Ian Puleston wrote:
There is something missing from the fix. I just had a problem where an
asynchronous connect was failing and the result was continuous looping with repeated calls to ldap_sasl_bind.
The problem is that if the connect fails, ldap_int_poll called from
ldap_int_check_async_open gets an error and returns -1, and then -1 is returned up the stack ldap_int_check_async_open -> ldap_send_initial_request -> ldap_sasl_bind. But when ldap_sasl_bind gets that -1 return from ldap_send_initial_request it returns ld->ld_errno, and nowhere have we set that. In fact, ld->ld_errno still contains LDAP_X_CONNECTING from when the initial connect was issued, and so ldap_sasl_bind returns LDAP_X_CONNECTING, and that is what leads to the infinite looping that I see.
Thanks for the analysis and fix. Now patched in master. In the future please send actual diffs, as noted http://www.openldap.org/devel/contributing.html
What is missing is that in ldap_int_check_async_open this:
default: return -1;
Should be changed to:
default: ld->ld_errno = LDAP_CONNECT_ERROR; return -1;