rbdtool: use canonical image name (pool/image)

rbdtool: can read data using snapshot

mds: fix up snap cap issue logic comments

mds: fix acquire_lock() calls in handle_client_open()

We want to acquire_locks exactly once per path through the function.
Don't look at mdr->* yet if we haven't locked yet.

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

mds: issue all snap_caps allowed given locks, not by current cap

For snap_caps, who cares if the live inode has a cap for this client.  We
always want to issue the maximum set of caps the inode locks allow for
a live inode or snapped inode with caps, or the max caps ever for a snapped
inode without caps.

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

fix FCGI detection and specfile

configure is building radosgw as soon as libfcgi is installed, even
if --without-radosgw is given. Here is a fix for configure.in and the
specfile.

Regards,
Christian

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

mon: avoid OOB m->cmd access in auth proprocess_command

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

msgr: brackets (not braces) for ipv6 addrs

Sigh

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

debug: only truncate log file on open if log_per_instance is set

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

mds: termiante client sessions before flushign log during shutdown

We need to close out client sessions, and caps, to empty the open_files
lists, so that we can properly trim the EOpen events from the log.

This fixes a crash like so:

mds/journal.cc: In function 'C_Gather* LogSegment::try_to_expire(MDS*)':
mds/journal.cc:160: FAILED assert(ls != this)
 1: (LogSegment::try_to_expire(MDS*)+0x11c1) [0x90a1df]
 2: (MDLog::try_expire(LogSegment*)+0x2d) [0x902c25]
 3: (MDLog::trim(int)+0x55b) [0x902b0b]
 4: (MDCache::shutdown_pass()+0xeb) [0x7d6aab]
 5: (MDS::_dispatch(Message*)+0x2dc1) [0x71597f]
 6: (MDS::ms_dispatch(Message*)+0x38) [0x712a44]
 7: (Messenger::ms_deliver_dispatch(Message*)+0x63) [0x6fe07b]
 8: (SimpleMessenger::dispatch_entry()+0x5fd) [0x6ef7e7]
 9: (SimpleMessenger::DispatchThread::entry()+0x2c) [0x6e44b8]
 10: (Thread::_entry_func(void*)+0x23) [0x6fd075]
 11: /lib64/libpthread.so.0() [0x3552e06a3a]
 12: (clone()+0x6d) [0x35526ddf3d]
 NOTE: a copy of the executable, or `objdump -rdS <executable>` is
needed to interpret this.

Reported-by: Henry C Chang <henry_c_chang@tcloudcomputing.com>
Signed-off-by: Sage Weil <sage@newdream.net>

ceph.spec.in: create /var/log/ceph directory

I've missed this one in my previous patch series.

Regards,
Christian

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

ceph.spec.in: add a build option for radosgw (--with radosgw)

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

ceph.spec.in: add /var/lib/ceph/

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

ceph.spec.in: add logrotate configuration

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

ceph.spec.in: set correct path for rbd classes

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

osd: print debug write before adding to transaction

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

msgr: return error and dethrottle if message decode fails

This was broken by the addition of the throttling stuff.

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

debian: make helper scripts remember most recent release

So you don't have to cut&paste it on the command line

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

osd: fix delayed pg creation use after free

Use local acting[], not the one stashed in creating_pgs[pgid].  And drop
the duplication .erase().

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

vstart: specify osd class tmp dir

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

monc: make hunt, ping rates configuration

3s for hunt rate, 10s for ping rate.

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

client: increase client mount timeout

This needs to be longer than the monclient hunt rate (currently 10s) so
that we succeed if we initially pick a down monitor.

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

filestore: improve lseek64 error checking, and use buffer::list::write_fd()

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

buffer: make write_fd() const

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

mount.ceph: handle ipv6 addrs, colons, braces

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

ceph.spec: require libedit

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

ceph.spec.in: require binutils

cclass needs /usr/bin/nm

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

debian: require binutils

cclass needs 'nm' to extract symbols from rados classes.

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

debian: explicit postinst/postrm for shlibs

The dh_shlibdeps sees the .so in 'ceph' and incorrectly adds ldconfig to
postinst and postrm.  Which is useless and upsets lintian.  So do our
postinst/postrm magic explicitly.

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

rbdtool: fix export

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

mon: create new pools using rjenkins hash

Otherwise they get -1, which always maps to an invalid (-1) pgid.

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

rbdtool: should support files larger than 2GB

debian: _do_ strip cosd

class: make class tmp directory configurable, not /tmp

/tmp usually gets mounted with 'noexec', making it a poor choice.

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

logger: shut up

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

config: can use '-' as a config option

msgr: remove fake orig_source_addr fakery

This makes get_orig_*() match get_*().

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

mon: fix message forwarding to use Connection to fake source addr

This make get_source_addr() work (not just get_orig_source_addr()).

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

debian: do not strip cosd

classhandler: print error when dlopen fails

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

ceph: drop version from ceph_fs.h, use autoconf VERSION macro instead

There's no reason to define this in the code anymore.  Since the kernel
side doesn't use it it just confuses matters.

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

shell script fixes

mds: fix logger init on startup, rank symlink creation

Busted by 83d1ea6636dd432dcbb6a0c6046d551bee7be5c6.

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

debian: do not strip rados classes

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

debian: add postinst, postrm for ceph package

This will keep debain from using the default, which thinks it needs to run
ldconfig.  Which it doesn't.

Makefile: cleanup

debian: put rbd classes in ceph, not librados1 package

The cluster classes are a server-side thing, not client lib thing.

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

osd: consolidate osd stats in _refresh_my_stats

Also keep local copy of osd_stat (send with pg stats update to monitor).

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

debian: include rados-classes

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

cclass: $libdir/rados-classes

logger: huge rewrite, cleanup

- name *.log
- name after name, not rank
- reopen on sighup
- remove pointer on shutdown
- clean up interface
- implement reset (truncate/reopen file)
- rotate logs

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

auth: rename supported_auth -> auth_supported

It should be 'auth supported' for naming consistency.

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

client: handle_client_session now careful opening mds sessions while unmounting

client: put MDSSession pointers in the msd_session map, rather than the sessions themselves.
This is a bit cleaner and clears the way to make it more closely mimic the kernel.

paxos: send summary if observer has fallen behind first_committed

This avoids a ceph crash like

mon/PGMap.h: In function 'void PGMap::apply_incremental(PGMap::Incremental&)':
mon/PGMap.h:77: FAILED assert(inc.version == version+1)
 1: (PGMap::apply_incremental(PGMap::Incremental&)+0x50) [0x5a5f22]
 2: (handle_notify(MMonObserveNotify*)+0x2c7) [0x567c53]
 3: (Admin::ms_dispatch(Message*)+0x52) [0x5aa042]
 4: (Messenger::ms_deliver_dispatch(Message*)+0x55) [0x5be389]
 5: (SimpleMessenger::dispatch_entry()+0x50f) [0x5ad3fb]
 6: (SimpleMessenger::DispatchThread::entry()+0x29) [0x597323]
 7: (Thread::_entry_func(void*)+0x20) [0x5b91c8]
 8: /lib/libpthread.so.0 [0x7f31185fd73a]
 9: (clone()+0x6d) [0x7f311782769d]

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

msgr: immediately drop messages with null dest

This avoids spawning a new pipe and then failing to connect to a bad
address.

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

mds: fix handle_dentry_unlink to handle remote unlink (no straydn)

The straydn is optional, and NULL if we're unlinking a remote dentry.

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

mds: be less noisy about memory usage in log

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

ceph: --in-file, not --in-data

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

osd: track pg_num by type in logger

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

librados: add pool set_snap_context

Set the snap context for a given pool handle.

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

client: take the first inode as root, rather than whatever comes out
with ino=1.

Also, translate root inode properly in fuse_ll.

ceph.spec: install /etc/logrotate.d/ceph

osd: document some of the logger stats

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

rbdtool: modify usage

config: allow optional parameters

mds: do not include snapped inodes in EOpen or LogSegment::open_files

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

mds: print useful error on malformed EOpen event

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

Makefile: drop unnecessary dist-hook for make_version

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

debian: simplify release.sh

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

Makefile: distribute check_version

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

radosgw: fix usage syntax

That fixes bug #256.

mount.ceph: ignore noauto, _netdev options

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

debian: include libcls_rbd.so.*

Signed-off-by: Fred Ar <ar.fred@yahoo.com>
Signed-off-by: Sage Weil <sage@newdream.net>

cclsinfo: fixes for dash

Signed-off-by: Fred Ar <ar.fred@yahoo.com>
Signed-off-by: Sage Weil <sage@newdream.net>

ceph.spec.in: make -j and unpackaged files

* now that make -j<numprocs> works out of the box, add it to the spec
* added new unpackaged files
* the test binaries get installed with "--without-debug". omitting
"--without-debug" fixes the problem.

Signed-off-by: Thomas Mueller <thomas@chaschperli.ch>
Signed-off-by: Sage Weil <sage@newdream.net>

Makefile: fix up ceph_ver.h autogeneration

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

Makefile: fix build when WITH_DEBUG not set

Signed-off-by: Fred Ar <ar.fred@yahoo.com>
Signed-off-by: Sage Weil <sage@newdream.net>

debug: don't create rank symlink if log-file

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

client: update ctime on setattr(no mask)

This makes pjd happy (chown -1 -1 should bump ctime, apparently).  The
kclient does the same trick: dirty any excl caps we have, otherwise do an
mds setattr(ctime) request.

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

client: return errors from setattr

Otherwise fuse just gets confused because the stat structure isn't filled
in and generates its own EIO.

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

client: fix lookup error codes

The fuse glue was making all errors ENOENT, and _lookup didn't return
ENAMETOOLONG.

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

client: better debug output

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

client: return ENAMETOOLONG where appropriate

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

qa: do untar_snap_rm on two different tarballs (medium and big)

client: fix f_namemax returned by statfs

This was fixed in the kernel last month.

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

Makefile: make more random crap only build with WITH_DEBUG

This should be everything not packaged.

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

rbdtool: implement import

still doesn't use fiemap

client: remove typo

client: synchronize metadata on fsync, when requested.

rbdtool: cleanup

osd: cleanup: move pools can into advance_map()

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

osd: fix null pg deref, put auid in PGPool, cleanup.

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

msgr: add config option to set stacksize of reader&writer threads

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

thread: force stacksize to be multiple of page size; clean up

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

thread: allow setting stacksize when calling create & check return value of pthread_create

Signed-off-by: Paul Chiang <paul_chiang@tcloudcomputing.com>
Signed-off-by: Sage Weil <sage@newdream.net>

update authors

rbdtool: implement export

shift feature supported/required specifications into Policy

This gets it out of the header, and also makes the previous
"my_features & their_features" logic explicit in the Policy definitions.

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

msgr: add set_policy_throttler

This lets us continue to use the helpers.

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