Use rcu to free rsb structs, and hold the rcu read lock
while looking up rsb structs. This allows us to avoid an
extra hash table lookup for an rsb. A new rsb flag HASHED
is added which is set while the rsb is in the hash table.
This flag is checked in place of repeating the hash table
lookup.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Remove a few calls to add_scan() and del_scan() in cases where
the rsb is a dir record, so the rsb should never be placed on
the scan list at all. Add WARN_ON to catch cases where this
is done.
Signed-off-by: David Teigland <teigland@redhat.com>
The old terminology of "toss" and "keep" is no longer an
accurate description of the rsb states and lists, so change
the names to "inactive" and "active". The old names had
also been copied into the scanning code, which is changed
back to use the "scan" name.
- "active" rsb structs have lkb's attached, and are ref counted.
- "inactive" rsb structs have no lkb's attached, are not ref counted.
- "scan" list is for rsb's that can be freed after a timeout period.
- "slow" lists are for infrequent iterations through active or
inactive rsb structs.
- inactive rsb structs that are directory records will not be put
on the scan list, since they are not freed based on timeouts.
- inactive rsb structs that are not directory records will be
put on the scan list to be freed, since they are not longer needed.
Signed-off-by: David Teigland <teigland@redhat.com>
According to kernel doc idr is deprecated and xarrays should be used
nowadays. This patch is moving the lkb idr implementation to xarrays.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch drops the own written rsb pre allocation mechanism as this is
already done by using kmem caches, we don't need another layer on top of
that to running some pre allocation scheme.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch removes a kref_init() that isn't necessary because the rsb is
created for toss list. Under toss list the rsb should not have any
reference counting logic. If in theory the rsb gets to into keep list
then a kref_init() for res_ref will be initiated.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch changes the orphans mutex to a spinlock since commit
c288745f1d ("dlm: avoid blocking receive at the end of recovery") is
using a rwlock_t to lock the DLM message receive path and do_purge() can
be called while this lock is held that forbids to sleep.
We need to use spin_lock_bh() because also a user context that calls
dlm_user_purge() can call do_purge() and since commit 92d59adfaf
("dlm: do message processing in softirq context") the DLM message
receive path is done under softirq context.
Fixes: c288745f1d ("dlm: avoid blocking receive at the end of recovery")
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Closes: https://lore.kernel.org/gfs2/9ad928eb-2ece-4ad9-a79c-d2bce228e4bc@moroto.mountain/
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Convert the lock for lkbidr to an rwlock. Most idr lookups will use
the read lock.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
The conversion to rhashtable introduced a hash table lock per lockspace,
in place of per bucket locks. To make this more scalable, switch to
using a rwlock for hash table access. The common case fast path uses
it as a read lock.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Currently the scand kthread acts like a garbage collection for expired
rsbs on toss list, to clean them up after a certain timeout. It triggers
every couple of seconds and iterates over the toss list while holding
ls_rsbtbl_lock for the whole hash bucket iteration.
To reduce the amount of time holding ls_rsbtbl_lock, we now handle the
disposal of expired rsbs using a per-lockspace timer that expires for the
earliest tossed rsb on the lockspace toss queue. This toss queue is
ordered according to the rsb res_toss_time with the earliest tossed rsb
as the first entry. The toss timer will only trylock() necessary locks,
since it is low priority garbage collection, and will rearm the timer
if trylock() fails. If the timer function does not find any expired
rsb's, it rearms the timer with the next earliest expired rsb.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
In the past we had problems when an rsb had a reference counter greater
than one while in the toss state. An rsb in the toss state is not
actively used for locking, and should not have any other references
apart from the single ref keeping it on the rsb hash. Shift to freeing
rsb's directly rather than using kref_put to free them, since the ref
counting is not meant to be used in this state. Add warnings if ref
counting is seen while an rsb is in the toss state.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Replace our own hash table with the more advanced rhashtable
for keeping rsb structs.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
To prepare for using rhashtable, add two rsb lists for iterating
through rsb's in two uncommon cases where this is necesssary:
- when dumping rsb state from debugfs, now using seq_list.
- when looking at all rsb's during recovery.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
There are several places where lock processing can perform two hash table
lookups, first in the "keep" list, and if not found, in the "toss" list.
This patch introduces a new rsb state flag "RSB_TOSS" to represent the
difference between the state of being on keep vs toss list, so that the
two lists can be combined. This avoids cases of two lookups.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Prepare to replace our own hash table with rhashtable by replacing
the per-bucket locks in our own hash table with a single lock.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Use spin_lock_bh for all spinlocks involved in message processing,
in preparation for softirq message processing. DLM lock requests
from user space involve dlm processing in user context, in addition
to the standard kernel context, necessitating bh variants.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Remove an explicit schedule() call in the message processing path,
in preparation for softirq message processing.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Convert ls_recv_active rw_semaphore to an rwlock to avoid
sleeping, in preparation for softirq message processing.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
The end of the recovery process transitioned to normal message
processing by temporarily blocking the receiving context,
processing saved messages, then unblocking the receiving
context. To avoid blocking the receiving context, the old
wait_queue and mutex are replaced by a new rwlock and the new
RECV_MSG_BLOCKED flag. Received messages are added to the
list of saved messages, protected by the rwlock, until the
flag is cleared, which happens when all saved messages have
been processed.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Convert the rsb struct res_lock from a mutex to a spinlock
in preparation for processing messages in softirq context.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Convert the waiters mutex to a spinlock in prepration for
processing messages in softirq context.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
The waiters_mutex no longer needs to be used in the waiters recovery
functions dlm_recover_waiters_pre() and dlm_recover_waiters_pre().
During recovery, ordinary locking operations are paused, and the
recovery thread is the only context accessing the waiters list,
so the lock is not needed.
Access to the waiters list from debugfs functions is avoided by
taking the top level recovery lock in the debugfs dump function.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Move the rsb root_list from the lockspace to a stack variable since
it is now only used by the ls_recover() function.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Add a new "masters_list" for master rsb structs, with a new
rwlock. The new list is created and used during the recovery
process to send the master rsb names to new nodes. With this
change, the current "root_list" can be used without locking.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Replace GFP_NOFS with GFP_ATOMIC. Also stop using idr_preload which
uses a non-bh spin_lock. This is further preparation for softirq
message processing.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Remove the context parameter for message allocations and
always use GFP_ATOMIC. This prepares for softirq message
processing.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Get rid of the unnecessary refcounting on callback structs.
Copy interesting callback info into the lkb struct rather
than maintaining pointers to callback structs from the lkb.
This goes back to the way things were done prior to
commit 61bed0baa4 ("fs: dlm: use a non-static queue for callbacks").
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch fixes the following issue:
node 1 is dir
node 2 is master
node 3 is other
1->2: unlock
2: put final lkb, rsb moved to toss
2->1: unlock_reply
1: queue lkb callback with EUNLOCK
2->1: remove
1: receive_remove ignored (rsb on keep because of queued lkb callback)
1: complete lkb callback, put_lkb, move rsb to toss
3->1: lookup
1->3: lookup_reply master=2
3->2: request
2->3: request_reply EBADR
In summary:
An unexpected lkb reference causes the rsb to remain on the wrong list.
The rsb being on the wrong list causes receive_remove to be ignored.
An ignored receive_remove causes inconsistent dir and master state.
This sequence requires an unusually long delay in delivering the unlock
callback, because the remove message from 2->1 usually happens after
some seconds. So, it's not known exactly how frequently this sequence
occurs in pratice. It's possible that the same end result could also
have another unknown cause.
The solution for this issue is to further separate callback state
from the lkb, so that an lkb reference (and from that, an rsb ref)
are not held while a callback remains queued. Then, within the
unlock_reply, the lkb will be freed and the rsb moved to the toss
list. So, the receive_remove will not be ignored.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
When a lock is waiting for a reply for a remote operation, and recovery
interrupts this "waiters" state, the remote operation is voided by the
recovery, and no reply will be processed. The lkb waiters state for the
remote operation is forcibly reset/cleared, so that the lock operation
can be restarted after recovery. Improve the comments describing this.
Signed-off-by: David Teigland <teigland@redhat.com>
Revert "fs: dlm: handle lkb wait count as atomic_t"
This reverts commit 75a7d60134.
This counter does not need to be atomic. As the comment in
the reverted commit mentions, the counter is protected by
the rsb lock.
Signed-off-by: David Teigland <teigland@redhat.com>
The dlm receive buffer should be never manipulated as DLM is the last
instance of parsing layer. This patch constify the whole receive buffer
so we are sure it never gets manipulated when it's being parsed.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Currently dlm_recover_master_copy() manipulates the receive buffer of an
rcom lock message and modifies it on the fly so a later memcpy() to a
new rcom message with the same message has those new values. This patch
avoids manipulating the received rcom message by store the values for
the new rcom message in paremter assigned with call by reference. Later
when dlm_send_rcom_lock() constructs a new message and memcpy() the
receive buffer those values will be set on the new constructed message.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch removes a read of the ls->ls_recover_seq uint64_t number in
_create_rcom(). If the ls->ls_recover_seq is readed the ls_recover_lock
need to held. However this number was always readed before when any rcom
message is received and it's not necessary to read it again from a per
lockspace variable to use it for the replying message. This patch will
pass the sequence number as parameter so another read of ls->ls_recover_seq
and holding the ls->ls_recover_lock is not required.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Currently the lkb_wait_count is locked by the rsb lock and it should be
fine to handle lkb_wait_count as non atomic_t value. However for the
overall process of reducing locking this patch converts it to an
atomic_t value.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch warns about messages which are received from nodes who
already left the lockspace resource signaled by the cluster manager.
Before commit 489d8e559c ("fs: dlm: add reliable connection if
reconnect") there was a synchronization issue with the socket
lifetime and the cluster event of leaving a lockspace and other
nodes did not stop of sending messages because the cluster manager has a
pending message to leave the lockspace. The reliable session layer for
dlm use sequence numbers to ensure dlm message were never being dropped.
If this is not corrected synchronized we have a problem, this patch will
use the filter case and turn it into a WARN_ON_ONCE() so we seeing such
issue on the kernel log because it should never happen now.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Commit 7175e131eb ("fs: dlm: fix invalid derefence of sb_lvbptr")
fixes an issue when the lkb->lkb_lvbptr set to an dangled pointer and an
followed memcpy() would fail. It was fixed by an additional check of
DLM_LKF_VALBLK flag. The mentioned commit forgot to add an additional check
if DLM_LKF_VALBLK is set for the additional amount of LVB data allocated
in a dlm message. This patch is changing the message allocation to check
additionally if DLM_LKF_VALBLK is set otherwise a dangled lkb->lkb_lvbptr
pointer would allocated zero LVB message data which not gets filled with
actual data.
This patch is however only a cleanup to reduce the amount of zero bytes
transmitted over network as receive_lvb() will only evaluates message LVB
data if DLM_LKF_VALBLK is set.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch moves lkb_sbflags handling to atomic bits ops. This should
prepare for a possible manipulating of lkb_sbflags flags at the same
time by concurrent execution.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch moves the rsb hash table handling to atomic flag operations.
The flag operations for DLM_RTF_SHRINK are protected by
ls->ls_rsbtbl[b].lock. However we switch to atomic ops if new possible
flags will be used in a different way and don't assume such lock
dependencies.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch will move the lkb_flags value to the recently introduced
lkb_iflags value. For lkb_iflags we use atomic bit operations because
some flags like DLM_IFL_CB_PENDING are used while non rsb lock is held
to avoid issues with other flag manipulations which might run at the
same time we switch to atomic bit operations. Snapshot the bit values to
an uint32_t value is only used for debugging/logging use cases and don't
need to be 100% correct.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Currently manipulating lkb_dflags assumes to held the rsb lock assigned
to the lkb. This is held by dlm message processing after certain
time to lookup the right rsb from the received lkb message id. For user
space locks flags, which is currently the only use case for lkb_dflags,
flags are also being set during dlm character device handling without
holding the rsb lock. To minimize the risk that bit operations are
getting corrupted we switch to atomic bit operations. This patch will
also introduce helpers to snapshot atomic bit values in an non atomic
way. There might be still issues with the flag handling e.g. running in
case of manipulating bit ops and snapshot them at the same time, but this
patch minimize them and will start to use atomic bit operations.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch stores lkb distributed flags value in an separate value
instead of sharing internal and distributed flags in lkb->lkb_flags value.
This has the advantage to not mask/write back flag values in
receive_flags() functionality. The dlm debug_fs does not provide the
distributed flags anymore, those can be added in future.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
The DLM_IFL_LOCAL_MS flag is an internal non shared flag but used in
m_flags of dlm messages. It is not shared because it is only used for
local messaging. Instead using DLM_IFL_LOCAL_MS in dlm messages we pass a
parameter around to signal local messaging or not. This patch is adding
the local parameter to signal local messaging.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch renames DLM_IFL_STUB_MS to DLM_IFL_LOCAL_MS flag. The
DLM_IFL_STUB_MS flag is somewhat misnamed, it means the dlm message is
used for local message transfer only. It is used by recovery to resolve
lock states if a node got fenced.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch removes code parts which was declared deprecated by
commit 6b0afc0cc3 ("fs: dlm: don't use deprecated timeout features by
default"). This contains the following dlm functionality:
- start a cancel of a dlm request did not complete after certain timeout:
The current way how dlm cancellation works and interfering with other
dlm requests triggered by the user can end in an overlapping and
returning in -EBUSY. The most user don't handle this case and are
unaware that DLM can return such errno in such situation. Due the
timeout the user are mostly unaware when this happens.
- start a netlink warning messages for user space if dlm requests did
not complete after certain timeout:
This feature was never being built in the only known dlm user space side.
As we are to remove the timeout cancellation feature we can directly
remove this feature as well.
There might be the possibility to bring the timeout cancellation feature
back. However the current way of handling the -EBUSY case which is only
a software limitation and not a hardware limitation should be changed.
We minimize the current code base in DLM cancellation feature to not have
to deal with those existing features while solving the DLM cancellation
feature in general.
UAPI define DLM_LSFL_TIMEWARN is commented as deprecated and reserved
value. We should avoid at first to give it a new meaning but let
possible users still compile by keeping this define. In far future we
can give this flag a new meaning. The same for the DLM_LKF_TIMEOUT lock
request flag.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch removes the ls_remove_wait waitqueue handling. The current
handling tries to wait before a lookup is send out for a identically
resource name which is going to be removed. Hereby the remove message
should be send out before the new lookup message. The reason is that
after a lookup request and response will actually use the specific
remote rsb. A followed remove message would delete the rsb on the remote
side but it's still being used.
To reach a similar behaviour we simple send the remove message out while
the rsb lookup lock is held and the rsb is removed from the toss list.
Other find_rsb() calls would never have the change to get a rsb back to
live while a remove message will be send out (without holding the lock).
This behaviour requires a non-sleepable context which should be provided
now and might be the reason why it was not implemented so in the first
place.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch allows to give the use control about the allocation context
based on a per message basis. Currently all messages forced to be
created under GFP_NOFS context.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch will introducde a queue implementation for callbacks by using
the Linux lists. The current callback queue handling is implemented by a
static limit of 6 entries, see DLM_CALLBACKS_SIZE. The sequence number
inside the callback structure was used to see if the entries inside the
static entry is valid or not. We don't need any sequence numbers anymore
with a dynamic datastructure with grows and shrinks during runtime to
offer such functionality.
We assume that every callback will be delivered to the DLM user if once
queued. Therefore the callback flag DLM_CB_SKIP was dropped and the
check for skipping bast was moved before worker handling and not skip
while the callback worker executes. This will reduce unnecessary queues
of the callback worker.
All last callback saves are pointers now and don't need to copied over.
There is a reference counter for callback structures which will care
about to free the callback structures at the right time if they are not
referenced anymore.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
There is no need to use a mutex in those hot path sections. We change it
to spin lock to serve callbacks more faster by not allowing schedule.
The locked sections will not be locked for a long time.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch adds tracepoints for send and recv cases of dlm messages and
dlm rcom messages. In case of send and dlm message we add the dlm rsb
resource name this dlm messages belongs to. This has the advantage to
follow dlm messages on a per lock basis. In case of recv message the
resource name can be extracted by follow the send message sequence
number.
The dlm message DLM_MSG_PURGE doesn't belong to a lock request and will
not set the resource name in a dlm_message trace. The same for all rcom
messages.
There is additional handling required for this debugging functionality
which is tried to be small as possible. Also the midcomms layer gets
aware of lock resource names, for now this is required to make a
connection between sequence number and lock resource names. It is for
debugging purpose only.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch removes the send repeat remove handling. This handling is
there to repeatingly DLM_MSG_REMOVE messages in cases the dlm stack
thinks it was not received at the first time. In cases of message drops
this functionality is necessary, but since the DLM midcomms layer
guarantees there are no messages drops between cluster nodes this
feature became not strict necessary anymore. Due message
delays/processing it could be that two send_repeat_remove() are sent out
while the other should be still on it's way. We remove the repeat remove
handling because we are sure that the message cannot be dropped due
communication errors.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
