Re: [PATCH v2 10/11] net/failsafe: fix sub-device ownership race

[Andrew Rybchenko <arybchenko@solarflare.com>]

On 05/10/2018 01:43 AM, Thomas Monjalon wrote:
> From: Matan Azrad <matan@mellanox.com>
>
> There is time between the sub-device port probing by the sub-device PMD
> to the sub-device port ownership taking by a fail-safe port.
>
> In this time, the port is available for the application usage. For
> example, the port will be exposed to the applications which use
> RTE_ETH_FOREACH_DEV iterator.
>
> Thus, ownership unaware applications may manage the port in this time
> what may cause a lot of problematic behaviors in the fail-safe
> sub-device initialization.
>
> Register to the ethdev NEW event to take the sub-device port ownership
> before it becomes exposed to the application.
>
> Fixes: a46f8d584eb8 ("net/failsafe: add fail-safe PMD")
> Cc: stable@dpdk.org
>
> Signed-off-by: Matan Azrad <matan@mellanox.com>
> Acked-by: Gaetan Rivet <gaetan.rivet@6wind.com>
> ---
>   drivers/net/failsafe/failsafe.c         | 22 ++++++++++---
>   drivers/net/failsafe/failsafe_eal.c     | 56 ++++++++++++++++++++++-----------
>   drivers/net/failsafe/failsafe_ether.c   | 23 ++++++++++++++
>   drivers/net/failsafe/failsafe_private.h |  4 +++
>   4 files changed, 82 insertions(+), 23 deletions(-)
>
> diff --git a/drivers/net/failsafe/failsafe.c b/drivers/net/failsafe/failsafe.c
> index fc989c4f5..c9d128de3 100644
> --- a/drivers/net/failsafe/failsafe.c
> +++ b/drivers/net/failsafe/failsafe.c
> @@ -204,16 +204,25 @@ fs_eth_dev_create(struct rte_vdev_device *vdev)
>   	}
>   	snprintf(priv->my_owner.name, sizeof(priv->my_owner.name),
>   		 FAILSAFE_OWNER_NAME);
> +	DEBUG("Failsafe port %u owner info: %s_%016"PRIX64, dev->data->port_id,
> +	      priv->my_owner.name, priv->my_owner.id);
> +	ret = rte_eth_dev_callback_register(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
> +					    failsafe_eth_new_event_callback,
> +					    dev);
> +	if (ret) {
> +		ERROR("Failed to register NEW callback");
> +		goto free_args;
> +	}
>   	ret = failsafe_eal_init(dev);
>   	if (ret)
> -		goto free_args;
> +		goto unregister_new_callback;
>   	ret = fs_mutex_init(priv);
>   	if (ret)
> -		goto free_args;
> +		goto unregister_new_callback;
>   	ret = failsafe_hotplug_alarm_install(dev);
>   	if (ret) {
>   		ERROR("Could not set up plug-in event detection");
> -		goto free_args;
> +		goto unregister_new_callback;
>   	}
>   	mac = &dev->data->mac_addrs[0];
>   	if (mac_from_arg) {
> @@ -226,7 +235,7 @@ fs_eth_dev_create(struct rte_vdev_device *vdev)
>   							       mac);
>   			if (ret) {
>   				ERROR("Failed to set default MAC address");
> -				goto free_args;
> +				goto unregister_new_callback;

It will fail to apply on next-net since there is cancel_alarm there now.
So, this hunk should be simply skipped.

>   			}
>   		}
>   	} else {
> @@ -261,6 +270,9 @@ fs_eth_dev_create(struct rte_vdev_device *vdev)
>   	};
>   	rte_eth_dev_probing_finish(dev);
>   	return 0;
> +unregister_new_callback:
> +	rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
> +					failsafe_eth_new_event_callback, dev);

There is cancel_alarm here in next-net.

>   free_args:
>   	failsafe_args_free(dev);
>   free_subs:
> @@ -280,6 +292,8 @@ fs_rte_eth_free(const char *name)
>   	dev = rte_eth_dev_allocated(name);
>   	if (dev == NULL)
>   		return -ENODEV;
> +	rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
> +					failsafe_eth_new_event_callback, dev);
>   	ret = failsafe_eal_uninit(dev);
>   	if (ret)
>   		ERROR("Error while uninitializing sub-EAL");
> diff --git a/drivers/net/failsafe/failsafe_eal.c b/drivers/net/failsafe/failsafe_eal.c
> index ce767703f..5672f3961 100644
> --- a/drivers/net/failsafe/failsafe_eal.c
> +++ b/drivers/net/failsafe/failsafe_eal.c
> @@ -18,8 +18,9 @@ fs_ethdev_portid_get(const char *name, uint16_t *port_id)
>   		return -EINVAL;
>   	}
>   	len = strlen(name);
> -	RTE_ETH_FOREACH_DEV(pid) {
> -		if (!strncmp(name, rte_eth_devices[pid].device->name, len)) {
> +	for (pid = 0; pid < RTE_MAX_ETHPORTS; pid++) {
> +		if (rte_eth_dev_is_valid_port(pid) &&
> +		    !strncmp(name, rte_eth_devices[pid].device->name, len)) {
>   			*port_id = pid;
>   			return 0;
>   		}
> @@ -41,6 +42,8 @@ fs_bus_init(struct rte_eth_dev *dev)
>   			continue;
>   		da = &sdev->devargs;
>   		if (fs_ethdev_portid_get(da->name, &pid) != 0) {
> +			struct rte_eth_dev_owner pid_owner;
> +
>   			ret = rte_eal_hotplug_add(da->bus->name,
>   						  da->name,
>   						  da->args);
> @@ -55,12 +58,26 @@ fs_bus_init(struct rte_eth_dev *dev)
>   				ERROR("sub_device %d init went wrong", i);
>   				return -ENODEV;
>   			}
> +			/*
> +			 * The NEW callback tried to take ownership, check
> +			 * whether it succeed or didn't.
> +			 */
> +			rte_eth_dev_owner_get(pid, &pid_owner);
> +			if (pid_owner.id != PRIV(dev)->my_owner.id) {
> +				INFO("sub_device %d owner(%s_%016"PRIX64") is not my,"
> +				     " owner(%s_%016"PRIX64"), will try again later",

Frankly speaking I don't understand what and why will change later.

> +				     i, pid_owner.name, pid_owner.id,
> +				     PRIV(dev)->my_owner.name,
> +				     PRIV(dev)->my_owner.id);
> +				continue;
> +			}
>   		} else {
> +			/* The sub-device port was found. */
>   			char devstr[DEVARGS_MAXLEN] = "";
>   			struct rte_devargs *probed_da =
>   					rte_eth_devices[pid].device->devargs;
>   
> -			/* Take control of device probed by EAL options. */
> +			/* Take control of probed device. */
>   			free(da->args);
>   			memset(da, 0, sizeof(*da));
>   			if (probed_da != NULL)
> @@ -77,22 +94,23 @@ fs_bus_init(struct rte_eth_dev *dev)
>   			}
>   			INFO("Taking control of a probed sub device"
>   			      " %d named %s", i, da->name);
> -		}
> -		ret = rte_eth_dev_owner_set(pid, &PRIV(dev)->my_owner);
> -		if (ret < 0) {
> -			INFO("sub_device %d owner set failed (%s),"
> -			     " will try again later", i, strerror(-ret));
> -			continue;
> -		} else if (strncmp(rte_eth_devices[pid].device->name, da->name,
> -			   strlen(da->name)) != 0) {
> -			/*
> -			 * The device probably was removed and its port id was
> -			 * reallocated before ownership set.
> -			 */
> -			rte_eth_dev_owner_unset(pid, PRIV(dev)->my_owner.id);
> -			INFO("sub_device %d was probably removed before taking"
> -			     " ownership, will try again later", i);
> -			continue;
> +			ret = rte_eth_dev_owner_set(pid, &PRIV(dev)->my_owner);
> +			if (ret < 0) {
> +				INFO("sub_device %d owner set failed (%s), "
> +				     "will try again later", i, strerror(-ret));
> +				continue;
> +			} else if (strncmp(rte_eth_devices[pid].device->name,
> +				   da->name, strlen(da->name)) != 0) {
> +				/*
> +				 * The device probably was removed and its port
> +				 * id was reallocated before ownership set.
> +				 */
> +				rte_eth_dev_owner_unset(pid,
> +							PRIV(dev)->my_owner.id);
> +				INFO("sub_device %d was removed before taking"
> +				     " ownership, will try again later", i);
> +				continue;
> +			}
>   		}
>   		ETH(sdev) = &rte_eth_devices[pid];
>   		SUB_ID(sdev) = i;
> diff --git a/drivers/net/failsafe/failsafe_ether.c b/drivers/net/failsafe/failsafe_ether.c
> index b414a7884..ebce87841 100644
> --- a/drivers/net/failsafe/failsafe_ether.c
> +++ b/drivers/net/failsafe/failsafe_ether.c
> @@ -463,3 +463,26 @@ failsafe_eth_lsc_event_callback(uint16_t port_id __rte_unused,
>   	else
>   		return 0;
>   }
> +
> +/* Take sub-device ownership before it becomes exposed to the application. */
> +int
> +failsafe_eth_new_event_callback(uint16_t port_id,
> +				enum rte_eth_event_type event __rte_unused,
> +				void *cb_arg, void *out __rte_unused)
> +{
> +	struct rte_eth_dev *fs_dev = cb_arg;
> +	struct sub_device *sdev;
> +	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
> +	size_t len = strlen(dev->device->name);
> +	uint8_t i;
> +
> +	FOREACH_SUBDEV_STATE(sdev, i, fs_dev, DEV_PARSED) {
> +		if (sdev->state >= DEV_PROBED)
> +			continue;
> +		if (strncmp(sdev->devargs.name, dev->device->name, len) != 0)

Shouldn't we ensure that name is exact match instead of checking
sdev->devargs.name prefix only.

> +			continue;
> +		rte_eth_dev_owner_set(port_id, &PRIV(fs_dev)->my_owner);

I think return value should be processed here.
If we really ignore it, it should be explained why.

> +		break;
> +	}
> +	return 0;
> +}
> diff --git a/drivers/net/failsafe/failsafe_private.h b/drivers/net/failsafe/failsafe_private.h
> index b54f8e336..cd8e0b8c9 100644
> --- a/drivers/net/failsafe/failsafe_private.h
> +++ b/drivers/net/failsafe/failsafe_private.h
> @@ -220,6 +220,10 @@ int failsafe_eth_rmv_event_callback(uint16_t port_id,
>   int failsafe_eth_lsc_event_callback(uint16_t port_id,
>   				    enum rte_eth_event_type event,
>   				    void *cb_arg, void *out);
> +int
> +failsafe_eth_new_event_callback(uint16_t port_id,
> +				enum rte_eth_event_type event,
> +				void *cb_arg, void *out);

Return value type and function name should be on the same line in
accordance with style used in this file. Just to be consistent.

[...]