Linux uses struct ucred to pass over socket credentials. Historically
this has always worked in recievemsg, if SO_PASSCRED was set on the socket,
even if the remote side didn't pass any credits. But this change broke that:
http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=
16e5726269611b71c930054ffe9b858c1cea88eb;hp=
a9e9fd7182332d0cf5f3e601df3e71dd431b70d7
However, it doesn't actually fail getting the credentials, it just returns
an "empty" one, as initialized by cred_to_ucred() at:
http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=net/core/sock.c;h=
b29ab61b029cf7f76fda992ecfcb8dcaa06b0483;#l756
So, we detect this and fail the credentials reading.
This actually happened in real life with gdbus acting as a server, as
gdbus expected an ucred but libdbus didn't send one.
ucred = data;
+ if (ucred->uid == (uid_t)-1 &&
+ ucred->gid == (gid_t)-1)
+ {
+ /* This happens if the remote side didn't pass the credentials */
+ goto out;
+ }
+
credentials = g_credentials_new ();
g_credentials_set_native (credentials, G_CREDENTIALS_TYPE_LINUX_UCRED, ucred);
message = g_unix_credentials_message_new_with_credentials (credentials);