mds: time out reconnect

kclient: another dumb typo

kclient: fix osdc kick_requests spinlocking.  and handle copy_to/from_user faults

mds: typo

mds: normalize _xlock_finish

kclient: create -> mknod sets S_IFREG.  debug mode.

mds: wake rdlock waiters on xlock_finish

kclient: fix up mdsc spinlocking a bit

mds: fix can_rdlock_soon vs xlock on FileLock

readme

kclient: use d_splice_alias in prepopulate

kclient: create fall back to mknod if no nameidata (as with nfs-kernel-server)

kclient: drop nameidata printk

kclient: remove unnecessary flush/truncate bits from flush_write_caps

kclient: drop r_old_dentry ref in put_request

kclient: remove weird igrab/iput pair in unlink

kclient: slight cleanup of do_lookup ENOENT handler

kclient: fill_trace now uses d_splice_alias

kclient: fix dentry ref leak in dn lease revocation

mds: stop gratuitously issuing dn leases

kclient: simplify unlink

kclient: clean out cap_delay_list on umount

kclient: sync stop osdc timeout timer on shutdown

kclient: fix page redirtying when writeback is canceled

kclient: osdc cancel timeout work _a_sync, to avoid deadlock

introduce time_warp_seq to EXCL utimes() + cap release vs file_cap/fill_inode races

kclient: fix page revocation for osdc reads

kclient: kmap in messenger read path

kclient: unwind lock on enomem

kclient: fix missed pages for redirty in writepages

mds: fix client session lru

mds: retry client messages indefinitely, since mds has its own timeout/markdown mechanism

todos

kclient: fix message skipping

kclient: handle revoked pages in read msg

kclient: clean up mtime/size attribute updates in fill_inode

kclient: turn down dirty page debug output

kclient: fixed up dirty page accounting.  hopefully.

kclient: fix leaking page refs in writepages?

kclient: dirty page accounting hell

disable ccgnu2 (for atomic_t) by default, as it seems to be buggy!

kclient: trivially clean up cap_trunc

kclient: more truncate tomfoolery

kclient: use vmtruncate; do cap trunc via wq to avoid i_mutex deadlock

kclient: only truncate fwd with EXCL; normal truncations require cleaning up of old objects etc. by the mds...

utime from float conversion fix

fakestore: fix error values

kclient: munging up umount a bit, not real change tho

mds: fix stale cap revoke/resume

kclient: check caps outside mdsc spinlock

kclient: improved dirty page accounting (supports mmap writes)

kclient: mucking with dirty page accounting

mds: check max size on truncate, in case we truncate forward

kclient: ping mds a bit more aggressively when we hit max_size

kclient: show outgoing msg seq #s

kclient: zero leftovers for partial page write

client: include lease msgs in mds session seq, so that umount is clean

client: cleaned up mds sessions a bit

fix cap stale/resume in mds, kclient.

kclient: misc cleanup

kclient: small mdsc cleanup

kclient: removed spurious unlock in mdsc forward handler

kclient: quiet down on unmount

kclient: put delayed caps on single queue, use existing mdsc delayed work handler

kclient: fixed some bugs in messenger shutdown

mds: fix hang on FileLock transition to LOCK

kclient: drop s_mutex in __ceph_mdsc_send_cap

we want to avoid ever touching the delayed work whie holding s_mutex:

[10124.715227] =======================================================
[10124.718705] [ INFO: possible circular locking dependency detected ]
[10124.718705] 2.6.25 #18
[10124.718705] -------------------------------------------------------
[10124.718705] events/1/10 is trying to acquire lock:
[10124.718705]  (&s->s_mutex){--..}, at: [<ffffffff880034a7>] ceph_check_caps+0x458/0x4f8 [ceph]
[10124.718705]
[10124.718705] but task is already holding lock:
[10124.718705]  (&(&ci->i_cap_dwork)->work){--..}, at: [<ffffffff80242fa7>] run_workqueue+0x9f/0x1f6
[10124.718705]
[10124.718705] which lock already depends on the new lock.
[10124.718705]
[10124.718705]
[10124.718705] the existing dependency chain (in reverse order) is:
[10124.718705]
[10124.718705] -> #1 (&(&ci->i_cap_dwork)->work){--..}:
[10124.718706]        [<ffffffff80251573>] __lock_acquire+0xa8b/0xc8a
[10124.718706]        [<ffffffff80251800>] lock_acquire+0x8e/0xb2
[10124.718706]        [<ffffffff802437af>] __cancel_work_timer+0xe8/0x1ea
[10124.718706]        [<ffffffff802438be>] cancel_delayed_work_sync+0xd/0xf
[10124.718706]        [<ffffffff8800ee26>] __ceph_mdsc_send_cap+0x247/0x264 [ceph]
[10124.718706]        [<ffffffff880034cd>] ceph_check_caps+0x47e/0x4f8 [ceph]
[10124.718706]        [<ffffffff88003610>] ceph_put_wrbuffer_cap_refs+0xc9/0xd2 [ceph]
[10124.718706]        [<ffffffff880090cc>] ceph_writepages+0x782/0x97a [ceph]
[10124.718706]        [<ffffffff8026d3f5>] do_writepages+0x2b/0x3a
[10124.718706]        [<ffffffff802aa8aa>] __writeback_single_inode+0x151/0x282
[10124.718706]        [<ffffffff802aade4>] sync_sb_inodes+0x1ab/0x26f
[10124.718706]        [<ffffffff802ab0cf>] writeback_inodes+0x85/0xe9
[10124.718706]        [<ffffffff8026dc9e>] wb_kupdate+0x9f/0x10d
[10124.718706]        [<ffffffff8026e1b9>] pdflush+0x134/0x1df
[10124.718706]        [<ffffffff8024663c>] kthread+0x49/0x79
[10124.718706]        [<ffffffff8020cd38>] child_rip+0xa/0x12
[10124.718706]        [<ffffffffffffffff>] 0xffffffffffffffff
[10124.718706]
[10124.718706] -> #0 (&s->s_mutex){--..}:
[10124.718706]        [<ffffffff80251475>] __lock_acquire+0x98d/0xc8a
[10124.718706]        [<ffffffff80251800>] lock_acquire+0x8e/0xb2
[10124.718706]        [<ffffffff8054f9aa>] mutex_lock_nested+0xed/0x273
[10124.718706]        [<ffffffff880034a7>] ceph_check_caps+0x458/0x4f8 [ceph]
[10124.718706]        [<ffffffff8800393e>] ceph_cap_delayed_work+0x101/0x14a [ceph]
[10124.718706]        [<ffffffff80242ff6>] run_workqueue+0xee/0x1f6
[10124.718706]        [<ffffffff80243b72>] worker_thread+0xdb/0xe8
[10124.718706]        [<ffffffff8024663c>] kthread+0x49/0x79
[10124.718706]        [<ffffffff8020cd38>] child_rip+0xa/0x12
[10124.718706]        [<ffffffffffffffff>] 0xffffffffffffffff
[10124.718706]
[10124.718706] other info that might help us debug this:
[10124.718706]
[10124.718706] 2 locks held by events/1/10:
[10124.718706]  #0:  (events){--..}, at: [<ffffffff80242fa7>] run_workqueue+0x9f/0x1f6
[10124.718706]  #1:  (&(&ci->i_cap_dwork)->work){--..}, at: [<ffffffff80242fa7>] run_workqueue+0x9f/0x1f6
[10124.718706]
[10124.718706] stack backtrace:
[10124.718706] Pid: 10, comm: events/1 Not tainted 2.6.25 #18
[10124.718706]
[10124.718706] Call Trace:
[10124.718706]  [<ffffffff8024f3ea>] print_circular_bug_tail+0x70/0x7b
[10124.718706]  [<ffffffff80251475>] __lock_acquire+0x98d/0xc8a
[10124.718706]  [<ffffffff80550a0f>] ? trace_hardirqs_on_thunk+0x35/0x3a
[10124.718706]  [<ffffffff8021287c>] ? native_sched_clock+0x4a/0x66
[10124.718706]  [<ffffffff880034a7>] ? :ceph:ceph_check_caps+0x458/0x4f8
[10124.718706]  [<ffffffff80251800>] lock_acquire+0x8e/0xb2
[10124.718706]  [<ffffffff880034a7>] ? :ceph:ceph_check_caps+0x458/0x4f8
[10124.718706]  [<ffffffff8054f9aa>] mutex_lock_nested+0xed/0x273
[10124.718706]  [<ffffffff880034a7>] ? :ceph:ceph_check_caps+0x458/0x4f8
[10124.718706]  [<ffffffff880034a7>] :ceph:ceph_check_caps+0x458/0x4f8
[10124.718706]  [<ffffffff8800383d>] ? :ceph:ceph_cap_delayed_work+0x0/0x14a
[10124.718706]  [<ffffffff8800393e>] :ceph:ceph_cap_delayed_work+0x101/0x14a
[10124.718706]  [<ffffffff80242ff6>] run_workqueue+0xee/0x1f6
[10124.718706]  [<ffffffff80243b72>] worker_thread+0xdb/0xe8
[10124.718706]  [<ffffffff8024675a>] ? autoremove_wake_function+0x0/0x38
[10124.718706]  [<ffffffff80243a97>] ? worker_thread+0x0/0xe8
[10124.718706]  [<ffffffff8024663c>] kthread+0x49/0x79
[10124.718706]  [<ffffffff8020cd38>] child_rip+0xa/0x12
[10124.718706]  [<ffffffff8020c44f>] ? restore_args+0x0/0x30
[10124.718706]  [<ffffffff802465f3>] ? kthread+0x0/0x79
[10124.718706]  [<ffffffff8020cd2e>] ? child_rip+0x0/0x12
[10124.718706]
[10175.822587] ceph_super: kill_sb ffff81010da60000

kclient: check for corrupt osd_op_reply

todos

kclient: typo

kclient: take spinlock while processing acks in msgr

increase default pg_bits

kclient: fix double mutex unlock in check_caps

automagic endian conversions for userland

final type munging in ceph_fs.h

fix up mds reply types

kclient: tear down /proc/fs/ceph on exit

type hell

kclient: switch mds sessions s_mutex to struct mutex

kclient: merged client.c and super.c

kclient: fix reset_connection

kclient: remove delay ack hack in msgr

kclient: use MSG_MORE flag appropriately with sendmsg

kclient: oops, be quiet

kclient: fixed various msgr bugs on incoming connections

kclient: fix mds reconnect for caps on dentryless inodes

simplify mds session caps stale/resume

kclient: deal with osdc abort vs kick races; revoke pages on msg dup

kclient: some msgr cleanups

kclient: msgr shutdown fix, warn on enomem

kclient: disable sysfs for < 2.6.25

kclient: beginnings of sysfs

kclient: cancel work while shutting down

kclient: osdc tid output

kclient: proc entry for debug_console

kclient: fixed do_lookup dentry refcount problem

kclient: fixed up page revocation

kclient: only do footer with data payloads

add message footer.  kclient safely revokes pages away from sending ceph_msg

synclient: fix size print