Skip to contentSkip to conversation

ipc: explicitly clear stack memory for shminfo

PATCH In review Kees Cook 3 participants 4 messages Nov 16, 2010

lore
Share & tools
lore

Kees Cook

author
#0 lore raw

PATCH 1/1 ipc: explicitly clear stack memory for shminfo

This fixes a kernel stack memory contents leak by explicitly clearing the shminfo structure on the kernel stack before it is populated and copied back to userspace.

---
 ipc/shm.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
ipc/shm.c +1 -0
index 7d3bb22..1d3d41f 100644
@@ -531,6 +531,7 @@ static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminf
 	    {
 		struct shminfo out;
 
+		memset(&out, 0, sizeof(out));
 		if(in->shmmax > INT_MAX)
 			out.shmmax = INT_MAX;
 		else
Signature · 1 line
1.7.2.3
#1 lore raw
Quoted 8 lines · Kees Cook

This fixes a kernel stack memory contents leak by explicitly clearing the shminfo structure on the kernel stack before it is populated and copied back to userspace.

ipc/shm.c +0 -0
index 7d3bb22..1d3d41f 100644

ipc/shm.c static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminf · line 537

Quoted 7 lines
ipc/shm.c +1 -0
@@ -531,6 +531,7 @@ static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminf
 	    {
 		struct shminfo out;
 
+		memset(&out, 0, sizeof(out));
 		if(in->shmmax > INT_MAX)
 			out.shmmax = INT_MAX;
 		else

Hi Kees.

Trivial size optimization:

Perhaps it's better to use

	struct type var = {};
instead of
	struct type var;
	memset(&var, 0, sizeof(var));

At least for x86, gcc produces very slightly smaller code when there are other automatic variables like:

Larger:

	struct type var;
	struct type var2;
	memset(&var, 0, sizeof(var));
Smaller:
	struct type var = {};
	struct type var2;

On the other hand, memset is more obvious.

#2 lore raw
In reply to Kees Cook

Hi Kees,

Quoted 24 lines · Kees Cook

This fixes a kernel stack memory contents leak by explicitly clearing the shminfo structure on the kernel stack before it is populated and copied back to userspace.

---
 ipc/shm.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
ipc/shm.c +1 -0
index 7d3bb22..1d3d41f 100644
@@ -531,6 +531,7 @@ static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminf
 	    {
 		struct shminfo out;
 
+		memset(&out, 0, sizeof(out));
 		if(in->shmmax > INT_MAX)
 			out.shmmax = INT_MAX;
 		else
Signature · 1 line
1.7.2.3

Can you please clarify what fields (padding bytes?) are uninitialized here? I see this struct shminfo definition:

    /* Obsolete, used only for backwards compatibility */
    struct	shminfo {
        int shmmax;
        int shmmin;
        int shmmni;
        int shmseg;
        int shmall;
    };

And this filling:

		if(in->shmmax > INT_MAX)
			out.shmmax = INT_MAX;
		else
			out.shmmax = (int)in->shmmax;

		out.shmmin	= in->shmmin;
		out.shmmni	= in->shmmni;
		out.shmseg	= in->shmseg;
		out.shmall	= in->shmall; 

		return copy_to_user(buf, &out, sizeof(out));

As I see all five fields are set anyway, no padding bytes here, correct?

Thanks,

Signature · 2 lines
Vasiliy Kulikov
http://www.openwall.com - bringing security into open computing environments
author
#3 lore raw

Hi Vasiliy,

Quoted 4 lines · Vasiliy Kulikov

Can you please clarify what fields (padding bytes?) are uninitialized here? I see this struct shminfo definition:
[snip]
As I see all five fields are set anyway, no padding bytes here, correct?

You are correct, I was looking at shminfo64. Nevermind!

-Kees

Signature · 2 lines
Kees Cook
Ubuntu Security Team