IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
I assumed that TALLOC_FREE(xconn->transport.fde) would close the socket,
but until now we didn't use tevent_fd_set_auto_close().
BUG: https://bugzilla.samba.org/show_bug.cgi?id=11898
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>
Pointer values can be reused (yes, I hit that during my testing!).
Introduce a channel_id to identify connections and also add
some timestamps to make debugging easier.
This makes smbXsrv_session_find_auth() much more robust.
This is a similar change as 0cec96526b:
"smb2_server: make sure we detect stale smbXsrv_connection pointers in smbXsrv_channel_global"
BUG: https://bugzilla.samba.org/show_bug.cgi?id=11898
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>
smbd_add_connection() may return a valid connection together with
NT_STATUS_NETWORK_ACCESS_DENIED.
We need additional cleanup for that case.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=11898
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>
We should not just overwrite the client->connections pointer if we
reject the connection.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=11898
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>
These will be used in the multi channel code in order to handle
public ip addresses, which can move arround ctdb nodes.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=11898
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>
This makes sure the lease/oplock break retry logic based on
missing TCP acks is tested.
We're still not able to run multichannel tests automatically,
as socket wrapper doesn't support fd-passing yet.
But this testing this with single channels is a good start.
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>
This will be used by smbtorture in order to simulate channel failures
without relying on iptables.
'smbd:FSCTL_SMBTORTURE = yes' is required in order to active this.
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>
This finally implements the retry of failed oplock/lease breaks.
Before smbd_smb2_break_send/recv completed directly after
sendmsg() passed the pdu to the kernel.
Now the completion is (at least) deferred until the
the next smbXsrv_connection_ack_checker() run happens
and smbd_smb2_send_queue_ack_bytes() found that
all bytes of the break notification left the kernel
send queue (and were TCP acked).
If the connection is disconnected all pending break
notifications are completed with an error, which is
then returned by smbd_smb2_break_recv().
smbXsrv_pending_break_submit() will then submit
another break notification via the next available
connection/channel.
The smbXsrv_connection_ack_checker() runs each
rto_usecs (between 0.2s and 1.0s). smbd_smb2_break_send()
will set a timeout of 6*rto_usecs (between 1.2s and 6s).
If smbXsrv_connection_ack_checker() detects via
smbd_smb2_send_queue_ack_bytes() that a pending break
notification is pending for more than its timeout
we'll disconnect the connection with NT_STATUS_IO_TIMEOUT.
This will be handled as any other disconnect and
will in turn also trigger the retry on the next 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>
For now it's safer to disable multi-channel without having support
for TIOCOUTQ/FIONWRITE on tcp sockets.
Using a fixed retransmission timeout (rto) of 1 second would be ok,
but we better require kernel support for requesting for unacked bytes
in the kernel send queue.
"force:server multi channel support = yes" can be used to overwrite
the compile time restriction (mainly for testing).
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>
This will be the core of the logic that allows
us to retry break notifications.
When we start the "pending break cycle" we ask for
the current retransmission timemout (rto) on the TCP connection
and remember how many unacked bytes are in the kernel's
send queue. Each time we send bytes into the kernel
we add them to the unacked bytes.
We use a timer using the rto interval in order
to check the amount of unacked bytes again.
The provides send_queu_entry.ack.req will be completed
with tevent_req_done() when everything is completely acked,
tevent_req_nterror(NT_STATUS_IO_TIMEOUT) when
send_queu_entry.ack.timeout is expired or
tevent_req_nterror(connection_error) when the connection
gets disconnected.
It works with support from the FreeBSD and Linux kernels.
For other platforms we just have a fixed rto of 1 second.
And pretend all bytes are acked when we recheck after 1 second.
So only a connection error could trigger tevent_req_nterror(),
but there's no timeout. A follow up commit will most likely
disable support for multi-channel if we don't have kernel support.
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>
For leases we need to use any available connection with the same
client_guid. That means all connections in the client->connections list.
We try the oldest connection first, as that's what windows is doing.
For oplocks we implement the same as that's what the specification
says. Windows behaves different and we have
'smb2 disable oplock break retry = yes' in order to behave like Windows.
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>
This will make it possible to detect errors in order to retry sending
the break on another connection.
For now we always report NT_STATUS_OK, when we delivered the break
notification to the kernel send queue. But that will change in
the following commits.
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>
The following patches require the size of the full sendfile() pdu.
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>
This prepares support for oplock/lease break replay from
the server to the client.
We need some state in order to do replays later.
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
