debian: /var/run/ceph should be owned by ceph:ceph

Fix upstart and sysvinit scripts to create /var/run/ceph properly.

Chown existing dir on upgrade.

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

upstart: setuser ceph

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

debian/ceph-common.dirs: install /var/lib/ceph

It is the ceph user's home dir and we need to set the ownership.

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

debian/ceph-common.postinst: fix adduser, addgroup

Drop the extra groups thing--it's broken anyway.

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

debian/ceph-common.postinst: fix /var/log/ceph permissions

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

debian: rename ceph-common.postinst

There was a trailing space!

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

PendingReleaseNotes: more notes about the 'ceph' user

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

global_init: ignore --set{user,group} if not root

Assume these options can only drop privileges; normal users can't setuid
root (and even if they could we wouldn't want to).

This makes ceph-disk behave when it sees the ceph user as 'root' and
invokes things with --setuser root but runs as a non-root user.

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

ceph-disk: fix get_ceph_user

getpwnam throws an exception

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

PendingReleaseNotes: some notes about upgrade and ceph user

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

systemd: use --setuser and --setgroup for all daemons

Allow all daemons drop privilege themselves, instead of letting
systemd do it.

Among other things, this means that admins can conditionally not
drop prives by setting

  setuser match path = /var/lib/ceph/$type/$cluster-$id

in their ceph.conf to ease the pain of upgrade.

Signed-off-by: Sage Weil <sage@redhat.com>
Reviewed-by: Boris Ranto <branto@redhat.com>

global: implement setuser_match_path

Allow the --setuser and --setgroup to be conditional on the
specified user/group matching the ownership of a given path.
This allows the ceph daemons to switch to user ceph for newly
deployed instances or stay as root depending on the ownership of
the data directory.

Signed-off-by: Sage Weil <sage@redhat.com>
Reviewed-by: Boris Ranto <branto@redhat.com>

ceph-disk: set owner of created files to ceph

Signed-off-by: Milan Broz <mbroz@redhat.com>

Set keys owner to ceph user if exists.

Also fix directory access rigths.

Signed-off-by: Milan Broz <mbroz@redhat.com>

Set Ceph device partitions owner to ceph user in udev.

Signed-off-by: Milan Broz <mbroz@redhat.com>

ceph-osd-prestart.sh: ensure data dir is root or ceph before start

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

init-ceph.in: Set ceph user and group when running the daemons

Signed-off-by: Boris Ranto <branto@redhat.com>

ceph.spec.in: User and group must be created in ceph-common pre-install script

The package ceph-common uses ceph user and group but they are both created in
ceph package %pre phase. We need to move the script to ceph-common %pre phase.

Signed-off-by: Boris Ranto <branto@redhat.com>

ceph.spec.in: /var/lib/ceph is owned by ceph package

Signed-off-by: Boris Ranto <branto@redhat.com>

ceph.spec.in: Fix ceph.limits.d path

Signed-off-by: Boris Ranto <branto@redhat.com>

ceph.spec.in: Fixup uid/gid setting

We need to set the variables in the %pre phase, otherwise they are not properly
evaluated. Also use /sbin/nolongin instead of /bin/false and set the default
uid/gid pair for fedora and rhel (these were already allocated). We can also
use them for older fedora releases as they are guaranteed to be free.

Signed-off-by: Boris Ranto <branto@redhat.com>

ceph.spec.in: install ceph.limits.d

limits.d expect the file to end with *.conf. Add the
installed file to package list of the ceph package.

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

ceph.spec.in: add ceph user/group

Add support of create ceph user/group.

TODO: Add correct number for SUSE.

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

set nofile ulimit in /etc/security/limits.d/ceph only

Specify the nofile ulimit in one standard place, where everyone expects it
to be.  Drop it from the ceph-osd unit file.

Leave upstart and sysvinit untouched for the time being to avoid compat
issues.

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

systemd: make ceph-osd setuid/gid to ceph:ceph

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

global: add --setuser and --setgroup options

These are done after reading config files/environment and before log files
are opened.  Allow a name or id to be specified.  In the case of --setuser,
also switch to that user's gid, unless --setgroup is also specified.

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

debian: fix /var/lib/ceph/* directory ownership

These dirs are owned by the package; make sure they are owend by the ceph
user.

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

systemd: chown ceph:ceph /var/run/ceph

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

systemd: run mon and mds as ceph:ceph

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

ceph.spec: chown and chmod /var/lib/ceph and /var/log/ceph

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

debian: chown -R ceph:ceph /var/log/ceph

The number of log files is generally bounded; safe to chown these.

Allow ceph group members to write to this dir.

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

debian: chown ceph:ceph /var/llib/ceph

Do not do it recursively--there may already be huge amounts of data
here.

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

debian: create ceph user and group

Use Debian base-passwd allocated UID/GID pair.

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

Merge pull request #5625 from liewegas/wip-memcpy

buffer: put inlined memcpy into a header; x86_64 only

Reviewed-by: Piotr Dałek <piotr.dalek@ts.fujitsu.com>
Reviewed-by: Steve Capper <steve.capper@linaro.org>
Reviewed-by: Yehuda Sadeh <yehuda@redhat.com>

Merge pull request #5585 from tchaikov/wip-4941

mon,osd: use GMT time for the object name of hitsets

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

Merge pull request #5671 from ceph/wip-msgr-fix-random

Messenger: Fix rand() generate the same sequence numbers

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

Merge pull request #5675 from ceph/wip-12791

ceph.spec.in: remove obsolete SUSE-specific code

ceph.spec.in: remove obsolete SUSE-specific code

http://tracker.ceph.com/issues/12791 Fixes: #12791

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

Merge pull request #5639 from rzarzynski/wip-12750

rgw: fix removal during iteration over Swift metadata in filter_out_temp_url()

Messenger: Fix rand() generate the same sequence numbers

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

Merge pull request #5629 from kylinstorage/wip-nocache_skip_promote-v3

Avoid cache pollution by rbd export etc v2

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

Merge pull request #5636 from liewegas/wip-12747

make EC plugin path static

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

Merge pull request #5646 from liewegas/wip-12766

osd: wait for cleanup from bench

Merge pull request #5650 from liewegas/wip-12736

uuid: use boost::random:random_device

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

include/inline_memcpy: use __builtin_memcpy instead of explicit ptr copies

The gcc version of this is smart enough to handle any alignment
issues that exist and *should* generate the fastest code that is
general and correct.

On x86_64 this performs identically for the uint64_t
microbenchmark.

Suggested-by: Steve Capper <steve.capper@linaro.org>
Signed-off-by: Sage Weil <sage@redhat.com>

include/inline_memcpy: make prototype resemble memcpy's

Suggested-by: Steve Capper <steve.capper@linaro.org>
Signed-off-by: Sage Weil <sage@redhat.com>

Merge pull request #5654 from ceph/wip-12614

config: skip lockdep for intentionally recursive md_config_t lock

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

Merge pull request #5647 from ceph/wip-12765

librbd: error closing image while set to invalid snapshot

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

Merge pull request #4677 from tchaikov/wip-fix-type-error-in-crush-rule

pybind: Avoid type error in crush rule (str, not an int)

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

Merge pull request #5659 from badone/doc_writeback_typo

doc: Fix typo in writeback throttle section

Merge pull request #5660 from badone/doc_snaps_typo

doc: Fix typo and redundant word in snaps section

Merge pull request #5648 from ceph/wip-10399

Objecter: pg_interval_t::is_new_interval needs pgid from previous pool

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

Merge pull request #5663 from ritz303/wip-doc-12771

doc: Added "Hammer" in the list of major releases.

doc: Added "Hammer" in the list of major releases.

Fixes: #12771
Signed-off-by: ritz303 <ritz_303@yahoo.com>

Merge pull request #5661 from SUSE/wip-suse-sysvinit-cleanup

ceph.spec.in: drop sysvinit-specific macros that run only on SUSE

Reviewed-by: Ken Dreyer <kdreyer@redhat.com>

ceph.spec.in: drop sysvinit-specific macros that run only on openSUSE/SLE

All sysvinit-based versions of openSUSE/SLE are EOL as far as upstream Ceph is
concerned.

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

Merge pull request #5657 from badone/master

doc: Fix typo in recovery reservation section

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

doc: Fix typo and redundant word in snaps section

Fix typo and remove identifying from "filtering identifying" in
doc/dev/osd_internals/snaps.rst

Signed-off-by: Brad Hubbard <bhubbard@redhat.com>

doc: Fix typo in writeback throttle section

Fix typo in doc/dev/osd_internals/wbthrottle.rst

Signed-off-by: Brad Hubbard <bhubbard@redhat.com>

doc: Fix typo in writeback throttle section

Fix typo in doc/dev/osd_internals/wbthrottle.rst

Signed-off-by: Brad Hubbard <bhubbard@redhat.com>

doc: Fix typo in recovery reservation section

Fix typo in doc/dev/osd_internals/recovery_reservation.rst

Signed-off-by: Brad Hubbard <bhubbard@redhat.com>

Merge pull request #5607 from rzarzynski/wip-12724

rgw: fix error handling during GET on object through Swift API.

Merge pull request #5563 from varadakari/wip-kvstore-keyformat

Change key format to preserve order of keys in backend db.

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

config: skip lockdep for intentionally recursive md_config_t lock

lockdep can't handle recursive locks, resulting in false positive
reports for certain set_val_or_die() calls, like via
md_config_t::parse_argv() passed "-m".

Fixes: #12614
Signed-off-by: Josh Durgin <jdurgin@redhat.com>

Merge pull request #5649 from ceph/wip-12764

tests: fixed rbd cli cram integration tests

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

buffer: modify inline memory ops to use packed structs

packed structs can be used to ensure alignment is not an issue.

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

uuid: use boost::random:random_device

The boost mt code uses uninitialized memory for extra randomness,
which is a bad idea in general but more importantly makes valgrind
unhappy.  Use /dev/urandom instead.

Unfortunately this introduces a link time dependency.. meh!

Fixes: #12736
Signed-off-by: Sage Weil <sage@redhat.com>

Merge pull request #5617 from rzarzynski/wip-12728

rgw: url_decode bucket name and prefix from X-Object-Manifest during GET on Swift DLO.

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

tests: fixed rbd cli cram integration tests

Fixes: #12764
Backport: infernalis
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Objecter: pg_interval_t::is_new_interval needs pgid from previous pool

When increasing the pg_num of a pool, an assert would fail since the
calculated pgid seed would be for the pool's new pg_num value instead
of the previous pg_num value.

Fixes: #10399
Backport: infernalis, hammer, firefly
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

librbd: error closing image while set to invalid snapshot

With cache disabled, closing the image will result in a flush
which might result in the image being refreshed.  If this happens
while the image is set to an invalid snapshot, an error will be
returned while closing the image.

Fixes: #12765
Backport: infernalis
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

osd: wait for cleanup from bench

We need to wait for cleanup to flush before we destroy the
sequencer or else we get a use-after-free.

Introduced e7bbafa3bfbd5e936a8be026a30b83a89f6121c3.

Fixes: #12766
Signed-off-by: Sage Weil <sage@redhat.com>

doc/release-notes: v9.0.3

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

Merge remote-tracking branch 'gh/next'

Merge pull request #5623 from yuyuyu101/wip-12729

Fix compile warning unused-result

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

set skip promote flag if NOCACHE or DONTNEED flag set

Signed-off-by: Yunchuan Wen <yunchuanwen@ubuntukylin.com>
Reviewed-by: Li Wang <liwang@ubuntukylin.com>
Reviewed-by: Sage Weil <sage@redhat.com>

osd: remove unused ctor

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

osd: remove useless hitset code

* we do not persist current hit set using get_hit_set_current_object()
  anymore, instead we always append current hitset into archive even
  !hitset.is_full(), see fbd9c15. so it's not necessary to remove the
  current hitset object before putting it to disk.

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

mon: print use_gmt_hitset in "ceph osd pool get"

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

mon: add "ceph osd pool set $pool use_gmt_hitset true" cmd

allow "ceph osd pool set $pool use_gmt_hitset <true|1>" as long as
the cluster supports gmt hitset.

Fixes: #9732
Signed-off-by: Kefu Chai <kchai@redhat.com>

osd: use GMT time for the object name of hitsets

* bump the encoding version of pg_hit_set_info_t to 2, so we can
  tell if the corresponding hit_set is named using localtime or
  GMT
* bump the encoding version of pg_pool_t to 20, so we can know
  if a pool is using GMT to name the hit_set archive or not. and
  we can tell if current cluster allows OSDs not support GMT
  mode or not.
* add an option named `osd_pool_use_gmt_hitset`. if enabled,
  the cluster will try to use GMT mode when creating a new pool
  if all the the up OSDs support GMT mode. if any of the
  pools in the cluster is using GMT mode, then only OSDs
  supporting GMT mode are allowed to join the cluster.

Fixes: #9732
Signed-off-by: Kefu Chai <kchai@redhat.com>

Merge pull request #5493 from dachary/wip-10887-shec

erasure-code: shec plugin feature

Reviewed-by: Takeshi Miyamae <miyamae.takeshi@jp.fujitsu.com>
Reviewed-by: Sage Weil <sage@redhat.com>

Merge pull request #5631 from wonzhq/finish-ctx

osd: consolidate encoding snaps into pg log code in finish_ctx

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

PendingReleaseNotes: make a note about KeyValueStore on-disk format change

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

Merge pull request #5611 from ceph/wip-newstore-misc

osd: final prereq patches for newstore

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

rgw: fix dangerous removal from STL map in filter_out_temp_url().

Fixes: #12750
Signed-off-by: Radoslaw Zarzynski <rzarzynski@mirantis.com>

test/erasure-code: drop directory from profile

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

do not include directory in ec profiles

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

osd: always load erasure plugins from the configured directory

Ignore the profile 'directory' field.

This ensures that we can always find plugins even when teh cluster
is installed across a mix of distros.

Rename the option to have no osd_ (or mon_) prefix since anybody
may use the ec factory/plugin code.

We still hard-code .libs in the unit tests... sigh.

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

9.0.3

erasure-code: shec plugin feature

There is one new plugin (shec). When upgrading a cluster, there
must be a protection against the following scenario:

   * the mon are upgraded but not the osd
   * a new pool is created using plugin shec
   * the osd fail to load the shec plugin because they have not been
     upgraded

A feature bit is added : PLUGINS_V3. The monitor will only agree to
create an erasure code profile for the shec plugin if all OSDs
supports PLUGINS_V3. Once such an erasure code profile is stored in the
OSDMap, an OSD can only boot if it supports the PLUGINS_V3 feature,
which means it is able to load the shec plugin.

The monitors will only activate the PLUGINS_V3 feature if all monitors
in the quorum support it. It protects against the following scenario:

   * the leader is upgraded the peons are not upgraded
   * the leader creates a pool with plugin=shec because all OSD have
     the PLUGINS_V3 feature
   * the leader goes down and a non upgraded peon becomes the leader
   * an old OSD tries to join the cluster
   * the new leader will let the OSD boot because it does not contain
     the logic that would excluded it
   * the old OSD will fail when required to load the plugin shec

This is going to be needed each time new plugins are added, which is
impractical. A more generic plugin upgrade support should be added
instead, as described in http://tracker.ceph.com/issues/7291.

See also 9687150ceac9cc7e506bc227f430d4207a6d7489 for the PLUGINS_V2
implementation.

http://tracker.ceph.com/issues/10887 Fixes: #10887

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

Merge pull request #5634 from ceph/wip-fix-FTBFS-boost

rbd: fix the FTBFS on old boost introduced by 2050d08

rbd: fix the FTBFS on old boost introduced by 2050d08

in boost 1.49, BOOST_SCOPE_EXIT() does not accept capture_tuple,
only `(capture) (capture) ...` is supported.

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

Merge pull request #5583 from ceph/wip-12676

mds: check mds up before calling MDSMap::get_mds_info

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

Merge pull request #5587 from caibo2014/master

rbd: remove the output file if rbd export-diff fails

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

Merge pull request #5632 from ceph/wip-fix-ftbfs-4998

osd: fix the FTBFS introduced by be28319

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

osd: fix the FTBFS introduced by be28319

the signature of Scrubber::write_blocked_by_scrub() was changed in #5470.

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

Merge pull request #4998 from wonzhq/hitset-remove-all

osd: implement hit_set_remove_all

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

osd: consolidate encoding snaps into pg log code in finish_ctx

'MODIFY', 'PROMOTE' and 'CLEAN' are sharing the same encoding snaps
code.

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

osd: only remove all the hit set objects on primary osd

Only the primary PG is allowed to remove all the hit set objects. And
the PG should be in the active or peered states.

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