I am working on a project in C to implement CBC mode on top of a skeleton code for DES with OpenSSL. We are not allowed to use a function that does the CBC mode automatically, in the sense that we must implement it ourselves. I am getting output but I have result files and my output is not matching up completely with the intended results. I also am stuck on figuring out how to pad the file to ensure all the blocks are of equal size, which is probably one of the reasons why I'm not receiving the correct output. Any help would be appreciated. Here's my modification of the skeleton code so far:
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <openssl/des.h>
#include <sys/time.h>
#include <unistd.h>
#define ENC 1
#define DEC 0
DES_key_schedule key;
int append(char*s, size_t size, char c) {
if(strlen(s) + 1 >= size) {
return 1;
}
int len = strlen(s);
s[len] = c;
s[len+1] = '\0';
return 0;
}
int getSize (char * s) {
char * t;
for (t = s; *t != '\0'; t++)
;
return t - s;
}
void strToHex(const_DES_cblock input, unsigned char *output) {
int arSize = 8;
unsigned int byte;
for(int i=0; i<arSize; i++) {
if(sscanf(input, "%2x", &byte) != 1) {
break;
}
output[i] = byte;
input += 2;
}
}
void doBitwiseXor(DES_LONG *xorValue, DES_LONG* data, const_DES_cblock roundOutput) {
DES_LONG temp[2];
memcpy(temp, roundOutput, 8*sizeof(unsigned char));
for(int i=0; i<2; i++) {
xorValue[i] = temp[i] ^ data[i];
}
}
void doCBCenc(DES_LONG *data, const_DES_cblock roundOutput, FILE *outFile) {
DES_LONG in[2];
doBitwiseXor(in, data, roundOutput);
DES_encrypt1(in,&key,ENC);
printf("ENCRYPTED\n");
printvalueOfDES_LONG(in);
printf("%s","\n");
fwrite(in, 8, 1, outFile);
memcpy(roundOutput, in, 2*sizeof(DES_LONG));
}
int main(int argc, char** argv)
{
const_DES_cblock cbc_key = {0x01,0x23,0x45,0x67,0x89,0xab,0xcd,0xef};
const_DES_cblock IV = {0x01,0x23,0x45,0x67,0x89,0xab,0xcd,0xef};
// Initialize the timing function
struct timeval start, end;
gettimeofday(&start, NULL);
int l;
if ((l = DES_set_key_checked(&cbc_key,&key)) != 0)
printf("\nkey error\n");
FILE *inpFile;
FILE *outFile;
inpFile = fopen("test.txt", "r");
outFile = fopen("test_results.txt", "wb");
if(inpFile && outFile) {
unsigned char ch;
// A char array that will hold all 8 ch values.
// each ch value is appended to this.
unsigned char eight_bits[8];
// counter for the loop that ensures that only 8 chars are done at a time.
int count = 0;
while(!feof(inpFile)) {
// read in a character
ch = fgetc(inpFile);
// print the character
printf("%c",ch);
// append the character to eight_bits
append(eight_bits,1,ch);
// increment the count so that we only go to 8.
count++;
const_DES_cblock roundOutput;
// When count gets to 8
if(count == 8) {
// for formatting
printf("%s","\n");
// Encrypt the eight characters and store them back in the char array.
//DES_encrypt1(eight_bits,&key,ENC);
doCBCenc(eight_bits, roundOutput, outFile);
// prints out the encrypted string
int k;
for(k = 0; k < getSize(eight_bits); k++){
printf("%c", eight_bits[k]);
}
// Sets count back to 0 so that we can do another 8 characters.
count = 0;
// so we just do the first 8. When everything works REMOVE THE BREAK.
//break;
}
}
} else {
printf("Error in opening file\n");
}
fclose(inpFile);
fclose(outFile);
// End the timing
gettimeofday(&end, NULL);
// Initialize seconds and micros to hold values for the time output
long seconds = (end.tv_sec - start.tv_sec);
long micros = ((seconds * 1000000) + end.tv_usec) - (start.tv_usec);
// Output the time
printf("The elapsed time is %d seconds and %d microseconds\n", seconds, micros);
}
Your crypto is at least half correct, but you have a lot of actual or potential other errors.
As you identified, raw CBC mode can only encrypt data which is a multiple of the block size, for DES 64 bits or 8 bytes (on most modern computers and all where you could use OpenSSL). In some applications this is okay; for example if the data is (always) an MD5 or SHA-256 or SHA-512 hash, or a GUID, or an IPv6 (binary) address, then it is a block multiple. But most applications want to handle at least any length in bytes, so they need to use some scheme to pad on encrypt and unpad on decrypt the last block (all blocks before the last already have the correct size). Many different schemes have been developed for this, so you need to know which to use. I assume this is a school assignment (since no real customer would set such a stupid and wasteful combination of requirements) and this should either have been specified or clearly left as a choice. One padding scheme very common today (although not for single-DES, because that is broken, unsafe, obsolete, and not common) is the one defined by PKCS5 and generalized by PKCS7 and variously called PKCS5, PKCS7, or PKCS5/7 padding, so I used that as an example.
Other than that:
you try to test feof(inpFile) before doing fgetc(inpFile). This doesn't work in C. It results in your code treating the low 8 bits of EOF (255 aka 0xFF on practically all implementations) as a valid data character added to the characters that were actually in the file. The common idiom is to store the return of getchar/getc/fgetc in a signed int and compare to EOF, but that would have required more changes so I used an alternate.
you don't initialize eight_bits which is a local-scope automatic duration variable, so its contents are undefined and depending on the implementation are often garbage, which means trying to 'append' to it by using strlen() to look for the end won't work right and might even crash. Although on some implementations at least some times it might happen to contain zero bytes, and 'work'. In addition it is possible in C for a byte read from a file (and stored here) to be \0 which will also make this work wrong, although if this file contains text, as its name suggests, it probably doesn't contain any \0 bytes.
once you fill eight_bits you write 'off-the-end' into element [8] which doesn't exist. Technically this is Undefined Behavior and anything at all can happen, traditionally expressed on Usenet as nasal demons. Plus after main finishes the first block it doesn't change anything in eight_bits so all further calls to append find it full and discard the new character.
while you could fix the above points separately, a much simple solution is available: you are already using count to count the number of bytes in the current block, so just use it as the subscript.
roundOutput is also an uninitialized local/auto variable within the loop, which is then used as the previous block for the CBC step, possibly with garbage or wrong value(s). And you don't use the IV at all, as is needed. You should allocate this before the loop (so it retains its value through all iterations) and initialize it to the IV, and then for each block in the loop your doCBCenc can properly XOR it to the new block and then leave the encrypted new block to be used next time.
your code labelled 'prints out the encrypted string' prints plaintext not ciphertext -- which is binary and shouldn't be printed directly anyway -- and is not needed because your file-read loop already echoes each character read. But if you do want to print a (validly null-terminated) string it's easier to just use fputs(s) or [f]printf([f,]"%s",s) or even fwrite(s,1,strlen(s),f).
your doCBCenc has a reference to printvalueofDES_LONG which isn't defined anywhere, and which along with two surrounding printf is clearly not needed.
you should use a cast to convert the first argument to doCBCenc -- this isn't strictly required but is good style and a good compiler (like mine) complains if you don't
finally, when an error occurs you usually print a message but then continue running, which will never work right and may produce symptoms that disguise the problem and make it hard to fix.
The below code fixes the above except that last (which would have been more work for less benefit) plus I removed routines that are now superfluous, and the timing code which is just silly: Unix already has builtin tools to measure and display process time more easily and reliably than writing code. Code I 'removed' is under #if 0 for reference, and code I added under #else or #if 1 except for the cast. The logic for PKCS5/7 padding is under #if MAYBE so it can be either selected or not. Some consider it better style to use sizeof(DES_block) or define a macro instead of the magic 8's, but I didn't bother -- especially since it would have required changes that aren't really necessary.
// SO70209636
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <openssl/des.h>
#include <sys/time.h>
#include <unistd.h>
#define ENC 1
#define DEC 0
DES_key_schedule key;
#if 0
int append(char*s, size_t size, char c) {
if(strlen(s) + 1 >= size) {
return 1;
}
int len = strlen(s);
s[len] = c;
s[len+1] = '\0';
return 0;
}
int getSize (char * s) {
char * t;
for (t = s; *t != '\0'; t++)
;
return t - s;
}
void strToHex(const_DES_cblock input, unsigned char *output) {
int arSize = 8;
unsigned int byte;
for(int i=0; i<arSize; i++) {
if(sscanf(input, "%2x", &byte) != 1) {
break;
}
output[i] = byte;
input += 2;
}
}
#endif
void doBitwiseXor(DES_LONG *xorValue, DES_LONG* data, const_DES_cblock roundOutput) {
DES_LONG temp[2];
memcpy(temp, roundOutput, 8*sizeof(unsigned char));
for(int i=0; i<2; i++) {
xorValue[i] = temp[i] ^ data[i];
}
}
void doCBCenc(DES_LONG *data, const_DES_cblock roundOutput, FILE *outFile) {
DES_LONG in[2];
doBitwiseXor(in, data, roundOutput);
DES_encrypt1(in,&key,ENC);
#if 0
printf("ENCRYPTED\n");
printvalueOfDES_LONG(in);
printf("%s","\n");
#endif
fwrite(in, 8, 1, outFile);
memcpy(roundOutput, in, 2*sizeof(DES_LONG));
}
int main(int argc, char** argv)
{
const_DES_cblock cbc_key = {0x01,0x23,0x45,0x67,0x89,0xab,0xcd,0xef};
const_DES_cblock IV = {0x01,0x23,0x45,0x67,0x89,0xab,0xcd,0xef};
#if 0
// Initialize the timing function
struct timeval start, end;
gettimeofday(&start, NULL);
#endif
int l;
if ((l = DES_set_key_checked(&cbc_key,&key)) != 0)
printf("\nkey error\n");
#if 1
DES_cblock roundOutput; // must be outside the loop
memcpy (roundOutput, IV, 8); // and initialized
#endif
FILE *inpFile;
FILE *outFile;
inpFile = fopen("test.txt", "r");
outFile = fopen("test.encrypt", "wb");
if(inpFile && outFile) {
unsigned char ch;
// A char array that will hold all 8 ch values.
// each ch value is appended to this.
unsigned char eight_bits[8];
// counter for the loop that ensures that only 8 chars are done at a time.
int count = 0;
#if 0
while(!feof(inpFile)) {
// read in a character
ch = fgetc(inpFile);
#else
while( ch = fgetc(inpFile), !feof(inpFile) ){
#endif
// print the character
printf("%c",ch);
#if 0
// append the character to eight_bits
append(eight_bits,1,ch);
// increment the count so that we only go to 8.
count++;
#else
eight_bits[count++] = ch;
#endif
#if 0
const_DES_cblock roundOutput;
#endif
// When count gets to 8
if(count == 8) {
// for formatting
printf("%s","\n");
// Encrypt the eight characters and store them back in the char array.
//DES_encrypt1(eight_bits,&key,ENC);
doCBCenc((DES_LONG*)eight_bits, roundOutput, outFile);
#if 0
// prints out the encrypted string
int k;
for(k = 0; k < getSize(eight_bits); k++){
printf("%c", eight_bits[k]);
}
#endif
// Sets count back to 0 so that we can do another 8 characters.
count = 0;
// so we just do the first 8. When everything works REMOVE THE BREAK.
//break;
}
}
#if MAYBE
memset (eight_bits+count, 8-count, 8-count); // PKCS5/7 padding
doCBCenc((DES_LONG*)eight_bits, roundOutput, outFile);
#endif
} else {
printf("Error in opening file\n");
}
fclose(inpFile);
fclose(outFile);
#if 0
// End the timing
gettimeofday(&end, NULL);
// Initialize seconds and micros to hold values for the time output
long seconds = (end.tv_sec - start.tv_sec);
long micros = ((seconds * 1000000) + end.tv_usec) - (start.tv_usec);
// Output the time
printf("The elapsed time is %d seconds and %d microseconds\n", seconds, micros);
#endif
}
PS: personally I wouldn't put the fwrite in doCBCenc; I would only do the encryption and let the caller do whatever I/O is appropriate which might in some cases not be fwrite. But what you have is not wrong for the requirements you apparently have.
I have a string creating function in C which accepts an array of structs as it's argument and outputs a string based on a predefined format (like a list of list in python).
Here's the function
typedef struct
{
PacketInfo_t PacketInfo;
char Gnss60[1900];
//and other stuff...
} Track_json_t;
typedef struct
{
double latitude;
double longitude;
} GPSPoint_t;
typedef struct
{
UInt16 GPS_StatusCode;
UInt32 fixtime;
GPSPoint_t point;
double altitude;
unsigned char GPS_Satilite_Num;
} GPS_periodic_t;
unsigned short SendTrack()
{
Track_json_t i_sTrack_S;
memset(&i_sTrack_S, 0x00, sizeof(Track_json_t));
getEvent_Track(&i_sTrack_S);
//Many other stuff added to the i_sTrack_S struct...
//Make a JSON format out of it
BuildTrackPacket_json(&i_sTrack_S, XPORT_MODE_GPRS);
}
Track_json_t *getEvent_Track(Track_json_t *trk)
{
GPS_periodic_t l_gps_60Sec[60];
memset(&l_gps_60Sec, 0x00,
sizeof(GPS_periodic_t) * GPS_PERIODIC_ARRAY_SIZE);
getLastMinGPSdata(l_gps_60Sec, o_gps_base);
get_gps60secString(l_gps_60Sec, trk->Gnss60);
return trk;
}
void get_gps60secString(GPS_periodic_t input[60], char *output)
{
int i = 0;
memcpy(output, "[", 1); ///< Copy the first char as [
char temp[31];
for (i = 0; i < 59; i++) { //Run for n-1 elements
memset(temp, 0, sizeof(temp));
snprintf(temp, sizeof(temp), "[%0.8f,%0.8f],",
input[i].point.latitude, input[i].point.longitude);
strncat(output, temp, sizeof(temp));
}
memset(temp, 0, sizeof(temp)); //assign last element
snprintf(temp, sizeof(temp), "[%0.8f,%0.8f]]",
input[i].point.latitude, input[i].point.longitude);
strncat(output, temp, sizeof(temp));
}
So the output of the function must be a string of format
[[12.12345678,12.12345678],[12.12345678,12.12345678],...]
But at times I get a string which looks like
[[12.12345678,12.12345678],[55.01[12.12345678,12.12345678],...]
[[21.28211567,84.13454083],[21.28211533,21.22[21.28211517,84.13454000],..]
Previously, I had a buffer overflow at the function get_gps60secString, I fixed that by using snprintf and strncat.
Note: This is an embedded application and this error occur once or twice a day (out of 1440 packets)
Question
1. Could this be caused by an interrupt during the snprintf/strncat process?
2. Could this be caused by a memory leak, overwriting the stack or some other segmentation issue caused else where?
Basically I would like to understand what might be causing a corrupt string.
Having a hard time finding the cause and fixing this bug.
EDIT:
I used chux's function. Below is the Minimal, Complete, and Verifiable Example
/*
* Test code for SO question https://stackoverflow.com/questions/5216413
* A Minimal, Complete, and Verifiable Example
*/
#include <stdio.h>
#include <stdlib.h>
#include <memory.h>
#include <stdbool.h>
#include <signal.h>
#include <unistd.h>
typedef unsigned short UInt16;
typedef unsigned long UInt32;
#define GPS_PERIODIC_ARRAY_SIZE 60
#define GPS_STRING_SIZE 1900
/* ---------------------- Data Structs --------------------------*/
typedef struct
{
char Gnss60[GPS_STRING_SIZE];
} Track_json_t;
typedef struct
{
double latitude;
double longitude;
} GPSPoint_t;
typedef struct
{
UInt16 GPS_StatusCode;
UInt32 fixtime;
GPSPoint_t point;
double altitude;
unsigned char GPS_Satilite_Num;
} GPS_periodic_t;
/* ----------------------- Global --------------------------------*/
FILE *fptr; //Global file pointer
int res = 0;
int g_last = 0;
GPS_periodic_t l_gps_60Sec[GPS_PERIODIC_ARRAY_SIZE];
/* ----------------------- Function defs --------------------------*/
/* At signal interrupt this function is called.
* Flush and close the file. And safly exit the program */
void userSignalInterrupt()
{
fflush(fptr);
fclose(fptr);
res = 1;
exit(0);
}
/* #brief From the array of GPS structs we create a string of the format
* [[lat,long],[lat,long],..]
* #param input The input array of GPS structs
* #param output The output string which will contain lat, long
* #param sz Size left in the output buffer
* #return 0 Successfully completed operation
* 1 Failed / Error
*/
int get_gps60secString(GPS_periodic_t input[GPS_PERIODIC_ARRAY_SIZE],
char *output, size_t sz)
{
int cnt = snprintf(output, sz, "[");
if (cnt < 0 || cnt >= sz)
return 1;
output += cnt;
sz -= cnt;
int i = 0;
for (i = 0; i < GPS_PERIODIC_ARRAY_SIZE; i++) {
cnt = snprintf(output, sz, "[%0.8f,%0.8f]%s",
input[i].point.latitude, input[i].point.longitude,
i + 1 == GPS_PERIODIC_ARRAY_SIZE ? "" : ",");
if (cnt < 0 || cnt >= sz)
return 1;
output += cnt;
sz -= cnt;
}
cnt = snprintf(output, sz, "]");
if (cnt < 0 || cnt >= sz)
return 1;
return 0; // no error
}
/* #brief Create a GPS struct with data for testing. It will populate the
* point field of GPS_periodic_t. Lat starts from 0.0 and increases by 1*10^(-8)
* and Long will dstart at 99.99999999 and dec by 1*10^(-8)
*
* #param o_gps_60sec Output array of GPS structs
*/
void getLastMinGPSdata(GPS_periodic_t *o_gps_60sec)
{
//Fill in GPS related data here
int i = 0;
double latitude = o_gps_60sec[0].point.latitude;
double longitude = o_gps_60sec[0].point.longitude;
for (i = 0; i < 60; i++)
{
o_gps_60sec[i].point.latitude = latitude + (0.00000001 * (float)g_last +
0.00000001 * (float)i);
o_gps_60sec[i].point.longitude = longitude - (0.00000001 * (float)g_last +
0.00000001 * (float)i);
}
g_last = 60;
}
/* #brief Get the GPS data and convert it into a string
* #param trk Track structure with GPS string
*/
int getEvent_Track(Track_json_t *trk)
{
getLastMinGPSdata(l_gps_60Sec);
get_gps60secString(l_gps_60Sec, trk->Gnss60, GPS_STRING_SIZE);
return 0;
}
int main()
{
fptr = fopen("gpsAno.txt", "a");
if (fptr == NULL) {
printf("Error!!\n");
exit(1);
}
//Quit at signal interrupt
signal(SIGINT, userSignalInterrupt);
Track_json_t trk;
memset(&l_gps_60Sec, 0x00, sizeof(GPS_periodic_t) * GPS_PERIODIC_ARRAY_SIZE);
//Init Points to be zero and 99.99999999
int i = 0;
for (i = 0; i < 60; i++) {
l_gps_60Sec[i].point.latitude = 00.00000000;
l_gps_60Sec[i].point.longitude = 99.99999999;
}
do {
memset(&trk, 0, sizeof(Track_json_t));
getEvent_Track(&trk);
//Write to file
fprintf(fptr, "%s", trk.Gnss60);
fflush(fptr);
sleep(1);
} while (res == 0);
//close and exit
fclose(fptr);
return 0;
}
Note: Error was not recreated in the above code.
Because this doesn't have the strcat pitfalls.
I tested this function in the embedded application.
Through this I was able to find that the snprintf returns an error and the string created ended up to be:
[17.42401750,78.46098717],[17.42402083,53.62
It ended there (because of the return 1).
Does this mean that the data which was passed to snprints corrupted? It's a float value. How can it get corrupted?
Solution
The error have not been seen since I changed the sprintf function with one that doesn't directly deal with 64 bits of data.
Here's the function modp_dtoa2
/** \brief convert a floating point number to char buffer with a
* variable-precision format, and no trailing zeros
*
* This is similar to "%.[0-9]f" in the printf style, except it will
* NOT include trailing zeros after the decimal point. This type
* of format oddly does not exists with printf.
*
* If the input value is greater than 1<<31, then the output format
* will be switched exponential format.
*
* \param[in] value
* \param[out] buf The allocated output buffer. Should be 32 chars or more.
* \param[in] precision Number of digits to the right of the decimal point.
* Can only be 0-9.
*/
void modp_dtoa2(double value, char* str, int prec)
{
/* if input is larger than thres_max, revert to exponential */
const double thres_max = (double)(0x7FFFFFFF);
int count;
double diff = 0.0;
char* wstr = str;
int neg= 0;
int whole;
double tmp;
uint32_t frac;
/* Hacky test for NaN
* under -fast-math this won't work, but then you also won't
* have correct nan values anyways. The alternative is
* to link with libmath (bad) or hack IEEE double bits (bad)
*/
if (! (value == value)) {
str[0] = 'n'; str[1] = 'a'; str[2] = 'n'; str[3] = '\0';
return;
}
if (prec < 0) {
prec = 0;
} else if (prec > 9) {
/* precision of >= 10 can lead to overflow errors */
prec = 9;
}
/* we'll work in positive values and deal with the
negative sign issue later */
if (value < 0) {
neg = 1;
value = -value;
}
whole = (int) value;
tmp = (value - whole) * pow10[prec];
frac = (uint32_t)(tmp);
diff = tmp - frac;
if (diff > 0.5) {
++frac;
/* handle rollover, e.g. case 0.99 with prec 1 is 1.0 */
if (frac >= pow10[prec]) {
frac = 0;
++whole;
}
} else if (diff == 0.5 && ((frac == 0) || (frac & 1))) {
/* if halfway, round up if odd, OR
if last digit is 0. That last part is strange */
++frac;
}
/* for very large numbers switch back to native sprintf for exponentials.
anyone want to write code to replace this? */
/*
normal printf behavior is to print EVERY whole number digit
which can be 100s of characters overflowing your buffers == bad
*/
if (value > thres_max) {
sprintf(str, "%e", neg ? -value : value);
return;
}
if (prec == 0) {
diff = value - whole;
if (diff > 0.5) {
/* greater than 0.5, round up, e.g. 1.6 -> 2 */
++whole;
} else if (diff == 0.5 && (whole & 1)) {
/* exactly 0.5 and ODD, then round up */
/* 1.5 -> 2, but 2.5 -> 2 */
++whole;
}
//vvvvvvvvvvvvvvvvvvv Diff from modp_dto2
} else if (frac) {
count = prec;
// now do fractional part, as an unsigned number
// we know it is not 0 but we can have leading zeros, these
// should be removed
while (!(frac % 10)) {
--count;
frac /= 10;
}
//^^^^^^^^^^^^^^^^^^^ Diff from modp_dto2
// now do fractional part, as an unsigned number
do {
--count;
*wstr++ = (char)(48 + (frac % 10));
} while (frac /= 10);
// add extra 0s
while (count-- > 0) *wstr++ = '0';
// add decimal
*wstr++ = '.';
}
// do whole part
// Take care of sign
// Conversion. Number is reversed.
do *wstr++ = (char)(48 + (whole % 10)); while (whole /= 10);
if (neg) {
*wstr++ = '-';
}
*wstr='\0';
strreverse(str, wstr-1);
}
Here's (part of) my unabashedly opinionated guide on safe string handling in C. Normally, I would promote dynamic memory allocation instead of fixed-length strings, but in this case I'm assuming that in the embedded environment that might be problematic. (Although assumptions like that should always be checked.)
So, first things first:
Any function which creates a string in a buffer must be told explicitly how long the buffer is. This is non-negotiable.
As should be obvious, it's impossible for a function filling a buffer to check for buffer overflow unless it knows where the buffer ends. "Hope that the buffer is long enough" is not a viable strategy. "Document the needed buffer length" would be fine if everyone carefully read the documentation (they don't) and if the required length never changes (it will). The only thing that's left is an extra argument, which should be of type size_t (because that's the type of buffer lengths in the C library functions which require lengths).
Forget that strncpy and strncat exist. Also forget about strcat. They are not your friends.
strncpy is designed for a specific use case: ensuring that an entire fixed-length buffer is initialised. It is not designed for normal strings, and since it doesn't guarantee that the output is NUL-terminated, it doesn't produce a string.
If you're going to NUL-terminate yourself anyway, you might as well use memmove, or memcpy if you know that the source and destination don't overlap, which should almost always be the case. Since you'll want the memmove to stop at the end of the string for short strings (which strncpy does not do), measure the string length first with strnlen: strnlen takes a maximum length, which is precisely what you want in the case that you are going move a maximum number of characters.
Sample code:
/* Safely copy src to dst where dst has capacity dstlen. */
if (dstlen) {
/* Adjust to_move will have maximum value dstlen - 1 */
size_t to_move = strnlen(src, dstlen - 1);
/* copy the characters */
memmove(dst, src, to_move);
/* NUL-terminate the string */
dst[to_move] = 0;
}
strncat has a slightly more sensible semantic, but it's practically never useful because in order to use it, you already have to know how many bytes you could copy. In order to know that, in practice, you need to know how much space is left in your output buffer, and to know that you need to know where in the output buffer the copy will start. [Note 1]. But if you already know where the copy will start, what's the point of searching through the buffer from the beginning to find the copy point? And if you do let strncat do the search, how sure are you that your previously computed start point is correct?
In the above code snippet, we already computed the length of the copy. We can extend that to do an append without rescanning:
/* Safely copy src1 and then src2 to dst where dst has capacity dstlen. */
/* Assumes that src1 and src2 are not contained in dst. */
if (dstlen) {
/* Adjust to_move will have maximum value dstlen - 1 */
size_t to_move = strnlen(src1, dstlen - 1);
/* Copy the characters from src1 */
memcpy(dst, src1, to_move);
/* Adjust the output pointer and length */
dst += to_move;
dstlen -= to_move;
/* Now safely copy src2 to just after src1. */
to_move = strnlen(src2, dstlen - 1);
memcpy(dst, src2, to_move);
/* NUL-terminate the string */
dst[to_move] = 0;
}
It might be that we want the original values of dst and dstlen after creating the string, and it might also be that we want to know how many bytes we inserted into dst in all. In that case, we would probably want to make copies of those variables before doing the copies, and save the cumulative sum of moves.
The above assumes that we're starting with an empty output buffer, but perhaps that isn't the case. Since we still need to know where the copy will start in order to know how many characters we can put at the end, we can still use memcpy; we just need to scan the output buffer first to find the copy point. (Only do this if there is no alternative. Doing it in a loop instead of recording the next copy point is Shlemiel the Painter's algorithm.)
/* Safely append src to dst where dst has capacity dstlen and starts
* with a string of unknown length.
*/
if (dstlen) {
/* The following code will "work" even if the existing string
* is not correctly NUL-terminated; the code will not copy anything
* from src, but it will put a NUL terminator at the end of the
* output buffer.
*/
/* Figure out where the existing string ends. */
size_t prefixlen = strnlen(dst, dstlen - 1);
/* Update dst and dstlen */
dst += prefixlen;
dstlen -= prefixlen;
/* Proceed with the append, as above. */
size_t to_move = strnlen(src, dstlen - 1);
memmove(dst, src, to_move);
dst[to_move] = 0;
}
Embrace snprintf. It really is your friend. But always check its return value.
Using memmove, as above, is slightly awkward. It requires you to manually check that the buffer's length is not zero (otherwise subtracting one would be disastrous since the length is unsigned), and it requires you to manually NUL-terminate the output buffer, which is easy to forget and the source of many bugs. It is very efficient, but sometimes it's worth sacrificing a little efficiency so that your code is easier to write and easier to read and verify.
And that leads us directly to snprintf. For example, you can replace:
if (dstlen) {
size_t to_move = strnlen(src, dstlen - 1);
memcpy(dst, src, to_move);
dst[to_move] = 0;
}
with the much simpler
int copylen = snprintf(dst, dstlen, "%s", src);
That does everything: checks that dstlen is not 0; only copies the characters from src which can fit in dst, and correctly NUL-terminates dst (unless dstlen was 0). And the cost is minimal; it takes very little time to parse the format string "%s" and most implementations are pretty well optimised for this case. [Note 2]
But snprintf is not a panacea. There are still a couple of really important warnings.
First, the documentation for snprintf makes clear that it is not permitted for any input argument to overlap the output range. (So it replaces memcpy but not memmove.) Remember that overlap includes NUL-terminators, so the following code which attempts to double the string in str instead leads to Undefined Behaviour:
char str[BUFLEN];
/* Put something into str */
get_some_data(str, BUFLEN);
/* DO NOT DO THIS: input overlaps output */
int result = snprintf(str, BUFLEN, "%s%s", str, str);
/* DO NOT DO THIS EITHER; IT IS STILL UB */
size_t len = strnlen(str, cap - 1);
int result = snprintf(str + len, cap - len, "%s", str);
The problem with the second invocation of snprintf is that the NUL which terminates str is precisely at str + len, the first byte of the output buffer. That's an overlap, so it's illegal.
The second important note about snprintf is that it returns a value, which must not be ignored. The value returned is not the length of the string created by snprintf. It's the length the string would have been had it not been truncated to fit in the output buffer.
If no truncation occurred, then the result is the length of the result, which must be strictly less than the size of the output buffer (because there must be room for a NUL terminator, which is not considered part of the length of the result.) You can use this fact to check whether truncation occurred:
if (result >= dstlen) /* Output was truncated */
This can be used, for example, to redo the snprintf with a larger, dynamically-allocated buffer (of size result + 1; never forget the need to NUL-terminate).
But remember that the result is an int -- that is, a signed value. That means that snprintf cannot cope with very long strings. That's not likely to be an issue in embedded code, but on systems where it's conceivable that strings exceed 2GB, you may not be able to safely use %s formats in snprintf. It also means that snprintf is allowed to return a negative value to indicate an error. Very old implementations of snprintf returned -1 to indicate truncation, or in response to being called with buffer length 0. That's not standard behaviour according to C99 (nor recent versions of Posix), but you should be prepared for it.
Standard-compliant implementations of snprintf will return a negative value if the buffer length argument is too big to fit in a (signed) int; it's not clear to me what the expected return value is if the buffer length is OK but the untruncated length is too big for an int. A negative value will also be returned if you used a conversion which resulted in an encoding error; for example, a %lc conversion whose corresponding argument contains an integer which cannot be converted to a multibyte (typically UTF-8) sequence.
In short, you should always check the return value of snprintf (recent gcc/glibc versions will produce a warning if you do not), and you should be prepared for it to be negative.
So, with all that behind us, let's write a function which produces a string of co-ordinate pairs:
/* Arguments:
* buf the output buffer.
* buflen the capacity of buf (including room for trailing NUL).
* points a vector of struct Point pairs.
* npoints the number of objects in points.
* Description:
* buf is overwritten with a comma-separated list of points enclosed in
* square brackets. Each point is output as a comma-separated pair of
* decimal floating point numbers enclosed in square brackets. No more
* than buflen - 1 characters are written. Unless buflen is 0, a NUL is
* written following the (possibly-truncated) output.
* Return value:
* If the output buffer contains the full output, the number of characters
* written to the output buffer, not including the NUL terminator.
* If the output was truncated, (size_t)(-1) is returned.
*/
size_t sprint_points(char* buf, size_t buflen,
struct Point const* points, size_t npoints)
{
if (buflen == 0) return (size_t)(-1);
size_t avail = buflen;
char delim = '['
while (npoints) {
int res = snprintf(buf, avail, "%c[%f,%f]",
delim, points->lat, points->lon);
if (res < 0 || res >= avail) return (size_t)(-1);
buf += res; avail -= res;
++points; --npoints;
delim = ',';
}
if (avail <= 1) return (size_t)(-1);
strcpy(buf, "]");
return buflen - (avail - 1);
}
Notes
You will often see code like this:
strncat(dst, src, sizeof(src)); /* NEVER EVER DO THIS! */
Telling strncat not to append more characters from src than can fit in src is obviously pointless (unless src is not correctly NUL-terminated, in which case you have a bigger problem). More importantly, it does absolutely nothing to protect you from writing beyond the end of the output buffer, since you have not done anything to check that dst has room for all those characters. So about all it does is get rid of compiler warnings about the unsafety of strcat. Since this code is exactly as unsafe as strcat was, you probably would be better off with the warning.
You might even find a compiler which understands snprintf will enough to parse the format string at compile time, so the convenience comes at no cost at all. (And if your current compiler doesn't do this, no doubt a future version will.) As with any use of the *printf family, you should never try to economize keystrokes by
leaving out the format string (snprintf(dst, dstlen, src) instead of snprintf(dst, dstlen, "%s", src).) That's unsafe (it has undefined behaviour if src contains an unduplicated %). And it's much slower because the library function has to parse the entire string to be copied looking for percent signs, instead of just copying it to the output.
Code is using functions that expect pointers to string, yet not always passing pointers to strings as arguments.
Stray characters seen at output of snprintf
A string must have a terminating null character.
strncat(char *, .... expects the first parameter to be a pointer to a string. memcpy(output, "[",1); does not insure that. #Jeremy
memcpy(output, "[",1);
...
strncat(output, temp,sizeof(temp));
This is a candidate source of stray characters.
strncat(...., ..., size_t size). itself is a problem as the size is the amount of space available for concatenating (minus the null character). The size available to char * output is not passed in. #Jonathan Leffler. Might as well do strcat() here.
Instead, pass in the size available to output to prevent buffer overflow.
#define N 60
int get_gps60secString(GPS_periodic_t input[N], char *output, size_t sz) {
int cnt = snprintf(output, sz, "[");
if (cnt < 0 || cnt >= sz)
return 1;
output += cnt;
sz -= cnt;
int i = 0;
for (i = 0; i < N; i++) {
cnt = snprintf(output, size, "[%0.8f,%0.8f]%s", input[i].point.latitude,
input[i].point.longitude, i + 1 == N ? "" : ",");
if (cnt < 0 || cnt >= sz)
return 1;
output += cnt;
sz -= cnt;
}
cnt = snprintf(output, sz, "]");
if (cnt < 0 || cnt >= sz)
return 1;
return 0; // no error
}
OP has posted more code - will review.
Apparently the buffer char *output is pre-filled with 0 before the get_gps60secString() so the missing null character from memcpy(output, "[",1); should not cause the issue - hmmmmmm
unsigned short SendTrack() does not return a value. 1) Using its result value is UB. 2) Enable all compiler warnings.
So I'm reading from a file descriptor which contains an int variable in its raw byte format.
So I'm doing:
char buffer[sizeof(int)];
ssize_t sizeOfFile = read(sock_fd, buffer, sizeof(int));
int extractedInt = ???;
How can I convert that buffer to an integer? I was thinking of memcpy but was wondering if there are better ways.
You could read directly an integer
int extractedInt;
ssize_t sizeOfFile = read(sock_fd, &extractedInt, sizeof(int));
read will read the size of an int bytes, and store them into extractedInt.
If your int is actually a string in a file you want to convert to an int, the procedure is a bit different.
#define SIZE 20
char buffer[SIZE]; // ensure there is enough space for a string containing an integer
ssize_t sizeOfFile = read(sock_fd, buffer, SIZE);
int extractedInt = atoi(buffer); // convert string to integer
I can guess from your code that you're reading from the network. This is then not portable to just read a int from the buffer, in your network protocol you chose a certain endianness but you cannot expect that all the platforms where your program will run to have the same, so it will lead to bad convertions.
And other proposed solutions of asking read to return an int will lead to the same problem.
So in your case, I can only advice to iterate through your array and compute the integer by progressively placing the bytes at the right place depending on the endianness of the platform.
You can detect the endianness of the build target platform by using the macro __BYTE_ORDER__in GCC.
There is an example for network data that is big endian:
// construct an `int` with the bytes in the given buffer
// considering the buffer contains the representation
// of an int in big endian
int buffer_to_int(char* buffer, int buffer_size) {
int result = 0;
int i;
char sign = buffer[0] & 0x80;
char * res_bytes = (char*)&result; // this pointer allows to access the int bytes
int offset = sizeof(int) - buffer_size;
if( sign != 0 )
sign = 0xFF;
if( offset < 0 ) {
// not representable with a `int` type
// we chose here to return the closest representable value
if( sign == 0 ) { //positive
return INT_MAX;
} else {
return INT_MIN;
}
}
#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
for(i=0; i<buffer_size; i++) {
res_bytes[i] = buffer[buffer_size-i-1]; // invert the bytes
}
for(i=0; i<offset; i++){
res_bytes[buffer_size+i] = sign;
}
#else
// same endianness, so simply copy bytes using memcpy
memcpy(&result + offset, buffer, buffer_size);
for(i=0; i<offset; i++){
res_bytes[i] = sign;
}
#endif
return result;
}