]>
git.apps.os.sepia.ceph.com Git - ceph.git/log
Sage Weil [Thu, 1 Oct 2009 20:11:45 +0000 (13:11 -0700)]
kclient: sync_page_range gone in 2.6.32; use vfs_fsync_range
This is what generic_write_sync does, but we also want to do the
sync if we are NEARFULL.
Sage Weil [Thu, 1 Oct 2009 20:05:36 +0000 (13:05 -0700)]
kclient: current_is_pdflush() gone in 2.6.32
Sage Weil [Thu, 1 Oct 2009 19:57:39 +0000 (12:57 -0700)]
todo
Sage Weil [Thu, 1 Oct 2009 19:57:17 +0000 (12:57 -0700)]
Revert "kclient: add smp_mb to sensitive bitops in msgr"
Don't think these are needed after all...
This reverts commit
fe11285c4b00fb9148251cfaaf28aa71a70b31c8 .
Sage Weil [Thu, 1 Oct 2009 18:21:54 +0000 (11:21 -0700)]
kclient: use smp_mb in place of spinlock for testing flag bit
Sage Weil [Thu, 1 Oct 2009 18:07:09 +0000 (11:07 -0700)]
kclient: add smp_mb to sensitive bitops in msgr
Sage Weil [Thu, 1 Oct 2009 17:59:10 +0000 (10:59 -0700)]
uclient: set new file layout to 0, which lets mds choose its default
Greg Farnum [Wed, 30 Sep 2009 17:58:49 +0000 (10:58 -0700)]
uclient: Use much saner/not insane defaults for layout settings
Sage Weil [Wed, 30 Sep 2009 21:10:46 +0000 (14:10 -0700)]
kclient: debugfs fix; better osdmap debug output
Sage Weil [Wed, 30 Sep 2009 21:10:23 +0000 (14:10 -0700)]
kclient: default pg_preferred to -1
Sage Weil [Wed, 30 Sep 2009 20:30:07 +0000 (13:30 -0700)]
mds: fix layout typo
Sage Weil [Wed, 30 Sep 2009 19:53:49 +0000 (12:53 -0700)]
kclient: include new files in patch set
Sage Weil [Wed, 30 Sep 2009 19:47:40 +0000 (12:47 -0700)]
osd: name args in ceph_osd_op union
Sage Weil [Wed, 30 Sep 2009 19:34:30 +0000 (12:34 -0700)]
document data structs; kill ttl_ms in caps msg
Sage Weil [Wed, 30 Sep 2009 19:24:31 +0000 (12:24 -0700)]
mds: kill old rdcaps_ttl business
This was from when caps independently timed out.
Sage Weil [Wed, 30 Sep 2009 19:09:11 +0000 (12:09 -0700)]
fix object_stripe_unit -> stripe_unit
Sage Weil [Wed, 30 Sep 2009 18:43:31 +0000 (11:43 -0700)]
kclient: try kmalloc, then vmalloc
"A still-lame-but-less-lame option here would be to attempt the kmalloc
(with __GFP_NOWARN) and if it failed, fall back to vmalloc." says akpm
Sage Weil [Wed, 30 Sep 2009 18:32:03 +0000 (11:32 -0700)]
kclient: audit inline
Sage Weil [Wed, 30 Sep 2009 18:01:37 +0000 (11:01 -0700)]
kclient: more fixes from akpm's review
Sage Weil [Wed, 30 Sep 2009 17:24:28 +0000 (10:24 -0700)]
uninline some header types
Sage Weil [Tue, 29 Sep 2009 21:53:07 +0000 (14:53 -0700)]
kclient: parse updated mdsmap
Sage Weil [Tue, 29 Sep 2009 21:51:43 +0000 (14:51 -0700)]
mdsmap: another unsized int
Sage Weil [Tue, 29 Sep 2009 21:38:10 +0000 (14:38 -0700)]
mdsmap: use sized int type
Sage Weil [Tue, 29 Sep 2009 20:29:07 +0000 (13:29 -0700)]
todo
Sage Weil [Fri, 25 Sep 2009 23:04:13 +0000 (16:04 -0700)]
kclient: reorder cap ops a bit (wrt invalidation)
Greg Farnum [Mon, 28 Sep 2009 23:54:58 +0000 (16:54 -0700)]
Hadoop: Set the local osd to be the preferred osd for all writes.
Greg Farnum [Tue, 29 Sep 2009 19:16:55 +0000 (12:16 -0700)]
uclient: Add methods for setting a default preferred osd.
Greg Farnum [Tue, 29 Sep 2009 19:08:52 +0000 (12:08 -0700)]
mds/ceph_fs: Add a preferred argument to ceph_mds_request_args and listen to it
Sage Weil [Tue, 29 Sep 2009 18:07:37 +0000 (11:07 -0700)]
mon: create lpg_num (not pg_num) lpg's
Sage Weil [Mon, 28 Sep 2009 22:28:16 +0000 (15:28 -0700)]
cosd: 3 monitors
Sage Weil [Tue, 29 Sep 2009 16:49:35 +0000 (09:49 -0700)]
uclient: open mds sessions to export targets
Sage Weil [Tue, 29 Sep 2009 04:16:02 +0000 (21:16 -0700)]
mds: add export targest to mds_info_t
Sage Weil [Mon, 28 Sep 2009 22:23:56 +0000 (15:23 -0700)]
osd: rev disk format
Sage Weil [Mon, 28 Sep 2009 22:23:01 +0000 (15:23 -0700)]
mon: create localized pgs
Sage Weil [Mon, 28 Sep 2009 22:22:42 +0000 (15:22 -0700)]
filestore: make collection dir have better name; move coll_t type around
Sage Weil [Mon, 28 Sep 2009 22:21:23 +0000 (15:21 -0700)]
objecter: send renew sub from maybe_request_map()
This broke with the last monc api change
Sage Weil [Mon, 28 Sep 2009 21:11:47 +0000 (14:11 -0700)]
osd: make pg_pool_t print clearer
Greg Farnum [Mon, 28 Sep 2009 20:55:03 +0000 (13:55 -0700)]
Merge branch 'unstable' of ceph.newdream.net:/git/ceph into unstable
Greg Farnum [Mon, 28 Sep 2009 20:54:55 +0000 (13:54 -0700)]
uclient: Return EFBIG rather than EBADF on too-large file sizes.
Sage Weil [Fri, 25 Sep 2009 23:26:34 +0000 (16:26 -0700)]
kclient: bug on bad msg in msgr
Sage Weil [Fri, 25 Sep 2009 18:19:08 +0000 (11:19 -0700)]
filestore: use btrfs TRANS_RESV_START ioctl if present
This will detect ENOSPC when starting the transaction, instead of
failing part way through.
Sage Weil [Fri, 25 Sep 2009 18:18:17 +0000 (11:18 -0700)]
osd: make shutdown msg higher priority
This makes cosd more responsive to SIGTERM.
Sage Weil [Mon, 28 Sep 2009 20:47:50 +0000 (13:47 -0700)]
msgr: be a bit more verbose during shutdown
At least until this bug is found:
09.09.28 15:16:54.678277
140100507507024 -- 10.10.10.92:6800/5504/0 >> 10.10.10.88:6789/0/0 pipe(0x186ae90 sd=61 pgs=0 cs=0 ltx=1).accept replacing existing (lossy) channel
msg/SimpleMessenger.cc: In function 'void SimpleMessenger::reaper()':
msg/SimpleMessenger.cc:1925: FAILED assert(lock.is_locked())
1: /usr/bin/cosd(_ZN15SimpleMessenger6reaperEv+0x59d) [0x45493d]
2: /usr/bin/cosd(_ZN15SimpleMessenger4waitEv+0x4d) [0x45574d]
3: /usr/bin/cosd(main+0xfe0) [0x44e770]
4: /lib/libc.so.6(__libc_start_main+0xe6) [0x7f6befc145a6]
5: /usr/bin/cosd(__gxx_personality_v0+0x1b9) [0x44d4e9]
NOTE: a copy of the executable, or `objdump -rdS <executable>` is needed to interpret this.
Sage Weil [Mon, 28 Sep 2009 20:47:22 +0000 (13:47 -0700)]
uclient: get_local_osd() to find an osd with the given ip
Sage Weil [Mon, 28 Sep 2009 17:58:51 +0000 (10:58 -0700)]
kclient: fix generic xattr calls
It compiles now :)
Sage Weil [Mon, 28 Sep 2009 17:56:05 +0000 (10:56 -0700)]
kclient: user.ceph.layout readonly for now
Sage Weil [Mon, 28 Sep 2009 17:09:23 +0000 (10:09 -0700)]
osdmaptool: more usage
Sage Weil [Mon, 28 Sep 2009 17:08:18 +0000 (10:08 -0700)]
config: allow object replication from 1-10 (instead of 2-3)
Sage Weil [Mon, 28 Sep 2009 17:06:48 +0000 (10:06 -0700)]
kclient: check xattr readonly flag on setxattr, removexattr
Sage Weil [Mon, 28 Sep 2009 16:54:54 +0000 (09:54 -0700)]
kclient: add helper to set allowed xattr prefixes
Sage Weil [Mon, 28 Sep 2009 16:37:28 +0000 (09:37 -0700)]
kclient: distinguish file vs dir xattrs; move the getvxattr_cb under i_lock
Sage Weil [Fri, 25 Sep 2009 22:58:16 +0000 (15:58 -0700)]
kclient: do not recurse into truncate|invalidate pages
A single open O_TRUNC on a file dirty pages would
-> truncate
-> put_wrbuffer_refs
-> check_caps
-> invalidate ... and deadlock
So, do not invalidate here if we have a pending truncate (which
remains elevated until after the truncate finishes).
Greg Farnum [Fri, 25 Sep 2009 22:55:04 +0000 (15:55 -0700)]
uclient: Pass in the right variables to update_inode_file_bits!
Greg Farnum [Fri, 25 Sep 2009 22:54:18 +0000 (15:54 -0700)]
uclient: Rework update_inode_file_bits to mimic kclient; better debugging
Greg Farnum [Thu, 24 Sep 2009 23:16:15 +0000 (16:16 -0700)]
uclient: Pass uid/gid from ll_create to _mknod
Add some debugging outputs
Greg Farnum [Thu, 24 Sep 2009 21:21:45 +0000 (14:21 -0700)]
uclient/filepath: Update filepath's empty() function and switch make_request_from_Meta to use that
Greg Farnum [Thu, 24 Sep 2009 18:48:01 +0000 (11:48 -0700)]
fuse: have FUSE enforce permissions
Greg Farnum [Thu, 24 Sep 2009 18:47:34 +0000 (11:47 -0700)]
uclient: Update ctime one setattr changes
Greg Farnum [Thu, 24 Sep 2009 17:19:15 +0000 (10:19 -0700)]
uclient/objecter: Add max_file_size checks and an accessor for max_file_size
Greg Farnum [Wed, 23 Sep 2009 21:58:34 +0000 (14:58 -0700)]
uclient: make_request_from_Meta will now create the path if none is given
Since this is currently unused, it should possibly be removed
since it introduces at least one extra if into the program flow.
Greg Farnum [Wed, 23 Sep 2009 19:03:50 +0000 (12:03 -0700)]
uclient: Add cap dropping to setattr.
Sage Weil [Fri, 25 Sep 2009 22:29:59 +0000 (15:29 -0700)]
mds: update mtime on truncation!
This didn't affect the kclient before for some reason, but it does
get the uclient. Fix it.
Sage Weil [Fri, 25 Sep 2009 21:51:22 +0000 (14:51 -0700)]
conf: don't run over g_conf options when setting them
Sage Weil [Fri, 25 Sep 2009 16:07:58 +0000 (09:07 -0700)]
configure: kill more commas
CentOS sucks!
Sage Weil [Fri, 25 Sep 2009 16:06:30 +0000 (09:06 -0700)]
configure: no commas?
Sage Weil [Fri, 25 Sep 2009 15:45:32 +0000 (08:45 -0700)]
configure: whitespace
Sage Weil [Fri, 25 Sep 2009 15:43:03 +0000 (08:43 -0700)]
configure: fix libexpat1, libfcgi checks (again)
Sage Weil [Thu, 24 Sep 2009 21:57:27 +0000 (14:57 -0700)]
mds: make the directory commit priority be low(er than journal)
I'd raise the journal priority to > default, but that means touching
more Filer.h etc code.
Sage Weil [Thu, 24 Sep 2009 21:56:39 +0000 (14:56 -0700)]
objecter: optionally set priority on ObjectOperations
This just sets the priority on the MOSDOp we send.
Sage Weil [Thu, 24 Sep 2009 21:10:57 +0000 (14:10 -0700)]
mds: do not attach mdr to NULL session
Sage Weil [Wed, 23 Sep 2009 17:11:56 +0000 (10:11 -0700)]
todo
Sage Weil [Wed, 23 Sep 2009 17:00:32 +0000 (10:00 -0700)]
mds: warn on dup inodes; don't crash
We just ignore the second instance of the inode. It can get
cleaned up transparently if/when the directory gets saved.
Sage Weil [Tue, 22 Sep 2009 22:12:18 +0000 (15:12 -0700)]
objecter: remove dead code
Sage Weil [Tue, 22 Sep 2009 22:11:47 +0000 (15:11 -0700)]
librados: pass resets to objecter, with proper locking
Sage Weil [Tue, 22 Sep 2009 22:11:34 +0000 (15:11 -0700)]
mds: pass resets to objecter, with proper locking
Sage Weil [Tue, 22 Sep 2009 22:11:25 +0000 (15:11 -0700)]
uclient: pass reset events to objecter, with proper locking
Sage Weil [Tue, 22 Sep 2009 22:11:10 +0000 (15:11 -0700)]
objecter: resend mon ops on handle mon connection reset
Sage Weil [Tue, 22 Sep 2009 22:37:51 +0000 (15:37 -0700)]
mds: fix inode load existence check
We can lookup using last snapid, not first. That's how the
inode_map is indexed.
Sage Weil [Tue, 22 Sep 2009 21:18:47 +0000 (14:18 -0700)]
todo
Sage Weil [Tue, 22 Sep 2009 21:17:50 +0000 (14:17 -0700)]
uclient: don't readdir first frag on opendir
This reorders things such that rm -r breaks. It was an old old
optimization that doesn't make sense anymore anyway.
Sage Weil [Tue, 22 Sep 2009 20:17:03 +0000 (13:17 -0700)]
objecter: fix uninitialized variable last_seen_version
Fixes hang on 'rados df'
Sage Weil [Tue, 22 Sep 2009 20:09:09 +0000 (13:09 -0700)]
librados: fix dispatch
Greg Farnum [Tue, 22 Sep 2009 00:27:59 +0000 (17:27 -0700)]
uclient: Better fix for insert_dentry_inode to work with pre-built Dentrys.
Greg Farnum [Tue, 22 Sep 2009 00:07:20 +0000 (17:07 -0700)]
uclient: get_or_create should return out-of-date lease Dentrys too
Greg Farnum [Tue, 22 Sep 2009 00:06:14 +0000 (17:06 -0700)]
uclient: encode_cap_releases shouldn't update caps if inode isn't in the cache
Greg Farnum [Tue, 22 Sep 2009 00:04:32 +0000 (17:04 -0700)]
uclient: First attempt at preemptive cache dropping in the functions.
Greg Farnum [Mon, 21 Sep 2009 21:13:19 +0000 (14:13 -0700)]
uclient: Don't *always* set_dentry_wanted() on MClientRequests.
Greg Farnum [Mon, 21 Sep 2009 20:46:37 +0000 (13:46 -0700)]
uclient: Only make the MClientRequest in send_request itself.
Add a few more functions to MetaRequest so we can use those in choose_mds.
Greg Farnum [Mon, 21 Sep 2009 20:11:52 +0000 (13:11 -0700)]
uclient: More debugging. May want to back off the inode printout to
not include caps_issued.
Greg Farnum [Mon, 21 Sep 2009 18:13:55 +0000 (11:13 -0700)]
uclient: changed link and insert_dentry_inode to behave with pre-built dentrys
Greg Farnum [Fri, 18 Sep 2009 22:44:45 +0000 (15:44 -0700)]
uclient: Attempt preemptive cap dropping on mknod.
Greg Farnum [Fri, 18 Sep 2009 21:21:58 +0000 (14:21 -0700)]
uclient: dout statements in new functions.
Greg Farnum [Fri, 18 Sep 2009 22:20:29 +0000 (15:20 -0700)]
uclient: Fixes for get_or_create
Greg Farnum [Fri, 18 Sep 2009 20:51:26 +0000 (13:51 -0700)]
uclient: Functions to create Dentrys prior to gettin caps from mds.
Sage Weil [Tue, 22 Sep 2009 19:29:33 +0000 (12:29 -0700)]
remove extra whitespace
Sage Weil [Tue, 22 Sep 2009 19:26:52 +0000 (12:26 -0700)]
kclient: fix bit test logic
This fixes the fault messages with <NULL> in them; we now print
the proper message, and in a timely fashion.
Sage Weil [Tue, 22 Sep 2009 19:15:33 +0000 (12:15 -0700)]
kclient: use pr_fmt to prefix module name
This doubles up the ceph:ceph: if we have CONFIG_DYNAMIC_DEBUG, but
that's not our fault.
Sage Weil [Tue, 22 Sep 2009 18:50:29 +0000 (11:50 -0700)]
kclient: prefix frag_ functions with ceph_
It was suggested this would avoid any potential conflict with TCP.
It's a good idea to keep the namespace clean anyway.
Sage Weil [Tue, 22 Sep 2009 18:27:32 +0000 (11:27 -0700)]
kclient: kill out_qlen
This is apparently buggy (not immediately obvious why) but it's
only really useful for debugging anyway (it makes it easier to
match up message sequence numbers). Rip it out.
Sage Weil [Tue, 22 Sep 2009 18:17:22 +0000 (11:17 -0700)]
make inline string functions non-inline