Creating a packet as a string and then extracted its fields in C - sockets

Question

I need to implement my own packets to send over UDP. I decided that I would do this by sending a char buffer which has the sequence number, checksum, size, and the data of the packet which is bytes from a file. The string i'm sending separates each field by a semicolon. Then, when I receive the string (which is my packet) I want to extract each felid, use them accordingly (the sequence number, size, and checksum) and write the bytes to a file. So far I have wrote a method to create 100 packets, and I'm trying to extract and write the bytes to a file (I'm not doing it in the receiver yet, first I'm testing the parsing in the sender). For some reason, the bytes written to my file are incorrect and I'm getting "JPEG DATATSTREAM CONTAINS NO IMAGE" error when I try to open it.
struct packetNode{
char packet[1052]; // this is the entire packet data including the header
struct packetNode *next;
};
This is how I'm creating my packets:
//populate initial window of size 100
for(i = 0; i < 100; i++){
memset(&data[0], 0, sizeof(data));
struct packetNode *p; // create packet node
p = (struct packetNode *)malloc(sizeof(struct packetNode));
bytes = fread(data, 1, sizeof(data), fp); // read 1024 bytes from file into data buffer
int b = fwrite(data, 1, bytes, fpNew);
printf("read: %d\n", bytes);
memset(&p->packet[0], 0, sizeof(p->packet));
sprintf(p->packet, "%d;%d;%d;%s", s, 0, numPackets, data); // create packet
//calculate checksum
int check = checksum8(p->packet, sizeof(p->packet));
sprintf(p->packet, "%d;%d;%d;%s", s, check, numPackets, data); //put checksum in packet
s++; //incremenet sequence number
if(i == 0){
head = p;
tail = p;
tail->next = NULL;
}
else{
tail->next = p;
tail = p;
tail->next = NULL;
}
}
fclose(fp);
and this is where I parse and write the bytes to a file:
void test(){
FILE *fpNew = fopen("test.jpg", "w");
struct packetNode *ptr = head;
char *tokens;
int s, c, size;
int i = 0;
char data[1024];
while(ptr != NULL){
memset(&data[0], 0, sizeof(data));
tokens = strtok(ptr->packet,";");
s = atoi(tokens);
tokens = strtok(NULL, ";");
c = atoi(tokens);
tokens = strtok(NULL, ";");
size = atoi(tokens);
tokens = strtok(NULL, ";");
if(tokens != NULL)
strcpy(data, tokens);
printf("sequence: %d, checksum: %d, size: %d\n", s,c,size);
int b = fwrite(data, 1, sizeof(data), fpNew);
ptr = ptr->next;
i++;
}
fclose(fpNew);
}

Answer 1

Since there is transfer of binary data, a JPEG stream, this data cannot be treated as a string. It's better to go all binary. For instance, instead of
sprintf(p->packet, "%d;%d;%d;%s", s, 0, numPackets, data); // create packet
you would do
sprintf(p->packet, "%d;%d;%d;", s, 0, numPackets);
memcpy(&p->packet[strlen(p->packet)], data, bytes);
but this leads to parsing problems: we would need to change this:
tokens = strtok(NULL, ";");
if(tokens != NULL)
strcpy(data, tokens);
to something like this:
tokens += 1 + ( size < 10 ? 1 : size < 100 ? 2 : size < 1000 ? 3 : size < 10000 ? 4 : 5 );
memcpy(data, tokens, sizeof(data));
#Binary Protocol
It's easier to use a binary packet:
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#pragma push(pack,1)
typedef struct Packet {
int seq, maxseq, size;
unsigned short cksum;
unsigned char payload[];
} Packet;
#pragma pop(pack)
typedef struct PacketNode{
struct PacketNode * next;
Packet packet;
} PacketNode;
PacketNode * allocPacketNode(int maxPayloadSize) {
void * ptr = malloc(sizeof(PacketNode) + maxPayloadSize); // FIXME: error checking
memset(ptr, 0, sizeof(PacketNode) + maxPayloadSize); // mallocz wouldn't cooperate
return (PacketNode*) ptr;
}
PacketNode * prepare(FILE * fp, int fsize, int chunksize)
{
PacketNode * head = allocPacketNode(chunksize);
PacketNode * pn = head;
int rd, seq = 0;
int maxseq = fsize / chunksize + ( fsize % chunksize ? 1 : 0 );
while ( ( rd = fread(pn->packet.payload, 1, chunksize, fp ) ) > 0 )
{
printf("read %d bytes\n", rd);
pn->packet.seq = seq++;
pn->packet.maxseq = maxseq;
pn->packet.size = rd + sizeof(Packet);
pn->packet.cksum = 0;
pn->packet.cksum = ~checksum(&pn->packet, pn->packet.size);
if ( rd == chunksize )
pn = pn->next = allocPacketNode(chunksize);
}
return head;
}
int checksum(unsigned char * data, int len)
{
int sum = 0, i;
for ( i = 0; i < len; i ++ )
sum += data[i];
if ( sum > 0xffff )
sum = (sum & 0xffff) + (sum>>16);
return sum;
}
void test( PacketNode * ptr ) {
FILE *fpNew = fopen("test.jpg", "w");
while (ptr != NULL)
{
printf("sequence: %d/%d, checksum: %04x, size: %d\n",
ptr->packet.seq,
ptr->packet.maxseq,
ptr->packet.cksum,
ptr->packet.size - sizeof(Packet)
);
int b = fwrite(ptr->packet.payload, ptr->packet.size - sizeof(Packet), 1, fpNew);
ptr = ptr->next;
}
fclose(fpNew);
}
void fatal( const char * msg ) { printf("FATAL: %s\n", msg); exit(1); }
int main(int argc, char** argv)
{
if ( ! argv[1] ) fatal( "missing filename argument" );
FILE * fp = fopen( argv[1], "r" );
if ( ! fp ) fatal( "cannot open file" );
fseek( fp, 0, SEEK_END );
long fsize = ftell(fp);
fseek( fp, 0, SEEK_SET );
printf("Filesize: %d\n", fsize );
test( prepare(fp, fsize, 1024) );
}
The #pragma push(pack,1) changes how the compiler aligns fields of the struct. We want them to be compact, for network transport. Using 1 is byte-aligned. The #pragma pop(pack) restores the previous setting of the pack pragma.
A note on the checksum method
First we calculate the sum of all the bytes in the packet:
int sum = 0, i;
for ( i = 0; i < len; i ++ )
sum += data[i];
Since the packet uses an unsigned short (16 bits, max value 65535 or 0xffff) to store the checksum, we make sure that the result will fit:
if ( sum > 0xffff ) // takes up more than 16 bits.
Getting the low 16 bits of this int is done using sum & 0xffff, masking out everything but the low 16 bits. We could simply return this value, but we would loose the information from higher checksum bits. So, we will add the upper 16 bits to the lower 16 bits. Accessing the higher 16 bits is done by shifting the int to the right 16 bits, like so: sum >> 16. This is the same as sum / 65536, since 65536 = 216 = 1 << 16.
sum = (sum & 0xffff) + (sum>>16); // add low 16 bits and high 16 bits
I should note that network packet checksums are usually computed 2 bytes (or 'octets' as they like to call them there) at a time. For that, the data should be cast to an unsigned short *, and len should be divided by 2. However! len may be odd, so in that case we'll need to take special care of the last byte. For instance, assuming that the maximum packet size is even, and that the len argument is always <= max_packet_size:
unsigned short * in = (unsigned short *) data;
if ( len & 1 ) data[len] = 0; // make sure last byte is 0
len = (len + 1) / 2;
The rest of the checksum method can remain the same, except that it should operate on in instead of data.