tests: add RBD fast diff feature to unit tests

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

pybind: add new diff_iterate features to RBD python bindings

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: snap_remove proxy requests require fast diff feature

Since older Ceph clients that support exclusive locking won't
support snap_remove, only attempt to proxy the command if
fast diff is enabled.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: relax diff_iterate discard test

The fast diff algorithm can only flag an object as deleted if
the full object is discarded -- not just a small section.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: implement fast diff algorithm

The fast diff algorithm reads all object maps between the from/to
snapshot ids and computes when the object was changed/removed.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: integrate with new object_map_snap_add/remove cls methods

When a snapshot is added, mark all HEAD image objects that exist as clean.
When a snapshot is removed, mark all clean objects that were dirty in the
deleted snapshot as dirty.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: correct update_features test for FAST_DIFF feature

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: ensure fast diff feature relies on object map

Throw an error if the fast diff feature is enabled w/o the
object map -- or if the object map is disabled while fast
diff is still enabled.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: test new RBD fast diff feature

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: move object map snapshot handling to ObjectMap class

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: new test cases for cls_rbd

Added test cases for new object_map_snap_add and object_map_snap_remove
methods.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

cls_rbd_client: added object_map_snap_add and object_map_snap_remove methods

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

cls_rbd: added object_map_snap_add and object_map_snap_remove methods

object_map_snap_add is invoked after creating a new snapshot to flag
all in-use HEAD objects within as clean.  object_map_snap_remove is
invoked after a snapshot is removed in order to clear the clean flag
on all in-use objects.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: move object map codes to common location

These codes will need to be accessible from cls_rbd and librbd.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

rbd: add support for new fast diff feature

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: add new FAST_DIFF feature

A new feature is required to track images that are properly tracking
object deltas between snapshots.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: add unit tests for proxied snap_remove operation

Verify handling of snap_remove proxied to the exclusive lock
owner.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: add snap_remove to image watcher RPC

In order to support merging dirty object metadata from
snapshot object maps, it is now required to coordinate this
activity through the exclusive lock owner.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

rbd: add new --object-extents option to diff / export-diff

The rbd CLI can now use the object map to compute the diffs between
two snapshots.  This option is enabled with the new '--object-extents'
option.

Fixes: #7787
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: add unit test for new diff_iterate2 librbd API method

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: add diff_iterate2 to API

New diff_iterate API supports skipping parent images and reporting
the object extents for an object that has been modified. The latter
will be helpful in providing an optimized diff method that can use
the object map to quickly compute object diffs between snapshots.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #4317 from ceph/wip-rpm-with-systemd-comparison

ceph.spec.in: set _with_systemd on RHEL 7 and Fedora

Merge pull request #4327 from dachary/wip-wheezy

install-deps.sh: Debian GNU/Linux wheezy needs backports

Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #4341 from dachary/wip-test-timeout

tests: display ceph report when stuck

Reviewed-by: Danny Al-Gaaf <danny.al-gaaf@bisect.de>

Merge pull request #4334 from smithfarm/master

do_autogen: add missing '--without-lttng' option to usage()

Reviewed-by: Loic Dachary <ldachary@redhat.com>

tests: display ceph report when stuck

When the cluster is stuck (wait_for_clean times out), display ceph
report to stderr for debugging purposes.

Signed-off-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #4239 from yuyuyu101/wip-librbd-conf

Wip librbd conf

Reviewed-by: Josh Durgin <jdurgin@redhat.com>

Merge pull request #4323 from ceph/wip-11363

librbd: ImageWatcher should cancel in-flight ops on watch error

Reviewed-by: Josh Durgin <jdurgin@redhat.com>

Merge pull request #4140 from ceph/wip-rbd-rebuild-object-map

rbd: add ability to rebuild corrupt/missing object maps

Reviewed-by: Josh Durgin <jdurgin@redhat.com>

Merge pull request #4325 from wonzhq/async-resize-2

librbd/AsyncResize: avoid dup incrementing refresh seq

Reviewed-by: Josh Durgin <jdurgin@redhat.com>

Merge pull request #4336 from ceph/wip-11369

librbd: failure to update the object map should always return success

Reviewed-by: Josh Durgin <jdurgin@redhat.com>

install-deps.sh: Debian GNU/Linux wheezy needs backports

It is not enough for the backports to be available, they also need to be
explicitly allowed to take precedence whenever a package is installed
indirectly. This is causing problems with libp11-kit-dev pulled by
libcurl4-gnutls-dev.

Signed-off-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #4319 from ceph/wip-11348

librbd: update default image format to version 2

Reviewed-by: Josh Durgin <jdurgin@redhat.com>

do_autogen.sh: add missing '--without-lttng' option to usage

Signed-off-by: Nathan Cutler <ncutler@suse.cz>

librbd: notify of header update after rebuilding object map

This will allow clients to detect that the object map is no longer
invalid and to reload the object map from disk.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: rebuilding object map shouldn't update piecemeal

The object map is now rebuilt in-memory and written back to the
OSDs as a single operation.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: add test case for cls_rbd object_map_save

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

cls_rbd_client: add object_map_save helper method

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

cls_rbd: add object_map_save method

Allow the object map rebuild process to build the full object map
in memory and save it to disk with one operation.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: add ImageWatcher test for new rebuild object map request

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: added rebuild_object_map test to test_librbd

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: librados_test_stub reads should deep-copy

If a client of librados_test_stub modified a bufferlist
retrieved via a read call, the client will actually be
changing the contents of the file.  Therefore, read calls
should deep-copy the contents of the buffer::ptrs.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: allow snapshot object maps to be updated

Rebuild will need to be able to update/resize the object maps
for image snapshots.  This was previously not permitted.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: update in-memory object map before on-disk update completes

When multiple IO updates are occurring within the same object, this
will allow only a single object map update request to be sent to the
OSD.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

cls_rbd: treat zero-byte object maps as missing

Acquiring the lock on a missing object map will create an
empty object. Treat the empty object as a non-existant
object map to support rebuilding corrupt object maps while
holding the exclusive lock.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

rbd: add object map rebuild command

An invalid object map can now be rebuilt via the rbd CLI.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: add rebuild_object_map to public API

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librados_test_stub: add another overload of aio_operate

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: connect async rebuild object map to state machine

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: require callers to ObjectMap::aio_update to acquire lock

This is needed to allow an atomic compare and update operation
from the rebuild object map utility.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: added RebuildObjectMapRequest state machine

It will verify the object map is properly sized, verify
the existence of each object within the image (snapshot),
and clear the invalid object map flag once complete.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: correct basic object map errors during refresh

If the object map is corrupt on-disk or too small for the image,
correct these basic issues as soon as possible.  The object map
is still flagged as invalid, but there will be less required repair
work if future IO is able to properly update the object map.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: use generic helper for issuing async requests

resize, flatten, and rebuild object map now use the same
bootstrap code for sending the request to the remote lock owner
or executing the request locally.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: connect ImageWatch rebuild request to async rebuild method

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: add preliminary methods for rebuilding object map

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: add hooks for rebuild object map to ImageWatcher

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: add new notify payload for rebuild object map

The request will be sent to the client which owns the exclusive
lock to rebuild the object map for the current image HEAD.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: failure to update the object map should always return success

If an object map update fails, the object map will be flagged as
invalid.  However, if a subsequent update failure occurs, the error
code will propagate back to the caller.

Fixes: #11369
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: librados_test_stub reads should deep-copy

If a client of librados_test_stub modified a bufferlist
retrieved via a read call, the client will actually be
changing the contents of the file.  Therefore, read calls
should deep-copy the contents of the buffer::ptrs.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #4330 from dachary/wip-hammer-release-notes

release-notes: draft v0.94.1 release notes

librbd: ImageWatcher should cancel in-flight ops on watch error

Upon an watch error notification from librados, pending requests
should be canceled and in-flight IO should be flushed prior to
unlocking the image.

Fixes: #11363
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #4333 from dachary/wip-fedora

tests: add Dockerfile for fedora 21

Reviewed-by: Kefu Chai <kchai@redhat.com>

tests: add Dockerfile for fedora 21

Signed-off-by: Loic Dachary <ldachary@redhat.com>

release-notes: draft v0.94.1 release notes

Signed-off-by: Loic Dachary <ldachary@redhat.com>

librbd: Fix image_watcher test

Signed-off-by: Haomai Wang <haomaiwang@gmail.com>

Librbd: Don't affect global md_config_t

Signed-off-by: Haomai Wang <haomaiwang@gmail.com>

Librbd: Don't apply to global configserver

Signed-off-by: Haomai Wang <haomaiwang@gmail.com>

librbd/AsyncResize: avoid dup incrementing refresh seq

The refresh_seq is incremented in notify_change when calling
notify_async_complete after the locker owner completes the resize
request.

Signed-off-by: Zhiqiang Wang <zhiqiang.wang@intel.com>

Merge pull request #4306 from dachary/wip-ec-troubleshooting

doc: an erasure code crush ruleset can be dynamically modified

Reviewed-by: Greg Farnum <gfarnum@redhat.com>

Merge pull request #4320 from liewegas/wip-crush-straw2-div0

crush/mapper: fix divide-by-0 in straw2

Reviewed-by: Yann Dupont <yann@objoo.org>
Reviewed-by: Loic Dachary <ldachary@redhat.com>

crush/mapper: fix divide-by-0 in straw2

If the item weight is 0 we don't want to divide; instead draw a minimal
value.

Fixes: #11357
Reported-by: Yann Dupont <yd@objoo.org>
Tested-by: Yann Dupont <yd@objoo.org>
Signed-off-by: Sage Weil <sage@redhat.com>

qa/workunits/rbd/copy.sh: removed deprecated --new-format option

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

tests: ensure old-format RBD tests still work

Override the RBD default image format back to version 1
to ensure tests properly cover the old format.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

rbd: deprecate --new-format command-line option

Image formats can be explicitly chosen via the --image-format X
command-line option.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

common/config_opts: updated rbd_default_format

New images will now default to format 2 RBD images.

Fixes: #11348
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librados_test_stub: added rados_ioctx_cct implementation

Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #4174 from trociny/wip-10976.master

osd: fix PG::all_unfound_are_queried_or_lost for non-existent osds

Reviewed-by: Kefu Chai <tchaikov@gmail.com
Reviewed-by: Samuel Just <sjust@redhat.com>

Merge pull request #4111 from wonzhq/hitsets-check

osd/ReplicatedPG: correct the checking if the promoting object is in other hit sets

Reviewed-by: Kefu Chai <tchaikov@gmail.com>

Merge pull request #4136 from XinzeChi/wip-pg-delete-sleep

osd: return fast if PG::deleting is true in snap_trimmer and PG::scrub

Reviewed-by: Sage Weil <sage@redhat.com>

Merge pull request #4178 from XinzeChi/wip-tracker-message

osd: more useful message to find out potential unhealth osd

Reviewed-by: Samuel Just <sjust@redhat.com>
Reviewed-by: Sage Weil <sage@redhat.com>

Merge pull request #4155 from wonzhq/prdop

osd/ReplicatedPG: don't check order in finish_proxy_read

Reviewed-by: Kefu Chai <tchaikov@gmail.com>
Reviewed-by: Sage Weil <sage@redhat.com>

Merge pull request #4139 from Ved-vampir/oscnt

Perf counters descriptions

Reviewed-by: Greg Farnum <gfarnum@redhat.com>
Reviewed-by: Samuel Just <sjust@redhat.com>

Merge pull request #3928 from dmitryya/rados_bench_with_iops_metric

rados bench: Add IOPS metric to `rados bench'

Reviewed-by: Loic Dachary <ldachary@redhat.com>

Merge pull request #4253 from majianpeng/scrub-fix2

Scrub bug fix

Reviewed-by: Samuel Just <sjust@redhat.com>

ceph.spec.in: set _with_systemd on RHEL 7 and Fedora

Commit 71a5090bca049a43e30a7f0cf99141950ef9c5dd added a "_with_systemd"
conditional to the RPMs, but I erred with the version comparison
operator, so this only applied to RHEL 8+, not RHEL 7+.

Adjust the conditional so that it will really apply to RHEL 7+. While
we're here, add Fedora as well.

Signed-off-by: Ken Dreyer <kdreyer@redhat.com>
Reported-by: Boris Ranto <branto@redhat.com>

Merge pull request #4224 from rzarzynski/wip-10682_bucketlisting

rgw: add support for end_marker parameter for GET on Swift container.

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>

Merge pull request #4305 from Ved-vampir/rgw

rgw: add perf counters descriptions

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>

Merge pull request #4240 from 9seconds/post_check_quotas

rgw: Quota not respected in POST object

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>

Merge pull request #4268 from guangyy/wip-11322

rgw : Issue AIO for next chunk first before flush the (cached) data.

Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>

Merge pull request #4315 from dachary/wip-workflow

doc: development workflows

Reviewed-by: Alistair Israel <aisrael@gmail.com>

doc: when and why publish a point release

Explain who decides when to cut a point release and the tradeoffs
involved.

Signed-off-by: Loic Dachary <ldachary@redhat.com>

doc: add Hammer to the release timeline

Signed-off-by: Loic Dachary <ldachary@redhat.com>

Librbd: aware_metadata_confs->apply_metadata_confs

Signed-off-by: Haomai Wang <haomaiwang@gmail.com>

Librbd: Use boost:assign to init aware_confs

Signed-off-by: Haomai Wang <haomaiwang@gmail.com>

Merge pull request #4233 from ceph/wip-librbd-aio-request

librbd: simplify AioRequest constructor parameters

Reviewed-by: Josh Durgin <jdurgin@redhat.com>

Merge remote-tracking branch 'origin/wip-2862'

Reviewed-by: Josh Durgin <jdurgin@redhat.com>
Conflicts:
src/common/ceph_argparse.cc
src/rbd.cc

Merge pull request #4307 from dzafman/wip-dz-cmd-fix

osd: Fix wrong usage for "ceph tell osd.* debug dump_missing"

Reviewed-by: Dan Mick <dmick@redhat.com>

osd: Fix wrong usage for "ceph tell osd.* debug dump_missing"

Signed-off-by: David Zafman <dzafman@redhat.com>

Merge pull request #4304 from Ved-vampir/mds

mds: add perf counters descriptions

Reviewed-by: John Spray <john.spray@redhat.com>

rgw: add perf counters descriptions

Sign-off-by: Alyona Kiseleva <akiselyova@mirantis.com>