netdev - Re: [RFC net-next v2 2/3] devlink: health: add remediation type

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 11 Mar 2021 16:32:44 +0200
From:   Eran Ben Elisha <eranbe@...dia.com>
To:     <netdev@...r.kernel.org>
CC:     <jiri@...nulli.us>, <saeedm@...dia.com>,
        <andrew.gospodarek@...adcom.com>, <jacob.e.keller@...el.com>,
        <guglielmo.morandin@...adcom.com>, <eugenem@...com>,
        <eranbe@...lanox.com>, Jakub Kicinski <kuba@...nel.org>
Subject: Re: [RFC net-next v2 2/3] devlink: health: add remediation type



On 3/11/2021 5:26 AM, Jakub Kicinski wrote:
> Currently devlink health does not give user any clear information
> of what kind of remediation ->recover callback will perform. This
> makes it difficult to understand the impact of enabling auto-
> -remediation, and the severity of the error itself.
> 
> To allow users to make more informed decision add a new remediation
> type attribute.
> 
> Note that we only allow one remediation type per reporter, this
> is intentional. devlink health is not built for mixing issues
> of different severity into one reporter since it only maintains
> one dump, of the first event and a single error counter.
> Nudging vendors towards categorizing issues beyond coarse
> groups is an added bonus.
> 
> Signed-off-by: Jakub Kicinski <kuba@...nel.org>
> ---
>   include/net/devlink.h        |  2 ++
>   include/uapi/linux/devlink.h | 25 +++++++++++++++++++++++++
>   net/core/devlink.c           |  7 ++++++-
>   3 files changed, 33 insertions(+), 1 deletion(-)
> 
> diff --git a/include/net/devlink.h b/include/net/devlink.h
> index b424328af658..72b37769761f 100644
> --- a/include/net/devlink.h
> +++ b/include/net/devlink.h
> @@ -659,6 +659,7 @@ struct devlink_health_reporter;
>   /**
>    * struct devlink_health_reporter_ops - Reporter operations
>    * @name: reporter name
> + * remedy: severity of the remediation required
>    * @recover: callback to recover from reported error
>    *           if priv_ctx is NULL, run a full recover
>    * @dump: callback to dump an object
> @@ -669,6 +670,7 @@ struct devlink_health_reporter;
>   
>   struct devlink_health_reporter_ops {
>   	char *name;
> +	enum devlink_health_remedy remedy;
>   	int (*recover)(struct devlink_health_reporter *reporter,
>   		       void *priv_ctx, struct netlink_ext_ack *extack);
>   	int (*dump)(struct devlink_health_reporter *reporter,
> diff --git a/include/uapi/linux/devlink.h b/include/uapi/linux/devlink.h
> index 41a6ea3b2256..8cd1508b525b 100644
> --- a/include/uapi/linux/devlink.h
> +++ b/include/uapi/linux/devlink.h
> @@ -534,6 +534,9 @@ enum devlink_attr {
>   	DEVLINK_ATTR_RELOAD_ACTION_STATS,       /* nested */
>   
>   	DEVLINK_ATTR_PORT_PCI_SF_NUMBER,	/* u32 */
> +
> +	DEVLINK_ATTR_HEALTH_REPORTER_REMEDY,	/* u32 */
> +
>   	/* add new attributes above here, update the policy in devlink.c */
>   
>   	__DEVLINK_ATTR_MAX,
> @@ -620,4 +623,26 @@ enum devlink_health_state {
>   	DL_HEALTH_STATE_ERROR,
>   };
>   
> +/**
> + * enum devlink_health_reporter_remedy - severity of remediation procedure
> + * @DL_HEALTH_REMEDY_NONE: transient error, no remediation required
> + * @DL_HEALTH_REMEDY_KICK: device stalled, processing will be re-triggered
> + * @DL_HEALTH_REMEDY_COMP_RESET: associated device component (e.g. device queue)
> + *			will be reset
> + * @DL_HEALTH_REMEDY_RESET: full device reset, will result in temporary
> + *			unavailability of the device, device configuration
> + *			should not be lost
> + * @DL_HEALTH_REMEDY_REINIT: device will be reinitialized and configuration lost
> + *
> + * Used in %DEVLINK_ATTR_HEALTH_REPORTER_REMEDY, categorizes the health reporter
> + * by the severity of the remediation.
> + */
> +enum devlink_health_remedy {
> +	DL_HEALTH_REMEDY_NONE = 1,

What is the reason zero is skipped?

> +	DL_HEALTH_REMEDY_KICK,
> +	DL_HEALTH_REMEDY_COMP_RESET,
> +	DL_HEALTH_REMEDY_RESET,
> +	DL_HEALTH_REMEDY_REINIT,
> +};
> +
>   #endif /* _UAPI_LINUX_DEVLINK_H_ */
> diff --git a/net/core/devlink.c b/net/core/devlink.c
> index 8e4e4bd7bb36..09d77d43ff63 100644
> --- a/net/core/devlink.c
> +++ b/net/core/devlink.c
> @@ -6095,7 +6095,8 @@ __devlink_health_reporter_create(struct devlink *devlink,
>   {
>   	struct devlink_health_reporter *reporter;
>   
> -	if (WARN_ON(graceful_period && !ops->recover))
> +	if (WARN_ON(graceful_period && !ops->recover) ||
> +	    WARN_ON(ops->recover && !ops->remedy))

It allows drivers to set recover callback and report DL_HEALTH_REMEDY_NONE.
Defining DL_HEALTH_REMEDY_NONE = 0  would make this if clause to catch it.

>   		return ERR_PTR(-EINVAL);
>   
>   	reporter = kzalloc(sizeof(*reporter), GFP_KERNEL);
> @@ -6265,6 +6266,10 @@ devlink_nl_health_reporter_fill(struct sk_buff *msg,
>   	if (nla_put_string(msg, DEVLINK_ATTR_HEALTH_REPORTER_NAME,
>   			   reporter->ops->name))
>   		goto reporter_nest_cancel;
> +	if (reporter->ops->remedy &&
> +	    nla_put_u32(msg, DEVLINK_ATTR_HEALTH_REPORTER_REMEDY,
> +			reporter->ops->remedy))
> +		goto reporter_nest_cancel;
>   	if (nla_put_u8(msg, DEVLINK_ATTR_HEALTH_REPORTER_STATE,
>   		       reporter->health_state))
>   		goto reporter_nest_cancel;
>