both a < b and b < a evaluate to true. This violates STL strict weak
ordering requirements which is a problem because GroupSnapshotNamespace
is used as a key in std::map (ictx->snap_ids at least), etc.
ee8887f4c0ff4f91117f31b621b95c8d08019130 was intended for adding
mpath devices support in ceph-volume but it has missed the lvm batch scenario.
This also fixes the zapping of mpath devices prepared with `ceph-volume raw`
Ilya Dryomov [Tue, 8 Feb 2022 09:11:49 +0000 (10:11 +0100)]
rbd: mark optional positional arguments as such in help output
Currently at least five commands have optional positional arguments.
Overloading po::value<std::string>()->default_value("") for this
is a bit sneaky but nothing better fits into the existing Shell.cc
framework.
Note that strictly speaking "[<interval>] [<start-time>]" should be
"[<interval> [<start-time>]]" but we aren't doing that here because
"ceph" command doesn't do it either.
Ilya Dryomov [Mon, 31 Jan 2022 13:08:26 +0000 (14:08 +0100)]
qa/suites/krbd: add legacy+rxbounce and crc+rxbounce coverage
For basic, rbd and rbd-nomount subsuites, replace legacy and crc
facets with "legacy or legacy+rxbounce" and "crc or crc+rxbounce"
facets (chosen at random).
For fsx, singleton and thrash subsuites, add legacy+rxbounce and
crc+rxbounce facets and drop prefer-crc facet. The expected behaviour
of the latter depends on cluster configuration and should be tested
separately.
where the numbers of scrubbed object, clones, dirty and omap are always
less than the total number of corresponding numbers, if the PG contains
object(s) whose hash happens to be 0xffffffff.
in this change, if the calculated hash of the upper bound is greater
than the maximum possible number represented by uint32_t, in addition to
setting the hash of the upper bound hobj to 0xffffffff, we also set the
nspace of hobj of the upper bound to "\xff", so that the upper bound
is greater than an hobj whose hash happens to be 0xfffffff. please note,
the nspace of "\xff" is not an ascii string, so it's not likely to be
less than a real-world nspace of an hobj.
with this new *greater* upper bound, we are able to include the previous
missing hobj when listing the objects in a PG. so the scrub won't be
annoyed when the number of objects does not match.
Adam Kupczyk [Tue, 2 Nov 2021 15:57:32 +0000 (16:57 +0100)]
os/bluestore/bluefs: Fix data corruption in truncate()
It is possible to create condition in which a BlueFS contains file that is corrupted.
It can happen when BlueFS replay log is on device A and we just wrote to device B and truncated file.
Scenario:
1) write to file h1 on SLOW device
2) flush h1 (initiate transfer, but no fdatasync yet)
3) truncate h1
4) write to file h2 on DB
5) fsync h2 (forces replay log to be written, after fdatasync to DB)
6) poweroff
Fixes: https://tracker.ceph.com/issues/53129 Signed-off-by: Adam Kupczyk <akupczyk@redhat.com>
(cherry picked from commit 49b7b44b3b5c94ee401562e603999e2b3bd8f9a2)
Ilya Dryomov [Fri, 21 Jan 2022 12:41:46 +0000 (13:41 +0100)]
rbd-mirror: fix races in snapshot-based mirroring deletion propagation
When remote image is deleted, rbd-mirror can encounter three cases:
1) no remote image id
2) no remote mirror metadata
3) MIRROR_IMAGE_STATE_DISABLING in remote mirror metadata
Commit d4c66ac5c615 ("rbd-mirror: fix issue with snapshot-based
mirroring deletion propagation") fixed case 1. Cases 2 and 3 remained
broken because for both of them finalize_snapshot_state_builder() would
populate not only remote_mirror_peer_uuid but also remote_image_id,
thus disabling ENOLINK logic in handle_prepare_remote_image() and
handle_bootstrap(). Commit ff60aec2d9ef ("rbd-mirror: fix bootstrap
sequence while the image is removed") touched on case 3, but it made
a difference only for journal-based mirroring.
Stop calling finalize_snapshot_state_builder() on errors. Instead,
align with journal-based mirroring by filling remote_mirror_peer_uuid
together with remote_mirror_uuid.
Make it clear that the local image non-primariness is asserted
independent of the mode; avoid the default implementation being
overridden but still relied on by both modes.
Ilya Dryomov [Wed, 19 Jan 2022 11:54:23 +0000 (12:54 +0100)]
rbd: add missing switch arguments for recognition by get_command_spec()
Currently this
$ rbd --all children img
doesn't work, while this
$ rbd children --all img
or this
$ rbd children img --all
does. The issue is that -a/--all isn't on the list of known switch
arguments. The "rbd children" example may seem contrived but for more
complicated commands such as "rbd device map" mixing switches and
positional arguments occurs naturally:
Conflicts:
src/tools/rbd/ArgumentTypes.h [ snapshot quiesce support
not in octopus ]
src/tools/rbd/action/Device.cc [ nbd cookie support not in
octopus ]
src/tools/rbd/action/Migration.cc [ import-only migration
not supported in octopus ]
src/tools/rbd/action/Wnbd.cc [ wnbd support not in octopus ]
Matt Benjamin [Tue, 4 Jan 2022 16:22:00 +0000 (11:22 -0500)]
rgwlc: remove lc entry on bucket delete
Buckets with lifecycle policies installed have a state entry that
must also be deleted when the bucket is removed.
Fixes: https://tracker.ceph.com/issues/46728
N.b., should really be generic, not specific to the RADOS store, but
there doesn't seem to be a clean model for implementing generic side
effects in Zipper, currently.
Igor Fedotov [Tue, 2 Nov 2021 12:03:39 +0000 (15:03 +0300)]
os/bluestore: avoid premature onode release.
This was observed when onode's removal is followed by reading
and the latter causes object release before the removal is finalized.
The root cause is an improper 'pinned' state assessment in Onode::get
More detailed overview is:
At some point Onode::get() might face the case when nref == 2 and pinned = true
which means parallel incomplete put is running on the onode - ref count is
decremented but pinned state is still unmodified (and even lock hasn't been
acquired yet).
This might finally result in two puts racing over the same onode with nref == 2
which finally results in a premature onode release:
// nref =3, pinned = 1
// Thread 1 Thread 2
// o->put() o->get()
// --nref(n = 2, pinned=1)
// nref++ (n=3, pinned = 1)
// return
// ...
// o->put()
// --nref(n = 2)
// pinned = 0,
// --nref(n = 1)
// ocs->_unpin_and_rm(o) -> o->put()
// ...
// --nref(n = 0)
// release o
// o->c->get_onode_cache()
// FAULT!
//
The suggested fix is to introduce additional atomic counter tracking
running put() functions. And permit onode release when both regular
nref and put_nref are both equal to zero.
Fixes: https://tracker.ceph.com/issues/53002 Signed-off-by: Igor Fedotov <igor.fedotov@croit.io>
(cherry picked from commit 96f0efe6d5307a55bea32f7216ef9511da0c5a47)
The recent changes from PR #43536 introduced a regeression preventing from
running ceph-volume in a containerized context on Ubuntu 18.04.
Given that the path for the binary `lvs` differs between CentOS 8 and Ubuntu 18.04.
(`/usr/sbin/lvs` and `/sbin/lvs` respictively). It means that ceph-volume running
in the container on CentOS 8 sees the `lvs` binary at `/usr/sbin/lvs` and try to
run it with `nsenter` on the host which is running Ubuntu 18.04.
Josh Durgin [Fri, 7 Jan 2022 18:37:13 +0000 (13:37 -0500)]
mon/OSDMonitor: avoid null dereference if stats are not available
Not confirmed yet whether this was the issue in the bug referenced
below, however it's a necessary defensive check for the
'osd pool get-quota' command.
All other uses of get_pool_stats() already handle this case.
Mykola Golub [Fri, 14 Jan 2022 18:21:29 +0000 (18:21 +0000)]
cls/journal: skip disconnected clients when finding min_commit_position
When a new journal client is registered, all already registered
clients are checked, and a client with min position is selected
as a position for the new client. Thus we may expect that
starting from the registered position all journal entries will be
available (not trimmed) for the new client.
But when looking for a min commit position, the client_register
function did not take into account that a registered client might
be in disconnected state, and in that case the journal entries
might be trimmed for this client.
Ilya Dryomov [Fri, 7 Jan 2022 12:31:08 +0000 (13:31 +0100)]
test/librbd: make diff-iterate clone tests exercise fast-diff mode
The fast-diff feature wasn't propagated to the clone so these tests
were exercising the slow list_snaps path no matter what RBD_FEATURES
value was supplied to ceph_test_librbd.
Ilya Dryomov [Wed, 5 Jan 2022 19:24:40 +0000 (20:24 +0100)]
librbd: restore diff-iterate include_parent functionality in fast-diff mode
Commit 4429ed4f3f4c ("librbd: switch diff iterate API to use new snaps
list dispatch methods") removed the recursive execute() call. The new
list_snaps method does indeed handle parent diffs internally but it is
not used in fast-diff mode. Nothing changed there -- we still need to
load the parent object map, calculate parent object_diff_state, etc.
Ilya Dryomov [Wed, 19 Jan 2022 20:08:01 +0000 (21:08 +0100)]
librbd: diff-iterate reports incorrect offsets if whole_object=true
It turns out that in octopus both fast-diff and list-snaps (slow)
modes were broken. As long as whole_object=true, the same incorrect
offset was reported in both modes. The fast-diff mode is fixed in
in previous commit.
This is an octopus-only patch for list-snaps mode. In pacific this
issue was addressed with 4429ed4f3f4c ("librbd: switch diff iterate
API to use new snaps list dispatch methods").
projects / ceph.git / log