[lustre-devel] [PATCH] staging: lustre: fix address space mismatches

[Greg KH]

On Tue, Dec 15, 2015 at 04:50:45PM +0000, Okash Khawaja wrote:
> This patch fixes address space warnings from sparse. Function
> lprocfs_write_helper() accepts user space buffer but was being 
> passed kernel space buffer by these functions:
> 
> contention_seconds_store()
> lockless_truncate_store()
> 
> Since these functions are used to implement show and store functions of
> lustre_attr object and since lustre_attr object is used to implement object
> inheritance through use of `container_of`, the address space warnings
> show up at multiple places inside driver's code base.
> 
> This patch creates a user space version of lustre_attr object lustre_attr_u.
> Keeping function names and signatures same - other than the __user attribute -
> ensures that object inheritance continues to work as it was, but address
> space discrepency is removed. That removes a whole bunch of address
> space warnings.
> 
> Signed-off-by: Okash Khawaja <okash.khawaja at gmail.com>
> ---
>  drivers/staging/lustre/lustre/include/lprocfs_status.h | 16 ++++++++++++++++
>  drivers/staging/lustre/lustre/osc/lproc_osc.c          | 12 ++++++------
>  2 files changed, 22 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/staging/lustre/lustre/include/lprocfs_status.h b/drivers/staging/lustre/lustre/include/lprocfs_status.h
> index f18c0c7..df6d9d5 100644
> --- a/drivers/staging/lustre/lustre/include/lprocfs_status.h
> +++ b/drivers/staging/lustre/lustre/include/lprocfs_status.h
> @@ -698,6 +698,22 @@ static struct lustre_attr lustre_attr_##name = __ATTR(name, mode, show, store)
>  #define LUSTRE_RO_ATTR(name) LUSTRE_ATTR(name, 0444, name##_show, NULL)
>  #define LUSTRE_RW_ATTR(name) LUSTRE_ATTR(name, 0644, name##_show, name##_store)
>  
> +struct lustre_attr_u {
> +	struct attribute attr;
> +	ssize_t (*show)(struct kobject *kobj, struct attribute *attr,
> +			char *buf);
> +	ssize_t (*store)(struct kobject *kobj, struct attribute *attr,
> +			 const char __user *buf, size_t len);

sysfs files do not have __user pointers, something is really wrong here
if that's the solution :(

See the other comments in the mailing list archives for how messed up
the __user and kernel pointers are in lustre, and how I'd not recommend
anyone trying to fix them, unless you are a lustre developer and can
test all of your changes...

sorry,

greg k-h