A compiler might insert padding bytes to a structure to ensure that structure members appear in the correct location. Initializing the members of the structure does not always initialize the padding bytes.

According to ISO/IEC 9899:1999 (C99), in 6.2.6.1, paragraph 6,

When a value is stored in an object of structure or union type, including in a member object, the bytes of the object representation that correspond to any padding bytes take unspecified values.⁴²⁾

42) Thus, for example, structure assignment may be implemented element-at-a-time or via memcpy.

As a result of being uninitialized, padding bytes could have random data (sensitive data). This structure could be passed to functions that do not have privilege.

For Example, There have been instances in Linux kernel where uninitialized stack bytes are leaked to unprivileged users as a result of copying structures to user space.

Some Compilers do not initialize the padding bytes if all the members of the structure are initialized. 

Noncompliant Code Example

In this example, the padding bytes after char b are left uninitialized and are leaked.

struct test{
  int a;
  char b;
  int c;
};
.
.
struct test arg = {.a=1,.b=2,.c=3};
.
.
//perform operations on arg
.
.
//copy arg to user space
copy_to_user(ptr, &arg, sizeof(arg));

The padding bytes could be explicitly initialized using memset to zero as shown below.

struct test{
  int a;
  char b;
  int c;
};
.
.
struct test arg;
// Initialize arg using memset such that padding bytes are initialized
memset_s(&arg,0,sizeof arg);
.
//perform operations on arg
.
arg.a =1;
arg.b =2;
.
//copy arg to user space
copy_to_user(ptr, &arg, sizeof(arg));

Here, the compiler could implement arg.b =2 in the following way,

• setting the low byte of a 32-bit register to 2, leaving the high bytes unchanged,
• storing all 32 bits of the register into memory,

Thus leaking stack bytes to unprivileged user.

This may not be the case with all compilers. But the compilers are free to implement it in their own way.

So the above example could leak data under some specific compiler. 

Compliant Solution 1(Structure Packing - GCC)

GCC allows specifying attributes of variables and structures using the keyword _attribute((packed_)).

This means that that GCC will not add any padding bytes (for memory alignment) and make variables or fields immediately next to each other.

struct test{
 int a;
 char b;
 int c;
} __attribute__((__packed__));
.
struct test arg = {.a=1,.b=2,.c=3};
.
.
// perform operation on arg
.
.
// Copy arg to user space
copy_to_user(ptr, &arg, sizeof(arg));

Compliant Solution 2(Structure Packing - MSVC)

In case of MSVC, use #pragma pack() instead of _packed_ attribute to ensure no padding bytes are added.

#pragma pack(1) // 1 byte
struct test{
 int a;
 char b;
 int c;
};
.
struct test arg = {.a=1,.b=2,.c=3};
.
.
// perform operation on arg
.
.
// Copy arg to user space
copy_to_user(ptr, &arg, sizeof(arg));

pack takes effect at the first struct declaration after the pragma is seen. The alignment of a member will be on a boundary that is a multiple of 1 byte.

Compliant Solution 3(Adding Padding bytes)

The padding bytes could be explicitly declared. This should be done carefully based on the memory architecture.

The following solution assumes to be implemented in an IA32 machine.

#define JOIN(x, y) JOIN_AGAIN(x, y)
#define JOIN_AGAIN(x, y) x ## y
#define static_assert(e) \
typedef char JOIN(assertion_failed_at_line_, __LINE__) [(e) ? 1 : -1]
struct test{
 int a;
 char b;
 char padding_1, padding_2, padding_3;
 int c;
};
static_assert(offsetof(struct test, c) ==  (offsetof(struct test, padding_3) + 1) );
struct test arg = {.a=1,.b=2,.c=3};
arg.padding_1 = 0;
arg.padding_2 = 0;
arg.padding_3 = 0;
// perform operation on arg
.
.
// Copy arg to user space
copy_to_user(ptr, &arg, sizeof(arg));

The static_assert() macro accepts a constant expression e, which is evaluated as the first operand to the conditional operator. If e evaluates to nonzero, an array type with a size of 1 is defined; otherwise, an array type with a size of -1 is defined. Because it is invalid to declare an array with a negative size, the resulting type definition will be flagged by the compiler. The name of the array is used to indicate the location of the failed assertion.

See recommendation ?DCL03-C. Use a static assertion to test the value of a constant expression

For Example, In an IA32 machine,

The padding bytes inserted should ensure that no padding bytes are added and thus the expression in static_assert should be true

i.e.

offsetof(struct test, c ) = 8

offsetof(struct test, padding_3 ) = 7

But, if there is padding bytes added then,

The memory would not be continuous and the expression would return 0. 

This approach ensures that no padding bytes are inserted.

Compliant Solution 4

If memset of the original structure to zero (as mentioned in the Non-Compliant example 2) does not work under some compilers, then copy the original structure to an unsigned char memory and pass that memory to the user as shown below.

struct test{
 int a;
 char b;
 int c;
};
struct test  arg = { .a = 1, .b = 2, .c = 3 };
unsigned char r[sizeof arg];
.
.
// Do all operations on arg
.
.
// just before passing arg to the function
memset(r,0,sizeof r); // initialize everything to zero
memcpy(r+offsetof(struct test,a),&arg.a,sizeof arg.a);
memcpy(r+offsetof(struct test,b),&arg.b,sizeof arg.b);
memcpy(r+offsetof(struct test,c),&arg.c,sizeof arg.c);
//now pass r to the function
 copy_to_user(ptr, r, sizeof(r));

This ensures that no uninitialized padding bytes are copied to unprivileged users.

Risk Assessment

Related Guidelines

ISO/IEC 9899:1999 Section 6.2.6.1 p6

Bibliography

DCL20-C. Always specify void even if a function accepts no arguments      02. Declarations and Initialization (DCL)