Symptom: messages similar to
"FIXME asender in bm_change_bits_to,
bitmap locked for 'write from resync_finished' by worker"
If a resync or verify is finished (or aborted), a full bitmap writeout
is triggered. If we have ongoing local IO, the bitmap may still change
during that writeout, pending and not yet processed acks may cause bits
to be cleared, while new writes may cause bits to be to be set.
To fix this, introduce the drbd_bm_write_copy_pages() variant.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
DRBD can freeze IO, due to fencing policy (fencing resource-and-stonith),
or because we lost access to data (on-no-data-accessible suspend-io).
Resuming from there (re-connect, or re-attach, or explicit admin
intervention) should "just work".
Unfortunately, if the re-attach/re-connect did not happen within
the timeout, since the commit
drbd: Implemented real timeout checking for request processing time
if so configured, the request_timer_fn() would timeout and
detach/disconnect virtually immediately.
This change tracks the most recent attach and connect, and does not
timeout within <configured timeout interval> after attach/connect.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Changes to the role and disk state should be delayed or rejected
while we establish a connection.
This is necessary, since the peer will base its resync decision
on the UUIDs and the state we sent in the drbd_connect() function.
The most prominent example for this race is becoming primary after
sending state and UUIDs and before the state changes to C_WF_CONNECTION.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
We used to have a barrier implementation where barrier_nr 0 was
reserved. That is long gone. Just use the full sequence space.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
We assumed only bios with bi_idx == 0 would end up
in drbd_make_request().
That is wrong.
At least device mapper, in __clone_and_map(), may submit
clones only covering a partial bio, but sharing
the original bvec, by adjusting bi_idx and relevant
other bio members of the clone.
We used __bio_for_each_segment() in various places,
even though that is documented as
* drivers should not use the __ version unless they _really_ want to
* run through the entire bio and not just pending pieces
Impact: we would send the full bio bvec, even for the clone
with bi_idx > 0, which will cause data corruption on the
peer (because we submit wrong data at the clone offset),
and will cause a DRBD protocol error, disconnect/reconnect
and resync (thus fixing the corruption),
because the next package header would be expected right
in the middle of the sent data, causing DRBD magic mismatch.
Fix: drop the assert, and use bio_for_each_segment()
instead of the __ version.
Conflicts:
drbd/drbd_tracing.c
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
If the backing device is already frozen during attach, we failed
to recognize that. The current disk-timeout code works on top
of the drbd_request objects. During attach we do not allow IO
and therefore never generate a drbd_request object but block
before that in drbd_make_request().
This patch adds the timeout to all drbd_md_sync_page_io().
Before this patch we used to go from D_ATTACHING directly
to D_DISKLESS if IO failed during attach. We can no longer
do this since we have to stay in D_FAILED until all IO
ops issued to the backing device returned.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
I.e. in C_WF_REPORT_PARAMS or in C_WF_CONNECTION.
Sending may already work in these cstates, but the peer still expects
the HandShake / ConnectionFeatures packet.
Actually triggered by the Testuite on kugel.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
If the asender thread, or request_timer_fn(), or some other part of
the code, decided to drop the connection (because of timeout or other),
but the receiver just now was processing a P_STATE packet, there was a
chance that receive_state() would do a hard state change
"re-establishing" an already failed connection without additional handshake.
Log excerpt:
Remote failed to finish a request within ko-count * timeout
peer( Secondary -> Unknown ) conn( Connected -> Timeout ) pdsk( UpToDate -> DUnknown )
asender terminated
...
peer( Unknown -> Secondary ) conn( Timeout -> Connected ) pdsk( DUnknown -> UpToDate ) peer_isp( 0 -> 1 )
...
Connection closed
peer( Secondary -> Unknown ) conn( Connected -> Unconnected ) pdsk( UpToDate -> DUnknown ) peer_isp( 1 -> 0 )
receiver terminated
Impact:
while the connection state is erroneously "Connected",
requests may be queued and even sent,
which would never be acknowledged,
and may have been missed by the cleanup.
These requests would never be completed.
The next drbd_suspend_io() will then lock up,
waiting forever for these requests to complete.
Fixed in several code paths:
Make sure the connection state is NetworkFailure or worse
before starting the cleanup in drbd_disconnect().
This should make sure the cleanup won't miss any requests.
Disallow receive_state() to "upgrade" the connection state
from an error state. This will make sure the "illegal" state
transition won't happen.
For all connection failure states,
relax the safe-guard in sanitize_state() again
to silently mask out those state changes
(e.g. Timeout -> Connected becomes Timeout -> Timeout).
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
DRBD state changes schedule after_state_ch() actions to a worker thread,
which decides on the old and new states of that change, whether to send
an informational state update packet (P_STATE) to the peer.
If it decides to drbd_send_state(), it would however always send the
_curent_ state, which, if a second state change happens before the
after_state_ch() of the first ran, may "fast-forward" the peer's view
about this node. In most cases that is harmless, but sometimes this can
confuse DRBD, for example into not actually starting a necessary resync
if you do a very tight detach/attach loop on a Connected Secondary.
Fix this by always sending the "new" state of the respective state
transition which scheduled this after_state_ch() work.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
If we get into the C_BROKEN_PIPE cstate once, the state engine set the
thi->t_state of the receiver thread to restarting. But with the while loop
in drbdd_init() a new connection gets established. After the call into
drbdd() returns immediately since the thi->t_state is not RUNNING. The
restart of drbd_init() then resets thi->t_state to RUNNING.
I.e. after entering C_BROKEN_PIPE once, the next successful established
connection gets wasted.
The two parts of the fix:
* Do not cause the thread to restart if we detect the issue
with the sockets while we are in C_WF_CONNECTION.
* Make sure that all actions that would have set us to C_BROKEN_PIPE
happen before the state change to C_WF_REPORT_PARAMS.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
...when the peer has inconsistent data. In that case we failed to
clear the susp_nod flag. When the local disk was attached again
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Now, the new edition of the clause only fires if a diskless
peer gets promoted.
This is a fixup for "drbd: Delayed creation of current-UUID".
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
When the disk-timeout is active, and it expires for a single request,
we consider the local disk as D_FAILED. Note: With this change,
I made both timeout based state transitions HARD state transitions.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
The new function drbd_md_get_buffer() aborts waiting for the buffer
in case the disk failes in the meantime.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Regression introduced with 8.3.11 commit:
drbd: Take a more conservative approach when deciding max_bio_size
Never ever tell an older drbd, that we support more than 32KiB
in a single data request (packet).
Never believe an older drbd, that is supports more than 32KiB
in a single data request (packet)
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
The reason for this change is that, with when doing
'drbdadm invalidate' on a disconnected resource caused
an "implicitly set pdsk from UpToDate to DUnknown" message,
which was missleading.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
module_param(bool) used to counter-intuitively take an int. In
fddd5201 (mid-2009) we allowed bool or int/unsigned int using a messy
trick.
It's time to remove the int/unsigned int option. For this version
it'll simply give a warning, but it'll break next kernel version.
Acked-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Found these with the help of ispell -l.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
The old (optimistic) implementation could shrink the bio size
on an primary device.
Shrinking the bio size on a primary device is bad. Since there
we might get BIOs with the old (bigger) size shortly after
we published the new size.
The new implementation is more conservative, and eventually
increases the max_bio_size on a primary device (which is valid).
It does so, when it knows the local limit AND the remote limit.
We cache the last seen max_bio_size of the peer in the meta
data, and rely on that, to make the operation of single
nodes more efficient.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
In case a write failes on the local disk, go into D_INCONSISTENT
disk state. That causes future reads of that block to be shipped
to the peer.
Read retry remote was already in place.
Actually the documentation needs to get fixed now. Since the
application is still shielded from the error. (as long as we have
only a single disk failing) The difference to detach is that
we keep the disk. And therefore might keep all the other, still
working sectors up to date.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
When we receive a barrier ack, we walk the ring list of drbd requests
in the transfer log of the respective epoch, do some housekeeping,
and free those objects.
We tried to keep epochs of mirrored and unmirrored drbd requests
separate, and assert that no local-only requests are present in a
barrier_acked epoch.
It turns out that this has quite a number of corner cases and would
add bloated code without functional benefit.
We now revert the (insufficient) commits
drbd: Fixed an issue with AHEAD -> SYNC_SOURCE transitions
drbd: Ensure that an epoch contains only requests of one kind
and instead fix the processing of barrier acks to cope with
a mix of local-only and mirrored requests.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
If we fail to send the information that we lost our disk,
we have no connection, and no disk: no access to data anymore.
That is either expected (deconfiguration), or there will be so much
noise in the logs that "Sending state failed" is not useful at all.
Drop it.
If the reason for a shorter than expected receive was a signal,
which we sent because we already decided to disconnect,
these additional log messages are confusing and useless.
This patch follows this pattern:
- dev_warn(DEV, "short read expecting header on sock: r=%d\n", r);
+ if (!signal_pending(current))
+ dev_warn(DEV, "short read expecting header on sock: r=%d\n", r);
Also make them all dev_warn for consistency.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Now that we do no longer in-place endian-swap the bitmap, we allow
selected bitmap operations (testing bits, sometimes even settting bits)
during some bulk operations.
This caused us to hit a lot of FIXME asserts similar to
FIXME asender in drbd_bm_count_bits,
bitmap locked for 'write from resync_finished' by worker
Which now is nonsense: looking at the bitmap is perfectly legal
as long as it is not being resized.
This cosmetic patch defines some flags to describe expectations in finer
detail, so the asserts in e.g. bm_change_bits_to() can be skipped if
appropriate.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
All decisions about sync, sync direction, and wether or not to
allow a connect or attach are based on our set of UUIDs to tag a
data generation.
Log changes to the UUIDs whenever they occur,
logging "new current UUID P:Q:R:S" is more useful
than "Creating new current UUID".
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
The "lazy writeout" of cleared bitmap pages happens during resync, and
should happen again once the resync finishes cleanly, or is aborted.
If resync finished cleanly, or was aborted because of peer disk
failure, we trigger the writeout from worker context in the after
state change work.
If resync was aborted because of connection failure, we should not
immediately trigger bitmap writeout, but rather postpone the
writeout to after the connection cleanup happened. We now do it
in the receiver context from drbd_disconnect().
If resync was aborted because of local disk failure, well, there
is nothing to write to anymore.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
This is a minor optimization and cleanup,
and also considerably reduces some harmless (but noisy) race with
the connection cleanup code.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
The assert in drbd_req.c:755 forces us to have only requests of
one kind in an epoch. The two kinds we distinguish here are:
local-only or mirrored.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
The test if rs_pending_cnt == 0 was too weak. Using Test for
unacked_cnt == 0 instead. Moved that into the worker.
Since unacked_cnt gets already increased when an P_RS_DATA_REQ
comes in.
Also using a timer to make Ahead -> SyncSource -> Ahead cycles
slower...
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
See also commit from 2009-08-15
"drbd_uuid_compare(): Do not full sync in case a P_SYNC_UUID packet gets lost."
We saw cases where the History UUIDs where not as expected. So the
detection of the special case did not trigger. With the sync UUID
no longer being a random number, but deducible from the previous
bitmap UUID, the detection of this special case becomes more
reliable.
The SyncUUID now is the previous bitmap UUID + 0x1000000000000.
Rule 5a:
Cs = H1p & H1p + Offset = Bp
Connection was lost before SyncUUID Packet came through.
Corrent (peer) UUIDs:
Bp = H1p
H1p = H2p
H2p = 0
Become Sync target.
Rule 7a:
Cp = H1s & H1s + Offset = Bs
Connection was lost before SyncUUID Packet came through.
Correct (own) UUIDs:
Bs = H1s
H1s = H2s
H2s = 0
Become Sync source.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Besides removed a few lines of code, this moves the inspection
of the state from before the queuing process to after the queuing.
I.e. more closely to the actual invocation of the work.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Create a new barrier when leaving the AHEAD mode.
Otherwise we trigger the assertion in req_mod(, barrier_acked)
D_ASSERT(req->rq_state & RQ_NET_SENT);
The new barrier is created by recycling the newest existing one.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
When on-no-data-accessible is set to suspend-io, also consider that
a Primary, SyncTarget node losses its connection.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Even though we now track the need for bitmap writeout per bitmap page,
there is no need to trigger the writeout while a resync is going on.
Once the resync is finished (or aborted),
we trigger bitmap writeout anyways.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
To improve the latency of IO requests during bitmap exchange,
we recently allowed writes while waiting for the bitmap, sending "set
out-of-sync" information packets for any newly dirtied bits.
We have to make sure that the new resync-uuid does not overtake
these "set oos" packets. Once the resync-uuid is received, the
sync target starts the resync process, and expects the bitmap to
only be cleared, not re-set.
If we use this protocol extension, we queue the generation and sending
of the resync-uuid on the worker, which naturally serializes with all
previously queued packets.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
If drbd used to have crypto digest algorithms configured, then is being
unconfigured (but not unloaded), it frees the algorithms, but does not
reset the config. If it then is reconfigured to use the very same
algorithm, it "forgot" to re-allocate the algorithms, thinking that the
config has not changed in that aspect.
It will then Oops on the first attempt to actually use those algorithms.
Fix this by resetting the config to defaults after cleanup.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
We must not call it directly from resync_finished,
as we may be in either receiver or worker context there.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
