2016-11-11 09:09:43

by Sachin Shukla

[permalink] [raw]
Subject: [PATCH] Kernel: Improvement in code readability when memdup_user_nul() fails.

From: "Sachin Shukla" <[email protected]>

There is no need to call kfree() if memdup_user_nul() fails, as no memory
was allocated and the error in the error-valued pointer should be returned.

Signed-off-by: Sachin Shukla <[email protected]>
---
kernel/user_namespace.c | 25 ++++++++++++++-----------
1 file changed, 14 insertions(+), 11 deletions(-)

diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
index 86b7854..a0ffbf0 100644
--- a/kernel/user_namespace.c
+++ b/kernel/user_namespace.c
@@ -672,28 +672,31 @@ static ssize_t map_write(struct file *file, const char __user *buf,
*/
mutex_lock(&userns_state_mutex);

- ret = -EPERM;
/* Only allow one successful write to the map */
- if (map->nr_extents != 0)
- goto out;
+ if (map->nr_extents != 0) {
+ mutex_unlock(&userns_state_mutex);
+ return -EPERM;
+ }

/*
* Adjusting namespace settings requires capabilities on the target.
*/
- if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
- goto out;
+ if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN)) {
+ mutex_unlock(&userns_state_mutex);
+ return -EPERM;
+ }

/* Only allow < page size writes at the beginning of the file */
- ret = -EINVAL;
- if ((*ppos != 0) || (count >= PAGE_SIZE))
- goto out;
+ if ((*ppos != 0) || (count >= PAGE_SIZE)) {
+ mutex_unlock(&userns_state_mutex);
+ return -EINVAL;
+ }

/* Slurp in the user data */
kbuf = memdup_user_nul(buf, count);
if (IS_ERR(kbuf)) {
- ret = PTR_ERR(kbuf);
- kbuf = NULL;
- goto out;
+ mutex_unlock(&userns_state_mutex);
+ return PTR_ERR(kbuf);
}

/* Parse the user data */
--
1.7.9.5


2016-11-11 10:02:29

by Vivek Gautam

[permalink] [raw]
Subject: Re: [PATCH] Kernel: Improvement in code readability when memdup_user_nul() fails.

On Fri, Nov 11, 2016 at 2:37 PM, Sachin Shukla <[email protected]> wrote:
> From: "Sachin Shukla" <[email protected]>
>
> There is no need to call kfree() if memdup_user_nul() fails, as no memory
> was allocated and the error in the error-valued pointer should be returned.
>
> Signed-off-by: Sachin Shukla <[email protected]>
> ---
> kernel/user_namespace.c | 25 ++++++++++++++-----------
> 1 file changed, 14 insertions(+), 11 deletions(-)
>
> diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
> index 86b7854..a0ffbf0 100644
> --- a/kernel/user_namespace.c
> +++ b/kernel/user_namespace.c
> @@ -672,28 +672,31 @@ static ssize_t map_write(struct file *file, const char __user *buf,
> */
> mutex_lock(&userns_state_mutex);
>
> - ret = -EPERM;
> /* Only allow one successful write to the map */
> - if (map->nr_extents != 0)
> - goto out;
> + if (map->nr_extents != 0) {
> + mutex_unlock(&userns_state_mutex);
> + return -EPERM;
> + }
>
> /*
> * Adjusting namespace settings requires capabilities on the target.
> */
> - if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
> - goto out;
> + if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN)) {
> + mutex_unlock(&userns_state_mutex);
> + return -EPERM;
> + }
>
> /* Only allow < page size writes at the beginning of the file */
> - ret = -EINVAL;
> - if ((*ppos != 0) || (count >= PAGE_SIZE))
> - goto out;
> + if ((*ppos != 0) || (count >= PAGE_SIZE)) {
> + mutex_unlock(&userns_state_mutex);
> + return -EINVAL;
> + }
>
> /* Slurp in the user data */
> kbuf = memdup_user_nul(buf, count);
> if (IS_ERR(kbuf)) {
> - ret = PTR_ERR(kbuf);
> - kbuf = NULL;
> - goto out;
> + mutex_unlock(&userns_state_mutex);
> + return PTR_ERR(kbuf);
> }

you may as well just move kfree() to a new error label, and
let 'out' do only mutex_unlock() and return ret.

Also remove the initialization of ret variable at the time of its
declaration. That doesn't make sense, since ret is initialized to
new values in this function anyways.


Thanks
Vivek

--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

2016-11-11 15:39:56

by Serge E. Hallyn

[permalink] [raw]
Subject: Re: [PATCH] Kernel: Improvement in code readability when memdup_user_nul() fails.

On Fri, Nov 11, 2016 at 02:37:15PM +0530, Sachin Shukla wrote:
> From: "Sachin Shukla" <[email protected]>
>
> There is no need to call kfree() if memdup_user_nul() fails, as no memory
> was allocated and the error in the error-valued pointer should be returned.

Hi,

in general, having a common exit path is considered more readable,
more easily reviewable, than having more exit paths. To this end,
initializing pointers to NULL and kfree()ing them at common exit
paths even when they may not have been alloc()ed yet is also often
seen as more readable.

> Signed-off-by: Sachin Shukla <[email protected]>

I do appreciate the work, and I recognize these things can be subjective,
but I would say

Nacked-by: Serge Hallyn <[email protected]>

> ---
> kernel/user_namespace.c | 25 ++++++++++++++-----------
> 1 file changed, 14 insertions(+), 11 deletions(-)
>
> diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
> index 86b7854..a0ffbf0 100644
> --- a/kernel/user_namespace.c
> +++ b/kernel/user_namespace.c
> @@ -672,28 +672,31 @@ static ssize_t map_write(struct file *file, const char __user *buf,
> */
> mutex_lock(&userns_state_mutex);
>
> - ret = -EPERM;
> /* Only allow one successful write to the map */
> - if (map->nr_extents != 0)
> - goto out;
> + if (map->nr_extents != 0) {
> + mutex_unlock(&userns_state_mutex);
> + return -EPERM;
> + }
>
> /*
> * Adjusting namespace settings requires capabilities on the target.
> */
> - if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
> - goto out;
> + if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN)) {
> + mutex_unlock(&userns_state_mutex);
> + return -EPERM;
> + }
>
> /* Only allow < page size writes at the beginning of the file */
> - ret = -EINVAL;
> - if ((*ppos != 0) || (count >= PAGE_SIZE))
> - goto out;
> + if ((*ppos != 0) || (count >= PAGE_SIZE)) {
> + mutex_unlock(&userns_state_mutex);
> + return -EINVAL;
> + }
>
> /* Slurp in the user data */
> kbuf = memdup_user_nul(buf, count);
> if (IS_ERR(kbuf)) {
> - ret = PTR_ERR(kbuf);
> - kbuf = NULL;
> - goto out;
> + mutex_unlock(&userns_state_mutex);
> + return PTR_ERR(kbuf);
> }
>
> /* Parse the user data */
> --
> 1.7.9.5

2016-11-11 16:58:51

by Eric W. Biederman

[permalink] [raw]
Subject: Re: [PATCH] Kernel: Improvement in code readability when memdup_user_nul() fails.

Sachin Shukla <[email protected]> writes:

> From: "Sachin Shukla" <[email protected]>
>
> There is no need to call kfree() if memdup_user_nul() fails, as no memory
> was allocated and the error in the error-valued pointer should be
> returned.

Nacked-by: "Eric W. Biederman" <[email protected]>

This is stupid, unnecesary gratituous change that makes the maintenance
of the code worse.


>
> Signed-off-by: Sachin Shukla <[email protected]>
> ---
> kernel/user_namespace.c | 25 ++++++++++++++-----------
> 1 file changed, 14 insertions(+), 11 deletions(-)
>
> diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
> index 86b7854..a0ffbf0 100644
> --- a/kernel/user_namespace.c
> +++ b/kernel/user_namespace.c
> @@ -672,28 +672,31 @@ static ssize_t map_write(struct file *file, const char __user *buf,
> */
> mutex_lock(&userns_state_mutex);
>
> - ret = -EPERM;
> /* Only allow one successful write to the map */
> - if (map->nr_extents != 0)
> - goto out;
> + if (map->nr_extents != 0) {
> + mutex_unlock(&userns_state_mutex);
> + return -EPERM;
> + }
>
> /*
> * Adjusting namespace settings requires capabilities on the target.
> */
> - if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
> - goto out;
> + if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN)) {
> + mutex_unlock(&userns_state_mutex);
> + return -EPERM;
> + }
>
> /* Only allow < page size writes at the beginning of the file */
> - ret = -EINVAL;
> - if ((*ppos != 0) || (count >= PAGE_SIZE))
> - goto out;
> + if ((*ppos != 0) || (count >= PAGE_SIZE)) {
> + mutex_unlock(&userns_state_mutex);
> + return -EINVAL;
> + }
>
> /* Slurp in the user data */
> kbuf = memdup_user_nul(buf, count);
> if (IS_ERR(kbuf)) {
> - ret = PTR_ERR(kbuf);
> - kbuf = NULL;
> - goto out;
> + mutex_unlock(&userns_state_mutex);
> + return PTR_ERR(kbuf);
> }
>
> /* Parse the user data */