Re: [PATCH ima-evm-utils v4] evmctl: Use secure heap for private keys and passwords

[Mimi Zohar <zohar@linux.ibm.com>]

On Sat, 2021-09-04 at 13:50 +0300, Vitaly Chikunov wrote:
> After CRYPTO_secure_malloc_init OpenSSL will automatically store private
> keys in secure heap. OPENSSL_secure_malloc(3):
> 
>   If a secure heap is used, then private key BIGNUM values are stored
>   there. This protects long-term storage of private keys, but will not
>   necessarily put all intermediate values and computations there.
> 
> Additionally, we try to keep user passwords in secure heap too.
> This facility is only available since OpenSSL_1_1_0-pre1.
> 
> Signed-off-by: Vitaly Chikunov <vt@altlinux.org>
> Reviewed-by: Bruno Meneguele <bmeneg@redhat.com>
> Cc: Stefan Berger <stefanb@linux.ibm.com>

Thanks, it looks good!  Just a few questions inline below.

> ---
> Change since v3:
> - Undo secure heap handling from (file2bin and) calc_evm_hmac since this
>   is debugging tool only. Add comment about this.
> - Since there is only code removals and new comments I keep Reviewed-by
>   tag.
> 
>  src/evmctl.c | 97 +++++++++++++++++++++++++++++++++++++++++++++-------
>  1 file changed, 85 insertions(+), 12 deletions(-)
> 
> diff --git a/src/evmctl.c b/src/evmctl.c
> index 5f7c2b8..7bd20f8 100644
> --- a/src/evmctl.c
> +++ b/src/evmctl.c
> @@ -59,6 +59,7 @@
>  #include <assert.h>
> 
>  #include <openssl/asn1.h>
> +#include <openssl/crypto.h>
>  #include <openssl/sha.h>
>  #include <openssl/pem.h>
>  #include <openssl/hmac.h>
> @@ -165,6 +166,24 @@ struct tpm_bank_info {
>  static char *pcrfile[MAX_PCRFILE];
>  static unsigned npcrfile;
> 
> +#if OPENSSL_VERSION_NUMBER <= 0x10100000
> +#warning Your OpenSSL version is too old to have OPENSSL_secure_malloc, \
> +	falling back to use plain OPENSSL_malloc.
> +#define OPENSSL_secure_malloc	  OPENSSL_malloc
> +#define OPENSSL_secure_free	  OPENSSL_free
> +/*
> + * Secure heap memory automatically cleared on free, but
> + * OPENSSL_secure_clear_free will be used in case of fallback
> + * to plain OPENSSL_malloc.
> + */
> +#define OPENSSL_secure_clear_free OPENSSL_clear_free
> +#define OPENSSL_clear_free(ptr, num)		\
> +	do {					\
> +		OPENSSL_cleanse(ptr, num);	\
> +		OPENSSL_free(ptr);		\
> +	} while (0)
> +#endif
> +
>  static int bin2file(const char *file, const char *ext, const unsigned char *data, int len)
>  {
>  	FILE *fp;
> @@ -1072,6 +1091,7 @@ static int cmd_setxattr_ima(struct command *cmd)
> 
>  #define MAX_KEY_SIZE 128
> 
> +/* This function is debugging tool and should not be used for real private data. */
>  static int calc_evm_hmac(const char *file, const char *keyfile, unsigned char *hash)
>  {
>          const EVP_MD *md;
> @@ -2596,15 +2616,41 @@ static struct option opts[] = {
> 
>  };
> 
> +/*
> + * Copy password from optarg into secure heap, so it could be
> + * freed in the same way as a result of get_password().
> + */

The reason given for copying the password to secure heap is technically
correct, but could we also include the reason that providing the
password on the command line is unsafe?  Perhaps a modified version of
the comment, below, should be included at the beginning of the function
comment.

> +static char *optarg_password(char *optarg)
> +{
> +	size_t len;
> +	char *keypass;
> +
> +	if (!optarg)
> +		return NULL;
> +	len = strlen(optarg);
> +	keypass = OPENSSL_secure_malloc(len + 1);
> +	if (keypass)
> +		memcpy(keypass, optarg, len + 1);
> +	else
> +		log_err("OPENSSL_secure_malloc(%zu) failed\n", len + 1);
> +	/*
> +	 * This memset does not add real security, just increases
> +	 * the chance of password being obscured in ps output.
> +	 */
> +	memset(optarg, 'X', len);
> +	return keypass;
> +}
> +
> +/* Read from TTY into secure heap. */
>  static char *get_password(void)
>  {
>  	struct termios flags, tmp_flags;
>  	char *password, *pwd;
> -	int passlen = 64;
> +	const int passlen = 64;
> 
> -	password = malloc(passlen);
> +	password = OPENSSL_secure_malloc(passlen);
>  	if (!password) {
> -		perror("malloc");
> +		log_err("OPENSSL_secure_malloc(%u) failed\n", passlen);
>  		return NULL;
>  	}
> 
> @@ -2614,8 +2660,8 @@ static char *get_password(void)
>  	tmp_flags.c_lflag |= ECHONL;
> 
>  	if (tcsetattr(fileno(stdin), TCSANOW, &tmp_flags) != 0) {
> -		perror("tcsetattr");
> -		free(password);
> +		log_err("tcsetattr: %s\n", strerror(errno));
> +		OPENSSL_secure_free(password);
>  		return NULL;
>  	}
> 
> @@ -2624,13 +2670,15 @@ static char *get_password(void)
> 
>  	/* restore terminal */
>  	if (tcsetattr(fileno(stdin), TCSANOW, &flags) != 0) {
> -		perror("tcsetattr");
> -		free(password);
> -		return NULL;
> +		log_err("tcsetattr: %s\n", strerror(errno));
> +		/*
> +		 * Password is already here, so there is no reason
> +		 * to stop working on this petty error.
> +		 */
>  	}
> 
>  	if (pwd == NULL) {
> -		free(password);
> +		OPENSSL_secure_free(password);
>  		return NULL;
>  	}
> 
> @@ -2643,6 +2691,7 @@ int main(int argc, char *argv[])
>  	ENGINE *eng = NULL;
>  	unsigned long keyid;
>  	char *eptr;
> +	char *keypass = NULL; /* @secure heap */
> 
>  #if !(OPENSSL_VERSION_NUMBER < 0x10100000)
>  	OPENSSL_init_crypto(
> @@ -2651,6 +2700,17 @@ int main(int argc, char *argv[])
>  #endif
>  			    OPENSSL_INIT_ENGINE_ALL_BUILTIN, NULL);
>  #endif
> +#if OPENSSL_VERSION_NUMBER > 0x10100000
> +	/*
> +	 * This facility is available since OpenSSL_1_1_0-pre1.
> +	 * 'Heap size' 8192 is chosen to be big enough, so that any single key
> +	 * data could fit. 'Heap minsize' 64 is just to be efficient for small
> +	 * buffers.
> +	 */

Assuming we aren't guaranteed that the heap size is allocated, should
we check (e.g. OPENSSL_secure_actual_size())?

> +	if (!CRYPTO_secure_malloc_init(8192, 64))
> +		log_err("CRYPTO_secure_malloc_init() failed\n");

Either change the test to "!= 1" or also log "and 2 if successful but
the heap could not be protected by memory mapping."

> +#endif
> +
>  	g_argv = argv;
>  	g_argc = argc;
> 
> @@ -2682,10 +2742,18 @@ int main(int argc, char *argv[])
>  			imaevm_params.hash_algo = optarg;
>  			break;
>  		case 'p':
> +			if (keypass)
> +				OPENSSL_secure_clear_free(keypass,
> +							  strlen(keypass));

Is this test needed in case the pasword is provided multiple times?  
If so, why not skip subsequent passwords?

>  			if (optarg)
> -				imaevm_params.keypass = optarg;
> +				keypass = optarg_password(optarg);
>  			else
> -				imaevm_params.keypass = get_password();
> +				keypass = get_password();
> +			if (!keypass) {
> +				log_err("Cannot read password\n");
> +				goto quit;
> +			}
> +			imaevm_params.keypass = keypass;
>  			break;
>  		case 'f':
>  			sigfile = 1;
> @@ -2841,7 +2909,9 @@ int main(int argc, char *argv[])
>  		if (err < 0)
>  			err = 125;
>  	}
> -
> +quit:

Stefan's patch introduced "error:".

> +	if (keypass)
> +		OPENSSL_secure_clear_free(keypass, strlen(keypass));
>  	if (eng) {
>  		ENGINE_finish(eng);
>  		ENGINE_free(eng);

and stores the the engine in imaevm_params.

> @@ -2849,6 +2919,9 @@ int main(int argc, char *argv[])
>  		ENGINE_cleanup();
>  #endif
>  	}
> +#if OPENSSL_VERSION_NUMBER > 0x10100000
> +	CRYPTO_secure_malloc_done();
> +#endif
>  	ERR_free_strings();
>  	EVP_cleanup();
>  	BIO_free(NULL);

What was the conclusion in terms of reading the password stored in the
environment variable?

thanks,

Mimi