View Full Version : problem using memcpy - should I even be using it...?

August 23rd, 2009, 02:50 PM
I have a structure defined by

struct str {
void *base;
size_t len;

str is passed to a function as an argument. It contains len bytes of data, which is stored in base. Within that function I use this ssl function to encrypt the data:

EVP_CipherUpdate(&psk_info->ctx, outbuf, &outlen, &str[i].base[offset*1024], inlen)

This works by taking a block of inlen bytes from str[i].base starting at the index offset*1024, encrypting it, and then the encrypted data is put into outbuf.

I run this function within a for loop, offset starting at 0 and increasing by 1 each iteration, so that all the data in str[i].base is eventually encrypted, in blocks of 1024 bytes.

In each iteration this line is run:


so that each time a block is encrypted and put into outbuf, the contents of outbuf are put into new_buffer at the right place. Therefore when all the data has been encrypted, new_buffer contains all of the encrypted data in the order it was encrypted.

I then have to copy all of the data in new_buffer into new_str, which is of exactly the same type as str (where the data is stored). I use memcpy to do this:


where templen is the total number of bytes of encrypted data. However this keeps seg faulting. I've been playing around with it for quite a while but can't see a way of achieving what I want to do.

So basically what I want to end up with is all the encrypted data stored in new_str[i].base....

Any help is much appreciated :)

August 23rd, 2009, 03:17 PM
Could you provide the complete source code?

August 23rd, 2009, 03:35 PM
void * user_arg,
const structure * str)
structuretemp * psk_info;
int inlen, remlen;
int outlen;
const structure * new_iovec;
int new_buffer_len = 0;
unsigned char * new_buffer;
int i, offset = 0, templen = 0;
unsigned char outbuf[1024 + EVP_MAX_BLOCK_LENGTH];

psk_info = (structuretemp *) user_arg;

new_iovec = my_calloc(iovec_count, sizeof(structure));

for(i=0; i<iovec_count; i++)
new_buffer_len = iovec[i].iov_len * 1024;
new_buffer = globus_malloc(new_buffer_len);
remlen = str[i].len;

EVP_CipherInit_ex(&psk_info->ctx, EVP_rc2_cbc(), NULL, NULL, NULL, 1);
EVP_CIPHER_CTX_set_key_length(&psk_info->ctx, 10);
EVP_CipherInit_ex(&psk_info->ctx, NULL, NULL, psk_info->key, psk_info->iv, 1);

while(remlen > 0)
inlen = remlen;
if (inlen > 1024)
inlen = 1024;
if(!EVP_CipherUpdate(&psk_info->ctx, outbuf, &outlen, &str[i].base[offset*1024], inlen))
/* Error */
return 0;
templen += outlen;
offset += 1;
remlen -= inlen;
if(!EVP_CipherFinal_ex(&psk_info->ctx, outbuf, &outlen))
/* Error */
printf("error encrypting final block\n");
return 0;
templen += outlen;
//memcpy(new_str[i].base,new_buffer,templen); - it is this line im having trouble with

August 23rd, 2009, 03:58 PM
Could you please post the complete source code of the program? There are some variables I have no idea about like iovec_count and a few others.

August 23rd, 2009, 04:37 PM
I think I found the problem - I defined new_str as being constant so I couldnt change the data within it, but now I have removed the const it seems to be working.


August 23rd, 2009, 05:41 PM
thats strange as the const should have given you a compilation warning/error but not a segmentation fault.
do you compile with -Wall?

August 23rd, 2009, 06:57 PM
What's the sugnature of EVP_CipherUpdate. You make pointer aritmetic with void* which is not the best thing in the world. I could imagin if it's implicitly converted to int what coud happen.