A COPY or MOVE onto an existing destination replaces it, which destroys
everything the destination collection held. Only the source subtree was
authorized per descendant, and the destination itself was authorized as an
update, so a rule restricting a subtree held against DELETE but not against
an overwrite of the collection above it. Removing a collection is now
authorized the way DELETE on it would be, on the collection and on every
path beneath it. Writing over a file stays an update.
Recursive COPY, MOVE, DELETE and PROPFIND were checked only against the
requested path, reaching descendants their rules deny. A broad rule shadowed
a narrower one naming a collection. LOCK was allowed on any permission, so a
read-only user could create files and block writers. Rules compared case
where the backing file system does not.
* fix: resolve dot segments before checking path rules (GHSA-chxv-mvjv-f92j)
* fix: match trailing-slash path rules against the bare collection
* fix: match destination rules against the URL path when no prefix is set
* fix: restrict collections named by a rule without granting access to them
* docs: cleanup
- Added logging for invalid username attempts to provide more detailed failure reasons.
- Removed "login attempt" log entries to reduce log volume and focus on final verification results.
- Retained logging for invalid password and successful user authorization for clarity.
- Added getRealRemoteIP function to retrieve the real client IP address when behind a reverse proxy.
- Updated authentication logging to use the extracted IP instead of r.RemoteAddr.
- Ensured compatibility for both proxy and non-proxy setups, falling back to r.RemoteAddr when X-Forwarded-For is not present.
Directory makes it more clear of what it is. In addition, this will make it easier when
allowing for multiple directories in the future, since we can just name it 'directories', which is more clear than 'scopes'.
Henrique Dias