clang reports warnings about uacme

[bmorel]

I took some time to reduce the number of warnings emitted by clang --analyzer, I am joining the patch to this comment.

To be honest, it's mostly marking where a patch would actually solve the issue. For several of those, I have no idea how to fix properly the problem.

There are still 2 potential double free and one garbage comparison that I could not find where to patch to avoid them.

Finally, I fixed a false positive using a C11 feature: _Noreturn. I do not know if you support it though.
some.analysis.patch.tar.gz

[ndilieto]

TL;DR waste of my time

Please note that I check the code for memory leaks, uninitialized variable assignments and other unpleasantries using valgrind, which unlike the analyzer you used actually runs the code instead of trying to figure it out by just looking at it.

I've checked the supposed "problems" you reported and they are all false alarms. All are notices, only some are warnings. As you have wasted enough of my scarce time and haven't even bothered to understand what the code does, I will only comment on the first warning for the sake of transparency. Here is the analysis:

crypto.c:2396:25: warning: Null pointer passed as an argument to a 'nonnull' parameter

The analyzer is ignoring context and can't figure out that the strcasecmp call it laments about happens only when when the earlier calls to gnutls_x509_crq_get_dn_by_oid or gnutls_x509_crq_get_subject_alt_name have returned a success code, which means buf is guaranteed not to be null to start with. The falltthrough from GNUTLS_SAN_IPADDRESS, which you'll see I marked as intentional, happens when buf is first freed then replaced by ip, which also is checked for null beforehand.

    do {
        r = cn ?
            gnutls_x509_crq_get_dn_by_oid(crq, GNUTLS_OID_X520_COMMON_NAME,
                    ncn, 0, buf, &size) :
            gnutls_x509_crq_get_subject_alt_name(crq, nsan, buf, &size,
                    NULL, NULL);

        switch (r) {
            case GNUTLS_E_REQUESTED_DATA_NOT_AVAILABLE:
                if (cn) {
                    cn = false;
                    r = GNUTLS_E_SUCCESS;
                }
                break;

            case GNUTLS_E_SHORT_MEMORY_BUFFER:
                buf = calloc(1, size);
                if (!buf) {
                    warn("csr_names: calloc failed");
                    goto out;
                }
                break;

            case GNUTLS_SAN_IPADDRESS:
                ip = calloc(1, INET6_ADDRSTRLEN);
                if (!ip) {
                    warnx("csr_names: calloc failed");
                    goto out;
                }
                if (!inet_ntop(size == 4 ? AF_INET : AF_INET6, buf, ip,
                            INET6_ADDRSTRLEN)) {
                    warnx("csr_names: invalid IP address in Subj Alt Name");
                    free(ip);
                    ip = NULL;
                    continue;
                }
                free(buf);
                buf = ip;
                ip = NULL;
                // intentional fallthrough
            case GNUTLS_E_SUCCESS:
            case GNUTLS_SAN_DNSNAME:
                for (i = 0; i < n; i++) {
                    if (strcasecmp(buf, names[i]) == 0)
                        break;
                }

Edit: and your supposed "patch" just contains the following garbage, no actual changes.

Sur la branche master
Votre branche est à jour avec 'origin/master'.

Modifications qui ne seront pas validées :
  (utilisez "git add <fichier>..." pour mettre à jour ce qui sera validé)
  (utilisez "git checkout -- <fichier>..." pour annuler les modifications dans la copie de travail)

        modifié :         crypto.c
        modifié :         json.c
        modifié :         uacme.c
        modifié :         ualpn.c

aucune modification n'a été ajoutée à la validation (utilisez "git add" ou "git commit -a")