msgr: avoid copying Pipe* xlist

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

mds: make avoid copying sessionmap by_state xlist

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

mds: avoid copying snap client_caps xlist

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

mon: do not copy subscription xlist

xlist is not copyable.

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

qa: clean up after snaptest-multiple-capsnaps.sh

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

Dispatcher: prevent copying

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

msgr: only SO_REUSEADDR when specific port is specified

In general SO_REUSEADDR is slightly dangerous, but avoids waiting for the
timeout when restarting servers.  This is important when binding to a
specific port.

When binding to a random port, it doesn't matter.  Also, it appears that
two processes can bind() to the same port with that flag set, and then one
will fail with EADDRINUSE on listen().  That's racy when starting up
daemons that should be binding to unique/random ports.

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

Merge branch 'wip-globalfoo'

mon: add 'log ....' command

Just sticks something in the central log.

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

Convert g_ceph_context to a pointer.

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

gitignore: add test_libcommon_build

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

inode_load_vec_t: de-globalize

Don't call ceph_clock_now from the constructor of inode_load_vec_t.
Update the classes that contain inode_load_vec_t to reflect this.

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

Create test_libcommon_build

Create test_libcommon_build to test for references to g_ceph_context and
other libglobal symbols from within libcommon.

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

DecayCounter: take a utime_t in ::decode

We need to initialize the last_decay time of the DecayCounter when
decoding it. This is not found in the encoded information, but instead
is set to the current time. We need to pass this in explicitly now
because of deglobalization.

This also reduces the number of calls to gettimeofday, which is good in
general.

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

Thread.cc: deglobalize

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

common_init: de-globalize

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

rgw: use adapting window for put obj

AuthNone: encode entity name in authorizer

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

include/atomic.h:fix build for non-atomic-ops case

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

rgw: put data using a window

MDSMap: de-globalize

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

assert: de-globalize

Implement a simple registration system so that assert output can still
show up non-interleaved in daemons and utilities. Library code will just
use dout_emergency, which is simple and deadlock-free, but might be
interleaved with ongoing messages sometimes.

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

Messenger: de-globalize

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

DoutStreambuf: de-globalize

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

OSDMap: de-globalize

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

auth: more deglobalization

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

hex.cc: de-globalize

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

MonClient: deglobalize

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

ProfLogger: de-globalize

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

tcp.cc: de-globalize

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

LogClient: remove accidental g_conf references

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

msg: remove globals

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

KeyRing: derr -> lderr

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

auth: AuthSupported: deglobalize

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

Create global_context.cc to house g_ceph_context

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

Move signal, pidfile to global

Move the global parts of signal and pidfile handling into global/

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

move common_init to global_init

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

Makefile: put LIBGLOBAL LDFLAGS into LIBGLOBAL_LDA

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

libclient: remove client/SyntheticClient.cc

client/SyntheticClient.cc is only of interest to csyn. The other users
of libclient don't need it.

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

mon/MonClient: use ldout instead of dout

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

mon/MonMap: deglobalize

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

SimpleMessenger: de-globalize

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

common/DecayCounter: deglobalize

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

ThreadPool: de-globalize

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

common: de-globalize LogClient, ProfLogger, Timer

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

common_init: misc globals cleanup

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

auth: remove remaining globals

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

auth: CephxKeyServer: de-globalize

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

auth: AuthAuthorizeHandler: deglobalize

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

auth: KeyRing, RotatingKeyRing: deglobalize

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

auth: CryptoKey, CryptoAES: deglobalize

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

Create ProfLoggerCollection to hold ProfLoggers

Rather than having global variables, create ProfLoggerCollection to
store the current set of ProfLoggers. This collection will be stored
inside of CephContext.

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

g_clock.now -> ceph_clock_now

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

common/Clock: add deglobalized clock stuff

Add clock functions which don't use globals. Remove Clock::recent_now.

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

auth: de-globlize CephXAuthorizer

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

auth: de-globalize AuthServiceHandler and pals

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

auth: de-globalize AuthClientHandler and friends

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

auth: de-globalize TicketManager, TicketHandler

de-globalize CephXTicketManager and CephXTicketHandler. Also
de-globalize two static functions.

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

run_cmd: ret empty string on success;err otherwise

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

include/atomic cleanup

* Don't allow copying of class atomic_t.
* Remove common/Spinlock.h because it's unecessary
* SimpleMessenger: use atomic var for qlen

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

include/atomic.h: whitespace cleanup

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

bufferlist: get rid of derr usage

bufferlist doesn't need to use dout/derr. Only one function used it, and
that was only to return an error string.

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

include/Completion: de-globalize

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

ObjectStore: clean up includes a little bit

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

include/Context.h: de-globalize

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

common/MemoryModel: de-globalize

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

common/Finisher: use ldout

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

lockdep: code cleanup and de-globalization

common/Mutex.cc: this is a duplicate of the code in common/lockdep.cc.
Delete the duplicate code.

common/lockdep.cc: require the programmer to register a CephContext to
use with lockdep. If we don't have one, we don't print anything out.

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

Thread.cc: clean up

Don't implement everything in the header file. It slows down compilation
and usually runtime.

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

Create de-globalized versions of the dout macro

The de-globalized versions are available in dout.h
Versions relying on globals continue to be available in debug.h

Split md_config_obs_t off into common/config_obs.h. This will make it
possible for DoutStreambuf.h to include just that structure, without
pulling in all of config.h.

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

Remove cdout

cdout just duplicates the functionality of dout.

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

lockdep: add lockdep_register_ceph_context,cleanup

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

signal: const cleanup, don't use derr

In signal.cc: const cleanup, don't use derr, add a comment.

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

rgw: put_obj_data uses aio

This is just a trivial fix, making the etag calculation done while the io
is pending. Still some room for improvement.

rgw: put_obj_data doesn't need mtime

filestore: fix fiemap

rbd: fix read_iterate for sparse read

qa: tiobench workunit: remove timeout

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

Merge remote branch 'origin/stable'

librados: return value on non void function

common: fix descriptinos for -i, -n too

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

update clitests with new usage

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

common: fix generic usage for -D vs -d

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

rgw: suspend/enable buckets through pool async api

librados: add async pool op operations

qa: organize workunits

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

secret.c: use safe_read when appropriate

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

mds: fix use-after-free on ESession replay

Reported by: Sam Lang <samlang@gmail.com>
Signed-off-by: Sage Weil <sage.weil@dreamhost.com>

qa: make snap-rm-diff workunit clean up

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

qa: wget -q

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

rgw: remove stray semicolon

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

rgw: cast RGW_SUSPENDED_USER_AUID

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

mds: CDir::log_mark_dirty needs to flush before waiting

We need to flush before waiting or else our completion (and mark_diryt())
will be out of order with respect to other dir updates.

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

mount.ceph: specify full path for /sbin/modprobe

Fixes: #1188
Signed-off-by: Sage Weil <sage.weil@dreamhost.com>

radosgw_admin: fix cli test

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

Merge branch 'stable'

osd: fix find_object_context debug output

Erroneously said get_object_context, confusing developers everywhere.

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

osd: fix snap_trimmer obc ref

As of fc4cc39 new_repop claims the obc reference (from the clonerange
branch) but this caller didn't get fixed.

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

mds: when prodding head inode on snapped behalf, avoid XSYN

If we try to rdlock a snapped inode in snap->sync state, we need to prod
the head inode into SYNC to get the data flushed.  Make sure we avoid
XSYN by passing as_anon=true to rdlock_start(head).

Also fix the waiting code.  If it succeeds, continue around the loop.  If
not, the rdlock_start will add the waiter for us--don't do it again!

Fixes workunit snaps/snaptest-snap-rm-cmp.sh.

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

mds: fix transitions away from XSYN state

If we are in XSYN, we need to transition first to EXCL before moving on.
Simply going to the other unstable state isn't enough, because if gather
is zero at the end of the method we end up jumping directly to the wrong
state without passing EXCL.   Instead, call file_excl(), and continue only
if we got there immediately.

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