Bug#29211078 DANGLING REFERENCE TO BUFFER OUT OF SCOPE IN QMGR::FAILREPORTLAB
Bug#29443052 : WL#12680: CLUSTER CRASHED DURING NODE RESTART WITH AN ERROR
Bug#29458648 : WL#12680: NODES CRASHED DURING SHUTDOWN OF CLUSTER WITH DIFFERENT ERRORS ..
Bug#29814551 WL#12680: "INITIAL START, WAITING 15 FOR " LIST UNUSED NODE ID

This patch changes the way a NodeBitMask object is sent across in signals.
The node bitmask should now be sent in a signal section.
Also, only the words until the last non-zero word from the left can be sent
which is made possible by adding a function- getPackedLengthInWords() to struct BitmaskPOD.
This eliminates the need to send unnecessary data and makes the signals shorter.

Also add function in ndb_version.h.in to check if the cluster version supports sending and
receiving node bitmasks in a signal section.

Reviewed-by: Mauritz Sundell <mauritz.sundell@oracle.com>
Reviewed-by: Mikael Ronström <mikael.ronstrom@oracle.com>
Reviewed-by: Sanjana DS <sanjana.ds@oracle.com>

Squashed comments
=================

WL #12564 : Adapt CM_REGREQ to deal with the node bitmask changes.

Send a dummy node bitmask with all zeroes in GSN_CM_REGREQ since it's not used
at the receiving end.
This is for backward compatibility.

WL #12564 : Modify CM_REGCONF protocol to deal with the node bitmask changes

Send node bitmask info in long section of the signal since it wont
fit in a normal signal if we increase the node bitmask size.
Also add relevant upgrade/downgrade code to cope up with these changes.

WL #12564 : Modify CM_REGREF to allow more nodes

Adapt CM_REGREF to deal with the node bitmask changes and add relevant
upgrade/downgrade code.

WL #12564 : Modify FAIL_REP to allow more nodes

Adapt FAIL_REP to deal with the node bitmask changes and add relevant
upgrade/downgrade code.

WL #12564 : Modify MASTER_GCPCONF to allow more nodes

Adapt MASTER_GCPCONF to deal with the node bitmask changes and add relevant
upgrade/downgrade code.

WL #12564 : Modify PREP_FAILREQ and PREP_FAILREF to allow more nodes

Adapt PREP_FAILREQ and PREP_FAILREF to deal with the node bitmask changes
and add relevant upgrade/downgrade code.

WL #12564 : Modify START_LCP_REQ to allow more nodes

Adapt START_LCP_REQ to deal with the node bitmask changes and add
relevant upgrade/downgrade code.

WL #12564 : Modify CNTR_START_CONF to allow more nodes

Adapt CNTR_START_CONF to deal with the node bitmask changes and
add relevant upgrade/downgrade code.

WL#12564: Remove HOT_SPAREREP signal entirely

WL#12564: Remove CM_INIT completely

WL#12564: Remove CNTR_MASTER* signals entirely

WL#12564: BACKUP_CONF, BACKUP_COMPLETE_REP

Removed Ndb node bitmasks from both BACKUP_CONF and BACKUP_COMPLETE_REP
signals. They were sent but not used at reception of the signal.

WL#12564: CHECKNODEGROUPSREQ/CONF

CHECKNODEGROUPSREQ and CHECKNODEGROUPSCONF has a Ndb node bitmask in its
signal. The signal is only sent locally in a node, so no upgrade logic
is required. In most cases the signal is sent as EXECUTE_DIRECT, in those
cases no special treatment is required. SUMA sends the signal and receives
it. SUMA only requires the node bitmask to be received, but to ensure
that new uses of this signal will be possible without changing DIH we
add support to receive the node bitmask in a section. In this case we
send it as cleared from SUMA.

WL#12564: CLOSE_COMREQ/CONF

CLOSE_COMREQ and CLOSE_COMCONF uses a bitmask to communicate to TRPMAN from
QMGR. Actually a bitmask is only needed in CLOSE_COMREQ since the bitmask
returned in CLOSE_COMCONF is always the same bitmask as sent in CLOSE_COMREQ.
For some reason we even use this to assign the same variable as we got the
bitmask from.

So changed signal to carry single node id for cases where no bitmask was
needed. In the case where a bitmask was needed it was only a bitmask for
data nodes and this is transported as usual in a section. Some extra code
was required in TrpmanProxy to ensure that sections could be carried in
the CLOSE_COMREQ signal.

WL#12564: ENABLE_COMREQ/CONF

ENABLE_COMREQ sends a bitmask with nodes to enable communication with.
In most cases only a single node is enabled at a time, only in one case
is a bitmask needed. This is solved in the same fashion as for
CLOSE_COMREQ/CONF. ENABLE_COMCONF needs only the single node sent
back, no need to send bitmask back in ENABLE_COMCONF.

ENABLE_COMREQ/CONF is only sent local in a node, so no need to
handle upgrade situations.

WL#12564: DIH_RESTARTREF/CONF

Bitmask is sent in response to DIH_RESTARTREQ. DIH_RESTART* are local
signals, so no need to handle upgrade situations. DIH_RESTARTREQ can
send a bitmask and array of GCIs, but this is only done in
EXECUTE_DIRECT. NDBCNTR receives DIH_RESTARTREF/CONF as well, but
don't bother with bitmasks, so it is enough to release the section.

WL#12564: READ_NODESCONF

READ_NODESCONF contains 5 data node bitmasks. These are put into a section
in unpacked format. All these signals are only sent in startup and is
local to the node, so no reason for upgrade code.

WL #12564 : Modify STOP_REQ to allow more nodes

Adapt STOP_REQ to deal with the node bitmask changes and
add relevant upgrade/downgrade code.

WL#12564: Remove EMPTY_LCP protocol support

Requires 7.4.3 to support upgrades now.

WL#12564: DIH_RESTARTCONF

Sent bitmask in a section from DIH as it was expected to be received
by both NDBCNTR and QMGR.

WL#12564: DIH_RESTARTCONF

In reception of DIH_RESTARTCONF in QMGR we used the wrong signal to
copy bitmask into, this led to overwrite the GCI value sent from
DIH that was then 0 and this led to President in QMGR not being the
same as the Master in NDBCNTR.

Added more documentation of QMGR behaviour.

Added a bit more printouts to be able to debug problems with allocation
of node ids that sometimes cause issues in NDB.

WL#12564: ISOLATE_ORD

Ensure that ISOLATE_ORD can handle longer data node bitmasks.

WL #12564 : Modify DEFINE_BACKUP_REQ to allow more nodes

Adapt DEFINE_BACKUP_REQ to deal with the node bitmask changes and
add relevant upgrade/downgrade code.

WL #12564 : Modify START_RECREQ to allow more nodes

Adapt START_RECREQ to deal with the node bitmask changes and add relevant
upgrade/downgrade code.

WL#12564: Support for EVENT_REP

Handled StartReport with 5 bitmasks, ConnectCheckStarted with 2 bitmasks,
InfoEvent extended to support lengths up to 4091 bytes and same for
WarningEvent. Limit of data in EVENT_REP signal section set to
1024 words. Limited code to handle MGM server in lower version than
data node.

Long info events must have type repeated in section.

In SavedEventBuffer::scan()
-  assert(data_len <= 25);
+  require(data_len <= MAX_EVENT_REP_SIZE_WORDS);

WL #12564 : Modify NODE_FAILREP to allow more nodes

Adapt NODE_FAILREP to deal with the node bitmask changes and add relevant
upgrade/downgrade code.

WL# 12564: FAIL_REP fine tuning

WL #12564 : Fix MASTER_GCPCONF

WL#12564: ISOLATE_ORD

WL# 12564: FAIL_REP fine tuning

WL# 12564: START_LCP_REQ bitmask transmission improvements

WL# 12564: PREP_FAILREQ and PREP_FAILREF cosmetic changes

WL#12564: EVENT_REP problem in warningEvent

WL# 12564: STOP_REQ improvements

WL# 12564: CM_REGREQ, CM_REGCONF, CM_REGREF, DEFINE_BACKUP_REQ code improvements

Use NdbNodeBitmask48::Size instead of 2.

WL# 12564: START_RECREQ improvements

WL# 12564: NODE_FAILREP cosmetic changes

WL#12564: Support for EVENT_REP

Handled StartReport with 5 bitmasks, ConnectCheckStarted with 2 bitmasks,
InfoEvent extended to support lengths up to 4091 bytes and same for
WarningEvent. Limit of data in EVENT_REP signal section set to
1024 words. Limited code to handle MGM server in lower version than
data node.

WL#12564 Pass node bitmask in section for CNTR_WAITREP:ZWAITPOINT_4_2

To old nodes continue to send it in signal.

Bug#29211078 DANGLING REFERENCE TO BUFFER OUT OF SCOPE IN QMGR::FAILREPORTLAB

Remove dangling reference to buffer out of scope by moving definition of
buffer extra to same scope as pointer msg.

WL#12564 Use TextLength constant for bitmask text buffer sizes.

General replace of size expression for character buffers used in call to
getText() to produce hexdump of node bitmasks.

Typically changing

  char buf[100];

to

  char buf[NdbNodeBitmask::TextLength + 1];

WL# 12564: Send and receive bitmask in NDB_STARTCONF through signal section

WL# 12564: Node bitmask in CONTINUEB in DBSPJ block through signal section

WL# 12564: Zero the node bitmask in GSN_EVENT_REP for ignorance

The node bitmask sent by the BACKUP block is not used at the receiver (mgm client).
Hence, zero those bits for ignorance.