Endianness and Socket Programming in C

I'm making a program that communicate with certain patient monitor using C sockets. I'm using connection-less sockets (UDP) to communicate with the device. But there is endianness mis-match between my computer and device and so far I was doing this to get parse response from the patient monitor:

recvfrom(int socket, char *buffer, size_t length, int flags,
             struct sockaddr *address, socklen_t *address_len);

Then I was casting the buffer directly to structure and using the ntohs and ntohl to change the byte ordering, for example:

struct A * a = (struct A *)buffer;
Struct A b;
b.v1 = ntohs(a->v1);
b.v2 = ntohl(a->v2);

After reading few examples over internet, I figured out that this may be wrong approach due to compiler dependent padding. But I'm not sure about it. I need simple way to un-marshall a buffer to a C structure with correct endiannes. The structure that I'm receiving can be of unpredictable length and little complex as well. I need fast and easy approach to do the un-marshalling.

I don't control sender. Sender is in network byte order. My question is only that:- Is is safe to cast a buffer to a structure and then use ntohs and ntohl on this casted structure to make host-byte order replica of this structure? Is it a fastest approach? If not, then what can be the fastest approach?


First of all, since a is apointer, your code should at minimum do this...

b.v1 = ntohs(a->v1);
b.v2 = ntohl(a->v2);

Second of all, the answer depends on circumstances and specification of the patient monitor. Who writes the code for the patient monitor? Is there a specification for it? What machine architecture is it using(in case you know), are you dealing with just one model or are there many versions of the monitor -- can you change the monitor, etc. etc.

Im going to assume that you cannot change the monitor, and that the byte order is documented somewhere -- and you may have to create your own unpack/pack routines, by doing byte addressing and bit manipulation -- that is unless you know that the format exactly matches that of "network" order -- and that padding of structs are the same in the network buffer.

So something like;

void unpack(struct *b, unsigned char *buffer)
{
   b->v1 = (buffer[0] <<8)|(buffer[1]);   
   b->v2 = (buffer[2] <<24)|(buffer[3] <<16)|(buffer[4] <<8)|(buffer[5]);
   etc....
}   

or like this if you prefer to you ntohX;

void unpack(struct *b, unsigned char *buffer)
{
   b->v1 = ntohs(buffer+0);   
   b->v2 = ntohl(buffer+2);
   etc....
}   

However if you do control the monitor code, then using a tool like protocol buffers would get rid of all the complexity of doing bit manipulation and worry about byte orders....

链接地址: http://www.djcxy.com/p/10616.html

上一篇: Heroku有哪些熵源可用?

下一篇: C语言中的字节顺序和Socket编程