rgw: radosgw_admin --help typo

rgw: implement atomic multipart upload, atomic copy

rgw: atomic put

rgw: radosgw_admin user rm --purge-data

config: remove unused iat options

Signed-off-by: Josh Durgin <josh.durgin@dreamhost.com>

testceph: actually call closedir on opened directory.

Otherwise testceph hangs because it's still got things open!
Fixes #1228

Signed-off-by: Greg Farnum <gregory.farnum@dreamhost.com>

qa: initial commit of file_layout.sh

This tests the file layout functionality. Right now you need to
check the comparisons yourself. Expect that to change shortly.

Signed-off-by: Greg Farnum <gregory.farnum@dreamhost.com>

filer: no need for tid_t here

Signed-off-by: Sage Weil <sage@newdream.net>

client: print cache size during trim for debug

Signed-off-by: Sage Weil <sage@newdream.net>

FileStore: fix error in ALIGN and ALIGN_UP

Signed-off-by: Samuel Just <samuel.just@dreamhost.com>

radsogw_admin: fix clitest

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

testrados_delete_pool_while_open: fix printf args

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

radosgw_admin: --help/-h

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

create testrados_delete_pool_while_open

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

client: fix trim_caps

We can't blindly remove caps from inodes because we need at least one cap
for any inode in our cache.  Try to trim non-essential caps (when there's
>1), otherwise try to drop referring dentries and indirectly release caps
that way.

Signed-off-by: Sage Weil <sage@newdream.net>

client: fix mds touch

client: prefer auth cap in caps_issued_mask()

If we have an auth cap, prefer to touch that one.  This helps consolidate
caps on a single mds and allows mds replicas to eventually recall their
state.

Signed-off-by: Sage Weil <sage@newdream.net>

client: touch cap on lookup even if we use the dentry lease

Touch the dir cap for the lease's mds even if we use the lease to traverse.
This makes the trim_caps() behave better because it keeps the dentry and
session cap LRUs in sync.

Signed-off-by: Sage Weil <sage@newdream.net>

client: touch dentries during path traversal

Signed-off-by: Sage Weil <sage@newdream.net>

testrados: add NUM_THREADS

We can now test librados access with multiple simultaneous threads from
a single process.

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

.gitignore: update for test_libceph_build

Signed-off-by: Josh Durgin <josh.durgin@dreamhost.com>

global_init: make global_init_set_globals static

Now that the libraries have been de-globalized, this funtion should only
be used by global_init itself.

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

CephFSInterface: de-globalize

CephFSInterface: de-globalize

liceph: add ceph_get_mount_context

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

Fix C_NotifyComplete::notify

C_NotifyComplete::notify was broken in an earlier change where I removed
a derr() without removing the if statement before it.

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

cleanup

mon: order degraded before unfound

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

osd: clean up pool/pg_stat_t structs slightly

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

FileStore: ensure that CLONE_RANGE ranges are aligned

Signed-off-by: Samuel Just <samuel.just@dreamhost.com>

PG: assemble_backlog - fix comparison error, tail is non-inclusive

Signed-off-by: Samuel Just <samuel.just@dreamhost.com>

ReplicatedPG: make sure that rollback target is not degraded

Signed-off-by: Samuel Just <samuel.just@dreamhost.com>

ReplicatedPG,OSD: clone srces need to be non-degraded

do_op: check that clone_range src sobjects are present on
replicas as well.  Also, clean up src_obc map if we encounter an error
before returning.

Signed-off-by: Samuel Just <samuel.just@dreamhost.com>

libceph: build without globals

libceph, libosdc, libclient: depend on libcommon rather than libglobal.
No more globals in libceph.

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

libceph: de-globalize

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

filestore: remove warning

Filer, Journaler: deglobalize

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

ObjectCacher.cc: de-globalize

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

rgw: restructure permissions verification

apparently 100-continue was broken

filestore: don't bother doing fiemap on small ranges

Merge remote branch 'origin/osd_metrics'

update .gitignore

client/Client.cc: deglobalize

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

qa: wget -q for kenrel_untar_build.sh

Signed-off-by: Sage Weil <sage@newdream.net>

librgw: eliminate globals

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

update .gitignore

librados, librbd, librgw: remove dep on libglobal

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

osd: reorder, cleanup prologger metrics

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

osd: rip out old peer stat sharing

This never really worked, and has bitrotted, and the stats it's calculating
are mostly redundant anyway.

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

osd: rename peer_stat_lock -> stat_lock

It protects osd_stat too.

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

osd: instrument osdmap updates

Messages, epochs, dup epochs.

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

osd: clean up numpg*, hb*

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

osd: instrument readable latency too

Time before a write is readable (not necessarily on disk).  Note that if we
get the commit first (e.g. writeahead journal) this value isn't calculated
or logged.

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

osd: remove lame rlsum/num

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

osd: fix loadavg check

Use the syscall, stupid.

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

osd: instrument pushes, pull requests

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

osd: add sub_op instrumentation

pull subop
push subop

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

osd: fix up log_op_stats to actually work

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

osd: rename push applied, committed completions

Be consistent.

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

osd: clean up client op stats

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

osd: fix compilation

Revert "osd: make it compile on older boost libraries"

This reverts commit 116241ef106e6a88a84482f7f0f21219a6d29e33.

test_ceph_argparse: fix build

When this was merged, libcommon.a was not changed to libcommon.la

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

push global_init_set_globals into preinit callers

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

osd: make it compile on older boost libraries

Merge branch 'next'

Conflicts:
src/librbd.cc

Merge remote branch 'origin/stable' into next

Conflicts:
src/librbd.cc

create test_librados_build

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

librados: fix end_block calculation for aio_read

Conflicts:

src/librbd.cc

librados: fix end_block calculation for aio_read

Conflicts:

src/librbd.cc

librbd: track block_ofs instead of ofs in aio_sparse_read callback

Add ceph_argparse unit test

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

ceph_argparse_witharg: fix dashes in args

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

common/Throttle: Remove unused return type on Throttle::get()

Signed-off-by: Jim Schutt <jaschut@sandia.gov>
Signed-off-by: Sage Weil <sage@newdream.net>

libceph: delogbalize, again

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

Makefile: add missing common/dout.h header to dist tarball

Signed-off-by: Sage Weil <sage@newdream.net>

librados: fix end_block calculation for aio_read

proflogger: only register one event for all loggers

If we already did logger_start(), don't schedule a new event.

Signed-off-by: Sage Weil <sage@newdream.net>

librados: deglobalize, again

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

OSDMap: kill some deadcode, deglobalize

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

librados: deglobalize

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

AnchorServer: fix return values in add().

I accidentally didn't set the return value in my new branch. (Not
that they're read anywhere.) While I'm at it, document what they mean.

Signed-off-by: Greg Farnum <gregory.farnum@dreamhost.com>

uclient: change snaprealm refcounting to erase at 0 refs.

This is more idiomatic and less confusing than having 0 refs be
still alive, and -1 be the point of destruction.

Signed-off-by: Greg Farnum <gregory.farnum@dreamhost.com>

mount.ceph, rbd: Output errors to stderr, not stdout.

Signed-off-by: Tommi Virtanen <tommi.virtanen@dreamhost.com>

mount.ceph, errno: Avoid printing binary, clarify message.

Payload is a binary key, printing it to stdout is not helpful.

Key name is often "client", make the message easier to understand.

Signed-off-by: Tommi Virtanen <tommi.virtanen@dreamhost.com>

mount.ceph, rbd: Avoid clobbering errno with printf and strerror calls.

Signed-off-by: Tommi Virtanen <tommi.virtanen@dreamhost.com>

run-cli-tests: Pass through CCACHE_DIR and such env vars.

Commit 7cd50f29d5cbf8deb64d00318b39c281119c0e03 makes the binaries
use libtool's "executable wrappers", which will transparently relink
the executables if they think that's needed. The test for that is
somewhat flawed: if the mtimes match, the binary might or might not
get relinked. In practise, this causes relinks on gitbuilder all the
time.

As of earlier commit 5a0bc6b78f2e40ec9255a1ea49f77ef9ea4690a6, we
started sanitizing the environment passed to the clitests. This meant
we also stripped away CCACHE_DIR and other settings, needed to
properly relink the binaries. Re-add CCACHE_DIR, CC, CXX to clitests
environment.

To handle the case where CCACHE_DIR etc are not set in the first
place, we need an extra wrapper script. Otherwise, ccache might see an
empty string as the env value, and naturally couldn't access a
directory by that name.

Signed-off-by: Tommi Virtanen <tommi.virtanen@dreamhost.com>

Remove "#include debug.h" from config.h

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

librbd: Remove derr uses

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

Merge branch 'next'

Conflicts:
src/librbd.cc

client: always use get_snap_context() accessor

There were a few places where we were using the member directly and not the
accessor, which rebuilds the internal value when needed.  This caused
inconsistent behavior based on whether debugging was enabled or not, since
we used the accessor to print the regenerated value.

Passes all snaps workunits with debugging off.

Signed-off-by: Sage Weil <sage@newdream.net>

librbd.cc: de-globalize

Signed-off-by: Colin McCabe <colin.mccabe@dreamhost.com>

AnchorServer: overwrite old Anchor backpointers when proper.

Sometimes when we do an AnchorTable update, it's because the inode
in question got moved. However, if the inode had multiple references,
its Anchor wasn't removed by decrementing its count, and so the add
function simply noted that it already had the Anchor and returned.

This obviously wasn't the proper behavior in cases where the inode
was getting moved -- we want to update its back pointer! So do so.

Signed-off-by: Greg Farnum <gregory.farnum@dreamhost.com>

uclient: Update statfs to match the kernel client and its block sizing.

Make it better match the kernel client, and its scheme to use a large
block size so we don't overflow 32-bit systems. This isn't presently
a serious concern since FUSE doesn't work on 32-bit systems anyway,
but the output of df should match even so.

Signed-off-by: Greg Farnum <gregory.farnum@dreamhost.com>

uclient: fix flush_caps(Inode*,int)

This function was just broken before. You need to be setting
flushing_cap_tids for the caps you're actually flushing, which
in this case is in->dirty_caps, not the horribly-named "flush"
variable.
Also, rename "flush" to "retain" since that's what it actually is!

Signed-off-by: Greg Farnum <gregory.farnum@dreamhost.com>

uclient: only change the auth_cap if the mseq is newer

Previously we just trusted the MDS' reported auth status, which meant
that even if the MDS was always right it could encode the stat while
auth, then export, and then have us decode the stat after getting
an IMPORT and set the auth cap right back!

Signed-off-by: Greg Farnum <gregory.farnum@dreamhost.com>

mds: Include the pinner in nested auth pin debugging output.

This will make identifying stuck pins a lot easier.

Signed-off-by: Greg Farnum <gregory.farnum@dreamhost.com>

librbd: track block_ofs instead of ofs in aio_sparse_read callback

Revert "debian: ceph-client-tools depends on librados2, librbd"

This reverts commit 96ef8a67bcf4a4a43f0a5c38224314abdd88a12c.

librbd: fix read_iterate for the case where block ofs is not 0

Conflicts:

src/librbd.cc

librbd: fix read_iterate for the case where block ofs is not 0

Merge remote branch 'origin/stable' into next

Conflicts:
src/librbd.cc