Howard Chu authored 4 years ago and Quanah Gibson-Mount committed 4 years ago

Fix a regression in delta-sync, was returning error on old
CSNs instead of ignoring them

Howard Chu authored 4 years ago and Quanah Gibson-Mount committed 4 years ago

Don't log cookiecsn at top, it was already logged on receipt.
Only log the "srs csn" and "too old" message once for each sid.
Fix log output for empty UUID.

Howard Chu authored 4 years ago and Quanah Gibson-Mount committed 4 years ago

Using cs_pmutex. Reverts the addition of cs_modmutex in ITS#9330,
use cs_pmutex for both delta and plain writes.

Note that plain syncrepl already used cs_pmutex when a cookie CSN
was present in the op. Now it is used for all writes, regardless
of presence of cookie.

Howard Chu authored 4 years ago and Quanah Gibson-Mount committed 4 years ago

Don't depend on accesslog overlay's serialization

Howard Chu authored 4 years ago and Quanah Gibson-Mount committed 4 years ago

The mutex_lock was being skipped in the lockout case,
but still calling mutex_unlock at the end.

This reverts commit 9f8a408c.

It unfortunately relies on logging only available in 2.5 and later.

Howard Chu authored 4 years ago and Quanah Gibson-Mount committed 4 years ago

Don't assume si_cookieState is always newer

Howard Chu authored 4 years ago and Quanah Gibson-Mount committed 4 years ago

In a standard Refresh present phase, the provider sends no cookie
since it is only listing the entries that existed as of the time
in the cookie the consumer sent. In this case the consumer only
needs to check entryCSNs against its last sent cookie.

Howard Chu authored 4 years ago and Quanah Gibson-Mount committed 4 years ago

libldap doesn't use a connected socket for UDP sessions, but 3rd
parties can, passed in with ldap_init_fd().

Howard Chu authored 4 years ago and Quanah Gibson-Mount committed 4 years ago

A bit uglier but more straightforward.