3 * Server support for kernels:
4 2.6.18-194.11.3.el5 (RHEL 5)
5 2.6.18-194.11.3.0.1.el5 (OEL 5)
6 * Client support for unpatched kernels:
7 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
8 2.6.18-194.11.3.el5 (RHEL 5),
9 2.6.18-194.11.3.0.1.el5 (OEL 5)
10 2.6.16.60-0.42.8 (SLES 10),
12 2.6.29.4-167.fc11 (FC11)
13 * Recommended e2fsprogs version: 1.41.10-sun2
14 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
15 Lustre filesystem with 4K stack may cause a stack overflow. For more
16 information, please refer to bugzilla 17630.
18 Severity : enhancement
20 Description: Update RHEL5.5 kernel to 2.6.18-194.11.3.el5 and OEL5.5 kernel to
21 2.6.18-194.11.3.0.1.el5.
23 --------------------------------------------------------------------------------
25 2010-07-15 Oracle, Inc.
27 * Server support for kernels:
28 2.6.18-164.11.1.el5 (RHEL 5)
29 2.6.18-164.11.1.0.1.el5 (OEL 5)
30 * Client support for unpatched kernels:
31 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
32 2.6.18-164.11.1.el5 (RHEL 5),
33 2.6.18-164.11.1.0.1.el5 (OEL 5)
34 2.6.16.60-0.42.8 (SLES 10),
36 2.6.29.4-167.fc11 (FC11)
37 * Recommended e2fsprogs version: 1.41.10-sun2
39 Severity : enhancement
41 Description: Distribute lfs_migrate script with lustre
42 Details : The lfs_migrate mentioned in the manual is now included in the
43 lustre tree in lustre/scripts and will distributed in lustre RPMs.
45 Severity : enhancement
47 Description: Failover nid list added to import proc file.
49 Severity : enhancement
51 Description: "som_preview" mount option required to use SOM
52 Details : The clients will need to use som_preview mount option
56 Description: Lustre does not do 1MB IOs to HW RAID
57 Details : Bump MAX_PHYS/HW_SEGMENTS and SG_ALL to 256 in the RHEL5 kernel.
59 Severity : enhancement
61 Description: lfs quota output is not very convenient for awk/sed-parsing
62 Details : Some positions in lfs quota output table could be empty or
63 non-empty which made it hard to parse it with scripts, now
64 a dash is put instead of space where there is not supposed
69 Description: Update to OFED 1.5.1
70 Details : OFED 1.5.1 will be used in environments that don't use
75 Description: Remove lnet. prefix from lctl params display
76 Details : Also renamed lnet.memused to lnet_memused to
81 Description: Use conf_param -d to remove permanant settings.
85 Description: Add list_param -R
86 Details : lctl list_param now supports recursive mode.
89 Bugzilla : 21829, 22312
90 Description: Fix llobdstat.
91 Details : 1. Now updating the write_bytes counter
92 2. Port script fixes from 1.8.
96 Description: decrease the usage of memory on clients.
97 Details : 1. On clients, recycle dentries and inodes unused.
98 2. Delete the code related to ll_deathrow(att 6215 in bug 1443). It
103 Description: llobdstat fix and enhancement
104 Details : add a counter to set a limit to how many samples will be returned
105 fix a wildcard in the path to limit to obdfilter stats only
109 Description: lfs check only allowed for root
110 Details : Code cleanups and added a regresion test of non-root use of
115 Description: MDS Oopses in ptlrpc_dispatch_difficult_reply()/__list_add()
116 Details : Fix hr->hr_index overload.
120 Description: Fix FIEMAP functionality.
124 Description: FC11 patchless client support added
128 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
129 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
133 Description: New acc-sm script and various test-framework enhancments.
137 Description: Previously landed patch was reverted.
141 Description: Update SLES10 kernel to 2.6.16.60-0.42.8 (patchless)
145 Description: Add ldiskfs maxdirsize mount option.
149 Description: Fix mkfs.lustre support for 16TB LUNs.
153 Description: Print mmp_check_interval
154 Details : print mmp_check_interval and make it possible to abort
155 mount operation in case it takes too long.
159 Description: Add cascading_rw to lustre mpi tests.
163 Description: Store SOM data in EA.
164 Details : Also several SOM related cleanups.
168 Description: Add refcount to md_open_data
172 Description: Use refcount on lov_request_set to prevent memory corruption.
176 Description: Statahead: use parent dir's i_mutex for synchronization.
180 Description: loadgen improvements.
184 Description: Enable peer health detection in ksocklnd and ko2iblnd.
188 Description: Control DCACHE_LUSTRE_INVALID flag with MDS_INODE_LOOKCUP lock
192 Description: Fix race on obd_devcie:md_stats
196 Description: Fix SLES 11 client builds.
197 Details : Resolved uninitalized pointer gcc warnings.
201 Description: Allow non-root access for "lfs check".
202 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
206 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
208 Severity : enhancement
210 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
211 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
215 Description: Sanity test_27p fails.
216 Details : OBD_FAIL_OST_ENOSPC has been moved to filter_statfs().
217 Sanity tests 27 were cleaned up and improved.
221 Description: Unify req format on client/servers
222 Details : Use new req_capsule API [almost] everywhere instead of old PTLRPC
223 buffers and swabbers approach..
228 Description: drop join file stuff.
230 Severity : enhancement
232 Description: Add 2.6.27 and SLES11 (patchless client) support
236 Description: Fix in ptlrpc_expire_one_request() to print the signed time
239 Severity : enhancement
241 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
244 Severity : enhancement
246 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
248 Severity : enhancement
250 Description: correctly handle big reply message.
251 Details : send LNet event if reply is bigger then buffer and adjust this buffer
256 Description: Drop unnecessary __GFP_NOMEMALLOC flag from filter_get_page()
258 Severity : enhancement
260 Description: Add support for OEL5.
264 Description: Conf-sanity.sh 50g test - deactivated OST should not cause a panic.
266 Severity : enhancement
268 Description: "weak-modules" support
269 Details : Implement "weak-modules" support which enables kernel modules
270 to be used with any kernel that implements the same kABI. In
271 order to achieve this modules are now installed in
272 /lib/modules/$(uname -r)/updates/kernel on all distributions.
276 Description: client could not reconnect to OST because of an active request.
277 Details : The client now retries to reconnect to the same server, if a
278 connect request failed with EBUSY or -EAGAIN. Test case 35b
279 conf-sanity.sh has been created.
283 Description: Don't allow make backward step on assignin osc next id.
284 Details : race between allocation next id and ll_sync thread can be cause
285 of set wrong osc next id and can be kill valid ost objects.
289 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
293 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
297 Description: Error handling in osc_statfs_interpret() has been improved.
298 Details : Check in osc_statfs_interpret() for EBADR.
302 Description: Do not send statfs() requests to OSTs disabled by administrator.
303 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
305 Severity : enhancement
307 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
309 Severity : enhancement
311 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
313 Severity : enhancement
315 Description: Enhancement for lfs(1) command to use numeric uid/gid.
319 Description: Improved cleanup procedure for conf-sanity test_46a.
323 Description: Do not start bulk transfer, if client tried to reconnect.
324 Details : Do not start bulk transfer, if client tried to reconnect,
325 but failed due to active rpcs. Check for exp_abort_active_req
326 has been implemented.
330 Description: Prevent inconsistences between linux and lustre mount structures.
331 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
332 Make the sleep interruptible.
334 Severity : enhancement
336 Description: provide server to client comms path
337 Details : server to client communications path via new LDLM_SET_INFO rpc
341 Description: truncate starts GFP_FS allocation under transaction causing deadlock
342 Details : ldiskfs_truncate calls grab_cache_page which may start page
343 allocation under an open transaction. This may lead to
344 calling prune_icache with consequent lustre reentrance.
348 Description: send correctly lsm on open replay
349 Details : MDS is trust to LSM size on replay open, but client can set wrong size
354 Description: enable client interface failover
355 Details : When a child reconnects from another NID, properly update export
356 nid hash position and ldlm reverse import.
358 Severity : enhancement
360 Description: Communicate OST degraded/readonly state via statfs to MDS
361 Details : Flags in the statfs returned from OSTs indicate whether the
362 OST is in a degraded RAID state, or if the filesystem has
363 turned read-only after a filesystem error is detected.
368 Description: don't panic if EPROTO was hit when reading symlink
369 Details : correctly handling request reference in error cases.
371 Severity : enhancement
373 Description: Add LustreNetLink, a kernel-userspace communcation path.
375 Severity : enhancement
377 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
382 Description: access to llog context before init.
383 Details : move handling CATALOGS file at osc layer and forbid access to llog
387 Frequency : rare, only when client is writing to OST on local node
389 Description: Avoid deadlock for local client writes
390 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
391 memory freeing context. This allows OST threads to set the
392 PF_MEMALLOC flag on task structures in order to allocate memory
393 from reserved pools and complete IO.
394 Use GFP_HIGHUSER for OST allocations for non-local client writes,
395 so that the OST threads generate memory pressure and allow
396 inactive pages to be reclaimed.
398 Severity : enhancement
400 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
405 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
406 Details : move ldlm namespace creation in setup phase to avoid grab
407 _lprocfs_lock with cli_sem held.
411 Description: Temporarily disable grant shrink.
412 Details : Disable the feature for debugging.
415 Frequency : only when formatting test filesystems
417 Description: Unable to run several mkfs.lustre on loop devices at the same time
418 Details : mkfs.lustre returns error 256 on the concurrent loop devices
419 formatting. The solution is to proper handle the error.
421 Severity : enhancement
423 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
425 Severity : enhancement
427 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
432 Description: resolve race between obd_disconnect and class_disconnect_exports
433 Details : if obd_disconnect will be called to already disconnected export he
434 forget release one reference and osc module can't unloaded.
436 Severity : enhancement
438 Description: Allow tuning service thread via /proc
439 Details : For each service a new
440 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
441 created that can be used to set min/max thread counts, and get the
442 current number of running threads.
447 Descriptoin: don't return error if have partially created objects for file.
448 Details : lov_update_create_set uses set->set_success as index for created
449 objects, so if some requests will be failed, they will have hole
450 at end of array and qos_shrink_lsm for allocate correct lsm.
452 Severity : enhancement
454 Description: Update OFED support to 1.4.1
459 Description: Reduce small size read RPC
460 Details : Set read-ahead limit for every file and only do read-ahead when
461 available read-ahead pages are bigger than 1M to avoid small size
464 Severity : enhancement
466 Description: move AT tunable parameters for more consistent usage
467 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
469 Severity : enhancement
471 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
472 Details : allow skip disconnected ost for send statfs request and hide error
478 Description: fix racy locking of mballoc block bitmaps causing BUG
479 Details : The locking of the mballoc buddy bitmap and the in-memory
480 block bitmap was using two different spin locks in some
481 cases. This made it possible to incorrectly access the
482 mballoc bitmap while another process was modifying it,
483 causing a sanity assertion to fail. While no on-disk corruption
484 was reported, there was some risk of this happening.
487 Frequency : rare, on llog test 6
489 Descriptoin: don't allow connect to already connected import
490 Details : allowing connect to already connected import is hide connecting problem.
495 Descriptoin: allow kill process which wait statahead result
496 Details : for some reasons 'ls' can stick in waiting result from statahead,
497 in this case need way for kill this process.
499 Severity : enhancement
501 Description: Add state history info file, enhance import info file
502 Details : Track import connection state changes in a new osc/mdc proc file;
503 add overview-type data to the osc/mdc import proc file.
505 Severity : enhancement
507 Description: MDS create should not wait for statfs RPC while holding DLM lock.
509 Severity : enhancement
511 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
516 Descriptoin: OST grant too much space to client even there are not enough space.
517 Details : Client will shrink its grant cache to OST if there are no write
518 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
519 this grant cache if there are already not enough avaible space
520 (left_space < total_clients * 32M).
523 Frequency : start MDS on uncleanly shutdowned MDS device
525 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
526 Details : stay in waiting mds<>ost recovery finished produce random bugs
527 due race between two ll_sync thread for one lov target. send
528 ACTIVATE event only if connect realy finished and import have
532 Frequency : rare, connect and disconnect target at same time
534 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
535 Details : don't call obd_disconnect under lov_lock. this long time
536 operation and can block ptlrpcd which answer to connect request.
541 Descriptoin: don't lose wakeup for imp_recovery_waitq
542 Details : recover_import_no_retry or invalidate_import and import_close can
543 both sleep on imp_recovery_waitq, but we was send only one wakeup
547 Frequency : always with long access acl
549 Descriptoin: mds can't pack reply with long acl.
550 Details : mds don't control size of acl but they limited by reint/getattr
553 Severity : enhancement
555 Description: Update to SLES10 kernel-2.6.16.60-0.33.
557 Severity : enhancement
559 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
562 Frequency : start MDS on uncleanly shutdowned MDS device
564 Descriptoin: aborting recovery hang on MDS
565 Details : don't throttle destroy RPCs for the MDT.
568 Frequency : on remount
570 Description: external journal device not working after the remount
571 Details : clear dev_rdonly flag for external journal devices in
577 Description: shutdown vs evict race
578 Details : client_disconnect_export vs connect request race.
579 if client will evicted at this time - we start invalidate
580 thread without referece to import and import can be freed
584 Frequency : rare, need acl's on inode.
586 Description: client can't handle ost additional correctly
587 Details : if ost was added after client connected to mds client can have
588 hit lnet_try_match_md ... to big messages to wide striped files.
589 in this case need teach client to handle config events about add
590 lov target and update client max ea size at that event.
592 Severity : enhancement
594 Description: Changelogs
595 Details : Changelogs are a lightweight mechanism to track filesystem
596 metadata and namespace changes. The changelog is recorded
597 permanently on the MDTs, and is periodically "consumed" / purged
598 when records are no longer needed.
600 Severity : enhancement
602 Description: compact fld format with extents
603 Details : Store range of seq rather than every seq in FLD. Seq
604 controller update FLD rather than clients. In Case of CMD, mdt0
605 has FLD, all other metadata server act as non persistent proxy
606 for FLD queries and cache fld entries in fld cache.
611 Description: don't skip ost target if they assigned to file
612 Details : Drop slow OSCs if we can, but not for requested start idx.
613 This means "if OSC is slow and it is not the requested
614 start OST, then it can be skipped, otherwise skip it only
615 if it is inactive/recovering/out-of-space.
619 Description: more cleanup in mds_lov
620 Details : not send LOV EA under replay, we can't know about they size at this
621 time. Don't allow client connect to mds before any ost connected,
622 for avoid problems with LOV EA size and returning EIO to client.
624 Severity : enhancement
626 Description: Interoperability at server side (Disk interoperability)
628 Severity : enhancement
630 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
632 Severity : enhancement
634 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
636 Severity : enhancement
638 Description: New client IO stack (CLIO).
640 Severity : enhancement
642 Description: Commit on sharing. Eliminate inter-client dependencies between
643 uncommitted transactions by doing transaction commits.
644 Thereby clients may recovery independently.
647 Frequency : Create a symlink file with a very long name
649 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
650 Details : If there is no extra space in the request for early cancels,
651 ldlm_req_handles_avail() returns 0 instead of a negative value.
653 Severity : enhancement
655 Description: Add /proc entry for import status
656 Details : The mdc, osc, and mgc import directories now have
657 an import directory that contains useful import data for debugging
660 Severity : enhancement
662 Description: Re-disable certain /proc logging
663 Details : Enable and disable client's offset_stats, extents_stats and
664 extents_stats_per_process stats logging on the fly.
667 Frequency : Only on FC kernels 2.6.22+
669 Description: oops in statahead
670 Details : Do not drop reference count for the dentry from VFS when lookup,
671 VFS will do that by itself.
673 Severity : enhancement
675 Description: Generic /proc file permissions
676 Details : Set /Proc file permissions in a more generic way to enable non-
677 root users operate on some /proc files.
681 Description: Hitting mdc_commit_close() ASSERTION
682 Details : Properly handle request reference release in
683 ll_release_openhandle().
687 Description: quota recovery deadlock during mds failover
688 Details : This patch includes att18982, att18236, att18237 in bz14840.
690 1. fix osts hang when mds does failover with quotaon
691 2. prevent watchdog storm when osts threads wait for the
696 Frequency : only patchless client
697 Description: add workaround for race between add/remove dentry from hash
699 Severity : enhancement
701 Description: Allow OST glimpses to return PW locks
705 Description: LBUG when llog conf file is full
706 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
710 Description: Prevent import from entering FULL state when server in recovery
714 Description: service mount cannot take device name with ":"
715 Details : Only when device name contains ":/" will mount treat it as
721 Description: replace ptlrpcd with the statahead thread to interpret the async
722 statahead RPC callback
726 Frequency : on recovery
727 Description: I/O failures after umount during fail back
728 Details : if client reconnected to restarted server we need join to recovery
729 instead of find server handler is changed and process self
730 eviction with cancel all locks.
732 Severity : enhancement
734 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
736 Severity : enhancement
738 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
740 Severity : enhancement
742 Description: Upcall on Lustre log has been dumped
743 Details : Allow for a user mode script to be called once a Lustre log has
744 been dumped. It passes the filename of the dumped log to the
745 script, the location of the script can be specified via
746 /proc/sys/lnet/debug_log_upcall.
751 Description: avoid idr_remove called for id which is not allocated.
752 Details : Move assigment s_dev for clustered nfs to end of initialization,
753 to avoid problem with error handling.
758 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
759 Details : When connection is reused this not moved from CONN_UNUSED_HASH
760 into CONN_USED_HASH and this prodice warning when put connection
761 again in unused hash.
764 Severity : enhancement
766 Description: File striping can now be set to use an arbitrary pool of OSTs.
768 Severity : enhancement
770 Description: Export bytes_read/bytes_write count on OSC/OST.
774 Description: Early reply size mismatch, MGC loses connection
775 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
776 the connect flags are properly negotiated.
781 Description: quotas are not honored with O_DIRECT
782 Details : all writes with the flag O_DIRECT will use grants which leads to
783 this problem. Now using OBD_BRW_SYNC to guard this.
787 Description: add quota statistics
788 Details : 1. sort out quota proc entries and proc code.
789 2. add quota statistics
791 Severity : enhancement
793 Description: enable quota support for HEAD.
797 Description: Properly propagate oinfo flags from lov to osc for statfs
798 Details : restore missing copy oi_flags to lov requests.
800 Severity : enhancement
802 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
803 : l_getgroups(8), lst(8), routerstat(8)
804 Details : included man pages for llobdstat(8), llstat(8),
805 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
807 Severity : enhancement
809 Description: configure's --enable-quota should check the
810 : kernel .config for CONFIG_QUOTA
811 Details : configure is terminated if --enable-quota is passed but
812 : no quota support is in kernel
816 Description: Remove portals compatibility
817 Details : Remove portals compatibility, not interoperable with releases
822 Description: Resolve device initialization race
823 Details : Prevent proc handler from accessing devices added to the
824 obd_devs array but yet be intialized.
826 Severity : enhancement
828 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
830 Severity : enhancement
832 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
837 Description: Using wrong pointer in osc_brw_prep_request
838 Details : Access to array[-1] can produce panic if kernel compiled with
839 CONFIG_PAGE_ALLOC enabled
843 Description: Client runs out of low memory
844 Details : Consider only lowmem when counting initial number of llap pages
848 Description: *optional* service tags registration
849 Details : if the "service tags" package is installed on a Lustre node
850 When the filesystem is mounted, a local-node service tag will
851 be created. See http://inventory.sun.com/ for more information
852 about the Service Tags asset management system.
856 Description: Kernel BUG tries to release flock
857 Details : Lustre does not destroy flock lock before last reference goes
858 away. So always drop flock locks when client is evicted and
859 perform unlock regardless of successfulness of speaking to MDS.
863 Description: add refcount protection for osc callbacks, avoid panic on shutdown
867 Description: sanity test 65a fails if stripecount of -1 is set
868 Details : handle -1 striping on filesystem in ll_dirstripe_verify
873 Description: ASSERTION(CheckWriteback(page,cmd)) failed
874 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
877 Severity : enhancement
879 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
883 Description: do not process already freed flock
884 Details : flock can possibly be freed by another thread before it reaches
885 to ldlm_flock_completion_ast.
889 Description: LBUG during stress test
890 Details : Need properly lock accesses the flock deadlock detection list.
894 Description: oops in page fault handler
895 Details : kernel page fault handler can return two special 'pages' in error
896 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
900 Description: timeout with invalidate import.
901 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
902 should be handled by ptlrpcd. This produce long age waiting and
903 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
905 Severity : enhancement
907 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
911 Description: enable MGS and MDT services start separately
912 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
913 a co-located MGS without starting the MGS, which is a complement
914 to 'nosvc' mount option.
918 Frequency : after recovery
919 Description: precreate to many object's after del orphan.
920 Details : del orphan st in oscc last_id == next_id and this triger growing
921 count of precreated objects. Set flag LOW to skip increase count
922 of precreated objects.
926 Frequency : rare, on clear nid stats
927 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
928 Details : when clean nid stats sometimes try destroy live entry,
929 and this produce panic in free.
933 Description: Stack overflow during MDS log replay
934 ease stack pressure by using a thread dealing llog_process.
938 Description: wait until IO finished before start new when do lock cancel.
939 Details : VM protocol want old IO finished before start new, in this case
940 need wait until PG_writeback is cleared until check dirty flag and
941 call writepages in lock cancel callback.
943 Severity : enhancement
945 Description: using special macro for print time and cleanup in includes.
949 Description: mds_mfd_close() ASSERTION(rc == 0)
950 Details : In mds_mfd_close(), we need protect inode's writecount change
951 within its orphan write semaphore to prevent possible races.
955 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
959 Frequency : rare, on shutdown ost
960 Description: don't hit live lock with umount ost.
961 Details : shrink_dcache_parent can be in long loop with destroy dentries,
962 use shrink_dcache_sb instead.
966 Description: don't panic with use echo client
967 Details : echo client pass NULL as client nid pointer and this produce null
972 Description: fix build on ppc32
973 Details : compile code with -m64 flag produce wrong object file for ppc32.
977 Description: add message levels for liblustreapi
981 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
982 Details : Move the 'good_osts' check before the 'total_bavail' check. This
983 will result in an -EAGAIN and in the exit call path we call
984 alloc_rr() which will with increasing aggressiveness attempt to
985 aquire precreated objects on the minimum number of required OSCs.
989 Description: Use old size assignment to avoid deadlock
990 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
991 the scheduling while holding a spinlock. We do not need locking
992 for size in ll_update_inode() because size is only updated from
993 the MDS for directories or files without objects, so there is no
994 other place to do the update, and concurrent access to such inodes
995 are protected by the inode lock.
999 Description: resolve "_IOWR redefined" build error on SLES10
1003 Description: dump the memory debugging after all modules are unloaded to
1004 suppress false negative in conf_sanity test 39
1006 Severity : enhancement
1008 Description: build kernel-ib packages for OFED 1.3 in our release cycle
1013 Description: fix SLES kernel versioning
1014 Details : the kernel version for our SLES 10 kernel did not include a "-"
1015 before the "smp" at the end. while this was not a problem in
1016 general, it did mean that software trying to use the kernel
1017 version to try to detect a vendor specific kernel would fail.
1018 this was most evident by the OFED build scripts.
1022 Description: Don't update lov_desc members until making sure they are valid
1023 Details : When updating lov_desc members via proc fs, need fix their
1024 validities before doing the real update.
1028 Description: don't put request into delay list while invalidate in flight.
1029 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
1030 import in flight. this produce timeout for invalidate and sometimes
1031 can cause stale data.
1035 Frequency : on ppc only
1036 Description: not convert ost objects for directory because it's not exist.
1037 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
1041 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
1043 Severity : enhancement
1045 Description: Update to support 2.6.22.14 vanilla kernel.
1049 Frequency : rare, on recovery
1050 Description: read procfs can produce deadlock in some situation
1051 Details : Holding lprocfs lock which send rpc can produce block for destroy
1052 obd objects and this also block reconnect with -EALREADY. This
1053 isn't fix all lprocfs bugs - but make it rare.
1055 Severity : enhancement
1057 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
1060 Frequency : frequent on X2 node
1062 Description: mdc_set_open_replay_data LBUG
1063 Details : Set replay data for requests that are eligible for replay.
1067 Description: lustre_mgs: operation 101 on unconnected MGS
1068 Details : When MGC is disconnected from MGS long enough, MGS will evict the
1069 MGC, and late on MGC cannot successfully connect to MGS and a lot
1070 of the error messages complaining that MGS is not connected.
1074 Frequency : on network error
1075 Description: panic with double free request if network error
1076 Details : mdc_finish_enqueue is finish request if any network error ocuring,
1077 but it's true only for synchronus enqueue, for async enqueue
1078 (via ptlrpcd) this incorrect and ptlrpcd want finish request
1081 Severity : enhancement
1083 Description: client-side metadata stat-ahead during readdir(directory readahead)
1084 Details : perform client-side metadata stat-ahead when the client detects
1085 readdir and sequential stat of dir entries therein
1088 Frequency : on start mds
1090 Description: Implement get_info(last_id) in obdfilter.
1093 Frequency : occasional
1095 Description: Correctly check stale fid, not start epoch if ost not support SOM
1096 Details : open with flag O_CREATE need set old fid in op_fid3 because
1097 op_fid2 was overwritten with new generated fid, but mds can answer
1098 with one of these two fids and both is not stale. Setattr
1099 incorrectly started an epoch and assume will be called
1100 done_writing, but without SOM done_writing ever being called.
1103 Frequency : rare, depends on device drivers and load
1105 Description: MDS or OSS nodes crash due to stack overflow
1106 Details : Code changes in 1.8.0 increased the stack usage of some functions.
1107 In some cases, in conjunction with device drivers that use a lot
1108 of stack the MDS (or possibly OSS) service threads could overflow
1109 the stack. One change which was identified to consume additional
1110 stack has been reworked to avoid the extra stack usage.
1113 Frequency : occasional
1115 Description: Do not fail import if osc_interpret_create gets -EAGAIN
1116 Details : If osc_interpret_create got -EAGAIN it immediately exits and
1117 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
1118 oscc_has_objects and see OSC has no objests and call
1119 oscc_internal_create to resend create request.
1121 Severity : enhancement
1123 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1125 Severity : enhancement
1127 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1130 Frequency : very rare
1132 Description: Fix replay if there is an un-replied request and open
1133 Details : In some cases, older replay request will revert the
1134 mcd->mcd_last_xid on MDS which is used to record the client's
1135 latest sent request.
1137 Severity : enhancement
1139 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1141 Severity : enhancement
1143 Description: Add rhel5 support to HEAD.
1145 Serverity : enhancement
1147 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1152 Description: Don't allow skipping OSTs if index has been specified.
1153 Details : Don't allow skipping OSTs if index has been specified, make
1154 locking in internal create lots better.
1158 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1159 Details : access to bitfield in structure is always rounded to long
1160 and this produce problem with not atomic change any bit.
1164 Description: Lustre make rpms failed.
1165 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1166 builds Lustre rpms from tarball.
1169 Frequency : rare on shutdown ost
1171 Description: If llog cancel was not send before clean_exports phase, this can
1172 produce deadlock in llog code.
1173 Details : If llog thread has last reference to obd and call class_import_put
1174 this produce deadlock because llog_cleanup_commit_master wait when
1175 last llog_commit_thread exited, but this never success because was
1176 called from llog_commit_thread.
1180 Description: allow userland application know is lost one of stripes.
1181 Details : fill lvb_blocks with error code on ost and return it to
1182 application if error flag found.
1186 Description: NULL lov_tgts causing MDS oops
1187 Details : more safe checks for NULL lov_tgts for avoid oops.
1189 Severity : enhancement
1191 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1195 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
1196 Details : If an OST is down the MDS will hang indefinitely in
1197 obd_statfs_rqset() waiting for the statfs data. While for
1198 MDS QOS usage of statfs, it should not stuck in waiting.
1200 Severity : enhancement
1202 Description: remote_acl support
1203 Details : Support ACL-based permission check for remote user.
1204 Support setfacl/getfacl for remote user with the utils
1205 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
1206 the system "{s,g}etfacl" utils.
1208 Severity : enhancement
1210 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
1212 Severity : enhancement
1214 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1218 Description: make mgs_setparam() handle fsname containing dash
1219 Details : fsname containing a dash does not work with lctl conf_param
1221 Severity : enhancement
1223 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1225 Severity : enhancement
1227 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1231 Frequency : rare, at shutdown
1232 Description: access already free / zero obd_namespace.
1233 Details : if client_disconnect_export was called without force flag set,
1234 and exist connect request in flight, this can produce access to
1235 NULL pointer (or already free pointer) when connect_interpret
1236 store ocd flags in obd_namespace.
1240 Frequency : only at startup
1241 Description: not alloc memory with spinlock held.
1242 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1243 if any spinlock held.
1245 Severity : enhancement
1247 Description: make lustre randomly fail allocating memory
1248 Details : Make lustre randomly failed allocating memory for testing purpose.
1250 Severity : enhancement
1252 Description: lost problems with lov objid file
1253 Details : Fixes some scability and access to not inited memory problems
1254 in work with lov objdid file.
1259 Description: lfs find does not continue on file error
1260 Details : Continue other files processing when a file/dir is absent.
1264 Description: Inconsistent usage of lustre_pack_reply()
1265 Details : Standardize the usage of lustre_pack_reply() such that it
1266 always generate a CERROR on failure.
1269 Frequency : occasional
1271 Description: MDS hang or stay in waiting lock
1272 Details : If client receive lock with CBPENDING flag ldlm need send lock
1273 cancel as separate rpc, to avoid situation when cancel request
1274 can't processed due all i/o threads stay in wait lock.
1278 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1282 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1286 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1290 Description: Update to RHEL4 latest kernel.
1292 Severity : enhancement
1294 Description: Build SLES10 patchless client fails
1295 Details : The configure was broken by run ./configure with
1296 --with-linux-obj=.... argument for patchless client. When the
1297 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1298 can't search header adequately. Use absolute path such as
1299 -I($LINUX)/include instead.
1303 Description: interrupt oig_wait produce painc on resend.
1304 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1305 private set, and this produce situation when rq_ptlrpcd_data not
1306 copyed to new allocated request and triggered LBUG on assert
1307 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1308 ptlrpcd set if request is changed and can be safety changed to use
1313 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1314 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1315 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1316 these macros and replace them with their difinitions instead.
1320 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1321 Details : Update which_patch & target file for SLES10 latest kernel.
1324 Description: Remove client patches from SLES 10 kernel.
1325 Details : This causes SLES 10 clients to behave as patchless clients
1326 even on a Lustre-patched (server) kernel.
1328 Severity : enhancement
1330 Description: self-adjustable client's lru lists
1331 Details : use adaptive algorithm for managing client cached locks lru
1332 lists according to current server load, other client's work
1333 pattern, memory activities, etc. Both, server and client
1334 side namespaces provide number of proc tunables for controlling
1337 Severity : enhancement
1339 Description: light-weight GSS support
1340 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1341 and reduce performance overhead.
1343 Severity : enhancement
1345 Description: Linux keyring support
1346 Details : Support using service of Linux keyring for Lustre GSS internal
1347 context refresh/cache mechanism.
1351 Description: Fix errors in lfs documentation
1352 Details : Fixes man pages
1356 Description: don't use GFP_* in generic Lustre code.
1357 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1361 Description: obdclass is limited by single OBD_ALLOC(idarray)
1362 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1366 Description: warnings with build patchless client with vanila 2.6.19 and up
1367 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1368 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1372 Frequency : Only for SLES
1373 Description: sanity_quota fail test_1
1374 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1375 file, which makes TSTID[2] inunique.
1379 Frequency : Only for SLES9
1380 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1381 Details : Update target/ChangeLog/which_patch .
1385 Frequency : Only for test_10 in sanity-quota.sh
1386 Description: a bug in quota test code
1387 Details : checking a test flag in a wrong place
1391 Frequency : Only for test_13 in sanity-quota.sh
1392 Description: a bug in quota test script
1393 Details : shouldn't check overfull allocated quota
1397 Description: jbd statistics
1398 Details : Port older jbd statistics patch for sles10
1400 Severity : enhancement
1402 Description: Add printing inode info into message about error in writepage.
1407 Description: reply_lock_interpret crash due to race with it and lock cancel.
1408 Details : Do not replay locks that are being cancelled. Do not reference
1409 locks by their address during replay, just by their handle.
1413 Frequency : When flocks are used.
1414 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1415 Details : Flock locks might destroy just granted lock if it could be merged
1416 with another existing flock, this is done in completion handler,
1417 so teach ldlm_cli_enquque_fini that this is a valid case for
1423 Description: Oops in read and write path when failing to allocate lock.
1424 Details : Check if lock allocation failed and return error back.
1428 Description: lstripe command fails for valid OST index
1429 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1430 instead of lov->desc.ld_active_tgt_count.
1434 Description: sanity.sh failed test 103
1435 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1440 Description: df doesn't work properly if diskfs blocksize != 4K
1441 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1445 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1446 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1448 Severity : enhancement
1450 Description: Adaptive timeouts
1451 Details : RPC timeouts adapt to changing server load and network
1452 conditions to reduce resend attempts and improve recovery time.
1456 Description: llapi_file_create() does not allow some changes
1457 Details : add llapi_file_open() that allows specifying the mode and
1458 open flags, and also returns an open file handle.
1462 Description: merge and cleanup kernel patches.
1463 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1467 Description: Add journal checksum support.(Kernel part)
1468 Details : The journal checksum feature adds two new flags i.e
1469 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1470 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1471 indicates that the commit block contains the checksum for
1472 the blocks described by the descriptor blocks. Now commit
1473 record can be sent to disk without waiting for descriptor
1474 blocks to be written to disk. This behavior is controlled
1475 using JBD2_FEATURE_ASYNC_COMMIT flag.
1479 Description: OSS needs mutliple precreate threads
1480 Details : Add ability to start more than one create thread per OSS.
1484 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1485 Details : execute lfs setstripe on client
1489 Description: mds_obd_create error creating tmp object
1490 Details : When the user sets quota on root, llog will be affected and can't
1491 create files and write files.
1494 Frequency : Always on ia64 patchless client, and possibly others.
1496 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1497 Details : This allows the patchless client to be loaded on architectures
1498 without this export.
1502 Description: RedHat Update kernel for RHEL5
1503 Details : Add the kernel config file for RHEL5.
1507 Description: RedHat Update kernel for RHEL5
1508 Details : Modify the kernel config file more closer RHEL5.
1512 Description: Build failure against Centos5 (RHEL5)
1513 Details : Define PAGE_SIZE when it isn't present.
1517 Description: test_7 on sanity-quota.sh can't be used on separate machines
1518 Details : add facet to handle it
1522 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1523 Details : with NFS, the anon dentry's parent was set to itself in
1524 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1525 corresponding dentry other than use rec->ur_name.
1527 Severity : enhancement
1529 Description: lfs setstripe enhancement
1530 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1534 Description: enable data checksumming by default
1535 Details : enable checksum by default, allow --disable-checksum
1536 configure option and "-o nochecksum" mount option. Checksums
1537 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1541 Description: lustre support for RHEL5
1542 Details : Add support for RHEL5.
1546 Description: Client eviction due to failover config
1547 Details : after a connection loss, the lustre client should attempt to
1548 reconnect to the last active server first before trying the
1549 other potential connections.
1553 Description: when mds and osts use different quota unit(32bit and 64bit),
1554 quota will be released repeatly.
1555 Details : void sending multiple quota reqs to mds, which will keep the
1556 status between the reqs.
1560 Description: osts not allocated evenly to files
1561 Details : change the condition to increase offset_idx
1564 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1566 Description: Data corruption for OSTs that are formatted larger than 2TB
1568 Details : When generating the bio request for lustre file writes the
1569 sector number would overflow a temporary variable before being
1570 used for the IO. The data reads correctly from Lustre (which
1571 will overflow in a similar manner) but other file data or
1572 filesystem metadata may be corrupted in some cases.
1576 Description: Tune the kernel for good SCSI performance.
1577 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1578 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1583 Description: rewrite ext2-derived code in obdclass/uuid.c
1584 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1585 from scratch preserving functionality.
1589 Description: "lfs find -obd UUID" prints directories
1590 Details : "lfs find -obd UUID" will return all directory names instead
1591 of just file names. It is incorrect because the directories
1592 do not reside on the OSTs.
1596 Description: use i_size_read and i_size_write in 2.6 port
1597 Details : replace inode->i_size access with i_size_read/write()
1600 Frequency : when removing large files
1602 Description: scheduling issue during removal of large Lustre files
1603 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1604 It causes scheduling issues when removing large files (17TB in the
1608 Frequency : only with liblustre clients on XT3
1610 Description: evictions taking too long
1611 Details : allow llrd to evict clients directly on OSTs
1614 Frequency : only on ppc
1616 Description: /proc/fs/lustre/devices broken on ppc
1617 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1618 We should make sure the type of 'index' is loff_t and then cast
1619 explicitly as needed below. Do not assign an explicitly cast
1623 Frequency : only for rhel5
1625 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1626 Details : Modify the target file & which_kernel.
1628 Severity : enhancement
1630 Description: omit set fsid for export NFS
1631 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1632 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1636 Frequency : Always, for kernels after 2.6.16
1637 Description: Fix warning idr_remove called for id=.. which is not allocated.
1638 Details : Last kernels save old s_dev before kill super and not allow
1639 to restore from callback - restore it before call kill_anon_super.
1643 Description: buffer overruns could theoretically occur
1644 Details : llapi_semantic_traverse() modifies the "path" argument by
1645 appending values to the end of the origin string, and a
1646 overrun may occur. Adding buffer overrun check in liblustreapi.
1650 Description: Fix error on 'ls .' at the top of the Lustre mount.
1651 Details : Don't revalidate dentry if it is a root dentry.
1655 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1656 Details : Modify vm-tunables-rhel4.patch.
1660 Description: Kernel config for 2.6.18-vanilla.
1661 Details : Modify targets/2.6-vanilla.target.in.
1662 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1663 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1664 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1665 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1669 Description: improve handling recoverable errors
1670 Details : if request processig with error which can be recoverable on server
1671 request should be resend, otherwise page released from cache and
1676 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1677 Details : Modify target file & which_patch.
1678 A flaw was found in the IA32 system call emulation provided
1679 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1680 value could be stored in the %RAX register, which could trigger an
1681 out-of-bounds system call table access. An untrusted local user
1682 could exploit this flaw to run code in the kernel
1683 (ie a root privilege escalation). (CVE-2007-4573).
1687 Description: O_DIRECT bypasses client statistics.
1688 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1689 read_ahead_stats were updated. copied the stats section from
1690 osc_send_oap_rpc() into async_internal().
1694 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1698 Description: change order of libsysio includes
1699 Details : '#include sysio.h' should always come before '#include xtio.h'
1703 Description: handle "serious error: objid * already exists" more gracefully
1704 Details : If LAST_ID value on disk is smaller than the objects existing in
1705 the O/0/d* directories, it indicates disk corruption and causes an
1706 LBUG(). If the object is 0-length, then we should use the existing
1707 object. This will help to avoid a full fsck in most cases.
1709 Severity : enhancement
1711 Description: adapt the lustre_config script to support the upgrade case
1712 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1713 server targets to 1.6 in parallel.
1717 Description: To avoid grant space > avaible space when the disk is almost
1718 full. Without this patch you might see the error "grant XXXX >
1719 available" or some LBUG about grant, when the disk is almost
1721 Details : In filter_check_grant, for non_grant cache write, we should
1722 check the left space by if (*left > ungranted + bytes), instead
1723 of (*left > ungranted), because only we are sure the left space
1724 is enough for another "bytes", then the ungrant space should be
1725 increase. In client, we should update cl_avail_grant only there
1726 is OBD_MD_FLGRANT in the reply.
1730 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1731 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1732 described in CVE-2007-4573. This problem could allow untrusted
1733 local users to gain root access.
1736 Frequency : when using O_DIRECT and quotas
1738 Description: Incorrect file ownership on O_DIRECT output files
1739 Details : block usage reported by 'lfs quota' does not take into account
1740 files that have been written with O_DIRECT.
1745 Description: touch file failed when fs is not full
1746 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1747 otherwise we can get ENOSP while fs is not full.
1751 Description: parallel lock callbacks
1752 Details : Instead of sending blocking and completion callbacks as separated
1753 requests, adding them to a set and sending in parallel.
1756 Frequency : only for Cray XT3
1757 Bugzilla : 12829/13455
1758 Description: Changing primary group doesn't change the group lustre assigns to
1760 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1761 overridden with the primary group provided by the group upcall,
1762 whereas the supplied fsgid can be trusted if it is in the list of
1763 supplementary groups returned by the group upcall.
1765 Severity : enhancement
1767 Description: Allow masking D_WARNING, D_ERROR messages from console
1768 Details : Console messages can now be disabled via lnet.printk.
1772 Description: User code with malformed file open parameter crashes client node
1773 Details : Before packing join_file req, all the related reference should be
1774 checked carefully in case some malformed flags cause fake
1775 join_file req on client.
1779 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1780 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1781 mode is equal to the granted mode. However, it should also check
1782 that LDLM_FL_CP_REQD is not set.
1786 Description: Heavy nfs access might result in deadlocks
1787 Details : After ELC code landed, it is now improper to enqueue any mds
1788 locks under och_sem, because enqueue might want to decide to
1789 cancel open locks for same inode we are holding och_sem for.
1793 Description: Client eviction while running blogbench
1794 Details : A lot of unlink operations with concurrent I/O can lead to a
1795 deadlock causing evictions. To address the problem, the number of
1796 oustanding OST_DESTROY requests is now throttled to
1797 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1798 callbacks are processed in priority.
1802 Description: enable ACLs on MDS by default
1803 Details : ACLs must be enabled on MDS by default.
1806 Frequency : PPC/PPC64 only
1808 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1809 Details : fix duplicated definitions between asm-ppc64/types.h and
1810 lustre_types.h on PPC.
1813 Frequency : PPC/PPC64 only
1815 Description: asm-ppc/segment.h does not exist
1816 Details : fix compile issue on PPC.
1820 Description: better handle error messages in extents code
1823 Frequency : RHEL4 only
1825 Description: mkfs is very slow on IA64/RHEL4
1826 Details : A performance regression has been discovered in the MPT Fusion
1827 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1828 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1829 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1831 Severity : enhancement
1833 Description: SNMP support enhancement
1834 Details : Adding total number of sampled request for an MDS node in snmp
1837 Severity : enhancement
1839 Description: Optimize ldlm waiting list processing for PR extent locks
1840 Details : When processing waiting list for read extent lock and meeting read
1841 lock that is same or wider to it that is not contended, skip
1842 processing rest of the list and immediatelly return current
1843 status of conflictness, since we are guaranteed there are no
1844 conflicting locks in the rest of the list.
1848 Description: Time out and refuse to reconnect
1849 Details : When the failover node is the primary node, it is possible
1850 to have two identical connections in imp_conn_list. We must
1851 compare not conn's pointers but NIDs, otherwise we can defeat
1852 connection throttling.
1856 Description: port llog fixes from b1_6 into HEAD
1857 Details : Port llog reference couting and some llog cleanups from b1_6
1858 (bug 10800) into HEAD, for protect from panic and access to already
1859 free llog structures.
1863 Description: Detect stride IO mode in read-ahead
1864 Details : When a client does stride read, read-ahead should detect that and
1865 read-ahead pages according to the detected stride pattern.
1869 Description: data checksumming impacts single node performance
1870 Details : add support for several checksum algorithm. Currently, only CRC32
1871 and Adler-32 are supported. The checksum type can be changed on
1872 the fly via /proc/fs/lustre/osc/*/checksum_type.
1876 Description: use adler32 for page checksums
1877 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1882 Description: build for x2 fails
1883 Details : fix compile issue on Cray systems.
1887 Description: Properly match for duplicate locks
1888 Details : Due to different lock order from skiplists code, we need to
1889 traverse entire list for now
1892 Frequency : only on PPC/SLES10
1894 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1895 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1896 be defined. Add a hack in mkfs_lustre.c to work around this compile
1901 Description: LASSERT on MDS when client holding flock lock dies
1902 Details : ldlm pool logic depends on number of granted locks equal to
1903 number of released locks which is not true for flock locks, so
1904 just exclude such locks from consideration.
1908 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1909 Details : Use fsfilt_sync() for both the whole filesystem sync and
1910 individual file sync to eliminate dangerous inode locking
1911 with I_LOCK that can lead to a deadlock.
1915 Description: performance in 1.6.3
1916 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1917 sure that 1MB requests are not fragmented by the block layer.
1919 Severity : enhancement
1921 Description: organize the server-side client stats on per-nid basis
1922 Details : Change the structure of stats under obdfilter and mds to
1930 The "uuid"s file would list the uuids of _active_ exports.
1931 And the clear entry is to clear all stats and stale nids.
1933 Severity : enhancement
1935 Description: eliminate client locks in face of contention
1936 Details : file contention detection and lockless i/o implementation
1937 for contended files.
1941 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1944 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1946 Description: MDS may lose file striping (and hence file data) in some cases
1947 Details : If there are additional extended attributes stored on the MDS,
1948 in particular ACLs, SELinux, or user attributes (if user_xattr
1949 is specified for the client mount options) then there is a risk
1950 of attribute loss. Additionally, the Lustre file striping
1951 needs to be larger than default (e.g. striped over all OSTs),
1952 and an additional attribute must be stored initially in the
1953 inode and then increase in size enough to be moved to the
1954 external attribute block (e.g. ACL growing in size) for file
1959 Description: skiplist implementation simplification
1960 Details : skiplists are used to group compatible locks on granted list
1961 that was implemented as tracking first and last lock of each
1962 lock group the patch changes that to using doubly linked lists
1966 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1967 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1968 because LASSERT on both the data supplied by a client, and the
1969 data on disk is dangerous and incorrect.
1971 Severity : enhancement
1973 Description: Slow truncate/writes to huge files at high offsets.
1974 Details : Directly associate cached pages to lock that protect those pages,
1975 this allows us to quickly find what pages to write and remove
1976 once lock callback is received.
1980 Description: more ldlm soft lockups
1981 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1982 starve other threads from the resource lock for a long time in
1983 case of long waiting queue, so change the debug level from
1984 D_OTHER to the less frequently used D_INFO.
1986 Severity : enhancement
1988 Description: add -gid, -group, -uid, -user options to lfs find
1992 Description: Hung threads in invalidate_inode_pages2_range
1993 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1994 once one is completed. As a result, some RPCs are stuck in the
1995 queue and are never sent.
1999 Description: filter threads hungs on waiting journal commit
2000 Details : Cleanup filter group llog code, then only filter group llog will
2001 be only created in the MDS/OST syncing process.
2005 Description: Procfs and llog threads access destoryed import sometimes.
2006 Details : Sync the import destoryed process with procfs and llog threads by
2007 the import refcount and semaphore.
2009 Severity : enhancement
2011 Description: openlock cache of b1_6 port to HEAD
2016 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
2017 Details : If insertion of an extent fails, then discard the inode
2018 preallocation and free data blocks else it can lead to duplicate
2023 Description: don't always update ctime in ext3_xattr_set_handle()
2024 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
2025 In some cases the ctime should not be updated, for example for
2026 2.0->1.8 compatibility it is necessary to delete an xattr and it
2027 should not update the ctime.
2031 Bugzilla : 15713/16362
2032 Description: Assertion in iopen_connect_dentry in 1.6.3
2033 Details : looking up an inode via iopen with the wrong generation number can
2034 populate the dcache with a disconneced dentry while the inode
2035 number is in the process of being reallocated. This causes an
2036 assertion failure in iopen since the inode's dentry list contains
2037 both a connected and disconnected dentry.
2041 Description: assertion failure in ldlm_handle2lock()
2042 Details : fix a race between class_handle_unhash() and class_handle2object()
2043 introduced in lustre 1.6.5 by bug 13622.
2048 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
2049 Details : remove the limit of 256 scsi disks in the sd_iostat patch
2054 Description: oops in sd_iostats_seq_show()
2055 Details : unloading/reloading the scsi low level driver triggers a kernel
2056 bug when trying to access the sd iostat file.
2061 Description: Kernel panics during QLogic driver reload
2062 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
2063 patch, causing memory corruption.
2068 Description: journal_dev option does not work in b1_6
2069 Details : pass mount option during pre-mount.
2071 Severity : enhancement
2073 Description: Add a FIEMAP(FIle Extent MAP) ioctl
2074 Details : FIEMAP ioctl will allow an application to efficiently fetch the
2075 extent information of a file. It can be used to map logical blocks
2076 in a file to physical blocks in the block device.
2080 Description: LDLM soft lockups - improvement
2081 Details : It is be possible to send the lock handle along with each read
2082 or write request because the client is already doing a lock match
2083 itself so there isn't any reason the OST should have to re-do that
2087 Frequency : only on Cray X2
2089 Description: X2 build failures
2090 Details : fix build failures on Cray X2.
2094 Description: xid & resent requests
2095 Details : Initialize RPC XID from clock at startup (randomly if clock is
2098 Severity : enhancement
2100 Description: Add lustre_start utility to start or stop multiple Lustre servers
2105 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
2106 Details : In case of memory pressure, list_del() can be called twice on
2107 req->rq_history_list, causing a kernel oops.
2111 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
2112 Details : fix stack overflow in the distributed lock manager by defering
2113 export eviction after a failed AST to the elt thread instead of
2114 handling it in the dlm interpret routine.
2118 Description: Convert some comments to new format.
2119 Details : Update documenting comments to match doxygen conventions.
2123 Description: Grammar fixes.
2124 Details : A couple of trivial spelling fixes.
2128 Description: OSD_COUNTERS-mandatory
2129 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
2130 to simplify the code.
2134 Description: simplify cmm_device freeing logic.
2135 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
2139 Description: Add lockdep support to dt_object_operations locking interface.
2140 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
2141 parameter indicating lock ordering. Update mdd code to use new
2146 Description: Introduce failloc constants for lockless IO tests.
2147 Details : Add two new failloc constants to test lockless IO. Only one of
2148 them in implemented---another is checked in yet to be landed
2153 Description: Add lockdep support for inode mutex.
2154 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
2155 in the situations where Lustre has to lock more than one inode
2160 Description: Add optional invariants checking support.
2161 Details : Add new LINVRNT() macro, optional on new --enable-invariants
2162 configure switch. This macro is to be used for consistency and
2163 sanity checks that are too expensive to be left in `production'
2168 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
2169 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
2174 Description: Add lockdep support for ldlm_lock and ldlm_resource.
2175 Details : Use spin_lock_nested() in (the only) situation where more than
2176 one ldlm_lock is locked simultaneously. Also, fix possible
2177 dead-lock in ldlm_lock_change_resource() by enforcing particular
2182 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
2183 Details : Instead of specifying each ldlm_lock call-back through separate
2184 parameter, wrap them into struct ldlm_callback_suite.
2188 Description: Kill join_lru obd method and its callers.
2189 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
2190 in memory---a requirement implemented through ->o_join_lru() obd
2191 method in HEAD. Remove this method and its users.
2195 Description: Add asynchronous ldlm ENQUEUE completion handler.
2196 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
2197 case that stock ldlm_completion_ast() cannot handle as it waits
2198 until lock is granted. Introduce new ldlm_completion_ast_async()
2199 for this. Also comment ldlm_completion_ast().
2203 Description: ldlm_error <-> errno conversion.
2204 Details : Add functions to map (rather arbitrary) between LDLM error codes
2205 and standard errno values. CLIO needs this to prevent LDLM specific
2206 constants from escaping ldlm and osc.
2210 Description: Kill unused ldlm_handle2lock_ns() function.
2211 Details : Kill unused ldlm_handle2lock_ns() function.
2215 Description: Add lu_ref support to ldlm_lock
2216 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
2217 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
2218 and ldlm_resource. LDLM interface has to be changed, because code that
2219 releases a reference on a lock, has to "know" what reference this is.
2220 In the most frequent case
2222 lock = ldlm_handle2lock(handle);
2224 LDLM_LOCK_PUT(lock);
2226 no changes are required. When any other reference (received _not_ from
2227 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
2228 instead of LDLM_LOCK_PUT().
2230 Arguably, changes are pervasive, and interface requires some discipline
2231 for proper use. On the other hand, it was very instrumental in finding
2232 a few leaked lock references.
2236 Description: Add ldlm_lock_addref_try().
2237 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
2238 attempts to addref a lock that might be being canceled
2243 Description: Add ldlm_weigh_callback().
2244 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
2245 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
2246 hard-coded `number of pages' logic.
2250 Description: Add lockdep annotations to llog code.
2251 Details : Use appropriately tagged _nested() locking calls in the places
2252 where llog takes more than one ->lgh_lock lock.
2256 Description: Add loi_kms_set().
2257 Details : Wrap kms updates into a helper function.
2261 Description: Constify instances of struct lsm_operations.
2262 Details : Constify instances of struct lsm_operations.
2266 Description: lu_conf support.
2267 Details : On a server, a file system object is uniquely identified
2268 by a fid, which is sufficient to locate and load all object
2269 state (inode). On a client, on the other hand, more data are
2270 necessary instantiate an object. Change lu_object_find() and
2271 friends to take additional `lu_conf' argument describing object.
2272 Typically this includes layout information.
2276 Description: lu_context fixes.
2277 Details : Introduce new lu_context functions that are needed on the client
2278 side, where some system threads (ptlrpcd) are shared by multiple
2279 modules, and so cannot be stopped during module shutdown.
2283 Description: Add start and stop methods to lu_device_type_operations.
2284 Details : Introduce two new methods in lu_device_type_operations, that are
2285 invoked when first instance of a given type is created and last one
2286 is destroyed respectively. This is need by CLIO.
2290 Description: Add lu_ref support to struct lu_device.
2291 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2292 track leaked references.
2296 Description: Introduce lu_kmem_descr.
2297 Details : lu_kmem_descr and its companion interface allow to create
2298 and destroy a number of kmem caches at once.
2302 Description: Fix lu_object finalization race.
2303 Details : Fix a race between lu_object_find() finding an object and its
2304 concurrent finalization. This race is (most likely) not possible
2305 on the server, but might happen on the client.
2309 Description: Introduce lu_ref interface.
2310 Details : lu_ref is a debugging module allowing to track references to
2311 a given object. It is quite cpu expensive, and has to be
2312 explicitly enabled with --enable-lu_ref. See usage description
2317 Description: Factor lu_site procfs stats into a separate function.
2318 Details : Separate lu_site stats printing code into a separate function
2319 that can be reused on a client.
2323 Description: Constify instances of struct {lu,dt,md}_device_operations.
2324 Details : Constify instances of struct {lu,dt,md}_device_operations.
2328 Description: Introduce struct md_site and move meta-data specific parts of
2329 struct lu_site here.
2330 Details : Move md-specific fields out of struct lu_site into special struct
2331 md_site, so that lu_site can be used on a client.
2335 Description: Kill mdd_lov_destroy().
2336 Details : Remove unused mdd code.
2340 Description: Add st_block checking to multistat.c.
2341 Details : Add st_block checking to multistat.c.
2345 Description: Add lu_ref support to struct obd_device.
2346 Details : Add lu_ref tracking to obd_device.
2350 Description: Kill obd_set_fail_loc().
2351 Details : Remove unused code.
2355 Description: Add special type for ptlrpc_request interpret functions.
2356 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2357 there. Actual usage will be in CLIO.
2361 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2362 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2363 doesn't work with lockdep.
2367 Description: Add rwv.c test program.
2368 Details : New testing program exercising readv(2) and writev(2) (Qian).
2372 Description: Add sendfile.c test program.
2373 Details : New testing program exercising sendfile(2) (Jay).
2377 Description: Ratelimit a message that can be very frequent.
2378 Details : Ratelimit a memory allocation failure message that can
2383 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2384 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2388 Description: Liblustre build fixes.
2389 Details : Liblustre build fixes.
2393 Description: libcfs: add cfs_{need,cond}_resched() interface.
2394 Details : libcfs: add cfs_{need,cond}_resched() definition and
2395 implementations for Linux, NT, and liblustre.
2397 Severity : enhancement
2399 Description: More exported tunables for mballoc
2400 Details : Add support for tunable preallocation window and new tunables for
2401 large/small requests
2405 Description: Detect corruption of block bitmap and checking for preallocations
2406 Details : Checks validity of on-disk block bitmap. Also it does better
2407 checking of number of applied preallocations. When corruption is
2408 found, it turns filesystem readonly to prevent further corruptions.
2412 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2413 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2414 window to 0. If it does hit cache miss, and read-pattern is still
2415 stride-io mode, does not reset the stride window, but also does not
2416 increase the stride window length in this case.
2420 Frequency : only for big-endian servers
2421 Description: Check if system is big-endian while mounting fs with extents feature
2422 Details : Mounting a filesystem with extents feature will fail on big-endian
2423 systems since ext3-based ldiskfs is not supported on big-endian
2424 systems. This can be over-riden with "bigendian_extents" mount option.
2426 Severity : enhancement
2428 Description: The root squash functionality
2429 Details : A security feature, which is to prevent users from being able
2430 to mount lustre on their desktop, run as root, and delete
2431 all of the files in the filesystem. The goal is accomplished by
2432 remapping user id (UID) and group id (GID) of the root user to
2433 a UID and GID specified by the system administartor via Lustre
2434 configuration management server (MGS). The functionality also
2435 allows to specify sets of clients for which the remapping does
2440 Description: Excessive recovery window
2441 Details : With AT enabled, the recovery window can be excessively long (6000+
2442 seconds). To address this problem, we no longer use
2443 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2444 timeout no longer depends on the service time, it is set to
2445 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2446 time via pb_service_time.
2450 Description: Watchdog triggered on MDS failover
2451 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2452 the OSTs know that the MDS "UUID" can be reused for the same export
2453 from a different NID, so we do not need to wait for the export to be
2457 Frequency : rare, only if using MMP with Linux RAID
2459 Description: MMP doesn't work with Linux RAID
2460 Details : While using HA for Lustre servers with Linux RAID, it is possible
2461 that MMP will not detect multiple mounts. To make this work we
2462 need to unplug the device queue in RAID when the MMP block is being
2463 written. Also while reading the MMP block, we should read it from
2464 disk and not the cached one.
2466 Severity : enhancement
2468 Description: open file using fid
2469 Details : A file can be opened using just its fid, like
2470 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2473 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2476 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2477 Details : add proper locking for creation of jbd_4k slab cache
2481 Description: MMP check in ext3_remount() fails without displaying any error
2482 Details : When multiple mount protection fails during remount, proper error
2485 Severity : enhancement
2487 Description: Allow stripe size to be up to 4G-64k
2488 Details : Fix math logic to allow large stripe sizes.
2492 Description: add check for >8TB ldiskfs filesystems
2493 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2494 Don't allow >8TB ldiskfs filesystems to be mounted without
2495 force_over_8tb mount option
2500 Descriptoin: OSS complains "too many blocks" and crashes
2501 Details : In filter_commitrw_write():
2502 1. lquota_chkquota();
2503 2. obdo_from_inode();
2504 3. lquota_pending_commit();
2505 "2." has changed uid and gid in this case, then caused
2506 lquota_chkquota() and lquota_pending_commit() to use
2507 different uid and gid. That is the root of the bug.
2511 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2512 Details : Detect and print out the 'stripe count = -1' case properly.
2516 Description: Out or order replies might be lost on replay
2517 Details : In ptlrpc_retain_replayable_request if we cannot find retained
2518 request with tid smaller then one currently being added, add it
2519 to the start, not end of the list.
2521 --------------------------------------------------------------------------------
2523 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2525 * Support for kernels:
2526 2.6.5-7.283 (SLES 9),
2527 2.6.9-55.EL (RHEL 4),
2528 2.6.16.46-0.14 (SLES 10),
2529 2.6.18.8 vanilla (kernel.org)
2530 * Client support for unpatched kernels:
2531 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2532 2.6.16 - 2.6.22 vanilla (kernel.org)
2533 * Due to recently discovered recovery problems, we do not recommend
2534 using patchless RHEL 4 clients with this or any earlier release.
2535 * Recommended e2fsprogs version: 1.39.cfs8
2536 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2537 * Starting with this release, the ldiskfs backing filesystem required
2538 by Lustre is now in its own package, lustre-ldiskfs. This package
2539 should be installed. It is versioned separately from Lustre and
2540 may be released separately in future.
2544 Description: block reactivating mgc import until all deactivates complete
2545 Details : Fix race when failing back MDT/MGS to itself (testing)
2548 Frequency : at statup only
2550 Description: mds_lov_synchronize race leads to various problems
2551 Details : simultaneous MDT->OST connections at startup can cause the
2552 sync to abort, leaving the OSC in a bad state.
2554 Severity : enhancement
2556 Description: add optional extra BUILD_VERSION info
2557 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2558 to override the lustre version.
2561 Frequency : 2.6.18 servers only
2563 Description: ll_kern_mount() doesn't release the module reference
2564 Details : The ldiskfs module reference count never drops down to 0
2565 because ll_kern_mount() doesn't release the module reference.
2570 Description: server LBUG when using old ost_num_threads parameter
2571 Details : Accept the old ost_num_threads parameter but warn that it
2572 is deprecated, and fix an off-by-one error that caused an LBUG.
2577 Description: Transient SCSI error results in persistent IO issue
2578 Details : iobuf->dr_error is not reinitialized to 0 between two
2582 Frequency : sometimes when underlying device returns I/O errors
2584 Description: OSTs not going read-only during write failures
2585 Details : OSTs are not remounted read-only when the journal commit threads
2586 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2587 instead of the ext3 wrappers.
2591 Description: poor connect scaling with increasing client count
2592 Details : Don't run filter_grant_sanity_check for more than 100 exports
2593 to improve scaling for large numbers of clients.
2596 Frequency : SLES10 only
2598 Description: sanity-quota.sh quotacheck failed: rc = -22
2599 Details : Quotas cannot be enabled on SLES10.
2602 Frequency : liblustre clients only
2604 Description: getdirentries does not give error when run on compute nodes
2605 Details : getdirentries does not fail when the size specified as an argument
2606 is too small to contain at least one entry
2608 Severity : enhancement
2610 Description: Add LNET router traceability for debug purposes
2611 Details : If a checksum failure occurs with a router as part of the
2612 IO path, the NID of the last router that forwarded the bulk data
2613 is printed so it can be identified.
2618 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2619 Details : Due to a race condition, liblustre clients were occasionally
2620 evicted incorrectly.
2622 Severity : enhancement
2624 Description: lfs setstripe use optional parameters instead of postional
2627 Severity : enhancement
2629 Description: Nanosecond timestamp support for ldiskfs
2630 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2631 resolution timestamps. There is not yet support for this at
2632 the Lustre filesystem level.
2635 Frequency : during server recovery
2637 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2638 Details : request with rq_no_resend flag not awake l_wait_event if they get
2642 Frequency : nfs export on patchless client
2644 Description: connectathon hang when test nfs export over patchless client
2645 Details : Disconnected dentry cannot be found with lookup, so we do not need
2646 to unhash it or make it invalid
2649 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2650 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2651 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2656 Description: open req refcounting wrong on reconnect
2657 Details : If reconnect happened between getting open reply from server and
2658 call to mdc_set_replay_data in ll_file_open, we will schedule
2659 replay for unreferenced request that we are about to free.
2660 Subsequent close will crash in variety of ways.
2661 Check that request is still eligible for replay in
2662 mdc_set_replay_data().
2667 Description: disable writes to filesystem when reading health_check file
2668 Details : the default for reading the health_check proc file has changed
2669 to NOT do a journal transaction and write to disk, because this
2670 can cause reads of the /proc file to hang and block HA state
2671 checking on a healthy but otherwise heavily loaded system. It
2672 is possible to return to the previous behaviour during configure
2673 with --enable-health-write.
2675 Severity : enhancement
2677 Description: 64-bit inode version
2678 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2679 to the inode. This will be required for version-based recovery.
2684 Description: MDS fails to start if a duplicate client export is detected
2685 Details : in some rare cases it was possible for a client to connect to
2686 an MDS multiple times. Upon recovery the MDS would detect this
2687 and fail during startup. Handle this more gracefully.
2689 Severity : enhancement
2691 Description: Add -o localflock option to simulate old noflock
2693 Details : This will achieve local-only flock/fcntl locks
2699 Description: log_commit_thread vs filter_destroy race leads to crash
2700 Details : Take import reference before releasing llog record semaphore
2705 Description: Wrong request locking in request set processing
2706 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2707 imp_delayed_list, in this place should be used imp_lock.
2710 Frequency : when reconnection
2712 Description: Grant Leak when osc reconnect to OST
2713 Details : When osc reconnect ost, OST(filter) should check whether it
2714 should grant more space to client by comparing fed_grant and
2715 cl_avail_grant, and return the granted space to client instead
2716 of "new granted" space, because client will call osc_init_grant
2717 to update the client grant space info.
2720 Frequency : when client reconnect to OST
2722 Description: Grant Leak when osc do resend and replay bulk write
2723 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2724 bulk write request, because the grant info while be sync between
2725 OSC and OST when reconnect, and we should ignore the grant info
2726 these of resend/replay write req.
2731 Description: Grant space more than avaiable left space sometimes.
2732 Details : When then OST is about to be full, if two bulk writing from
2733 different clients came to OST. Accord the avaliable space of the
2734 OST, the first req should be permitted, and the second one
2735 should be denied by ENOSPC. But if the seconde arrived before
2736 the first one is commited. The OST might wrongly permit second
2737 writing, which will cause grant space > avaiable space.
2740 Frequency : when client is evicted
2742 Description: Grant might be wrongly erased when osc is evicted by OST
2743 Details : when the import is evicted by server, it will fork another
2744 thread ptlrpc_invalidate_import_thread to invalidate the
2745 import, where the grant will be set to 0. While the original
2746 thread will update the grant it got when connecting. So if
2747 the former happened latter, the grant will be wrongly errased
2748 because of this race.
2753 Description: Checking Stale with correct fid
2754 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2755 to check whether it is stale, because sometimes, we want the
2756 enqueue happened anyway, and op_data.fid2 will not be initialized.
2758 Severity : enhancement
2760 Description: update patchless client
2761 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2764 Frequency : only with 2.4 kernel
2766 Description: random memory corruption
2767 Details : size of struct ll_inode_info is to big for union inode.u and this
2768 can be cause of random memory corruption.
2773 Description: Memory leak in recovery
2774 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2775 It should also check obd_fail before fsfilt_start, otherwise if
2776 fsfilt_start return -EROFS,(failover mds during mds recovery).
2777 then the req will return with repmsg->transno = 0 and rc = EROFS.
2778 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2779 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2780 be freed no matter whether fsfilt_commit success or not.
2783 Frequency : only with huge count clients
2785 Description: Prevents from taking the superblock lock in llap_from_page for
2787 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2788 for llap_from_page call in ll_removepage prevents from taking the
2789 superblock lock for a soon died page.
2794 Description: Not check open intent error before release open handle
2795 Details : in some rare cases, the open intent error is not checked before
2796 release open handle, which may cause
2797 ASSERTION(open_req->rq_transno != 0), because it tries to release
2798 the failed open handle.
2803 Description: Set cat log bitmap only after create log success.
2804 Details : in some rare cases, the cat log bitmap is set too early. and it
2805 should be set only after create log success.
2809 Description: Accessing a block bevice can re-enable I/O when Lustre is
2810 tearing down a device.
2811 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2816 Description: service threads may hog cpus when there are a lot of requests
2817 Details : Insert cond_resched to give other threads a chance to use some CPU
2822 Description: the cat log was not initialized in recovery
2823 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2824 unlink log on mds will not be initialized until mds post
2825 recovery. And also in mds post recovery, the unlink log will
2826 initialization will be done asynchronausly, so there will be race
2827 between add unlink log and unlink log initialization.
2831 Description: brw_stats were being printed incorrectly
2832 Details : brw_stats were being printed as log2 but all of them were not
2833 recorded as log2. Also remove some code duplication arising from
2834 filter_tally_{read,write}.
2838 Frequency : rare, only in recovery.
2839 Description: ASSERTION(req->rq_type != LI_POISON) failed
2840 Details : imp_lock should be held while iterating over imp_sending_list for
2841 prevent destroy request after get timeout in ptlrpc_queue_wait.
2845 Description: replay-single.sh test 52 fails
2846 Details : A lock's skiplist need to be cleanup when it being unlinked
2847 from its resource list.
2851 Description: Short directio read returns full requested size rather than
2853 Details : Direct I/O operations should return actual amount of bytes
2854 transferred rather than requested size.
2856 Severity : enhancement
2858 Description: metadata RPC reduction (e.g. for rm performance)
2859 Details : decrease the amount of synchronous RPC between clients and servers
2860 by canceling conflicing lock before the operation on the client
2861 and packing thier handles into the main operation RPC to server.
2863 Severity : enhancement
2865 Description: Async OSC create to avoid the blocking unnecessarily.
2866 Details : If a OST has no remain object, system will block on the creating
2867 when need to create a new object on this OST. Now, ways use
2868 pre-created objects when available, instead of blocking on an
2869 empty osc while others are not empty. If we must block, we block
2870 for the shortest possible period of time.
2874 Description: improve handling recoverable errors
2875 Details : if request processig with error which can be recoverable on server
2876 request should be resend, otherwise page released from cache and
2879 Severity : enhancement
2881 Description: refine locking for avoid write wrong info into lov_objid file
2882 Details : fix possible races with add new target and write/update data in
2885 --------------------------------------------------------------------------------
2887 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2892 Frequency : on some architectures
2894 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2895 Details : Uninitialized flags sometimes cause configuration commands to
2899 Frequency : patchless clients only
2901 Description: missing __iget() symbol export
2902 Details : The __iget() symbol export is missing. To avoid the need for
2903 this on patchless clients the deathrow inode reaper is turned
2904 off, and we depend on the VM to clean up old inodes. This
2905 dependency was during via the fix for bug 12181.
2907 --------------------------------------------------------------------------------
2909 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2911 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2912 INTEROPERATE with older versions automatically. In many cases a
2913 special upgrade step is needed. Please read the
2914 user documentation before upgrading any part of a live system.
2915 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2916 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2917 * WARNING: Lustre configuration and startup changes are required with
2918 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2922 Severity : enhancement
2924 Description: MountConf
2925 Details : Lustre configuration is now managed via mkfs and mount
2926 commands instead of lmc and lconf. New obd types (MGS, MGC)
2927 are added for dynamic configuration management. See
2928 https://mail.clusterfs.com/wikis/lustre/MountConf for
2931 Severity : enhancement
2933 Description: dynamic OST addition
2934 Details : OSTs can now be added to a live filesystem
2936 Severity : enhancement
2938 Description: startup order invariance
2939 Details : MDTs and OSTs can be started in any order. Clients only
2940 require the MDT to complete startup.
2942 Severity : enhancement
2944 Description: parallel, asynchronous orphan cleanup
2945 Details : orphan cleanup is now performed in separate threads for each
2946 OST, allowing parallel non-blocking operation.
2948 Severity : enhancement
2950 Description: optimized stripe assignment
2951 Details : stripe assignments are now made based on ost space available,
2952 ost previous usage, and OSS previous usage, in order to try
2953 to optimize storage space and networking resources.
2955 Severity : enhancement
2957 Description: Permanently set tunables
2958 Details : All writable /proc/fs/lustre tunables can now be permanently
2959 set on a per-server basis, at mkfs time or on a live system.
2961 Severity : enhancement
2963 Description: Lustre message v2
2964 Details : Add lustre message format v2.
2966 Severity : enhancement
2968 Description: client OST exclusion list
2969 Details : Clients can be started with a list of OSTs that should be
2970 declared "inactive" for known non-responsive OSTs.
2974 Description: SPEC SFS validation failure on NFS v2 over lustre.
2975 Details : Changes the blocksize for regular files to be 2x RPC size,
2976 and not depend on stripe size.
2978 Severity : enhancement
2980 Description: Multiple MD RPCs in flight.
2981 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2982 To avoid overly-overloading MDS, introduce a limit on number of
2983 MDS RPCs in flight for a single client and add /proc controls
2984 to adjust this limit.
2986 Severity : enhancement
2988 Description: client read/write statistics
2989 Details : Add client read/write call usage stats for performance
2990 analysis of user processes.
2991 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2992 file access. extents_stats shows chunk size distribution.
2993 extents_stats_per_process show chunk size distribution per
2996 Severity : enhancement
2998 Description: mds statistics
2999 Details : Add detailed mds operations statistics in
3000 /proc/fs/lustre/mds/*/stats.
3004 Description: Failure of copying files with lustre special EAs.
3005 Details : Client side always return success for setxattr call for lustre
3006 special xattr (currently only "trusted.lov").
3011 Description: Refcount LNET uuids
3012 Details : The global LNET uuid list grew linearly with every startup;
3013 refcount repeated list entries instead of always adding to
3016 Severity : enhancement
3018 Description: Dynamic service threads
3019 Details : Within a small range, start extra service threads
3020 automatically when the request queue builds up.
3023 Frequency : mixed-endian client/server environments
3025 Description: mixed-endian crashes
3026 Details : The new msg_v2 system had some failures in mixed-endian
3029 Severity : enhancement
3031 Description: Easy OST removal
3032 Details : OSTs can be permanently deactivated with e.g. 'lctl
3033 conf_param lustre-OST0001.osc.active=0'
3035 Severity : enhancement
3037 Description: MGS proc entries
3038 Details : Added basic proc entries for the MGS showing what filesystems
3041 Severity : enhancement
3043 Description: provide MGS failover
3044 Details : Added config lock reacquisition after MGS server failover.
3046 Severity : enhancement
3048 Description: add Linux 2.4 support
3049 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
3053 Description: a large application tries to do I/O to the same resource and dies
3054 in the middle of it.
3055 Details : Check the req->rq_arrival time after the call to
3056 ost_brw_lock_get(), but before we do anything about
3057 processing it & sending the BULK transfer request. This
3058 should help move old stale pending locks off the queue as
3059 quickly as obd_timeout.
3062 Frequency : when an incorrect nid is specified during startup
3064 Description: ptlrpc connect to non-existant node causes kernel crash
3065 Details : LNET can't be re-entered from an event callback, which
3066 happened when we expire a message after the export has been
3067 cleaned up. Instead, hand the zombie cleanup off to another
3070 Severity : enhancement
3072 Description: plain/inodebits lock performance improvement
3073 Details : Grouping plain/inodebits in granted list by their request modes
3074 and bits policy, thus improving the performance of search through
3078 Frequency : only if OST filesystem is corrupted
3080 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3081 Details : for a short time RPCs with bulk IO are in the replay list,
3082 but replay of bulk IOs is unimplemented. If the OST filesystem
3083 is corrupted due to disk cache incoherency and then replay is
3084 started it is possible to trip an assertion. Avoid putting
3085 committed RPCs into the replay list at all to avoid this issue.
3088 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3090 Description: System hang on startup
3091 Details : This bug allowed the liblustre (e.g. catamount) client to
3092 return to the app before handling all startup RPCs. This
3093 could leave the node unresponsive to lustre network traffic
3094 and manifested as a server ptllnd timeout.
3096 Severity : enhancement
3098 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3099 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
3100 boolean to control whether to print peer debug info when a
3101 client's RPC times out.
3104 Frequency : only for kernels with patches from Lustre below 1.4.3
3106 Description: Remove old rdonly API
3107 Details : Remove old rdonly API which unused from at least lustre 1.4.3
3110 Frequency : only for devices with external journals
3112 Description: Set external device read-only also
3113 Details : During a commanded failover stop, we set the disk device
3114 read-only while the server shuts down. We now also set any
3115 external journal device read-only at the same time.
3118 Frequency : when upgrading from 1.4 while trying to change parameters
3120 Description: The wrong (new) MDC name was used when setting parameters for
3121 upgraded MDT's. Also allows changing of OSC (and MDC)
3122 parameters if --writeconf is specified at tunefs upgrade time.
3125 Frequency : when setting specific ost indicies
3127 Description: QOS code breaks on skipped indicies
3128 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3129 created with --index need not be sequential.
3131 Severity : enhancement
3133 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3134 Details : The uninit_groups feature works in conjunction with the kernel
3135 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3136 up the pass1 processing of e2fsck. This is a read-only feature
3137 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3138 mount filesystems that have had this feature enabled.
3140 Severity : enhancement
3142 Description: Improve multi-block allocation algorithm to avoid fragmentation
3143 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3144 allocation locality and avoid filesystem fragmentation.
3146 ------------------------------------------------------------------------------
3148 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3150 * Support for kernels:
3151 2.4.21-47.0.1.EL (RHEL 3)
3152 2.6.5-7.283 (SLES 9)
3153 2.6.9-42.0.10.EL (RHEL 4)
3154 2.6.12.6 vanilla (kernel.org)
3155 2.6.16.27-0.9 (SLES 10)
3156 * Recommended e2fsprogs version: 1.39.cfs5
3158 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3162 Frequency : occasional, depends on client load and configuration
3163 Bugzilla : 12181, 12203
3164 Description: data loss for recently-modified files
3166 Details : In some cases it is possible that recently written or created
3167 files may not be written to disk in a timely manner (this should
3168 normally be within 30s unless client IO load is very high).
3169 The problem appears as zero-length files or files that are a
3170 multiple of 1MB in size after a client crash or client eviction
3171 that are missing data at the end of the file.
3173 This problem is more likely to be hit on clients where files are
3174 repeatedly created and unlinked in the same directory, clients
3175 have a large amount of RAM, have many CPUs, the filesystem has
3176 many OSTs, the clients are rebooted frequently, and/or the files
3177 are not accessed by other nodes after being written.
3179 The presence of the problem can be detected by looking at
3180 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
3181 smaller than the second (nr_unused) then dirty files will not
3182 be flushed automatically to disk. "sync; sleep 10" should be
3183 run several times on the node before unmounting it to update
3184 Lustre (this is also safe to run on nodes without this problem).
3186 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
3187 that can cause this same problem, so customers using that kernel
3188 also need to update the kernel in addition to Lustre. In order
3189 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3191 It is normal that files written just before a client crash (less
3192 than 30s) may not yet have been flushed to disk, even for local
3196 Frequency : frequent on thin XT3 nodes
3198 Description: UUID collision on thin XT3 Linux nodes
3199 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3200 randomly, since we relied on an insufficiently-seeded PRNG.
3205 Description: OSS hangs after "All ost request buffers busy"
3206 Details : A deadlock between quota and journal operations caused OSS
3207 hangs after printing "All ost request buffers busy."
3210 Frequency : always on liblustre builds
3212 Description: Cleanup compiler warnings on liblustre
3215 Frequency : always on liblustre builds on XT3
3217 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3221 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3222 > 32000 subdirectories
3223 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3224 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3225 subdirectory count crosses 32000. This will aid e2fsck to
3226 correctly handle more than 32000 subdirectories.
3229 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3231 Description: System hang on startup
3232 Details : This bug allowed the liblustre (e.g. catamount) client to
3233 return to the app before handling all startup RPCs. This
3234 could leave the node unresponsive to lustre network traffic
3235 and manifested as a server ptllnd timeout.
3237 Severity : enhancement
3239 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3240 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3241 boolean to control whether to print peer debug info when a
3242 client's RPC times out.
3247 Description: make O_SYNC working on 2.6 kernels
3248 Details : 2.6 kernels use different method for mark pages for write,
3249 so need add a code to lustre for O_SYNC work.
3254 Description: Failure to close file and release space on NFS
3255 Details : Put inode details into lock acquired in ll_intent_file_open.
3256 Use mdc_intent_lock in ll_intent_open to properly
3257 detect all kind of errors unhandled by mdc_enqueue.
3262 Description: proc file read during shutdown sometimes raced obd removal,
3264 Details : Add lock to prevent obd access after proc file removal.
3267 Frequency : Only for files larger than 4GB on 32-bit clients.
3269 Description: improperly doing page alignment of locks
3270 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3271 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3276 Description: RPCs being resent when they shouldn't be
3277 Details : Some RPCs that should not be resent are being resent. This
3278 can cause inconsistencies in the RPC state machine. Do not
3279 resend such requests.
3282 Frequency : rare, only with NFS export
3284 Description: Crash on NFS re-export node
3285 Details : under very unusual load conditions an assertion is hit in
3286 ll_intent_file_open()
3289 Frequency : only if OST filesystem is corrupted
3291 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3292 Details : for a short time RPCs with bulk IO are in the replay list,
3293 but replay of bulk IOs is unimplemented. If the OST filesystem
3294 is corrupted due to disk cache incoherency and then replay is
3295 started it is possible to trip an assertion. Avoid putting
3296 committed RPCs into the replay list at all to avoid this issue.
3301 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3302 Details : Large single O_DIRECT read and write calls can fail to allocate
3303 a sufficiently large buffer to process the request. In case of
3304 allocation failure the allocation is retried with a smaller
3305 buffer and broken into smaller requests.
3307 Severity : enhancement
3309 Description: Add -o localflock option to simulate old noflock behaviour.
3310 Details : This will achieve local-only flock/fcntl locks coherentness.
3315 Description: versioning check is incomplete
3316 Details : Checking the version difference of client vs. server, report
3317 error if the gap is too big.
3322 Description: add support PG_writeback bit
3323 Details : add support for PG_writeback bit for Lustre, for more carefull
3324 work with page cache in 2.6 kernel. This also fix some deadlocks
3325 and remove hack for work O_SYNC with 2.6 kernel.
3327 Severity : enhancement
3329 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3330 Details : The uninit_groups feature works in conjunction with the kernel
3331 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3332 up the pass1 processing of e2fsck. This is a read-only feature
3333 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3334 mount filesystems that have had this feature enabled.
3336 Severity : enhancement
3338 Description: Improve multi-block allocation algorithm to avoid fragmentation
3339 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3340 allocation locality and avoid filesystem fragmentation.
3342 ------------------------------------------------------------------------------
3344 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3346 * Support for kernels:
3347 2.6.9-42.0.3.EL (RHEL 4)
3348 2.6.5-7.276 (SLES 9)
3349 2.4.21-47.0.1.EL (RHEL 3)
3350 2.6.12.6 vanilla (kernel.org)
3351 2.6.16.21-0.8 (SLES10)
3352 * Recommended e2fsprogs version: 1.39.cfs2-0
3354 * The backwards-compatible /proc/sys/portals symlink has been removed
3355 in this release. Before upgrading, please ensure that you change
3356 any configuration scripts or /etc/sysctl.conf files that access
3357 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3358 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3359 in advance of the upgrade on any system running Lustre 1.4.6 or
3360 newer, since /proc/sys/lnet was added in that version.
3361 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3365 Frequency : only when quota is used
3367 Description: avoid scanning export list for quota master
3368 Details : Change the algorithms to avoid scanning export list in order
3369 to improve the efficiency.
3372 Frequency : MDS failover only, very rarely
3374 Description: "went back in time" messages on mds failover
3375 Details : The greatest transno may be lost when the current operation
3376 finishes with an error (transno==0) and the client's last_rcvd
3377 record is over-written. Save the greatest transno in the
3378 mds_last_transno for this case.
3381 Frequency : always for specific kernels and striping counts
3383 Description: client may get "Matching packet too big" without ACL support
3384 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3385 when trying to access files in certain configurations. The
3386 clients should in fact be denied when mounting because they do
3387 not understand ACLs.
3390 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3392 Description: many clients connecting with IO in progress causes connect timeouts
3393 Details : Avoid synchronous journal commits to avoid delays caused by many
3394 clients connecting/disconnecting when bulk IO is in progress.
3395 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3396 OST_IO_PORTAL to avoid delays behind potentially many pending
3400 Frequency : occasionally with multiple writers to a single file
3402 Description: shared writes to file may result in wrong size reported by stat()
3403 Details : Allow growing of kms when extent lock is cancelled
3406 Frequency : always with random mmap IO to multi-striped file
3408 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3409 Details : Only if the hole is at the end of OST object so that kms is too
3410 small. Fix is to increase kms accordingly in ll_nopage.
3413 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3415 Description: writes to a missing object would leak memory on the OST
3416 Details : If there is an inconsistency between the MDS and OST filesystems,
3417 such that the MDS references an object that doesn't exist, writes
3418 to that object will leak memory due to incorrect cleanup in the
3419 error handling path, eventually running out of memory on the OST.
3424 Description: Creating too long symlink causes lustre errors
3425 Details : Check symlink and name lengths before sending requests to MDS.
3428 Frequency : only if flock is enabled (not on by default)
3430 Description: posix locks not released on fd closure on 2.6.9+
3431 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3432 kernels, this caused such locks not to be released on fd close and
3433 then assertions on fs unmount about still used locks.
3436 Frequency : MDS failover only, very rarely
3438 Description: clients may get ASSERTION(granted_lock != NULL)
3439 Details : When request was taking a long time, and a client was resending
3440 a getattr by name lock request. The were multiple lock requests
3441 with the same client lock handle and
3442 mds_getattr_name->fixup_handle_for_resent_request found one of the
3443 lock handles but later failed with ASSERTION(granted_lock != NULL).
3448 Description: handle->h_buffer_credits > 0, assertion failure
3449 Details : h_buffer_credits is zero after truncate, causing assertion
3450 failure. This patch extends the transaction or creates a new
3454 Frequency : NFS re-export or patchless client
3455 Bugzilla : 11179, 10796
3456 Description: Crash on NFS re-export node (__d_move)
3457 Details : We do not want to hash the dentry if we don't have a lock.
3458 But if this dentry is later used in d_move, we'd hit uninitialised
3459 list head d_hash, so we just do this to init d_hash field but
3460 leave dentry unhashed.
3463 Frequency : NFS re-export or patchless client
3465 Description: NFS exports has problem with symbolic link
3466 Details : lustre client didn't properly install dentry when re-exported
3467 to NFS or running patchless client.
3470 Frequency : NFS re-export or patchless client
3472 Description: Various nfs/patchless fixes.
3473 Details : fixes reuse disconected alias for lookup process - this fixes
3474 warning "find_exported_dentry: npd != pd",
3475 fix permission error with open files at nfs.
3476 fix apply umask when do revalidate.
3479 Frequency : occasional
3481 Description: Crash on NFS re-export node
3482 Details : calling clear_page() on the wrong pointer triggered oops in
3483 generic_mapping_read().
3486 Frequency : rarely, using O_DIRECT IO
3488 Description: unaligned directio crashes client with LASSERT
3489 Details : check for unaligned buffers before trying any requests.
3492 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3494 Description: stale data returned from RAID cache
3495 Details : If only a small amount of IO is done to the RAID device before
3496 reading it again it is possible to get stale data from the RAID
3497 cache instead of reading it from disk.
3500 Frequency : always for sles10 kernel
3502 Description: sles10 support
3503 Details : ll_follow_link: compile fixes and using of nd_set_link
3504 under newer kernels.
3507 Frequency : depends on arch, kernel and compiler version, always on sles10
3510 Description: recursive or deep enough symlinks cause stack overflow
3511 Details : getting rid of large stack-allocated variable in
3515 Frequency : depends on hardware
3517 Description: lustre write performance loss in the SLES10 kernel
3518 Details : the performance loss is caused by using of write barriers in the
3519 ext3 code. The SLES10 kernel turns barrier support on by
3520 default. The fix is to undo that change for ldiskfs.
3522 ------------------------------------------------------------------------------
3524 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3526 * Support for kernels:
3527 2.6.9-42.0.3EL (RHEL 4)
3528 2.6.5-7.276 (SLES 9)
3529 2.4.21-47.0.1.EL (RHEL 3)
3530 2.6.12.6 vanilla (kernel.org)
3534 Frequency : quota enabled and large files being deleted
3536 Description: releasing more than 4GB of quota at once hangs OST
3537 Details : If a user deletes more than 4GB of files on a single OST it
3538 will cause the OST to spin in an infinite loop. Release
3539 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3544 Description: statfs data retrieved from /proc may be stale or zero
3545 Details : When reading per-device statfs data from /proc, in the
3546 {kbytes,files}_{total,free,avail} files, it may appear
3547 as zero or be out of date.
3550 Frequency : systems with MD RAID1 external journal devices
3552 Description: lconf's call to blkid is confused by RAID1 journal devices
3553 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3554 of returning all block devices that match the journal UUID.
3557 Frequency : always, for aggregate stripe size over 4GB
3559 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3560 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3561 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3564 Frequency : always if "lfs find" used on a local file/directory
3566 Description: "lfs find" segfaults if used on a local file/directory
3567 Details : The case where a directory component was not specified wasn't
3568 handled correctly. Handle this properly.
3571 Frequency : always on ppc64
3573 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3574 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3575 perform endianness conversion.
3578 Frequency : rarely (truncate to non-zero file size after write under load)
3579 Bugzilla : 10730, 10687
3580 Description: Files padded with zeros to next 4K multiple
3581 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3582 it is possible that files that are truncated to a non-zero size
3583 immediately after being written are filled with zero bytes beyond
3584 the truncated size. No file data is lost.
3586 Severity : enhancement
3588 Description: Allow recovery/failover for liblustre clients.
3589 Details : liblustre clients were unaware of failover configurations until
3592 Severity : enhancement
3594 Description: user file locks should fail when not mounting with flock option
3595 Details : Set up an error-returning stub in ll_file_operations.lock field
3596 to prevent incorrect behaviour when client is mounted without
3597 flock option. Also, set up properly f_op->flock field for
3601 Frequency : always on ia64
3603 Description: "lfs df" loops on printing out MDS statfs information
3604 Details : The obd_ioctl_data was not initialized and in some systems
3605 this caused a failure during the ioctl that did not return
3606 an error. Initialize the struct and return an error on failure.
3609 Frequency : SLES 9 only
3611 Description: Error of copying files with lustre special EAs as root
3612 Details : Client side always return success for setxattr call for lustre
3613 special xattr (currently only "trusted.lov").
3616 Frequency : rarely on clusters with both ia64+i386 clients
3618 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3619 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3620 clients are doing concurrent writes to the same file it is
3621 possible that the ia64 clients may LASSERT because the OST
3622 extent locks are not PAGE_SIZE aligned. Ensure that grown
3623 locks are always aligned on the request boundary.
3626 Frequency : specific use, occasional
3628 Description: Overwriting in use executable truncates on-disk binary image
3629 Details : If one node attempts to overwrite an executable in use by
3630 another node, we now correctly return ETXTBSY instead of
3631 truncating the file.
3633 Severity : enhancement
3635 Description: Async OSC create to avoid the blocking unnecessarily.
3636 Details : If a OST has no remain object, system will block on the creating
3637 when need to create a new object on this OST. Now, ways use
3638 pre-created objects when available, instead of blocking on an
3639 empty osc while others are not empty. If we must block, we block
3640 for the shortest possible period of time.
3645 Description: chmod on Lustre root is propagated to other clients
3646 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3647 invalid by the follow_mount time.
3650 Frequency : liblustre clients only
3652 Description: Race in 'instant cancel' lock handling could lead to such locks
3653 never to be granted in case of SMP MDS
3654 Details : Do not destroy not yet granted but cbpending locks in
3658 Frequency : replay/resend of open
3660 Description: non null lock assetion failure in mds_intent_policy
3661 Details : Trying to replay/resend lockless open requests resulted in
3662 mds_open() returning 0 with no lock. Now it sets a flag if
3663 a lock is going to be returned.
3665 Severity : enhancement
3667 Description: Checksum enhancements
3668 Details : New checksum enhancements allow for resending RPCs that failed
3671 Severity : enhancement
3673 Description: Tunables on number of dirty pages in cacche
3674 Details : Allow to set limit on number of dirty pages cached.
3679 Description: client crash on unmount - lock still has references
3680 Details : In some error handling cases it was possible to leak a lock
3681 reference on a client while accessing a file. This was not
3682 harmful to the client during operation, but would cause the
3683 client to crash when the filesystem is unmounted.
3686 Frequency : specific case, rare
3688 Description: ETXTBSY on mds though file not in use by client
3689 Details : ETXTBSY is no longer incorrectly returned when attempting to
3690 chmod or chown a directory that the user previously tried to
3691 execute or a currently-executing binary.
3694 Frequency : extremely rare except on liblustre-based clients
3696 Description: Lustre space not freed when files are deleted
3697 Details : Clean up open-unlinked files after client eviction. Previously
3698 the unlink was skipped and the files remained as orphans.
3703 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3704 Details : In some cases it was possible to send a blocking callback to
3705 a client doing a glimpse, even though that client didn't get
3706 a lock granted. When the glimpse lock is cancelled on the OST
3707 the freed lock is left on the waiting list and corrupted the list.
3710 Frequency : all core dumps
3712 Description: Broke core dumps to lustre
3713 Details : Negative dentry may be unhashed if parent does not have UPDATE
3714 lock, but some callers, e.g. do_coredump, expect dentry to be
3715 hashed after successful create, hash it in ll_create_it.
3717 ------------------------------------------------------------------------------
3719 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3721 * Support for kernels:
3722 2.6.9-42.0.2.EL (RHEL 4)
3723 2.6.5-7.276 (SLES 9)
3724 2.4.21-40.EL (RHEL 3)
3725 2.6.12.6 vanilla (kernel.org)
3729 Frequency : always on RHEL 3
3731 Description: Number of open files grows over time
3732 Details : The number of open files grows over time, whether or not
3733 Lustre is started. This was due to a filp leak introduced
3734 by one of our kernel patches.
3736 ------------------------------------------------------------------------------
3738 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3740 * Support for kernels:
3741 2.6.9-42.EL (RHEL 4)
3742 2.6.5-7.276 (SLES 9)
3743 2.4.21-40.EL (RHEL 3)
3744 2.6.12.6 vanilla (kernel.org)
3749 Bugzilla : 5719, 9635, 9792, 9684
3750 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3751 Details : If a server is under heavy load and cannot reply to new
3752 connection requests before the client resends the (re)connect,
3753 the connection handling code can behave badly if two service
3754 threads are concurrently handing separate (re)connections from
3755 the same client. Add better locking to the connection handling
3756 code, and ensure that only a single connection will be processed
3757 for a given client UUID, even if the lock is dropped.
3759 Severity : enhancement
3761 Description: add TCP zero-copy support to kernel
3762 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3763 sends if the hardware supports scatter-gather and checksumming.
3764 This allows socklnd to do client-write and server-read more
3765 efficiently and reduce CPU utilization from skbuf copying.
3768 Frequency : only if NFS exporting from client
3770 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3771 Details : If setting attributes on a file created under NFS that had
3772 never been opened it would be possible to oops the client
3773 if the file had no objects.
3776 Frequency : always for liblustre
3778 Description: liblustre client does MDS+OSTs setattr RPC for each write
3779 Details : When doing a write from a liblustre client, the client
3780 incorrectly issued an RPC to the MDS and each OST the file was
3781 striped over in order to update the timestamps. When writing
3782 with small chunks and many clients this could overwhelm the MDS
3783 with RPCs. In all cases it would slow down the write because
3784 these RPCs are unnecessary.
3786 Severity : enhancement
3788 Description: allow number of MDS service threads to be changed at module load
3789 Details : It is now possible to change the number of MDS service threads
3790 running. Adding "options mds mds_num_threads={N}" to the MDS's
3791 /etc/modprobe.conf will set the number of threads for the next
3792 time Lustre is restarted (assuming the "mds" module is also
3793 reloaded at that time). The default number of threads will
3794 stay the same, 32 for most systems.
3799 Description: OST crash if filesystem is unformatted or corrupt
3800 Details : If an OST is started on a device that has never been formatted
3801 or if the filesystem is corrupt and cannot even mount then the
3802 error handling cleanup routines would dereference a NULL pointer.
3807 Description: NULL pointer deref in llap_from_page.
3808 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3809 incorrect) mapping if the page was truncated/reclaimed while it was
3810 searched for. Check for this condition and skip such pages when
3811 doing readahead. Introduce extra check to llap_from_page() to
3812 verify page->mapping->host is non-NULL (so page is not anonymous).
3815 Frequency : Sometimes when using sys_sendfile
3817 Description: "page not covered by a lock" warnings from ll_readpage
3818 Details : sendfile called ll_readpage without right page locks present.
3819 Now we introduced ll_file_sendfile that does necessary locking
3820 around call to generic_file_sendfile() much like we do in
3824 Frequency : with certain MDS communication failures at client mount time
3826 Description: NULL pointer deref after failed client mount
3827 Details : a client connection request may delayed by the network layer
3828 and not be sent until after the PTLRPC layer has timed out the
3829 request. If the client fails the mount immediately it will try
3830 to clean up before the network times out the request. Add a
3831 reference from the request import to the obd device and delay
3832 the cleanup until the network drops the request.
3835 Frequency : occasionally during client (re)connect
3837 Description: assertion failure during client (re)connect
3838 Details : processing a client connection request may be delayed by the
3839 client or server longer than the client connect timeout. This
3840 causes the client to resend the connection request. If the
3841 original connection request is replied in this interval, the
3842 client may trip an assertion failure in ptlrpc_connect_interpret()
3843 which thought it would be the only running connect process.
3846 Frequency : only with obd_echo servers and clients that are rebooted
3848 Description: kernel BUG accessing uninitialized data structure
3849 Details : When running an obd_echo server it did not start the ping_evictor
3850 thread, and when a client was evicted an uninitialized data
3851 structure was accessed. Start the ping_evictor in the RPC
3852 service startup instead of the OBD startup.
3854 Severity : enhancement
3855 Bugzilla : 10193 (patchless)
3856 Description: Remove dependency on various unexported kernel interfaces.
3857 Details : No longer need reparent_to_init, exit_mm, exit_files,
3858 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3861 Frequency : rare (only users of deprecated and unsupported LDAP config)
3863 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3864 Details : LDAP apparently contains 'lustreName' attributes instead of
3865 'name'. A simple remapping of the name is sufficient.
3868 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3870 Description: waiting_locks_callback trips kernel BUG if client is evicted
3871 Details : Running with the dump_on_timeout debug flag turned on makes
3872 it possible that the waiting_locks_callback() can try to dump
3873 the Lustre kernel debug logs from an interrupt handler. Defer
3874 this log dumping to the expired_lock_main() thread.
3876 Severity : enhancement
3878 Description: Support NFS exporting on 2.6 kernels.
3879 Details : Implement non-rawops metadata methods for NFS server to use without
3880 changing NFS server code.
3883 Frequency : very rare (synthetic metadata workload only)
3885 Description: two racing renames might cause an MDS thread to deadlock
3886 Details : Running the "racer" program may cause one MDS thread to rename
3887 a file from being the source of a rename to being the target of
3888 a rename at exactly the same time that another thread is doing
3889 so, and the second thread has already enqueued these locks after
3890 doing a lookup of the target and is trying to relock them in
3891 order. Ensure that we don't try to re-lock the same resource.
3894 Frequency : only very large systems with liblustre clients
3896 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3897 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3898 all OSTs in parallel. This allows the eviction of stale liblustre
3899 clients to proceed much faster than if they were done in series,
3900 and also offers similar improvements for other set_info RPCs.
3905 Description: excessive CPU usage during initial read phase on client
3906 Details : During the initial read phase on a client, it would agressively
3907 retry readahead on the file, consuming too much CPU and impacting
3908 performance (since 1.4.5.8). Improve the readahead algorithm
3909 to avoid this, and also improve some other common cases (read
3910 of small files in particular, where "small" is files smaller than
3911 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3916 Description: MDS crash when receiving packet with unknown intent.
3917 Details : Do not LBUG in unknown intent case, just return -EFAULT
3919 Severity : enhancement
3920 Bugzilla : 9293, 9385
3921 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3922 Details : Do not serialize getattr (non-intent version) and statfs.
3925 Frequency : occasional, when OST network is overloaded/intermittent
3927 Description: client evicted by OST after bulk IO timeout
3928 Details : If a client sends a bulk IO request (read or write) the OST
3929 may evict the client if it is unresposive to its data GET/PUT
3930 request. This is incorrect if the network is overloaded (takes
3931 too long to transfer the RPC data) or dropped the OST GET/PUT
3932 request. There is no need to evict the client at all, since
3933 the pinger and/or lock callbacks will handle this, and the
3934 client can restart the bulk request.
3937 Frequency : Always when mmapping file with no objects
3939 Description: client crashes when mmapping file with no objects
3940 Details : Check that we actually have objects in a file before doing any
3941 operations on objects in ll_vm_open, ll_vm_close and
3947 Description: Request leak when working with deleted CWD
3948 Details : Introduce advanced request refcount tracking for requests
3949 referenced from lustre intent.
3951 Severity : Enhancement
3953 Description: Cache open file handles on client.
3954 Details : MDS now will return special lock along with openhandle, if
3955 requested and client is allowed to hold openhandle, even if unused,
3956 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3957 closing files for every read/write openration.
3959 Severity : Enhancement
3961 Description: Cache open negative dentries on client when possible.
3962 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3963 negative dentries on lock revocation.
3968 Description: Remounting a client read-only wasn't possible with a zconf mount
3969 Details : It wasn't possible to remount a client read-only with llmount.
3971 Severity : enhancement
3972 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3973 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3974 MPICH is included to add Lustre-specific ADIO interfaces.
3975 This is based closely on the UFS ADIO layer and only differs
3976 in file creation, in order to allow the OST striping to be set.
3977 This is user-contributed code and not supported by CFS.
3982 Description: extended inode attributes (immutable, append-only) work improperly
3983 when 2.4 and 2.6 kernels are used on client/server or vice versa
3984 Details : Introduce kernel-independent values for these flags.
3986 Severity : enhancement
3989 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3990 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3991 granularity. Now, Fractional MB granularity is be supported,
3992 this is very useful for low memory system.
3994 Severity : enhancement
3996 Description: Getattr by fid
3997 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3998 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3999 have locks and avoids getattr rpc requests for it on every stat.
4002 Frequency : Always, for filesystems larger than 2TB
4004 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
4005 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
4006 filesystems larger than 2TB to be mounted with mballoc enabled.
4009 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
4011 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
4012 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
4013 filesystems larger than 2TB is mounted then the kernel will
4014 silently corrupt the start of the filesystem. CONFIG_LBD is
4015 enabled for all CFS-supported kernels, but the possibility of
4016 this happening with a modified kernel config exists.
4018 Severity : enhancement
4020 Description: add client O_DIRECT support for 2.6 kernels
4021 Details : It is now possible to do O_DIRECT reads and writes to files
4022 in the Lustre client mountpoint on 2.6 kernel clients.
4024 Severity : enhancement
4026 Description: parallel glimpse, setattr, statfs, punch, destroy requests
4027 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
4028 parallel, not waiting for response from every OST before sending
4029 a rpc to the next OST.
4034 Description: setattr vs write race when updating file timestamps
4035 Details : Client processes that update a file timestamp into the past
4036 right after writing to the file (e.g. tar) it is possible that
4037 the updated file modification time can be reset to the current
4038 time due to a race between processing the setattr and write RPC.
4040 Severity : enhancement
4042 Description: Bring 'lfs find' closer in line with regular Linux find.
4043 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
4044 -print0 options and obtains all the needed info through the lustre
4047 Severity : enhancement
4049 Description: support up to 1024 configured devices on one node
4050 Details : change obd_dev array from statically allocated to dynamically
4051 allocated structs as they are first used to reduce memory usage
4056 Description: Flush dirty partially truncated pages during truncate
4057 Details : Immediatelly flush partially truncated pages in filter_setattr,
4058 this way we completely avoid having any pages in page cache on OST
4059 and can retire ugly workarounds during writes to flush such pages.
4064 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
4065 Details : protect inode from truncation within vfs_unlink() context
4066 just take a reference before calling vfs_unlink() and release it
4067 when parent's i_sem is free.
4072 Description: last_id value checked outside lock on OST caused LASSERT failure
4073 Details : If there were multiple MDS->OST object precreate requests in
4074 flight, it was possible that the OST's last object id was checked
4075 outside a lock and incorrectly tripped an assertion. Move checks
4076 inside locks, and discard old precreate requests.
4079 Frequency : always, if extents are used on OSTs
4081 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
4082 Details : OSTs using the extents format would not zero the high 16 bits of
4083 the index physical block number. This is not a problem for any
4084 OST filesystems smaller than 16TB, and no kernels support ext3
4085 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
4086 new/modified files) and can be fixed for existing filesystems
4087 with e2fsprogs-1.39-cfs1.
4092 Description: import connection selection may be incorrect if timer wraps
4093 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
4094 import connections to be ignored if the 32-bit jiffies counter
4095 wraps. Use a 64-bit jiffies counter.
4098 Frequency : very large clusters immediately after boot
4100 Description: LNET request buffers exhausted under heavy short-term load
4101 Details : If a large number of client requests are generated on a service
4102 that has previously never seen so many requests it is possible
4103 that the request buffer growth cannot keep up with the spike in
4104 demand. Instead of dropping incoming requests, they are held in
4105 the LND until the RPC service can accept more requests.
4108 Frequency : Sometimes during replay
4110 Description: Assertion failure in ll_local_open after replay.
4111 Details : If replay happened on an open request reply before we were able
4112 to set replay handler, reply will become not swabbed tripping the
4113 assertion in ll_local_open. Now we set the handler right after
4114 recognising of open request
4117 Frequency : very rare
4119 Description: kernel reports "badness in vsnprintf"
4120 Details : Reading from the "recovery_status" /proc file in small chunks
4121 may cause a negative length in lprocfs_obd_rd_recovery_status()
4122 call to vsnprintf() (which is otherwise harmless). Exit early
4123 if there is no more space in the output buffer.
4125 Severity : enhancement
4127 Description: clear OBD RPC statistics by writing to them
4128 Details : It is now possible to clear the OBD RPC statistics by writing
4129 to the "stats" file.
4134 Description: Client mtime is not the same on different clients after utimes
4135 Details : In some cases, the client was using the utimes() syscall on
4136 a file cached on another node. The clients now validate the
4137 ctime from the MDS + OSTs to determine which one is right.
4142 Description: Inability to activate failout mode
4143 Details : lconf script incorrectly assumed that in pythong string's numeric
4144 value is used in comparisons.
4147 Frequency : always with multiple stripes per file
4149 Description: Inefficient object allocation for mutli-stripe files
4150 Details : When selecting which OSTs to stripe files over, for files with
4151 a stripe count that divides evenly into the number of OSTs,
4152 the MDS is always picking the same starting OST for each file.
4153 Return the OST selection heuristic to the original design.
4158 Description: mount failures may take full timeout to return an error
4159 Details : Under some heavy load conditions it is possible that a
4160 failed mount can wait for the full obd_timeout interval,
4161 possibly several minutes, before reporting an error.
4162 Instead return an error as soon as the status is known.
4164 Frequency : quota enabled and large files being deleted
4166 Description: releasing more than 4GB of quota at once hangs OST
4167 Details : If a user deletes more than 4GB of files on a single OST it
4168 will cause the OST to spin in an infinite loop. Release
4169 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4174 Description: statfs data retrieved from /proc may be stale or zero
4175 Details : When reading per-device statfs data from /proc, in the
4176 {kbytes,files}_{total,free,avail} files, it may appear
4177 as zero or be out of date.
4180 Frequency : systems with MD RAID1 external journal devices
4182 Description: lconf's call to blkid is confused by RAID1 journal devices
4183 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4184 of returning all block devices that match the journal UUID.
4187 Frequency : always, for aggregate stripe size over 4GB
4189 Description: assertion fails when trying to use 4GB stripe size
4190 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
4191 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
4194 Frequency : always on ppc64
4196 Description: the first write on an ext3 filesystem with mballoc got stuck
4197 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4198 perform endianness conversion.
4200 ------------------------------------------------------------------------------
4202 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
4204 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4205 INTEROPERATE with older versions automatically. Please read the
4206 user documentation before upgrading any part of a live system.
4207 * WARNING: Lustre networking configuration changes are required with
4208 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4211 * Support for kernels:
4212 2.6.9-22.0.2.EL (RHEL 4)
4213 2.6.5-7.244 (SLES 9)
4214 2.6.12.6 vanilla (kernel.org)
4217 Severity : enhancement
4218 Bugzilla : 7981/8208
4219 Description: Introduced Lustre Networking (LNET)
4220 Details : LNET is new networking infrastructure for Lustre, it includes
4221 a reorganized network configuration mode (see the user
4222 documentation for full details) as well as support for routing
4223 between different network fabrics. Lustre Networking Devices
4224 (LNDs) for the supported network fabrics have also been
4225 created for this new infrastructure.
4227 Severity : enhancement
4228 Description: Introduced Access control lists
4229 Details : clients can set ACLs on files and directories in order to have
4230 more fine-grained permissions than the standard Unix UGO+RWX.
4231 The MDS must be started with the "-o acl" mount option.
4233 Severity : enhancement
4234 Description: Introduced filesystem quotas
4235 Details : Administrators may now establish per-user quotas on the
4238 Severity : enhancement
4240 Description: Configuration change for the XT3
4241 The PTLLND is now used to run Lustre over Portals on the XT3
4242 The configure option(s) --with-cray-portals are no longer used.
4243 Rather --with-portals=<path-to-portals-includes> is used to
4244 enable building on the XT3. In addition to enable XT3 specific
4245 features the option --enable-cray-xt3 must be used.
4250 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4251 Details : A race between allocating a new llcd and re-getting the llcd_lock
4252 allowed another thread to grab newly-allocated llcd.
4254 Severity : enhancement
4256 Description: 2.6 OST async journal commit and locking fix to improve performance
4257 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4258 2.6 kernels are now async as they already were in 2.4 kernels so
4259 that they can commit concurrently with the network bulk transfer.
4260 For block-allocated files the filter allocation semaphore is held
4261 to avoid filesystem fragmentation during allocation. BKL lock
4262 removed for 2.6 xattr operations where it is no longer needed.
4267 Description: lconf incorrectly determined whether two IP networks could talk
4268 Details : In some more complicated routing and multiple-network
4269 configurations, lconf will avoid trying to make a network
4270 connection to a disjoint part of the IP space. It was doing the
4271 math incorrectly for one set of cases.
4276 Description: Fix for potential infinite loop processing records in an llog.
4277 Details : If an llog record is corrupted/zeroed, it is possible to loop
4278 forever in llog_process(). Validate the llog record length
4279 and skip the remainder of the block on error.
4282 Frequency : occasional (liblustre only)
4284 Description: liblustre could not open files whose last component is a symlink
4285 Details : sysio_path_walk() would incorrectly pass the open intent to
4286 intermediate path components.
4289 Frequency : rare (liblustre only with non-standard tuning)
4290 Bugzilla : 7201 (7350)
4291 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4292 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4293 after it has been released, only do this for referenced locks
4295 Severity : enhancement
4297 Description: specifying an (invalid) directory default stripe_size of -1
4298 would reset the directory default striping
4299 Details : stripe_size -1 was used internally to signal directory stripe
4300 removal, now use "all default" to signal dir stripe removal
4301 as a directory striping of "all default" is not useful
4304 Frequency : common for large clusters running liblustre clients
4306 Description: doing an ls when liblustre clients are running is slow
4307 Details : sending a glimpse AST to a liblustre client waits for every AST
4308 to time out, as liblustre clients will not respond. Since they
4309 cannot cache data we refresh the OST lock LVB from disk instead.
4311 Severity : enhancement
4313 Description: doing an ls at the same time as file IO can be slow
4314 Details : enqueue and other "small" requests can be blocked behind many
4315 large IO requests. Create a new OST IO portal for non-IO
4316 requests so they can be processed faster.
4319 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4321 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4322 created file with mode 000
4323 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4324 allow an ftruncate() as a non-root user to a file with mode 000.
4325 Lustre now respects this flag to disable mode checks when
4326 truncating a file owned by the user
4329 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4332 Description: Revoking locks from clients that went dead or catatonic might take
4334 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4335 cancellation of such locks instant on servers without waiting for
4336 any reply from clients. Clients drops these locks when cancel
4337 notification from server is received without replying.
4340 Frequency : liblustre-only, when liblustre client dies or becomes busy
4342 Description: Doing ls on Linux clients can take a long time with active
4344 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4345 granting such locks to it in the first place if possible. Locks
4346 are taken by proxy on the OST during the read or write and
4347 dropped immediately afterward. Add connect flags handling, do
4348 not grant locks to liblustre clients for glimpse ASTs.
4350 Severity : enhancement
4352 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4353 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4354 sequential reads and behaves badly with applications doing random
4355 reads. Improve it by reading ahead at least read region, and
4356 avoiding excessive large RPC for small reads.
4358 Severity : enhancement
4360 Description: Creating more than 1000 files for a single job may cause a load
4361 imbalance on the OSTs if there are also a large number of OSTs.
4362 Details : qos_prep_create() uses an OST index reseed value that is an
4363 even multiple of the number of available OSTs so that if the
4364 reseed happens in the middle of the object allocation it will
4365 still utilize the OSTs as uniformly as possible.
4370 Description: OST or MDS may oops in ping_evictor_main()
4371 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4372 but doesn't restart at beginning of obd_exports_timed list
4375 Severity : enhancement
4377 Description: improve by-nid export eviction on the MDS and OST
4378 Details : allow multiple exports with the same NID to be evicted at one
4379 time without re-searching the exports list.
4382 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4384 Description: MDS may oops in groups_free()
4385 Details : in rare race conditions a newly allocated group_info struct is
4386 freed again, and this can be NULL. The 2.4 compatibility code
4387 for groups_free() checked for a NULL pointer, but 2.6 did not.
4390 Frequency : common for liblustre clients doing little filesystem IO
4391 Bugzilla : 9352, 7313
4392 Description: server may evict liblustre clients accessing contended locks
4393 Details : if a client is granted a lock or receives a completion AST
4394 with a blocking AST already set it would not reply to the AST
4395 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4398 Frequency : lfs setstripe, only systems with more than 160 OSTs
4400 Description: unable to set striping with a starting offset beyond OST 160
4401 Details : llapi_create_file() incorrectly limited the starting stripe
4402 index to the maximum single-file stripe count.
4405 Frequency : LDAP users only
4407 Description: lconf did not handle in-kernel recovery with LDAP properly
4408 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4410 Severity : enhancement
4412 Description: bind OST threads to NUMA nodes to improve performance
4413 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4414 node and do their allocations there to localize memory access
4416 Severity : enhancement
4418 Description: llmount can determine client NID directly from Myrinet (GM)
4419 Details : the client NID code from gmnalnid was moved directly into
4420 llmount, removing the need to use this or specifying the
4421 client NID explicitly when mounting GM clients with zeroconf
4424 Frequency : if client is started with down MDS
4426 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4427 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4428 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4429 ptlrpc_import_delayed_req() because we didn't check if the
4430 request was interrupted, and we also didn't break out of the
4431 event loop if there was no timeout
4436 Description: data loss during non-page-aligned writes to a single file from
4437 both multiple nodes and multiple threads on one node at same time
4438 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4439 inconsistency led to false short-reads, that were cached and later
4440 used by ->prepare_write() to fill in partially written page,
4441 leading to data loss.
4444 Frequency : always, if lconf --abort_recovery used
4446 Description: lconf --abort_recovery fails with 'Operation not supported'
4447 Details : lconf was attempting to abort recovery on the MDT device and not
4450 Severity : enhancement
4452 Description: remove cleanup logs
4453 Details : replace lconf-generated cleanup logs with lustre internal
4454 cleanup routines. Eliminates the need for client-cleanup and
4457 Severity : enhancement
4459 Description: add support for EAs (user and system) on lustre filesystems
4460 Details : it is now possible to store extended attributes in the Lustre
4461 client filesystem, and with the user_xattr mount option it
4462 is possible to allow users to store EAs on their files also
4464 Severity : enhancement
4466 Description: Add possibility (config option) to show minimal available OST free
4468 Details : When compiled with --enable-mindf configure option, statfs(2)
4469 (and so, df) will return least minimal free space available from
4470 all OSTs as amount of free space on FS, instead of summary of
4471 free spaces of all OSTs.
4473 Severity : enhancement
4475 Description: do not expand extent locks acquired on OST-side
4476 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4477 by server: they are not cached anyway.
4480 Frequency : when mmap is used/binaries executed from Lustre
4482 Description: Unmmap pages before throwing them away from read cache.
4483 Details : llap_shrink cache now attempts to unmap pages before discarding
4484 them (if unmapping failed - do not discard). SLES9 kernel has
4485 extra checks that trigger if this unmapping is not done first.
4490 Description: lconf didn't resolve symlinks before checking to see whether a
4491 given mountpoint was already in use
4494 Frequency : when migrating failover services
4495 Bugzilla : 6395, 9514
4496 Description: When migrating a subset of services from a node (e.g. failback
4497 from a failover service node) the remaining services would
4498 time out and evict clients.
4499 Details : lconf --force (implied by --failover) sets the global obd_timeout
4500 to 5 seconds in order to quickly disconnect, but this caused
4501 other RPCs to time out too quickly. Do not change the global
4502 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4504 Severity : enhancement
4505 Frequency : if MDS is started with down OST
4506 Bugzilla : 9439,5706
4507 Description: Allow startup/shutdown of an MDS without depending on the
4508 availability of the OSTs.
4509 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4510 Add appropriate locking and lov-osc refcounts for safe
4511 cleaning. Add osc abort_inflight calls in case the
4512 synchronize never started.
4515 Frequency : occasional (Cray XT3 only)
4517 Description: root not authorized to access files in CRAY_PORTALS environment
4518 Details : The client process capabilities were not honoured on the MDS in
4519 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4520 been accessed by an authorized user then root was able to access
4521 the file on the local client also. The root user capabilities
4522 are now allowed on the MDS, as this environment has secure UID.
4525 Frequency : occasional
4527 Description: ldiskfs "too long searching" message happens too often
4528 Details : A debugging message (otherwise harmless) prints too often on
4529 the OST console. This has been reduced to only happen when
4530 there are fragmentation problems on the filesystem.
4535 Description: Division by zero in statfs when all OSCs are inactive
4536 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4537 lov_statfs divides by value returned by lov_get_stripecnt().
4541 Bugzilla : 9489, 3273
4542 Description: First write from each client to each OST was only 4kB in size,
4543 to initialize client writeback cache, which caused sub-optimal
4544 RPCs and poor layout on disk for the first writen file.
4545 Details : Clients now request an initial cache grant at (re)connect time
4546 and so that they can start streaming writes to the cache right
4547 away and always do full-sized RPCs if there is enough data.
4548 If the OST is rebooted the client also re-establishes its grant
4549 so that client cached writes will be honoured under the grant.
4554 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4555 Details : Now I/O RPCs go to different portal number and (presumably) fast
4556 lock requests (and glimses) and other RPCs get their own service
4557 threads pool that should be able to service those RPCs
4560 Severity : enhancement
4562 Description: Ability to exchange lustre version between client and servers and
4563 issue warnings at client side if client is too old. Also for
4564 liblustre clients there is ability to refuse connection of too old
4566 Details : New 'version' field is added to connect data structure that is
4567 filled with version info. That info is later checked by server and
4571 Frequency : rare, liblustre only.
4572 Bugzilla : 9296, 9581
4573 Description: Two simultaneous writes from liblustre at offset within same page
4574 might proceed at the same time overwriting eachother with stale
4576 Details : I/O lock withing llu_file_prwv was released too early, before data
4577 actually was hitting the wire. Extended lock-holding time until
4578 server acknowledges receiving data.
4581 Frequency : extremely rare. Never observed in practice.
4583 Description: avoid generating lustre_handle cookie of 0.
4584 Details : class_handle_hash() generates handle cookies by incrementing
4585 global counter, and can hit 0 occasionaly (this is unlikely, but
4586 not impossible, because initial value of cookie counter is
4587 selected randonly). Value of 0 is used as a sentinel meaning
4588 "unassigned handle" --- avoid it. Also coalesce two critical
4589 sections in this function into one.
4591 Severity : enhancement
4593 Description: allow liblustre clients to delegate truncate locking to OST
4594 Details : To avoid overhead of locking, liblustre client instructs OST to
4595 take extent lock in ost_punch() on client's behalf. New connection
4596 flag is added to handle backward compatibility.
4598 Severity : enhancement
4599 Bugzilla : 4928, 7341, 9758
4600 Description: allow number of OST service threads to be specified
4601 Details : a module parameter allows the number of OST service threads
4602 to be specified via "options ost ost_num_threads={N}" in the
4603 OSS's /etc/modules.conf or /etc/modprobe.conf.
4607 Bugzilla : 6146, 9635, 9895
4608 Description: servers crash with bad pointer in target_handle_connect()
4609 Details : In rare cases when a client is reconnecting it was possible that
4610 the connection request was the last reference for that export.
4611 We would temporarily drop the export reference and get a new
4612 one, but this may have been the last reference and the export
4613 was just destroyed. Get new reference before dropping old one.
4615 Severity : enhancement
4616 Frequency : if client is started with failover MDS
4618 Description: Allow multiple MDS hostnames in the mount command
4619 Details : Try to read the configuration from all specified MDS
4620 hostnames during a client mount in case the "primary"
4623 Severity : enhancement
4625 Description: Stop sending data to evicted clients as soon as possible.
4626 Details : Check if the client we are about to send or are sending data to
4627 was evicted already. (Check is done every second of waiting,
4628 for which l_wait_event interface was extended to allow checking
4629 of exit condition at specified intervals).
4632 Frequency : rare, normally only when NFS exporting is done from client
4634 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4636 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4639 Frequency : always, if srand() is called before liblustre initialization
4641 Description: Liblustre uses system PRNG disturbing its usage by user application
4642 Details : Introduce internal to lustre fast and high-quality PRNG for
4643 lustre usage and make liblustre and some other places in generic
4644 lustre code to use it.
4646 Severity : enhancement
4647 Bugzilla : 9477, 9557, 9870
4648 Description: Verify that the MDS configuration logs are updated when xml is
4649 Details : Check if the .xml configuration logs are newer than the config
4650 logs stored on the MDS and report an error if this is the case.
4651 Request --write-conf, or allow starting with --old_conf.
4653 Severity : enhancement
4655 Description: Handle symlinks in the path when checking if Lustre is mounted.
4656 Details : Resolve intermediate symlinks when checking if a client has
4657 mounted a filesystem to avoid duplicate client mounts.
4662 Description: lconf can hit an error exception but still return success.
4663 Details : The lconf command catches the Command error exception at the top
4664 level script context and will exit with the associated exit
4665 status, but doesn't ensure that this exit status is non-zero.
4670 Description: failure of ptlrpc thread startup can cause oops
4671 Details : Starting a ptlrpc service thread can fail if there are a large
4672 number of threads or the server memory is very fragmented.
4673 Handle this without oopsing.
4676 Frequency : always, only if liblustre and non-default acceptor port was used
4678 Description: liblustre cannot connect to servers with non-default acceptor port
4679 Details : tcpnal_set_default_params() was not called and was therefore
4680 ignoring the environment varaible TCPNAL_PORT, as well as other
4681 TCPNAL_ environment variables
4686 Description: two objects could be created on the same OST for a single file
4687 Details : If an OST is down, in some cases it was possible to create two
4688 objects on a single OST for a single file. No problems other
4689 than potential performance impact and spurious error messages.
4693 Bugzilla : 5681, 9562
4694 Description: Client may oops in ll_unhash_aliases
4695 Details : Client dcache may become inconsistent in race condition.
4696 In some cases "getcwd" can fail if the current directory is
4702 Description: Inode refcounting problems in NFS export code
4703 Details : link_raw functions used to call d_instantiate without obtaining
4704 extra inode reference first.
4708 Bugzilla : 9942, 9903
4709 Description: Referencing freed requests leading to crash, memleaks with NFS.
4710 Details : We used to require that call to ll_revalidate_it was always
4711 followed by ll_lookup_it. Also with revalidate_special() it is
4712 possible to call ll_revalidate_it() twice for the same dentry
4713 even if first occurence returned success. This fix changes semantic
4714 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4715 reference on a request referred from the intent.
4716 ll_intent_release() then releases such a request.
4719 Frequency : rare, normally benchmark loads only
4721 Description: unlinked inodes were kept in memory on the client
4722 Details : If a client is repeatedly creating and unlinking files it
4723 can accumulate a lot of stale inodes in the inode slab cache.
4724 If there is no other client load running this can cause the
4725 client node to run out of memory. Instead flush old inodes
4726 from client cache that have the same inode number as a new inode.
4729 Frequency : SLES9 2.6.5 kernel and long filenames only
4730 Bugzilla : 9969, 10379
4731 Description: utime reports stale NFS file handle
4732 Details : SLES9 uses out-of-dentry names in some cases, which confused
4733 the lustre dentry revalidation. Change it to always use the
4737 Frequency : rare, unless heavy write-truncate concurrency is continuous
4738 Bugzilla : 4180, 6984, 7171, 9963, 9331
4739 Description: OST becomes very slow and/or deadlocked during object unlink
4740 Details : filter_destroy() was holding onto the parent directory lock
4741 while truncating+unlinking objects. For very large objects this
4742 may block other threads for a long time and slow overall OST
4743 responsiveness. It may also be possible to get a lock ordering
4744 deadlock in this case, or run out of journal credits because of
4745 the combined truncate+unlink. Solution is to do object truncate
4746 first in one transaction without parent lock, and then do the
4747 final unlink in a new transaction with the parent lock. This
4748 reduces the lock hold time dramatically.
4751 Frequency : rare, 2.4 kernels only
4753 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4754 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4755 before unmounting their filesystem in order to avoid deadlock.
4756 The kernel_locked() function in 2.4 kernels only checks whether
4757 the kernel lock is held, not whether it is this process that is
4758 holding it as 2.6 kernels do.
4763 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4764 Details : The client ptlrpc code may be trying to reconnect to a down
4765 server before a previous connection attempt has timed out.
4766 Increase the reconnect interval to be longer than the connection
4767 timeout interval to avoid sending duplicate connections to
4771 Frequency : echo_client brw_test command
4773 Description: fix echo_client to work with OST preallocated code
4774 Details : OST preallocation code (5137) didn't take echo_client IO path
4775 into account: echo_client calls filter methods outside of any
4776 OST thread and, hence, there is no per-thread preallocated
4777 pages and buffers to use. Solution: hijack pga pages for IO. As
4778 a byproduct, this avoids unnecessary data copying.
4782 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4783 Description: Client can oops in mdc_commit_close() after open replay
4784 Details : It was possible for the MDS to return an open request with no
4785 transaction number in mds_finish_transno() if the client was
4786 evicted, but without actually returning an error. Clients
4787 would later try to replay that open and may trip an assertion
4788 Simplify the client close codepath, and always return an error
4789 from the MDS in case the open is not successful.
4792 Frequency : rare, 2.6 OSTs only
4794 Description: OST may deadlock under high load on fragmented files
4795 Details : If there was a heavy load and highly-fragmented OST filesystems
4796 it was possible to have all the OST threads deadlock waiting on
4797 allocation of biovecs, because the biovecs were not released
4798 until the entire RPC IO was completed. Instead, release biovecs
4799 as soon as they are complete to ensure forward IO progress.
4801 Severity : enhancement
4803 Description: Support for specifying external journal device at mount
4804 Details : If an OST or MDS device is formatted with an external journal
4805 device, this device major/minor is stored in the ext3 superblock
4806 and may not be valid for failover. Allow detecting and
4807 specifying the external journal at mount time.
4812 Description: Mounting an MDS with pending unlinked files may cause oops
4813 Details : target_finish_recovery() calls mds_postrecov() which returned
4814 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4815 considers this an error and immediately begins cleaning up the
4816 lov, just after starting the mds_lov process
4818 Severity : enhancement
4820 Description: Implement 'lfs df' to report actual free space on per-OST basis
4821 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4822 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4823 usage of objects; '-h' to report in human readable format.
4825 ------------------------------------------------------------------------------
4827 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4834 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4835 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4836 at startup during orphan unlinking. Instead, skip buddy update
4837 before setup, as it will be regenerated after recovery is complete.
4842 Description: If an OST is inactive, its locks might reference stale inodes.
4843 Details : lov_change_cbdata() must iterate over all namespaces, even if
4844 they are inactive to clear inode references from the lock.
4846 Severity : enhancement
4847 Frequency : occasional, if non-standard max_dirty_mb used
4849 Description: Client will block write RPCs if not enough grant
4850 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4851 then the client will block writes while waiting for another RPC
4852 to complete instead of consuming its dirty limit. With change
4853 we get improved performance when max_dirty_mb is small.
4855 Severity : enhancement
4856 Bugzilla : 3389, 6253
4857 Description: Add support for supplementary groups on the MDS.
4858 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4859 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4860 lookups for user supplementary groups into a cache.
4864 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4865 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4866 for liblustre O_EXCL is enforced by MDS.
4870 Description: readdir never returns NULL in liblustre.
4871 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4872 dentry in struct dirent.
4876 Frequency : liblustre only, depends on application IO pattern
4877 Description: liblustre clients evicted if not contacting servers
4878 Details : Don't put liblustre clients into the ping_evictor list, so
4879 they will not be evicted by the pinger ever.
4881 Severity : enhancement
4883 Description: Add ability to evict clients by NID from MDS.
4884 Details : By echoing "nid:$NID" string into
4885 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4886 $NID would be instantly evicted from this MDS and from all active
4887 OSTs connected to it.
4891 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4892 Details : lookup_it_finish() used to query file size from OSTs that was not
4897 Description: service threads change working directory to that of init
4898 Details : Starting lustre service threads may pin the working directory
4899 of the parent thread, making that filesystem busy. Threads
4900 now change to the working directory of init to avoid this.
4904 Frequency : during shutdown only
4905 Description: shutdown with a failed MDS or OST can cause unmount to hang
4906 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4911 Frequency : 2.6 only
4912 Description: chmod/chown may include an extra supplementary group
4913 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4914 supplementary group and if none is specified this is used.
4917 Bugzilla : 5479 (6816)
4919 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4920 Details : If lookup is in progress on a file that is unlinked we might try
4921 to revalidate the inode and fail in revalidate after lookup is
4922 complete and ll_file_open() enqueues the open again but
4923 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4926 Frequency : always, if lconf --abort_recovery used
4928 Description: lconf --abort_recovery fails with 'Operation not supported'
4929 Details : lconf was attempting to abort recovery on the MDT device and not
4932 ------------------------------------------------------------------------------
4934 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4939 Frequency : rare (only unsupported configurations with a node running as an
4941 Bugzilla : 6514, 5137
4942 Description: Mounting a Lustre file system on a node running as an OST could
4944 Details : OSTs now preallocates memory needed to write out data at
4945 startup, instead of when needed, to avoid having to
4946 allocate memory in possibly low memory situations.
4947 Specifically, if the file system is mounted on on OST,
4948 memory pressure could force it to try to write out data,
4949 which it needed to allocate memory to do. Due to the low
4950 memory, it would be unable to do so and the node would
4951 become unresponsive.
4953 Severity : enhancement
4955 Description: Addition of lconf --service command line option
4956 Details : lconf now accepts a '--service <arg>' option, which is
4957 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4959 Severity : enhancement
4961 Description: Failover mode is now the default for OSTs.
4962 Details : By default, OSTs will now run in failover mode. To return to
4963 the old behaviour, add '--failout' to the lmc line for OSTs.
4965 Severity : enhancement
4967 Description: Health checks are now provided for MDS and OSTs
4968 Details : Additional detailed health check information on MSD and OSTs
4969 is now provided through the procfs health_check value.
4972 Frequency : occasional, depends on IO load
4974 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4975 after numerous create/delete cycles
4976 Details : The ext3 inode allocation policy would not allocate new inodes
4977 very well on the OSTs because there are no new directories
4978 being created. Instead we look for groups with free space if
4979 the parent directories are nearly full.
4984 Description: Network or server problems during mount may cause partially
4985 mounted clients instead of returning an error.
4986 Details : The config llog parsing code may overwrite the error return
4987 code during mount error handling, returning success instead
4993 Description: MDS can fail to allocate large reply buffers
4994 Details : After long uptimes the MDS can fail to allocate large reply
4995 buffers (e.g. zconf client mount config records) due to memory
4996 fragmentation or consumption by the buffer cache. Preallocate
4997 some large reply buffers so that these replies can be sent even
4998 under memory pressure.
5002 Frequency : rare (liblustre)
5003 Description: fsx running with liblustre complained that using truncate() to
5004 extend the file doesn't work. This patch corrects that issue.
5005 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
5006 ATTR_SIZE and lsm use in llu_setattr_raw.
5010 Frequency : rare, only 2.6 kernels
5011 Description: Unusual file access patterns on the MDS may result in inode
5012 data being lost in very rare circumstances.
5013 Details : Bad interaction between the ea-in-inode patch and the "no-read"
5014 code in the 2.6 kernel caused the inode and/or EA data not to
5015 be read from disk, causing single-file corruption.
5019 Frequency : rare, only 2.6 filesystems using extents
5020 Description: Heavy concurrent write and delete load may cause data corruption.
5021 Details : It was possible under high-load situations to have an extent
5022 metadata block in the block device cache from a just-unlinked
5023 file overwrite a newly-allocated data block. We now unmap any
5024 metadata buffers that alias just-allocated data blocks.
5028 Frequency : filesystems with default stripe_count larger than 77
5029 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
5030 Details : lconf specifies an inode size of 4096 bytes when the default
5031 stripe_count is larger than 77. This conflicts with the default
5032 inode density of 1 per 4096 bytes. Allocate smaller inodes in
5033 this case to avoid pinning too much memory for large EAs.
5035 ------------------------------------------------------------------------------
5037 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
5042 Frequency : rare (extremely heavy IO load with hundreds of clients)
5044 Description: Client is evicted, gets IO error writing to file
5045 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
5046 caused clients to be evicted for AST timeouts. The fixes in
5047 bug 5192 mean we no longer need to have such short AST timeouts
5048 so ldlm_timeout has been increased.
5051 Frequency : occasional during --force or --failover shutdown under load
5052 Bugzilla : 5949, 4834
5053 Description: Server oops/LBUG if stopped with --force or --failover under load
5054 Details : a collection of import/export refcount and cleanup ordering
5055 issues fixed for safer force cleanup
5058 Frequency : only filesystems larger than 120 OSTs
5059 Bugzilla : 5990, 6223
5060 Description: lfs getstripe would oops on a very large filesystem
5061 Details : lov_getconfig used kfree on vmalloc'd memory
5064 Frequency : only filesystems exporting via NFS to Solaris 10 clients
5065 Bugzilla : 6242, 6243
5066 Description: reading from files that had been truncated to a non-zero size
5067 but never opened returned no data
5068 Details : ll_file_read() reads zeros from no-object files to EOF
5073 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
5074 Details : There's little harm in aborting MDS/OSS recovery and letting it
5075 try again, so I removed the LASSERT and return an error instead.
5077 Severity : enhancement
5079 Description: New debugging infrastructure for tracking down data corruption
5080 Details : The I/O checksum code was replaced to: (a) control it at runtime,
5081 (b) cover more of the client-side code path, and (c) try to narrow
5082 down where problems occurred
5086 Bugzilla : 3819, 4364, 4397, 6313
5087 Description: Racing close and eviction MDS could cause assertion in mds_close
5088 Details : It was possible to get multiple mfd references during close and
5089 client eviction, leading to one thread referencing a freed mfd.
5091 Severity: : enhancement
5092 Bugzilla : 3262, 6359
5093 Description: Attempts to reconnect to servers are now more aggressive.
5094 Details : This builds on the enhanced upcall-less recovery that was added
5095 in 1.4.2. When trying to reconnect to servers, clients will
5096 now try each server in the failover group every 10 seconds. By
5097 default, clients would previously try one server every 25 seconds.
5102 Description: After recovery, certain operations trigger a failed
5103 assertion on a client.
5104 Details : Failing over an mds, using lconf -d --failover, while a
5105 client was doing a readdir() call would cause the client to
5106 LBUG after recovery completed and the readdir() was resent.
5108 Severity : enhancement
5110 Description: Default groups are now added by lconf
5111 Details : You can now run lconf --group <servicename> without having to
5112 manually add groups with lmc.
5115 Frequency : occasional
5117 Description: Nodes with an elan id of 0 trigger a failed assertion
5120 Frequency : always when accessing e.g. tty/console device nodes
5122 Description: tty and some other devices nodes cannot be used on lustre
5123 Details : file's private_data field is used by device data and lustre
5124 values in there got lost. New field was added to struct file to
5125 store fs-specific private data.
5128 Frequency : when exporting Lustre via NFS
5130 Description: NFSD failed occasionally when looking up a path component
5131 Details : NFSD is looking up ".." which was broken in ext3 directories
5132 that had grown large enough to become hashed.
5135 Frequency : Clusters with multiple interfaces not on the same subnet
5137 Description: Nodes will repeatedly try to reconnect to an interface which it
5138 cannot reach and report an error to the log.
5139 Details : Extra peer list entries will be created by lconf with some peers
5140 unreachable. lconf now validates the peer before adding it.
5143 Frequency : Only if a default stripe is set on the filesystem root.
5145 Description: Setting a default stripe on the filesystem root prevented the
5146 filesystem from being remounted.
5147 Details : The client was sending extra request flags in the root getattr
5148 request and did not allocate a reply buffer for the dir EA.
5151 Frequency : occasional, higher if lots of files are accessed by one client
5152 Bugzilla : 6159, 6097
5153 Description: Client trips assertion regarding lsm mismatch/magic
5154 Details : While revalidating inodes the VFS looks up inodes with ifind()
5155 and in rare cases can find an inode that is being freed.
5156 The ll_test_inode() code will free the lsm during ifind()
5157 when it finds an existing inode and then the VFS later attaches
5158 this free lsm to a new inode.
5162 Bugzilla : 6422, 7030
5163 Description: MDS deadlock between mkdir and client eviction
5164 Details : Creating a new file via mkdir or mknod (starting a transaction
5165 and getting the ns lock) can deadlock with client eviction
5166 (gets ns lock and trying to finish a synchronous transaction).
5169 Frequency : occasional
5170 Description: While starting a server, the fsfilt_ext3 module could not be
5172 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5173 kernels. Previously, lconf would still use the ext3 name
5174 when trying to load modules. Now, it will correctly use
5175 ext3 on 2.4 and ldiskfs on 2.6.
5177 Severity : enhancement
5178 Description: The default stripe count has been changed to 1
5179 Details : The interpretation of the default stripe count (0, to lfs
5180 or lmc) has been changed to mean striping across a single
5181 OST, rather than all available. For general usage we have
5182 found a stripe count of 1 or 2 works best.
5184 Severity : enhancement
5185 Description: Add support for compiling against Cray portals.
5186 Details : Conditional compiling for some areas that are different
5190 Frequency : occasional
5191 Bugzilla : 6409, 6834
5192 Description: Creating files with an explicit stripe count may lead to
5193 a failed assertion on the MDS
5194 Details : If some OSTs are full or unavailable, creating files may
5195 trigger a failed assertion on the MDS. Now, Lustre will
5196 try to use other servers or return an error to the
5200 Frequency : occasional
5202 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5203 caused lock timeout during readahead (since 1.4.2).
5204 Details : Processes doing readahead might match a lock that hasn't been
5205 granted yet if there are overlapping and conflicting lock
5206 requests. The readahead process waits on ungranted lock
5207 (original lock is CBPENDING), while OST waits for that process
5208 to cancel CBPENDING read lock and eventually evicts client.
5210 Severity : enhancement
5212 Description: Initial enabling of flock support for clients
5213 Details : Implements fcntl advisory locking and file status functions.
5214 This feature is provided as an optional mount flag (default
5215 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5216 locking are implemented yet, and those that are are not guaranteed
5217 to be completely correct in production environments.
5218 mount -t lustre -o [flock|noflock] ...
5221 Frequency : occasional
5223 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5224 assertion in the ext3 JBD (journaling) layer.
5225 Details : The b_committed_data struct is protected by the big kernel lock
5226 in 2.4 kernels, serializing journal_commit_transaction() and
5227 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5228 access to this struct. In 2.6 kernels there is finer grained
5229 locking to improve SMP performance of the JBD layer.
5233 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5238 Description: merge and cleanup kernel patches.
5241 -----------------------------------------------------------------------------
5243 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5245 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5246 versions of Lustre. Please update all servers and clients to
5247 version 1.4.2 or later at the same time. You must also run
5248 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5249 to update the configuration logs.
5251 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5252 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5253 - swab llog records of type '0' so we get proper header size/idx (5861)
5254 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5255 - fix rename of one directory over another leaking an inode (5953)
5256 - avoid SetPageDirty on 2.6 (5981)
5257 - don't re-add just-being-destroyed locks to the waiting list (5653)
5258 - when creating new directories, inherit the parent's custom
5259 striping settings if present parent (3048)
5260 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5261 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5262 - don't hold client locks on temporary worklist from l_lru (5666)
5263 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5264 - automatically evict dead clients (5921)
5265 - Update file size properly in create+truncate+fstat case (6196)
5266 - Do not unhash mountpoint dentries, do not allow removal of
5268 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5269 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5270 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5271 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5272 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5273 - Correctly handle reads of files with no objects (6243)
5274 - lctl recover will also mark a device active if deactivate used (5933)
5276 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5277 - allow --write-conf on an MDS with different nettype than client (5619)
5278 - don't write config llogs to MDS for mounts not from that MDS (5617)
5279 - lconf should create multiple TCP connections from a client (5201)
5280 - init scripts are now turned off by default; run chkconfig --on
5281 lustre and chkconfig --on lustrefs to use them
5282 - upcalls are no longer needed for clients to recover to failover
5284 - add --abort-recovery option to lconf to abort recovery on device
5286 - add support for an arbitrary number of OSTs (3026)
5287 - Quota support protocol changes.
5288 - forward compatibility changes to wire structs (6007)
5289 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5290 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5291 - improved "lustre status" script
5292 - initialize blocksize for non-regular files (6062)
5293 - added --disable-server and --disable-client configure options (5782)
5294 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5295 - Vanilla 2.4.29 support
5296 - increase maximum number of obd devices to 520 (6242)
5297 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5298 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5299 SLES 9 2.6.5 kernels
5300 - limit stripes per file to 160 (the maximum EA size) (6093)
5302 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5305 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5306 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5307 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5308 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5309 - small ext3 extents cleanups and fixes (5733)
5310 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5311 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5312 - avoid needless client->OST connect, fix handle mismatch (5317)
5313 - fix DLM error path that led to out-of-sync client, long delays (5779)
5314 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5315 - fix several locking issues related to i_size (5492,5624,5654,5672)
5316 - don't move pending lock onto export if it is already evicted (5683)
5317 - fix kernel oops when creating .foo in unlinked directory (5548)
5318 - fix deadlock in obdfilter statistics vs. object create (5811)
5319 - use time_{before,after} to avoid timer jiffies wrap (5882)
5320 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5321 - Do not leak request if server was not able to process it (5154)
5322 - If mds_open unable to find parent dir, make that negative lookup(5154)
5323 - don't create new directories with extent-mapping (5909, 5936)
5325 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5326 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5327 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5328 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5329 - support for RHEL4 kernel on i686 (5773)
5330 - provide error messages when incompatible logs are encountered (5898)
5332 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5333 * version 1.4.0.10 (1.4.1 release candidate 1)
5335 - don't keep a lock reference when lock is not granted (4238)
5336 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5337 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5338 - drop import inflight refcount on signal_completed_replay error (5255)
5339 - unlock page after async write error during send (3677)
5340 - handle missing objects in filter_preprw_read properly (5265)
5341 - no transno return for symlink open, don't save no-trasno open (3440)
5342 - don't try to complete elan receive that already failed (4012)
5343 - free RPC server reply state on error (5406)
5344 - clean up thread from ptlrpc_start_thread() on error (5160)
5345 - readahead could read extra page into cache that wasn't ejected (5388)
5346 - prevent races in class_attach/setup/cleanup/detach (5260)
5347 - don't dereference de->d_inode after l_dput of de (5458)
5348 - use "int" for stripe value returned from lock_to_stripe (5544)
5349 - mballoc allocation and error-checking fixes in 2.6 (5504)
5350 - block device patches to fix I/O request sizes in 2.6 (5482)
5351 - look up hostnames for IB nals (5602)
5352 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5353 - don't start multiple acceptors for the same port (5277)
5354 - fix incorrect LASSERT in mds_getattr_name (5635)
5355 - export a proc file for general "ping" checking (5628)
5356 - fix "lfs check" to not block when the MDS is down (5628)
5358 - service request history (4965)
5359 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5360 - create an "evict_client" /proc entry on OSTs, like the MDS has
5361 - fix mount usage message, return errors per mount(8) (5168)
5362 - change grep [] to grep "[]" in tests so they work in more UMLs
5363 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5364 - remove ancient LOV_MAGIC_V0 EA support (5047)
5365 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5366 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5367 - allow readahead allocations to fail when low on memory (5383)
5368 - mmap locking landed again, after considerable improvement (2828)
5369 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5371 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5374 - send OST transaction number in read/write reply to free req (4966)
5375 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5376 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5377 - the watchdog thread now runs as interruptible (5246)
5378 - flock/lockf fixes (but it's still disabled, pending 5135)
5379 - don't use EXT3 constants in llite code (5094)
5380 - memory shortage at startup could cause assertion (5176)
5382 - reorganization of lov code
5383 - single portals codebase
5385 - add extents/mballoc support (5025)
5386 - direct I/O reads in the obdfilter (4048)
5387 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5389 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5392 - send OST transaction number in read/write reply to free req (4966)
5393 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5394 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5395 - the watchdog thread now runs as interruptible (5246)
5396 - handle missing objects in filter_preprw_read properly (5265)
5397 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5398 - drop import inflight refcount on signal_completed_replay error (5255)
5399 - unlock page after async write error during send (3677)
5400 - return original error code on reconstructed replies (3761)
5401 - no transno return for symlink open, don't save no-trasno open (3440)
5403 - add pid to ldlm debugging output (4922)
5404 - bump the watchdog timeouts -- we can't handle 30sec yet
5405 - extra debugging for orphan dentry/inode bug (5259)
5407 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5410 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5411 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5412 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5413 - fix NULL dereference in /proc/sys/portals/routes (4827)
5414 - allow failed mdc_close() operations to be interrupted (4561)
5415 - stop precreate on OST before MDS would time out on it (4778)
5416 - don't send partial-page writes before EOF from client (4410)
5417 - discard client grant for sub-page writes on large-page clients (4520)
5418 - don't free dentries not owned by NFS code, check generation (4806)
5419 - fix lsm leak if mds_create_objects() fails (4801)
5420 - limit debug_daemon file size, always print CERROR messages (4789)
5421 - use transno after validating reply (3892)
5422 - process timed out requests if import state changes (3754)
5423 - update mtime on OST during writes, return in glimpse (4829)
5424 - add mkfsoptions to LDAP (4679)
5425 - use ->max_readahead method instead of zapping global ra (5039)
5426 - don't interrupt __l_wait_event() during strace
5428 - add software watchdogs to catch hung threads quickly (4941)
5429 - make lustrefs init script start after nfs is mounted
5430 - fix CWARN/ERROR duplication (4930)
5431 - return async write errors to application if possible (2248)
5432 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5433 - print NAL number in %x format (4645)
5434 - update barely-supported suse-2.4.21-171 series (4842)
5435 - support for sles 9 %post scripts
5436 - support for building 2.6 kernel-source packages
5437 - support for sles km_* packages
5439 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5442 - ignore -ENOENT errors in osc_destroy (3639)
5443 - notify osc create thread that OSC is being cleaned up (4600)
5444 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5445 - reconstruct ost_handle() like mds_handle() (4657)
5446 - create a new thread to do import eviction to avoid deadlock (3969)
5447 - let lconf resolve symlinked-to devices (4629)
5448 - don't unlink "objects" from directory with default EA (4554)
5449 - hold socknal file ref over connect in case target is down (4394)
5450 - allow more than 32000 subdirectories in a single directory (3244)
5451 - fix blocks count for O_DIRECT writes (3751)
5452 - OST returns ENOSPC from object create when no space left (4539)
5453 - don't send truncate RPC if file size isn't changing (4410)
5454 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5455 - bind to privileged port in socknal and tcpnal (3689)
5457 - rate limit CERROR/CWARN console message to avoid overload (4519)
5458 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5459 - basic mmap support (3918)
5460 - kernel patch series update from b1_4 (4711)
5462 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5465 - avoid crash during MDS cleanup with OST shut down (2775)
5466 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5467 - don't use bad inodes on the MDS (3744)
5468 - dynamic object preallocation to improve recovery speed (4236)
5469 - don't hold spinlock over lock dumping or change debug flags (4401)
5470 - don't zero obd_dev when it is force cleaned (3651)
5471 - print grants to console if they go negative (4431)
5472 - "lctl deactivate" will stop automatic recovery attempts (3406)
5473 - look for existing locks in ldlm_handle_enqueue() (3764)
5474 - don't resolve lock handle twice in recovery avoiding race (4401)
5475 - revalidate should check working dir is a directory (4134)
5477 - don't always mark "slow" obdfilter messages as errors (4418)
5479 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5482 - don't close LustreDB during write_conf until it is done (3860)
5483 - fix typo in lconf for_each_profile (3821)
5484 - allow dumping logs from multiple threads at one time (3820)
5485 - don't allow multiple threads in OSC recovery (3812)
5486 - fix debug_size parameters (3864)
5487 - fix mds_postrecov to initialize import for llog ctxt (3121)
5488 - replace config semaphore with spinlock (3306)
5489 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5490 - don't allow enqueue to complete on a destroyed export (3822)
5491 - down write_lock before checking llog header bitmap (3825)
5492 - recover from lock replay timeout (3764)
5493 - up llog sem before sending rpc (3652)
5494 - reduce ns lock hold times when setting kms (3267)
5495 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5496 - fix NULL deref and obd_dev leak on setup error (3312)
5497 - replace some LBUG about llog ops with error handling (3841)
5498 - don't match INVALID dentries from d_lookup and spin (3784)
5499 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5500 - fix invalid assertion in ptlrpc_set_wait (3880)
5502 - add libwrap support for the TCP acceptor (3996)
5503 - add /proc/sys/portals/routes for non-root route listing (3994)
5504 - allow setting MDS UUID in .xml (2580)
5505 - print the stack of a process that LBUGs (4228)
5507 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5510 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5511 - make sure to unset replay flag from failed open requests (3440)
5512 - if default stripe count is 0, use OST count for inode size (3636)
5513 - update parent mtime/ctime on client for create/unlink (2611)
5514 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5515 - free recovery state on server during a forced cleanup (3571)
5516 - unregister_reply for resent reqs (3063)
5517 - loop back devices mounting and status check on 2.6 (3563)
5518 - fix resource-creation race that can provoke i_size == 0 (3513)
5519 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5520 - more debugging for page-accounting assertion (3746)
5521 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5522 - avoid deadlock after precreation failure (3758)
5523 - fix race and lock order deadlock in orphan handling (3450, 3750)
5524 - add validity checks when grabbing inodes from l_ast_data (3599)
5526 - add /proc/.../recovery_status to obdfilter (3428)
5527 - lightweight CDEBUG infrastructure, debug daemon (3668)
5528 - change default OSC RPC parameters to be better on small clusters
5529 - turn off OST read cache for files smaller than 32MB
5530 - install man pages and include them in rpms (3100)
5531 - add new init script for (un)mounting lustre filesystems (2593)
5532 - run chkconfig in %post for init scripts (3701)
5533 - drop scimac NAL (unmaintained)
5535 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5538 - clean kiobufs before and after use (3485)
5539 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5540 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5541 - change init script to fail more clearly if not run as root (1528)
5542 - allow clients to reconnect during replay (1742)
5543 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5544 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5545 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5546 - fix cond_resched (3554)
5547 - only evict unfinished clients after recovery (3515)
5548 - allow bulk resend, prevent data loss (3570)
5549 - dynamic ptlrpc request buffer allocation (2102)
5550 - don't allow unlinking open directory if it isn't empty (2904)
5551 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5552 - remove extraneous obd dereference causing LASSERT failure (3334)
5553 - don't use get_cycles() when creating temp. files on the mds (3156)
5554 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5555 - handle EEXIST for set-stripe, set proper directory name (3336)
5557 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5558 - fix ksocknal_fmb_callback() error messages (2918)
5560 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5563 - don't copy lvb into (possibly NULL) reply on error (2983)
5564 - don't deref dentry after dput, don't free lvb on error (2922)
5565 - use the kms to determine writeback rpc length (2947)
5566 - increment oti_logcookies when osc is inactive (2948)
5567 - update client's i_blocks count via lvb messages (2543)
5568 - handle intent open/close of special files properly (1557)
5569 - mount MDS with errors=remount-ro, like obdfilter (2009)
5570 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5571 - don't use cancelling-locks' kms values (2947)
5572 - use highest lock extent for kms, not last one (2925)
5573 - don't dereference ERR_PTR() dentry in error handling path (3107)
5574 - fix thread race in portals_debug_dumplog() (3122)
5575 - create lprocfs device entries at setup instead of at attach (1519)
5576 - common AST error handler, don't evict client on completion race (3145)
5577 - zero nameidata in detach_mnt in 2.6 (3118)
5578 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5579 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5580 - initialize RPC timeout timer earlier for 2.6 (3219)
5581 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5582 - print nal/nid for unknown nid (3258)
5583 - additional checks for oscc recovery before doing precreate (3284)
5584 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5585 - don't crash in mdc_close for bad permissions on open (3285)
5586 - zero i_rdev for non-device files (3147)
5587 - clear page->private before handing to FS, better assertion (3119)
5588 - tune the read pipeline (3236)
5589 - fix incorrect decref of invalidated dentry (2350)
5590 - provide read-ahead stats and refine rpc in flight stats (3328)
5591 - don't hold journal transaction open across create RPC (3313)
5592 - update atime on MDS at close time (3265)
5593 - close LDAP connection when recovering to avoid server load (3315)
5594 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5595 - don't leak open file on MDS after open resend (3325)
5596 - serialize filter_precreate and filter_destroy_precreated (3329)
5597 - loop device shouldn't call sync_dev() for nul device (3092)
5598 - clear page cache after eviction (2766)
5599 - resynchronize MDS->OST in background (2824)
5600 - refuse to mount the same filesystem twice on same mountpoint (3394)
5601 - allow llmount to create routes for mounting behind routers (3320)
5602 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5603 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5604 - fix rare problem with rename on htree directories (3417)
5606 - allow default OST striping configuration per directory (1414)
5607 - fix compilation for qswnal for 2.6 kernels (3125)
5608 - increase maximum number of MDS request buffers for large systems
5609 - change liblustreapi to be useful for external progs like lfsck (3098)
5610 - increase local configuration timeout for slow disks (3353)
5611 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5613 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5616 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5617 - don't overwrite extent policy data in reply if lock was blocked (2901)
5618 - drop filter export grants atomically with removal from device (2663)
5619 - del obd_self_export from work_list in class_disconnect_exports (2908)
5620 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5621 - swab reply message in mdc_close, other PPC fixes (2464)
5622 - fix destroying of named logs (2325)
5623 - overwrite old logs when running lconf --write_conf (2264)
5624 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5625 - fix race in target_handle_connect (2898)
5626 - mds_reint_create() should take same inode create lock (2926)
5627 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5628 - don't close files for self_export to avoid uninitialized obd (2936)
5629 - allow MDS with the same name as client node (2939)
5630 - hold dentry reference for closed log files for unlink (2325)
5631 - reserve space for all logs during transactions (2059)
5632 - don't evict page beyond end of stripe extent (2925)
5633 - don't oops on a deleted current working directory (2399)
5634 - handle hard links to targets without a parent properly (2517)
5635 - don't dereference NULL lock when racing during eviction (2867)
5636 - don't grow lock extents when lots of conflicting locks (2919)
5638 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5641 - account for cache space usage on clients to avoid data loss (974)
5642 - lfsck support in lustre kernel code (2349)
5643 - reduce journal credits needed for BRW writes (2370)
5644 - orphan handling to avoid losing space on client/server crashes
5645 - ptlrpcd can be blocked, stopping ALL progress (2477)
5646 - use lock value blocks to assist in proper KMS, faster stat (1021)
5647 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5648 - recovery for initial connections (2355)
5649 - fixes for mds_cleanup_orphans (1934)
5650 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5651 - block all file creations until orphan recovery completes (1901)
5652 - client remove rq_connection from request struct (2423)
5653 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5654 - recovery timer race (2670)
5655 - mdc_close recovey bug (2532)
5656 - ptlrpc cleanup bug (2710)
5657 - mds timeout on local locks (2588)
5658 - namespace lock held during RPCs (2431)
5659 - handle interrupted sync write properly (2503)
5660 - don't try to handle a message that hasn't been replied to (2699)
5661 - client assert failure during cleanup after abort recovery (2701)
5662 - leak mdc device after failed mount (2712)
5663 - ptlrpc_check_set allows timedout requests to complete (2714)
5664 - wait for inflight reqs when ptlrpcd finishes (2710)
5665 - make sure unregistered services are removed from the srv_list
5666 - reset bulk XID's when resending them (caught by 1138 test)
5667 - unregister_bulk after timeout
5668 - fix lconf error (2694)
5669 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5670 - readahead locks pages, leaves pending causing memory pressure (2673)
5671 - increase OST request buffers to 4096 on large machines (2729)
5672 - fix up permission of existing directories in simple_mkdir (2661)
5673 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5674 - don't assign transno to errored transactions (2742)
5675 - don't delete objects on OST if given a bogus objid from MDS (2751)
5676 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5677 - if rq_no_resend, then timeout request after recovery (2432)
5678 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5679 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5680 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5681 - MDS recovery shouldn't delete objects that it has given out (2730)
5682 - if enqueue arrives after completion, don't clobber LVB (2819)
5683 - don't unlock pages twice when trigger_group_io returns error (2814)
5684 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5685 - don't write pages to disk if there was an error (1450)
5686 - don't ping imports that have recovery disabled (2676)
5687 - take buffered bytes into account when balancing socknal conn (2817)
5688 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5689 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5690 - remove little-endian swabbing of llog records (1987)
5691 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5692 - retry reposting request buffers if they fail (1191)
5693 - grow extent at grant time to avoid granting a revoked lock (2809)
5694 - lock revoke doesn't evict page if covered by a second lock (2765)
5695 - disable VM readahead to avoid reading outside lock extents (2805)
5697 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5698 - updated kernel patches for hp-2.4.20 kernel (2681)
5700 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5703 - fix truncated write corruption (2366)
5704 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5706 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5707 - ensure len(uuid) < 37 in lmc (1171)
5708 - fix ia64 OOPS in llog_test (2255)
5709 - zero end of page at obdfilter for partial page writes (2648)
5710 - don't leave stale dentries around after renames (bug 2428)
5711 - fix timeouts when evicting a client with a single lock held (2642)
5712 - set deadline for the initial HELLO message to drain (2634)
5713 - print out dotted-quad IP addresses in the socknal (2302)
5715 - additional debugging for MDS client eviction problem (2443)
5716 - fix mkfsoptions support for osts (2603, 2604)
5718 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5721 - add series for the vanilla 2.6.0 kernel
5722 - add series for the vanilla 2.4.24 kernel
5723 - add series for a cray x86/64 UL kernel drop
5724 - fix xattr patches for the vanilla 2.4.19 series
5726 - generate true UUIDs in lmc (1171)
5727 - have portals stack dumping break in UML (2466)
5728 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5729 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5730 - wake callback waiting threads on client eviction (2460)
5731 - Add --ptldebug and --subsystem to lmc (1719)
5732 - update assertion to allow safe interrupt allocation
5733 - set rq_no_resend for cancel requests (2432)
5734 - recalculate ptlrpcd timeout after resend (2494)
5735 - call vfs_rmdir when removing pending directories (2368)
5736 - fix renaming a file to itself (2429)
5737 - lmc creates a default one-stripe lov (2454)
5738 - expand procfs space to handle large clusters (2326)
5739 - increase UML stack to avoid overflow
5740 - update lconf's list of debug and subsystem masks
5741 - fix lfs find --obd (2510)
5742 - /proc tunable for disabling filter read caching (2591)
5743 - stop rpm packages from altering slapd.conf (2301)
5744 - disable nagle in the socknal under 0conf (2578)
5745 - choose mds inode size based on stripe count (2572)
5746 - fix kernel-source rpm problems (2516)
5748 - add --disable-doc to avoid pdf generation (2421)
5749 - update documentation, tests, type-os, comments
5750 - avoid format warnings on ia64
5751 - remove the TOE NAL
5752 - tiny code cleanups by removing unused fields
5754 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5757 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5758 - remove the most blatant lies from BUILDING file (2371)
5759 - change default debug level to reasonable production setting
5760 - reduce client side cache size to reduce cache flush time
5761 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5762 - make TCP zerocopy and pinger support enabled by default (2476)
5763 - sync writes completed after process exits caused crashes (2319)
5764 - maintain correct mount count on the MDS (2356)
5765 - backout 1557, because 2316 wasn't really fixed
5766 - better file I/O statistics gathering in /proc
5767 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5768 - another kernel patch to fix zero-copy TCP function export
5769 - don't take duplicate lock when processing re-sent getattr (2420)
5770 - lctl uses obd_self_export instead of creating new conn (2353)
5771 - MDS/OST recovery case which requires object creation asserted (2425)
5772 - move lfs from /usr/sbin to /usr/bin in packages
5773 - fix race between mds_client_add and mds_client_free (2417)
5774 - use kmalloc instead of slabs in portals (2430)
5775 - don't create duplicate records when a failover MDS is present (2442)
5776 - remove unnecessary mount age check (2332)
5777 - don't remove directory inodes from locks prematurely (2451)
5778 - don't break if MDS service name is the same as hostname (2103)
5779 - fix races in client write RPC generation when cache full (2482)
5781 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5784 - remove now-unused request->rq_obd (278)
5785 - if an allocation fails, print out how much memory we've used (1933)
5786 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5787 - add the "configurable stack size" patch to most series files (1256)
5788 - ability to write large log records, for 100+ OST configs (2306)
5789 - fix NULL deref when filter_prep fails (2314)
5790 - fix operator precedence error in filter_sync
5791 - dynamic allocation of socknal TX descriptors (2315)
5792 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5793 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5794 - ensure that CWARN messages in Portals always get to the syslog
5795 - __init/__exit are not for prototype decls (ldlm_init/exit)
5796 - x86-64 compile warning fixes
5797 - fix gateway LMC keyword conflict (2318)
5798 - fix MDS lock inversions in getattr/reint paths (1844)
5799 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5800 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5801 - DLM race condition prevented some lock evictions (2328)
5802 - ENOMEM detection and retry on socknal sends (2230)
5803 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5804 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5805 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5806 - better error messages when a client is rejected during recovery (1505)
5807 - avoid cancelling locks which were never granted, after failure (2330)
5808 - fix i_sem/journal inversion in mds_client_add (2333)
5809 - fix truncate/getattr lock cycle deadlock (2334)
5810 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5811 - fix two rare exit paths which could leak an l_lock() ref (2321)
5812 - fencepost error in MDS/OST orphan recovery (2226)
5813 - make log record alignment 8 bytes (1988)
5814 - lstripe now fails when requested offset > ost_count (2237)
5815 - ensure that all kernel series have a complete list.h (1607)
5816 - fix crashes in special-file operations (2316)
5817 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5819 - allow configurable automake binary, for testing new versions
5820 - small update to the lfs documentation
5822 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5824 * fix negative export reference count in fsfilt_sync (2312)
5826 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5827 * release candidate 0.9.1
5829 - orphans are moved into the PENDING directory for possible recovery
5830 - replayed opens now open by fid for orphan/rename safety (1042)
5831 - last close of an orphan inode generates a transno (683)
5832 - chdir() and mount() now pin the directory entry (1020)
5833 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5834 - discard very old requests without processing them (1502)
5835 - remove some common, well-understood CERRORs (1505)
5836 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5837 - clear "grant" flags in OST replies until OST grant code lands (1644)
5838 - fix read performance by not clobbering i_blksize on client (1598)
5839 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5840 - make LRU size a /proc tunable, clears locks when reduced (707)
5841 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5842 - server locks take references on exports now (1558)
5843 - build fixes for 2.4.20-rh trees (1663)
5844 - return an error from lov_create if all OSCs are inactive (1751)
5845 - fix import levels when a reconnect happens without a timeout (1597)
5846 - exit early from mds_open if we get a lookup error (1749)
5847 - partial page read at EOF wouldn't wait for disk before sending (1642)
5848 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5849 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5850 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5851 - write/truncate lock inversion (1639)
5852 - Don't auto-load obdclass, portals modules during cleanup (1495)
5853 - fix timestamps from jumping to "now" (1763)
5854 - extra journal assertions (1648)
5855 - add an extra multiunlink test (1771)
5856 - fix read_record/write_record API (1776)
5857 - fix leak of offset_extent, possible incorrect i_size later (1772)
5858 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5859 - Debugging for the kqswnal_get_idle_tx problems (1820)
5860 - Allow recovery to be attempted multiple times (1536)
5861 - Write out MDS last_rcvd file after it is first created (1600)
5862 - Fix tx_descriptor leak in failed transmit situations (1827)
5863 - ext3 journaling fixes for assertion failure after IO error (1871)
5864 - class_export_put() on freed export after completion AST error (1896)
5865 - Fix revalidate looping in VFS (1322)
5866 - Don't access a freed export during MDS_REINT timeout (1521)
5867 - Add open-unlink recovery support on the MDS (1673,1764)
5868 - Return an error if no MDS data was read from last_rcvd (1946)
5869 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5870 - Don't setup a disk device that doesn't match exported UUID (317)
5871 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5872 - avoid committing NULL handle in force close
5873 - local.sh is now a one-stripe LOV configuration
5874 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5875 - fix ext3 htree duplicate directory entry corruption (1516)
5876 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5877 - update to new LOV EA format (2097)
5878 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5880 2003-06-15 Phil Schwan <phil@clusterfs.com>
5883 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5884 - per-import/export recovery handling (958, 931, 959)
5885 - multiple last-rcvd slots, for serving multiple FSes (949)
5886 - connections are again shared between multiple imp/exports (963, 964)
5887 - "umount -f" would hang if any requests needed to be sent (393, 978)
5888 - avoid pinning large req buffer by copying for queued messages (989)
5889 - add "uuid" to "lctl device" command to help upcalls (991)
5890 - "open" RPCs with transnos would confuse recovery counters (1037)
5891 - do proper endian conversion of all wire messages (288, 340, 891)
5892 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5893 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5894 - fix ptlrpc_connection leak in target_handle_connect (1174)
5895 - fix import refcounting bug in OST and MDS cleanup (1134)
5896 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5897 - fix very unlikely obd_types race condition (501)
5898 - remove osc_open hack for echo_client (1187)
5899 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5900 - a failure in read_inode2 leads to deadlock (1139)
5901 - cancel ack-locks as soon as transaction is committed (1072)
5902 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5903 - make sure to commitrw after any preprw to avoid deadlock (1162)
5904 - failing to execute a file in a lustre FS would lock inode (1203)
5905 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5906 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5907 - obd_uuid2tgt would walk off the end of the list (1255)
5908 - on IA64 the osc would give portals incorrect bulk size (1258)
5909 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5910 - fix lock inversion caused by new llite matching code (1282)
5911 - limit the number of dirty pages on a client to 10MB (1286)
5912 - timed out locks were not being corrected cancelled (1289)
5913 - fix O_DIRECT above 4GB on IA-32 (1292)
5914 * major user-visible changes
5915 - fail out/fail over policy now controlled by the upcall (993)
5917 - add OBD_PING to check server availability and failure (954)
5918 - lustre messages are now sent in sending host order (288, 340, 891)
5919 - add eadatalen to MDS getattr reply (340)
5920 - OST read replies may contain second buffer, with per-page status (593)
5922 2003-03-11 Phil Schwan <phil@clusterfs.com>
5925 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5926 - failed open()s could cause deadlock; fixed (867, 869)
5927 - stop cancelling OST locks when files are closed (481)
5928 - overlapping XID spaces caused network corruption (851, 853)
5929 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5930 - setattr_raw added, to do single-RPC, server-side setattrs
5931 - lmc/lconf syntax change for OST UUIDs
5932 - fix crashy race condition between ptlrpc_free_req and osc_close
5933 - don't use request in mdc_enqueue if we hit a timeout (889)
5934 - don't set the inode i_size for regular files from the MDS (896)
5935 - handle out of order completion AST (842)
5936 - don't LBUG if a lock request times out after receiving AST (913)
5937 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5938 - if a bad lock AST arrives, send an error instead of dropping entirely
5939 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5940 - fix leak in bulk IO when only partially completed (899, 900, 926)
5941 - fix O_DIRECT for ia64 (55)
5942 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5943 - C-z after timeout could hang a process forever; fixed (977)
5945 - client-side I/O cache (678, 924, 929, 941, 970)
5947 - READPAGE and SETATTRs which don't take server-side locks get
5950 2003-02-11 Phil Schwan <phil@clusterfs.com>
5953 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5954 - Fix fsfilt_extN_readpage() to read a full page of directory
5955 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5956 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5957 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5958 - Most intent operations were removed, in favour of a new RPC mode
5959 that does a single RPC to the server and bypasses most of the VFS
5960 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5961 - Aggressively cancel local locks on DLM servers
5962 - mds_reint_unlink sends EA to the client if it's the last nlink.
5963 client uses that EA to unlink OST objects.
5964 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5965 - recursive symlinks were fixed (439)
5966 - fixed NULL deref in DEBUG_REQ
5967 - filter_update_lastobjid no longer calls sync, which annoyed extN
5968 - fixed multi-client small-writes to a single file problem (445)
5969 - fixed mtime updates during file writes (607)
5970 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5971 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5972 - fixed Linux OST size reporting problem (444, 656)
5973 - OST now updates object mtime with writes or setattr (607, 619)
5974 - client verifies file size before zeroing page past EOF (445)
5975 - OST now writes last allocated objid to disk with allocation (108)
5976 - LOV on echo now works (409)
5978 - mds_reint_unlink sends a new buffer, with the EA included. this
5979 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5980 set if a regular file was being unlinked, and it was the last link
5981 - use PtlGet from the target for bulk writes (315)
5982 - OST now updates object mtime with writes or setattr (607, 619)
5983 - LDLM now has a grant-time callback to revalidate locked items, if
5985 - Many MDS operations were reorganized to combat race conditions
5987 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5988 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5990 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5993 - Fully reactivate OST imports after reconnection (512, others)
5994 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5995 - More graceful error handling for truncating on dead OST (515)
5996 - Don't error out unless we're actually accessing dead stripes (474)
5997 - Fix garbage sizes when stripes are missing (410)
5998 - LRU counters were broken, causing constant lock purge (433, 432)
5999 - garbage on read from stripes with failed OSTs (441)
6000 - mark OSCs as active before reconnecting during recovery (438)
6001 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
6002 - lfind did not preserve OST order in output (443)
6003 - symlinks cause hung clients, incorrect data (439)
6004 - stop dereferencing request after dropping refcount (457)
6005 - don't LASSERT(spin_is_locked) on non-SMP (455)
6006 - fixes for many rename() bugs
6007 - fstat didn't correctly synchronize attributes (399)
6008 - server must handle lock cancellation during blocking AST prep (487)
6009 - bulk descriptors were free()d too soon (511)
6010 - fix paths in lconf, which would load incorrect modules (451, 507)
6011 - fix confusing lconf 'host not found' error message (386)
6012 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
6013 - fix race condition in mdc_blocking_ast() for inode access (526)
6014 - fix lov_unpackmd() unpacking wrong number of stripes (537)
6015 - fix lov_set_osc_active() marking wrong OSC inactive (440)
6016 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
6017 - fix multiple writes of stripe MD to MDS (358, maybe 519)
6018 - fix lstripe in several ways (kernel side) (527)
6019 - fix request leak in ldlm_cli_enqueue (262)
6020 - incorrect OSC was marked inactive after OST failure
6021 - call mds_fs_cleanup before unmounting filesystem (524)
6022 - fix races between taking ns_lock and ldlm_lock_change_resource
6023 - fix races updating LOV export open file list
6024 - fix lov_enqueue error path, avoid decref-ing bad lock handle
6025 - fix recovery NULL deref in ldlm_cli_cancel_unused
6026 - fix some DLM races by using new hash table for lock handles (419)
6027 - permit the client to specify desired inodes, at replay
6028 - duplicate requests when we queue them for replay reintegration
6029 - fix last_rcvd offset calculation
6030 - sync after each recovered transaction, so we always make progress
6031 - never, not always, ERESTART requests without transnos
6032 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
6034 - skip replay if the MDS says that the client is already connected
6035 - don't check for a recovery-enabled export to match lctl's UUID
6036 - don't INC_USE_COUNT for phantom exports
6037 - don't crash when cleaning up phantom exports (567)
6038 - don't double-finish or set replay data for errored mdc_open requests
6039 - abort requests when they time out, so we don't get old replies
6040 - send/receive replies for AST messages again
6041 - if the client says that it doesn't have the lock, cancel it on the
6043 - if we timeout during I/O, don't try to cancel an in-use lock; instead
6044 mark it as destroyed, it will all work out when decref is called
6045 - fix module use counts (22, 581)
6047 - ASTs now expect a reply (server cancels lock on error reply)
6049 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
6052 - fix many simultaneous client startup (392)
6053 - fix dentry->d_it clobbering
6054 - credentials weren't being shipped for readdir/getattr operations
6055 - remove invalid assertions triggered during some concurrent MD
6057 - proper Lustre versions added (336, 389)
6058 - fix memory leak for create error case (398)
6059 - fix LOV locking bug that would get cli/srv out of sync
6060 - fix echo client over LOV (409)
6061 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
6062 - fix double-O_EXCL intent crash (424)
6063 - avoid sending multiple lock CANCELs (352)
6065 - MDS can do multi-client recovery (modulo bugs in new code)
6067 - many updates, edits, cleanups
6069 2002-11-18 Phil Schwan <phil@clusterfs.com>
6072 - fix null d_it dereference (346)
6073 - fix full OST/dbench hang (333)
6074 - fix permission problem with file removal (286)
6075 - fix removal of OSCs from LOV when they fail
6076 - fix NULL deref during bulk timeout (214)
6077 - fix problems related to multiple filesystems on one MDS (241)
6078 - fixed serious subtle metadata locking bugs
6079 - free locks on clients when inodes are removed due to memory
6081 - fix inode pointer in lock data (285)
6082 - partial support for multiple MDS on a single host (241)
6083 - data locks weren't cancelled at clear_inode time (290, 311)
6084 - intent locks could lead to unbounded lock growth (205)
6085 - added a maximum lock count, an LRU list, and a flusher
6086 - fix multiple rename (365)
6087 - properly abstracted the echo client
6088 - OSC locked 1 byte too many; fixed
6089 - rewrote brw callback code:
6090 - fixed recovery bugs related to LOVs (306)
6091 - fixed too-many-pages-in-one-write crash (191)
6092 - fixed (again) crash in sync_io_timeout (214)
6093 - probably fixed callback-related race (385)
6095 - Add capability to MDS protocol
6096 - LDLM cancellations and callbacks on different portals
6098 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
6101 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
6102 - defer open object destruction to close time (601981, 138)
6103 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
6104 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
6105 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
6106 - don't call obd_setattr in ll_file_release for destroyed objects
6107 * protocol change to lustre_msg: move |version| and add |flags|
6108 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
6109 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
6110 * added replay of create, unlink, link and rename operations during
6111 MDS failover; recovery should be much more robust now
6112 * remove failed OSCs from LOVs (only lov_create uses this so far)
6113 * the lustre-HOWTO was brought (more) up to date (582544)
6115 2002-10-23 Phil Schwan <phil@clusterfs.com>
6118 - in-use dentries weren't being reused properly (617851)
6119 - prevent multiple LDLM setup (599178)
6120 - fix LOV size calculations for truncate (617853)
6121 - fix client handling of MDS intent errors (POSIX)
6122 - fix permission bug in lovstripe.c test (624321)
6123 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6124 - truncate past end of file could corrupt data
6125 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6126 - a race in recovery could return ETIMEDOUT to apps (623947)
6127 - building outside the source directory was fixed
6128 * the lustre-HOWTO was brought (more) up to date (582544)
6129 * major progress was made on recovery functionality
6131 2002-10-10 Phil Schwan <phil@clusterfs.com>
6134 - recovery deadlock fix
6135 - rm -rf causes LBUG fix (617817)
6136 - file open by multiple tasks fix (618962)
6137 - directory permissions bugs (602707 and 620007)
6138 - journal_stop fixed with locking (611313)
6139 - O_APPEND failures resolved (618273, perhaps 614459)
6140 - lconf PATH fix (619770)
6141 - IA64 build fix (621450)
6142 - RPC buffer sizes scale with amount of memory
6144 2002-10-01 Phil Schwan <phil@clusterfs.com>
6147 - locks would be cancelled without throwing away data pages,
6148 resulting in inconsistent data (605627)
6149 - inode attributes were not always being refreshed (605627, 612449)
6150 - lconf now continues to cleanup after lctl reports an error
6151 - MDS now enforces user permissions (602707)
6152 - lprocfs cleanup fixed, but not yet enabled (614157)
6153 - fixed infinite server hang, should a client not respond to an AST
6154 - avoid going into recovery if user calls readlink() with a buffer
6155 that's too small (613941)
6156 - AST RPCs no longer require replies (614867) -- this may be changed
6157 - don't crash server if client sends an IOV that's too big (611336)
6158 - fixed lock conversion deadlock (611892)
6159 - fixed the following of symlinks (614622)
6160 * recovery: the server can remove locks from a client that dies, other
6161 clients can make progress
6162 * more extN patch fixes
6163 * compile-time configurable ptlrpc buffer allocations
6165 - collaborative read cache document
6166 - Lustre Lite Performance CDR document-in-progress
6168 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6171 - fix typo in patch-2.4.18
6173 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6176 - clear ptlrpc request each time in handle_incoming_request()
6177 - unlink of files now destroys the object on the OST
6179 2002-09-19 Peter Braam <braam@clusterfs.com>
6181 * add hard link support
6182 * change obdfile creation method
6183 * kernel patch changed
6185 2002-09-19 Peter Braam <braam@clusterfs.com>
6188 - stack overflow bug in extN fixed
6190 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6192 * documentation updates
6193 - add man pages for config tools
6194 - update tests/README to describe testing with new config tools
6195 - finish metadata API descriptions
6196 * bug fixes and cleanups
6197 - statfs workaround for 16TB limit
6198 - LOV stripe allocation improved, can stripe on subset of OSTs
6199 - LOV file size/IO offset was wrong for files > 4GB in size
6200 - object EA data was being dropped, caused files to be unreadable
6201 - memory overflow with non-LOV OST caused memory corruption
6202 - fixed regression tests to work with new config tools, obdfilter
6203 - fixed bug when directory size became larger than 1 block
6204 - fixed bug (for single client case) when PWD was deleted
6205 - invalidate local directory pages when doing intent-based ops
6206 - avoid LDLM oops when lock callback contained bad data
6208 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6210 * documentation updates
6211 * bug fixes and cleanups
6212 - configuration tools
6215 - 64-bit compile warnings
6216 - 64-bit internal statfs data
6218 * test_brw on persistent OST devices
6220 * lprocfs (disabled)
6222 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6224 * documentation updates
6225 * bug fixes and cleanups
6226 * configuration tools
6228 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6231 * many small fixes to 0_5_4
6232 * io/network handling
6233 * thinkos in MDS operations
6235 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6238 * crucial basic fixes to 0.5.3
6239 * IOR, Iozone work over Elan
6242 2002-08-07 Phil Schwan <phil@clusterfs.com>
6243 * version 0_5_3, our first alpha
6244 * we use the new Portals iovs
6245 * documentation updates
6246 * bug fixes and cleanups
6247 * small changes in the DLM wire protocol
6249 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6250 * version 0_5_1 with some initial stability,
6251 * locking on MD and file I/O.
6252 * documentation updates
6253 * several bug fixes since 0.5.0
6254 * small changes in wire protocol
6256 2002-07-18 Phil Schwan <phil@clusterfs.com>
6258 * delivered as Lustre Light Alpha
6259 * fixed a crash after handling invalid MDS requests
6260 * fixed directory pages for architectures with non-4k pages sizes
6262 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6263 * release version v0_4_4
6264 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6266 * Fixes a number of interruption problems with OST operations.
6267 * Update documentation for portals header changes
6268 * Move all wire protocol structs/defines to lustre_idl.h
6269 * Fixes symlink length bug.
6270 * Add tcpdump to repository.
6272 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6273 * release version v0_4_3
6274 * Fixes statfs for inodes on extN.
6275 * Fixes bug in runtests which would delete /etc/hosts.
6276 * Use 64-bit object IDs wherever possible (not into VFS though)
6277 Remove ost_get_info, which is unused by lustre, and out of date.
6279 2002-07-03 Peter Braam <braam@clusterfs.com>
6280 * release version v0_4_2 Fixes a lookup error (type not passed)
6281 * move forward to head of Portals
6282 * move forward to latest Lustre kernel
6284 2002-06-25 Peter Braam <braam@clusterfs.com>
6285 * release version v0_4_1. Hopefully stable on single node use.