With the multi-queue support we could fail at setting up
some of the rings and fail the connection. That meant that
all resources tied to rings[0..n-1] (where n is the ring
that failed to be setup). Eventually the frontend will switch
to the states and we will call xen_blkif_disconnect.
However we do not want to be at the mercy of the frontend
deciding when to change states. This allows us to do the
cleanup right away and freeing resources.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Make pool of persistent grants and free pages per-queue/ring instead of
per-device to get better scalability.
Test was done based on null_blk driver:
dom0: v4.2-rc8 16vcpus 10GB "modprobe null_blk"
domu: v4.2-rc8 16vcpus 10GB
[test]
rw=read
direct=1
ioengine=libaio
bs=4k
time_based
runtime=30
filename=/dev/xvdb
numjobs=16
iodepth=64
iodepth_batch=64
iodepth_batch_complete=64
group_reporting
Results:
iops1: After patch "xen/blkfront: make persistent grants per-queue".
iops2: After this patch.
Queues: 1 4 8 16
Iops orig(k): 810 1064 780 700
Iops1(k): 810 1230(~20%) 1024(~20%) 850(~20%)
Iops2(k): 810 1410(~35%) 1354(~75%) 1440(~100%)
With 4 queues after this commit we can get ~75% increase in IOPS, and
performance won't drop if increasing queue numbers.
Please find the respective chart in this link:
https://www.dropbox.com/s/agrcy2pbzbsvmwv/iops.png?dl=0
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Backend advertises "multi-queue-max-queues" to front, also get the negotiated
number from "multi-queue-num-queues" written by blkfront.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Preparatory patch for multiple hardware queues (rings). The number of
rings is unconditionally set to 1, larger number will be enabled in
"xen/blkback: get the number of hardware queues/rings from blkfront".
Signed-off-by: Arianna Avanzini <avanzini.arianna@gmail.com>
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
---
v2: Align variables in the structures.
Split per ring information to an new structure "xen_blkif_ring", so that one vbd
device can be associated with one or more rings/hardware queues.
Introduce 'pers_gnts_lock' to protect the pool of persistent grants since we
may have multi backend threads.
This patch is a preparation for supporting multi hardware queues/rings.
Signed-off-by: Arianna Avanzini <avanzini.arianna@gmail.com>
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
---
v2: Align the variables in the structure.
According to this piece code:
"
pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
"
if xen_blkif_max_ring_order is bigger that XENBUS_MAX_RING_GRANT_ORDER,
need to set xen_blkif_max_ring_order using XENBUS_MAX_RING_GRANT_ORDER,
but not 0.
Signed-off-by: Peng Fan <van.freenix@gmail.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: "Roger Pau Monné" <roger.pau@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Make persistent grants per-queue/ring instead of per-device, so that we can
drop the 'dev_lock' and get better scalability.
Test was done based on null_blk driver:
dom0: v4.2-rc8 16vcpus 10GB "modprobe null_blk"
domu: v4.2-rc8 16vcpus 10GB
[test]
rw=read
direct=1
ioengine=libaio
bs=4k
time_based
runtime=30
filename=/dev/xvdb
numjobs=16
iodepth=64
iodepth_batch=64
iodepth_batch_complete=64
group_reporting
Queues: 1 4 8 16
Iops orig(k): 810 1064 780 700
Iops patched(k): 810 1230(~20%) 1024(~20%) 850(~20%)
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
We do the same exact operations a bit earlier in the
function.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
The max number of hardware queues for xen/blkfront is set by parameter
'max_queues'(default 4), while it is also capped by the max value that the
xen/blkback exposes through XenStore key 'multi-queue-max-queues'.
The negotiated number is the smaller one and would be written back to xenstore
as "multi-queue-num-queues", blkback needs to read this negotiated number.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
After patch "xen/blkfront: separate per ring information out of device
info", per-ring data is protected by a per-device lock ('io_lock').
This is not a good way and will effect the scalability, so introduce a
per-ring lock ('ring_lock').
The old 'io_lock' is renamed to 'dev_lock' which protects the ->grants list and
->persistent_gnts_c which are shared by all rings.
Note that in 'blkfront_probe' the 'blkfront_info' is setup via kzalloc
so setting ->persistent_gnts_c to zero is not needed.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Preparatory patch for multiple hardware queues (rings). The number of
rings is unconditionally set to 1, larger number will be enabled in
patch "xen/blkfront: negotiate number of queues/rings to be used with backend"
so as to make review easier.
Note that blkfront_gather_backend_features does not call
blkfront_setup_indirect anymore (as that needs to be done per ring).
That means that in blkif_recover/blkif_connect we have to do it in a loop
(bounded by nr_rings).
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Split per ring information to a new structure "blkfront_ring_info".
A ring is the representation of a hardware queue, every vbd device can associate
with one or more rings depending on how many hardware queues/rings to be used.
This patch is a preparation for supporting real multi hardware queues/rings.
We also add a backpointer to 'struct blkfront_info' (dev_info) which
is not needed (we could use containers_of) but further patch
("xen/blkfront: pseudo support for multi hardware queues/rings")
will make allocation of 'blkfront_ring_info' dynamic.
Signed-off-by: Arianna Avanzini <avanzini.arianna@gmail.com>
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Commit 8182503df1 used monotonic time, but if the adapter is
using the seconds for logging entries, then we'll get duplicate
entries if the system is rebooted. Use real time instead.
Reported-by: Arnd Bergmann <arnd@arndb.de>
Fixes: 8182503df1 ("block: sx8.c: Replace timeval with ktime_t")
Signed-off-by: Jens Axboe <axboe@fb.com>
32-bit systems using 'struct timeval' will break in the year 2038,
in order to avoid that replace the code with more appropriate types.
This patch replaces timeval with 64 bit ktime_t which is y2038 safe.
Since st->timestamp is only interested in seconds, directly using
time64_t here. Function ktime_get_seconds is used since it uses
monotonic instead of real time and thus will not cause overflow.
Signed-off-by: Shraddha Barke <shraddha.6596@gmail.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
In case the lower level device size changed, but some other internal
details of the resize did not work out, drbd_determine_dev_size() would
try to restore the previous settings, trusting
drbd_md_set_sector_offsets() to "do the right thing", but overlooked
that this internally may set the meta data base offset based on device size.
This could end up with incomplete on-disk meta data layout change, and
ultimately lead to data corruption (if the failure was not noticed or
ignored by the operator, and other things go wrong as well).
Just remember all meta data related offsets/sizes,
and on error restore them all.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
During handshake communication, we also reconsider our device size,
using drbd_determine_dev_size(). Just in case we need to change the
offsets or layout of our on-disk metadata, we lock out application
and other meta data IO, and wait for the activity log to be "idle"
(no more referenced extents).
If this handshake happens just after a connection loss, with a fencing
policy of "resource-and-stonith", we have frozen IO.
If, additionally, the activity log was "starving" (too many incoming
random writes at that point in time), it won't become idle, ever,
because of the frozen IO, and this would be a lockup of the receiver
thread, and consquentially of DRBD.
Previous logic (re-)initialized with a special "empty" transaction
block, which required the activity log to fully drain first.
Instead, write out some standard activity log transactions.
Using lc_try_lock_for_transaction() instead of lc_try_lock() does not
care about pending activity log references, avoiding the potential
deadlock.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
To be able to "force out" an activity log transaction,
even if there are no pending updates.
This will be used to relocate the on-disk activity log,
if the on-disk offsets have to be changed,
without the need to empty the activity log first.
While at it, move the definition,
so we can drop the forward declaration of a static helper.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Avoid to prematurely resume application IO: don't set/clear a single
bit, but inc/dec an atomic counter.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Don't remember a DRBD request as ack_pending, if it is not.
In protocol A, we usually clear RQ_NET_PENDING at the same time we set
RQ_NET_SENT, so when deciding to remember it as ack_pending,
mod_rq_state needs to look at the current request state,
not at the previous state before the current modification was applied.
This should prevent advance_conn_req_ack_pending() from walking the full
transfer log just to find NULL in protocol A, which would cause serious
performance degradation with many "in-flight" requests, e.g. when
working via DRBD-proxy, or with a huge bandwidth-delay product.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
new_disk_conf could be leaked if the follow on checks fail,
so make sure to free it on error if it was not assigned yet.
Found with smatch.
Signed-off-by: Oleg Drokin <green@linuxhacker.ru>
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Disconnect should wait for pending bitmap IO.
But if that bitmap IO is not happening, because it is waiting for
pending application IO, and there is no progress, because the fencing
policy suspended application IO because of the disconnect,
then we deadlock.
The bitmap writeout in this case does not care for concurrent
application IO, so there is no point waiting for it.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
lsblk should be able to pick up stacking device driver relations
involving DRBD conveniently.
Even though upstream kernel since 2011 says
"DON'T USE THIS UNLESS YOU'RE ALREADY USING IT."
a new user has been added since (bcache),
which sets the precedences for us to use it as well.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
We cannot possibly support SECDISCARD, even if all backend devices would
support it: if our peer is currently unreachable, some instance of the
data may obviously still be recoverable.
We did not set discard_granularity at all. We don't really care (yet),
we only pass them on, so for now, set our granularity to one sector.
blkdev_stack_limits() takes care of the rest.
If we decide we cannot support discards,
not only clear the (not user visible) QUEUE_FLAG_DISCARD,
but set both (user visible) discard_granularity and max_discard_sectors
to zero, to avoid confusion with e.g. lsblk -D.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
When accessing out meta data area on disk, we double check the
plausibility of the requested sector offsets, and are very noisy about
it if they look suspicious.
During initial read of our "superblock", for "external" meta data,
this triggered because the range estimate returned by
drbd_md_last_sector() was still wrong.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Suggested by Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Apparently we now implicitly get definitions for BITS_PER_PAGE and
BITS_PER_PAGE_MASK from the pid_namespace.h
Instead of renaming our defines, I chose to define only if not yet
defined, but to double check the value if already defined.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Since kernel 3.3, we can use snprintf-style arguments
to create a workqueue.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
The effective data generation ID may be interesting for debugging
purposes of scenarios involving diskless states.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
In a multiple error scenario, we may end up with a "frozen" Primary,
that has no access to any data (no local disk, no replication link).
If we then resume-io, we try to generate a new data generation id,
which will fail if there is no longer a local disk.
Double check for available local data,
which prevents the NULL pointer deref.
If we are diskless, turn the resume-io in this situation
into the first stage of a "force down", by bumping the "effective" data
gen id, which will prevent later attach or connect to the former data
set without first being demoted (deconfigured).
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
The intention is to reduce CPU utilization. Recent measurements
unveiled that the current performance bottleneck is CPU utilization
on the receiving node. The asender thread became CPU limited.
One of the main points is to eliminate the idr_for_each_entry() loop
from the sending acks code path.
One exception in that is sending back ping_acks. These stay
in the ack-receiver thread. Otherwise the logic becomes too
complicated for no added value.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
This prepares the next patch where the sending on the meta (or
control) socket is moved to a dedicated workqueue.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
A D_FAILED disk transitions as quickly as possible to
D_DISKLESS. But in the "unresponsive local disk" case,
there remains a time window where a administrative detach command could
find the disk already failed, but some internal meta data IO against the
unresponsive local disk still pending.
In that case, drbd_md_get_buffer() will return NULL.
Don't unconditionally call drbd_md_put_buffer(), or it will cause
refcount imbalance, and prevent any further re-attach on this volume
(until it is deleted and re-created).
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
The recent (not yet released) backport of the extended state broadcasts
to support the "events2" subcommand of drbdsetup had some glitches.
remember_old_state() would first count all connections with a
net_conf != NULL, then allocate a suitable array, then populate that
array with all connections found to have net_conf != NULL.
This races with the state change to C_STANDALONE,
and the NULL assignment there.
remember_new_state() then iterates over said connection array,
assuming that it would be fully populated.
But rcu_lock() just makes sure the thing some pointer points to,
if any, won't go away. It does not make the pointer itself immutable.
In fact there is no need to "filter" connections based on whether or not
they have a currently valid configuration. Just record them always, if
they don't have a config, that's fine, there will be no change then.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Don't blame the peer for being unresponsive,
if we did not even ask the question yet.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
The only way to make DRBD intentionally call panic is to
set a disk timeout, have that trigger, "abort" some request and complete
to upper layers, then have the backend IO subsystem later complete these
requests successfully regardless.
As the attached IO pages have been recycled for other purposes
meanwhile, this will cause unexpected random memory changes.
To prevent corruption, we rather panic in that case.
Make it obvious from stack traces that this was the case by introducing
drbd_panic_after_delayed_completion_of_aborted_request().
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Even though we really want to get the state information about our bad
disk to the peer as soon as possible, it is useful to first call the
local-io-error handler.
People may chose to hard-reset the box from there.
If that looks and behaves exactly like a "regular node crash", without
bumping the data generation UUIDs on the peer in between, it makes it
easier to deal with.
If you intend to return from the local-io-error handler, then better
return as quickly as possible to avoid triggering other timeouts.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
If for some reason the primary lost its disk *and* the replication link
before it is able to communicate the disk loss, probably blocked IO,
then later is able to re-establish the connection, the peer needs to
bump its UUIDs just like it does when peer only loses the disk
and is able to communicate this in time.
Otherwise, a later re-attach of the disk on the primary may start a
resync in the "wrong" direction.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
When detaching, we make sure no application IO is in-flight
by internally suspending IO, then trigger the state change,
wait for the result, and finally internally resume IO again.
Once we triggered the stat change to "Failed",
we expect it to change from Failed to Diskless.
(To avoid races, we actually wait for it to leave "Failed").
On an unresponsive local IO backend, this may not happen, ever.
Don't have a "hung" detach block IO "forever", but resume IO
before waiting for the state change to Diskless.
We may well be able to continue IO to and from a healthy peer.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
(You should not use disk-timeout anyways,
see the man page for why...)
We add incoming requests to the tail of some ring list.
On local completion, requests are removed from that list.
The timer looks only at the head of that ring list,
so is supposed to only see the oldest request.
All protected by a spinlock.
The request object is created with timestamps zeroed out.
The timestamp was only filled in just before the actual submit.
But to actually submit the request, we need to give up the spinlock.
If you are unlucky, there is no older still pending request, the timer
looks at a new request with timestamp still zero (before it even was
submitted), and 0 + timeout is most likely older than "now".
Better assign the timestamp right when we put the
request object on said ring list.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
GFP_NOWAIT has a value of 0. I.e. functionality not changed.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
The lc_destroy() function tests whether its argument is NULL and then
returns immediately. Thus the test around the call is not needed.
This issue was detected by using the Coccinelle software.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Roland Kammerer <roland.kammerer@linbit.com>
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
The status command originates the drbd9 code base. While for now we
keep the status information in /proc/drbd available, this commit
allows the user base to gracefully migrate their monitoring
infrastructure to the new status reporting interface.
In drbd9 no status information is exposed through /proc/drbd.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
The events2 command originates from drbd-9 development. It features
more information but requires a incompatible change in output
format.
Therefore the previous events command continues to exist, the new
improved events2 command becomes available now.
This prepares the user-base for a later switch to the complete
drbd9 code base.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Instead of using a rwlock for synchronizing state changes across
resources, take the request locks of all resources for global state
changes. Use resources_mutex to serialize global state changes.
This means that taking the request lock of a resource is now enough to
prevent changes of that resource. (Previously, a read lock on the
global state lock was needed as well.)
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Also change the enum values to all-capital letters.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
There is no need to have these two as inline functions. In addition,
drbd_should_send_out_of_sync() is only used in a single place, anyway.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
