ldb_tdb: replace strange dn_list_cmp() in index code

This replaces dn_list_cmp() with functions that do not attempt to
to care about string termination.  All index values are case sensitive
and correctly length-bound already, even for a DN index
so just use a length check and memcmp()

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

ldb_tdb: Do not allow a modification of the GUID_index_attribute (objectGUID)

This would totally break our index scheme if this could be modified.

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

ldb_tdb: Add GUID_index_attribute to ltdb_private

This will be used to determine if we are in GUID index mode

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

ldb_tdb: Provide struct ltdb_private to index routines

This will make it easier to switch the GUID index mode on and off

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

ldb_tdb: Use a more complete error mapping in ltdb_search_key()

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

ldb_tdb: Add ltdb_search_key()

This allows us to slowly split out the tdb key in the DB from being the DN

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

ldb_tdb: Use ltdb_key_msg() in re_index()

This will allow changing to a GUID tdb key in the future

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

ldb_tdb: provide ldb_key_dn() and ldb_key_msg()

This will in time allow us to generate a TDB key from
the msg, eg from an objectGUID.

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

ldb_tdb: Add helper function ltdb_search_and_return_base()

This avoids an extra DB lookup for the base, when that is the only
record we will return, and avoids going into the index code for
a base search, as that won't work for special DNs once the GUID
index mode is enabled.

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

ldb_tdb: Make ldb_match_message() available to ldb_tdb

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

ctdb-tests: Add tests for client with multiple connections

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Volker Lendecke <vl@samba.org>
Autobuild-User(master): Volker Lendecke <vl@samba.org>
Autobuild-Date(master): Fri Sep 22 20:58:46 CEST 2017 on sn-devel-144

ctdb-tests: Add support for multiple ctdb connections in dummy_client

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Volker Lendecke <vl@samba.org>

ctdb-tests: Check all connections from a process in CHECK_PID_SRVID control

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Volker Lendecke <vl@samba.org>

ctdb-daemon: Check all connections from a process in CHECK_PID_SRVID control

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Volker Lendecke <vl@samba.org>

selftest: Also run smbtorture smb2.compound with aio enabled

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13047

Signed-off-by: Christof Schmitt <cs@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Fri Sep 22 09:49:30 CEST 2017 on sn-devel-144

torture: Add testcase for compound CREATE-WRITE-CLOSE request

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13047

Signed-off-by: Christof Schmitt <cs@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

smbd/aio: Do not go async for SMB2 compound requests

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13047

Signed-off-by: Christof Schmitt <cs@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

smbd: Move check for SMB2 compound request to new function

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13047

Signed-off-by: Christof Schmitt <cs@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

s3: Fix a small spelling mistake in smbcacls.

Signed-off-by: Richard Sharpe <realrichardsharpe@gmail.com>
Reviewed-by: Volker Lendecke <vl@samba.org>
Autobuild-User(master): Volker Lendecke <vl@samba.org>
Autobuild-Date(master): Fri Sep 22 05:44:09 CEST 2017 on sn-devel-144

auth3: Remove auth_domain

If you're a domain member, use winbind. Auth_domain is from times when we did
not have winbind. It has served its purpose, but we should move on.

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Fri Sep 22 00:02:29 CEST 2017 on sn-devel-144

ctdb-tests: Add tests for PID with srvid

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
Autobuild-User(master): Martin Schwenke <martins@samba.org>
Autobuild-Date(master): Thu Sep 21 12:56:30 CEST 2017 on sn-devel-144

ctdb-tests: Implement control CHECK_PID_SRVID in fake daemon

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

ctdb-tests: Reimplement message handlers using srvid abstraction

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

This is required for CHECK_PID_SRVID control implementation.

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

ctdb-tool: Update process-exists command to pass optional srvid

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

ctdb-client: Add client code for control CHECK_PID_SRVID

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

ctdb-daemon: Add implementation of control CHECK_PID_SRVID

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

ctdb-protocol: Add marshalling for control CHECK_PID_SRVID

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

ctdb-protocol: Add marshalling for struct ctdb_pid_srvid

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

ctdb-protocol: Add new control CTDB_CONTROL_CHECK_PID_SRVID

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

ctdb-common: Extend srvid_exists() check to support optional private_data

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13042

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

testsuite: Fix the 32-bit test build

samba_init_module returns 32-bit. For some reason on my
32-bit lxc "return 0" was converted to something but
NT_STATUS_OK, making initialization fail.

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Thu Sep 21 02:49:32 CEST 2017 on sn-devel-144

WHATSNEW: Mention code removal from "net" and "rpcclient"

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

libnet: Remove libnet_samsync

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

net: Don't depend on libnet_samsync anymore

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

net: Remove NT4-based vampire keytab

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

net: Remove NT4-based rpc vampire ldif

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

net: Remove rpc vampire from NT4 domains

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

net: Remove rpc samdump

This uses the NT4 replication commands. Samba does not have a server
for this, no tests, and whoever needs to migrate a native domain can
use an old Samba version

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

rpcclient: Remove sam_sync related commands

These three commands don't use the netlogon credential chain
correctly. They are missing the netlogon_creds_store after the dcerpc
call, so they destroy the correct use of the netlogon creds.

The only valid server for these calls that I know of would be NT4, and
that should be gone long ago.

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

Make sure smbtorture tests can run if someone has set their min protocol above NT1.

This code is SMB1 only, and already modifies
maxprotocol, so this change is appropriate.

Signed-off-by: Richard Sharpe <richard.sharpe@primarydata.com>
Reviewed-by: Jeremy Allison <jra@samba.org>

CVE-2017-12163: s3:smbd: Prevent client short SMB1 write from writing server memory to file.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13020

Signed-off-by: Jeremy Allison <jra@samba.org>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Autobuild-User(master): Karolin Seeger <kseeger@samba.org>
Autobuild-Date(master): Wed Sep 20 17:06:23 CEST 2017 on sn-devel-144

CVE-2017-12150: s3:libsmb: only fallback to anonymous if authentication was not requested

With forced encryption or required signing we should also don't fallback.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12997

Signed-off-by: Stefan Metzmacher <metze@samba.org>

CVE-2017-12150: libcli/smb: add smbXcli_conn_signing_mandatory()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12997

Signed-off-by: Stefan Metzmacher <metze@samba.org>

CVE-2017-12150: auth/credentials: cli_credentials_authentication_requested() should check for NTLM_CCACHE/SIGN/SEAL

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12997

Signed-off-by: Stefan Metzmacher <metze@samba.org>

CVE-2017-12150: libgpo: make use of SMB_SIGNING_REQUIRED in gpo_connect_server()

It's important that we use a signed connection to get the GPOs!

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12997

Signed-off-by: Stefan Metzmacher <metze@samba.org>

CVE-2017-12150: s3:pylibsmb: make use of SMB_SIGNING_DEFAULT for 'samba.samba3.libsmb_samba_internal'

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12997

Signed-off-by: Stefan Metzmacher <metze@samba.org>

CVE-2017-12150: s3:lib: get_cmdline_auth_info_signing_state smb_encrypt SMB_SIGNING_REQUIRED

This is an addition to the fixes for CVE-2015-5296.

It applies to smb2mount -e, smbcacls -e and smbcquotas -e.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12997

Signed-off-by: Stefan Metzmacher <metze@samba.org>

CVE-2017-12150: s3:popt_common: don't turn a guessed username into a specified one

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12997

Signed-off-by: Stefan Metzmacher <metze@samba.org>

CVE-2017-12151: s3:libsmb: make use of cli_state_is_encryption_on()

This will keep enforced encryption across dfs referrals.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12996

Signed-off-by: Stefan Metzmacher <metze@samba.org>

CVE-2017-12151: s3:libsmb: add cli_state_is_encryption_on() helper function

This allows to check if the current cli_state uses encryption
(either via unix extentions or via SMB3).

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12996

Signed-off-by: Stefan Metzmacher <metze@samba.org>

dsdb: Only trigger a re-index once per @INDEXLIST modification

A modify of both @INDEXLIST and @ATTRIBUTES will still trigger two re-index passes
but that is a task for later.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=9527

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
Autobuild-User(master): Andrew Bartlett <abartlet@samba.org>
Autobuild-Date(master): Wed Sep 20 12:29:49 CEST 2017 on sn-devel-144

selftest: sort dbcheck output to avoid sort order impacting results

The GUID index code will change the returned results order

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

s4-dnsserver: Check for too many DNS results

If we had this check in when the wildcard DNS tests were written, we would have
noticed that the name needed to be escaped (see previous commit).

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12994
Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

s4-dnsserver: Always encode user-supplied names when looking up DNS records

BUG: https://bugzilla.samba.org/show_bug.cgi?id=12994

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

kcc: Remove unused, untested KCC code

This code tries to implement the full KCC algorithm, but never
actually worked correctly.

Removing this doesn't affect the full-mesh KCC. This code only
attempted to calculate a graph using the "proper" algorithm, though it
neglected to write its results back into the database. The full-mesh
calculation occurs elsewhere.

Signed-off-by: Garming Sam <garming@catalyst.net.nz>
Reviewed-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Autobuild-User(master): Douglas Bagnall <dbagnall@samba.org>
Autobuild-Date(master): Wed Sep 20 06:28:07 CEST 2017 on sn-devel-144

ldap_server: Plumb ldb error string from a failed connect to ldapsrv_terminate_connection()

However, do not plumb it to the client-seen error string, as it could contain server paths.

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

samdb: Rework samdb_connect_url() to return LDB error code and an error string

This allows debugging of why the LDB failed to start up.

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

ldb: Release 1.2.3

 * Bug #13033 LDB open with LDB_FLG_RDONLY can cause the database
   to fail to open

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13033

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

ldb: Add tests for read only behaviour

As the kernel is no longer enforcing the read-only DB
add some tests.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13033

Signed-off-by: Gary Lockyer <gary@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

ldb_tdb: Change ltdb_connect() NOT to request a kernel-level read only TDB

We support opening and LDB multiple times in a process, but do not support this in tdb.

As we can open the ldb with different flags, we must ensure a later read-write
open is possible.

Additionally, a read-only TDB will refuse the all-record lock, preventing
the ldb from even loading.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13033

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

ldb_tdb: Give a debug message as well as setting the error string if prepare_commit() fails

This is a serious condition, and should be logged.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13033

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

ldb_tdb: Map TDB error codes into LDB error codes in ltdb_lock_read()

The ltdb_lock_read() routine did not return an LDB error code, but -1.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=13033

Signed-off-by: Gary Lockyer <gary@catalyst.net.nz>
Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

ctdb-tools: Fix a typo for a talloc context

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
Autobuild-User(master): Martin Schwenke <martins@samba.org>
Autobuild-Date(master): Tue Sep 19 17:31:18 CEST 2017 on sn-devel-144

ctdb-tools: Add debug to ctdb_killtcp

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-tools: Move special case of 0 connections into computation

This avoids other potential users from unnecessarily setting up file
descriptors and such.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-tools: Rework killtcp logic into a tevent_req-based computation

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-tools: New function ctdb_kill_tcp_init()

This replaces ctdb_killtcp(), which did the initialisation inside a
loop.  The new logic is inverted, making it more natural.

The variable containing all the state is called "state" in
anticipation of the next commit that will convert this to a tevent_req
computation.  This will mean less churn.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-tools: Improve error handling

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-tools: Drop global variable prog

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-tools: Use db_hash in ctdb_killtcp

One less use of trbt_tree_t.  The code is easier to read and is
significantly smaller.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-tools: Use ctdb_connection and ctdb_connection_list structs

Also use new connection and sock addr utilities.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-protocol: Add ctdb_connection_list utilities

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-protocol: Add marshalling for ctdb_connection_list

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-protocol: Add new data structure ctdb_connection_list

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-protocol: Add ctdb_connection utilities

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-protocol: Factor out static function ctdb_sock_addr_cmp_family()

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-protocol: Add ctdb_sock_addr_from_string()

This and the supporting functions duplicate functionality (parse_ip()
and parse_ip_port()) from common/system_util.c.  The old functions
will be removed at a later time.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-protocol: Optionally print port for address printing functions

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-protocol: Add utility function ctdb_sock_addr_to_buf()

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-protocol: Add ctdb_sock_addr_port() and sock_addr_set_port()

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-protocol: Add server and client aliases in ctdb_connection

The current code is ambiguous in its use of src and dst.  This allows
new code to use server and client for clarity.

Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>

ctdb-common: Initialise socket addresses before reading into them

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-build: Split protocol-util as a separate subsystem

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

ctdb-build: Fix dependency for ctdbd

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

s4-torture: move lease break handler outside the lease testsuite.

Guenther

Signed-off-by: Guenther Deschner <gd@samba.org>
Pair-Programmed-With: Jose A. Rivera <jarrpa@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>
Autobuild-User(master): Ralph Böhme <slow@samba.org>
Autobuild-Date(master): Tue Sep 19 09:36:40 CEST 2017 on sn-devel-144

selftest: Avoid a build started just before midnight failing

It looks like commit 070f24bc9d2 was incomplete because it doesn't
match on the trailing fullstop and newline.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Autobuild-User(master): Martin Schwenke <martins@samba.org>
Autobuild-Date(master): Tue Sep 19 05:32:22 CEST 2017 on sn-devel-144

Revert "ctdb-daemon: Don't explicitly stop monitoring during shutdown"

This reverts commit 19318d28354190a9cfafa1696e41c90351f85af9.

With this commit, a shutdown that occurs while the startup event is
running can cause an abort because the startup callback will try to
decrease the run state from SHUTDOWN to RUNNING.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

Revert "ctdb-daemon: Remove unused function ctdb_stop_monitoring()"

This reverts commit b119104267635d96b1f99a2e98d35438841585bf.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

selftest: Add some tests for linked attribute conflicts

Currently we have tests that check we can resolve object conflicts, but
these don't test anything related to conflicting linked attributes.
This patch adds some basic tests that checks that Samba can resolve
conflicting linked attributes.

This highlights some problems with Samba, as the following tests
currently fail:
- test_conflict_single_valued_link: Samba currently can't resolve a
  conflicting targets for a single-valued linked attribute - the
  replication exits with an error.
- test_link_deletion_conflict: If 2 DCs add the same linked attribute,
  currently when they resolve this conflict the RMD_VERSION for the
  linked attribute incorrectly gets incremented. This means the version
  numbers get out of step and subsequent changes to the linked attribute
  can be dropped/ignored.
- test_full_sync_link_conflict: fails for the same reason as above.

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
Autobuild-User(master): Garming Sam <garming@samba.org>
Autobuild-Date(master): Mon Sep 18 09:56:41 CEST 2017 on sn-devel-144

getncchanges.c: max_links calculation didn't work well in some cases

The max_links calculation didn't work particularly well if max_links was
set to a value lower than max_objects.

As soon as repl_chunk->object_count exceeded repl_chunk->max_links, the
chunk would be deemed full, even if there was only one link to send (or
even worse, no links to send). For example, if max_objects=100 and
max_links=10, then it would send back chunks of 10 objects (or 9 objects
and 1 link).

I believe the historic reason this logic exists is to avoid overfilling
the response message. It's hard to tell what the appropriate limit would
be because the total message size would depend on how many attributes
each object has.

I couldn't think of logic that would be suitable for all cases. I toyed
with the idea of working out a percentage of how full the message is.
However, adjusting the max_links doesn't really make sense when the
settings are small enough, e.g. max_objects=100 and max_links=100 is
never going to overfill the message, so there's no reason to alter the
values.

In the end I went with:
- If the user is using non-default values, just use those.
- In the default value case, just use the historic calculation

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12972
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

replmd: Avoid duplicated debug/warnings

We display warnings if a target object is missing but it's still OK to
continue the replication. Currently we need to check the target twice -
once to verify it when we first receive it, and once when we actually
commit it (we can't skip the 2nd check altogether because in the join
case, they could occur quite far apart).

One annoying side-effect is we get the same warning message coming out
twice in these special cases.

In the cases where we're checking the dsdb_repl_flags, we can actually
just bypass the verification checks for the target object (if it doesn't
exist we still continue anyway). This may save us a tiny bit of
unnecessary work.

For cross-partition links, we can limit logging these warnings to when
the objects are actually being committed. This avoids spurious warnings
in the join case (i.e. we receive the link before we receive the target
object's partition, but we have received all partitions by the time we
actually commit the objects).

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12972
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

replmd: Allow missing targets if GET_TGT has already been set

While running the selftests, I noticed a case where DC replication
unexpectedly sends a linked attribute for a deleted object (created in
the drs.ridalloc_exop tests). The problem is due to the
msDS-NC-Replica-Locations attribute, which is a (known) one-way link.
Because it is a one-way link, when the test demotes the DC and deletes
the link target, there is no backlink to delete the link from the source
object.

After much debate and head-scratching, we decided that there wasn't an
ideal way to resolve this problem. Any automated intervention could
potentially do the wrong thing, especially if the link spans partitions.
Running dbcheck will find this problem and is able to fix it (providing
the deleted object is still a tombstone). So the recommendation is to
run dbcheck on your DCs every 6 months (or more frequently if using a
lower tombstone lifetime setting).

However, it does highlight a problem with the current GET_TGT
implementation. If the tombstone object had been expunged and you
upgraded to 4.8, then you would be stuck - replication would fail
because the target object can't be resolved, even with GET_TGT, and
dbcheck would not be able to fix the hanging link. The solution is to
not fail the replication for an unknown target if GET_TGT has already
been set (i.e. the dsdb_repl_flags contains
DSDB_REPL_FLAG_TARGETS_UPTODATE).

It's debatable whether we should add a hanging link in this case or
ignore/drop the link. Some cases to consider:
- If you're talking to a DC that still sends all the links last, you
  could still get object deletion between processing the source object's
  links and sending the target (GET_TGT just restarts the replication
  cycle from scratch). Adding a hanging link in this case would be
  incorrect and would add spurious information to the DB.
- Suppose there's a bug in Samba that incorrectly results in an object
  disappearing. If other DCs then remove any links that pointed to that
  object, it makes recovering from the problem harder. However, simply
  ignoring the link shouldn't result in data loss, i.e. replication won't
  remove the existing link information from other DCs. Data loss in this
  case would only occur if a new DC were brought online, or if it were a
  new link that was affected.
Based on this, I think ignoring the link does the least harm.

This problem also highlights that we should really be using the same
logic in both the unknown target and the deleted target cases.
Combining the logic and moving it into a common
replmd_allow_missing_target() function fixes the problem. (This also has
the side-effect of fixing another logic flaw - in the deleted object
case we would unnecessarily retry with GET_TGT if the target object was
in another partition. This is pointless work, because GET_TGT won't
resolve the target).

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12972
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

getncchanges.c: Support GET_TGT better with large numbers of links

A source object can potentially link to thousands of target objects.
We have to be careful not to overfill the GetNCChanges response message
with more data than it's possible to send. We also don't want the client
to timeout while we're busy checking the linked attributes. The GET_TGT
support added so far is fairly dumb - this patch extends it to better
handle larger numbers of links.

To do so, this extends the repl_chunk usage so that it also works out if
the current chunk is full of links. Now as soon as the chunk is full of
either links or objects, we stop and send it back.

These changes now mean that we need to also check:
- that all the links for the last source object in the previous chunk
  have been sent, before we move on and send the next object. This only
  takes effect when immediate_link_sync is configured. It also means
  that a chunk in the middle of the replication cycle can now contain
  only links, and no objects.
- when GET_TGT is used, we only send back the links that we've verified
  the target object for. i.e. if we stop checking links because we timed
  out, we only send back the links whose targets were checked.

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

getncchanges.c: Refactor to track more state using repl_chunk

To prepare GET_TGT to deal with a large number of links better, there
is now a 'repl_chunk' struct to help keep track of all the factors
relating to the current chunk of replication data (i.e. how many
objects/links we can send and how many we've already processed). This
means we can have a consistent way of working out whether the current
chunk is full (whether that be due to objects, links, or just too much
time taken).

This patch should not alter functionality. This is just a refactor to
add the basic framework, which will be used in the next patch.

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Signed-off-by: Garming Sam <garming@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

getncchanges.py: Add a multi-valued linked attribute test

Add a test where a source object links to multiple different targets.
First we do the replication without GET_TGT and check that the server
can handle sending a chunk containing only links (in the middle of the
replication). Then we repeat the replication forcing GET_TGT to be used.

To avoid having to create 1500 objects/links, I've lowered the 'max
link sync' setting on the vampire_dc testenv to 250.

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

getncchanges.py: Add a test for dropped cross-partition links

Samba would drop linked attributes that span partitions if it didn't
know about the target object. This patch adds a test that exposes the
problem.

I've re-used the code from the previous re-animation test to do this.
I've also added a very basic DcConnection helper class that basically
stores the connection state information the drs_base.py uses for
replication. This allows us to switch the DC we want to replicate from
easily. This approach could potentially be retro-fitted to some of the
existing test cases, as it allows us to test both the DRS client code
and server code at the same time.

Note this test case relates to the code change for commit
fae5df891c11f642cb.

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12972
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

getncchanges.py: Add test for replicating reanimated objects

Reading between the lines, this scenario seems to be the main reason
that Microsoft added the GET_TGT flag. MS AD can handle getting links
for unknown targets OK, but if it receives links for a deleted/recycled
target then it would tend to drop the received links. Samba client also
used to drop the links if talking to a Microsoft DC (or a Samba server
with GET_TGT support).

The specific scenario is the client side already knows about a deleted
object. That object is then re-animated and used as the target for a
linked attribute. *Then* the target object gets updated again so it gets
sent in a later replication chunk to the linked attribute, i.e. the
client receives the link before it learns that the target object has
been re-animated.

In this test we're interested in particular at how the client behaves
when it receives a linked attribute for a deleted object. (It *should*
retry with GET_TGT to make sure the target is up-to-date. However, it
was just dropping the linked attribute).

To exercise the client-side, we disable replication, setup the
links/objects on one DC the way we want them, then force a replication
to the second DC. We then check that when we query each DC, they both
tell us about the links/objects we're expecting (i.e. no links got
lost).

Note that this wasn't a problem with older versions of Samba-to-Samba
because sending the links last guaranteed that the target objects were
always up-to-date.

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

drs: Add basic GET_TGT support

This adds basic DRS_GET_TGT support. If the GET_TGT flag is specified
then the server will use the object cache to store the objects it sends
back. If the target object for a linked attribute is not in the cache
(i.e. it has not been sent already), then it is added to the response
message.

Note that large numbers of linked attributes will not be handled well
yet - the server could potentially try to send more than will fit in a
single repsonse message.

Also note that the client can sometimes set the GET_TGT flag even if the
server is still sending the links last. In this case, we know the client
supports GET_TGT so it's safe to send the links interleaved with the
source objects (the alternative of fetching the target objects but not
sending the links until last doesn't really make any sense).

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

getncchanges.py: Add tests for object deletion during replication

Add tests that delete the source and target objects for linked
attributes in the middle of a replication cycle.

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>

getnc_exop.py: Extend EXOP_REPL_OBJ test case to use GET_TGT

We already check that when we use GET_ANC that we still only receive a
single object when EXOP_REPL_OBJ is used. This extends the test to also
check that only a single object is returned when GET_TGT is used.

Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>