Correct, portable way to interpret buffer as a struct

c++networkingportabilitystrict-aliasing

The context of my problem is in network programming. Say I want to send messages over the network between two programs. For simplicity, let's say messages look like this, and byte-order is not a concern. I want to find a correct, portable, and efficient way to define these messages as C structures. I know of four approaches to this: explicit casting, casting through a union, copying, and marshaling.

struct message {
    uint16_t logical_id;
    uint16_t command;
};

Explicit Casting:

void send_message(struct message *msg) {
    uint8_t *bytes = (uint8_t *) msg;
    /* call to write/send/sendto here */
}

void receive_message(uint8_t *bytes, size_t len) {
    assert(len >= sizeof(struct message);
    struct message *msg = (struct message*) bytes;
    /* And now use the message */
    if (msg->command == SELF_DESTRUCT)
        /* ... */
}

My understanding is that send_message does not violate aliasing rules, because a byte/char pointer may alias any type. However, the converse is not true, and so receive_message violates aliasing rules and thus has undefined behavior.

Casting Through a Union:

union message_u {
    struct message m;
    uint8_t bytes[sizeof(struct message)];
};

void receive_message_union(uint8_t *bytes, size_t len) {
    assert(len >= sizeof(struct message);
    union message_u *msgu = bytes;
    /* And now use the message */
    if (msgu->m.command == SELF_DESTRUCT)
        /* ... */
}

However, this seems to violate the idea that a union only contains one of its members at any given time. Additionally, this seems like it could lead to alignment issues if the source buffer isn't aligned on a word/half-word boundary.

Copying:

void receive_message_copy(uint8_t *bytes, size_t len) {
    assert(len >= sizeof(struct message);
    struct message msg;
    memcpy(&msg, bytes, sizeof msg);
    /* And now use the message */
    if (msg.command == SELF_DESTRUCT)
        /* ... */
}

This seems guaranteed to produce the correct result, but of course I would greatly prefer to not have to copy the data.

Marshaling

void send_message(struct message *msg) {
    uint8_t bytes[4];
    bytes[0] = msg.logical_id >> 8;
    bytes[1] = msg.logical_id & 0xff;
    bytes[2] = msg.command >> 8;
    bytes[3] = msg.command & 0xff;
    /* call to write/send/sendto here */
}

void receive_message_marshal(uint8_t *bytes, size_t len) {
    /* No longer relying on the size of the struct being meaningful */
    assert(len >= 4);    
    struct message msg;
    msg.logical_id = (bytes[0] << 8) | bytes[1];    /* Big-endian */
    msg.command = (bytes[2] << 8) | bytes[3];
    /* And now use the message */
    if (msg.command == SELF_DESTRUCT)
        /* ... */
}

Still have to copy, but now decoupled from the representation of the struct. But now we need be explicit with the position and size of each member, and endian-ness is a much more obvious issue.

Related info:

What is the strict aliasing rule?

Aliasing array with pointer-to-struct without violating the standard

When is char* safe for strict pointer aliasing?

http://blog.llvm.org/2011/05/what-every-c-programmer-should-know.html

Real World Example

I've been looking for examples of networking code to see how this situation is handled elsewhere. The light-weight ip has a few similar cases. In the udp.c file lies the following code:

/**
 * Process an incoming UDP datagram.
 *
 * Given an incoming UDP datagram (as a chain of pbufs) this function
 * finds a corresponding UDP PCB and hands over the pbuf to the pcbs
 * recv function. If no pcb is found or the datagram is incorrect, the
 * pbuf is freed.
 *
 * @param p pbuf to be demultiplexed to a UDP PCB (p->payload pointing to the UDP header)
 * @param inp network interface on which the datagram was received.
 *
 */
void
udp_input(struct pbuf *p, struct netif *inp)
{
  struct udp_hdr *udphdr;

  /* ... */

  udphdr = (struct udp_hdr *)p->payload;

  /* ... */
}

where struct udp_hdr is a packed representation of a udp header and p->payload is of type void *. Going on my understanding and this answer, this is definitely [edit- not] breaking strict-aliasing and thus has undefined behavior.

Best Solution

I guess this is what I've been trying to avoid, but I finally went and took a look at the C99 standard myself. Here's what I've found (emphasis added):
§6.3.2.2 void

1 The (nonexistent) value of a void expression (an expression that has type void) shall not be used in any way, and implicit or explicit conversions (except to void) shall not be applied to such an expression. If an expression of any other type is evaluated as a void expression, its value or designator is discarded. (A void expression is evaluated for its side effects.)

§6.3.2.3 Pointers

1 A pointer to void may be converted to or from a pointer to any incomplete or object type. A pointer to any incomplete or object type may be converted to a pointer to void and back again; the result shall compare equal to the original pointer.

And §3.14

1 object
region of data storage in the execution environment, the contents of which can represent values

§6.5

An object shall have its stored value accessed only by an lvalue expression that has one of the following types:
a type compatible with the effective type of the object,
— a qualified version of a type compatible with the effective type of the object,
— a type that is the signed or unsigned type corresponding to the effective type of the object,
— a type that is the signed or unsigned type corresponding to a qualified version of the effective type of the object,
— an aggregate or union type that includes one of the aforementioned types among its
members (including, recursively, a member of a subaggregate or contained union), or
— a character type.

§6.5

The effective type of an object for an access to its stored value is the declared type of the
object, if any. If a value is stored into an object having no declared type through an lvalue having a type that is not a character type, then the type of the lvalue becomes the effective type of the object for that access and for subsequent accesses that do not modify the stored value. If a value is copied into an object having no declared type using memcpy or memmove, or is copied as an array of character type, then the effective type of the modified object for that access and for subsequent accesses that do not modify the value is the effective type of the object from which the value is copied, if it has one. For all other accesses to an object having no declared type, the effective type of the object is simply the type of the lvalue used for the access.

§J.2 Undefined Behavior

— An attempt is made to use the value of a void expression, or an implicit or explicit conversion (except to void) is applied to a void expression (6.3.2.2).

Conclusion

It is ok (well-defined) to cast to-and-from a void*, but not ok to use a value of type void in C99. Therefore the "real world example" is not undefined behavior. Therefore, the explicit casting method can be used with the following modification, as long as alignment, padding, and byte-order is taken care of:

void receive_message(void *bytes, size_t len) {
    assert(len >= sizeof(struct message);
    struct message *msg = (struct message*) bytes;
    /* And now use the message */
    if (msg->command == SELF_DESTRUCT)
        /* ... */
}
Related Question