osd/ReplicatedPG: remove stray debug line

This snuck in

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit ff9600a6cef613d40e875597b6392778df1bb04c)

Merge pull request #6330 from dachary/wip-13550-hammer

qemu workunit refers to apt-mirror.front.sepia.ceph.com

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

qa: Use public qemu repo

This would allow some external tests outside of sepia lab

Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>
(cherry picked from commit 4731c1e35539c9506ff3fe3141553fad099d0eee)

0.94.4

Merge pull request #6287 from jcsp/wip-test-fixes

git path backports for hammer

use git://git.ceph.com

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 34c467ba06345eacd2fd0e79c162dfd0b22f0486)

qa: http://ceph.com/qa -> http://download.ceph.com/qa

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit d4d65fbd6ad15320339751c604613ac63511e645)

Merge pull request #6161 from dachary/wip-13227-hammer

With root as default user, unable to have multiple RGW instances running

Reviewed-by: Sage Weil <sage@redhat.com>
Reviewed-by: Yehuda Sadeh <ysadehwe@redhat.com>
Reviewed-by: Ken Dreyer <kdreyer@redhat.com>

init-radosgw.sysv: remove

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 9d8c115b0ab05cfc1db6670d1d367737e61fe0ea)

Merge pull request #6166 from dachary/wip-13046-hammer

RGW : setting max number of buckets for user via ceph.conf option

Merge pull request #6039 from ceph/wip-13015-hammer

rgw: don't preserve acls when copying object

Merge pull request #6042 from dreamhost/wip-hammer-crossftp

rgw: remove trailing :port from host for purposes of subdomain matching

Merge pull request #6160 from dachary/wip-13228-hammer

rgw: segments are read during HEAD on Swift DLO

Merge pull request #6162 from dachary/wip-13226-hammer

Keystone Fernet tokens break auth

Merge pull request #6163 from dachary/wip-13054-hammer

rgw: region data still exist in region-map after region-map update

Merge pull request #6164 from dachary/wip-13053-hammer

rgw: GWWatcher::handle_error -> common/Mutex.cc: 95: FAILED assert(r == 0)

Merge pull request #5718 from dachary/wip-12859-hammer

testGetContentType and testHead failed

Merge pull request #5860 from lebauce/wip-12960-hammer

rgw: add delimiter to prefix only when path is specified

Merge pull request #6165 from dachary/wip-13052-hammer

rgw: init_rados failed leads to repeated delete

Merge pull request #6201 from dachary/wip-13410-hammer

TEST_crush_rule_create_erasure consistently fails on i386 builder

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

Merge pull request #5885 from Abhishekvrshny/wip-13034-hammer

osd: copy-from doesn't preserve truncate_{seq,size}

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

tests: robust test for the pool create crushmap test

The test that goes with f1e86be589803596e86acc964ac5c5c03b4038d8 to
verify that a bugous crush ruleset will prevent the creation of a pool
trying to use it was fragile. I depends on the implementation of the
erasure code lrc plugin and turns out to not work on i386.

The test is modified to use a fake crushtool that always returns false
and validate that it prevents the creation of a pool, which demonstrate
it is used for crushmap validation prior to the pool creation.

http://tracker.ceph.com/issues/12419 Fixes: #12419

Signed-off-by: Loic Dachary <ldachary@redhat.com>
(cherry picked from commit e47903727b5d53f795014bf29b417d357f2bd65a)

Merge pull request #6192 from dachary/wip-13401-hammer

mon: fix crush testing for new pools

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

crush/CrushTester: test fewer inputs when running crushtool

If there are a lot of crush rules (say, 100) then the test can
take a long time.  100 values per rule should be enough to catch
most issues.

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 0f82f461b33d93d868e185912a2c7e4074d06900)

Conflicts:
src/crush/CrushTester.cc
  in hammer the crushtool validation is via a shell
       and not via an internal subprocess utility

tests: update to match crushmap validation message

http://tracker.ceph.com/issues/13182 Fixes: #13182

Signed-off-by: Loic Dachary <ldachary@redhat.com>
(cherry picked from commit 21a1e75d8a7bad89a48cd9d36902c5d609be5015)

Conflicts:
qa/workunits/cephtool/test.sh
  the test structure in hammer is different, adapt
  the relevant test.

mon/OSDMonitor: fix crush injection error message

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 1b3090d50e5bd5ca3e6e396b23d2d9826896c718)

Conflicts:
src/mon/OSDMonitor.cc
  hammer has a separate error message for EINTR
  which is in the context of the fix for error message

mon/OSDMonitor: only test crush ruleset for the newly created pool

Otherwise, we test *all* crush rules.. which might be a lot, and
which is a big waste of time and effort.

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 524b0bdcc45c2f4b95f2239c988e93250f337f3d)

Conflicts:
src/mon/OSDMonitor.cc : trivial resolution

crush/CrushTester: allow testing by ruleset

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit aa238e5ed50f44a94caf84567267e4f6be8732a2)

Conflicts:
src/crush/CrushTester.cc
          in hammer the crushtool validation is via a shell
  and not via an internal subprocess utility
src/tools/crushtool.cc
  ceph_argparse_withint is preferred to ceph_argparse_witharg

Merge pull request #5887 from Abhishekvrshny/wip-13044-hammer

LibCephFS.GetPoolId failure

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

Merge pull request #6172 from dachary/wip-13354-hammer

qa/workunits/cephtool/test.sh: don't assume crash_replay_interval=45

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

qa/workunits/cephtool/test.sh: don't assume crash_replay_interval=45

e.g., it's 5 in teuthology's ceph.conf.

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 80b7237e5e74d12b9d8b1f96ea535c96bdff9c6f)

rgw:add --reset-regions for regionmap update

Fixes: #12964
Signed-off-by: Weijun Duan <duanweijun@h3c.com>
(cherry picked from commit 95685c19d6f1eab50b903e61273b5351bedc2980)

Conflicts:
src/rgw/rgw_admin.cc: usage is on cerr in hammer, not cout
        src/test/cli/radosgw-admin/help.t: remove extra empty line

Merge pull request #6156 from dachary/wip-13170-hammer

doc: update docs to point to download.ceph.com

Reviewed-by: Abhishek Varshney <abhishek.varshney@flipkart.com>

rgw : setting max number of buckets for users via ceph.conf option

This patch adds a new option "rgw_user_max_buckets" for setting
max number of buckets for users via ceph.conf.

Fixes #12714

Signed-off-by: Vikhyat Umrao <vumrao@redhat.com>
(cherry picked from commit f65267c96cbd4cd25036b6bf399692e77bbb9436)

Conflicts:
src/rgw/rgw_admin.cc: usage is on cerr in hammer, not cout

rgw: init_rados failed leads to repeated delete

Fixes: #12978
Signed-off-by: Xiaowei Chen <chen.xiaowei@h3c.com>
(cherry picked from commit ab4232baa7bcc86e90746e13312ac9bda1772872)

rgw: delete finisher only after finalizing watches

Fixes: #12208
The watch error path might try to schedule a finisher work, delete finisher
only after watch destruction.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 97aed59fb9e865d30d31d2b7f4e93fc9727c96fa)

rgw: be more flexible with iso8601 timestamps

make parsing 8601 more flexible by not restricting the length of seconds
to 5, this allows timestamp to be specified both as ms or us. Newer
keystone backends such as fernet token backend default to microseconds
when publishing iso8601 timestamps, so this allows these timestamps to
be allowed when specifying the token expiry time.

Fixes: #12761
Reported-by: Ian Unruh <ianunruh@gmail.com>
Signed-off-by: Abhishek Lekshmanan <abhishek.lekshmanan@ril.com>
(cherry picked from commit 136242b5612b8bbf260910b1678389361e86d22a)

init-radosgw: specify pid file to start-stop-daemon

This fixes restart when multiple instances are running.

Fixes: #12407
Tested-by: Pavan Rallabhandi <pavan.rallabhandi@sandisk.com>
Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit e17c8e1a7a06db0e2680dedc7e954021610320d1)

rgw: fix radosgw start-up script.

radosgw init script is unable to start radosgw daemon.
as it is relies on requiretty being disabled.
once init script start this daemon with sudo it fails
to start the daemon.

changing 'sudo' to 'su', it will fix this issue and
will also help running radosgw daemon with our new
"ceph" UID project.

Fixes: #10927
Signed-off-by: Vikhyat Umrao <vumrao@redhat.com>
(cherry picked from commit c83542296aeb8f5dd96f5d9e09327e7a89370742)

init-radosgw: unify init-radosgw[.sysv]

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 1c45f512a91ee49e9253ba6fd3857af9bd2ae57a)

init-radosgw: look in /var/lib/ceph/radosgw

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit a4bb992426b39188e750fce9198052ca0d510117)

doc: rgw: fix typo in comments

Signed-off-by: Nathan Cutler <ncutler@suse.cz>
(cherry picked from commit 05cf0db22dc3a9cb2bf4b86b8f240683c23f41ac)

rgw: init script waits until the radosgw stops

Fixes: #11140
Init script waits in stop action until the radowgw daemons stop.

Signed-off-by: Dmitry Yatsushkevich <dyatsushkevich@mirantis.com>
(cherry picked from commit 1cca0c1e91a1e18f82f4d22855e96b2fc947f5ea)

rgw: don't read actual data on user manifest HEAD

Fixes: #12780
We unconditionally read all the data, which is not needed
when doing HEAD operation on user manifest objects.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
Signed-off-by: Radoslaw Zarzynski <rzarzynski@mirantis.com>
(cherry picked from commit 4a8b08d1dd2511e28e999f1625103fb919fb5465)

doc: remove mention of ceph-extra as a requirement

Signed-off-by: Alfredo Deza <adeza@redhat.com>
(cherry picked from commit a6f07e9fa3bd5e50a2d8cfa387b44a39d383fa65)

doc: remove ceph-extras

Signed-off-by: Alfredo Deza <adeza@redhat.com>
(cherry picked from commit c0ef84fd22aac58b6ec8e58eb4f8dffed306d39e)

doc: correct links to download.ceph.com

Signed-off-by: Alfredo Deza <adeza@redhat.com>
(cherry picked from commit 387d7800359154950431d0984c756f43f21dd9b4)

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

Fixes: #12771
Signed-off-by: ritz303 <ritz_303@yahoo.com>
(cherry picked from commit 582f0f64455079290ad3b4ae7338b716e170911f)

rgw: set default value for env->get() call

Fixes: #13239
This fixes a regression introduced at commit abe4ec293d08b0314bf5c081ace2456073f3a22c.
The host var is a string, env->get() returns a char pointer, shouldn't
pass in NULL.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit 0d5730efc0054334897315dc23ba04f30548e36e)
Signed-off-by: Robin H. Johnson <robin.johnson@dreamhost.com>

osd/ReplicatedPG: tolerate promotion completion with stopped agent

We may start a promotion, then get a pool update that disables the
agent, and then complete it.  We should not segfault in this case.

Fixes: #13190
Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit fd9ce66e8b76787f69c5e337332d57aa72c8b8d5)

rgw: remove trailing :port from host for purposes of subdomain matching

Some clients (ahem, CrossFTP) include the :port in the HTTP_HOST header.
Strip it out.

Switch req_info field to a std::string and avoid copying it in preprocess.

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit abe4ec293d08b0314bf5c081ace2456073f3a22c)

Merge pull request #5715 from dachary/wip-12592-hammer

RGW returns requested bucket name raw in Bucket response header

Merge pull request #5719 from dachary/wip-12855-hammer

segmentation fault when rgw_gc_max_objs > HASH_PRIME

Merge pull request #5720 from dachary/wip-12854-hammer

the arguments 'domain' should not be assigned when return false

rgw: preserve all attrs if intra-zone copy

Fixes: #13015
Intra zone copy requires that all objects' attributes are preserved.

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

rgw: don't preserve acls when copying object

Fixes: #12370
When copying an object we need to use the acls we calculated earlier,
and not the source acls.
This was broken at e41d97c8e38bb60d7e09e9801c0179efe7af1734.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit fb020247449d3daf033885bab5b3a10c0e5df4a5)

Merge pull request #5930 from liewegas/wip-11798-hammer

upstart: limit respawn to 3 in 30 mins (instead of 5 in 30s)

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

Merge pull request #5908 from liewegas/wip-simple-connect-hammer

Pipe: Drop connect_seq increase line

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

upstart: limit respawn to 3 in 30 mins (instead of 5 in 30s)

It may take tens of seconds to restart each time, so 5 in 30s does not stop
the crash on startup respawn loop in many cases.  In particular, we'd like
to catch the case where the internal heartbeats fail.

This should be enough for all but the most sluggish of OSDs and capture
many cases of failure shortly after startup.

Fixes: #11798
Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit eaff6cb24ef052c54dfa2131811758e335f19939)

Pipe: Drop connect_seq increase line

Revert commit 0fc47e267b6f8dcd4511d887d5ad37d460374c25.

When accepting and "connect.connect_seq == existing->connect_seq",
existing->state maybe STATE_OPEN, STATE_STANDBY or STANDY_CONNECTING.
This commit only fix partial problem and want to assert
"(existing->state == STATE_CONNECTING)".

So later we added codes to catch
"(existing->state == STATE_OPEN || existing->state == STATE_STANDBY)"
before asserting.

Backport: dumpling, firefly, giant

Signed-off-by: Haomai Wang <haomaiwang@gmail.com>
(cherry picked from commit 67225cb3ee1f6d274a02293724942bdb25cec8ca)

Merge pull request #5767 from dachary/wip-12597-hammer

Crash during shutdown after writeback blocked by IO errors

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

Merge pull request #5687 from liewegas/wip-hammer-feature-hammer

include/ceph_features: define HAMMER_0_94_4 feature

Merge pull request #5892 from ceph/wip-13060-hammer

osd: allow peek_map_epoch to return an error

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

osd/PG: peek_map_epoch: skip legacy PGs if infos object is missing

- pg is removed
- osd is stopped before pg is fully removed
- on restart, we ignore/skip the pg because its epoch is too old
- we upgrade to hammer and convert other pgs, skipping this one, and then
  remove the legacy infos object
- hammer starts, tries to parse the legacy pg, and fails because the infos
  object is gone, crashing.

The fix is to continue ignoring the zombie pg.

Fixes: #16030
Signed-off-by: Sage Weil <sage@redhat.com>

osd: allow peek_map_epoch to return an error

Allow PG::peek_map_epoch to return an error indicating the PG
should be skipped.

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit f15d9585edc5a12ac2d076951076247253b897c2)

[fixed *pepoch default of 0]
[fixed other return paths in peek_map_epoch]

crypto: fix unbalanced ceph::crypto::init/ceph::crypto:shutdown

we may create a CephContext without calling common_init_finish(), then
delete the CephContext. In this case, ceph::crypto:init() is not called,
so CephContext::~CephContext() should not call ceph::crypto::shutdown().

Fixes: #12598
Signed-off-by: Yan, Zheng <zyan@redhat.com>
(cherry picked from commit 98a85ec6464d8ec3fc7f0c647ac97c8cf002ebe2)

ReplicatedPG,Objecter: copy_get should include truncate_seq and size

Otherwise, we break CephFS over cache tiers.

Fixes: #12551
Signed-off-by: Samuel Just <sjust@redhat.com>
(cherry picked from commit 6f9ee7961eee9ee3a61a07cbe0d8d289ee98fa9a)

Conflicts:
src/osd/ReplicatedPG.cc

Merge pull request #5769 from dachary/wip-12850-hammer

Crash during TestInternal.MultipleResize

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

Merge pull request #5768 from dachary/wip-12849-hammer

[  FAILED  ] TestLibRBD.BlockingAIO

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

rgw: fix assignment of copy obj attributes

Fixes: #11563
Clarify the confusing usage of set_copy_attrs() by switching the source and
destinatiion params (attrs, src_attrs). Switch to use attrs instead of
src_attrs afterwards. In one of the cases we originally used the wrong
variable.

Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit e41d97c8e38bb60d7e09e9801c0179efe7af1734)

rgw: add delimiter to prefix only when path is specified

http://tracker.ceph.com/issues/12960
Fixes: #12960
Signed-off-by: Sylvain Baubeau <sbaubeau@redhat.com>
(cherry picked from commit 27cf257248ea55f8f4bc7851c3956611828bcae2)

Merge pull request #5697 from tchaikov/wip-12638-hammer

mon: add a cache layer over MonitorDBStore

Reviewed-by: Joao Eduardo Luis <joao@suse.de>

Merge pull request #5381 from dachary/wip-12499-hammer

ceph-fuse 0.94.2-1trusty segfaults / aborts

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

Merge pull request #5757 from dachary/wip-12836-hammer

WBThrottle::clear_object: signal on cond when we reduce throttle values

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

Merge pull request #5759 from dachary/wip-12841-hammer

recursive lock of md_config_t (0)

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

Merge pull request #5761 from dachary/wip-12843-hammer

long standing slow requests: connection->session->waiting_for_map->connection ref cycle

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

Merge pull request #5762 from dachary/wip-12844-hammer

osd suicide timeout during peering - search for missing objects

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

Merge pull request #5763 from dachary/wip-12846-hammer

osd/PGLog.cc: 732: FAILED assert(log.log.size() == log_keys_debug.size())

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

Merge pull request #5764 from dachary/wip-12847-hammer

common: do not insert emtpy ptr when rebuild emtpy bufferlist

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

Merge pull request #5373 from dachary/wip-12489-hammer

pg_interval_t::check_new_interval - for ec pool, should not rely on min_size to determine if the PG was active at the interval

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

Merge pull request #5383 from dachary/wip-12504-hammer

rest-bench common/WorkQueue.cc: 54: FAILED assert(_threads.empty())

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

Merge pull request #5765 from dachary/wip-12883-hammer

cache agent is idle although one object is left in the cache

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

Merge pull request #5754 from dachary/wip-12588-hammer

Change radosgw pools default crush ruleset

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

Merge pull request #5377 from dachary/wip-12396-hammer

register_new_pgs() should check ruleno instead of its index

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

Merge pull request #5758 from dachary/wip-12839-hammer

Mutex Assert from PipeConnection::try_get_pipe

tests: tiering agent and proxy read

Verify that an object promoted to a cache tier because of a proxy read
is evicted as expected.

http://tracker.ceph.com/issues/12673 Refs: #12673

Signed-off-by: Loic Dachary <ldachary@redhat.com>
(cherry picked from commit 7924231930732bd297d3bd034c8295e96cb81088)

osd: trigger the cache agent after a promotion

When a proxy read happens, the object promotion is done in parallel. The
agent_choose_mode function must be called to reconsider the situation
to protect against the following scenario:

  * proxy read
  * agent_choose_mode finds no object exists and the agent
    goes idle
  * object promotion happens
  * the agent does not reconsider and eviction does not happen
    although it should

http://tracker.ceph.com/issues/12673 Fixes: #12673

Signed-off-by: Loic Dachary <ldachary@redhat.com>
(cherry picked from commit e1f58feb9b1d20b72f2eb2eefdea5982e0cddccd)

Merge pull request #5276 from dachary/wip-11824-hammer

implicit erasure code crush ruleset is not validated

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

Merge pull request #5382 from dachary/wip-12500-hammer

segfault launching ceph-fuse with bad --name

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

Merge pull request #5367 from dachary/wip-12311-hammer

read on chunk-aligned xattr not handled

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

lockdep: allow lockdep to be dynamically enabled/disabled

librbd test cases attempt to enable lockdep coverage via the librados
API.  Use a configuration observer to register/unregister lockdep
support.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit af0cade0293be22e74a3293765676c7b227b06c4)

tests: librbd API test cannot use private md_config_t struct

Remove all depencencies on md_config_t and instead use librados API
methods to get/set configuration values.

Fixes: #12479
Backport: hammer
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 2c51aadb447d3447b5dace19c5cb6268e94247c8)

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>
(cherry picked from commit 3b95edb5bd373bd05f83516eccfd30a6b5cf4693)

librados_test_stub: implement conf get/set API methods

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 4d03c664f2a9362009c7e3077ed65e2989f0f21b)

crypto: use NSS_InitContext/NSS_ShutdownContex to avoid memory leak

Switched to context-aware NSS init/shutdown functions to avoid conflicts
with parent application.  Use a reference counter to properly shutdown the
NSS crypto library when the last CephContext is destroyed.  This avoids
memory leaks with the NSS library from users of librados.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 9fa0112dd9286178da1d76020158503b2062d252)

auth: use crypto_init_mutex to protect NSS_Shutdown()

Signed-off-by: Yan, Zheng <zyan@redhat.com>
(cherry picked from commit cf058174fe6be26129eb52f4fb24d0f7146ef7af)

auth: reinitialize NSS modules after fork()

Fixes: #11128
Signed-off-by: Yan, Zheng <zyan@redhat.com>
(cherry picked from commit 4c24d0cc074462ae258b5cf901cd884bb4f50a53)

librbd: prevent race condition between resize requests

It was possible that the same resize request could be sent twice
if a completed resize op started a newly created resize op while
it was also being concurrently started by another thread.

Fixes: #12664
Backport: hammer
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 6ec431bac1cf9446f3af90209d193dfcf003f2c5)

librbd: Add a paramter:purge_on_error in ImageCtx::invalidate_cache().

If bh_write met error, it will try again. For closing image, if met this
issue, it will trigger a assert:
>>2015-02-03 15:22:49.198292 7ff62d537800 -1 osdc/ObjectCacher.cc: In
function 'ObjectCacher::~ObjectCacher()' thread 7ff62d537800 time
>>2015-02-03 15:22:49.195927osdc/ObjectCacher.cc: 551: FAILED
>>assert(i->empty())

Now add purge_on_error, when shutdown_cache it set true.
In ImageCtx::invalidate_cache, if met error and purge_on_error is true,
purge the dirty bh.

Signed-off-by: Jianpeng Ma <jianpeng.ma@intel.com>
(cherry picked from commit 35def5c81f7fc83d55d18320e4860c6a63d4c7f5)

Conflicts:
src/librbd/ImageCtx.cc : trivial resolution

librbd: Remvoe unused func ImageCtx::read_from_cache.

Signed-off-by: Jianpeng Ma <jianpeng.ma@intel.com>
(cherry picked from commit 101440a41253680770f94bc380af7072c7adaebf)