librbd: read ops require journal replay

When journaling is enabled, a read op should acquire the exclusive
lock and replay the journal in case uncommitted writes are stored
in the journal.

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

Merge pull request #7371 from theanalyst/mon/quota-msg-cleanup

mon: cleanup set-quota error msg

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

Merge pull request #7708 from jdurgin/wip-cython-cleanup

pybind: replace __del__ with __dealloc__ for rbd

Reviewed-by: Hector Martin <marcan@marcan.st>
Reviewed-by: Mehdi Abaakouk <sileht@sileht.net>

mon: cleanup set-quota error msg

When we send in an unrecognized field for set-quota, specify that the
options should be `max_bytes` or `max_objects`. This doesn't affect the
ceph-cli as the choices are caught at an earlier stage, however the
error message is seen while reaching the cluster via rados api for eg.

Signed-off-by: Abhishek Lekshmanan <abhishek@suse.com>

Merge pull request #7668 from dillaman/wip-librbd-replay-locks

librbd: use async librados notifications

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

.gitignore: include recently added rgw programs

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

pybind: replace __del__ with __dealloc__ for rbd

Cython extension types like these don't call __del__, but use
__dealloc__ instead:

http://docs.cython.org/src/userguide/special_methods.html#finalization-method-dealloc

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

Merge pull request #7706 from sileht/sileht/rados-cython

pybind: remove next() on iterators

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

pybind: remove next() on iterator

Cython recommends to implement only __next__() methods.

http://docs.cython.org/src/userguide/special_methods.html#the-next-method

Signed-off-by: Mehdi Abaakouk <sileht@redhat.com>

Merge pull request #7641 from batrick/vstart-usage

Clarify usage on starting single osd/mds/mon.

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

Merge pull request #7692 from sileht/sileht/rados-cython

pybind/rados: use __dealloc__ since __del__ is ignored by cython

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

librbd: don't hold owner lock while refreshing image synchronously

The write lock will be taken when the new state is applied.

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

librbd: need owner write lock when disabling exclusive lock

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

librbd: correct memory leaks discovered via valgrind

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

librbd: notifications should be flushed between exclusive lock states

Avoid leaving in-flight notification messages when transitioning lock
states.

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

librbd: update image name upon rename operations

Previously the image could not have been renamed twice without
re-opening the image.

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

librbd: missing owner lock on snap rollback cache invalidation

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

librbd: journal replay should execute ops in clean context

lockdep will complain about loop cycles that won't cause an
issue in reality as replay and record are two different
journal states.

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

librbd: use AIO notifications to prevent blocking ops

If two or more images share the same CephContext, notifications
from one image can block the work queue which will potentially
block acknowledging the notification until after it times out.

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

librbd: cleanup header update notifications

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

Revert "librbd: lock notifications should be executed outside librados thread"

This reverts commit d898995b0e3ea301b1325f68a0532d57afa3c816.

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

librbd: support for AIO notifications

The header update and lock notifications might be invoked
from the librados AIO thread.  Update the close state
machine to flush any potential AIO notifications.

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

rstats: enable by default on client; move test workunit to its own dir

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

Merge pull request #7323 from ukernel/wip-client-dirsize

client: add option to control how directory size is calculated

This lets you disable rstats if your workload is unhappy about directories
changing size (eg, tar of recently-moved/created/untarred files).

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

Merge pull request #7457 from weiqiaomiao/w1

Rgw: check the return value when call fe->run()

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

Merge pull request #7666 from rzarzynski/wip-rgw-sysreq-over-swift

rgw: add support for system requests over Swift API

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

Merge pull request #7678 from linuxbox2/rgw-sigusr1

rgw: move signal.h dependency from rgw_front.h

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

Merge branch 'greg-pr-7034'

Merge pull request #7679 from linuxbox2/rgw-ssconst

sstring.hh: return type from str_len(...) need not be const

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

pybind: rados cleanup

This change uses:
* a pystring to store ioctx state like other state attributes and like
  previous binding
* use __dealloc__ instead of __del__ to ensure Object are freed correctly

Signed-off-by: Mehdi Abaakouk <sileht@redhat.com>

Merge pull request #7684 from ceph/wip-pybind-install

pybind: use correct subdir for rados install-exec rule

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

Merge branch 'wip-sharded-scan' of git://github.com/jcsp/ceph into greg-pr-7034

Conflicts:
src/osdc/Objecter.cc

pybind: use correct subdir for rados install-exec rule

This fixes package builds and 'make install'

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

Merge pull request #7621 from sileht/sileht/rados-cython

python binding of librados with cython

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

rgw: remove unnecessary overloads in RGWHTTPArgs.

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

sstring.hh: return type from str_len(...) need not be const

Removes a compile warning.

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

rgw: move signal.h dependency from rgw_front.h

Missed in refactoring of rgw_main.cc.

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

Update the documentation

Signed-off-by: Mehdi Abaakouk <sileht@redhat.com>

librbdpy: Use new rados lib

Signed-off-by: Mehdi Abaakouk <sileht@redhat.com>

ceph.in: Use new python rados module

This implements run_in_thread inside the ceph command itself.

And fixes the ceph command bootstrap when it run inside the
source tree to correctly load the new rados python module.

Signed-off-by: Mehdi Abaakouk <sileht@redhat.com>

Fix rpm/deb packaging

Signed-off-by: Mehdi Abaakouk <sileht@redhat.com>

Remove old rados pybinding

Signed-off-by: Mehdi Abaakouk <sileht@redhat.com>

Cython Rados module

Notable changes:

* run_in_thread have disapeared
* timeout argument of some methods are ignored
* rados_create_write_op/rados_create_read_op returns WriteOp/ReadOp
  instead of the pointer address
* rados_monitor_log callback 'arg' arguments was broken in previous python
  binding (callback was called with the pointer address instead pointed object)
* object attributes that was pointer addresses are now private and not accessible in python

Some tests have been added to cover all methods

Signed-off-by: Mehdi Abaakouk <sileht@redhat.com>

Merge pull request #7603 from roidayan/xio_fixes

Xio fixes

Reviewed-by: Matt Benjamin <mbenjamin@redhat.com>
Reviewed-by: Casey Bodley <cbodley@redhat.com>

rgw: support system requests over Swift API.

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

rgw: enable access to system arguments of RGWHTTPArgs.

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

Move pybind rbd module into it own directory

To allow to create a autonomous rados module with cython.
We move the current librbdpy to the rbd sub directory.

Signed-off-by: Mehdi Abaakouk <sileht@redhat.com>

Merge pull request #7520 from renhwsky/renhw-wip-mon-monitor

mon/monitor: some clean up

Reviewed-by: Joao Eduardo Luis <joao@suse.de>
Reviewed-by: Kefu Chai <kchai@redhat.com>

Merge pull request #7629 from ceph/wip-14750-op-timeouts

Objecter: detect laggy ops with objecter_timeout, not osd_timeout

Reviewed-by: Adam C. Emerson <aemerson@redhat.com>

Merge pull request #7655 from gregsfortytwo/wip-metarequest-lock

client: removed unused Mutex from MetaRequest

Reviewed-by: Yan, Zheng <zyan@redhat.com>

Merge pull request #7616 from gregsfortytwo/wip-timer-14697

mds: don't double-shutdown the timer when suiciding

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

Merge pull request #7490 from x11562/mds_mdstype_ref

mds: function parameter 'df' should be passed by reference

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

Merge pull request #7610 from dillaman/wip-bit-vector

common/bit_vector: use hard-coded value for block size

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

Merge pull request #7653 from dillaman/wip-librbd-race-conditions

librbd: several race conditions discovered under single CPU environment

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

Merge pull request #7408 from ceph/wip-14549

rgw: don't use s->bucket for metadata api path entry

Reviewed-by: Orit Wasserman <owasserm@redhat.com>

Merge pull request #7633 from zaitcev/cleanup-various-1

rgw: cleanups to comments and messages

Reviewed-by: Orit Wasserman <owasserm@redhat.com>

vstart.sh: clarify usage on single osd/mds/mon.

vstart.sh expects the literal string "osd"/"mds"/"mon". To me, it was confusing
as I interpreted the usage hint as an optional count for each daemon.

Signed-off-by: Patrick Donnelly <batrick@batbytes.com>

librados_test_stub: watch_flush should block until notifies complete

It shouldn't block until all the acks are received as that is not
how librados handles the method.

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

librbd: lock notifications should be executed outside librados thread

Otherwise it's possible that the notification will be blocked if
the librados AIO thread isn't available to invoke the notification
handler.

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

librbd: potential race on image close

If multiple states are pending, it's possible for an image to
be closed while the state lock is unlocked during the callback.

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

librbd: potential deadlock during AIO image re-open

The AIO image open might complete under the librados AIO callback.
Attempting to close the old image handle under that thread will
deadlock under the new librados_test_stub flush handling.

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

librados_test_stub: ensure AIO callback is also flushed

There are sporadic librbd unit test failures due to the
mock being destroyed while an active librados_test_stub
AIO callback is completing.

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

Merge branch 'wip-commit_position-tests' of git://github.com/trociny/ceph

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

tests: librbd: commit position updated after journal replay

Signed-off-by: Mykola Golub <mgolub@mirantis.com>

common/bit_vector: use hard-coded value for block size

The CEPH_PAGE_SIZE is not actually a constant. On aarch64 platforms
the page size is 64K instead of the assumed 4K.  The CRC block
computations require a true constant.

Fixes: #14747
Signed-off-by: Jason Dillaman <dillaman@redhat.com>

Merge pull request #7646 from dachary/wip-ceph-disk-flake8

ceph-disk: flake8 fixes

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

ceph-disk: flake8 fixes

Signed-off-by: Loic Dachary <loic@dachary.org>

rgw: cleanups to comments and messages

These are minor cleanups that do not affect the code in a
material way:
- Forgot to add --tenant to usage message
- Make the help slightly less misleading
- Fix a strange error message "could not add remove caps"
- Spelling in a comment
- The vstart.sh can be run with civetweb

Signed-off-by: Pete Zaitcev <zaitcev@redhat.com>

client: removed unused Mutex from MetaRequest

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

Merge pull request #7634 from ceph/rgw-nfs-5

nfs for rgw (Matt Benjamin, Orit Wasserman)

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

Objecter: detect laggy ops with objecter_timeout, not osd_timeout

Fixes: 14750
Signed-off-by: Greg Farnum <gfarnum@redhat.com>

Merge pull request #7350 from ddiss/tracker14499_enotconn_rados_monitor_log

librados: check connection state in rados_monitor_log

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

Merge pull request #7584 from ifed01/wip-14511

osd: fix lack of object unblock when flush fails

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

Merge pull request #7593 from ifed01/wip-no-blockedpromote-return-fix

osd: fix return value from maybe_handle_cache_detail()

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

Merge pull request #7587 from cbodley/wip-pageset-align

osd: memstore: fix alignment of Page for test_pageset

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

Merge pull request #7595 from liewegas/wip-hobject-stringify

osd: ghobject_t: use ! instead of @ as a separator

Merge pull request #7596 from dachary/wip-erasure-code-corpus

tests: sync ceph-erasure-code-corpus for mktemp -d

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

Merge pull request #7608 from liewegas/wip-warnings

tests: fix a few build warnings

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

librgw: fix rpm packaging

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: fix new Debian control entries

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: remove junk files

Files accidentally added.

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: don't compare a string_ref with "" (use empty())

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: fix boost::basic_string_ref template instantiation

The commit adding Seastar sstring.h replaced a part of the interface
which used an internal string-view type, with an equivalent using
basic_string_ref from Boost.  I relied on a default template argument
not present in older versions.

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: add debian and rpm packaging

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: consolidate S3 bucket name validation

There is more cleanup potential here, but it is intrusive, as the
validators for bucket and object names have been factored through
the handler inheritance hierarchy.

This change just deletes the S3 validate_bucket_name for now, as
that method was large.  The corresponding object name validator was
not defined in RGWHandler_REST_S3, so I've left it as is, for now.

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

Merge pull request #7626 from tchaikov/wip-fix-automake

Makefile: workaround an automake bug for "make check"

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

librgw: fix copyright in src/include/rados/rgw_file.h

This file originated at Red Hat.

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: initialize RGWLib pointer members

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: remove slo, dlo, and supplied-md5 logic in RGWWriteRequest

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: remove RGWLibRequest "magic" numbers added for debugging

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: comment rgw_nfs config_ops.h tunables

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

cohort_lru: remove #if 0 section

The declaration being skipped was meant as a placeholder for future
optimization work, moved to a comment.

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

{rgw,common}/Makefile.am: add a few more missing .h files

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

common: add xxhash.h to Makefile.am

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

rgw: fix null dereference

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

librgw/rgw: fix librgw.la linkage

Ensure that key symbols from global are satisfied in the library.

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw/rgw: reorganize autotools linkage

Adds the RGW-NFS library to the radosgw build, with unit tests.
Adds -fPIC to compile flags to satisfy the linker, but should have
been evident to automake/libtool.

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: call obj_rec::sync() on dirs1_b

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw_file_nfsns: create readf_out_name, if needed

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>

librgw: remove deprecated/unused librgw personality

Signed-off-by: Matt Benjamin <mbenjamin@redhat.com>