s3:smbd: avoid dereferencing client->connections

There're typically better ways to get the same information.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s3:smbd: setup client->global->client_guid even without multichannel support

It's too confusing if client->global->client_guid and
client->connections->smb2.client.guid don't have the same value.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s3:smbd: pass down smbXsrv_client to smbd_smb2_send_{oplock,lease}_break()

Which connection is actually used should not matter to the main logic.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s3:smbd: pass smbXsrv_client to downgrade_lease()

This prepares for multichannel support, where breaks are not bound
to a single connection.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s3:smbd: add smbd_server_disconnect_client[_ex]()

With multichannel things may not happen only on one connection.
We may need to disconnect all connections of a client, when something
bad happens.

The first users of this will be the lease/oplock break code,
if they are not able allocate memory or something similar
we need to bail out.

Having a special smbXsrv_client based function is better than
calling exit_server*() directly.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s3:smbd: remove unused session,tcon parameters from smbd_smb2_send_oplock_break()

They are no longer used. However we'll make use of
op->compat->vuid in the next commits, as the session id should be part
of oplock breaks.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s3:smbd: pass down session_id to smbd_smb2_send_break()

Oplock break should contain a valid session id of the open file handle,
as file handles are relative to a session.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s3:smbd: remove dead code from smbd_smb2_send_break()

Starting with commit 0a924d13cf4bb570cce3955cf0de9d8678b37dbe
("smbd: Send SMB2 oplock breaks unencrypted") we always passed in
session=NULL and tcon=NULL.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:torture/smb2: add smb2.multichannel.oplocks.test3{_windows,specification}

This is similar to the smb2.multichannel.leases.test5,
but it tests the oplock case instead of leases.

With Oplocks Windows only sends a single break on the latest channel,
this is not what the spec says...

Maybe we should have a similar test that would expect the
behavior from the [MS-SMB2] (3/4/2020 rev 60.0)
"3.3.4.6 Object Store Indicates an Oplock Break":

  ...

  If the server implements the SMB 3.x dialect family, SMB2 Oplock Break
  Notification MUST be sent to the client using the first available
  connection in Open.Session.ChannelList where Channel.Connection is not
  NULL. If the server fails to send the notification to the client, the
  server MUST retry the send using an alternate connection, if available,
  in Open.Session.ChannelList.
  ...

Here I add one test that demonstrates the Windows behavior:
 smb2.multichannel.oplocks.test3_windows
and a 2nd test that demonstrates the behavior from MS-SMB2.
 smb2.multichannel.oplocks.test3_specification

Note that Windows 10 seems to behave differently and it's not
possible to open all 32 channel used by this test.

Against remote servers it's required to run iptables as root:

 #> smbtorture //server/torture -Uu%p \
    --option="torture:use_iptables=yes" \
    --option="torture:iptables_command=sudo /sbin/iptables" \
    smb2.multichannel.oplocks.test3_windows

 #> smbtorture //server/torture -Uu%p \
    --option="torture:use_iptables=yes" \
    --option="torture:iptables_command=sudo /sbin/iptables" \
    smb2.multichannel.oplocks.test3_specification

The test will also work against a Samba server
with 'smbd:FSCTL_SMBTORTURE = yes', and won't require iptables
in that case.

Samba will get a "smb2 disable oplock break retry" configuration
option to switch between both behaviors, as it's much more common with Samba
that leases are not supported and clients will fallback to
oplocks together with multichannel.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:torture/smb2: (re-)add smb2.multichannel.leases.test4

This tests 32 channels, which is the maximum Windows Server
versions support. (Note that Windows 10 (a Client OS as SMB server,
seems to support only 20 channels and may differ in other aspects,
so we ignore that for now).

This works at least against Windows Server 2019
and we see lease break notification retries every ~ 1.3 seconds
with ~ 5 TCP retransmissions. At that rate we see the remaining
5 retries after the conflicting SMB2 Create already returned.

Older Windows Server versions use much longer timeouts in the TCP-stack,
they send lease break notification retries less often and only 4 in
total, all other channels get TCP-RST packets because of missing
TCP keepalive packets before they're used.
The intervals between lease break notification retries are
~19 seconds for 2012[_R2] and ~25 seconds for 2016.
It means that only ~2 lease break notifications arrive before
the open returns after ~35 seconds.

Note that Windows 10 seems to behave differently and it's not
possible to open all 32 channel used by this test.

Against remote servers it's required to run iptables as root:

 #> smbtorture //server/torture -Uu%p \
    --option="torture:use_iptables=yes" \
    --option="torture:iptables_command=sudo /sbin/iptables" \
    smb2.multichannel.leases.test4

The test will also work against a Samba server
with 'smbd:FSCTL_SMBTORTURE = yes', and won't require iptables
in that case.

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:torture/smb2: remove useless 'smb2.multichannel.leases.test4'

Having a test that would only pass against Samba makes things way
to complex, they're already complex and we should try to behave
like windows as much as possible.

The next commit will add a better test that will work against Windows
Servers and the future Samba servers.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:torture/smb2: fix smb2.multichannel.leases.test2 against windows

We still receive the break on the blocked channel,
it's only the response ACKs, which we are blocking (or simulate to
block).

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:torture/smb2: split smb2.oplock.batch22 into a and b

batch22a tests the timeout on a valid connection
and batch22b tests the timeout on a broken/blocked connection.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:torture/smb2: move smb2_transport blocking to the generic block.[ch]

We may want to use this in other places too, not only multichannel.c

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:torture/smb2: make use of FSCTL_SMBTORTURE_FORCE_UNACKED_TIMEOUT

This is a way to test without being able to use iptables.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:torture/smb2: refactor block.c to block the OUTPUT path

In order to create useful tests, we should block the outgoing
tcp packets only. That means we're able to see incoming
break notifications, but prevent outgoing TCP ACKs to be delivered
to the server.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:torture/smb2: add break_info.oplock_skip_ack

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:torture/smb2: move interface_info test to smb2.multichannel.generic

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Guenther Deschner <gd@samba.org>
Reviewed-by: Stefan Metzmacher <metze@samba.org>

s4:torture/smb2: make use of transport_options.only_negprot for multichannel connections

This avoid useless session setups and tree connects on the wire.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:torture/smb2: simplify code to generate list of smb2 channels

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Pair-Programmed-With: Stefan Metzmacher <metze@samba.org>

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

s4:torture/smb2: add const to options for test_multichannel_create_channel()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:libcli/smb2: add const to struct smbcli_options *options for smb2_connect()

It will just be passed to smb2_connect_ext(), which already takes a
const pointer.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:libcli/smb2: allow smb2_connect*() to fake session and tcon

For multichannel connection we want a way to have just a connection
with a negprot finished.

For now we just fake a tcon and session in order to avoid changes in the
caller. We can clean that up later if needed.

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:param: use struct initializer in lpcfg_smbcli_session_options()

We should zero all fields not initialiazed explicitly.

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s4:param: use struct initializer in lpcfg_smbcli_options()

We should zero all fields not initialiazed explicitly.

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

libcli/smb: define FSCTL_SMBTORTURE_FORCE_UNACKED_TIMEOUT

This will be used by smbtorture in order to simulate channel failures
without relying on iptables.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

s3:includes: change OPLOCK_BREAK_TIMEOUT from 30 to 35 seconds

This is what windows is using for normal oplock and lease breaks.

Note that windows uses higher values for persistent handles,
they use 60 seconds for oplocks and 180 seconds for leases.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>

lib/util: allow to set TCP_USER_TIMEOUT socket option

BUG: https://bugzilla.samba.org/show_bug.cgi?id=11897

Signed-off-by: Guenther Deschner <gd@samba.org>
Reviewed-by: Stefan Metzmacher <metze@samba.org>

s3:dbwrap_watch: avoid recursion into dbwrap_do_locked() from dbwrap_watched_do_locked_{storev,delete}()

This avoids a lot of overhead!

Using smbtorture3 //foo/bar -U% local-g-lock-ping-pong -o 500000
under valgrind --tool=callgrind...

This change replaces this:

 6,877,542,529  PROGRAM TOTALS

   590,000,773  lib/tdb/common/lock.c:tdb_lock_list
   479,000,608  lib/tdb/common/lock.c:tdb_unlock
   446,500,532  lib/tdb/common/io.c:tdb_read
   364,000,824  lib/tdb/common/hash.c:tdb_jenkins_hash
   285,000,532  lib/tdb/common/io.c:tdb_write
   262,054,669  /x86_64/multiarch/memmove-vec-unaligned-erms.S:__memcpy_avx_unaligned_erms
   206,500,496  lib/tdb/common/mutex.c:tdb_mutex_lock
   193,000,176  lib/tdb/common/tdb.c:tdb_find
   160,000,256  lib/talloc/talloc.c:_talloc_get_type_abort
   148,500,297  lib/tdb/common/tdb.c:tdb_storev
   140,000,196  lib/tdb/common/lock.c:tdb_lock
   130,000,858  lib/util/debug.c:debuglevel_get_class
   128,003,722  lib/talloc/talloc.c:_talloc_free
   128,000,118  lib/tdb/common/tdb.c:tdb_parse_record
   126,000,576  lib/tdb/common/lock.c:tdb_brlock.part.3
   121,000,272  lib/tdb/common/mutex.c:tdb_mutex_unlock
   118,000,225  /nptl/pthread_mutex_lock.c:__pthread_mutex_lock_full
   112,750,222  lib/tdb/common/freelist.c:tdb_allocate_from_freelist
   108,500,168  lib/tdb/common/io.c:tdb_ofs_read
   102,500,000  lib/tdb/common/io.c:tdb_parse_data

by this:

 5,706,522,398  PROGRAM TOTALS

   434,000,617  lib/tdb/common/lock.c:tdb_lock_list
   389,500,494  lib/tdb/common/io.c:tdb_read
   359,000,488  lib/tdb/common/lock.c:tdb_unlock
   285,000,532  lib/tdb/common/io.c:tdb_write
   237,554,655  /x86_64/multiarch/memmove-vec-unaligned-erms.S:__memcpy_avx_unaligned_erms
   208,000,668  lib/tdb/common/hash.c:tdb_jenkins_hash
   206,500,496  lib/tdb/common/mutex.c:tdb_mutex_lock
   160,000,256  lib/talloc/talloc.c:_talloc_get_type_abort
   148,500,297  lib/tdb/common/tdb.c:tdb_storev
   136,000,132  lib/tdb/common/tdb.c:tdb_find
   130,000,858  lib/util/debug.c:debuglevel_get_class
   126,000,576  lib/tdb/common/lock.c:tdb_brlock.part.3
   121,000,272  lib/tdb/common/mutex.c:tdb_mutex_unlock
   118,000,225  /nptl/pthread_mutex_lock.c:__pthread_mutex_lock_full
   112,750,222  lib/tdb/common/freelist.c:tdb_allocate_from_freelist
   112,000,168  lib/tdb/common/lock.c:tdb_lock
    94,500,154  lib/tdb/common/io.c:tdb_ofs_read
    94,000,188  /nptl/pthread_mutex_unlock.c:__pthread_mutex_unlock_full
    86,000,086  lib/dbwrap/dbwrap.c:dbwrap_lock_order_lock
    83,000,083  lib/dbwrap/dbwrap_tdb.c:db_tdb_do_locked

time smbtorture3 //foo/bar -U% local-g-lock-ping-pong -o 5000000

gives:

  902834 locks/sec

 real    0m11,103s
 user    0m8,233s
 sys     0m2,868s

vs.

 1037262 locks/sec

 real    0m9,685s
 user    0m6,788s
 sys     0m2,896s

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Volker Lendecke <vl@samba.org>
Autobuild-User(master): Volker Lendecke <vl@samba.org>
Autobuild-Date(master): Wed Jul  8 11:02:39 UTC 2020 on sn-devel-184

s3:locking: convert share_mode_lock.c to generate_unique_u64()

Instead of a sequence number that gets incremented we just
need a value that's not reused.

The is a similar change like the commit before at the g_lock.c
layer.

I expect a similar performance improvement here, but
I don't know a specific benchmark test to check.

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Volker Lendecke <vl@samba.org>

s3:g_lock: avoid very expensive generate_random_buffer() in g_lock_parse()

We don't require a sequence number that is incremented,
we just need a value that's not reused.
We use the new generate_unique_u64(), which is much cheaper!

Using smbtorture3 //foo/bar -U% local-g-lock-ping-pong -o 500000
under valgrind --tool=callgrind...

This change replaces this:

 13,129,925,659  PROGRAM TOTALS

  4,125,752,958  ???:_nettle_sha256_compress [/usr/lib/x86_64-linux-gnu/libnettle.so.6.4]
  1,257,005,866  ???:_nettle_aes_encrypt [/usr/lib/x86_64-linux-gnu/libnettle.so.6.4]
    590,000,773  bin/default/../../lib/tdb/common/lock.c:tdb_lock_list
    571,503,429  ???:_nettle_aes_set_key [/usr/lib/x86_64-linux-gnu/libnettle.so.6.4]
    479,000,608  bin/default/../../lib/tdb/common/lock.c:tdb_unlock
    ...

by this:

  6,877,826,377  PROGRAM TOTALS

    590,000,773  bin/default/../../lib/tdb/common/lock.c:tdb_lock_list
    479,000,608  bin/default/../../lib/tdb/common/lock.c:tdb_unlock
    ...
     12,500,033  bin/default/../../lib/util/genrand_util.c:generate_unique_u64
    ...
     8,996,970  ???:_nettle_sha256_compress [/usr/lib/x86_64-linux-gnu/libnettle.so.6.4]

time smbtorture3 //foo/bar -U% local-g-lock-ping-pong -o 5000000

gives:

   537426 locks/sec

  real    0m19,071s
  user    0m15,061s
  sys     0m3,999s

vs.

   900956 locks/sec

  real    0m11,155s
  user    0m8,293s
  sys     0m2,860s

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Volker Lendecke <vl@samba.org>

lib/util: add generate_unique_u64() helper function

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Volker Lendecke <vl@samba.org>

s3:smbcacls: Add support for DFS path

smbcacls does not handle DFS paths correctly. This is beacuse once the
command encounters a path which returns STATUS_PATH_NOT_COVERED, it does
not attempt a GET REFERRAL.

We use cli_resolve_path API to perform a DFS path resolution to solve
the above problem.

Additionally this removes the known fail against smbcacls tests
Signed-off-by: Anubhav Rakshit <anubhav.rakshit@gmail.com>
Reviewed-by: Noel Power <noel.power@suse.com>
Reviewed-by: Jeremy Allison <jra@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Tue Jul  7 23:03:00 UTC 2020 on sn-devel-184

selftest: run smbcacls test against a share with a DFS link

The commit creates a dfs link in existing 'fileserver' env
share msdfs_share. Additionally we create a new dfs target in
a new share (with associated directory)

Additionally add a known fail as smbcacls doesn't not yet navigate DFS links.
A subsequent commit will fix smcacls to handle DFS (and remove the
knownfail)

Signed-off-by: Noel Power <noel.power@suse.com>
Reviewed-by: Jeremy Allison <jra@samba.org>

selftest: Add basic smbcacls test(s)

Signed-off-by: Noel Power <noel.power@suse.com>
Reviewed-by: Jeremy Allison <jra@samba.org>

s4:rpc_server: Implement epmapper LookupHandleFree

Signed-off-by: Samuel Cabrero <scabrero@samba.org>
Reviewed-by: David Mulder <dmulder@suse.com>
Reviewed-by: Andreas Schneider <asn@samba.org>
Autobuild-User(master): Samuel Cabrero <scabrero@samba.org>
Autobuild-Date(master): Tue Jul  7 13:43:14 UTC 2020 on sn-devel-184

s4:torture: Do not skip epmapper lookup terminate test in samba4

Signed-off-by: Samuel Cabrero <scabrero@samba.org>
Reviewed-by: David Mulder <dmulder@suse.com>
Reviewed-by: Andreas Schneider <asn@samba.org>

selftest: Split samba4 epmapper expected failures into individual tests

Signed-off-by: Samuel Cabrero <scabrero@samba.org>
Reviewed-by: David Mulder <dmulder@suse.com>
Reviewed-by: Andreas Schneider <asn@samba.org>

s4:selftest: test for smbtorture subunit names with and without --fullname

We check the output with both --fullname and with the default shortname
to ensure it works as expected.

We also do tests for each level and test relative names are used.

Pair-Programmed-With: Stefan Metzmacher <metze@samba.org>

Signed-off-by: Sachin Prabhu <sprabhu@redhat.com>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>
Autobuild-User((no branch)): Stefan Metzmacher <metze@samba.org>
Autobuild-Date((no branch)): Tue Jul  7 12:16:34 UTC 2020 on sn-devel-184

s4:torture/local: Add new test to check smbtorture --fullname

Add new always pass test to test smbtorture --fullname.

Since we test the printing of the fullname of the test, the test is
placed at the bottom of several levels of test suites.

test : local.smbtorture.level1.level2.level3.always_pass

Signed-off-by: Sachin Prabhu <sprabhu@redhat.com>
Reviewed-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>

s4:torture: Add command line parameter --fullname

This argument is used to print out the the full name which includes the
name of the test suite.

For example, the name in the output for the test smb2.read.dir is "dir".
By using the --fullname parameter, the name used will be
"smb2.read.dir".

The default continues to be to use the shortname.

Pair-Programmed-With: Stefan Metzmacher <metze@samba.org>

Signed-off-by: Sachin Prabhu <sprabhu@redhat.com>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>

s4:torture: avoid multiple recursions into the test lists

torture_run_suite_restricted() and torture_run_tcase_restricted()
already handle recursion. If we call them from smbtorture
we should avoid our own recursion.

Without this passing 'smb2.multichannel.*' results in running
tests more than once.

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>

s4:torture: use prefix_name() for all names in run_matching()

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>

lib/torture: fix subunit names of nested suites

E.g. passing 'smb2.multichannel.generic' to smbtorture results in

- interface_info
- num_channels

While passing 'smb2.multichannel' to smbtorture results in:

- generic.interface_info
- genetic.num_channels
- oplocks.test1
...
- leases.test1
...

Before we got this:

- interface_info
- num_channels
- test1
...
- test1

That made it impossible to add knownfail entries for

leases.test1 vs. oplocks.test1

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>

lib/torture: add back links to the parent objects

This makes would allow the full name of a test to be constructed
if needed in future.

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>

lib/torture: make torture_subunit_test_name() public

This makes it easier to modify.

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>

selftest/Samba4: split out _setup_ad_dc()

setup_ad_dc() was used for more than 'ad_dc'.
Setting up 'ad_dc' means not all missing arguments are
passed as undef.

The following had no effect

        if (!defined($conf_opts)) {
                $conf_opts = "";
        }

and $conf_opts results in a string of 'HASH(....)'

So we better pass explicit undef arguments from setup_ad_dc() to
_setup_ad_dc() now.

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>

s3:wscript: vfs_gpfs needs kernel oplock support

It uses symbols, which are only available if we have
HAVE_KERNEL_OPLOCKS_LINUX defined.

This is not the case when building withing the
Windows Subsystem for Liux (WSL). So we better don't try to
build the vfs_gpfs module there.

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Volker Lendecke <vl@samba.org>
Autobuild-User(master): Stefan Metzmacher <metze@samba.org>
Autobuild-Date(master): Tue Jul  7 09:37:37 UTC 2020 on sn-devel-184

s3:smbd: check for stale pid in delay_for_oplock_fn() when leases_db_get() fails

If leases_db_get() failed the leases_db record might have been cleaned up for
stale processes. Check if the share-mode-entry owner is stale in this case and
return ignore the entry. In any other case, log a debug messages and panic.

Commit 05d4466a6d1ad048fa86aea09ec0a56a7b961369
"smbd: check for stale pid in get_lease_type()" fixed only one half of
this.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14428

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Tue Jul  7 02:47:46 UTC 2020 on sn-devel-184

s3:leases: log errors with level 0 in leases_db_do_locked_fn()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14428

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>

dsdb/password_hash: don't double assign variables (CID 1363048, 1034720)

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andreas Schneider <asn@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>
Autobuild-User(master): Andreas Schneider <asn@cryptomilk.org>
Autobuild-Date(master): Mon Jul  6 13:56:17 UTC 2020 on sn-devel-184

pypidl: check the result of py_dcerpc_ndr_pointer_deref()

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andreas Schneider <asn@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>

policy/gp: always close a file (CID 1445119)

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andreas Schneider <asn@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>

s4/winbind/idmap: check the right variable (CID 1272950)

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andreas Schneider <asn@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>

smbd: check for stale pid in get_lease_type()

If leases_db_get() failed the leases_db record might have been cleaned up for
stale processes. Check if the share-mode-entry owner is stale in this case and
return a 0 lease state. In any other case, log a debug messages and panic.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14428

Signed-off-by: Ralph Boehme <slow@samba.org>
Autobuild-User(master): Stefan Metzmacher <metze@samba.org>
Autobuild-Date(master): Thu Jul  2 16:45:42 UTC 2020 on sn-devel-184

smbd: let get_lease_type() take a non-const share_mode_entry

We're going to add a call to share_entry_stale_pid(share_mode_entry) which takes
a non-const pointer (in order to eventually set e->state = true).

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14428

Signed-off-by: Ralph Boehme <slow@samba.org>
Reviewed-by: Stefan Metzmacher <metze@samba.org>

smbd: inverse if/else logic in get_lease_type()

No change in behaviour.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14428

Signed-off-by: Ralph Boehme <slow@samba.org>
Reviewed-by: Stefan Metzmacher <metze@samba.org>

s3/leases: log NDR decoding failure with level 0 in leases_db_get_fn()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14428

Signed-off-by: Ralph Boehme <slow@samba.org>

smbd: increase loglevel when leases_db_del() with anything then NT_STATUS_NOT_FOUND

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14428

Signed-off-by: Ralph Boehme <slow@samba.org>
Reviewed-by: Stefan Metzmacher <metze@samba.org>

s3:smbd: make sure vfs_ChDir() always sets conn->cwd_fsp->fh->fd = AT_FDCWD

This is what all consumers of conn->cwd_fsp->fh->fd expect!

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14427

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>

s3:smbd: reformat if statement for caching in vfs_ChDir()

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14427

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>

s4:torture/smb2: add smb2.delete-on-close-perms.BUG14427

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14427

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>

s3:winbind:idmap_ad - make failure to get attrnames for schema mode fatal

Add check for failure to resolve the OID array for the schema mode into
names.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14425

Signed-off-by: Andrew <awalker@ixsystems.com>
Reviewed-by: Stefan Metzmacher <metze@samba.org>

lib/debug: set the correct default backend loglevel to MAX_DEBUG_LEVEL

This fixed a regression introduced by commit
c83ce5f4f99aef94530411ec82cc03e9935b352d for bug #14345.

The backend loglevel globally restricts logging of a particular backend. If this
value is smaller then any explicitly configured logging class, logging for this
class is skipped.

Eg, given the following logging config in smb.conf:

  log level = 1 auth_json_audit:3@/var/log/samba/samba_auth_audit.log

the default class loglevel of 1 (dbgc_config[DBGC_ALL].loglevel) will be
assigned to the backend loglevel.

So even though the logging class auth_json_audit is configured at level 3, this
doesn't become effective as the file backend drops all log messages with a level
below 1.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14426

Signed-off-by: Ralph Boehme <slow@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>
Autobuild-User(master): Andreas Schneider <asn@cryptomilk.org>
Autobuild-Date(master): Thu Jul  2 13:25:29 UTC 2020 on sn-devel-184

lib/debug: assert file backend

The debug file backend is a built-in default, if it's missing we're totally
screwed.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14426

Signed-off-by: Ralph Boehme <slow@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>

ldap_client: Make ldap_parse_basic_url() IPv6-address aware

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Alexander Bokovoy <ab@samba.org>
Autobuild-User(master): Volker Lendecke <vl@samba.org>
Autobuild-Date(master): Thu Jul  2 12:01:06 UTC 2020 on sn-devel-184

ldap_client: Align integer types

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Alexander Bokovoy <ab@samba.org>

ldap_client: Make ldap_parse_basic_url take care of ldapi as well

SUSV4's sscanf has the %m modifier, which allocates the right
amount. Remove those SMB_ASSERTS for string buffers.

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Alexander Bokovoy <ab@samba.org>

pdb_dsdb: Fix typos

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Alexander Bokovoy <ab@samba.org>

ldb_ldap: Fix a memleak

Don't allocate a temporary value on a long-term context

Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: David Mulder <dmulder@samba.org>

CVE-2020-14303: s4 nbt: fix busy loop on empty UDP packet

An empty UDP packet put the nbt server into a busy loop that consumes
100% of a cpu.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14417

Signed-off-by: Gary Lockyer <gary@catalyst.net.nz>
Autobuild-User(master): Karolin Seeger <kseeger@samba.org>
Autobuild-Date(master): Thu Jul  2 10:26:24 UTC 2020 on sn-devel-184

CVE-2020-14303 Ensure an empty packet will not DoS the NBT server

Signed-off-by: Andrew Bartlett <abartlet@samba.org>

CVE-2020-10760 dsdb: Add tests for paged_results and VLV over the Global Catalog port

This should avoid a regression.

Signed-off-by: Andrew Bartlett <abartlet@samba.org>

CVE-2020-10760 dsdb: Ensure a proper talloc tree for saved controls

Otherwise a paged search on the GC port will fail as the ->data was
not kept around for the second page of searches.

An example command to produce this is
 bin/ldbsearch --paged -H ldap://$SERVER:3268 -U$USERNAME%$PASSWORD

This shows up later in the partition module as:

ERROR: AddressSanitizer: heap-use-after-free on address 0x60b00151ef20 at pc 0x7fec3f801aac bp 0x7ffe8472c270 sp 0x7ffe8472c260
READ of size 4 at 0x60b00151ef20 thread T0 (ldap(0))
    #0 0x7fec3f801aab in talloc_chunk_from_ptr ../../lib/talloc/talloc.c:526
    #1 0x7fec3f801aab in __talloc_get_name ../../lib/talloc/talloc.c:1559
    #2 0x7fec3f801aab in talloc_check_name ../../lib/talloc/talloc.c:1582
    #3 0x7fec1b86b2e1 in partition_search ../../source4/dsdb/samdb/ldb_modules/partition.c:780

or

smb_panic_default: PANIC (pid 13287): Bad talloc magic value - unknown value
(from source4/dsdb/samdb/ldb_modules/partition.c:780)

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14402

Signed-off-by: Andrew Bartlett <abartlet@samba.org>

CVE-2020-10745: ndr/dns-utils: prepare for NBT compatibility

NBT has a funny thing where it sometimes needs to send a trailing dot as
part of the last component, because the string representation is a user
name. In DNS, "example.com", and "example.com." are the same, both
having three components ("example", "com", ""); in NBT, we want to treat
them differently, with the second form having the three components
("example", "com.", "").

This retains the logic of e6e2ec0001fe3c010445e26cc0efddbc1f73416b.

Also DNS compression cannot be turned off for NBT.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14378

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

CVE-2020-10745: dns_util/push: forbid names longer than 255 bytes

As per RFC 1035.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14378

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

CVE-2020-10745: ndr_dns: do not allow consecutive dots

The empty subdomain component is reserved for the root domain, which we
should only (and always) see at the end of the list. That is, we expect
"example.com.", but never "example..com".

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14378

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

CVE-2020-10745: ndr/dns_utils: correct a comment

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14378

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

CVE-2020-10745: ndr_dns: move ndr_push_dns_string core into sharable function

This is because ndr_nbt.c does almost exactly the same thing with
almost exactly the same code, and they both do it wrong. Soon they
will both be using the better version that this will become. Though in
this patch we just move the code, not fix it.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14378

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

CVE-2020-10745: librpc/tests: cmocka tests of dns and ndr strings

These time the push and pull function in isolation.

Timing should be under 0.0001 seconds on even quite old hardware; we
assert it must be under 0.2 seconds.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14378

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

CVE-2020-10745: pytests: hand-rolled invalid dns/nbt packet tests

The client libraries don't allow us to make packets that are broken in
certain ways, so we need to construct them as byte strings.

These tests all fail at present, proving the server is rendered
unresponsive, which is the crux of CVE-2020-10745.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14378

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>

CVE-2020-10730: lib ldb: Check if ldb_lock_backend_callback called twice

Prevent use after free issues if ldb_lock_backend_callback is called
twice, usually due to ldb_module_done being called twice. This can happen if a
module ignores the return value from function a function that calls
ldb_module_done as part of it's error handling.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14364

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

CVE-2020-10730: s4 dsdb vlv_pagination: Prevent repeat call of ldb_module_done

Check the return code from vlv_results, if it is not LDB_SUCCESS
ldb_module_done has already been called, and SHOULD NOT be called again.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14364

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

CVE-2020-10730: s4 dsdb paged_results: Prevent repeat call of ldb_module_done

Check the return code from paged_results, if it is not LDB_SUCCESS
ldb_module_done has already been called, and SHOULD NOT be called again.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14364

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

CVE-2020-10730: dsdb: Ban the combination of paged_results and VLV

This (two different paging controls) makes no sense and fails against
Windows Server 1709.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14364

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

CVE-2020-10730: dsdb: Fix crash when vlv and paged_results are combined

The GUID is not returned in the DN for some reason in this (to be banned)
combination.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14364

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

CVE-2020-10730: selftest: Add test to show that VLV and paged_results are incompatible

As tested against Windows Server 1709

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14364

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

CVE-2020-10730: vlv: Another workaround for mixing ASQ and VLV

This is essentially an alternative patch, but without the correct
behaviour.  Instead this just avoids a segfault.

Included in case we have something simialr again in
another module.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14364

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

CVE-2020-10730: selftest: Add test to confirm VLV interaction with ASQ

Tested against Windows 1709.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14364

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

CVE-2020-10730: vlv: Do not re-ASQ search the results of an ASQ search with VLV

This is a silly combination, but at least try and keep the results sensible
and avoid a double-dereference.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14364

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

CVE-2020-10730: vlv: Use strcmp(), not strncmp() checking the NULL terminated control OIDs

The end result is the same, as sizeof() includes the trailing NUL, but this
avoids having to think about that.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14364

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

s3: libsmb: Fix SMB2 client rename bug to a Windows server.

Fix bug where renaming to a target name of one
UCS2 character (name length 2 bytes) fails to
a Windows 10 SMB2 server.

The Windows 10 SMB2 server has a minimum length
for a SMB2_FILE_RENAME_INFORMATION buffer of
24 bytes. It returns NT_STATUS_INFO_LENGTH_MISMATCH
if the length is less. This isn't an alignment
issue as Windows client happily 2-byte align
for larget target name sizes. Also the Windows 10
SMB1 server doesn't have this restriction.

If the name length is too short, pad out with
zeros to 24 bytes.

Hard to add a test for this as we don't want to
add this silly restriction to the Samba server
as it would break all non-Windows clients.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14403

Signed-off-by: Jeremy Allison <jra@samba.org>
Reviewed-by: Volker Lendecke <vl@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Wed Jul  1 18:59:53 UTC 2020 on sn-devel-184

tls: Use NORMAL:-VERS-SSL3.0 as the default configuration

This seems to be really broken in GnuTLS and the documentation is also
not correct.

This partially reverts 53e3a959b958a3b099df6ecc5f6e294e96bd948e

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14408

Signed-off-by: Andreas Schneider <asn@samba.org>
Reviewed-by: Alexander Bokovoy <ab@samba.org>
Autobuild-User(master): Andrew Bartlett <abartlet@samba.org>
Autobuild-Date(master): Wed Jul  1 14:56:33 UTC 2020 on sn-devel-184

selftest: Run test of how userPassword / crypt() style passwords are stored in quicktest

This ensures that the crypt_r()/crypt_rn()/crypt() behaviour is tested in all
the samba-o3 builds and so is checked on RHEL7 in GitLab CI.

https://bugzilla.samba.org/show_bug.cgi?id=14424

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Alexander Bokovoy <ab@samba.org>

selftest: Split samba.tests.samba_tool.user_virtualCryptSHA into GPG and not GPG parts

This allows the userPassword (not GPG) part of the test to run on hosts without
python3-gpg (eg RHEL7) while still testing the userPassword handling.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14424

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Alexander Bokovoy <ab@samba.org>

dsdb: Allow "password hash userPassword schemes = CryptSHA256" to work on RHEL7

On RHEL7 crypt_r() will set errno.  This is a problem because the implementation of crypt_r()
in RHEL8 and elsewhere in libcrypt will return non-NULL but set errno on failure.

The workaround is to use crypt_rn(), provided only by libcrypt, which will return NULL
on failure, and so avoid checking errno in the non-failure case.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14424

Signed-off-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Alexander Bokovoy <ab@samba.org>

share_mode_lock.c: initialize out param

detected by covscan:
source3/locking/share_mode_lock.c:1563:6: warning: Branch condition evaluates to a garbage value

Signed-off-by: Isaac Boukris <iboukris@samba.org>
Reviewed-by: David Mulder <dmulder@suse.com>
Autobuild-User(master): Isaac Boukris <iboukris@samba.org>
Autobuild-Date(master): Tue Jun 30 09:42:33 UTC 2020 on sn-devel-184

libgpo: only install group policy admx files when building as DC

Guenther

Signed-off-by: Guenther Deschner <gd@samba.org>
Reviewed-by: David Mulder <dmulder@samba.org>
Reviewed-by: Anoop C S <anoopcs@samba.org>
Autobuild-User(master): Anoop C S <anoopcs@samba.org>
Autobuild-Date(master): Tue Jun 30 07:51:48 UTC 2020 on sn-devel-184

util: fix build on AIX by fixing the order of replace.h include

BUG: https://bugzilla.samba.org/show_bug.cgi?id=14422

Signed-off-by: Bjoern Jacke <bjacke@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>
Autobuild-User(master): Björn Jacke <bjacke@samba.org>
Autobuild-Date(master): Mon Jun 29 18:19:04 UTC 2020 on sn-devel-184

s3:smbd: fix the handling of the durable_v2_timeout

Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Sat Jun 27 05:42:05 UTC 2020 on sn-devel-184