[ Upstream commit bb34f24c7d ]
We should not handle migrate lockres if we are already in
'DLM_CTXT_IN_SHUTDOWN', as that will cause lockres remains after leaving
dlm domain. At last other nodes will get stuck into infinite loop when
requsting lock from us.
The problem is caused by concurrency umount between nodes. Before
receiveing N1's DLM_BEGIN_EXIT_DOMAIN_MSG, N2 has picked up N1 as the
migrate target. So N2 will continue sending lockres to N1 even though
N1 has left domain.
N1 N2 (owner)
touch file
access the file,
and get pr lock
begin leave domain and
pick up N1 as new owner
begin leave domain and
migrate all lockres done
begin migrate lockres to N1
end leave domain, but
the lockres left
unexpectedly, because
migrate task has passed
[piaojun@huawei.com: v3]
Link: http://lkml.kernel.org/r/5A9CBD19.5020107@huawei.com
Link: http://lkml.kernel.org/r/5A99F028.2090902@huawei.com
Signed-off-by: Jun Piao <piaojun@huawei.com>
Reviewed-by: Yiwen Jiang <jiangyiwen@huawei.com>
Reviewed-by: Joseph Qi <jiangqi903@gmail.com>
Reviewed-by: Changwei Ge <ge.changwei@h3c.com>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 1c01967116 upstream.
When a node dies, other live nodes have to choose a new master for an
existed lock resource mastered by the dead node.
As for ocfs2/dlm implementation, this is done by function -
dlm_move_lockres_to_recovery_list which marks those lock rsources as
DLM_LOCK_RES_RECOVERING and manages them via a list from which DLM
changes lock resource's master later.
So without invoking dlm_move_lockres_to_recovery_list, no master will be
choosed after dlm recovery accomplishment since no lock resource can be
found through ::resource list.
What's worse is that if DLM_LOCK_RES_RECOVERING is not marked for lock
resources mastered a dead node, it will break up synchronization among
nodes.
So invoke dlm_move_lockres_to_recovery_list again.
Fixs: 'commit ee8f7fcbe6 ("ocfs2/dlm: continue to purge recovery lockres when recovery master goes down")'
Link: http://lkml.kernel.org/r/63ADC13FD55D6546B7DECE290D39E373CED6E0F9@H3CMLB14-EX.srv.huawei-3com.com
Signed-off-by: Changwei Ge <ge.changwei@h3c.com>
Reported-by: Vitaly Mayatskih <v.mayatskih@gmail.com>
Tested-by: Vitaly Mayatskikh <v.mayatskih@gmail.com>
Cc: Mark Fasheh <mfasheh@versity.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The value of 'stage' must be between 1 and 2, so the switch can't reach
the default case.
Link: http://lkml.kernel.org/r/57FB5EB2.7050002@huawei.com
Signed-off-by: Jun Piao <piaojun@huawei.com>
Cc: Mark Fasheh <mfasheh@versity.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We found a dlm-blocked situation caused by continuous breakdown of
recovery masters described below. To solve this problem, we should
purge recovery lock once detecting recovery master goes down.
N3 N2 N1(reco master)
go down
pick up recovery lock and
begin recoverying for N2
go down
pick up recovery
lock failed, then
purge it:
dlm_purge_lockres
->DROPPING_REF is set
send deref to N1 failed,
recovery lock is not purged
find N1 go down, begin
recoverying for N1, but
blocked in dlm_do_recovery
as DROPPING_REF is set:
dlm_do_recovery
->dlm_pick_recovery_master
->dlmlock
->dlm_get_lock_resource
->__dlm_wait_on_lockres_flags(tmpres,
DLM_LOCK_RES_DROPPING_REF);
Fixes: 8c03439681 ("ocfs2/dlm: clear DROPPING_REF flag when the master goes down")
Link: http://lkml.kernel.org/r/578453AF.8030404@huawei.com
Signed-off-by: Jun Piao <piaojun@huawei.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Reviewed-by: Jiufei Xue <xuejiufei@huawei.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When master handles convert request, it queues ast first and then
returns status. This may happen that the ast is sent before the request
status because the above two messages are sent by two threads. And
right after the ast is sent, if master down, it may trigger BUG in
dlm_move_lockres_to_recovery_list in the requested node because ast
handler moves it to grant list without clear lock->convert_pending. So
remove BUG_ON statement and check if the ast is processed in
dlmconvert_remote.
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Reported-by: Yiwen Jiang <jiangyiwen@huawei.com>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Tariq Saeed <tariq.x.saeed@oracle.com>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
when o2hb detect a node down, it first set the dead node to recovery map
and create ocfs2rec which will replay journal for dead node. o2hb
thread then call dlm_do_local_recovery_cleanup() to delete the lock for
dead node. After the lock of dead node is gone, locks for other nodes
can be granted and may modify the meta data without replaying journal of
the dead node. The detail is described as follows.
N1 N2 N3(master)
modify the extent tree of
inode, and commit
dirty metadata to journal,
then goes down.
o2hb thread detects
N1 goes down, set
recovery map and
delete the lock of N1.
dlm_thread flush ast
for the lock of N2.
do not detect the death
of N1, so recovery map is
empty.
read inode from disk
without replaying
the journal of N1 and
modify the extent tree
of the inode that N1
had modified.
ocfs2rec recover the
journal of N1.
The modification of N2
is lost.
The modification of N1 and N2 are not serial, and it will lead to
read-only file system. We can set recovery_waiting flag to the lock
resource after delete the lock for dead node to prevent other node from
getting the lock before dlm recovery. After dlm recovery, the recovery
map on N2 is not empty, ocfs2_inode_lock_full_nested() will wait for ocfs2
recovery.
Signed-off-by: Jiufei Xue <xuejiufei@huawei.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If master migrate this lock resource to node when it happened to purge
it, a new lock resource will be created and inserted into hash list. If
then master goes down, the lock resource being purged is recovered, so
there exist two lock resource with different owner. So return error to
master if the lock resource is in DROPPING state, master will retry to
migrate this lock resource.
Signed-off-by: xuejiufei <xuejiufei@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If the master goes down after return in-progress for deref message. The
lock resource on non-master node can not be purged. Clear the
DROPPING_REF flag and recovery it.
Signed-off-by: xuejiufei <xuejiufei@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When recovery master down, dlm_do_local_recovery_cleanup() only remove
the $RECOVERY lock owned by dead node, but do not clear the refmap bit.
Which will make umount thread falling in dead loop migrating $RECOVERY
to the dead node.
Signed-off-by: xuejiufei <xuejiufei@huawei.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Found this when do patch review, remove to make it clear and save a
little cpu time.
Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Joseph Qi <joseph.qi@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We found a race between purge and migration when doing code review.
Node A put lockres to purgelist before receiving the migrate message
from node B which is the master. Node A call dlm_mig_lockres_handler to
handle this message.
dlm_mig_lockres_handler
dlm_lookup_lockres
>>>>>> race window, dlm_run_purge_list may run and send
deref message to master, waiting the response
spin_lock(&res->spinlock);
res->state |= DLM_LOCK_RES_MIGRATING;
spin_unlock(&res->spinlock);
dlm_mig_lockres_handler returns
>>>>>> dlm_thread receives the response from master for the deref
message and triggers the BUG because the lockres has the state
DLM_LOCK_RES_MIGRATING with the following message:
dlm_purge_lockres:209 ERROR: 6633EB681FA7474A9C280A4E1A836F0F: res
M0000000000000000030c0300000000 in use after deref
Signed-off-by: Jiufei Xue <xuejiufei@huawei.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Reviewed-by: Yiwen Jiang <jiangyiwen@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A node can mount multiple ocfs2 volumes. And if thread names are same for
each volume/domain, it will bring inconvenience when analyzing problems
because we have to identify which volume/domain the messages belong to.
Since thread name will be printed to messages, so add volume uuid or dlm
name to thread name can benefit problem analysis.
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Gang He <ghe@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
dlm_lockres_put will call dlm_lockres_release if it is the last
reference, and then it may call dlm_print_one_lock_resource and
take lockres spinlock.
So unlock lockres spinlock before dlm_lockres_put to avoid deadlock.
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The order of the following three spinlocks should be:
dlm_domain_lock < dlm_ctxt->spinlock < dlm_lock_resource->spinlock
But dlm_dispatch_assert_master() is called while holding
dlm_ctxt->spinlock and dlm_lock_resource->spinlock, and then it calls
dlm_grab() which will take dlm_domain_lock.
Once another thread (for example, dlm_query_join_handler) has already
taken dlm_domain_lock, and tries to take dlm_ctxt->spinlock deadlock
happens.
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: "Junxiao Bi" <junxiao.bi@oracle.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Revert commit f83c7b5e9f ("ocfs2/dlm: use list_for_each_entry instead
of list_for_each").
list_for_each_entry() will dereference its `pos' argument, which can be
NULL in dlm_process_recovery_data().
Reported-by: Julia Lawall <julia.lawall@lip6.fr>
Reported-by: Fengguang Wu <fengguang.wu@gmail.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Use list_for_each_entry instead of list_for_each to simplify code.
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When the recovery master is down, the owner of $RECOVERY calls
dlm_do_local_recovery_cleanup() to prune any $RECOVERY entries for dead
nodes. The lock is in the granted list and the refcount must be 2. We
should put twice to remove this lock. Otherwise, it will lead to a memory
leak.
Signed-off-by: joyce.xue <xuejiufei@huawei.com>
Reported-by: yangwenfang <vicky.yangwenfang@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In dlm_process_recovery_data, only when dlm_new_lock failed the ret will
be set to -ENOMEM. And in this case, newlock is definitely NULL. So
test newlock is meaningless, remove it.
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Reviewed-by: Alex Chen <alex.chen@huawei.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Do not BUG() if GFP_ATOMIC allocation fails in dlm_dispatch_assert_master.
Instead, return -ENOMEM to the sender and then retry.
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Reviewed-by: Alex Chen <alex.chen@huawei.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
dlm_lockres_put() should be called without &res->spinlock, otherwise a
deadlock case may happen.
spin_lock(&res->spinlock)
...
dlm_lockres_put
->dlm_lockres_release
->dlm_print_one_lock_resource
->spin_lock(&res->spinlock)
Signed-off-by: Alex Chen <alex.chen@huawei.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When workqueue is delayed, it may occur that a lockres is purged while it
is still queued for master assert. it may trigger BUG() as follows.
N1 N2
dlm_get_lockres()
->dlm_do_master_requery
is the master of lockres,
so queue assert_master work
dlm_thread() start running
and purge the lockres
dlm_assert_master_worker()
send assert master message
to other nodes
receiving the assert_master
message, set master to N2
dlmlock_remote() send create_lock message to N2, but receive DLM_IVLOCKID,
if it is RECOVERY lockres, it triggers the BUG().
Another BUG() is triggered when N3 become the new master and send
assert_master to N1, N1 will trigger the BUG() because owner doesn't
match. So we should not purge lockres when it is queued for assert
master.
Signed-off-by: joyce.xue <xuejiufei@huawei.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We found there is a conversion deadlock when the owner of lockres
happened to crash before send DLM_PROXY_AST_MSG for a downconverting
lock. The situation is as follows:
Node1 Node2 Node3
the owner of lockresA
lock_1 granted at EX mode
and call ocfs2_cluster_unlock
to decrease ex_holders.
converting lock_3 from
NL to EX
send DLM_PROXY_AST_MSG
to Node1, asking Node 1
to downconvert.
receiving DLM_PROXY_AST_MSG,
thread ocfs2dc send
DLM_CONVERT_LOCK_MSG
to Node2 to downconvert
lock_1(EX->NL).
lock_1 can be granted and
put it into pending_asts
list, return DLM_NORMAL.
then something happened
and Node2 crashed.
received DLM_NORMAL, waiting
for DLM_PROXY_AST_MSG.
selected as the recovery
master, receving migrate
lock from Node1, queue
lock_1 to the tail of
converting list.
After dlm recovery, converting list in the master of lockresA(Node3)
will be: converting list head <-> lock_3(NL->EX) <->lock_1(EX<->NL).
Requested mode of lock_3 is not compatible with the granted mode of
lock_1, so it can not be granted. and lock_1 can not downconvert
because covnerting queue is strictly FIFO. So a deadlock is created.
We think function dlm_process_recovery_data() should queue_ast for
lock_1 or alter the order of lock_1 and lock_3, so dlm_thread can
process lock_1 first. And if there are multiple downconverting locks,
they must convert form PR to NL, so no need to sort them.
Signed-off-by: joyce.xue <xuejiufei@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There is a race window in dlm_do_recovery() between dlm_remaster_locks()
and dlm_reset_recovery() when the recovery master nearly finish the
recovery process for a dead node. After the master sends FINALIZE_RECO
message in dlm_remaster_locks(), another node may become the recovery
master for another dead node, and then send the BEGIN_RECO message to
all the nodes included the old master, in the handler of this message
dlm_begin_reco_handler() of old master, dlm->reco.dead_node and
dlm->reco.new_master will be set to the second dead node and the new
master, then in dlm_reset_recovery(), these two variables will be reset
to default value. This will cause new recovery master can not finish
the recovery process and hung, at last the whole cluster will hung for
recovery.
old recovery master: new recovery master:
dlm_remaster_locks()
become recovery master for
another dead node.
dlm_send_begin_reco_message()
dlm_begin_reco_handler()
{
if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
return -EAGAIN;
}
dlm_set_reco_master(dlm, br->node_idx);
dlm_set_reco_dead_node(dlm, br->dead_node);
}
dlm_reset_recovery()
{
dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
dlm_set_reco_master(dlm, O2NM_INVALID_NODE_NUM);
}
will hang in dlm_remaster_locks() for
request dlm locks info
Before send FINALIZE_RECO message, recovery master should set
DLM_RECO_STATE_FINALIZE for itself and clear it after the recovery done,
this can break the race windows as the BEGIN_RECO messages will not be
handled before DLM_RECO_STATE_FINALIZE flag is cleared.
A similar race may happen between new recovery master and normal node
which is in dlm_finalize_reco_handler(), also fix it.
Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
Reviewed-by: Srinivas Eeda <srinivas.eeda@oracle.com>
Reviewed-by: Wengang Wang <wen.gang.wang@oracle.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This issue was introduced by commit 800deef3f6 ("ocfs2: use
list_for_each_entry where benefical") in 2007 where it replaced
list_for_each with list_for_each_entry. The variable "lock" will point
to invalid data if "tmpq" list is empty and a panic will be triggered
due to this. Sunil advised reverting it back, but the old version was
also not right. At the end of the outer for loop, that
list_for_each_entry will also set "lock" to an invalid data, then in the
next loop, if the "tmpq" list is empty, "lock" will be an stale invalid
data and cause the panic. So reverting the list_for_each back and reset
"lock" to NULL to fix this issue.
Another concern is that this seemes can not happen because the "tmpq"
list should not be empty. Let me describe how.
old lock resource owner(node 1): migratation target(node 2):
image there's lockres with a EX lock from node 2 in
granted list, a NR lock from node x with convert_type
EX in converting list.
dlm_empty_lockres() {
dlm_pick_migration_target() {
pick node 2 as target as its lock is the first one
in granted list.
}
dlm_migrate_lockres() {
dlm_mark_lockres_migrating() {
res->state |= DLM_LOCK_RES_BLOCK_DIRTY;
wait_event(dlm->ast_wq, !dlm_lockres_is_dirty(dlm, res));
//after the above code, we can not dirty lockres any more,
// so dlm_thread shuffle list will not run
downconvert lock from EX to NR
upconvert lock from NR to EX
<<< migration may schedule out here, then
<<< node 2 send down convert request to convert type from EX to
<<< NR, then send up convert request to convert type from NR to
<<< EX, at this time, lockres granted list is empty, and two locks
<<< in the converting list, node x up convert lock followed by
<<< node 2 up convert lock.
// will set lockres RES_MIGRATING flag, the following
// lock/unlock can not run
dlm_lockres_release_ast(dlm, res);
}
dlm_send_one_lockres()
dlm_process_recovery_data()
for (i=0; i<mres->num_locks; i++)
if (ml->node == dlm->node_num)
for (j = DLM_GRANTED_LIST; j <= DLM_BLOCKED_LIST; j++) {
list_for_each_entry(lock, tmpq, list)
if (lock) break; <<< lock is invalid as grant list is empty.
}
if (lock->ml.node != ml->node)
BUG() >>> crash here
}
I see the above locks status from a vmcore of our internal bug.
Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
Reviewed-by: Wengang Wang <wen.gang.wang@oracle.com>
Cc: Sunil Mushran <sunil.mushran@gmail.com>
Reviewed-by: Srinivas Eeda <srinivas.eeda@oracle.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A parallel umount on 4 nodes triggered a bug in
dlm_process_recovery_date(). Here's the situation:
Receiving MIG_LOCKRES message, A node processes the locks in migratable
lockres. It copys lvb from migratable lockres when processing the first
valid lock.
If there is a lock in the blocked list with the EX level, it triggers the
BUG. Since valid lvbs are set when locks are granted with EX or PR
levels, locks in the blocked list cannot have valid lvbs. Therefore I
think we should skip the locks in the blocked list.
Signed-off-by: Xuejiufei <xuejiufei@huawei.com>
Signed-off-by: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
dlm_do_local_recovery_cleanup() should force clean refmap if the owner of
lockres is UNKNOWN. Otherwise node may hang when umounting filesystems.
Here's the situation:
Node1 Node2
dlmlock()
-> dlm_get_lock_resource()
send DLM_MASTER_REQUEST_MSG to
other nodes.
trying to master this lockres,
return MAYBE.
selected as the master of lockresA,
set mle->master to Node1,
and do assert_master,
send DLM_ASSERT_MASTER_MSG to Node2.
Node 2 has interest on lockresA
and return
DLM_ASSERT_RESPONSE_MASTERY_REF
then something happened and
Node2 crashed.
Receiving DLM_ASSERT_RESPONSE_MASTERY_REF, set Node2 into refmap, and keep
sending DLM_ASSERT_MASTER_MSG to other nodes
o2hb found node2 down, calling dlm_hb_node_down() -->
dlm_do_local_recovery_cleanup() the master of lockresA is still UNKNOWN,
no need to call dlm_free_dead_locks().
Set the master of lockresA to Node1, but Node2 stills remains in refmap.
When Node1 umount, it found that the refmap of lockresA is not empty and
attempted to migrate it to Node2, But Node2 is already down, so umount
hang, trying to migrate lockresA again and again.
Signed-off-by: joyce <xuejiufei@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Jie Liu <jeff.liu@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
dlm_request_all_locks() should deal with the status sent from target node
if DLM_LOCK_REQUEST_MSG is sent successfully, or recovery master will fall
into endless loop, waiting for other nodes to send locks and
DLM_RECO_DATA_DONE_MSG to me.
NodeA NodeB
selected as recovery master
dlm_remaster_locks()
->dlm_request_all_locks()
send DLM_LOCK_REQUEST_MSG to nodeA
It happened that NodeA cannot alloc memory when it processes this
message. dlm_request_all_locks_handler() do not queue
dlm_request_all_locks_worker and returns -ENOMEM. It will never send
locks and DLM_RECO_DATA_DONE_MSG to NodeB.
NodeB do not deal with the status
sent from nodeA, and will fall in
endless loop waiting for the
recovery state of NodeA to be
changed.
Signed-off-by: joyce <xuejiufei@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Jeff Liu <jeff.liu@oracle.com>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
dlm_begin_reco_handler() returns without putting dlm when dlm recovery
state is DLM_RECO_STATE_FINALIZE.
Signed-off-by: joyce <xuejiufei@huawei.com>
Reviewed-by: Jie Liu <jeff.liu@oracle.com>
Acked-by: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In dlm_request_all_locks, ret is type enum. But o2net_send_message
returns a type int value. Then it will never run into the following
error branch. So we should change the ret type from enum to int.
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Acked-by: Sunil Mushran <sunil.mushran@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Below 3 functions have already been declared in dlmcommon.h, so we have
no need to declare them again in dlmrecovery.c:
dlm_complete_recovery_thread
dlm_launch_recovery_thread
dlm_kick_recovery_thread
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Acked-by: Sunil Mushran <sunil.mushran@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
dlm_mig_lockres_handler() is missing a dlm_lockres_put() on an error path.
Signed-off-by: joyce <xuejiufei@huawei.com>
Reviewed-by: shencanquan <shencanquan@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kfree on a NULL pointer is a no-op. Remove the redundant null pointer
check.
Signed-off-by: Sachin Kamat <sachin.kamat@linaro.org>
Acked-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I'm not sure why, but the hlist for each entry iterators were conceived
list_for_each_entry(pos, head, member)
The hlist ones were greedy and wanted an extra parameter:
hlist_for_each_entry(tpos, pos, head, member)
Why did they need an extra pos parameter? I'm not quite sure. Not only
they don't really need it, it also prevents the iterator from looking
exactly like the list iterator, which is unfortunate.
Besides the semantic patch, there was some manual work required:
- Fix up the actual hlist iterators in linux/list.h
- Fix up the declaration of other iterators based on the hlist ones.
- A very small amount of places were using the 'node' parameter, this
was modified to use 'obj->member' instead.
- Coccinelle didn't handle the hlist_for_each_entry_safe iterator
properly, so those had to be fixed up manually.
The semantic patch which is mostly the work of Peter Senna Tschudin is here:
@@
iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host;
type T;
expression a,c,d,e;
identifier b;
statement S;
@@
-T b;
<+... when != b
(
hlist_for_each_entry(a,
- b,
c, d) S
|
hlist_for_each_entry_continue(a,
- b,
c) S
|
hlist_for_each_entry_from(a,
- b,
c) S
|
hlist_for_each_entry_rcu(a,
- b,
c, d) S
|
hlist_for_each_entry_rcu_bh(a,
- b,
c, d) S
|
hlist_for_each_entry_continue_rcu_bh(a,
- b,
c) S
|
for_each_busy_worker(a, c,
- b,
d) S
|
ax25_uid_for_each(a,
- b,
c) S
|
ax25_for_each(a,
- b,
c) S
|
inet_bind_bucket_for_each(a,
- b,
c) S
|
sctp_for_each_hentry(a,
- b,
c) S
|
sk_for_each(a,
- b,
c) S
|
sk_for_each_rcu(a,
- b,
c) S
|
sk_for_each_from
-(a, b)
+(a)
S
+ sk_for_each_from(a) S
|
sk_for_each_safe(a,
- b,
c, d) S
|
sk_for_each_bound(a,
- b,
c) S
|
hlist_for_each_entry_safe(a,
- b,
c, d, e) S
|
hlist_for_each_entry_continue_rcu(a,
- b,
c) S
|
nr_neigh_for_each(a,
- b,
c) S
|
nr_neigh_for_each_safe(a,
- b,
c, d) S
|
nr_node_for_each(a,
- b,
c) S
|
nr_node_for_each_safe(a,
- b,
c, d) S
|
- for_each_gfn_sp(a, c, d, b) S
+ for_each_gfn_sp(a, c, d) S
|
- for_each_gfn_indirect_valid_sp(a, c, d, b) S
+ for_each_gfn_indirect_valid_sp(a, c, d) S
|
for_each_host(a,
- b,
c) S
|
for_each_host_safe(a,
- b,
c, d) S
|
for_each_mesh_entry(a,
- b,
c, d) S
)
...+>
[akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c]
[akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c]
[akpm@linux-foundation.org: checkpatch fixes]
[akpm@linux-foundation.org: fix warnings]
[akpm@linux-foudnation.org: redo intrusive kvm changes]
Tested-by: Peter Senna Tschudin <peter.senna@gmail.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch adds a new dlm message DLM_BEGIN_EXIT_DOMAIN_MSG and ups the dlm
protocol to 1.2.
o2dlm sends this new message in dlm_unregister_domain() to mark the beginning
of the exit domain. This message is sent to all nodes in the domain.
Currently o2dlm has no way of informing other nodes of its impending exit.
This information is useful as the other nodes could disregard the exiting
node in certain operations. For example, in resource migration. If two or
more nodes were umounting in parallel, it would be more efficient if o2dlm
were to choose a non-exiting node to be the new master node rather than an
exiting one.
Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: Joel Becker <jlbec@evilplan.org>
mlog_exit is used to record the exit status of a function.
But because it is added in so many functions, if we enable it,
the system logs get filled up quickly and cause too much I/O.
So actually no one can open it for a production system or even
for a test.
This patch just try to remove it or change it. So:
1. if all the error paths already use mlog_errno, it is just removed.
Otherwise, it will be replaced by mlog_errno.
2. if it is used to print some return value, it is replaced with
mlog(0,...).
mlog_exit_ptr is changed to mlog(0.
All those mlog(0,...) will be replaced with trace events later.
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
ENTRY is used to record the entry of a function.
But because it is added in so many functions, if we enable it,
the system logs get filled up quickly and cause too much I/O.
So actually no one can open it for a production system or even
for a test.
So for mlog_entry_void, we just remove it.
for mlog_entry(...), we replace it with mlog(0,...), and they
will be replace by trace event later.
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
In the following situation, there remains an incorrect bit in refmap on the
recovery master. Finally the recovery master will fail at purging the lockres
due to the incorrect bit in refmap.
1) node A has no interest on lockres A any longer, so it is purging it.
2) the owner of lockres A is node B, so node A is sending de-ref message
to node B.
3) at this time, node B crashed. node C becomes the recovery master. it recovers
lockres A(because the master is the dead node B).
4) node A migrated lockres A to node C with a refbit there.
5) node A failed to send de-ref message to node B because it crashed. The failure
is ignored. no other action is done for lockres A any more.
For mormal, re-send the deref message to it to recovery master can fix it. Well,
ignoring the failure of deref to the original master and not recovering the lockres
to recovery master has the same effect. And the later is simpler.
Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Acked-by: Srinivas Eeda <srinivas.eeda@oracle.com>
Cc: stable@kernel.org
Signed-off-by: Joel Becker <joel.becker@oracle.com>
dlm->recovery_map is defined as
unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
We should treat O2NM_MAX_NODES as the bit map size in bits.
This patches fixes a bit operation that takes O2NM_MAX_NODES + 1 as bitmap size.
Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Print the node number of a peer node if sending it a message failed.
Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
If a node down event happens while dlm shutdown in progress, dlm recovery
should be done before dlm is shutdown. We can't migrate unrecovered locks,
obviously. But dlm_reco_thread only does recovery if the dlm_state is
in DLM_CTXT_JOINED.
dlm_reco_thread should do recovery if dlm_state is in DLM_CTXT_JOINED or
DLM_CTXT_IN_SHUTDOWN.
Signed-off-by: Srinivas Eeda <srinivas.eeda@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
During recovery, the dlm frees the locks for the dead node. If it finds a
lock in a resource for the dead node, it expects that node to also have a
ref in that lock resource. If not, it BUGs.
ossbz#1175 was filed with the above BUG. Now, while it is correct that we
should be expecting the ref, I see no reason why we have to BUG. After all,
we are freeing up the lock and clearing the ref.
This patch replaces the BUG_ON with a printk(). Hopefully, that will give
us more clues next time this happens.
http://oss.oracle.com/bugzilla/show_bug.cgi?id=1175
Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Acked-by: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Mainline commit aad1b15310 made the
dlm_begin_reco_handler() return -EAGAIN instead of EAGAIN.
As this error is transmitted over the wire, we want the receiver,
dlm_send_begin_reco_message(), to understand both the older EAGAIN and
the newer -EAGAIN, to allow rolling upgrade of the cluster nodes.
Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
When a lock resource is migrated, the dlm compares the migrated
locks with that that was already existing on the new node. If the
comparison fails, it BUGs. This patch prints more messages when the
comparison fails inorder to help with the root cause analyis.
http://oss.oracle.com/bugzilla/show_bug.cgi?id=1206
This does not fix bz1206. However, if we run into it again, we will
have more information to chew on.
Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
During lock resource migration, o2dlm fills the packet with a LVB from the
first valid lock. For sanity, it ensures that the other valid locks have the
same LVB. If not, it BUGs.
The valid locks are ones that have granted EX or PR lock levels and are either
on the Granted or Converting lists. Locks in the Blocked list cannot have a
valid LVB.
This patch ensures that we skip the locks in the Blocked list.
Fixes oss bugzilla#1202
http://oss.oracle.com/bugzilla/show_bug.cgi?id=1202
Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
We used to return positive EAGAIN to indicate a retry action
is needed in dlm_begin_reco_handler(). Now we return negative
-EAGAIN to erase the confusion caused by this error code.
Signed-off-by: Tiger Yang <tiger.yang@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Jun Piao
Changwei Ge
Joseph Qi
Jiufei Xue
Junxiao Bi
Andrew Morton
alex chen
Sachin Kamat
Sasha Levin
Sunil Mushran
Tao Ma
Wengang Wang
Srinivas Eeda
Tiger Yang