNEL_STP - stop hello and hold timers */
> + del_timer(&br->hello_timer);
> + list_for_each_entry(p, &br->port_list, list)
> + del_timer(&p->hold_timer);
I'm thinking, what if the timers are running when deleting them ?
del_timer may not be going to delete it, and still have to stop itself
next time when br->stp_enabled = BR_NO_STP.
So do you think it's better to do nothing here and just leave it to be
stopped by itself when checking br->stp_enabled in
br_hello_timer_expired ?
> }
>
> br->stp_enabled = BR_NO_STP;
> --
> 2.13.0
>
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [Bridge] [PATCH net] bridge: fix hello and hold timers starting/stopping
2017-05-19 16:51 ` Xin Long
@ 2017-05-19 16:55 ` Ivan Vecera
-1 siblings, 0 replies; 14+ messages in thread
From: Ivan Vecera @ 2017-05-19 16:55 UTC (permalink / raw)
To: Xin Long; +Cc: Nikolay Aleksandrov, network dev, bridge, davem
2017-05-19 18:51 GMT+02:00 Xin Long <lucien.xin@gmail.com>:
> On Sat, May 20, 2017 at 12:25 AM, Ivan Vecera <cera@cera.cz> wrote:
>> Current bridge code incorrectly handles starting/stopping of hello and
>> hold timers during STP enable/disable.
>>
>> 1. Timers are stopped in br_stp_start() during NO_STP->USER_STP
>> transition. This is not correct as the timers are stopped in NO_STP
>> case.
>>
>> 2. Timers are started in br_stp_stop() during USER_STP->NO_STP transition.
>> This is not also correct as the timers should be stopped in NO_STP
>> state.
>>
>> 3. Timers are NOT stopped in br_stp_stop() during KERNEL_STP->NO_STP
>> transition. They should be stopped as they are running in KERNEL_STP
>> state and should not run in NO_STP case.
>>
>> The patch is a follow-up for "bridge: start hello_timer when enabling
>> KERNEL_STP in br_stp_start" patch from Xin Long.
>>
>> Cc: davem@davemloft.net
>> Cc: sashok@cumulusnetworks.com
>> Cc: stephen@networkplumber.org
>> Cc: bridge@lists.linux-foundation.org
>> Cc: lucien.xin@gmail.com
>> Cc: nikolay@cumulusnetworks.com
>> Signed-off-by: Ivan Vecera <cera@cera.cz>
>> ---
>> net/bridge/br_stp_if.c | 15 +++++----------
>> 1 file changed, 5 insertions(+), 10 deletions(-)
>>
>> diff --git a/net/bridge/br_stp_if.c b/net/bridge/br_stp_if.c
>> index 0db8102995a5..f137ebf27755 100644
>> --- a/net/bridge/br_stp_if.c
>> +++ b/net/bridge/br_stp_if.c
>> @@ -150,7 +150,6 @@ static int br_stp_call_user(struct net_bridge *br, char *arg)
>>
>> static void br_stp_start(struct net_bridge *br)
>> {
>> - struct net_bridge_port *p;
>> int err = -ENOENT;
>>
>> if (net_eq(dev_net(br->dev), &init_net))
>> @@ -169,11 +168,6 @@ static void br_stp_start(struct net_bridge *br)
>> if (!err) {
>> br->stp_enabled = BR_USER_STP;
>> br_debug(br, "userspace STP started\n");
>> -
>> - /* Stop hello and hold timers */
>> - del_timer(&br->hello_timer);
>> - list_for_each_entry(p, &br->port_list, list)
>> - del_timer(&p->hold_timer);
>> } else {
>> br->stp_enabled = BR_KERNEL_STP;
>> br_debug(br, "using kernel STP\n");
>> @@ -197,13 +191,14 @@ static void br_stp_stop(struct net_bridge *br)
>> br_err(br, "failed to stop userspace STP (%d)\n", err);
>>
>> /* To start timers on any ports left in blocking */
>> - mod_timer(&br->hello_timer, jiffies + br->hello_time);
>> - list_for_each_entry(p, &br->port_list, list)
>> - mod_timer(&p->hold_timer,
>> - round_jiffies(jiffies + BR_HOLD_TIME));
>> spin_lock_bh(&br->lock);
>> br_port_state_selection(br);
>> spin_unlock_bh(&br->lock);
>> + } else {
>> + /* BR_KERNEL_STP - stop hello and hold timers */
>> + del_timer(&br->hello_timer);
>> + list_for_each_entry(p, &br->port_list, list)
>> + del_timer(&p->hold_timer);
> I'm thinking, what if the timers are running when deleting them ?
> del_timer may not be going to delete it, and still have to stop itself
> next time when br->stp_enabled = BR_NO_STP.
>
> So do you think it's better to do nothing here and just leave it to be
> stopped by itself when checking br->stp_enabled in
> br_hello_timer_expired ?
Yes, this kind of "lazy stopping" could be safer.
I.
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [PATCH net] bridge: fix hello and hold timers starting/stopping
@ 2017-05-19 16:55 ` Ivan Vecera
0 siblings, 0 replies; 14+ messages in thread
From: Ivan Vecera @ 2017-05-19 16:55 UTC (permalink / raw)
To: Xin Long
Cc: network dev, davem, sashok, Stephen Hemminger, bridge,
Nikolay Aleksandrov
2017-05-19 18:51 GMT+02:00 Xin Long <lucien.xin@gmail.com>:
> On Sat, May 20, 2017 at 12:25 AM, Ivan Vecera <cera@cera.cz> wrote:
>> Current bridge code incorrectly handles starting/stopping of hello and
>> hold timers during STP enable/disable.
>>
>> 1. Timers are stopped in br_stp_start() during NO_STP->USER_STP
>> transition. This is not correct as the timers are stopped in NO_STP
>> case.
>>
>> 2. Timers are started in br_stp_stop() during USER_STP->NO_STP transition.
>> This is not also correct as the timers should be stopped in NO_STP
>> state.
>>
>> 3. Timers are NOT stopped in br_stp_stop() during KERNEL_STP->NO_STP
>> transition. They should be stopped as they are running in KERNEL_STP
>> state and should not run in NO_STP case.
>>
>> The patch is a follow-up for "bridge: start hello_timer when enabling
>> KERNEL_STP in br_stp_start" patch from Xin Long.
>>
>> Cc: davem@davemloft.net
>> Cc: sashok@cumulusnetworks.com
>> Cc: stephen@networkplumber.org
>> Cc: bridge@lists.linux-foundation.org
>> Cc: lucien.xin@gmail.com
>> Cc: nikolay@cumulusnetworks.com
>> Signed-off-by: Ivan Vecera <cera@cera.cz>
>> ---
>> net/bridge/br_stp_if.c | 15 +++++----------
>> 1 file changed, 5 insertions(+), 10 deletions(-)
>>
>> diff --git a/net/bridge/br_stp_if.c b/net/bridge/br_stp_if.c
>> index 0db8102995a5..f137ebf27755 100644
>> --- a/net/bridge/br_stp_if.c
>> +++ b/net/bridge/br_stp_if.c
>> @@ -150,7 +150,6 @@ static int br_stp_call_user(struct net_bridge *br, char *arg)
>>
>> static void br_stp_start(struct net_bridge *br)
>> {
>> - struct net_bridge_port *p;
>> int err = -ENOENT;
>>
>> if (net_eq(dev_net(br->dev), &init_net))
>> @@ -169,11 +168,6 @@ static void br_stp_start(struct net_bridge *br)
>> if (!err) {
>> br->stp_enabled = BR_USER_STP;
>> br_debug(br, "userspace STP started\n");
>> -
>> - /* Stop hello and hold timers */
>> - del_timer(&br->hello_timer);
>> - list_for_each_entry(p, &br->port_list, list)
>> - del_timer(&p->hold_timer);
>> } else {
>> br->stp_enabled = BR_KERNEL_STP;
>> br_debug(br, "using kernel STP\n");
>> @@ -197,13 +191,14 @@ static void br_stp_stop(struct net_bridge *br)
>> br_err(br, "failed to stop userspace STP (%d)\n", err);
>>
>> /* To start timers on any ports left in blocking */
>> - mod_timer(&br->hello_timer, jiffies + br->hello_time);
>> - list_for_each_entry(p, &br->port_list, list)
>> - mod_timer(&p->hold_timer,
>> - round_jiffies(jiffies + BR_HOLD_TIME));
>> spin_lock_bh(&br->lock);
>> br_port_state_selection(br);
>> spin_unlock_bh(&br->lock);
>> + } else {
>> + /* BR_KERNEL_STP - stop hello and hold timers */
>> + del_timer(&br->hello_timer);
>> + list_for_each_entry(p, &br->port_list, list)
>> + del_timer(&p->hold_timer);
> I'm thinking, what if the timers are running when deleting them ?
> del_timer may not be going to delete it, and still have to stop itself
> next time when br->stp_enabled = BR_NO_STP.
>
> So do you think it's better to do nothing here and just leave it to be
> stopped by itself when checking br->stp_enabled in
> br_hello_timer_expired ?
Yes, this kind of "lazy stopping" could be safer.
I.
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [Bridge] [PATCH net] bridge: fix hello and hold timers starting/stopping
2017-05-19 16:25 ` Ivan Vecera
@ 2017-05-19 16:55 ` Nikolay Aleksandrov
-1 siblings, 0 replies; 14+ messages in thread
From: Nikolay Aleksandrov @ 2017-05-19 16:55 UTC (permalink / raw)
To: Ivan Vecera, netdev; +Cc: bridge, lucien.xin, davem
On 5/19/17 7:25 PM, Ivan Vecera wrote:
> Current bridge code incorrectly handles starting/stopping of hello and
> hold timers during STP enable/disable.
>
> 1. Timers are stopped in br_stp_start() during NO_STP->USER_STP
> transition. This is not correct as the timers are stopped in NO_STP
> case.
This really is a noop, but ok.
>
> 2. Timers are started in br_stp_stop() during USER_STP->NO_STP transition.
> This is not also correct as the timers should be stopped in NO_STP
> state.
Indeed, but the actual end result is almost as them being stopped because
in the timers there are specific checks if the STP == KERNEL_STP (see
br_transmit_config()) and the hold_timers will simply expire and not rearm
in any other mode. The only real problem is the hello_timer which continues
to rearm itself, but with Xin's earlier patch that is taken care of too.
>
> 3. Timers are NOT stopped in br_stp_stop() during KERNEL_STP->NO_STP
> transition. They should be stopped as they are running in KERNEL_STP
> state and should not run in NO_STP case.
Same comment as for point 2.
>
> The patch is a follow-up for "bridge: start hello_timer when enabling
> KERNEL_STP in br_stp_start" patch from Xin Long.
>
I'd say this is more of a cleanup/improvement after Xin's patch and thus would
suggest targeting net-next. The only real issue is fixed by his patch.
> Cc: davem@davemloft.net
> Cc: sashok@cumulusnetworks.com
> Cc: stephen@networkplumber.org
> Cc: bridge@lists.linux-foundation.org
> Cc: lucien.xin@gmail.com
> Cc: nikolay@cumulusnetworks.com
> Signed-off-by: Ivan Vecera <cera@cera.cz>
> ---
> net/bridge/br_stp_if.c | 15 +++++----------
> 1 file changed, 5 insertions(+), 10 deletions(-)
>
> diff --git a/net/bridge/br_stp_if.c b/net/bridge/br_stp_if.c
> index 0db8102995a5..f137ebf27755 100644
> --- a/net/bridge/br_stp_if.c
> +++ b/net/bridge/br_stp_if.c
> @@ -150,7 +150,6 @@ static int br_stp_call_user(struct net_bridge *br, char *arg)
>
> static void br_stp_start(struct net_bridge *br)
> {
> - struct net_bridge_port *p;
> int err = -ENOENT;
>
> if (net_eq(dev_net(br->dev), &init_net))
> @@ -169,11 +168,6 @@ static void br_stp_start(struct net_bridge *br)
> if (!err) {
> br->stp_enabled = BR_USER_STP;
> br_debug(br, "userspace STP started\n");
> -
> - /* Stop hello and hold timers */
> - del_timer(&br->hello_timer);
> - list_for_each_entry(p, &br->port_list, list)
> - del_timer(&p->hold_timer);
> } else {
> br->stp_enabled = BR_KERNEL_STP;
> br_debug(br, "using kernel STP\n");
> @@ -197,13 +191,14 @@ static void br_stp_stop(struct net_bridge *br)
> br_err(br, "failed to stop userspace STP (%d)\n", err);
>
> /* To start timers on any ports left in blocking */
> - mod_timer(&br->hello_timer, jiffies + br->hello_time);
> - list_for_each_entry(p, &br->port_list, list)
> - mod_timer(&p->hold_timer,
> - round_jiffies(jiffies + BR_HOLD_TIME));
> spin_lock_bh(&br->lock);
> br_port_state_selection(br);
> spin_unlock_bh(&br->lock);
> + } else {
> + /* BR_KERNEL_STP - stop hello and hold timers */
> + del_timer(&br->hello_timer);
> + list_for_each_entry(p, &br->port_list, list)
> + del_timer(&p->hold_timer);
> }
>
> br->stp_enabled = BR_NO_STP;
>
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [PATCH net] bridge: fix hello and hold timers starting/stopping
@ 2017-05-19 16:55 ` Nikolay Aleksandrov
0 siblings, 0 replies; 14+ messages in thread
From: Nikolay Aleksandrov @ 2017-05-19 16:55 UTC (permalink / raw)
To: Ivan Vecera, netdev; +Cc: davem, sashok, stephen, bridge, lucien.xin
On 5/19/17 7:25 PM, Ivan Vecera wrote:
> Current bridge code incorrectly handles starting/stopping of hello and
> hold timers during STP enable/disable.
>
> 1. Timers are stopped in br_stp_start() during NO_STP->USER_STP
> transition. This is not correct as the timers are stopped in NO_STP
> case.
This really is a noop, but ok.
>
> 2. Timers are started in br_stp_stop() during USER_STP->NO_STP transition.
> This is not also correct as the timers should be stopped in NO_STP
> state.
Indeed, but the actual end result is almost as them being stopped because
in the timers there are specific checks if the STP == KERNEL_STP (see
br_transmit_config()) and the hold_timers will simply expire and not rearm
in any other mode. The only real problem is the hello_timer which continues
to rearm itself, but with Xin's earlier patch that is taken care of too.
>
> 3. Timers are NOT stopped in br_stp_stop() during KERNEL_STP->NO_STP
> transition. They should be stopped as they are running in KERNEL_STP
> state and should not run in NO_STP case.
Same comment as for point 2.
>
> The patch is a follow-up for "bridge: start hello_timer when enabling
> KERNEL_STP in br_stp_start" patch from Xin Long.
>
I'd say this is more of a cleanup/improvement after Xin's patch and thus would
suggest targeting net-next. The only real issue is fixed by his patch.
> Cc: davem@davemloft.net
> Cc: sashok@cumulusnetworks.com
> Cc: stephen@networkplumber.org
> Cc: bridge@lists.linux-foundation.org
> Cc: lucien.xin@gmail.com
> Cc: nikolay@cumulusnetworks.com
> Signed-off-by: Ivan Vecera <cera@cera.cz>
> ---
> net/bridge/br_stp_if.c | 15 +++++----------
> 1 file changed, 5 insertions(+), 10 deletions(-)
>
> diff --git a/net/bridge/br_stp_if.c b/net/bridge/br_stp_if.c
> index 0db8102995a5..f137ebf27755 100644
> --- a/net/bridge/br_stp_if.c
> +++ b/net/bridge/br_stp_if.c
> @@ -150,7 +150,6 @@ static int br_stp_call_user(struct net_bridge *br, char *arg)
>
> static void br_stp_start(struct net_bridge *br)
> {
> - struct net_bridge_port *p;
> int err = -ENOENT;
>
> if (net_eq(dev_net(br->dev), &init_net))
> @@ -169,11 +168,6 @@ static void br_stp_start(struct net_bridge *br)
> if (!err) {
> br->stp_enabled = BR_USER_STP;
> br_debug(br, "userspace STP started\n");
> -
> - /* Stop hello and hold timers */
> - del_timer(&br->hello_timer);
> - list_for_each_entry(p, &br->port_list, list)
> - del_timer(&p->hold_timer);
> } else {
> br->stp_enabled = BR_KERNEL_STP;
> br_debug(br, "using kernel STP\n");
> @@ -197,13 +191,14 @@ static void br_stp_stop(struct net_bridge *br)
> br_err(br, "failed to stop userspace STP (%d)\n", err);
>
> /* To start timers on any ports left in blocking */
> - mod_timer(&br->hello_timer, jiffies + br->hello_time);
> - list_for_each_entry(p, &br->port_list, list)
> - mod_timer(&p->hold_timer,
> - round_jiffies(jiffies + BR_HOLD_TIME));
> spin_lock_bh(&br->lock);
> br_port_state_selection(br);
> spin_unlock_bh(&br->lock);
> + } else {
> + /* BR_KERNEL_STP - stop hello and hold timers */
> + del_timer(&br->hello_timer);
> + list_for_each_entry(p, &br->port_list, list)
> + del_timer(&p->hold_timer);
> }
>
> br->stp_enabled = BR_NO_STP;
>
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [Bridge] [PATCH net] bridge: fix hello and hold timers starting/stopping
2017-05-19 16:55 ` Nikolay Aleksandrov
@ 2017-05-19 17:06 ` Ivan Vecera
-1 siblings, 0 replies; 14+ messages in thread
From: Ivan Vecera @ 2017-05-19 17:06 UTC (permalink / raw)
To: Nikolay Aleksandrov; +Cc: Xin Long, network dev, bridge, David Miller
2017-05-19 18:55 GMT+02:00 Nikolay Aleksandrov <nikolay@cumulusnetworks.com>:
> On 5/19/17 7:25 PM, Ivan Vecera wrote:
>>
>> Current bridge code incorrectly handles starting/stopping of hello and
>> hold timers during STP enable/disable.
>>
>> 1. Timers are stopped in br_stp_start() during NO_STP->USER_STP
>> transition. This is not correct as the timers are stopped in NO_STP
>> case.
>
>
> This really is a noop, but ok.
Yes, stopping of stopped timers are safe but confusing.
>> 2. Timers are started in br_stp_stop() during USER_STP->NO_STP transition.
>> This is not also correct as the timers should be stopped in NO_STP
>> state.
>
>
> Indeed, but the actual end result is almost as them being stopped because
> in the timers there are specific checks if the STP == KERNEL_STP (see
> br_transmit_config()) and the hold_timers will simply expire and not rearm
> in any other mode. The only real problem is the hello_timer which continues
> to rearm itself, but with Xin's earlier patch that is taken care of too.
Yes, this is clean-up as well. The starting of timers are more
confusing than dangerous
but from a reader's point of view the starting of timers is non-sense
when STP is
going to be disabled.
>>
>> 3. Timers are NOT stopped in br_stp_stop() during KERNEL_STP->NO_STP
>> transition. They should be stopped as they are running in KERNEL_STP
>> state and should not run in NO_STP case.
>
>
> Same comment as for point 2.
This can be removed... and leave hello_timer handler to stop itself.
>> The patch is a follow-up for "bridge: start hello_timer when enabling
>> KERNEL_STP in br_stp_start" patch from Xin Long.
>>
>
> I'd say this is more of a cleanup/improvement after Xin's patch and thus
> would
> suggest targeting net-next. The only real issue is fixed by his patch.
Agree... will send resend against net-next.
Thanks for comments,
Ivan
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [PATCH net] bridge: fix hello and hold timers starting/stopping
@ 2017-05-19 17:06 ` Ivan Vecera
0 siblings, 0 replies; 14+ messages in thread
From: Ivan Vecera @ 2017-05-19 17:06 UTC (permalink / raw)
To: Nikolay Aleksandrov; +Cc: Xin Long, network dev, bridge, David Miller
2017-05-19 18:55 GMT+02:00 Nikolay Aleksandrov <nikolay@cumulusnetworks.com>:
> On 5/19/17 7:25 PM, Ivan Vecera wrote:
>>
>> Current bridge code incorrectly handles starting/stopping of hello and
>> hold timers during STP enable/disable.
>>
>> 1. Timers are stopped in br_stp_start() during NO_STP->USER_STP
>> transition. This is not correct as the timers are stopped in NO_STP
>> case.
>
>
> This really is a noop, but ok.
Yes, stopping of stopped timers are safe but confusing.
>> 2. Timers are started in br_stp_stop() during USER_STP->NO_STP transition.
>> This is not also correct as the timers should be stopped in NO_STP
>> state.
>
>
> Indeed, but the actual end result is almost as them being stopped because
> in the timers there are specific checks if the STP == KERNEL_STP (see
> br_transmit_config()) and the hold_timers will simply expire and not rearm
> in any other mode. The only real problem is the hello_timer which continues
> to rearm itself, but with Xin's earlier patch that is taken care of too.
Yes, this is clean-up as well. The starting of timers are more
confusing than dangerous
but from a reader's point of view the starting of timers is non-sense
when STP is
going to be disabled.
>>
>> 3. Timers are NOT stopped in br_stp_stop() during KERNEL_STP->NO_STP
>> transition. They should be stopped as they are running in KERNEL_STP
>> state and should not run in NO_STP case.
>
>
> Same comment as for point 2.
This can be removed... and leave hello_timer handler to stop itself.
>> The patch is a follow-up for "bridge: start hello_timer when enabling
>> KERNEL_STP in br_stp_start" patch from Xin Long.
>>
>
> I'd say this is more of a cleanup/improvement after Xin's patch and thus
> would
> suggest targeting net-next. The only real issue is fixed by his patch.
Agree... will send resend against net-next.
Thanks for comments,
Ivan
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [Bridge] [PATCH net] bridge: fix hello and hold timers starting/stopping
2017-05-19 16:25 ` Ivan Vecera
@ 2017-05-19 17:26 ` Xin Long
-1 siblings, 0 replies; 14+ messages in thread
From: Xin Long @ 2017-05-19 17:26 UTC (permalink / raw)
To: Ivan Vecera; +Cc: Nikolay Aleksandrov, network dev, bridge, davem
On Sat, May 20, 2017 at 12:25 AM, Ivan Vecera <cera@cera.cz> wrote:
[...]
> @@ -197,13 +191,14 @@ static void br_stp_stop(struct net_bridge *br)
> br_err(br, "failed to stop userspace STP (%d)\n", err);
>
> /* To start timers on any ports left in blocking */
> - mod_timer(&br->hello_timer, jiffies + br->hello_time);
> - list_for_each_entry(p, &br->port_list, list)
> - mod_timer(&p->hold_timer,
> - round_jiffies(jiffies + BR_HOLD_TIME));
> spin_lock_bh(&br->lock);
> br_port_state_selection(br);
> spin_unlock_bh(&br->lock);
> + } else {
> + /* BR_KERNEL_STP - stop hello and hold timers */
> + del_timer(&br->hello_timer);
> + list_for_each_entry(p, &br->port_list, list)
> + del_timer(&p->hold_timer);
> }
>
> br->stp_enabled = BR_NO_STP;
I have a question here, br->stp_enabled is not atomic, and it is being
changed without holding br->lock here, while it may be checked in
br_hello_timer_expired, is it safe ?
(sorry if I misunderstood or overthought about it)
> --
> 2.13.0
>
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [PATCH net] bridge: fix hello and hold timers starting/stopping
@ 2017-05-19 17:26 ` Xin Long
0 siblings, 0 replies; 14+ messages in thread
From: Xin Long @ 2017-05-19 17:26 UTC (permalink / raw)
To: Ivan Vecera; +Cc: Nikolay Aleksandrov, network dev, bridge, davem
On Sat, May 20, 2017 at 12:25 AM, Ivan Vecera <cera@cera.cz> wrote:
[...]
> @@ -197,13 +191,14 @@ static void br_stp_stop(struct net_bridge *br)
> br_err(br, "failed to stop userspace STP (%d)\n", err);
>
> /* To start timers on any ports left in blocking */
> - mod_timer(&br->hello_timer, jiffies + br->hello_time);
> - list_for_each_entry(p, &br->port_list, list)
> - mod_timer(&p->hold_timer,
> - round_jiffies(jiffies + BR_HOLD_TIME));
> spin_lock_bh(&br->lock);
> br_port_state_selection(br);
> spin_unlock_bh(&br->lock);
> + } else {
> + /* BR_KERNEL_STP - stop hello and hold timers */
> + del_timer(&br->hello_timer);
> + list_for_each_entry(p, &br->port_list, list)
> + del_timer(&p->hold_timer);
> }
>
> br->stp_enabled = BR_NO_STP;
I have a question here, br->stp_enabled is not atomic, and it is being
changed without holding br->lock here, while it may be checked in
br_hello_timer_expired, is it safe ?
(sorry if I misunderstood or overthought about it)
> --
> 2.13.0
>
^ permalink raw reply [flat|nested] 14+ messages in thread