If we detach due to local read-error (which sets a bit in the bitmap),
stay Primary, and then re-attach (which re-reads the bitmap from disk),
we potentially lost the "out-of-sync" (or, "bad block") information in
the bitmap.
Always (try to) write out the changed bitmap pages before going diskless.
That way, we don't lose the bit for the bad block,
the next resync will fetch it from the peer, and rewrite
it locally, which may result in block reallocation in some
lower layer (or the hardware), and thereby "heal" the bad blocks.
If the bitmap writeout errors out as well, we will (again: try to)
mark the "we need a full sync" bit in our super block,
if it was a READ error; writes are covered by the activity log already.
If that superblock does not make it to disk either, we are sorry.
Maybe we just lost an entire disk or controller (or iSCSI connection),
and there actually are no bad blocks at all, so we don't need to
re-fetch from the peer, there is no "auto-healing" necessary.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
The intention of force-detach is to be able to deal with a completely
unresponsive lower level IO stack, which does not even deliver error
completions anymore, but no completion at all.
In all other cases, we must still wait for the meta data IO completion.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
This has not yet been observed, but conceivably, when using GFP_KERNEL
allocations from drbd_md_sync(), drbd_flush_after_epoch() or
receive_SyncParam(), we could trigger additional IO to our own device,
or an other device in a criss-cross setup, and end up in a local
deadlock, or potentially a distributed deadlock in a criss-cross setup
involving the peer blocked in a similar way waiting for us to make
progress.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
The former comment arguing that GFP_KERNEL was good enough was wrong: it
did not take resize into account at all, and assumed the only path
leading here was the normal attach on a still secondary device, so no
deadlock would be possible.
Both resize on a Primary, or attach on a diskless Primary,
could potentially deadlock.
drbd_bm_resize() is called while IO to the respective device is
suspended, so we must use GFP_NOIO to avoid potential deadlock.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Using list_move_tail() instead of list_del() + list_add_tail().
spatch with a semantic match is used to found this problem.
(http://coccinelle.lip6.fr/)
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
"aborting" requests, or force-detaching the disk, is intended for
completely blocked/hung local backing devices which do no longer
complete requests at all, not even do error completions. In this
situation, usually a hard-reset and failover is the only way out.
By "aborting", basically faking a local error-completion,
we allow for a more graceful swichover by cleanly migrating services.
Still the affected node has to be rebooted "soon".
By completing these requests, we allow the upper layers to re-use
the associated data pages.
If later the local backing device "recovers", and now DMAs some data
from disk into the original request pages, in the best case it will
just put random data into unused pages; but typically it will corrupt
meanwhile completely unrelated data, causing all sorts of damage.
Which means delayed successful completion,
especially for READ requests,
is a reason to panic().
We assume that a delayed *error* completion is OK,
though we still will complain noisily about it.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
is_valid_transition() might return SS_NOTHING_TO_DO.
The condition function _req_st_cond() returned SS_NOTHING_TO_DO, which
caused the wait_event to abort too early. Therefore drbd_req_state()
did not consume the next CL_ST_CHG_SUCCESS or SS_CW_FAILED_BY_PEER
causing serve disruption of the state machine logic...
Detaching from a single volue was one way to trigger this bug.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
We use the RQ_POSTPONED flag to mark a request for several reasons.
It may be a conflicting request in a dual-primary setup,
where conflict detection and resolution on the peer decided that
this request needs to be re-submitted, it needs to re-enter
drbd_make_request() to fix the data divergence caused by these
conflicting, partially overlapping, quasi-simultaneous requests.
In this case we need to mark the corresponding area as out-of-sync,
before we call drbd_al_complete_io().
We also use the RQ_POSTPONED flag to just "push back" a request,
before even processing it, if IO is suspended for some reason.
In this case, as this request was neither submitted nor sent yet,
we must not touch the bitmap.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
A postponed request might has RQ_IN_ACT_LOG already set, but
is POSTPONED before it gets something in the RQ_LOCAL_MASK
set. Up to now this caused a left-over active extent.
Fix that by only testing for the RQ_IN_ACT_LOG bit in drbd_req_destroy()
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Without this, the meta-data gets updates after 5 seconds by the
md_sync_timer. Better to do it immeditaly after a state change.
If the asender detects a network failure, it may take a bit until
the worker processes the according after-conn-state-change work item.
The worker might be blocked in sending something, i.e. it
takes until it gets into its timeout. That is 6 seconds by
default which is longer than the 5 seconds of the md_sync_timer.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
* Postponed requests should not set or clear out-of-sync marks
* When a request gets postponed we need to drop its reference
mdev->local_cnt (put_ldev()).
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
With merging the commit
'drbd: Delay/reject other state changes while establishing a connection'
the condition check for clearing the flag was wrong.
Move the bit clearing to the __drbd_set_state() function
in order to have it already cleared for the other parts of
the function. I.e. clearing the susp_fen in the after_state_ch() function.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
When _conn_requests_state() is used to change other parts of the state
than the connection, do not check for a valid connection transition.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
The previous way of doing the state change was also okay since the
state change on the susp flag gets propagated from the mdev
to the tconn.
Fortunately all this goes away in drbd-9.0
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
If the md_sync_timer triggers a second time,
while the work queued during the first time is still pending,
this could result in list_add() of an already added item,
and corrupt the work item list.
This likely only triggered because of the erroneous
batch-dequeueing of work items fixed with
drbd: dequeue single work items in wait_for_work()
Still, skip queueing if md_sync_work is already queued.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
As long as we still use drbd_queue_work_front(),
we must only dequeue the single first item during normal operation.
The comment in drbd_worker() even says so,
but bc8a5a1 drbd: remove struct drbd_tl_epoch objects (barrier works)
introduced the batch dequeueing again via list_splice_init() in
wait_for_work().
Change back to list_move() of the first item, if any.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Documentation of mutex_unlock says
we must not use it in interrupt context.
So do not call it while holding the spin_lock_irq,
but give up the spinlock temporarily.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
If the preconditions for a state change change after the wait_event() we
might hit the BUG() statement in conn_set_state().
With holding the spin_lock while evaluating the condition AND until the
actual state change we ensure the the preconditions can not change anymore.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
drbd_adm_disk_opts() does
wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
drbd_al_shrink(mdev);
If the device is very busy, this can take a very long time to succeed.
Fix this by temporarily suspending IO,
then quickly change the settings, and resume.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
We must only send P_BARRIER for epochs we actually sent P_DATA in.
If we (re-)establish a connection, we reinitialized the
send.current_epoch_nr, but forgot to reset send.current_epoch_writes.
This could result in a spurious P_BARRIER with stale epoch information,
and a disconnect/reconnect cycle once the then "unexpected"
P_BARRIER_ACK is received:
BAD! BarrierAck #28823 received, expected #28829!
Introduce re_init_if_first_write() and maybe_send_barrier() helpers,
and call them appropriately for read/write/set-out-of-sync requests.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
drbd_disconnected() is supposed to clear the resync lru cache,
by calling drbd_rs_cancel_all().
We must do so before we call drbd_flush_workqueue(), as at least the
callback w_restart_disk_io() may wait for resync progres, and would
otherwise deadlock.
drbd_finish_peer_reqs() may again populate that cache, which will
then potentially be stale after the next resync handshake and bitmap
exchange, we have to do it again after that.
A stale resync lru cache causes no harm but ugly messages like this:
BAD! sector=196608s enr=6 rs_left=-256 rs_failed=0 count=256 cstate=SyncTarget
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Disconnecting is a cluster wide state change. In case the peer node agrees
to the state transition, it sends back the fact on the meta-data connection
and closes both sockets.
In case the node node that initiated the state transfer sees the closing
action on the data-socket, before the P_STATE_CHG_REPLY packet, it was
going into one of the network failure states.
At least with the fencing option set to something else thatn "dont-care",
the unclean shutdown of the connection causes a short IO freeze or
a fence operation.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
There is at least the worker context, the receiver context, the context of
receiving netlink packts.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
We need to write the whole bitmap after we moved the meta data
due to an online resize operation.
With the support for one peta byte devices bitmap IO was optimized
to only write out touched pages. This optimization must be turned
off when writing the bitmap after an online resize.
This issue was introduced with drbd-8.3.10.
The impact of this bug is that after an online resize, the next
resync could become larger than expected.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
In various places (E.g. CONNECTION_LOST_WHILE_PENDING) the
RQ_COMPLETION_SUSP mask is passed in the clear set to mod_rq_state().
The issue was that it tried to clear the RQ_COMPLETION_SUSP bit
out of the state mask first, and eventuelly set it afterwards,
in the drbd_req_put_completion_ref() function.
Fixed that by moving the reference getting out of
drbd_req_put_completion_ref() into the mod_rq_state(), before the place
where the extra reference might be put.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
If for some reason (typically "split-brained" cluster manager)
drbd replica data has diverged, we can chose a victim,
and reconnect using "--discard-my-data", causing the victim
to become sync-target, fetching all changed blocks from the peer.
If we are Primary, we are potentially in use, and we refuse to
"roll back" changes to the data below the page cache and other users.
Rename the error symbol for this to ERR_DISCARD_IMPOSSIBLE.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
We don't discard anything here, really.
We resolve conflicting, concurrent writes to overlapping data blocks.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
To avoid confusion with REQ_DISCARD aka TRIM, rename our
"discard concurrent write acks" from P_DISCARD_WRITE to P_SUPERSEDED.
At the same time, rename the drbd request event DISCARD_WRITE
to CONFLICT_RESOLVED. It already triggers both successful completion
or restart of the request, depending on our RQ_POSTPONED flag.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Don't drop a request from the transfer log just because it was NEG_ACKED.
We need it around to be able to verify P_BARRIER_ACKs against the
transver log.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Almost all code paths calling start_new_tl_epoch() guarded it with
if (... current_tle_writes > 0 ... ).
Just move that inside start_new_tl_epoch().
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Requests of an acked epoch are stored on the barrier_acked_requests list. In
case the private bio of such a request completes while IO on the drbd device
is suspended [req_mod(completed_ok)] then the request stays there.
When thawing IO because the fence_peer handler returned, then we use
tl_clear() to apply the connection_lost_while_pending event to all requests
on the transfer-log and the barrier_acked_requests list.
Up to now the connection_lost_while_pending event was not applied
on requests on the barrier_acked_requests list. Fixed that.
I.e. now the connection_lost_while_pending and resend events are
applied to requests on the barrier_acked_requests list. For that
it is necessary that the resend event finishes (local only)
READS correctly.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
The DISCARD_CONCURRENT flag should be set on one node and cleared on the
other node.
As the code was before it was theoretical possible that a node accepts the
meta socket, but has to close it later on, and keeps the DISCARD_CONCURRENT
flag.
Correct this by moving the clear_bit(DISCARD_CONCURRENT) where the packet
gets sent.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
DRBD has a concept of request epochs or reorder-domains,
which are separated on the wire by P_BARRIER packets.
Older DRBD is not able to handle zero-sized requests at all,
so we need to map empty flushes to these drbd barriers.
These are the equivalent of empty flushes, and
by default trigger flushes on the receiving side anyways
(unless not supported or explicitly disabled),
so there is no need to handle this differently in newer drbd either.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Since the listening socket is open all the time, it was possible to
get into stable "initial packet S crossed" loops.
* when both sides realize in the drbd_socket_okay() call at the end
of the loop that the other side closed the main socket you had
the chance to get into a stable loop with repeated "packet S crossed"
messages.
* when both sides do not realize with the drbd_socket_okay() call at the end
of the loop that the other side closed the main socket you had
the chance to get into a stable loop with alternating "packet S crossed"
"packet M crossed" messages.
In order to break out these stable loops randomize the behaviour if
such a crossing of P_INITIAL_DATA or P_INITIAL_META packets is detected.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Since now our listening socket is open all the time we will get
connection tries of the peer always in. No need to try it three
times.
This is valid when connecting to older peers as well, it simply
increases the probability that the new version DRBD will accept
a connection instead that it will establish one.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Since the drbd_socket_okay() function itself tests if the the
socket is NULL, the explicit test "if (sock.socket && &msock.socket)"
was redundent.
Apart from that the address opperator ('&') before msock.socket rendered
the test pointless.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
In 8.4, we may have bios spanning two activity log extents.
Fixup drbd_al_begin_io() and drbd_al_complete_io() to deal with zero sized bios.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
We now can schedule only a specific range of sectors for online verify,
or interrupt a running verify without interrupting the connection.
Had to bump the protocol version differently, we are now 101.
Added verify_can_do_stop_sector() { protocol >= 97 && protocol != 100; }
Also, the return value convention for worker callbacks has changed,
we returned "true/false" for "keep the connection up" in 8.3,
we return 0 for success and <= for failure in 8.4.
Affected: receive_state()
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
If you do back to back wait-sync/invalidate on a Primary in a tight loop,
during application IO load, you could trigger a race:
kernel: block drbd6: FIXME going to queue 'set_n_write from StartingSync'
but 'write from resync_finished' still pending?
Fix this by changing the order of the drbd_queue_work() and
the wake_up() in dec_ap_pending(), and adding the additional
drbd_flush_workqueue() before requesting the full sync.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
In case we want to hard-reset from the local-io-error handler,
we need to call it before notifying the peer or aborting local IO.
Otherwise the peer will advance its data generation UUIDs even
if secondary.
This way, local io error looks like a "regular" node crash,
which reduces the number of different failure cases.
This may be useful in a bigger picture where crashed or otherwise
"misbehaving" nodes are automatically re-deployed.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Fix asserts like
block drbd0: in got_BlockAck:4634: rs_pending_cnt = -35 < 0 !
We reset the resync lru cache and related information (rs_pending_cnt),
once we successfully finished a resync or online verify, or if the
replication connection is lost.
We also need to reset it if a resync or online verify is aborted
because a lower level disk failed.
In that case the replication link is still established,
and we may still have packets queued in the network buffers
which want to touch rs_pending_cnt.
We do not have any synchronization mechanism to know for sure when all
such pending resync related packets have been drained.
To avoid this counter to go negative (and violate the ASSERT that it
will always be >= 0), just do not reset it when we lose a disk.
It is good enough to make sure it is re-initialized before the next
resync can start: reset it when we re-attach a disk.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Philipp Reisner