Next round; closer to 1.9.6 release

Hi,

To break the silence, here is a taste of what's been brewing since the
last CVS checkin...

CFS:
- De-ssify unused locks_delete_block().
- Enable gathered sync writes. (#ifdef CFS_GATHERWRITES)
- Fix [ ssic-linux-Bugs-686748 ] filesystem stacking deadlock.
  - BDI write congestion livelock. (#ifdef CFS_FILE_WRITE_LIVELOCK_FIX)
- Regression:
  - stale dentry cache at client. (#ifdef CFS_ZAP_CACHES) (#ifndef
CFS_FILE_RW_NOLOCK)
  - inode revalidation for token REVOKE. (#ifdef CFS_ZAP_CACHES)
    - revert SU token; stale remote write cache. (#ifdef SSI_CFSTOK_SU)
  - export ops unexpected -ESTALE; stale ssidev from NFS client after
SSI cluster reboot. (#ifdef CFS_FH_TO_DENTRY_ANON)
  - dentry memory leak at server. (#ifdef CFS_FH_TO_DENTRY_ANON)
  - invalid pagevec length on SB retry. (#ifdef CFS_PAGEVEC)

IPVS:
- Scrub existing SSI code; style.
- Fix inet socket release traversing clusterwide IPVS delete() code
path for non-registered LVS ports. (#ifdef IPVS_SERVICE_RACE_FIX)
- Various fixes for register_ports()... (#ifdef IPVS_REGISTER_PORTS_FIX)
  - Traversing clusterwide IPVS add() code path for non-registered LVS ports.
  - Adding real server when sockets not listening on CVIP.
  - Setting port weight to 0 did not remove real server from LVS directors.

SSI:
- Fix ssi_wait_event() race missed signal if woken up early. (#ifdef
LINUX_SSI_EVENT)
- Refactor ssi_wait/signal/broadcast_event() with Linux atomic
wait-for-completion. (#ifdef LINUX_SSI_EVENT)
- Reduce contention for ssidev_do_poll/do_select() SMP performance:
  - Lock free ssidev_poll_cli_list. (#ifdef RCU_SSIDEV_POLL_CLI_LIST)
  - Granular locking of ssidev_poll_cli_t structure. (#ifdef
SSIDEV_POLL_CLI_LOCK)
  - Lock free ssidev_poll_svr_list. (#ifdef RCU_SSIDEV_POLL_SVR_LIST)
  - Granular locking of ssidev_poll_svr_t structure. (#ifdef
SSIDEV_POLL_SVR_LOCK)

VPROC:
- Remove redundant "svrnode" in fb_data structure. (#ifdef
REOP_EXPORT_PATH_SVRNODE)
- Comment out unused variable "giveback". (#ifdef VPROC_LIST_SPINLOCK)
- Fix pvpop_add_to_originalchild_list() -EAGAIN (EDEADLK) not handled.
(#ifdef VPROC_RMV_FROM_ORIGINALCHILD_LIST_CANT_FIND_CHILD_FIX)
- New macro VPROC_RELE_PID(). (#ifdef VPROC_RELEASE__REFCNT_RACE_FIX)
- Fix [ ssic-linux-Bugs-2310316 ] Migration sometimes fails:
add_thread_group:Cannot move proc
  - add_thread_group() error -22 process partially shared when not
shared. (#ifdef VPROC_MIGRATE_THREAD_GROUP_FIX)
- PVPOP_REAP() downgrade VPROC LOCK except when code path possibly
__ptrace_unlink. (#ifdef VPROC_RW_LOCK)
- Fix incorrect lock ordering may cause vod_rwlock hang. (#ifdef
ND_ORIGIN_DEADLOCK_FIX)
- Regression:
  - [ ssic-linux-bugs-2000692 calling exec from a thread crashes node
]. (#ifdef DE_THREAD__OOPS_FIX)
  - proc_pid_lookup() livelock on migrate server. (#ifdef VPROC_IS_ALIVE)
  - process migration pointing to wrong SHM server node. (#ifdef
REOP_EXPORT_PATH_SVRNODE)
  - PVPOP_SIGPROC() callers have shared VPROC LOCK. (#ifdef VPROC_RW_LOCK)
  - PVPOP_REAP() deadlock with parent in SIGCONT code path. (#ifdef
VPROC_RW_LOCK)
  - proc_pid_ttynode() oops reading zombies. (#ifdef VPROC_HOLD_ZERO_GET_TASK)
  - do_task_stat() oops when controlling tty is remote. (#ifdef
VPROC_HOLD_ZERO_GET_TASK)
  - procfs -ENOENT when process is migrating back to client. (#ifdef
VPROC_HOLD_ZERO_GET_TASK)
  - PVP_RECLAIM_CHILD() livelock in VPOP_WAIT() code path when ptrace.
(#ifdef VPROC_RW_LOCK)
  - Possible uninitialized vproc_origin_data structure. (#ifdef VOD_HLIST)



---------- Forwarded message ----------
From: John Hughes <[email protected]>
Date: Dec, 2008
To: Roger Tsang <[email protected]>


Roger Tsang wrote:
>
> Are you running with the vod_rwlock bug fix patch?
>

I'm up to date with current CVS

+ cfs_zap_caches_fix.patch
+ return_migrate_import_procfs_path_fix.patch
+ cluster_util_nsc_scalls.c.patch
+ nd_origin_deadlock_fix.patch

Ah, it's that last one:

Index: linux/cluster/ssi/vproc/nd_origin.c
===================================================================
--- linux.orig/cluster/ssi/vproc/nd_origin.c
+++ linux/cluster/ssi/vproc/nd_origin.c
@@ -116,7 +116,11 @@ vproc_origin_create(
                       error = 0;
               }

+#ifdef ND_ORIGIN_DEADLOCK_FIX
+               UNLOCK_SHR_RW_LOCK(&vodp->vod_rwlock);
+#else
               UNLOCK_EXCL_RW_LOCK(&vodp->vod_rwlock);
+#endif
               goto out;
       }

Yup, I've got that

------------------------------------------------------------------------------
SF.Net email is Sponsored by MIX09, March 18-20, 2009 in Las Vegas, Nevada.
The future of the web can't happen without you.  Join us at MIX09 to help
pave the way to the Next Web now. Learn more and register at
http://ad.doubleclick.net/clk;208669438;13503038;i?http://2009.visitmix.com/