1 tbd Sun Microsystems, Inc.
3 * Server support for kernels:
4 2.6.18-164.11.1.el5 (RHEL 5)
5 2.6.18-164.11.1.0.1.el5 (OEL 5)
6 * Client support for unpatched kernels:
7 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
8 2.6.18-164.11.1.el5 (RHEL 5),
9 2.6.18-164.11.1.0.1.el5 (OEL 5)
10 2.6.16.60-0.42.8 (SLES 10),
12 * Recommended e2fsprogs version: 1.41.6.sun1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
18 Description: lfs check only allowed for root
19 Details : Code cleanups and added a regresion test of non-root use of
24 Description: MDS Oopses in ptlrpc_dispatch_difficult_reply()/__list_add()
25 Details : Fix hr->hr_index overload.
29 Description: Fix FIEMAP functionality.
33 Description: FC11 patchless client support added
37 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
38 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
42 Description: New acc-sm script and various test-framework enhancments.
46 Description: Previously landed patch was reverted.
50 Description: Update SLES10 kernel to 2.6.16.60-0.42.8 (patchless)
54 Description: Add ldiskfs maxdirsize mount option.
58 Description: Fix mkfs.lustre support for 16TB LUNs.
62 Description: Add cascading_rw to lustre mpi tests.
66 Description: Store SOM data in EA.
67 Details : Also several SOM related cleanups.
71 Description: Add refcount to md_open_data
75 Description: Use refcount on lov_request_set to prevent memory corruption.
79 Description: Statahead: use parent dir's i_mutex for synchronization.
83 Description: loadgen improvements.
87 Description: Enable peer health detection in ksocklnd and ko2iblnd.
91 Description: Control DCACHE_LUSTRE_INVALID flag with MDS_INODE_LOOKCUP lock
95 Description: Fix race on obd_devcie:md_stats
99 Description: Fix SLES 11 client builds.
100 Details : Resolved uninitalized pointer gcc warnings.
104 Description: Allow non-root access for "lfs check".
105 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
109 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
111 Severity : enhancement
113 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
114 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
118 Description: Sanity test_27p fails.
119 Details : OBD_FAIL_OST_ENOSPC has been moved to filter_statfs().
120 Sanity tests 27 were cleaned up and improved.
124 Description: Unify req format on client/servers
125 Details : Use new req_capsule API [almost] everywhere instead of old PTLRPC
126 buffers and swabbers approach..
131 Description: drop join file stuff.
133 Severity : enhancement
135 Description: Add 2.6.27 and SLES11 (patchless client) support
139 Description: Fix in ptlrpc_expire_one_request() to print the signed time
142 Severity : enhancement
144 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
147 Severity : enhancement
149 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
151 Severity : enhancement
153 Description: correctly handle big reply message.
154 Details : send LNet event if reply is bigger then buffer and adjust this buffer
159 Description: Drop unnecessary __GFP_NOMEMALLOC flag from filter_get_page()
161 Severity : enhancement
163 Description: Add support for OEL5.
167 Description: Conf-sanity.sh 50g test - deactivated OST should not cause a panic.
171 Description: client could not reconnect to OST because of an active request.
172 Details : The client now retries to reconnect to the same server, if a
173 connect request failed with EBUSY or -EAGAIN. Test case 35b
174 conf-sanity.sh has been created.
178 Description: Don't allow make backward step on assignin osc next id.
179 Details : race between allocation next id and ll_sync thread can be cause
180 of set wrong osc next id and can be kill valid ost objects.
184 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
188 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
192 Description: Error handling in osc_statfs_interpret() has been improved.
193 Details : Check in osc_statfs_interpret() for EBADR.
197 Description: Do not send statfs() requests to OSTs disabled by administrator.
198 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
200 Severity : enhancement
202 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
204 Severity : enhancement
206 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
208 Severity : enhancement
210 Description: Enhancement for lfs(1) command to use numeric uid/gid.
214 Description: Improved cleanup procedure for conf-sanity test_46a.
218 Description: Do not start bulk transfer, if client tried to reconnect.
219 Details : Do not start bulk transfer, if client tried to reconnect,
220 but failed due to active rpcs. Check for exp_abort_active_req
221 has been implemented.
225 Description: Prevent inconsistences between linux and lustre mount structures.
226 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
227 Make the sleep interruptible.
229 Severity : enhancement
231 Description: provide server to client comms path
232 Details : server to client communications path via new LDLM_SET_INFO rpc
236 Description: truncate starts GFP_FS allocation under transaction causing deadlock
237 Details : ldiskfs_truncate calls grab_cache_page which may start page
238 allocation under an open transaction. This may lead to
239 calling prune_icache with consequent lustre reentrance.
243 Description: send correctly lsm on open replay
244 Details : MDS is trust to LSM size on replay open, but client can set wrong size
249 Description: enable client interface failover
250 Details : When a child reconnects from another NID, properly update export
251 nid hash position and ldlm reverse import.
253 Severity : enhancement
255 Description: Communicate OST degraded/readonly state via statfs to MDS
256 Details : Flags in the statfs returned from OSTs indicate whether the
257 OST is in a degraded RAID state, or if the filesystem has
258 turned read-only after a filesystem error is detected.
263 Description: don't panic if EPROTO was hit when reading symlink
264 Details : correctly handling request reference in error cases.
266 Severity : enhancement
268 Description: Add LustreNetLink, a kernel-userspace communcation path.
270 Severity : enhancement
272 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
277 Description: access to llog context before init.
278 Details : move handling CATALOGS file at osc layer and forbid access to llog
282 Frequency : rare, only when client is writing to OST on local node
284 Description: Avoid deadlock for local client writes
285 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
286 memory freeing context. This allows OST threads to set the
287 PF_MEMALLOC flag on task structures in order to allocate memory
288 from reserved pools and complete IO.
289 Use GFP_HIGHUSER for OST allocations for non-local client writes,
290 so that the OST threads generate memory pressure and allow
291 inactive pages to be reclaimed.
293 Severity : enhancement
295 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
300 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
301 Details : move ldlm namespace creation in setup phase to avoid grab
302 _lprocfs_lock with cli_sem held.
306 Description: Temporarily disable grant shrink.
307 Details : Disable the feature for debugging.
310 Frequency : only when formatting test filesystems
312 Description: Unable to run several mkfs.lustre on loop devices at the same time
313 Details : mkfs.lustre returns error 256 on the concurrent loop devices
314 formatting. The solution is to proper handle the error.
316 Severity : enhancement
318 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
320 Severity : enhancement
322 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
327 Description: resolve race between obd_disconnect and class_disconnect_exports
328 Details : if obd_disconnect will be called to already disconnected export he
329 forget release one reference and osc module can't unloaded.
331 Severity : enhancement
333 Description: Allow tuning service thread via /proc
334 Details : For each service a new
335 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
336 created that can be used to set min/max thread counts, and get the
337 current number of running threads.
342 Descriptoin: don't return error if have partially created objects for file.
343 Details : lov_update_create_set uses set->set_success as index for created
344 objects, so if some requests will be failed, they will have hole
345 at end of array and qos_shrink_lsm for allocate correct lsm.
347 Severity : enhancement
349 Description: Update OFED support to 1.4.1
354 Description: Reduce small size read RPC
355 Details : Set read-ahead limit for every file and only do read-ahead when
356 available read-ahead pages are bigger than 1M to avoid small size
359 Severity : enhancement
361 Description: move AT tunable parameters for more consistent usage
362 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
364 Severity : enhancement
366 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
367 Details : allow skip disconnected ost for send statfs request and hide error
373 Description: fix racy locking of mballoc block bitmaps causing BUG
374 Details : The locking of the mballoc buddy bitmap and the in-memory
375 block bitmap was using two different spin locks in some
376 cases. This made it possible to incorrectly access the
377 mballoc bitmap while another process was modifying it,
378 causing a sanity assertion to fail. While no on-disk corruption
379 was reported, there was some risk of this happening.
382 Frequency : rare, on llog test 6
384 Descriptoin: don't allow connect to already connected import
385 Details : allowing connect to already connected import is hide connecting problem.
390 Descriptoin: allow kill process which wait statahead result
391 Details : for some reasons 'ls' can stick in waiting result from statahead,
392 in this case need way for kill this process.
394 Severity : enhancement
396 Description: Add state history info file, enhance import info file
397 Details : Track import connection state changes in a new osc/mdc proc file;
398 add overview-type data to the osc/mdc import proc file.
400 Severity : enhancement
402 Description: MDS create should not wait for statfs RPC while holding DLM lock.
404 Severity : enhancement
406 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
411 Descriptoin: OST grant too much space to client even there are not enough space.
412 Details : Client will shrink its grant cache to OST if there are no write
413 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
414 this grant cache if there are already not enough avaible space
415 (left_space < total_clients * 32M).
418 Frequency : start MDS on uncleanly shutdowned MDS device
420 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
421 Details : stay in waiting mds<>ost recovery finished produce random bugs
422 due race between two ll_sync thread for one lov target. send
423 ACTIVATE event only if connect realy finished and import have
427 Frequency : rare, connect and disconnect target at same time
429 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
430 Details : don't call obd_disconnect under lov_lock. this long time
431 operation and can block ptlrpcd which answer to connect request.
436 Descriptoin: don't lose wakeup for imp_recovery_waitq
437 Details : recover_import_no_retry or invalidate_import and import_close can
438 both sleep on imp_recovery_waitq, but we was send only one wakeup
442 Frequency : always with long access acl
444 Descriptoin: mds can't pack reply with long acl.
445 Details : mds don't control size of acl but they limited by reint/getattr
448 Severity : enhancement
450 Description: Update to SLES10 kernel-2.6.16.60-0.33.
452 Severity : enhancement
454 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
457 Frequency : start MDS on uncleanly shutdowned MDS device
459 Descriptoin: aborting recovery hang on MDS
460 Details : don't throttle destroy RPCs for the MDT.
463 Frequency : on remount
465 Description: external journal device not working after the remount
466 Details : clear dev_rdonly flag for external journal devices in
472 Description: shutdown vs evict race
473 Details : client_disconnect_export vs connect request race.
474 if client will evicted at this time - we start invalidate
475 thread without referece to import and import can be freed
479 Frequency : rare, need acl's on inode.
481 Description: client can't handle ost additional correctly
482 Details : if ost was added after client connected to mds client can have
483 hit lnet_try_match_md ... to big messages to wide striped files.
484 in this case need teach client to handle config events about add
485 lov target and update client max ea size at that event.
487 Severity : enhancement
489 Description: Changelogs
490 Details : Changelogs are a lightweight mechanism to track filesystem
491 metadata and namespace changes. The changelog is recorded
492 permanently on the MDTs, and is periodically "consumed" / purged
493 when records are no longer needed.
495 Severity : enhancement
497 Description: compact fld format with extents
498 Details : Store range of seq rather than every seq in FLD. Seq
499 controller update FLD rather than clients. In Case of CMD, mdt0
500 has FLD, all other metadata server act as non persistent proxy
501 for FLD queries and cache fld entries in fld cache.
506 Description: don't skip ost target if they assigned to file
507 Details : Drop slow OSCs if we can, but not for requested start idx.
508 This means "if OSC is slow and it is not the requested
509 start OST, then it can be skipped, otherwise skip it only
510 if it is inactive/recovering/out-of-space.
514 Description: more cleanup in mds_lov
515 Details : not send LOV EA under replay, we can't know about they size at this
516 time. Don't allow client connect to mds before any ost connected,
517 for avoid problems with LOV EA size and returning EIO to client.
519 Severity : enhancement
521 Description: Interoperability at server side (Disk interoperability)
523 Severity : enhancement
525 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
527 Severity : enhancement
529 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
531 Severity : enhancement
533 Description: New client IO stack (CLIO).
535 Severity : enhancement
537 Description: Commit on sharing. Eliminate inter-client dependencies between
538 uncommitted transactions by doing transaction commits.
539 Thereby clients may recovery independently.
542 Frequency : Create a symlink file with a very long name
544 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
545 Details : If there is no extra space in the request for early cancels,
546 ldlm_req_handles_avail() returns 0 instead of a negative value.
548 Severity : enhancement
550 Description: Add /proc entry for import status
551 Details : The mdc, osc, and mgc import directories now have
552 an import directory that contains useful import data for debugging
555 Severity : enhancement
557 Description: Re-disable certain /proc logging
558 Details : Enable and disable client's offset_stats, extents_stats and
559 extents_stats_per_process stats logging on the fly.
562 Frequency : Only on FC kernels 2.6.22+
564 Description: oops in statahead
565 Details : Do not drop reference count for the dentry from VFS when lookup,
566 VFS will do that by itself.
568 Severity : enhancement
570 Description: Generic /proc file permissions
571 Details : Set /Proc file permissions in a more generic way to enable non-
572 root users operate on some /proc files.
576 Description: Hitting mdc_commit_close() ASSERTION
577 Details : Properly handle request reference release in
578 ll_release_openhandle().
582 Description: quota recovery deadlock during mds failover
583 Details : This patch includes att18982, att18236, att18237 in bz14840.
585 1. fix osts hang when mds does failover with quotaon
586 2. prevent watchdog storm when osts threads wait for the
591 Frequency : only patchless client
592 Description: add workaround for race between add/remove dentry from hash
594 Severity : enhancement
596 Description: Allow OST glimpses to return PW locks
600 Description: LBUG when llog conf file is full
601 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
605 Description: Prevent import from entering FULL state when server in recovery
609 Description: service mount cannot take device name with ":"
610 Details : Only when device name contains ":/" will mount treat it as
616 Description: replace ptlrpcd with the statahead thread to interpret the async
617 statahead RPC callback
621 Frequency : on recovery
622 Description: I/O failures after umount during fail back
623 Details : if client reconnected to restarted server we need join to recovery
624 instead of find server handler is changed and process self
625 eviction with cancel all locks.
627 Severity : enhancement
629 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
631 Severity : enhancement
633 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
635 Severity : enhancement
637 Description: Upcall on Lustre log has been dumped
638 Details : Allow for a user mode script to be called once a Lustre log has
639 been dumped. It passes the filename of the dumped log to the
640 script, the location of the script can be specified via
641 /proc/sys/lnet/debug_log_upcall.
646 Description: avoid idr_remove called for id which is not allocated.
647 Details : Move assigment s_dev for clustered nfs to end of initialization,
648 to avoid problem with error handling.
653 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
654 Details : When connection is reused this not moved from CONN_UNUSED_HASH
655 into CONN_USED_HASH and this prodice warning when put connection
656 again in unused hash.
659 Severity : enhancement
661 Description: File striping can now be set to use an arbitrary pool of OSTs.
663 Severity : enhancement
665 Description: Export bytes_read/bytes_write count on OSC/OST.
669 Description: Early reply size mismatch, MGC loses connection
670 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
671 the connect flags are properly negotiated.
676 Description: quotas are not honored with O_DIRECT
677 Details : all writes with the flag O_DIRECT will use grants which leads to
678 this problem. Now using OBD_BRW_SYNC to guard this.
682 Description: add quota statistics
683 Details : 1. sort out quota proc entries and proc code.
684 2. add quota statistics
686 Severity : enhancement
688 Description: enable quota support for HEAD.
692 Description: Properly propagate oinfo flags from lov to osc for statfs
693 Details : restore missing copy oi_flags to lov requests.
695 Severity : enhancement
697 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
698 : l_getgroups(8), lst(8), routerstat(8)
699 Details : included man pages for llobdstat(8), llstat(8),
700 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
702 Severity : enhancement
704 Description: configure's --enable-quota should check the
705 : kernel .config for CONFIG_QUOTA
706 Details : configure is terminated if --enable-quota is passed but
707 : no quota support is in kernel
711 Description: Remove portals compatibility
712 Details : Remove portals compatibility, not interoperable with releases
717 Description: Resolve device initialization race
718 Details : Prevent proc handler from accessing devices added to the
719 obd_devs array but yet be intialized.
721 Severity : enhancement
723 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
725 Severity : enhancement
727 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
732 Description: Using wrong pointer in osc_brw_prep_request
733 Details : Access to array[-1] can produce panic if kernel compiled with
734 CONFIG_PAGE_ALLOC enabled
738 Description: Client runs out of low memory
739 Details : Consider only lowmem when counting initial number of llap pages
743 Description: *optional* service tags registration
744 Details : if the "service tags" package is installed on a Lustre node
745 When the filesystem is mounted, a local-node service tag will
746 be created. See http://inventory.sun.com/ for more information
747 about the Service Tags asset management system.
751 Description: Kernel BUG tries to release flock
752 Details : Lustre does not destroy flock lock before last reference goes
753 away. So always drop flock locks when client is evicted and
754 perform unlock regardless of successfulness of speaking to MDS.
758 Description: add refcount protection for osc callbacks, avoid panic on shutdown
762 Description: sanity test 65a fails if stripecount of -1 is set
763 Details : handle -1 striping on filesystem in ll_dirstripe_verify
768 Description: ASSERTION(CheckWriteback(page,cmd)) failed
769 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
772 Severity : enhancement
774 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
778 Description: do not process already freed flock
779 Details : flock can possibly be freed by another thread before it reaches
780 to ldlm_flock_completion_ast.
784 Description: LBUG during stress test
785 Details : Need properly lock accesses the flock deadlock detection list.
789 Description: oops in page fault handler
790 Details : kernel page fault handler can return two special 'pages' in error
791 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
795 Description: timeout with invalidate import.
796 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
797 should be handled by ptlrpcd. This produce long age waiting and
798 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
800 Severity : enhancement
802 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
806 Description: enable MGS and MDT services start separately
807 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
808 a co-located MGS without starting the MGS, which is a complement
809 to 'nosvc' mount option.
813 Frequency : after recovery
814 Description: precreate to many object's after del orphan.
815 Details : del orphan st in oscc last_id == next_id and this triger growing
816 count of precreated objects. Set flag LOW to skip increase count
817 of precreated objects.
821 Frequency : rare, on clear nid stats
822 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
823 Details : when clean nid stats sometimes try destroy live entry,
824 and this produce panic in free.
828 Description: Stack overflow during MDS log replay
829 ease stack pressure by using a thread dealing llog_process.
833 Description: wait until IO finished before start new when do lock cancel.
834 Details : VM protocol want old IO finished before start new, in this case
835 need wait until PG_writeback is cleared until check dirty flag and
836 call writepages in lock cancel callback.
838 Severity : enhancement
840 Description: using special macro for print time and cleanup in includes.
844 Description: mds_mfd_close() ASSERTION(rc == 0)
845 Details : In mds_mfd_close(), we need protect inode's writecount change
846 within its orphan write semaphore to prevent possible races.
850 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
854 Frequency : rare, on shutdown ost
855 Description: don't hit live lock with umount ost.
856 Details : shrink_dcache_parent can be in long loop with destroy dentries,
857 use shrink_dcache_sb instead.
861 Description: don't panic with use echo client
862 Details : echo client pass NULL as client nid pointer and this produce null
867 Description: fix build on ppc32
868 Details : compile code with -m64 flag produce wrong object file for ppc32.
872 Description: add message levels for liblustreapi
876 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
877 Details : Move the 'good_osts' check before the 'total_bavail' check. This
878 will result in an -EAGAIN and in the exit call path we call
879 alloc_rr() which will with increasing aggressiveness attempt to
880 aquire precreated objects on the minimum number of required OSCs.
884 Description: Use old size assignment to avoid deadlock
885 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
886 the scheduling while holding a spinlock. We do not need locking
887 for size in ll_update_inode() because size is only updated from
888 the MDS for directories or files without objects, so there is no
889 other place to do the update, and concurrent access to such inodes
890 are protected by the inode lock.
894 Description: resolve "_IOWR redefined" build error on SLES10
898 Description: dump the memory debugging after all modules are unloaded to
899 suppress false negative in conf_sanity test 39
901 Severity : enhancement
903 Description: build kernel-ib packages for OFED 1.3 in our release cycle
908 Description: fix SLES kernel versioning
909 Details : the kernel version for our SLES 10 kernel did not include a "-"
910 before the "smp" at the end. while this was not a problem in
911 general, it did mean that software trying to use the kernel
912 version to try to detect a vendor specific kernel would fail.
913 this was most evident by the OFED build scripts.
917 Description: Don't update lov_desc members until making sure they are valid
918 Details : When updating lov_desc members via proc fs, need fix their
919 validities before doing the real update.
923 Description: don't put request into delay list while invalidate in flight.
924 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
925 import in flight. this produce timeout for invalidate and sometimes
926 can cause stale data.
930 Frequency : on ppc only
931 Description: not convert ost objects for directory because it's not exist.
932 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
936 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
938 Severity : enhancement
940 Description: Update to support 2.6.22.14 vanilla kernel.
944 Frequency : rare, on recovery
945 Description: read procfs can produce deadlock in some situation
946 Details : Holding lprocfs lock which send rpc can produce block for destroy
947 obd objects and this also block reconnect with -EALREADY. This
948 isn't fix all lprocfs bugs - but make it rare.
950 Severity : enhancement
952 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
955 Frequency : frequent on X2 node
957 Description: mdc_set_open_replay_data LBUG
958 Details : Set replay data for requests that are eligible for replay.
962 Description: lustre_mgs: operation 101 on unconnected MGS
963 Details : When MGC is disconnected from MGS long enough, MGS will evict the
964 MGC, and late on MGC cannot successfully connect to MGS and a lot
965 of the error messages complaining that MGS is not connected.
969 Frequency : on network error
970 Description: panic with double free request if network error
971 Details : mdc_finish_enqueue is finish request if any network error ocuring,
972 but it's true only for synchronus enqueue, for async enqueue
973 (via ptlrpcd) this incorrect and ptlrpcd want finish request
976 Severity : enhancement
978 Description: client-side metadata stat-ahead during readdir(directory readahead)
979 Details : perform client-side metadata stat-ahead when the client detects
980 readdir and sequential stat of dir entries therein
983 Frequency : on start mds
985 Description: Implement get_info(last_id) in obdfilter.
988 Frequency : occasional
990 Description: Correctly check stale fid, not start epoch if ost not support SOM
991 Details : open with flag O_CREATE need set old fid in op_fid3 because
992 op_fid2 was overwritten with new generated fid, but mds can answer
993 with one of these two fids and both is not stale. Setattr
994 incorrectly started an epoch and assume will be called
995 done_writing, but without SOM done_writing ever being called.
998 Frequency : rare, depends on device drivers and load
1000 Description: MDS or OSS nodes crash due to stack overflow
1001 Details : Code changes in 1.8.0 increased the stack usage of some functions.
1002 In some cases, in conjunction with device drivers that use a lot
1003 of stack the MDS (or possibly OSS) service threads could overflow
1004 the stack. One change which was identified to consume additional
1005 stack has been reworked to avoid the extra stack usage.
1008 Frequency : occasional
1010 Description: Do not fail import if osc_interpret_create gets -EAGAIN
1011 Details : If osc_interpret_create got -EAGAIN it immediately exits and
1012 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
1013 oscc_has_objects and see OSC has no objests and call
1014 oscc_internal_create to resend create request.
1016 Severity : enhancement
1018 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1020 Severity : enhancement
1022 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1025 Frequency : very rare
1027 Description: Fix replay if there is an un-replied request and open
1028 Details : In some cases, older replay request will revert the
1029 mcd->mcd_last_xid on MDS which is used to record the client's
1030 latest sent request.
1032 Severity : enhancement
1034 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1036 Severity : enhancement
1038 Description: Add rhel5 support to HEAD.
1040 Serverity : enhancement
1042 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1047 Description: Don't allow skipping OSTs if index has been specified.
1048 Details : Don't allow skipping OSTs if index has been specified, make
1049 locking in internal create lots better.
1053 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1054 Details : access to bitfield in structure is always rounded to long
1055 and this produce problem with not atomic change any bit.
1059 Description: Lustre make rpms failed.
1060 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1061 builds Lustre rpms from tarball.
1064 Frequency : rare on shutdown ost
1066 Description: If llog cancel was not send before clean_exports phase, this can
1067 produce deadlock in llog code.
1068 Details : If llog thread has last reference to obd and call class_import_put
1069 this produce deadlock because llog_cleanup_commit_master wait when
1070 last llog_commit_thread exited, but this never success because was
1071 called from llog_commit_thread.
1075 Description: allow userland application know is lost one of stripes.
1076 Details : fill lvb_blocks with error code on ost and return it to
1077 application if error flag found.
1081 Description: NULL lov_tgts causing MDS oops
1082 Details : more safe checks for NULL lov_tgts for avoid oops.
1084 Severity : enhancement
1086 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1090 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
1091 Details : If an OST is down the MDS will hang indefinitely in
1092 obd_statfs_rqset() waiting for the statfs data. While for
1093 MDS QOS usage of statfs, it should not stuck in waiting.
1095 Severity : enhancement
1097 Description: remote_acl support
1098 Details : Support ACL-based permission check for remote user.
1099 Support setfacl/getfacl for remote user with the utils
1100 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
1101 the system "{s,g}etfacl" utils.
1103 Severity : enhancement
1105 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
1107 Severity : enhancement
1109 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1113 Description: make mgs_setparam() handle fsname containing dash
1114 Details : fsname containing a dash does not work with lctl conf_param
1116 Severity : enhancement
1118 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1120 Severity : enhancement
1122 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1126 Frequency : rare, at shutdown
1127 Description: access already free / zero obd_namespace.
1128 Details : if client_disconnect_export was called without force flag set,
1129 and exist connect request in flight, this can produce access to
1130 NULL pointer (or already free pointer) when connect_interpret
1131 store ocd flags in obd_namespace.
1135 Frequency : only at startup
1136 Description: not alloc memory with spinlock held.
1137 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1138 if any spinlock held.
1140 Severity : enhancement
1142 Description: make lustre randomly fail allocating memory
1143 Details : Make lustre randomly failed allocating memory for testing purpose.
1145 Severity : enhancement
1147 Description: lost problems with lov objid file
1148 Details : Fixes some scability and access to not inited memory problems
1149 in work with lov objdid file.
1154 Description: lfs find does not continue on file error
1155 Details : Continue other files processing when a file/dir is absent.
1159 Description: Inconsistent usage of lustre_pack_reply()
1160 Details : Standardize the usage of lustre_pack_reply() such that it
1161 always generate a CERROR on failure.
1164 Frequency : occasional
1166 Description: MDS hang or stay in waiting lock
1167 Details : If client receive lock with CBPENDING flag ldlm need send lock
1168 cancel as separate rpc, to avoid situation when cancel request
1169 can't processed due all i/o threads stay in wait lock.
1173 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1177 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1181 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1185 Description: Update to RHEL4 latest kernel.
1187 Severity : enhancement
1189 Description: Build SLES10 patchless client fails
1190 Details : The configure was broken by run ./configure with
1191 --with-linux-obj=.... argument for patchless client. When the
1192 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1193 can't search header adequately. Use absolute path such as
1194 -I($LINUX)/include instead.
1198 Description: interrupt oig_wait produce painc on resend.
1199 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1200 private set, and this produce situation when rq_ptlrpcd_data not
1201 copyed to new allocated request and triggered LBUG on assert
1202 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1203 ptlrpcd set if request is changed and can be safety changed to use
1208 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1209 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1210 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1211 these macros and replace them with their difinitions instead.
1215 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1216 Details : Update which_patch & target file for SLES10 latest kernel.
1219 Description: Remove client patches from SLES 10 kernel.
1220 Details : This causes SLES 10 clients to behave as patchless clients
1221 even on a Lustre-patched (server) kernel.
1223 Severity : enhancement
1225 Description: self-adjustable client's lru lists
1226 Details : use adaptive algorithm for managing client cached locks lru
1227 lists according to current server load, other client's work
1228 pattern, memory activities, etc. Both, server and client
1229 side namespaces provide number of proc tunables for controlling
1232 Severity : enhancement
1234 Description: light-weight GSS support
1235 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1236 and reduce performance overhead.
1238 Severity : enhancement
1240 Description: Linux keyring support
1241 Details : Support using service of Linux keyring for Lustre GSS internal
1242 context refresh/cache mechanism.
1246 Description: Fix errors in lfs documentation
1247 Details : Fixes man pages
1251 Description: don't use GFP_* in generic Lustre code.
1252 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1256 Description: obdclass is limited by single OBD_ALLOC(idarray)
1257 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1261 Description: warnings with build patchless client with vanila 2.6.19 and up
1262 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1263 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1267 Frequency : Only for SLES
1268 Description: sanity_quota fail test_1
1269 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1270 file, which makes TSTID[2] inunique.
1274 Frequency : Only for SLES9
1275 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1276 Details : Update target/ChangeLog/which_patch .
1280 Frequency : Only for test_10 in sanity-quota.sh
1281 Description: a bug in quota test code
1282 Details : checking a test flag in a wrong place
1286 Frequency : Only for test_13 in sanity-quota.sh
1287 Description: a bug in quota test script
1288 Details : shouldn't check overfull allocated quota
1292 Description: jbd statistics
1293 Details : Port older jbd statistics patch for sles10
1295 Severity : enhancement
1297 Description: Add printing inode info into message about error in writepage.
1302 Description: reply_lock_interpret crash due to race with it and lock cancel.
1303 Details : Do not replay locks that are being cancelled. Do not reference
1304 locks by their address during replay, just by their handle.
1308 Frequency : When flocks are used.
1309 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1310 Details : Flock locks might destroy just granted lock if it could be merged
1311 with another existing flock, this is done in completion handler,
1312 so teach ldlm_cli_enquque_fini that this is a valid case for
1318 Description: Oops in read and write path when failing to allocate lock.
1319 Details : Check if lock allocation failed and return error back.
1323 Description: lstripe command fails for valid OST index
1324 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1325 instead of lov->desc.ld_active_tgt_count.
1329 Description: sanity.sh failed test 103
1330 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1335 Description: df doesn't work properly if diskfs blocksize != 4K
1336 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1340 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1341 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1343 Severity : enhancement
1345 Description: Adaptive timeouts
1346 Details : RPC timeouts adapt to changing server load and network
1347 conditions to reduce resend attempts and improve recovery time.
1351 Description: llapi_file_create() does not allow some changes
1352 Details : add llapi_file_open() that allows specifying the mode and
1353 open flags, and also returns an open file handle.
1357 Description: merge and cleanup kernel patches.
1358 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1362 Description: Add journal checksum support.(Kernel part)
1363 Details : The journal checksum feature adds two new flags i.e
1364 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1365 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1366 indicates that the commit block contains the checksum for
1367 the blocks described by the descriptor blocks. Now commit
1368 record can be sent to disk without waiting for descriptor
1369 blocks to be written to disk. This behavior is controlled
1370 using JBD2_FEATURE_ASYNC_COMMIT flag.
1374 Description: OSS needs mutliple precreate threads
1375 Details : Add ability to start more than one create thread per OSS.
1379 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1380 Details : execute lfs setstripe on client
1384 Description: mds_obd_create error creating tmp object
1385 Details : When the user sets quota on root, llog will be affected and can't
1386 create files and write files.
1389 Frequency : Always on ia64 patchless client, and possibly others.
1391 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1392 Details : This allows the patchless client to be loaded on architectures
1393 without this export.
1397 Description: RedHat Update kernel for RHEL5
1398 Details : Add the kernel config file for RHEL5.
1402 Description: RedHat Update kernel for RHEL5
1403 Details : Modify the kernel config file more closer RHEL5.
1407 Description: Build failure against Centos5 (RHEL5)
1408 Details : Define PAGE_SIZE when it isn't present.
1412 Description: test_7 on sanity-quota.sh can't be used on separate machines
1413 Details : add facet to handle it
1417 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1418 Details : with NFS, the anon dentry's parent was set to itself in
1419 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1420 corresponding dentry other than use rec->ur_name.
1422 Severity : enhancement
1424 Description: lfs setstripe enhancement
1425 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1429 Description: enable data checksumming by default
1430 Details : enable checksum by default, allow --disable-checksum
1431 configure option and "-o nochecksum" mount option. Checksums
1432 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1436 Description: lustre support for RHEL5
1437 Details : Add support for RHEL5.
1441 Description: Client eviction due to failover config
1442 Details : after a connection loss, the lustre client should attempt to
1443 reconnect to the last active server first before trying the
1444 other potential connections.
1448 Description: when mds and osts use different quota unit(32bit and 64bit),
1449 quota will be released repeatly.
1450 Details : void sending multiple quota reqs to mds, which will keep the
1451 status between the reqs.
1455 Description: osts not allocated evenly to files
1456 Details : change the condition to increase offset_idx
1459 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1461 Description: Data corruption for OSTs that are formatted larger than 2TB
1463 Details : When generating the bio request for lustre file writes the
1464 sector number would overflow a temporary variable before being
1465 used for the IO. The data reads correctly from Lustre (which
1466 will overflow in a similar manner) but other file data or
1467 filesystem metadata may be corrupted in some cases.
1471 Description: Tune the kernel for good SCSI performance.
1472 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1473 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1478 Description: rewrite ext2-derived code in obdclass/uuid.c
1479 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1480 from scratch preserving functionality.
1484 Description: "lfs find -obd UUID" prints directories
1485 Details : "lfs find -obd UUID" will return all directory names instead
1486 of just file names. It is incorrect because the directories
1487 do not reside on the OSTs.
1491 Description: use i_size_read and i_size_write in 2.6 port
1492 Details : replace inode->i_size access with i_size_read/write()
1495 Frequency : when removing large files
1497 Description: scheduling issue during removal of large Lustre files
1498 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1499 It causes scheduling issues when removing large files (17TB in the
1503 Frequency : only with liblustre clients on XT3
1505 Description: evictions taking too long
1506 Details : allow llrd to evict clients directly on OSTs
1509 Frequency : only on ppc
1511 Description: /proc/fs/lustre/devices broken on ppc
1512 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1513 We should make sure the type of 'index' is loff_t and then cast
1514 explicitly as needed below. Do not assign an explicitly cast
1518 Frequency : only for rhel5
1520 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1521 Details : Modify the target file & which_kernel.
1523 Severity : enhancement
1525 Description: omit set fsid for export NFS
1526 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1527 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1531 Frequency : Always, for kernels after 2.6.16
1532 Description: Fix warning idr_remove called for id=.. which is not allocated.
1533 Details : Last kernels save old s_dev before kill super and not allow
1534 to restore from callback - restore it before call kill_anon_super.
1538 Description: buffer overruns could theoretically occur
1539 Details : llapi_semantic_traverse() modifies the "path" argument by
1540 appending values to the end of the origin string, and a
1541 overrun may occur. Adding buffer overrun check in liblustreapi.
1545 Description: Fix error on 'ls .' at the top of the Lustre mount.
1546 Details : Don't revalidate dentry if it is a root dentry.
1550 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1551 Details : Modify vm-tunables-rhel4.patch.
1555 Description: Kernel config for 2.6.18-vanilla.
1556 Details : Modify targets/2.6-vanilla.target.in.
1557 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1558 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1559 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1560 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1564 Description: improve handling recoverable errors
1565 Details : if request processig with error which can be recoverable on server
1566 request should be resend, otherwise page released from cache and
1571 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1572 Details : Modify target file & which_patch.
1573 A flaw was found in the IA32 system call emulation provided
1574 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1575 value could be stored in the %RAX register, which could trigger an
1576 out-of-bounds system call table access. An untrusted local user
1577 could exploit this flaw to run code in the kernel
1578 (ie a root privilege escalation). (CVE-2007-4573).
1582 Description: O_DIRECT bypasses client statistics.
1583 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1584 read_ahead_stats were updated. copied the stats section from
1585 osc_send_oap_rpc() into async_internal().
1589 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1593 Description: change order of libsysio includes
1594 Details : '#include sysio.h' should always come before '#include xtio.h'
1598 Description: handle "serious error: objid * already exists" more gracefully
1599 Details : If LAST_ID value on disk is smaller than the objects existing in
1600 the O/0/d* directories, it indicates disk corruption and causes an
1601 LBUG(). If the object is 0-length, then we should use the existing
1602 object. This will help to avoid a full fsck in most cases.
1604 Severity : enhancement
1606 Description: adapt the lustre_config script to support the upgrade case
1607 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1608 server targets to 1.6 in parallel.
1612 Description: To avoid grant space > avaible space when the disk is almost
1613 full. Without this patch you might see the error "grant XXXX >
1614 available" or some LBUG about grant, when the disk is almost
1616 Details : In filter_check_grant, for non_grant cache write, we should
1617 check the left space by if (*left > ungranted + bytes), instead
1618 of (*left > ungranted), because only we are sure the left space
1619 is enough for another "bytes", then the ungrant space should be
1620 increase. In client, we should update cl_avail_grant only there
1621 is OBD_MD_FLGRANT in the reply.
1625 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1626 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1627 described in CVE-2007-4573. This problem could allow untrusted
1628 local users to gain root access.
1631 Frequency : when using O_DIRECT and quotas
1633 Description: Incorrect file ownership on O_DIRECT output files
1634 Details : block usage reported by 'lfs quota' does not take into account
1635 files that have been written with O_DIRECT.
1640 Description: touch file failed when fs is not full
1641 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1642 otherwise we can get ENOSP while fs is not full.
1646 Description: parallel lock callbacks
1647 Details : Instead of sending blocking and completion callbacks as separated
1648 requests, adding them to a set and sending in parallel.
1651 Frequency : only for Cray XT3
1652 Bugzilla : 12829/13455
1653 Description: Changing primary group doesn't change the group lustre assigns to
1655 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1656 overridden with the primary group provided by the group upcall,
1657 whereas the supplied fsgid can be trusted if it is in the list of
1658 supplementary groups returned by the group upcall.
1660 Severity : enhancement
1662 Description: Allow masking D_WARNING, D_ERROR messages from console
1663 Details : Console messages can now be disabled via lnet.printk.
1667 Description: User code with malformed file open parameter crashes client node
1668 Details : Before packing join_file req, all the related reference should be
1669 checked carefully in case some malformed flags cause fake
1670 join_file req on client.
1674 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1675 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1676 mode is equal to the granted mode. However, it should also check
1677 that LDLM_FL_CP_REQD is not set.
1681 Description: Heavy nfs access might result in deadlocks
1682 Details : After ELC code landed, it is now improper to enqueue any mds
1683 locks under och_sem, because enqueue might want to decide to
1684 cancel open locks for same inode we are holding och_sem for.
1688 Description: Client eviction while running blogbench
1689 Details : A lot of unlink operations with concurrent I/O can lead to a
1690 deadlock causing evictions. To address the problem, the number of
1691 oustanding OST_DESTROY requests is now throttled to
1692 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1693 callbacks are processed in priority.
1697 Description: enable ACLs on MDS by default
1698 Details : ACLs must be enabled on MDS by default.
1701 Frequency : PPC/PPC64 only
1703 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1704 Details : fix duplicated definitions between asm-ppc64/types.h and
1705 lustre_types.h on PPC.
1708 Frequency : PPC/PPC64 only
1710 Description: asm-ppc/segment.h does not exist
1711 Details : fix compile issue on PPC.
1715 Description: better handle error messages in extents code
1718 Frequency : RHEL4 only
1720 Description: mkfs is very slow on IA64/RHEL4
1721 Details : A performance regression has been discovered in the MPT Fusion
1722 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1723 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1724 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1726 Severity : enhancement
1728 Description: SNMP support enhancement
1729 Details : Adding total number of sampled request for an MDS node in snmp
1732 Severity : enhancement
1734 Description: Optimize ldlm waiting list processing for PR extent locks
1735 Details : When processing waiting list for read extent lock and meeting read
1736 lock that is same or wider to it that is not contended, skip
1737 processing rest of the list and immediatelly return current
1738 status of conflictness, since we are guaranteed there are no
1739 conflicting locks in the rest of the list.
1743 Description: Time out and refuse to reconnect
1744 Details : When the failover node is the primary node, it is possible
1745 to have two identical connections in imp_conn_list. We must
1746 compare not conn's pointers but NIDs, otherwise we can defeat
1747 connection throttling.
1751 Description: port llog fixes from b1_6 into HEAD
1752 Details : Port llog reference couting and some llog cleanups from b1_6
1753 (bug 10800) into HEAD, for protect from panic and access to already
1754 free llog structures.
1758 Description: Detect stride IO mode in read-ahead
1759 Details : When a client does stride read, read-ahead should detect that and
1760 read-ahead pages according to the detected stride pattern.
1764 Description: data checksumming impacts single node performance
1765 Details : add support for several checksum algorithm. Currently, only CRC32
1766 and Adler-32 are supported. The checksum type can be changed on
1767 the fly via /proc/fs/lustre/osc/*/checksum_type.
1771 Description: use adler32 for page checksums
1772 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1777 Description: build for x2 fails
1778 Details : fix compile issue on Cray systems.
1782 Description: Properly match for duplicate locks
1783 Details : Due to different lock order from skiplists code, we need to
1784 traverse entire list for now
1787 Frequency : only on PPC/SLES10
1789 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1790 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1791 be defined. Add a hack in mkfs_lustre.c to work around this compile
1796 Description: LASSERT on MDS when client holding flock lock dies
1797 Details : ldlm pool logic depends on number of granted locks equal to
1798 number of released locks which is not true for flock locks, so
1799 just exclude such locks from consideration.
1803 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1804 Details : Use fsfilt_sync() for both the whole filesystem sync and
1805 individual file sync to eliminate dangerous inode locking
1806 with I_LOCK that can lead to a deadlock.
1810 Description: performance in 1.6.3
1811 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1812 sure that 1MB requests are not fragmented by the block layer.
1814 Severity : enhancement
1816 Description: organize the server-side client stats on per-nid basis
1817 Details : Change the structure of stats under obdfilter and mds to
1825 The "uuid"s file would list the uuids of _active_ exports.
1826 And the clear entry is to clear all stats and stale nids.
1828 Severity : enhancement
1830 Description: eliminate client locks in face of contention
1831 Details : file contention detection and lockless i/o implementation
1832 for contended files.
1836 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1839 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1841 Description: MDS may lose file striping (and hence file data) in some cases
1842 Details : If there are additional extended attributes stored on the MDS,
1843 in particular ACLs, SELinux, or user attributes (if user_xattr
1844 is specified for the client mount options) then there is a risk
1845 of attribute loss. Additionally, the Lustre file striping
1846 needs to be larger than default (e.g. striped over all OSTs),
1847 and an additional attribute must be stored initially in the
1848 inode and then increase in size enough to be moved to the
1849 external attribute block (e.g. ACL growing in size) for file
1854 Description: skiplist implementation simplification
1855 Details : skiplists are used to group compatible locks on granted list
1856 that was implemented as tracking first and last lock of each
1857 lock group the patch changes that to using doubly linked lists
1861 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1862 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1863 because LASSERT on both the data supplied by a client, and the
1864 data on disk is dangerous and incorrect.
1866 Severity : enhancement
1868 Description: Slow truncate/writes to huge files at high offsets.
1869 Details : Directly associate cached pages to lock that protect those pages,
1870 this allows us to quickly find what pages to write and remove
1871 once lock callback is received.
1875 Description: more ldlm soft lockups
1876 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1877 starve other threads from the resource lock for a long time in
1878 case of long waiting queue, so change the debug level from
1879 D_OTHER to the less frequently used D_INFO.
1881 Severity : enhancement
1883 Description: add -gid, -group, -uid, -user options to lfs find
1887 Description: Hung threads in invalidate_inode_pages2_range
1888 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1889 once one is completed. As a result, some RPCs are stuck in the
1890 queue and are never sent.
1894 Description: filter threads hungs on waiting journal commit
1895 Details : Cleanup filter group llog code, then only filter group llog will
1896 be only created in the MDS/OST syncing process.
1900 Description: Procfs and llog threads access destoryed import sometimes.
1901 Details : Sync the import destoryed process with procfs and llog threads by
1902 the import refcount and semaphore.
1904 Severity : enhancement
1906 Description: openlock cache of b1_6 port to HEAD
1911 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1912 Details : If insertion of an extent fails, then discard the inode
1913 preallocation and free data blocks else it can lead to duplicate
1918 Description: don't always update ctime in ext3_xattr_set_handle()
1919 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1920 In some cases the ctime should not be updated, for example for
1921 2.0->1.8 compatibility it is necessary to delete an xattr and it
1922 should not update the ctime.
1926 Bugzilla : 15713/16362
1927 Description: Assertion in iopen_connect_dentry in 1.6.3
1928 Details : looking up an inode via iopen with the wrong generation number can
1929 populate the dcache with a disconneced dentry while the inode
1930 number is in the process of being reallocated. This causes an
1931 assertion failure in iopen since the inode's dentry list contains
1932 both a connected and disconnected dentry.
1936 Description: assertion failure in ldlm_handle2lock()
1937 Details : fix a race between class_handle_unhash() and class_handle2object()
1938 introduced in lustre 1.6.5 by bug 13622.
1943 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1944 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1949 Description: oops in sd_iostats_seq_show()
1950 Details : unloading/reloading the scsi low level driver triggers a kernel
1951 bug when trying to access the sd iostat file.
1956 Description: Kernel panics during QLogic driver reload
1957 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1958 patch, causing memory corruption.
1963 Description: journal_dev option does not work in b1_6
1964 Details : pass mount option during pre-mount.
1966 Severity : enhancement
1968 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1969 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1970 extent information of a file. It can be used to map logical blocks
1971 in a file to physical blocks in the block device.
1975 Description: LDLM soft lockups - improvement
1976 Details : It is be possible to send the lock handle along with each read
1977 or write request because the client is already doing a lock match
1978 itself so there isn't any reason the OST should have to re-do that
1982 Frequency : only on Cray X2
1984 Description: X2 build failures
1985 Details : fix build failures on Cray X2.
1989 Description: xid & resent requests
1990 Details : Initialize RPC XID from clock at startup (randomly if clock is
1993 Severity : enhancement
1995 Description: Add lustre_start utility to start or stop multiple Lustre servers
2000 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
2001 Details : In case of memory pressure, list_del() can be called twice on
2002 req->rq_history_list, causing a kernel oops.
2006 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
2007 Details : fix stack overflow in the distributed lock manager by defering
2008 export eviction after a failed AST to the elt thread instead of
2009 handling it in the dlm interpret routine.
2013 Description: Convert some comments to new format.
2014 Details : Update documenting comments to match doxygen conventions.
2018 Description: Grammar fixes.
2019 Details : A couple of trivial spelling fixes.
2023 Description: OSD_COUNTERS-mandatory
2024 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
2025 to simplify the code.
2029 Description: simplify cmm_device freeing logic.
2030 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
2034 Description: Add lockdep support to dt_object_operations locking interface.
2035 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
2036 parameter indicating lock ordering. Update mdd code to use new
2041 Description: Introduce failloc constants for lockless IO tests.
2042 Details : Add two new failloc constants to test lockless IO. Only one of
2043 them in implemented---another is checked in yet to be landed
2048 Description: Add lockdep support for inode mutex.
2049 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
2050 in the situations where Lustre has to lock more than one inode
2055 Description: Add optional invariants checking support.
2056 Details : Add new LINVRNT() macro, optional on new --enable-invariants
2057 configure switch. This macro is to be used for consistency and
2058 sanity checks that are too expensive to be left in `production'
2063 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
2064 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
2069 Description: Add lockdep support for ldlm_lock and ldlm_resource.
2070 Details : Use spin_lock_nested() in (the only) situation where more than
2071 one ldlm_lock is locked simultaneously. Also, fix possible
2072 dead-lock in ldlm_lock_change_resource() by enforcing particular
2077 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
2078 Details : Instead of specifying each ldlm_lock call-back through separate
2079 parameter, wrap them into struct ldlm_callback_suite.
2083 Description: Kill join_lru obd method and its callers.
2084 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
2085 in memory---a requirement implemented through ->o_join_lru() obd
2086 method in HEAD. Remove this method and its users.
2090 Description: Add asynchronous ldlm ENQUEUE completion handler.
2091 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
2092 case that stock ldlm_completion_ast() cannot handle as it waits
2093 until lock is granted. Introduce new ldlm_completion_ast_async()
2094 for this. Also comment ldlm_completion_ast().
2098 Description: ldlm_error <-> errno conversion.
2099 Details : Add functions to map (rather arbitrary) between LDLM error codes
2100 and standard errno values. CLIO needs this to prevent LDLM specific
2101 constants from escaping ldlm and osc.
2105 Description: Kill unused ldlm_handle2lock_ns() function.
2106 Details : Kill unused ldlm_handle2lock_ns() function.
2110 Description: Add lu_ref support to ldlm_lock
2111 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
2112 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
2113 and ldlm_resource. LDLM interface has to be changed, because code that
2114 releases a reference on a lock, has to "know" what reference this is.
2115 In the most frequent case
2117 lock = ldlm_handle2lock(handle);
2119 LDLM_LOCK_PUT(lock);
2121 no changes are required. When any other reference (received _not_ from
2122 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
2123 instead of LDLM_LOCK_PUT().
2125 Arguably, changes are pervasive, and interface requires some discipline
2126 for proper use. On the other hand, it was very instrumental in finding
2127 a few leaked lock references.
2131 Description: Add ldlm_lock_addref_try().
2132 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
2133 attempts to addref a lock that might be being canceled
2138 Description: Add ldlm_weigh_callback().
2139 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
2140 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
2141 hard-coded `number of pages' logic.
2145 Description: Add lockdep annotations to llog code.
2146 Details : Use appropriately tagged _nested() locking calls in the places
2147 where llog takes more than one ->lgh_lock lock.
2151 Description: Add loi_kms_set().
2152 Details : Wrap kms updates into a helper function.
2156 Description: Constify instances of struct lsm_operations.
2157 Details : Constify instances of struct lsm_operations.
2161 Description: lu_conf support.
2162 Details : On a server, a file system object is uniquely identified
2163 by a fid, which is sufficient to locate and load all object
2164 state (inode). On a client, on the other hand, more data are
2165 necessary instantiate an object. Change lu_object_find() and
2166 friends to take additional `lu_conf' argument describing object.
2167 Typically this includes layout information.
2171 Description: lu_context fixes.
2172 Details : Introduce new lu_context functions that are needed on the client
2173 side, where some system threads (ptlrpcd) are shared by multiple
2174 modules, and so cannot be stopped during module shutdown.
2178 Description: Add start and stop methods to lu_device_type_operations.
2179 Details : Introduce two new methods in lu_device_type_operations, that are
2180 invoked when first instance of a given type is created and last one
2181 is destroyed respectively. This is need by CLIO.
2185 Description: Add lu_ref support to struct lu_device.
2186 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2187 track leaked references.
2191 Description: Introduce lu_kmem_descr.
2192 Details : lu_kmem_descr and its companion interface allow to create
2193 and destroy a number of kmem caches at once.
2197 Description: Fix lu_object finalization race.
2198 Details : Fix a race between lu_object_find() finding an object and its
2199 concurrent finalization. This race is (most likely) not possible
2200 on the server, but might happen on the client.
2204 Description: Introduce lu_ref interface.
2205 Details : lu_ref is a debugging module allowing to track references to
2206 a given object. It is quite cpu expensive, and has to be
2207 explicitly enabled with --enable-lu_ref. See usage description
2212 Description: Factor lu_site procfs stats into a separate function.
2213 Details : Separate lu_site stats printing code into a separate function
2214 that can be reused on a client.
2218 Description: Constify instances of struct {lu,dt,md}_device_operations.
2219 Details : Constify instances of struct {lu,dt,md}_device_operations.
2223 Description: Introduce struct md_site and move meta-data specific parts of
2224 struct lu_site here.
2225 Details : Move md-specific fields out of struct lu_site into special struct
2226 md_site, so that lu_site can be used on a client.
2230 Description: Kill mdd_lov_destroy().
2231 Details : Remove unused mdd code.
2235 Description: Add st_block checking to multistat.c.
2236 Details : Add st_block checking to multistat.c.
2240 Description: Add lu_ref support to struct obd_device.
2241 Details : Add lu_ref tracking to obd_device.
2245 Description: Kill obd_set_fail_loc().
2246 Details : Remove unused code.
2250 Description: Add special type for ptlrpc_request interpret functions.
2251 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2252 there. Actual usage will be in CLIO.
2256 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2257 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2258 doesn't work with lockdep.
2262 Description: Add rwv.c test program.
2263 Details : New testing program exercising readv(2) and writev(2) (Qian).
2267 Description: Add sendfile.c test program.
2268 Details : New testing program exercising sendfile(2) (Jay).
2272 Description: Ratelimit a message that can be very frequent.
2273 Details : Ratelimit a memory allocation failure message that can
2278 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2279 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2283 Description: Liblustre build fixes.
2284 Details : Liblustre build fixes.
2288 Description: libcfs: add cfs_{need,cond}_resched() interface.
2289 Details : libcfs: add cfs_{need,cond}_resched() definition and
2290 implementations for Linux, NT, and liblustre.
2292 Severity : enhancement
2294 Description: More exported tunables for mballoc
2295 Details : Add support for tunable preallocation window and new tunables for
2296 large/small requests
2300 Description: Detect corruption of block bitmap and checking for preallocations
2301 Details : Checks validity of on-disk block bitmap. Also it does better
2302 checking of number of applied preallocations. When corruption is
2303 found, it turns filesystem readonly to prevent further corruptions.
2307 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2308 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2309 window to 0. If it does hit cache miss, and read-pattern is still
2310 stride-io mode, does not reset the stride window, but also does not
2311 increase the stride window length in this case.
2315 Frequency : only for big-endian servers
2316 Description: Check if system is big-endian while mounting fs with extents feature
2317 Details : Mounting a filesystem with extents feature will fail on big-endian
2318 systems since ext3-based ldiskfs is not supported on big-endian
2319 systems. This can be over-riden with "bigendian_extents" mount option.
2321 Severity : enhancement
2323 Description: The root squash functionality
2324 Details : A security feature, which is to prevent users from being able
2325 to mount lustre on their desktop, run as root, and delete
2326 all of the files in the filesystem. The goal is accomplished by
2327 remapping user id (UID) and group id (GID) of the root user to
2328 a UID and GID specified by the system administartor via Lustre
2329 configuration management server (MGS). The functionality also
2330 allows to specify sets of clients for which the remapping does
2335 Description: Excessive recovery window
2336 Details : With AT enabled, the recovery window can be excessively long (6000+
2337 seconds). To address this problem, we no longer use
2338 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2339 timeout no longer depends on the service time, it is set to
2340 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2341 time via pb_service_time.
2345 Description: Watchdog triggered on MDS failover
2346 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2347 the OSTs know that the MDS "UUID" can be reused for the same export
2348 from a different NID, so we do not need to wait for the export to be
2352 Frequency : rare, only if using MMP with Linux RAID
2354 Description: MMP doesn't work with Linux RAID
2355 Details : While using HA for Lustre servers with Linux RAID, it is possible
2356 that MMP will not detect multiple mounts. To make this work we
2357 need to unplug the device queue in RAID when the MMP block is being
2358 written. Also while reading the MMP block, we should read it from
2359 disk and not the cached one.
2361 Severity : enhancement
2363 Description: open file using fid
2364 Details : A file can be opened using just its fid, like
2365 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2368 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2371 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2372 Details : add proper locking for creation of jbd_4k slab cache
2376 Description: MMP check in ext3_remount() fails without displaying any error
2377 Details : When multiple mount protection fails during remount, proper error
2380 Severity : enhancement
2382 Description: Allow stripe size to be up to 4G-64k
2383 Details : Fix math logic to allow large stripe sizes.
2387 Description: add check for >8TB ldiskfs filesystems
2388 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2389 Don't allow >8TB ldiskfs filesystems to be mounted without
2390 force_over_8tb mount option
2395 Descriptoin: OSS complains "too many blocks" and crashes
2396 Details : In filter_commitrw_write():
2397 1. lquota_chkquota();
2398 2. obdo_from_inode();
2399 3. lquota_pending_commit();
2400 "2." has changed uid and gid in this case, then caused
2401 lquota_chkquota() and lquota_pending_commit() to use
2402 different uid and gid. That is the root of the bug.
2406 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2407 Details : Detect and print out the 'stripe count = -1' case properly.
2411 Description: Out or order replies might be lost on replay
2412 Details : In ptlrpc_retain_replayable_request if we cannot find retained
2413 request with tid smaller then one currently being added, add it
2414 to the start, not end of the list.
2416 --------------------------------------------------------------------------------
2418 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2420 * Support for kernels:
2421 2.6.5-7.283 (SLES 9),
2422 2.6.9-55.EL (RHEL 4),
2423 2.6.16.46-0.14 (SLES 10),
2424 2.6.18.8 vanilla (kernel.org)
2425 * Client support for unpatched kernels:
2426 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2427 2.6.16 - 2.6.22 vanilla (kernel.org)
2428 * Due to recently discovered recovery problems, we do not recommend
2429 using patchless RHEL 4 clients with this or any earlier release.
2430 * Recommended e2fsprogs version: 1.39.cfs8
2431 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2432 * Starting with this release, the ldiskfs backing filesystem required
2433 by Lustre is now in its own package, lustre-ldiskfs. This package
2434 should be installed. It is versioned separately from Lustre and
2435 may be released separately in future.
2439 Description: block reactivating mgc import until all deactivates complete
2440 Details : Fix race when failing back MDT/MGS to itself (testing)
2443 Frequency : at statup only
2445 Description: mds_lov_synchronize race leads to various problems
2446 Details : simultaneous MDT->OST connections at startup can cause the
2447 sync to abort, leaving the OSC in a bad state.
2449 Severity : enhancement
2451 Description: add optional extra BUILD_VERSION info
2452 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2453 to override the lustre version.
2456 Frequency : 2.6.18 servers only
2458 Description: ll_kern_mount() doesn't release the module reference
2459 Details : The ldiskfs module reference count never drops down to 0
2460 because ll_kern_mount() doesn't release the module reference.
2465 Description: server LBUG when using old ost_num_threads parameter
2466 Details : Accept the old ost_num_threads parameter but warn that it
2467 is deprecated, and fix an off-by-one error that caused an LBUG.
2472 Description: Transient SCSI error results in persistent IO issue
2473 Details : iobuf->dr_error is not reinitialized to 0 between two
2477 Frequency : sometimes when underlying device returns I/O errors
2479 Description: OSTs not going read-only during write failures
2480 Details : OSTs are not remounted read-only when the journal commit threads
2481 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2482 instead of the ext3 wrappers.
2486 Description: poor connect scaling with increasing client count
2487 Details : Don't run filter_grant_sanity_check for more than 100 exports
2488 to improve scaling for large numbers of clients.
2491 Frequency : SLES10 only
2493 Description: sanity-quota.sh quotacheck failed: rc = -22
2494 Details : Quotas cannot be enabled on SLES10.
2497 Frequency : liblustre clients only
2499 Description: getdirentries does not give error when run on compute nodes
2500 Details : getdirentries does not fail when the size specified as an argument
2501 is too small to contain at least one entry
2503 Severity : enhancement
2505 Description: Add LNET router traceability for debug purposes
2506 Details : If a checksum failure occurs with a router as part of the
2507 IO path, the NID of the last router that forwarded the bulk data
2508 is printed so it can be identified.
2513 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2514 Details : Due to a race condition, liblustre clients were occasionally
2515 evicted incorrectly.
2517 Severity : enhancement
2519 Description: lfs setstripe use optional parameters instead of postional
2522 Severity : enhancement
2524 Description: Nanosecond timestamp support for ldiskfs
2525 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2526 resolution timestamps. There is not yet support for this at
2527 the Lustre filesystem level.
2530 Frequency : during server recovery
2532 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2533 Details : request with rq_no_resend flag not awake l_wait_event if they get
2537 Frequency : nfs export on patchless client
2539 Description: connectathon hang when test nfs export over patchless client
2540 Details : Disconnected dentry cannot be found with lookup, so we do not need
2541 to unhash it or make it invalid
2544 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2545 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2546 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2551 Description: open req refcounting wrong on reconnect
2552 Details : If reconnect happened between getting open reply from server and
2553 call to mdc_set_replay_data in ll_file_open, we will schedule
2554 replay for unreferenced request that we are about to free.
2555 Subsequent close will crash in variety of ways.
2556 Check that request is still eligible for replay in
2557 mdc_set_replay_data().
2562 Description: disable writes to filesystem when reading health_check file
2563 Details : the default for reading the health_check proc file has changed
2564 to NOT do a journal transaction and write to disk, because this
2565 can cause reads of the /proc file to hang and block HA state
2566 checking on a healthy but otherwise heavily loaded system. It
2567 is possible to return to the previous behaviour during configure
2568 with --enable-health-write.
2570 Severity : enhancement
2572 Description: 64-bit inode version
2573 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2574 to the inode. This will be required for version-based recovery.
2579 Description: MDS fails to start if a duplicate client export is detected
2580 Details : in some rare cases it was possible for a client to connect to
2581 an MDS multiple times. Upon recovery the MDS would detect this
2582 and fail during startup. Handle this more gracefully.
2584 Severity : enhancement
2586 Description: Add -o localflock option to simulate old noflock
2588 Details : This will achieve local-only flock/fcntl locks
2594 Description: log_commit_thread vs filter_destroy race leads to crash
2595 Details : Take import reference before releasing llog record semaphore
2600 Description: Wrong request locking in request set processing
2601 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2602 imp_delayed_list, in this place should be used imp_lock.
2605 Frequency : when reconnection
2607 Description: Grant Leak when osc reconnect to OST
2608 Details : When osc reconnect ost, OST(filter) should check whether it
2609 should grant more space to client by comparing fed_grant and
2610 cl_avail_grant, and return the granted space to client instead
2611 of "new granted" space, because client will call osc_init_grant
2612 to update the client grant space info.
2615 Frequency : when client reconnect to OST
2617 Description: Grant Leak when osc do resend and replay bulk write
2618 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2619 bulk write request, because the grant info while be sync between
2620 OSC and OST when reconnect, and we should ignore the grant info
2621 these of resend/replay write req.
2626 Description: Grant space more than avaiable left space sometimes.
2627 Details : When then OST is about to be full, if two bulk writing from
2628 different clients came to OST. Accord the avaliable space of the
2629 OST, the first req should be permitted, and the second one
2630 should be denied by ENOSPC. But if the seconde arrived before
2631 the first one is commited. The OST might wrongly permit second
2632 writing, which will cause grant space > avaiable space.
2635 Frequency : when client is evicted
2637 Description: Grant might be wrongly erased when osc is evicted by OST
2638 Details : when the import is evicted by server, it will fork another
2639 thread ptlrpc_invalidate_import_thread to invalidate the
2640 import, where the grant will be set to 0. While the original
2641 thread will update the grant it got when connecting. So if
2642 the former happened latter, the grant will be wrongly errased
2643 because of this race.
2648 Description: Checking Stale with correct fid
2649 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2650 to check whether it is stale, because sometimes, we want the
2651 enqueue happened anyway, and op_data.fid2 will not be initialized.
2653 Severity : enhancement
2655 Description: update patchless client
2656 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2659 Frequency : only with 2.4 kernel
2661 Description: random memory corruption
2662 Details : size of struct ll_inode_info is to big for union inode.u and this
2663 can be cause of random memory corruption.
2668 Description: Memory leak in recovery
2669 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2670 It should also check obd_fail before fsfilt_start, otherwise if
2671 fsfilt_start return -EROFS,(failover mds during mds recovery).
2672 then the req will return with repmsg->transno = 0 and rc = EROFS.
2673 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2674 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2675 be freed no matter whether fsfilt_commit success or not.
2678 Frequency : only with huge count clients
2680 Description: Prevents from taking the superblock lock in llap_from_page for
2682 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2683 for llap_from_page call in ll_removepage prevents from taking the
2684 superblock lock for a soon died page.
2689 Description: Not check open intent error before release open handle
2690 Details : in some rare cases, the open intent error is not checked before
2691 release open handle, which may cause
2692 ASSERTION(open_req->rq_transno != 0), because it tries to release
2693 the failed open handle.
2698 Description: Set cat log bitmap only after create log success.
2699 Details : in some rare cases, the cat log bitmap is set too early. and it
2700 should be set only after create log success.
2704 Description: Accessing a block bevice can re-enable I/O when Lustre is
2705 tearing down a device.
2706 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2711 Description: service threads may hog cpus when there are a lot of requests
2712 Details : Insert cond_resched to give other threads a chance to use some CPU
2717 Description: the cat log was not initialized in recovery
2718 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2719 unlink log on mds will not be initialized until mds post
2720 recovery. And also in mds post recovery, the unlink log will
2721 initialization will be done asynchronausly, so there will be race
2722 between add unlink log and unlink log initialization.
2726 Description: brw_stats were being printed incorrectly
2727 Details : brw_stats were being printed as log2 but all of them were not
2728 recorded as log2. Also remove some code duplication arising from
2729 filter_tally_{read,write}.
2733 Frequency : rare, only in recovery.
2734 Description: ASSERTION(req->rq_type != LI_POISON) failed
2735 Details : imp_lock should be held while iterating over imp_sending_list for
2736 prevent destroy request after get timeout in ptlrpc_queue_wait.
2740 Description: replay-single.sh test 52 fails
2741 Details : A lock's skiplist need to be cleanup when it being unlinked
2742 from its resource list.
2746 Description: Short directio read returns full requested size rather than
2748 Details : Direct I/O operations should return actual amount of bytes
2749 transferred rather than requested size.
2751 Severity : enhancement
2753 Description: metadata RPC reduction (e.g. for rm performance)
2754 Details : decrease the amount of synchronous RPC between clients and servers
2755 by canceling conflicing lock before the operation on the client
2756 and packing thier handles into the main operation RPC to server.
2758 Severity : enhancement
2760 Description: Async OSC create to avoid the blocking unnecessarily.
2761 Details : If a OST has no remain object, system will block on the creating
2762 when need to create a new object on this OST. Now, ways use
2763 pre-created objects when available, instead of blocking on an
2764 empty osc while others are not empty. If we must block, we block
2765 for the shortest possible period of time.
2769 Description: improve handling recoverable errors
2770 Details : if request processig with error which can be recoverable on server
2771 request should be resend, otherwise page released from cache and
2774 Severity : enhancement
2776 Description: refine locking for avoid write wrong info into lov_objid file
2777 Details : fix possible races with add new target and write/update data in
2780 --------------------------------------------------------------------------------
2782 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2787 Frequency : on some architectures
2789 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2790 Details : Uninitialized flags sometimes cause configuration commands to
2794 Frequency : patchless clients only
2796 Description: missing __iget() symbol export
2797 Details : The __iget() symbol export is missing. To avoid the need for
2798 this on patchless clients the deathrow inode reaper is turned
2799 off, and we depend on the VM to clean up old inodes. This
2800 dependency was during via the fix for bug 12181.
2802 --------------------------------------------------------------------------------
2804 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2806 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2807 INTEROPERATE with older versions automatically. In many cases a
2808 special upgrade step is needed. Please read the
2809 user documentation before upgrading any part of a live system.
2810 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2811 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2812 * WARNING: Lustre configuration and startup changes are required with
2813 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2817 Severity : enhancement
2819 Description: MountConf
2820 Details : Lustre configuration is now managed via mkfs and mount
2821 commands instead of lmc and lconf. New obd types (MGS, MGC)
2822 are added for dynamic configuration management. See
2823 https://mail.clusterfs.com/wikis/lustre/MountConf for
2826 Severity : enhancement
2828 Description: dynamic OST addition
2829 Details : OSTs can now be added to a live filesystem
2831 Severity : enhancement
2833 Description: startup order invariance
2834 Details : MDTs and OSTs can be started in any order. Clients only
2835 require the MDT to complete startup.
2837 Severity : enhancement
2839 Description: parallel, asynchronous orphan cleanup
2840 Details : orphan cleanup is now performed in separate threads for each
2841 OST, allowing parallel non-blocking operation.
2843 Severity : enhancement
2845 Description: optimized stripe assignment
2846 Details : stripe assignments are now made based on ost space available,
2847 ost previous usage, and OSS previous usage, in order to try
2848 to optimize storage space and networking resources.
2850 Severity : enhancement
2852 Description: Permanently set tunables
2853 Details : All writable /proc/fs/lustre tunables can now be permanently
2854 set on a per-server basis, at mkfs time or on a live system.
2856 Severity : enhancement
2858 Description: Lustre message v2
2859 Details : Add lustre message format v2.
2861 Severity : enhancement
2863 Description: client OST exclusion list
2864 Details : Clients can be started with a list of OSTs that should be
2865 declared "inactive" for known non-responsive OSTs.
2869 Description: SPEC SFS validation failure on NFS v2 over lustre.
2870 Details : Changes the blocksize for regular files to be 2x RPC size,
2871 and not depend on stripe size.
2873 Severity : enhancement
2875 Description: Multiple MD RPCs in flight.
2876 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2877 To avoid overly-overloading MDS, introduce a limit on number of
2878 MDS RPCs in flight for a single client and add /proc controls
2879 to adjust this limit.
2881 Severity : enhancement
2883 Description: client read/write statistics
2884 Details : Add client read/write call usage stats for performance
2885 analysis of user processes.
2886 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2887 file access. extents_stats shows chunk size distribution.
2888 extents_stats_per_process show chunk size distribution per
2891 Severity : enhancement
2893 Description: mds statistics
2894 Details : Add detailed mds operations statistics in
2895 /proc/fs/lustre/mds/*/stats.
2899 Description: Failure of copying files with lustre special EAs.
2900 Details : Client side always return success for setxattr call for lustre
2901 special xattr (currently only "trusted.lov").
2906 Description: Refcount LNET uuids
2907 Details : The global LNET uuid list grew linearly with every startup;
2908 refcount repeated list entries instead of always adding to
2911 Severity : enhancement
2913 Description: Dynamic service threads
2914 Details : Within a small range, start extra service threads
2915 automatically when the request queue builds up.
2918 Frequency : mixed-endian client/server environments
2920 Description: mixed-endian crashes
2921 Details : The new msg_v2 system had some failures in mixed-endian
2924 Severity : enhancement
2926 Description: Easy OST removal
2927 Details : OSTs can be permanently deactivated with e.g. 'lctl
2928 conf_param lustre-OST0001.osc.active=0'
2930 Severity : enhancement
2932 Description: MGS proc entries
2933 Details : Added basic proc entries for the MGS showing what filesystems
2936 Severity : enhancement
2938 Description: provide MGS failover
2939 Details : Added config lock reacquisition after MGS server failover.
2941 Severity : enhancement
2943 Description: add Linux 2.4 support
2944 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2948 Description: a large application tries to do I/O to the same resource and dies
2949 in the middle of it.
2950 Details : Check the req->rq_arrival time after the call to
2951 ost_brw_lock_get(), but before we do anything about
2952 processing it & sending the BULK transfer request. This
2953 should help move old stale pending locks off the queue as
2954 quickly as obd_timeout.
2957 Frequency : when an incorrect nid is specified during startup
2959 Description: ptlrpc connect to non-existant node causes kernel crash
2960 Details : LNET can't be re-entered from an event callback, which
2961 happened when we expire a message after the export has been
2962 cleaned up. Instead, hand the zombie cleanup off to another
2965 Severity : enhancement
2967 Description: plain/inodebits lock performance improvement
2968 Details : Grouping plain/inodebits in granted list by their request modes
2969 and bits policy, thus improving the performance of search through
2973 Frequency : only if OST filesystem is corrupted
2975 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2976 Details : for a short time RPCs with bulk IO are in the replay list,
2977 but replay of bulk IOs is unimplemented. If the OST filesystem
2978 is corrupted due to disk cache incoherency and then replay is
2979 started it is possible to trip an assertion. Avoid putting
2980 committed RPCs into the replay list at all to avoid this issue.
2983 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2985 Description: System hang on startup
2986 Details : This bug allowed the liblustre (e.g. catamount) client to
2987 return to the app before handling all startup RPCs. This
2988 could leave the node unresponsive to lustre network traffic
2989 and manifested as a server ptllnd timeout.
2991 Severity : enhancement
2993 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2994 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2995 boolean to control whether to print peer debug info when a
2996 client's RPC times out.
2999 Frequency : only for kernels with patches from Lustre below 1.4.3
3001 Description: Remove old rdonly API
3002 Details : Remove old rdonly API which unused from at least lustre 1.4.3
3005 Frequency : only for devices with external journals
3007 Description: Set external device read-only also
3008 Details : During a commanded failover stop, we set the disk device
3009 read-only while the server shuts down. We now also set any
3010 external journal device read-only at the same time.
3013 Frequency : when upgrading from 1.4 while trying to change parameters
3015 Description: The wrong (new) MDC name was used when setting parameters for
3016 upgraded MDT's. Also allows changing of OSC (and MDC)
3017 parameters if --writeconf is specified at tunefs upgrade time.
3020 Frequency : when setting specific ost indicies
3022 Description: QOS code breaks on skipped indicies
3023 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3024 created with --index need not be sequential.
3026 Severity : enhancement
3028 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3029 Details : The uninit_groups feature works in conjunction with the kernel
3030 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3031 up the pass1 processing of e2fsck. This is a read-only feature
3032 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3033 mount filesystems that have had this feature enabled.
3035 Severity : enhancement
3037 Description: Improve multi-block allocation algorithm to avoid fragmentation
3038 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3039 allocation locality and avoid filesystem fragmentation.
3041 ------------------------------------------------------------------------------
3043 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3045 * Support for kernels:
3046 2.4.21-47.0.1.EL (RHEL 3)
3047 2.6.5-7.283 (SLES 9)
3048 2.6.9-42.0.10.EL (RHEL 4)
3049 2.6.12.6 vanilla (kernel.org)
3050 2.6.16.27-0.9 (SLES 10)
3051 * Recommended e2fsprogs version: 1.39.cfs5
3053 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3057 Frequency : occasional, depends on client load and configuration
3058 Bugzilla : 12181, 12203
3059 Description: data loss for recently-modified files
3061 Details : In some cases it is possible that recently written or created
3062 files may not be written to disk in a timely manner (this should
3063 normally be within 30s unless client IO load is very high).
3064 The problem appears as zero-length files or files that are a
3065 multiple of 1MB in size after a client crash or client eviction
3066 that are missing data at the end of the file.
3068 This problem is more likely to be hit on clients where files are
3069 repeatedly created and unlinked in the same directory, clients
3070 have a large amount of RAM, have many CPUs, the filesystem has
3071 many OSTs, the clients are rebooted frequently, and/or the files
3072 are not accessed by other nodes after being written.
3074 The presence of the problem can be detected by looking at
3075 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
3076 smaller than the second (nr_unused) then dirty files will not
3077 be flushed automatically to disk. "sync; sleep 10" should be
3078 run several times on the node before unmounting it to update
3079 Lustre (this is also safe to run on nodes without this problem).
3081 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
3082 that can cause this same problem, so customers using that kernel
3083 also need to update the kernel in addition to Lustre. In order
3084 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3086 It is normal that files written just before a client crash (less
3087 than 30s) may not yet have been flushed to disk, even for local
3091 Frequency : frequent on thin XT3 nodes
3093 Description: UUID collision on thin XT3 Linux nodes
3094 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3095 randomly, since we relied on an insufficiently-seeded PRNG.
3100 Description: OSS hangs after "All ost request buffers busy"
3101 Details : A deadlock between quota and journal operations caused OSS
3102 hangs after printing "All ost request buffers busy."
3105 Frequency : always on liblustre builds
3107 Description: Cleanup compiler warnings on liblustre
3110 Frequency : always on liblustre builds on XT3
3112 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3116 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3117 > 32000 subdirectories
3118 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3119 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3120 subdirectory count crosses 32000. This will aid e2fsck to
3121 correctly handle more than 32000 subdirectories.
3124 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3126 Description: System hang on startup
3127 Details : This bug allowed the liblustre (e.g. catamount) client to
3128 return to the app before handling all startup RPCs. This
3129 could leave the node unresponsive to lustre network traffic
3130 and manifested as a server ptllnd timeout.
3132 Severity : enhancement
3134 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3135 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3136 boolean to control whether to print peer debug info when a
3137 client's RPC times out.
3142 Description: make O_SYNC working on 2.6 kernels
3143 Details : 2.6 kernels use different method for mark pages for write,
3144 so need add a code to lustre for O_SYNC work.
3149 Description: Failure to close file and release space on NFS
3150 Details : Put inode details into lock acquired in ll_intent_file_open.
3151 Use mdc_intent_lock in ll_intent_open to properly
3152 detect all kind of errors unhandled by mdc_enqueue.
3157 Description: proc file read during shutdown sometimes raced obd removal,
3159 Details : Add lock to prevent obd access after proc file removal.
3162 Frequency : Only for files larger than 4GB on 32-bit clients.
3164 Description: improperly doing page alignment of locks
3165 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3166 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3171 Description: RPCs being resent when they shouldn't be
3172 Details : Some RPCs that should not be resent are being resent. This
3173 can cause inconsistencies in the RPC state machine. Do not
3174 resend such requests.
3177 Frequency : rare, only with NFS export
3179 Description: Crash on NFS re-export node
3180 Details : under very unusual load conditions an assertion is hit in
3181 ll_intent_file_open()
3184 Frequency : only if OST filesystem is corrupted
3186 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3187 Details : for a short time RPCs with bulk IO are in the replay list,
3188 but replay of bulk IOs is unimplemented. If the OST filesystem
3189 is corrupted due to disk cache incoherency and then replay is
3190 started it is possible to trip an assertion. Avoid putting
3191 committed RPCs into the replay list at all to avoid this issue.
3196 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3197 Details : Large single O_DIRECT read and write calls can fail to allocate
3198 a sufficiently large buffer to process the request. In case of
3199 allocation failure the allocation is retried with a smaller
3200 buffer and broken into smaller requests.
3202 Severity : enhancement
3204 Description: Add -o localflock option to simulate old noflock behaviour.
3205 Details : This will achieve local-only flock/fcntl locks coherentness.
3210 Description: versioning check is incomplete
3211 Details : Checking the version difference of client vs. server, report
3212 error if the gap is too big.
3217 Description: add support PG_writeback bit
3218 Details : add support for PG_writeback bit for Lustre, for more carefull
3219 work with page cache in 2.6 kernel. This also fix some deadlocks
3220 and remove hack for work O_SYNC with 2.6 kernel.
3222 Severity : enhancement
3224 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3225 Details : The uninit_groups feature works in conjunction with the kernel
3226 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3227 up the pass1 processing of e2fsck. This is a read-only feature
3228 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3229 mount filesystems that have had this feature enabled.
3231 Severity : enhancement
3233 Description: Improve multi-block allocation algorithm to avoid fragmentation
3234 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3235 allocation locality and avoid filesystem fragmentation.
3237 ------------------------------------------------------------------------------
3239 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3241 * Support for kernels:
3242 2.6.9-42.0.3.EL (RHEL 4)
3243 2.6.5-7.276 (SLES 9)
3244 2.4.21-47.0.1.EL (RHEL 3)
3245 2.6.12.6 vanilla (kernel.org)
3246 2.6.16.21-0.8 (SLES10)
3247 * Recommended e2fsprogs version: 1.39.cfs2-0
3249 * The backwards-compatible /proc/sys/portals symlink has been removed
3250 in this release. Before upgrading, please ensure that you change
3251 any configuration scripts or /etc/sysctl.conf files that access
3252 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3253 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3254 in advance of the upgrade on any system running Lustre 1.4.6 or
3255 newer, since /proc/sys/lnet was added in that version.
3256 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3260 Frequency : only when quota is used
3262 Description: avoid scanning export list for quota master
3263 Details : Change the algorithms to avoid scanning export list in order
3264 to improve the efficiency.
3267 Frequency : MDS failover only, very rarely
3269 Description: "went back in time" messages on mds failover
3270 Details : The greatest transno may be lost when the current operation
3271 finishes with an error (transno==0) and the client's last_rcvd
3272 record is over-written. Save the greatest transno in the
3273 mds_last_transno for this case.
3276 Frequency : always for specific kernels and striping counts
3278 Description: client may get "Matching packet too big" without ACL support
3279 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3280 when trying to access files in certain configurations. The
3281 clients should in fact be denied when mounting because they do
3282 not understand ACLs.
3285 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3287 Description: many clients connecting with IO in progress causes connect timeouts
3288 Details : Avoid synchronous journal commits to avoid delays caused by many
3289 clients connecting/disconnecting when bulk IO is in progress.
3290 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3291 OST_IO_PORTAL to avoid delays behind potentially many pending
3295 Frequency : occasionally with multiple writers to a single file
3297 Description: shared writes to file may result in wrong size reported by stat()
3298 Details : Allow growing of kms when extent lock is cancelled
3301 Frequency : always with random mmap IO to multi-striped file
3303 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3304 Details : Only if the hole is at the end of OST object so that kms is too
3305 small. Fix is to increase kms accordingly in ll_nopage.
3308 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3310 Description: writes to a missing object would leak memory on the OST
3311 Details : If there is an inconsistency between the MDS and OST filesystems,
3312 such that the MDS references an object that doesn't exist, writes
3313 to that object will leak memory due to incorrect cleanup in the
3314 error handling path, eventually running out of memory on the OST.
3319 Description: Creating too long symlink causes lustre errors
3320 Details : Check symlink and name lengths before sending requests to MDS.
3323 Frequency : only if flock is enabled (not on by default)
3325 Description: posix locks not released on fd closure on 2.6.9+
3326 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3327 kernels, this caused such locks not to be released on fd close and
3328 then assertions on fs unmount about still used locks.
3331 Frequency : MDS failover only, very rarely
3333 Description: clients may get ASSERTION(granted_lock != NULL)
3334 Details : When request was taking a long time, and a client was resending
3335 a getattr by name lock request. The were multiple lock requests
3336 with the same client lock handle and
3337 mds_getattr_name->fixup_handle_for_resent_request found one of the
3338 lock handles but later failed with ASSERTION(granted_lock != NULL).
3343 Description: handle->h_buffer_credits > 0, assertion failure
3344 Details : h_buffer_credits is zero after truncate, causing assertion
3345 failure. This patch extends the transaction or creates a new
3349 Frequency : NFS re-export or patchless client
3350 Bugzilla : 11179, 10796
3351 Description: Crash on NFS re-export node (__d_move)
3352 Details : We do not want to hash the dentry if we don't have a lock.
3353 But if this dentry is later used in d_move, we'd hit uninitialised
3354 list head d_hash, so we just do this to init d_hash field but
3355 leave dentry unhashed.
3358 Frequency : NFS re-export or patchless client
3360 Description: NFS exports has problem with symbolic link
3361 Details : lustre client didn't properly install dentry when re-exported
3362 to NFS or running patchless client.
3365 Frequency : NFS re-export or patchless client
3367 Description: Various nfs/patchless fixes.
3368 Details : fixes reuse disconected alias for lookup process - this fixes
3369 warning "find_exported_dentry: npd != pd",
3370 fix permission error with open files at nfs.
3371 fix apply umask when do revalidate.
3374 Frequency : occasional
3376 Description: Crash on NFS re-export node
3377 Details : calling clear_page() on the wrong pointer triggered oops in
3378 generic_mapping_read().
3381 Frequency : rarely, using O_DIRECT IO
3383 Description: unaligned directio crashes client with LASSERT
3384 Details : check for unaligned buffers before trying any requests.
3387 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3389 Description: stale data returned from RAID cache
3390 Details : If only a small amount of IO is done to the RAID device before
3391 reading it again it is possible to get stale data from the RAID
3392 cache instead of reading it from disk.
3395 Frequency : always for sles10 kernel
3397 Description: sles10 support
3398 Details : ll_follow_link: compile fixes and using of nd_set_link
3399 under newer kernels.
3402 Frequency : depends on arch, kernel and compiler version, always on sles10
3405 Description: recursive or deep enough symlinks cause stack overflow
3406 Details : getting rid of large stack-allocated variable in
3410 Frequency : depends on hardware
3412 Description: lustre write performance loss in the SLES10 kernel
3413 Details : the performance loss is caused by using of write barriers in the
3414 ext3 code. The SLES10 kernel turns barrier support on by
3415 default. The fix is to undo that change for ldiskfs.
3417 ------------------------------------------------------------------------------
3419 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3421 * Support for kernels:
3422 2.6.9-42.0.3EL (RHEL 4)
3423 2.6.5-7.276 (SLES 9)
3424 2.4.21-47.0.1.EL (RHEL 3)
3425 2.6.12.6 vanilla (kernel.org)
3429 Frequency : quota enabled and large files being deleted
3431 Description: releasing more than 4GB of quota at once hangs OST
3432 Details : If a user deletes more than 4GB of files on a single OST it
3433 will cause the OST to spin in an infinite loop. Release
3434 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3439 Description: statfs data retrieved from /proc may be stale or zero
3440 Details : When reading per-device statfs data from /proc, in the
3441 {kbytes,files}_{total,free,avail} files, it may appear
3442 as zero or be out of date.
3445 Frequency : systems with MD RAID1 external journal devices
3447 Description: lconf's call to blkid is confused by RAID1 journal devices
3448 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3449 of returning all block devices that match the journal UUID.
3452 Frequency : always, for aggregate stripe size over 4GB
3454 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3455 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3456 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3459 Frequency : always if "lfs find" used on a local file/directory
3461 Description: "lfs find" segfaults if used on a local file/directory
3462 Details : The case where a directory component was not specified wasn't
3463 handled correctly. Handle this properly.
3466 Frequency : always on ppc64
3468 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3469 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3470 perform endianness conversion.
3473 Frequency : rarely (truncate to non-zero file size after write under load)
3474 Bugzilla : 10730, 10687
3475 Description: Files padded with zeros to next 4K multiple
3476 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3477 it is possible that files that are truncated to a non-zero size
3478 immediately after being written are filled with zero bytes beyond
3479 the truncated size. No file data is lost.
3481 Severity : enhancement
3483 Description: Allow recovery/failover for liblustre clients.
3484 Details : liblustre clients were unaware of failover configurations until
3487 Severity : enhancement
3489 Description: user file locks should fail when not mounting with flock option
3490 Details : Set up an error-returning stub in ll_file_operations.lock field
3491 to prevent incorrect behaviour when client is mounted without
3492 flock option. Also, set up properly f_op->flock field for
3496 Frequency : always on ia64
3498 Description: "lfs df" loops on printing out MDS statfs information
3499 Details : The obd_ioctl_data was not initialized and in some systems
3500 this caused a failure during the ioctl that did not return
3501 an error. Initialize the struct and return an error on failure.
3504 Frequency : SLES 9 only
3506 Description: Error of copying files with lustre special EAs as root
3507 Details : Client side always return success for setxattr call for lustre
3508 special xattr (currently only "trusted.lov").
3511 Frequency : rarely on clusters with both ia64+i386 clients
3513 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3514 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3515 clients are doing concurrent writes to the same file it is
3516 possible that the ia64 clients may LASSERT because the OST
3517 extent locks are not PAGE_SIZE aligned. Ensure that grown
3518 locks are always aligned on the request boundary.
3521 Frequency : specific use, occasional
3523 Description: Overwriting in use executable truncates on-disk binary image
3524 Details : If one node attempts to overwrite an executable in use by
3525 another node, we now correctly return ETXTBSY instead of
3526 truncating the file.
3528 Severity : enhancement
3530 Description: Async OSC create to avoid the blocking unnecessarily.
3531 Details : If a OST has no remain object, system will block on the creating
3532 when need to create a new object on this OST. Now, ways use
3533 pre-created objects when available, instead of blocking on an
3534 empty osc while others are not empty. If we must block, we block
3535 for the shortest possible period of time.
3540 Description: chmod on Lustre root is propagated to other clients
3541 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3542 invalid by the follow_mount time.
3545 Frequency : liblustre clients only
3547 Description: Race in 'instant cancel' lock handling could lead to such locks
3548 never to be granted in case of SMP MDS
3549 Details : Do not destroy not yet granted but cbpending locks in
3553 Frequency : replay/resend of open
3555 Description: non null lock assetion failure in mds_intent_policy
3556 Details : Trying to replay/resend lockless open requests resulted in
3557 mds_open() returning 0 with no lock. Now it sets a flag if
3558 a lock is going to be returned.
3560 Severity : enhancement
3562 Description: Checksum enhancements
3563 Details : New checksum enhancements allow for resending RPCs that failed
3566 Severity : enhancement
3568 Description: Tunables on number of dirty pages in cacche
3569 Details : Allow to set limit on number of dirty pages cached.
3574 Description: client crash on unmount - lock still has references
3575 Details : In some error handling cases it was possible to leak a lock
3576 reference on a client while accessing a file. This was not
3577 harmful to the client during operation, but would cause the
3578 client to crash when the filesystem is unmounted.
3581 Frequency : specific case, rare
3583 Description: ETXTBSY on mds though file not in use by client
3584 Details : ETXTBSY is no longer incorrectly returned when attempting to
3585 chmod or chown a directory that the user previously tried to
3586 execute or a currently-executing binary.
3589 Frequency : extremely rare except on liblustre-based clients
3591 Description: Lustre space not freed when files are deleted
3592 Details : Clean up open-unlinked files after client eviction. Previously
3593 the unlink was skipped and the files remained as orphans.
3598 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3599 Details : In some cases it was possible to send a blocking callback to
3600 a client doing a glimpse, even though that client didn't get
3601 a lock granted. When the glimpse lock is cancelled on the OST
3602 the freed lock is left on the waiting list and corrupted the list.
3605 Frequency : all core dumps
3607 Description: Broke core dumps to lustre
3608 Details : Negative dentry may be unhashed if parent does not have UPDATE
3609 lock, but some callers, e.g. do_coredump, expect dentry to be
3610 hashed after successful create, hash it in ll_create_it.
3612 ------------------------------------------------------------------------------
3614 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3616 * Support for kernels:
3617 2.6.9-42.0.2.EL (RHEL 4)
3618 2.6.5-7.276 (SLES 9)
3619 2.4.21-40.EL (RHEL 3)
3620 2.6.12.6 vanilla (kernel.org)
3624 Frequency : always on RHEL 3
3626 Description: Number of open files grows over time
3627 Details : The number of open files grows over time, whether or not
3628 Lustre is started. This was due to a filp leak introduced
3629 by one of our kernel patches.
3631 ------------------------------------------------------------------------------
3633 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3635 * Support for kernels:
3636 2.6.9-42.EL (RHEL 4)
3637 2.6.5-7.276 (SLES 9)
3638 2.4.21-40.EL (RHEL 3)
3639 2.6.12.6 vanilla (kernel.org)
3644 Bugzilla : 5719, 9635, 9792, 9684
3645 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3646 Details : If a server is under heavy load and cannot reply to new
3647 connection requests before the client resends the (re)connect,
3648 the connection handling code can behave badly if two service
3649 threads are concurrently handing separate (re)connections from
3650 the same client. Add better locking to the connection handling
3651 code, and ensure that only a single connection will be processed
3652 for a given client UUID, even if the lock is dropped.
3654 Severity : enhancement
3656 Description: add TCP zero-copy support to kernel
3657 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3658 sends if the hardware supports scatter-gather and checksumming.
3659 This allows socklnd to do client-write and server-read more
3660 efficiently and reduce CPU utilization from skbuf copying.
3663 Frequency : only if NFS exporting from client
3665 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3666 Details : If setting attributes on a file created under NFS that had
3667 never been opened it would be possible to oops the client
3668 if the file had no objects.
3671 Frequency : always for liblustre
3673 Description: liblustre client does MDS+OSTs setattr RPC for each write
3674 Details : When doing a write from a liblustre client, the client
3675 incorrectly issued an RPC to the MDS and each OST the file was
3676 striped over in order to update the timestamps. When writing
3677 with small chunks and many clients this could overwhelm the MDS
3678 with RPCs. In all cases it would slow down the write because
3679 these RPCs are unnecessary.
3681 Severity : enhancement
3683 Description: allow number of MDS service threads to be changed at module load
3684 Details : It is now possible to change the number of MDS service threads
3685 running. Adding "options mds mds_num_threads={N}" to the MDS's
3686 /etc/modprobe.conf will set the number of threads for the next
3687 time Lustre is restarted (assuming the "mds" module is also
3688 reloaded at that time). The default number of threads will
3689 stay the same, 32 for most systems.
3694 Description: OST crash if filesystem is unformatted or corrupt
3695 Details : If an OST is started on a device that has never been formatted
3696 or if the filesystem is corrupt and cannot even mount then the
3697 error handling cleanup routines would dereference a NULL pointer.
3702 Description: NULL pointer deref in llap_from_page.
3703 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3704 incorrect) mapping if the page was truncated/reclaimed while it was
3705 searched for. Check for this condition and skip such pages when
3706 doing readahead. Introduce extra check to llap_from_page() to
3707 verify page->mapping->host is non-NULL (so page is not anonymous).
3710 Frequency : Sometimes when using sys_sendfile
3712 Description: "page not covered by a lock" warnings from ll_readpage
3713 Details : sendfile called ll_readpage without right page locks present.
3714 Now we introduced ll_file_sendfile that does necessary locking
3715 around call to generic_file_sendfile() much like we do in
3719 Frequency : with certain MDS communication failures at client mount time
3721 Description: NULL pointer deref after failed client mount
3722 Details : a client connection request may delayed by the network layer
3723 and not be sent until after the PTLRPC layer has timed out the
3724 request. If the client fails the mount immediately it will try
3725 to clean up before the network times out the request. Add a
3726 reference from the request import to the obd device and delay
3727 the cleanup until the network drops the request.
3730 Frequency : occasionally during client (re)connect
3732 Description: assertion failure during client (re)connect
3733 Details : processing a client connection request may be delayed by the
3734 client or server longer than the client connect timeout. This
3735 causes the client to resend the connection request. If the
3736 original connection request is replied in this interval, the
3737 client may trip an assertion failure in ptlrpc_connect_interpret()
3738 which thought it would be the only running connect process.
3741 Frequency : only with obd_echo servers and clients that are rebooted
3743 Description: kernel BUG accessing uninitialized data structure
3744 Details : When running an obd_echo server it did not start the ping_evictor
3745 thread, and when a client was evicted an uninitialized data
3746 structure was accessed. Start the ping_evictor in the RPC
3747 service startup instead of the OBD startup.
3749 Severity : enhancement
3750 Bugzilla : 10193 (patchless)
3751 Description: Remove dependency on various unexported kernel interfaces.
3752 Details : No longer need reparent_to_init, exit_mm, exit_files,
3753 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3756 Frequency : rare (only users of deprecated and unsupported LDAP config)
3758 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3759 Details : LDAP apparently contains 'lustreName' attributes instead of
3760 'name'. A simple remapping of the name is sufficient.
3763 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3765 Description: waiting_locks_callback trips kernel BUG if client is evicted
3766 Details : Running with the dump_on_timeout debug flag turned on makes
3767 it possible that the waiting_locks_callback() can try to dump
3768 the Lustre kernel debug logs from an interrupt handler. Defer
3769 this log dumping to the expired_lock_main() thread.
3771 Severity : enhancement
3773 Description: Support NFS exporting on 2.6 kernels.
3774 Details : Implement non-rawops metadata methods for NFS server to use without
3775 changing NFS server code.
3778 Frequency : very rare (synthetic metadata workload only)
3780 Description: two racing renames might cause an MDS thread to deadlock
3781 Details : Running the "racer" program may cause one MDS thread to rename
3782 a file from being the source of a rename to being the target of
3783 a rename at exactly the same time that another thread is doing
3784 so, and the second thread has already enqueued these locks after
3785 doing a lookup of the target and is trying to relock them in
3786 order. Ensure that we don't try to re-lock the same resource.
3789 Frequency : only very large systems with liblustre clients
3791 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3792 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3793 all OSTs in parallel. This allows the eviction of stale liblustre
3794 clients to proceed much faster than if they were done in series,
3795 and also offers similar improvements for other set_info RPCs.
3800 Description: excessive CPU usage during initial read phase on client
3801 Details : During the initial read phase on a client, it would agressively
3802 retry readahead on the file, consuming too much CPU and impacting
3803 performance (since 1.4.5.8). Improve the readahead algorithm
3804 to avoid this, and also improve some other common cases (read
3805 of small files in particular, where "small" is files smaller than
3806 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3811 Description: MDS crash when receiving packet with unknown intent.
3812 Details : Do not LBUG in unknown intent case, just return -EFAULT
3814 Severity : enhancement
3815 Bugzilla : 9293, 9385
3816 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3817 Details : Do not serialize getattr (non-intent version) and statfs.
3820 Frequency : occasional, when OST network is overloaded/intermittent
3822 Description: client evicted by OST after bulk IO timeout
3823 Details : If a client sends a bulk IO request (read or write) the OST
3824 may evict the client if it is unresposive to its data GET/PUT
3825 request. This is incorrect if the network is overloaded (takes
3826 too long to transfer the RPC data) or dropped the OST GET/PUT
3827 request. There is no need to evict the client at all, since
3828 the pinger and/or lock callbacks will handle this, and the
3829 client can restart the bulk request.
3832 Frequency : Always when mmapping file with no objects
3834 Description: client crashes when mmapping file with no objects
3835 Details : Check that we actually have objects in a file before doing any
3836 operations on objects in ll_vm_open, ll_vm_close and
3842 Description: Request leak when working with deleted CWD
3843 Details : Introduce advanced request refcount tracking for requests
3844 referenced from lustre intent.
3846 Severity : Enhancement
3848 Description: Cache open file handles on client.
3849 Details : MDS now will return special lock along with openhandle, if
3850 requested and client is allowed to hold openhandle, even if unused,
3851 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3852 closing files for every read/write openration.
3854 Severity : Enhancement
3856 Description: Cache open negative dentries on client when possible.
3857 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3858 negative dentries on lock revocation.
3863 Description: Remounting a client read-only wasn't possible with a zconf mount
3864 Details : It wasn't possible to remount a client read-only with llmount.
3866 Severity : enhancement
3867 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3868 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3869 MPICH is included to add Lustre-specific ADIO interfaces.
3870 This is based closely on the UFS ADIO layer and only differs
3871 in file creation, in order to allow the OST striping to be set.
3872 This is user-contributed code and not supported by CFS.
3877 Description: extended inode attributes (immutable, append-only) work improperly
3878 when 2.4 and 2.6 kernels are used on client/server or vice versa
3879 Details : Introduce kernel-independent values for these flags.
3881 Severity : enhancement
3884 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3885 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3886 granularity. Now, Fractional MB granularity is be supported,
3887 this is very useful for low memory system.
3889 Severity : enhancement
3891 Description: Getattr by fid
3892 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3893 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3894 have locks and avoids getattr rpc requests for it on every stat.
3897 Frequency : Always, for filesystems larger than 2TB
3899 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3900 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3901 filesystems larger than 2TB to be mounted with mballoc enabled.
3904 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3906 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3907 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3908 filesystems larger than 2TB is mounted then the kernel will
3909 silently corrupt the start of the filesystem. CONFIG_LBD is
3910 enabled for all CFS-supported kernels, but the possibility of
3911 this happening with a modified kernel config exists.
3913 Severity : enhancement
3915 Description: add client O_DIRECT support for 2.6 kernels
3916 Details : It is now possible to do O_DIRECT reads and writes to files
3917 in the Lustre client mountpoint on 2.6 kernel clients.
3919 Severity : enhancement
3921 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3922 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3923 parallel, not waiting for response from every OST before sending
3924 a rpc to the next OST.
3929 Description: setattr vs write race when updating file timestamps
3930 Details : Client processes that update a file timestamp into the past
3931 right after writing to the file (e.g. tar) it is possible that
3932 the updated file modification time can be reset to the current
3933 time due to a race between processing the setattr and write RPC.
3935 Severity : enhancement
3937 Description: Bring 'lfs find' closer in line with regular Linux find.
3938 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3939 -print0 options and obtains all the needed info through the lustre
3942 Severity : enhancement
3944 Description: support up to 1024 configured devices on one node
3945 Details : change obd_dev array from statically allocated to dynamically
3946 allocated structs as they are first used to reduce memory usage
3951 Description: Flush dirty partially truncated pages during truncate
3952 Details : Immediatelly flush partially truncated pages in filter_setattr,
3953 this way we completely avoid having any pages in page cache on OST
3954 and can retire ugly workarounds during writes to flush such pages.
3959 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3960 Details : protect inode from truncation within vfs_unlink() context
3961 just take a reference before calling vfs_unlink() and release it
3962 when parent's i_sem is free.
3967 Description: last_id value checked outside lock on OST caused LASSERT failure
3968 Details : If there were multiple MDS->OST object precreate requests in
3969 flight, it was possible that the OST's last object id was checked
3970 outside a lock and incorrectly tripped an assertion. Move checks
3971 inside locks, and discard old precreate requests.
3974 Frequency : always, if extents are used on OSTs
3976 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3977 Details : OSTs using the extents format would not zero the high 16 bits of
3978 the index physical block number. This is not a problem for any
3979 OST filesystems smaller than 16TB, and no kernels support ext3
3980 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3981 new/modified files) and can be fixed for existing filesystems
3982 with e2fsprogs-1.39-cfs1.
3987 Description: import connection selection may be incorrect if timer wraps
3988 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3989 import connections to be ignored if the 32-bit jiffies counter
3990 wraps. Use a 64-bit jiffies counter.
3993 Frequency : very large clusters immediately after boot
3995 Description: LNET request buffers exhausted under heavy short-term load
3996 Details : If a large number of client requests are generated on a service
3997 that has previously never seen so many requests it is possible
3998 that the request buffer growth cannot keep up with the spike in
3999 demand. Instead of dropping incoming requests, they are held in
4000 the LND until the RPC service can accept more requests.
4003 Frequency : Sometimes during replay
4005 Description: Assertion failure in ll_local_open after replay.
4006 Details : If replay happened on an open request reply before we were able
4007 to set replay handler, reply will become not swabbed tripping the
4008 assertion in ll_local_open. Now we set the handler right after
4009 recognising of open request
4012 Frequency : very rare
4014 Description: kernel reports "badness in vsnprintf"
4015 Details : Reading from the "recovery_status" /proc file in small chunks
4016 may cause a negative length in lprocfs_obd_rd_recovery_status()
4017 call to vsnprintf() (which is otherwise harmless). Exit early
4018 if there is no more space in the output buffer.
4020 Severity : enhancement
4022 Description: clear OBD RPC statistics by writing to them
4023 Details : It is now possible to clear the OBD RPC statistics by writing
4024 to the "stats" file.
4029 Description: Client mtime is not the same on different clients after utimes
4030 Details : In some cases, the client was using the utimes() syscall on
4031 a file cached on another node. The clients now validate the
4032 ctime from the MDS + OSTs to determine which one is right.
4037 Description: Inability to activate failout mode
4038 Details : lconf script incorrectly assumed that in pythong string's numeric
4039 value is used in comparisons.
4042 Frequency : always with multiple stripes per file
4044 Description: Inefficient object allocation for mutli-stripe files
4045 Details : When selecting which OSTs to stripe files over, for files with
4046 a stripe count that divides evenly into the number of OSTs,
4047 the MDS is always picking the same starting OST for each file.
4048 Return the OST selection heuristic to the original design.
4053 Description: mount failures may take full timeout to return an error
4054 Details : Under some heavy load conditions it is possible that a
4055 failed mount can wait for the full obd_timeout interval,
4056 possibly several minutes, before reporting an error.
4057 Instead return an error as soon as the status is known.
4059 Frequency : quota enabled and large files being deleted
4061 Description: releasing more than 4GB of quota at once hangs OST
4062 Details : If a user deletes more than 4GB of files on a single OST it
4063 will cause the OST to spin in an infinite loop. Release
4064 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4069 Description: statfs data retrieved from /proc may be stale or zero
4070 Details : When reading per-device statfs data from /proc, in the
4071 {kbytes,files}_{total,free,avail} files, it may appear
4072 as zero or be out of date.
4075 Frequency : systems with MD RAID1 external journal devices
4077 Description: lconf's call to blkid is confused by RAID1 journal devices
4078 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4079 of returning all block devices that match the journal UUID.
4082 Frequency : always, for aggregate stripe size over 4GB
4084 Description: assertion fails when trying to use 4GB stripe size
4085 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
4086 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
4089 Frequency : always on ppc64
4091 Description: the first write on an ext3 filesystem with mballoc got stuck
4092 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4093 perform endianness conversion.
4095 ------------------------------------------------------------------------------
4097 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
4099 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4100 INTEROPERATE with older versions automatically. Please read the
4101 user documentation before upgrading any part of a live system.
4102 * WARNING: Lustre networking configuration changes are required with
4103 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4106 * Support for kernels:
4107 2.6.9-22.0.2.EL (RHEL 4)
4108 2.6.5-7.244 (SLES 9)
4109 2.6.12.6 vanilla (kernel.org)
4112 Severity : enhancement
4113 Bugzilla : 7981/8208
4114 Description: Introduced Lustre Networking (LNET)
4115 Details : LNET is new networking infrastructure for Lustre, it includes
4116 a reorganized network configuration mode (see the user
4117 documentation for full details) as well as support for routing
4118 between different network fabrics. Lustre Networking Devices
4119 (LNDs) for the supported network fabrics have also been
4120 created for this new infrastructure.
4122 Severity : enhancement
4123 Description: Introduced Access control lists
4124 Details : clients can set ACLs on files and directories in order to have
4125 more fine-grained permissions than the standard Unix UGO+RWX.
4126 The MDS must be started with the "-o acl" mount option.
4128 Severity : enhancement
4129 Description: Introduced filesystem quotas
4130 Details : Administrators may now establish per-user quotas on the
4133 Severity : enhancement
4135 Description: Configuration change for the XT3
4136 The PTLLND is now used to run Lustre over Portals on the XT3
4137 The configure option(s) --with-cray-portals are no longer used.
4138 Rather --with-portals=<path-to-portals-includes> is used to
4139 enable building on the XT3. In addition to enable XT3 specific
4140 features the option --enable-cray-xt3 must be used.
4145 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4146 Details : A race between allocating a new llcd and re-getting the llcd_lock
4147 allowed another thread to grab newly-allocated llcd.
4149 Severity : enhancement
4151 Description: 2.6 OST async journal commit and locking fix to improve performance
4152 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4153 2.6 kernels are now async as they already were in 2.4 kernels so
4154 that they can commit concurrently with the network bulk transfer.
4155 For block-allocated files the filter allocation semaphore is held
4156 to avoid filesystem fragmentation during allocation. BKL lock
4157 removed for 2.6 xattr operations where it is no longer needed.
4162 Description: lconf incorrectly determined whether two IP networks could talk
4163 Details : In some more complicated routing and multiple-network
4164 configurations, lconf will avoid trying to make a network
4165 connection to a disjoint part of the IP space. It was doing the
4166 math incorrectly for one set of cases.
4171 Description: Fix for potential infinite loop processing records in an llog.
4172 Details : If an llog record is corrupted/zeroed, it is possible to loop
4173 forever in llog_process(). Validate the llog record length
4174 and skip the remainder of the block on error.
4177 Frequency : occasional (liblustre only)
4179 Description: liblustre could not open files whose last component is a symlink
4180 Details : sysio_path_walk() would incorrectly pass the open intent to
4181 intermediate path components.
4184 Frequency : rare (liblustre only with non-standard tuning)
4185 Bugzilla : 7201 (7350)
4186 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4187 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4188 after it has been released, only do this for referenced locks
4190 Severity : enhancement
4192 Description: specifying an (invalid) directory default stripe_size of -1
4193 would reset the directory default striping
4194 Details : stripe_size -1 was used internally to signal directory stripe
4195 removal, now use "all default" to signal dir stripe removal
4196 as a directory striping of "all default" is not useful
4199 Frequency : common for large clusters running liblustre clients
4201 Description: doing an ls when liblustre clients are running is slow
4202 Details : sending a glimpse AST to a liblustre client waits for every AST
4203 to time out, as liblustre clients will not respond. Since they
4204 cannot cache data we refresh the OST lock LVB from disk instead.
4206 Severity : enhancement
4208 Description: doing an ls at the same time as file IO can be slow
4209 Details : enqueue and other "small" requests can be blocked behind many
4210 large IO requests. Create a new OST IO portal for non-IO
4211 requests so they can be processed faster.
4214 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4216 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4217 created file with mode 000
4218 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4219 allow an ftruncate() as a non-root user to a file with mode 000.
4220 Lustre now respects this flag to disable mode checks when
4221 truncating a file owned by the user
4224 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4227 Description: Revoking locks from clients that went dead or catatonic might take
4229 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4230 cancellation of such locks instant on servers without waiting for
4231 any reply from clients. Clients drops these locks when cancel
4232 notification from server is received without replying.
4235 Frequency : liblustre-only, when liblustre client dies or becomes busy
4237 Description: Doing ls on Linux clients can take a long time with active
4239 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4240 granting such locks to it in the first place if possible. Locks
4241 are taken by proxy on the OST during the read or write and
4242 dropped immediately afterward. Add connect flags handling, do
4243 not grant locks to liblustre clients for glimpse ASTs.
4245 Severity : enhancement
4247 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4248 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4249 sequential reads and behaves badly with applications doing random
4250 reads. Improve it by reading ahead at least read region, and
4251 avoiding excessive large RPC for small reads.
4253 Severity : enhancement
4255 Description: Creating more than 1000 files for a single job may cause a load
4256 imbalance on the OSTs if there are also a large number of OSTs.
4257 Details : qos_prep_create() uses an OST index reseed value that is an
4258 even multiple of the number of available OSTs so that if the
4259 reseed happens in the middle of the object allocation it will
4260 still utilize the OSTs as uniformly as possible.
4265 Description: OST or MDS may oops in ping_evictor_main()
4266 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4267 but doesn't restart at beginning of obd_exports_timed list
4270 Severity : enhancement
4272 Description: improve by-nid export eviction on the MDS and OST
4273 Details : allow multiple exports with the same NID to be evicted at one
4274 time without re-searching the exports list.
4277 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4279 Description: MDS may oops in groups_free()
4280 Details : in rare race conditions a newly allocated group_info struct is
4281 freed again, and this can be NULL. The 2.4 compatibility code
4282 for groups_free() checked for a NULL pointer, but 2.6 did not.
4285 Frequency : common for liblustre clients doing little filesystem IO
4286 Bugzilla : 9352, 7313
4287 Description: server may evict liblustre clients accessing contended locks
4288 Details : if a client is granted a lock or receives a completion AST
4289 with a blocking AST already set it would not reply to the AST
4290 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4293 Frequency : lfs setstripe, only systems with more than 160 OSTs
4295 Description: unable to set striping with a starting offset beyond OST 160
4296 Details : llapi_create_file() incorrectly limited the starting stripe
4297 index to the maximum single-file stripe count.
4300 Frequency : LDAP users only
4302 Description: lconf did not handle in-kernel recovery with LDAP properly
4303 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4305 Severity : enhancement
4307 Description: bind OST threads to NUMA nodes to improve performance
4308 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4309 node and do their allocations there to localize memory access
4311 Severity : enhancement
4313 Description: llmount can determine client NID directly from Myrinet (GM)
4314 Details : the client NID code from gmnalnid was moved directly into
4315 llmount, removing the need to use this or specifying the
4316 client NID explicitly when mounting GM clients with zeroconf
4319 Frequency : if client is started with down MDS
4321 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4322 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4323 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4324 ptlrpc_import_delayed_req() because we didn't check if the
4325 request was interrupted, and we also didn't break out of the
4326 event loop if there was no timeout
4331 Description: data loss during non-page-aligned writes to a single file from
4332 both multiple nodes and multiple threads on one node at same time
4333 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4334 inconsistency led to false short-reads, that were cached and later
4335 used by ->prepare_write() to fill in partially written page,
4336 leading to data loss.
4339 Frequency : always, if lconf --abort_recovery used
4341 Description: lconf --abort_recovery fails with 'Operation not supported'
4342 Details : lconf was attempting to abort recovery on the MDT device and not
4345 Severity : enhancement
4347 Description: remove cleanup logs
4348 Details : replace lconf-generated cleanup logs with lustre internal
4349 cleanup routines. Eliminates the need for client-cleanup and
4352 Severity : enhancement
4354 Description: add support for EAs (user and system) on lustre filesystems
4355 Details : it is now possible to store extended attributes in the Lustre
4356 client filesystem, and with the user_xattr mount option it
4357 is possible to allow users to store EAs on their files also
4359 Severity : enhancement
4361 Description: Add possibility (config option) to show minimal available OST free
4363 Details : When compiled with --enable-mindf configure option, statfs(2)
4364 (and so, df) will return least minimal free space available from
4365 all OSTs as amount of free space on FS, instead of summary of
4366 free spaces of all OSTs.
4368 Severity : enhancement
4370 Description: do not expand extent locks acquired on OST-side
4371 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4372 by server: they are not cached anyway.
4375 Frequency : when mmap is used/binaries executed from Lustre
4377 Description: Unmmap pages before throwing them away from read cache.
4378 Details : llap_shrink cache now attempts to unmap pages before discarding
4379 them (if unmapping failed - do not discard). SLES9 kernel has
4380 extra checks that trigger if this unmapping is not done first.
4385 Description: lconf didn't resolve symlinks before checking to see whether a
4386 given mountpoint was already in use
4389 Frequency : when migrating failover services
4390 Bugzilla : 6395, 9514
4391 Description: When migrating a subset of services from a node (e.g. failback
4392 from a failover service node) the remaining services would
4393 time out and evict clients.
4394 Details : lconf --force (implied by --failover) sets the global obd_timeout
4395 to 5 seconds in order to quickly disconnect, but this caused
4396 other RPCs to time out too quickly. Do not change the global
4397 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4399 Severity : enhancement
4400 Frequency : if MDS is started with down OST
4401 Bugzilla : 9439,5706
4402 Description: Allow startup/shutdown of an MDS without depending on the
4403 availability of the OSTs.
4404 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4405 Add appropriate locking and lov-osc refcounts for safe
4406 cleaning. Add osc abort_inflight calls in case the
4407 synchronize never started.
4410 Frequency : occasional (Cray XT3 only)
4412 Description: root not authorized to access files in CRAY_PORTALS environment
4413 Details : The client process capabilities were not honoured on the MDS in
4414 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4415 been accessed by an authorized user then root was able to access
4416 the file on the local client also. The root user capabilities
4417 are now allowed on the MDS, as this environment has secure UID.
4420 Frequency : occasional
4422 Description: ldiskfs "too long searching" message happens too often
4423 Details : A debugging message (otherwise harmless) prints too often on
4424 the OST console. This has been reduced to only happen when
4425 there are fragmentation problems on the filesystem.
4430 Description: Division by zero in statfs when all OSCs are inactive
4431 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4432 lov_statfs divides by value returned by lov_get_stripecnt().
4436 Bugzilla : 9489, 3273
4437 Description: First write from each client to each OST was only 4kB in size,
4438 to initialize client writeback cache, which caused sub-optimal
4439 RPCs and poor layout on disk for the first writen file.
4440 Details : Clients now request an initial cache grant at (re)connect time
4441 and so that they can start streaming writes to the cache right
4442 away and always do full-sized RPCs if there is enough data.
4443 If the OST is rebooted the client also re-establishes its grant
4444 so that client cached writes will be honoured under the grant.
4449 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4450 Details : Now I/O RPCs go to different portal number and (presumably) fast
4451 lock requests (and glimses) and other RPCs get their own service
4452 threads pool that should be able to service those RPCs
4455 Severity : enhancement
4457 Description: Ability to exchange lustre version between client and servers and
4458 issue warnings at client side if client is too old. Also for
4459 liblustre clients there is ability to refuse connection of too old
4461 Details : New 'version' field is added to connect data structure that is
4462 filled with version info. That info is later checked by server and
4466 Frequency : rare, liblustre only.
4467 Bugzilla : 9296, 9581
4468 Description: Two simultaneous writes from liblustre at offset within same page
4469 might proceed at the same time overwriting eachother with stale
4471 Details : I/O lock withing llu_file_prwv was released too early, before data
4472 actually was hitting the wire. Extended lock-holding time until
4473 server acknowledges receiving data.
4476 Frequency : extremely rare. Never observed in practice.
4478 Description: avoid generating lustre_handle cookie of 0.
4479 Details : class_handle_hash() generates handle cookies by incrementing
4480 global counter, and can hit 0 occasionaly (this is unlikely, but
4481 not impossible, because initial value of cookie counter is
4482 selected randonly). Value of 0 is used as a sentinel meaning
4483 "unassigned handle" --- avoid it. Also coalesce two critical
4484 sections in this function into one.
4486 Severity : enhancement
4488 Description: allow liblustre clients to delegate truncate locking to OST
4489 Details : To avoid overhead of locking, liblustre client instructs OST to
4490 take extent lock in ost_punch() on client's behalf. New connection
4491 flag is added to handle backward compatibility.
4493 Severity : enhancement
4494 Bugzilla : 4928, 7341, 9758
4495 Description: allow number of OST service threads to be specified
4496 Details : a module parameter allows the number of OST service threads
4497 to be specified via "options ost ost_num_threads={N}" in the
4498 OSS's /etc/modules.conf or /etc/modprobe.conf.
4502 Bugzilla : 6146, 9635, 9895
4503 Description: servers crash with bad pointer in target_handle_connect()
4504 Details : In rare cases when a client is reconnecting it was possible that
4505 the connection request was the last reference for that export.
4506 We would temporarily drop the export reference and get a new
4507 one, but this may have been the last reference and the export
4508 was just destroyed. Get new reference before dropping old one.
4510 Severity : enhancement
4511 Frequency : if client is started with failover MDS
4513 Description: Allow multiple MDS hostnames in the mount command
4514 Details : Try to read the configuration from all specified MDS
4515 hostnames during a client mount in case the "primary"
4518 Severity : enhancement
4520 Description: Stop sending data to evicted clients as soon as possible.
4521 Details : Check if the client we are about to send or are sending data to
4522 was evicted already. (Check is done every second of waiting,
4523 for which l_wait_event interface was extended to allow checking
4524 of exit condition at specified intervals).
4527 Frequency : rare, normally only when NFS exporting is done from client
4529 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4531 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4534 Frequency : always, if srand() is called before liblustre initialization
4536 Description: Liblustre uses system PRNG disturbing its usage by user application
4537 Details : Introduce internal to lustre fast and high-quality PRNG for
4538 lustre usage and make liblustre and some other places in generic
4539 lustre code to use it.
4541 Severity : enhancement
4542 Bugzilla : 9477, 9557, 9870
4543 Description: Verify that the MDS configuration logs are updated when xml is
4544 Details : Check if the .xml configuration logs are newer than the config
4545 logs stored on the MDS and report an error if this is the case.
4546 Request --write-conf, or allow starting with --old_conf.
4548 Severity : enhancement
4550 Description: Handle symlinks in the path when checking if Lustre is mounted.
4551 Details : Resolve intermediate symlinks when checking if a client has
4552 mounted a filesystem to avoid duplicate client mounts.
4557 Description: lconf can hit an error exception but still return success.
4558 Details : The lconf command catches the Command error exception at the top
4559 level script context and will exit with the associated exit
4560 status, but doesn't ensure that this exit status is non-zero.
4565 Description: failure of ptlrpc thread startup can cause oops
4566 Details : Starting a ptlrpc service thread can fail if there are a large
4567 number of threads or the server memory is very fragmented.
4568 Handle this without oopsing.
4571 Frequency : always, only if liblustre and non-default acceptor port was used
4573 Description: liblustre cannot connect to servers with non-default acceptor port
4574 Details : tcpnal_set_default_params() was not called and was therefore
4575 ignoring the environment varaible TCPNAL_PORT, as well as other
4576 TCPNAL_ environment variables
4581 Description: two objects could be created on the same OST for a single file
4582 Details : If an OST is down, in some cases it was possible to create two
4583 objects on a single OST for a single file. No problems other
4584 than potential performance impact and spurious error messages.
4588 Bugzilla : 5681, 9562
4589 Description: Client may oops in ll_unhash_aliases
4590 Details : Client dcache may become inconsistent in race condition.
4591 In some cases "getcwd" can fail if the current directory is
4597 Description: Inode refcounting problems in NFS export code
4598 Details : link_raw functions used to call d_instantiate without obtaining
4599 extra inode reference first.
4603 Bugzilla : 9942, 9903
4604 Description: Referencing freed requests leading to crash, memleaks with NFS.
4605 Details : We used to require that call to ll_revalidate_it was always
4606 followed by ll_lookup_it. Also with revalidate_special() it is
4607 possible to call ll_revalidate_it() twice for the same dentry
4608 even if first occurence returned success. This fix changes semantic
4609 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4610 reference on a request referred from the intent.
4611 ll_intent_release() then releases such a request.
4614 Frequency : rare, normally benchmark loads only
4616 Description: unlinked inodes were kept in memory on the client
4617 Details : If a client is repeatedly creating and unlinking files it
4618 can accumulate a lot of stale inodes in the inode slab cache.
4619 If there is no other client load running this can cause the
4620 client node to run out of memory. Instead flush old inodes
4621 from client cache that have the same inode number as a new inode.
4624 Frequency : SLES9 2.6.5 kernel and long filenames only
4625 Bugzilla : 9969, 10379
4626 Description: utime reports stale NFS file handle
4627 Details : SLES9 uses out-of-dentry names in some cases, which confused
4628 the lustre dentry revalidation. Change it to always use the
4632 Frequency : rare, unless heavy write-truncate concurrency is continuous
4633 Bugzilla : 4180, 6984, 7171, 9963, 9331
4634 Description: OST becomes very slow and/or deadlocked during object unlink
4635 Details : filter_destroy() was holding onto the parent directory lock
4636 while truncating+unlinking objects. For very large objects this
4637 may block other threads for a long time and slow overall OST
4638 responsiveness. It may also be possible to get a lock ordering
4639 deadlock in this case, or run out of journal credits because of
4640 the combined truncate+unlink. Solution is to do object truncate
4641 first in one transaction without parent lock, and then do the
4642 final unlink in a new transaction with the parent lock. This
4643 reduces the lock hold time dramatically.
4646 Frequency : rare, 2.4 kernels only
4648 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4649 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4650 before unmounting their filesystem in order to avoid deadlock.
4651 The kernel_locked() function in 2.4 kernels only checks whether
4652 the kernel lock is held, not whether it is this process that is
4653 holding it as 2.6 kernels do.
4658 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4659 Details : The client ptlrpc code may be trying to reconnect to a down
4660 server before a previous connection attempt has timed out.
4661 Increase the reconnect interval to be longer than the connection
4662 timeout interval to avoid sending duplicate connections to
4666 Frequency : echo_client brw_test command
4668 Description: fix echo_client to work with OST preallocated code
4669 Details : OST preallocation code (5137) didn't take echo_client IO path
4670 into account: echo_client calls filter methods outside of any
4671 OST thread and, hence, there is no per-thread preallocated
4672 pages and buffers to use. Solution: hijack pga pages for IO. As
4673 a byproduct, this avoids unnecessary data copying.
4677 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4678 Description: Client can oops in mdc_commit_close() after open replay
4679 Details : It was possible for the MDS to return an open request with no
4680 transaction number in mds_finish_transno() if the client was
4681 evicted, but without actually returning an error. Clients
4682 would later try to replay that open and may trip an assertion
4683 Simplify the client close codepath, and always return an error
4684 from the MDS in case the open is not successful.
4687 Frequency : rare, 2.6 OSTs only
4689 Description: OST may deadlock under high load on fragmented files
4690 Details : If there was a heavy load and highly-fragmented OST filesystems
4691 it was possible to have all the OST threads deadlock waiting on
4692 allocation of biovecs, because the biovecs were not released
4693 until the entire RPC IO was completed. Instead, release biovecs
4694 as soon as they are complete to ensure forward IO progress.
4696 Severity : enhancement
4698 Description: Support for specifying external journal device at mount
4699 Details : If an OST or MDS device is formatted with an external journal
4700 device, this device major/minor is stored in the ext3 superblock
4701 and may not be valid for failover. Allow detecting and
4702 specifying the external journal at mount time.
4707 Description: Mounting an MDS with pending unlinked files may cause oops
4708 Details : target_finish_recovery() calls mds_postrecov() which returned
4709 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4710 considers this an error and immediately begins cleaning up the
4711 lov, just after starting the mds_lov process
4713 Severity : enhancement
4715 Description: Implement 'lfs df' to report actual free space on per-OST basis
4716 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4717 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4718 usage of objects; '-h' to report in human readable format.
4720 ------------------------------------------------------------------------------
4722 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4729 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4730 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4731 at startup during orphan unlinking. Instead, skip buddy update
4732 before setup, as it will be regenerated after recovery is complete.
4737 Description: If an OST is inactive, its locks might reference stale inodes.
4738 Details : lov_change_cbdata() must iterate over all namespaces, even if
4739 they are inactive to clear inode references from the lock.
4741 Severity : enhancement
4742 Frequency : occasional, if non-standard max_dirty_mb used
4744 Description: Client will block write RPCs if not enough grant
4745 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4746 then the client will block writes while waiting for another RPC
4747 to complete instead of consuming its dirty limit. With change
4748 we get improved performance when max_dirty_mb is small.
4750 Severity : enhancement
4751 Bugzilla : 3389, 6253
4752 Description: Add support for supplementary groups on the MDS.
4753 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4754 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4755 lookups for user supplementary groups into a cache.
4759 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4760 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4761 for liblustre O_EXCL is enforced by MDS.
4765 Description: readdir never returns NULL in liblustre.
4766 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4767 dentry in struct dirent.
4771 Frequency : liblustre only, depends on application IO pattern
4772 Description: liblustre clients evicted if not contacting servers
4773 Details : Don't put liblustre clients into the ping_evictor list, so
4774 they will not be evicted by the pinger ever.
4776 Severity : enhancement
4778 Description: Add ability to evict clients by NID from MDS.
4779 Details : By echoing "nid:$NID" string into
4780 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4781 $NID would be instantly evicted from this MDS and from all active
4782 OSTs connected to it.
4786 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4787 Details : lookup_it_finish() used to query file size from OSTs that was not
4792 Description: service threads change working directory to that of init
4793 Details : Starting lustre service threads may pin the working directory
4794 of the parent thread, making that filesystem busy. Threads
4795 now change to the working directory of init to avoid this.
4799 Frequency : during shutdown only
4800 Description: shutdown with a failed MDS or OST can cause unmount to hang
4801 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4806 Frequency : 2.6 only
4807 Description: chmod/chown may include an extra supplementary group
4808 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4809 supplementary group and if none is specified this is used.
4812 Bugzilla : 5479 (6816)
4814 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4815 Details : If lookup is in progress on a file that is unlinked we might try
4816 to revalidate the inode and fail in revalidate after lookup is
4817 complete and ll_file_open() enqueues the open again but
4818 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4821 Frequency : always, if lconf --abort_recovery used
4823 Description: lconf --abort_recovery fails with 'Operation not supported'
4824 Details : lconf was attempting to abort recovery on the MDT device and not
4827 ------------------------------------------------------------------------------
4829 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4834 Frequency : rare (only unsupported configurations with a node running as an
4836 Bugzilla : 6514, 5137
4837 Description: Mounting a Lustre file system on a node running as an OST could
4839 Details : OSTs now preallocates memory needed to write out data at
4840 startup, instead of when needed, to avoid having to
4841 allocate memory in possibly low memory situations.
4842 Specifically, if the file system is mounted on on OST,
4843 memory pressure could force it to try to write out data,
4844 which it needed to allocate memory to do. Due to the low
4845 memory, it would be unable to do so and the node would
4846 become unresponsive.
4848 Severity : enhancement
4850 Description: Addition of lconf --service command line option
4851 Details : lconf now accepts a '--service <arg>' option, which is
4852 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4854 Severity : enhancement
4856 Description: Failover mode is now the default for OSTs.
4857 Details : By default, OSTs will now run in failover mode. To return to
4858 the old behaviour, add '--failout' to the lmc line for OSTs.
4860 Severity : enhancement
4862 Description: Health checks are now provided for MDS and OSTs
4863 Details : Additional detailed health check information on MSD and OSTs
4864 is now provided through the procfs health_check value.
4867 Frequency : occasional, depends on IO load
4869 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4870 after numerous create/delete cycles
4871 Details : The ext3 inode allocation policy would not allocate new inodes
4872 very well on the OSTs because there are no new directories
4873 being created. Instead we look for groups with free space if
4874 the parent directories are nearly full.
4879 Description: Network or server problems during mount may cause partially
4880 mounted clients instead of returning an error.
4881 Details : The config llog parsing code may overwrite the error return
4882 code during mount error handling, returning success instead
4888 Description: MDS can fail to allocate large reply buffers
4889 Details : After long uptimes the MDS can fail to allocate large reply
4890 buffers (e.g. zconf client mount config records) due to memory
4891 fragmentation or consumption by the buffer cache. Preallocate
4892 some large reply buffers so that these replies can be sent even
4893 under memory pressure.
4897 Frequency : rare (liblustre)
4898 Description: fsx running with liblustre complained that using truncate() to
4899 extend the file doesn't work. This patch corrects that issue.
4900 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4901 ATTR_SIZE and lsm use in llu_setattr_raw.
4905 Frequency : rare, only 2.6 kernels
4906 Description: Unusual file access patterns on the MDS may result in inode
4907 data being lost in very rare circumstances.
4908 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4909 code in the 2.6 kernel caused the inode and/or EA data not to
4910 be read from disk, causing single-file corruption.
4914 Frequency : rare, only 2.6 filesystems using extents
4915 Description: Heavy concurrent write and delete load may cause data corruption.
4916 Details : It was possible under high-load situations to have an extent
4917 metadata block in the block device cache from a just-unlinked
4918 file overwrite a newly-allocated data block. We now unmap any
4919 metadata buffers that alias just-allocated data blocks.
4923 Frequency : filesystems with default stripe_count larger than 77
4924 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4925 Details : lconf specifies an inode size of 4096 bytes when the default
4926 stripe_count is larger than 77. This conflicts with the default
4927 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4928 this case to avoid pinning too much memory for large EAs.
4930 ------------------------------------------------------------------------------
4932 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4937 Frequency : rare (extremely heavy IO load with hundreds of clients)
4939 Description: Client is evicted, gets IO error writing to file
4940 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4941 caused clients to be evicted for AST timeouts. The fixes in
4942 bug 5192 mean we no longer need to have such short AST timeouts
4943 so ldlm_timeout has been increased.
4946 Frequency : occasional during --force or --failover shutdown under load
4947 Bugzilla : 5949, 4834
4948 Description: Server oops/LBUG if stopped with --force or --failover under load
4949 Details : a collection of import/export refcount and cleanup ordering
4950 issues fixed for safer force cleanup
4953 Frequency : only filesystems larger than 120 OSTs
4954 Bugzilla : 5990, 6223
4955 Description: lfs getstripe would oops on a very large filesystem
4956 Details : lov_getconfig used kfree on vmalloc'd memory
4959 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4960 Bugzilla : 6242, 6243
4961 Description: reading from files that had been truncated to a non-zero size
4962 but never opened returned no data
4963 Details : ll_file_read() reads zeros from no-object files to EOF
4968 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4969 Details : There's little harm in aborting MDS/OSS recovery and letting it
4970 try again, so I removed the LASSERT and return an error instead.
4972 Severity : enhancement
4974 Description: New debugging infrastructure for tracking down data corruption
4975 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4976 (b) cover more of the client-side code path, and (c) try to narrow
4977 down where problems occurred
4981 Bugzilla : 3819, 4364, 4397, 6313
4982 Description: Racing close and eviction MDS could cause assertion in mds_close
4983 Details : It was possible to get multiple mfd references during close and
4984 client eviction, leading to one thread referencing a freed mfd.
4986 Severity: : enhancement
4987 Bugzilla : 3262, 6359
4988 Description: Attempts to reconnect to servers are now more aggressive.
4989 Details : This builds on the enhanced upcall-less recovery that was added
4990 in 1.4.2. When trying to reconnect to servers, clients will
4991 now try each server in the failover group every 10 seconds. By
4992 default, clients would previously try one server every 25 seconds.
4997 Description: After recovery, certain operations trigger a failed
4998 assertion on a client.
4999 Details : Failing over an mds, using lconf -d --failover, while a
5000 client was doing a readdir() call would cause the client to
5001 LBUG after recovery completed and the readdir() was resent.
5003 Severity : enhancement
5005 Description: Default groups are now added by lconf
5006 Details : You can now run lconf --group <servicename> without having to
5007 manually add groups with lmc.
5010 Frequency : occasional
5012 Description: Nodes with an elan id of 0 trigger a failed assertion
5015 Frequency : always when accessing e.g. tty/console device nodes
5017 Description: tty and some other devices nodes cannot be used on lustre
5018 Details : file's private_data field is used by device data and lustre
5019 values in there got lost. New field was added to struct file to
5020 store fs-specific private data.
5023 Frequency : when exporting Lustre via NFS
5025 Description: NFSD failed occasionally when looking up a path component
5026 Details : NFSD is looking up ".." which was broken in ext3 directories
5027 that had grown large enough to become hashed.
5030 Frequency : Clusters with multiple interfaces not on the same subnet
5032 Description: Nodes will repeatedly try to reconnect to an interface which it
5033 cannot reach and report an error to the log.
5034 Details : Extra peer list entries will be created by lconf with some peers
5035 unreachable. lconf now validates the peer before adding it.
5038 Frequency : Only if a default stripe is set on the filesystem root.
5040 Description: Setting a default stripe on the filesystem root prevented the
5041 filesystem from being remounted.
5042 Details : The client was sending extra request flags in the root getattr
5043 request and did not allocate a reply buffer for the dir EA.
5046 Frequency : occasional, higher if lots of files are accessed by one client
5047 Bugzilla : 6159, 6097
5048 Description: Client trips assertion regarding lsm mismatch/magic
5049 Details : While revalidating inodes the VFS looks up inodes with ifind()
5050 and in rare cases can find an inode that is being freed.
5051 The ll_test_inode() code will free the lsm during ifind()
5052 when it finds an existing inode and then the VFS later attaches
5053 this free lsm to a new inode.
5057 Bugzilla : 6422, 7030
5058 Description: MDS deadlock between mkdir and client eviction
5059 Details : Creating a new file via mkdir or mknod (starting a transaction
5060 and getting the ns lock) can deadlock with client eviction
5061 (gets ns lock and trying to finish a synchronous transaction).
5064 Frequency : occasional
5065 Description: While starting a server, the fsfilt_ext3 module could not be
5067 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5068 kernels. Previously, lconf would still use the ext3 name
5069 when trying to load modules. Now, it will correctly use
5070 ext3 on 2.4 and ldiskfs on 2.6.
5072 Severity : enhancement
5073 Description: The default stripe count has been changed to 1
5074 Details : The interpretation of the default stripe count (0, to lfs
5075 or lmc) has been changed to mean striping across a single
5076 OST, rather than all available. For general usage we have
5077 found a stripe count of 1 or 2 works best.
5079 Severity : enhancement
5080 Description: Add support for compiling against Cray portals.
5081 Details : Conditional compiling for some areas that are different
5085 Frequency : occasional
5086 Bugzilla : 6409, 6834
5087 Description: Creating files with an explicit stripe count may lead to
5088 a failed assertion on the MDS
5089 Details : If some OSTs are full or unavailable, creating files may
5090 trigger a failed assertion on the MDS. Now, Lustre will
5091 try to use other servers or return an error to the
5095 Frequency : occasional
5097 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5098 caused lock timeout during readahead (since 1.4.2).
5099 Details : Processes doing readahead might match a lock that hasn't been
5100 granted yet if there are overlapping and conflicting lock
5101 requests. The readahead process waits on ungranted lock
5102 (original lock is CBPENDING), while OST waits for that process
5103 to cancel CBPENDING read lock and eventually evicts client.
5105 Severity : enhancement
5107 Description: Initial enabling of flock support for clients
5108 Details : Implements fcntl advisory locking and file status functions.
5109 This feature is provided as an optional mount flag (default
5110 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5111 locking are implemented yet, and those that are are not guaranteed
5112 to be completely correct in production environments.
5113 mount -t lustre -o [flock|noflock] ...
5116 Frequency : occasional
5118 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5119 assertion in the ext3 JBD (journaling) layer.
5120 Details : The b_committed_data struct is protected by the big kernel lock
5121 in 2.4 kernels, serializing journal_commit_transaction() and
5122 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5123 access to this struct. In 2.6 kernels there is finer grained
5124 locking to improve SMP performance of the JBD layer.
5128 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5133 Description: merge and cleanup kernel patches.
5136 -----------------------------------------------------------------------------
5138 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5140 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5141 versions of Lustre. Please update all servers and clients to
5142 version 1.4.2 or later at the same time. You must also run
5143 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5144 to update the configuration logs.
5146 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5147 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5148 - swab llog records of type '0' so we get proper header size/idx (5861)
5149 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5150 - fix rename of one directory over another leaking an inode (5953)
5151 - avoid SetPageDirty on 2.6 (5981)
5152 - don't re-add just-being-destroyed locks to the waiting list (5653)
5153 - when creating new directories, inherit the parent's custom
5154 striping settings if present parent (3048)
5155 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5156 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5157 - don't hold client locks on temporary worklist from l_lru (5666)
5158 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5159 - automatically evict dead clients (5921)
5160 - Update file size properly in create+truncate+fstat case (6196)
5161 - Do not unhash mountpoint dentries, do not allow removal of
5163 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5164 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5165 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5166 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5167 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5168 - Correctly handle reads of files with no objects (6243)
5169 - lctl recover will also mark a device active if deactivate used (5933)
5171 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5172 - allow --write-conf on an MDS with different nettype than client (5619)
5173 - don't write config llogs to MDS for mounts not from that MDS (5617)
5174 - lconf should create multiple TCP connections from a client (5201)
5175 - init scripts are now turned off by default; run chkconfig --on
5176 lustre and chkconfig --on lustrefs to use them
5177 - upcalls are no longer needed for clients to recover to failover
5179 - add --abort-recovery option to lconf to abort recovery on device
5181 - add support for an arbitrary number of OSTs (3026)
5182 - Quota support protocol changes.
5183 - forward compatibility changes to wire structs (6007)
5184 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5185 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5186 - improved "lustre status" script
5187 - initialize blocksize for non-regular files (6062)
5188 - added --disable-server and --disable-client configure options (5782)
5189 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5190 - Vanilla 2.4.29 support
5191 - increase maximum number of obd devices to 520 (6242)
5192 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5193 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5194 SLES 9 2.6.5 kernels
5195 - limit stripes per file to 160 (the maximum EA size) (6093)
5197 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5200 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5201 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5202 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5203 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5204 - small ext3 extents cleanups and fixes (5733)
5205 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5206 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5207 - avoid needless client->OST connect, fix handle mismatch (5317)
5208 - fix DLM error path that led to out-of-sync client, long delays (5779)
5209 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5210 - fix several locking issues related to i_size (5492,5624,5654,5672)
5211 - don't move pending lock onto export if it is already evicted (5683)
5212 - fix kernel oops when creating .foo in unlinked directory (5548)
5213 - fix deadlock in obdfilter statistics vs. object create (5811)
5214 - use time_{before,after} to avoid timer jiffies wrap (5882)
5215 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5216 - Do not leak request if server was not able to process it (5154)
5217 - If mds_open unable to find parent dir, make that negative lookup(5154)
5218 - don't create new directories with extent-mapping (5909, 5936)
5220 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5221 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5222 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5223 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5224 - support for RHEL4 kernel on i686 (5773)
5225 - provide error messages when incompatible logs are encountered (5898)
5227 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5228 * version 1.4.0.10 (1.4.1 release candidate 1)
5230 - don't keep a lock reference when lock is not granted (4238)
5231 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5232 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5233 - drop import inflight refcount on signal_completed_replay error (5255)
5234 - unlock page after async write error during send (3677)
5235 - handle missing objects in filter_preprw_read properly (5265)
5236 - no transno return for symlink open, don't save no-trasno open (3440)
5237 - don't try to complete elan receive that already failed (4012)
5238 - free RPC server reply state on error (5406)
5239 - clean up thread from ptlrpc_start_thread() on error (5160)
5240 - readahead could read extra page into cache that wasn't ejected (5388)
5241 - prevent races in class_attach/setup/cleanup/detach (5260)
5242 - don't dereference de->d_inode after l_dput of de (5458)
5243 - use "int" for stripe value returned from lock_to_stripe (5544)
5244 - mballoc allocation and error-checking fixes in 2.6 (5504)
5245 - block device patches to fix I/O request sizes in 2.6 (5482)
5246 - look up hostnames for IB nals (5602)
5247 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5248 - don't start multiple acceptors for the same port (5277)
5249 - fix incorrect LASSERT in mds_getattr_name (5635)
5250 - export a proc file for general "ping" checking (5628)
5251 - fix "lfs check" to not block when the MDS is down (5628)
5253 - service request history (4965)
5254 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5255 - create an "evict_client" /proc entry on OSTs, like the MDS has
5256 - fix mount usage message, return errors per mount(8) (5168)
5257 - change grep [] to grep "[]" in tests so they work in more UMLs
5258 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5259 - remove ancient LOV_MAGIC_V0 EA support (5047)
5260 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5261 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5262 - allow readahead allocations to fail when low on memory (5383)
5263 - mmap locking landed again, after considerable improvement (2828)
5264 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5266 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5269 - send OST transaction number in read/write reply to free req (4966)
5270 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5271 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5272 - the watchdog thread now runs as interruptible (5246)
5273 - flock/lockf fixes (but it's still disabled, pending 5135)
5274 - don't use EXT3 constants in llite code (5094)
5275 - memory shortage at startup could cause assertion (5176)
5277 - reorganization of lov code
5278 - single portals codebase
5280 - add extents/mballoc support (5025)
5281 - direct I/O reads in the obdfilter (4048)
5282 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5284 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5287 - send OST transaction number in read/write reply to free req (4966)
5288 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5289 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5290 - the watchdog thread now runs as interruptible (5246)
5291 - handle missing objects in filter_preprw_read properly (5265)
5292 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5293 - drop import inflight refcount on signal_completed_replay error (5255)
5294 - unlock page after async write error during send (3677)
5295 - return original error code on reconstructed replies (3761)
5296 - no transno return for symlink open, don't save no-trasno open (3440)
5298 - add pid to ldlm debugging output (4922)
5299 - bump the watchdog timeouts -- we can't handle 30sec yet
5300 - extra debugging for orphan dentry/inode bug (5259)
5302 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5305 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5306 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5307 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5308 - fix NULL dereference in /proc/sys/portals/routes (4827)
5309 - allow failed mdc_close() operations to be interrupted (4561)
5310 - stop precreate on OST before MDS would time out on it (4778)
5311 - don't send partial-page writes before EOF from client (4410)
5312 - discard client grant for sub-page writes on large-page clients (4520)
5313 - don't free dentries not owned by NFS code, check generation (4806)
5314 - fix lsm leak if mds_create_objects() fails (4801)
5315 - limit debug_daemon file size, always print CERROR messages (4789)
5316 - use transno after validating reply (3892)
5317 - process timed out requests if import state changes (3754)
5318 - update mtime on OST during writes, return in glimpse (4829)
5319 - add mkfsoptions to LDAP (4679)
5320 - use ->max_readahead method instead of zapping global ra (5039)
5321 - don't interrupt __l_wait_event() during strace
5323 - add software watchdogs to catch hung threads quickly (4941)
5324 - make lustrefs init script start after nfs is mounted
5325 - fix CWARN/ERROR duplication (4930)
5326 - return async write errors to application if possible (2248)
5327 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5328 - print NAL number in %x format (4645)
5329 - update barely-supported suse-2.4.21-171 series (4842)
5330 - support for sles 9 %post scripts
5331 - support for building 2.6 kernel-source packages
5332 - support for sles km_* packages
5334 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5337 - ignore -ENOENT errors in osc_destroy (3639)
5338 - notify osc create thread that OSC is being cleaned up (4600)
5339 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5340 - reconstruct ost_handle() like mds_handle() (4657)
5341 - create a new thread to do import eviction to avoid deadlock (3969)
5342 - let lconf resolve symlinked-to devices (4629)
5343 - don't unlink "objects" from directory with default EA (4554)
5344 - hold socknal file ref over connect in case target is down (4394)
5345 - allow more than 32000 subdirectories in a single directory (3244)
5346 - fix blocks count for O_DIRECT writes (3751)
5347 - OST returns ENOSPC from object create when no space left (4539)
5348 - don't send truncate RPC if file size isn't changing (4410)
5349 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5350 - bind to privileged port in socknal and tcpnal (3689)
5352 - rate limit CERROR/CWARN console message to avoid overload (4519)
5353 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5354 - basic mmap support (3918)
5355 - kernel patch series update from b1_4 (4711)
5357 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5360 - avoid crash during MDS cleanup with OST shut down (2775)
5361 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5362 - don't use bad inodes on the MDS (3744)
5363 - dynamic object preallocation to improve recovery speed (4236)
5364 - don't hold spinlock over lock dumping or change debug flags (4401)
5365 - don't zero obd_dev when it is force cleaned (3651)
5366 - print grants to console if they go negative (4431)
5367 - "lctl deactivate" will stop automatic recovery attempts (3406)
5368 - look for existing locks in ldlm_handle_enqueue() (3764)
5369 - don't resolve lock handle twice in recovery avoiding race (4401)
5370 - revalidate should check working dir is a directory (4134)
5372 - don't always mark "slow" obdfilter messages as errors (4418)
5374 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5377 - don't close LustreDB during write_conf until it is done (3860)
5378 - fix typo in lconf for_each_profile (3821)
5379 - allow dumping logs from multiple threads at one time (3820)
5380 - don't allow multiple threads in OSC recovery (3812)
5381 - fix debug_size parameters (3864)
5382 - fix mds_postrecov to initialize import for llog ctxt (3121)
5383 - replace config semaphore with spinlock (3306)
5384 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5385 - don't allow enqueue to complete on a destroyed export (3822)
5386 - down write_lock before checking llog header bitmap (3825)
5387 - recover from lock replay timeout (3764)
5388 - up llog sem before sending rpc (3652)
5389 - reduce ns lock hold times when setting kms (3267)
5390 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5391 - fix NULL deref and obd_dev leak on setup error (3312)
5392 - replace some LBUG about llog ops with error handling (3841)
5393 - don't match INVALID dentries from d_lookup and spin (3784)
5394 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5395 - fix invalid assertion in ptlrpc_set_wait (3880)
5397 - add libwrap support for the TCP acceptor (3996)
5398 - add /proc/sys/portals/routes for non-root route listing (3994)
5399 - allow setting MDS UUID in .xml (2580)
5400 - print the stack of a process that LBUGs (4228)
5402 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5405 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5406 - make sure to unset replay flag from failed open requests (3440)
5407 - if default stripe count is 0, use OST count for inode size (3636)
5408 - update parent mtime/ctime on client for create/unlink (2611)
5409 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5410 - free recovery state on server during a forced cleanup (3571)
5411 - unregister_reply for resent reqs (3063)
5412 - loop back devices mounting and status check on 2.6 (3563)
5413 - fix resource-creation race that can provoke i_size == 0 (3513)
5414 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5415 - more debugging for page-accounting assertion (3746)
5416 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5417 - avoid deadlock after precreation failure (3758)
5418 - fix race and lock order deadlock in orphan handling (3450, 3750)
5419 - add validity checks when grabbing inodes from l_ast_data (3599)
5421 - add /proc/.../recovery_status to obdfilter (3428)
5422 - lightweight CDEBUG infrastructure, debug daemon (3668)
5423 - change default OSC RPC parameters to be better on small clusters
5424 - turn off OST read cache for files smaller than 32MB
5425 - install man pages and include them in rpms (3100)
5426 - add new init script for (un)mounting lustre filesystems (2593)
5427 - run chkconfig in %post for init scripts (3701)
5428 - drop scimac NAL (unmaintained)
5430 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5433 - clean kiobufs before and after use (3485)
5434 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5435 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5436 - change init script to fail more clearly if not run as root (1528)
5437 - allow clients to reconnect during replay (1742)
5438 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5439 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5440 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5441 - fix cond_resched (3554)
5442 - only evict unfinished clients after recovery (3515)
5443 - allow bulk resend, prevent data loss (3570)
5444 - dynamic ptlrpc request buffer allocation (2102)
5445 - don't allow unlinking open directory if it isn't empty (2904)
5446 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5447 - remove extraneous obd dereference causing LASSERT failure (3334)
5448 - don't use get_cycles() when creating temp. files on the mds (3156)
5449 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5450 - handle EEXIST for set-stripe, set proper directory name (3336)
5452 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5453 - fix ksocknal_fmb_callback() error messages (2918)
5455 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5458 - don't copy lvb into (possibly NULL) reply on error (2983)
5459 - don't deref dentry after dput, don't free lvb on error (2922)
5460 - use the kms to determine writeback rpc length (2947)
5461 - increment oti_logcookies when osc is inactive (2948)
5462 - update client's i_blocks count via lvb messages (2543)
5463 - handle intent open/close of special files properly (1557)
5464 - mount MDS with errors=remount-ro, like obdfilter (2009)
5465 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5466 - don't use cancelling-locks' kms values (2947)
5467 - use highest lock extent for kms, not last one (2925)
5468 - don't dereference ERR_PTR() dentry in error handling path (3107)
5469 - fix thread race in portals_debug_dumplog() (3122)
5470 - create lprocfs device entries at setup instead of at attach (1519)
5471 - common AST error handler, don't evict client on completion race (3145)
5472 - zero nameidata in detach_mnt in 2.6 (3118)
5473 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5474 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5475 - initialize RPC timeout timer earlier for 2.6 (3219)
5476 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5477 - print nal/nid for unknown nid (3258)
5478 - additional checks for oscc recovery before doing precreate (3284)
5479 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5480 - don't crash in mdc_close for bad permissions on open (3285)
5481 - zero i_rdev for non-device files (3147)
5482 - clear page->private before handing to FS, better assertion (3119)
5483 - tune the read pipeline (3236)
5484 - fix incorrect decref of invalidated dentry (2350)
5485 - provide read-ahead stats and refine rpc in flight stats (3328)
5486 - don't hold journal transaction open across create RPC (3313)
5487 - update atime on MDS at close time (3265)
5488 - close LDAP connection when recovering to avoid server load (3315)
5489 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5490 - don't leak open file on MDS after open resend (3325)
5491 - serialize filter_precreate and filter_destroy_precreated (3329)
5492 - loop device shouldn't call sync_dev() for nul device (3092)
5493 - clear page cache after eviction (2766)
5494 - resynchronize MDS->OST in background (2824)
5495 - refuse to mount the same filesystem twice on same mountpoint (3394)
5496 - allow llmount to create routes for mounting behind routers (3320)
5497 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5498 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5499 - fix rare problem with rename on htree directories (3417)
5501 - allow default OST striping configuration per directory (1414)
5502 - fix compilation for qswnal for 2.6 kernels (3125)
5503 - increase maximum number of MDS request buffers for large systems
5504 - change liblustreapi to be useful for external progs like lfsck (3098)
5505 - increase local configuration timeout for slow disks (3353)
5506 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5508 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5511 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5512 - don't overwrite extent policy data in reply if lock was blocked (2901)
5513 - drop filter export grants atomically with removal from device (2663)
5514 - del obd_self_export from work_list in class_disconnect_exports (2908)
5515 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5516 - swab reply message in mdc_close, other PPC fixes (2464)
5517 - fix destroying of named logs (2325)
5518 - overwrite old logs when running lconf --write_conf (2264)
5519 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5520 - fix race in target_handle_connect (2898)
5521 - mds_reint_create() should take same inode create lock (2926)
5522 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5523 - don't close files for self_export to avoid uninitialized obd (2936)
5524 - allow MDS with the same name as client node (2939)
5525 - hold dentry reference for closed log files for unlink (2325)
5526 - reserve space for all logs during transactions (2059)
5527 - don't evict page beyond end of stripe extent (2925)
5528 - don't oops on a deleted current working directory (2399)
5529 - handle hard links to targets without a parent properly (2517)
5530 - don't dereference NULL lock when racing during eviction (2867)
5531 - don't grow lock extents when lots of conflicting locks (2919)
5533 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5536 - account for cache space usage on clients to avoid data loss (974)
5537 - lfsck support in lustre kernel code (2349)
5538 - reduce journal credits needed for BRW writes (2370)
5539 - orphan handling to avoid losing space on client/server crashes
5540 - ptlrpcd can be blocked, stopping ALL progress (2477)
5541 - use lock value blocks to assist in proper KMS, faster stat (1021)
5542 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5543 - recovery for initial connections (2355)
5544 - fixes for mds_cleanup_orphans (1934)
5545 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5546 - block all file creations until orphan recovery completes (1901)
5547 - client remove rq_connection from request struct (2423)
5548 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5549 - recovery timer race (2670)
5550 - mdc_close recovey bug (2532)
5551 - ptlrpc cleanup bug (2710)
5552 - mds timeout on local locks (2588)
5553 - namespace lock held during RPCs (2431)
5554 - handle interrupted sync write properly (2503)
5555 - don't try to handle a message that hasn't been replied to (2699)
5556 - client assert failure during cleanup after abort recovery (2701)
5557 - leak mdc device after failed mount (2712)
5558 - ptlrpc_check_set allows timedout requests to complete (2714)
5559 - wait for inflight reqs when ptlrpcd finishes (2710)
5560 - make sure unregistered services are removed from the srv_list
5561 - reset bulk XID's when resending them (caught by 1138 test)
5562 - unregister_bulk after timeout
5563 - fix lconf error (2694)
5564 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5565 - readahead locks pages, leaves pending causing memory pressure (2673)
5566 - increase OST request buffers to 4096 on large machines (2729)
5567 - fix up permission of existing directories in simple_mkdir (2661)
5568 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5569 - don't assign transno to errored transactions (2742)
5570 - don't delete objects on OST if given a bogus objid from MDS (2751)
5571 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5572 - if rq_no_resend, then timeout request after recovery (2432)
5573 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5574 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5575 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5576 - MDS recovery shouldn't delete objects that it has given out (2730)
5577 - if enqueue arrives after completion, don't clobber LVB (2819)
5578 - don't unlock pages twice when trigger_group_io returns error (2814)
5579 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5580 - don't write pages to disk if there was an error (1450)
5581 - don't ping imports that have recovery disabled (2676)
5582 - take buffered bytes into account when balancing socknal conn (2817)
5583 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5584 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5585 - remove little-endian swabbing of llog records (1987)
5586 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5587 - retry reposting request buffers if they fail (1191)
5588 - grow extent at grant time to avoid granting a revoked lock (2809)
5589 - lock revoke doesn't evict page if covered by a second lock (2765)
5590 - disable VM readahead to avoid reading outside lock extents (2805)
5592 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5593 - updated kernel patches for hp-2.4.20 kernel (2681)
5595 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5598 - fix truncated write corruption (2366)
5599 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5601 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5602 - ensure len(uuid) < 37 in lmc (1171)
5603 - fix ia64 OOPS in llog_test (2255)
5604 - zero end of page at obdfilter for partial page writes (2648)
5605 - don't leave stale dentries around after renames (bug 2428)
5606 - fix timeouts when evicting a client with a single lock held (2642)
5607 - set deadline for the initial HELLO message to drain (2634)
5608 - print out dotted-quad IP addresses in the socknal (2302)
5610 - additional debugging for MDS client eviction problem (2443)
5611 - fix mkfsoptions support for osts (2603, 2604)
5613 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5616 - add series for the vanilla 2.6.0 kernel
5617 - add series for the vanilla 2.4.24 kernel
5618 - add series for a cray x86/64 UL kernel drop
5619 - fix xattr patches for the vanilla 2.4.19 series
5621 - generate true UUIDs in lmc (1171)
5622 - have portals stack dumping break in UML (2466)
5623 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5624 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5625 - wake callback waiting threads on client eviction (2460)
5626 - Add --ptldebug and --subsystem to lmc (1719)
5627 - update assertion to allow safe interrupt allocation
5628 - set rq_no_resend for cancel requests (2432)
5629 - recalculate ptlrpcd timeout after resend (2494)
5630 - call vfs_rmdir when removing pending directories (2368)
5631 - fix renaming a file to itself (2429)
5632 - lmc creates a default one-stripe lov (2454)
5633 - expand procfs space to handle large clusters (2326)
5634 - increase UML stack to avoid overflow
5635 - update lconf's list of debug and subsystem masks
5636 - fix lfs find --obd (2510)
5637 - /proc tunable for disabling filter read caching (2591)
5638 - stop rpm packages from altering slapd.conf (2301)
5639 - disable nagle in the socknal under 0conf (2578)
5640 - choose mds inode size based on stripe count (2572)
5641 - fix kernel-source rpm problems (2516)
5643 - add --disable-doc to avoid pdf generation (2421)
5644 - update documentation, tests, type-os, comments
5645 - avoid format warnings on ia64
5646 - remove the TOE NAL
5647 - tiny code cleanups by removing unused fields
5649 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5652 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5653 - remove the most blatant lies from BUILDING file (2371)
5654 - change default debug level to reasonable production setting
5655 - reduce client side cache size to reduce cache flush time
5656 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5657 - make TCP zerocopy and pinger support enabled by default (2476)
5658 - sync writes completed after process exits caused crashes (2319)
5659 - maintain correct mount count on the MDS (2356)
5660 - backout 1557, because 2316 wasn't really fixed
5661 - better file I/O statistics gathering in /proc
5662 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5663 - another kernel patch to fix zero-copy TCP function export
5664 - don't take duplicate lock when processing re-sent getattr (2420)
5665 - lctl uses obd_self_export instead of creating new conn (2353)
5666 - MDS/OST recovery case which requires object creation asserted (2425)
5667 - move lfs from /usr/sbin to /usr/bin in packages
5668 - fix race between mds_client_add and mds_client_free (2417)
5669 - use kmalloc instead of slabs in portals (2430)
5670 - don't create duplicate records when a failover MDS is present (2442)
5671 - remove unnecessary mount age check (2332)
5672 - don't remove directory inodes from locks prematurely (2451)
5673 - don't break if MDS service name is the same as hostname (2103)
5674 - fix races in client write RPC generation when cache full (2482)
5676 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5679 - remove now-unused request->rq_obd (278)
5680 - if an allocation fails, print out how much memory we've used (1933)
5681 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5682 - add the "configurable stack size" patch to most series files (1256)
5683 - ability to write large log records, for 100+ OST configs (2306)
5684 - fix NULL deref when filter_prep fails (2314)
5685 - fix operator precedence error in filter_sync
5686 - dynamic allocation of socknal TX descriptors (2315)
5687 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5688 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5689 - ensure that CWARN messages in Portals always get to the syslog
5690 - __init/__exit are not for prototype decls (ldlm_init/exit)
5691 - x86-64 compile warning fixes
5692 - fix gateway LMC keyword conflict (2318)
5693 - fix MDS lock inversions in getattr/reint paths (1844)
5694 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5695 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5696 - DLM race condition prevented some lock evictions (2328)
5697 - ENOMEM detection and retry on socknal sends (2230)
5698 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5699 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5700 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5701 - better error messages when a client is rejected during recovery (1505)
5702 - avoid cancelling locks which were never granted, after failure (2330)
5703 - fix i_sem/journal inversion in mds_client_add (2333)
5704 - fix truncate/getattr lock cycle deadlock (2334)
5705 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5706 - fix two rare exit paths which could leak an l_lock() ref (2321)
5707 - fencepost error in MDS/OST orphan recovery (2226)
5708 - make log record alignment 8 bytes (1988)
5709 - lstripe now fails when requested offset > ost_count (2237)
5710 - ensure that all kernel series have a complete list.h (1607)
5711 - fix crashes in special-file operations (2316)
5712 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5714 - allow configurable automake binary, for testing new versions
5715 - small update to the lfs documentation
5717 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5719 * fix negative export reference count in fsfilt_sync (2312)
5721 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5722 * release candidate 0.9.1
5724 - orphans are moved into the PENDING directory for possible recovery
5725 - replayed opens now open by fid for orphan/rename safety (1042)
5726 - last close of an orphan inode generates a transno (683)
5727 - chdir() and mount() now pin the directory entry (1020)
5728 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5729 - discard very old requests without processing them (1502)
5730 - remove some common, well-understood CERRORs (1505)
5731 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5732 - clear "grant" flags in OST replies until OST grant code lands (1644)
5733 - fix read performance by not clobbering i_blksize on client (1598)
5734 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5735 - make LRU size a /proc tunable, clears locks when reduced (707)
5736 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5737 - server locks take references on exports now (1558)
5738 - build fixes for 2.4.20-rh trees (1663)
5739 - return an error from lov_create if all OSCs are inactive (1751)
5740 - fix import levels when a reconnect happens without a timeout (1597)
5741 - exit early from mds_open if we get a lookup error (1749)
5742 - partial page read at EOF wouldn't wait for disk before sending (1642)
5743 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5744 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5745 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5746 - write/truncate lock inversion (1639)
5747 - Don't auto-load obdclass, portals modules during cleanup (1495)
5748 - fix timestamps from jumping to "now" (1763)
5749 - extra journal assertions (1648)
5750 - add an extra multiunlink test (1771)
5751 - fix read_record/write_record API (1776)
5752 - fix leak of offset_extent, possible incorrect i_size later (1772)
5753 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5754 - Debugging for the kqswnal_get_idle_tx problems (1820)
5755 - Allow recovery to be attempted multiple times (1536)
5756 - Write out MDS last_rcvd file after it is first created (1600)
5757 - Fix tx_descriptor leak in failed transmit situations (1827)
5758 - ext3 journaling fixes for assertion failure after IO error (1871)
5759 - class_export_put() on freed export after completion AST error (1896)
5760 - Fix revalidate looping in VFS (1322)
5761 - Don't access a freed export during MDS_REINT timeout (1521)
5762 - Add open-unlink recovery support on the MDS (1673,1764)
5763 - Return an error if no MDS data was read from last_rcvd (1946)
5764 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5765 - Don't setup a disk device that doesn't match exported UUID (317)
5766 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5767 - avoid committing NULL handle in force close
5768 - local.sh is now a one-stripe LOV configuration
5769 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5770 - fix ext3 htree duplicate directory entry corruption (1516)
5771 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5772 - update to new LOV EA format (2097)
5773 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5775 2003-06-15 Phil Schwan <phil@clusterfs.com>
5778 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5779 - per-import/export recovery handling (958, 931, 959)
5780 - multiple last-rcvd slots, for serving multiple FSes (949)
5781 - connections are again shared between multiple imp/exports (963, 964)
5782 - "umount -f" would hang if any requests needed to be sent (393, 978)
5783 - avoid pinning large req buffer by copying for queued messages (989)
5784 - add "uuid" to "lctl device" command to help upcalls (991)
5785 - "open" RPCs with transnos would confuse recovery counters (1037)
5786 - do proper endian conversion of all wire messages (288, 340, 891)
5787 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5788 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5789 - fix ptlrpc_connection leak in target_handle_connect (1174)
5790 - fix import refcounting bug in OST and MDS cleanup (1134)
5791 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5792 - fix very unlikely obd_types race condition (501)
5793 - remove osc_open hack for echo_client (1187)
5794 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5795 - a failure in read_inode2 leads to deadlock (1139)
5796 - cancel ack-locks as soon as transaction is committed (1072)
5797 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5798 - make sure to commitrw after any preprw to avoid deadlock (1162)
5799 - failing to execute a file in a lustre FS would lock inode (1203)
5800 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5801 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5802 - obd_uuid2tgt would walk off the end of the list (1255)
5803 - on IA64 the osc would give portals incorrect bulk size (1258)
5804 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5805 - fix lock inversion caused by new llite matching code (1282)
5806 - limit the number of dirty pages on a client to 10MB (1286)
5807 - timed out locks were not being corrected cancelled (1289)
5808 - fix O_DIRECT above 4GB on IA-32 (1292)
5809 * major user-visible changes
5810 - fail out/fail over policy now controlled by the upcall (993)
5812 - add OBD_PING to check server availability and failure (954)
5813 - lustre messages are now sent in sending host order (288, 340, 891)
5814 - add eadatalen to MDS getattr reply (340)
5815 - OST read replies may contain second buffer, with per-page status (593)
5817 2003-03-11 Phil Schwan <phil@clusterfs.com>
5820 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5821 - failed open()s could cause deadlock; fixed (867, 869)
5822 - stop cancelling OST locks when files are closed (481)
5823 - overlapping XID spaces caused network corruption (851, 853)
5824 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5825 - setattr_raw added, to do single-RPC, server-side setattrs
5826 - lmc/lconf syntax change for OST UUIDs
5827 - fix crashy race condition between ptlrpc_free_req and osc_close
5828 - don't use request in mdc_enqueue if we hit a timeout (889)
5829 - don't set the inode i_size for regular files from the MDS (896)
5830 - handle out of order completion AST (842)
5831 - don't LBUG if a lock request times out after receiving AST (913)
5832 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5833 - if a bad lock AST arrives, send an error instead of dropping entirely
5834 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5835 - fix leak in bulk IO when only partially completed (899, 900, 926)
5836 - fix O_DIRECT for ia64 (55)
5837 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5838 - C-z after timeout could hang a process forever; fixed (977)
5840 - client-side I/O cache (678, 924, 929, 941, 970)
5842 - READPAGE and SETATTRs which don't take server-side locks get
5845 2003-02-11 Phil Schwan <phil@clusterfs.com>
5848 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5849 - Fix fsfilt_extN_readpage() to read a full page of directory
5850 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5851 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5852 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5853 - Most intent operations were removed, in favour of a new RPC mode
5854 that does a single RPC to the server and bypasses most of the VFS
5855 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5856 - Aggressively cancel local locks on DLM servers
5857 - mds_reint_unlink sends EA to the client if it's the last nlink.
5858 client uses that EA to unlink OST objects.
5859 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5860 - recursive symlinks were fixed (439)
5861 - fixed NULL deref in DEBUG_REQ
5862 - filter_update_lastobjid no longer calls sync, which annoyed extN
5863 - fixed multi-client small-writes to a single file problem (445)
5864 - fixed mtime updates during file writes (607)
5865 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5866 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5867 - fixed Linux OST size reporting problem (444, 656)
5868 - OST now updates object mtime with writes or setattr (607, 619)
5869 - client verifies file size before zeroing page past EOF (445)
5870 - OST now writes last allocated objid to disk with allocation (108)
5871 - LOV on echo now works (409)
5873 - mds_reint_unlink sends a new buffer, with the EA included. this
5874 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5875 set if a regular file was being unlinked, and it was the last link
5876 - use PtlGet from the target for bulk writes (315)
5877 - OST now updates object mtime with writes or setattr (607, 619)
5878 - LDLM now has a grant-time callback to revalidate locked items, if
5880 - Many MDS operations were reorganized to combat race conditions
5882 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5883 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5885 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5888 - Fully reactivate OST imports after reconnection (512, others)
5889 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5890 - More graceful error handling for truncating on dead OST (515)
5891 - Don't error out unless we're actually accessing dead stripes (474)
5892 - Fix garbage sizes when stripes are missing (410)
5893 - LRU counters were broken, causing constant lock purge (433, 432)
5894 - garbage on read from stripes with failed OSTs (441)
5895 - mark OSCs as active before reconnecting during recovery (438)
5896 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5897 - lfind did not preserve OST order in output (443)
5898 - symlinks cause hung clients, incorrect data (439)
5899 - stop dereferencing request after dropping refcount (457)
5900 - don't LASSERT(spin_is_locked) on non-SMP (455)
5901 - fixes for many rename() bugs
5902 - fstat didn't correctly synchronize attributes (399)
5903 - server must handle lock cancellation during blocking AST prep (487)
5904 - bulk descriptors were free()d too soon (511)
5905 - fix paths in lconf, which would load incorrect modules (451, 507)
5906 - fix confusing lconf 'host not found' error message (386)
5907 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5908 - fix race condition in mdc_blocking_ast() for inode access (526)
5909 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5910 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5911 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5912 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5913 - fix lstripe in several ways (kernel side) (527)
5914 - fix request leak in ldlm_cli_enqueue (262)
5915 - incorrect OSC was marked inactive after OST failure
5916 - call mds_fs_cleanup before unmounting filesystem (524)
5917 - fix races between taking ns_lock and ldlm_lock_change_resource
5918 - fix races updating LOV export open file list
5919 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5920 - fix recovery NULL deref in ldlm_cli_cancel_unused
5921 - fix some DLM races by using new hash table for lock handles (419)
5922 - permit the client to specify desired inodes, at replay
5923 - duplicate requests when we queue them for replay reintegration
5924 - fix last_rcvd offset calculation
5925 - sync after each recovered transaction, so we always make progress
5926 - never, not always, ERESTART requests without transnos
5927 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5929 - skip replay if the MDS says that the client is already connected
5930 - don't check for a recovery-enabled export to match lctl's UUID
5931 - don't INC_USE_COUNT for phantom exports
5932 - don't crash when cleaning up phantom exports (567)
5933 - don't double-finish or set replay data for errored mdc_open requests
5934 - abort requests when they time out, so we don't get old replies
5935 - send/receive replies for AST messages again
5936 - if the client says that it doesn't have the lock, cancel it on the
5938 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5939 mark it as destroyed, it will all work out when decref is called
5940 - fix module use counts (22, 581)
5942 - ASTs now expect a reply (server cancels lock on error reply)
5944 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5947 - fix many simultaneous client startup (392)
5948 - fix dentry->d_it clobbering
5949 - credentials weren't being shipped for readdir/getattr operations
5950 - remove invalid assertions triggered during some concurrent MD
5952 - proper Lustre versions added (336, 389)
5953 - fix memory leak for create error case (398)
5954 - fix LOV locking bug that would get cli/srv out of sync
5955 - fix echo client over LOV (409)
5956 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5957 - fix double-O_EXCL intent crash (424)
5958 - avoid sending multiple lock CANCELs (352)
5960 - MDS can do multi-client recovery (modulo bugs in new code)
5962 - many updates, edits, cleanups
5964 2002-11-18 Phil Schwan <phil@clusterfs.com>
5967 - fix null d_it dereference (346)
5968 - fix full OST/dbench hang (333)
5969 - fix permission problem with file removal (286)
5970 - fix removal of OSCs from LOV when they fail
5971 - fix NULL deref during bulk timeout (214)
5972 - fix problems related to multiple filesystems on one MDS (241)
5973 - fixed serious subtle metadata locking bugs
5974 - free locks on clients when inodes are removed due to memory
5976 - fix inode pointer in lock data (285)
5977 - partial support for multiple MDS on a single host (241)
5978 - data locks weren't cancelled at clear_inode time (290, 311)
5979 - intent locks could lead to unbounded lock growth (205)
5980 - added a maximum lock count, an LRU list, and a flusher
5981 - fix multiple rename (365)
5982 - properly abstracted the echo client
5983 - OSC locked 1 byte too many; fixed
5984 - rewrote brw callback code:
5985 - fixed recovery bugs related to LOVs (306)
5986 - fixed too-many-pages-in-one-write crash (191)
5987 - fixed (again) crash in sync_io_timeout (214)
5988 - probably fixed callback-related race (385)
5990 - Add capability to MDS protocol
5991 - LDLM cancellations and callbacks on different portals
5993 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5996 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5997 - defer open object destruction to close time (601981, 138)
5998 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5999 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
6000 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
6001 - don't call obd_setattr in ll_file_release for destroyed objects
6002 * protocol change to lustre_msg: move |version| and add |flags|
6003 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
6004 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
6005 * added replay of create, unlink, link and rename operations during
6006 MDS failover; recovery should be much more robust now
6007 * remove failed OSCs from LOVs (only lov_create uses this so far)
6008 * the lustre-HOWTO was brought (more) up to date (582544)
6010 2002-10-23 Phil Schwan <phil@clusterfs.com>
6013 - in-use dentries weren't being reused properly (617851)
6014 - prevent multiple LDLM setup (599178)
6015 - fix LOV size calculations for truncate (617853)
6016 - fix client handling of MDS intent errors (POSIX)
6017 - fix permission bug in lovstripe.c test (624321)
6018 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6019 - truncate past end of file could corrupt data
6020 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6021 - a race in recovery could return ETIMEDOUT to apps (623947)
6022 - building outside the source directory was fixed
6023 * the lustre-HOWTO was brought (more) up to date (582544)
6024 * major progress was made on recovery functionality
6026 2002-10-10 Phil Schwan <phil@clusterfs.com>
6029 - recovery deadlock fix
6030 - rm -rf causes LBUG fix (617817)
6031 - file open by multiple tasks fix (618962)
6032 - directory permissions bugs (602707 and 620007)
6033 - journal_stop fixed with locking (611313)
6034 - O_APPEND failures resolved (618273, perhaps 614459)
6035 - lconf PATH fix (619770)
6036 - IA64 build fix (621450)
6037 - RPC buffer sizes scale with amount of memory
6039 2002-10-01 Phil Schwan <phil@clusterfs.com>
6042 - locks would be cancelled without throwing away data pages,
6043 resulting in inconsistent data (605627)
6044 - inode attributes were not always being refreshed (605627, 612449)
6045 - lconf now continues to cleanup after lctl reports an error
6046 - MDS now enforces user permissions (602707)
6047 - lprocfs cleanup fixed, but not yet enabled (614157)
6048 - fixed infinite server hang, should a client not respond to an AST
6049 - avoid going into recovery if user calls readlink() with a buffer
6050 that's too small (613941)
6051 - AST RPCs no longer require replies (614867) -- this may be changed
6052 - don't crash server if client sends an IOV that's too big (611336)
6053 - fixed lock conversion deadlock (611892)
6054 - fixed the following of symlinks (614622)
6055 * recovery: the server can remove locks from a client that dies, other
6056 clients can make progress
6057 * more extN patch fixes
6058 * compile-time configurable ptlrpc buffer allocations
6060 - collaborative read cache document
6061 - Lustre Lite Performance CDR document-in-progress
6063 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6066 - fix typo in patch-2.4.18
6068 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6071 - clear ptlrpc request each time in handle_incoming_request()
6072 - unlink of files now destroys the object on the OST
6074 2002-09-19 Peter Braam <braam@clusterfs.com>
6076 * add hard link support
6077 * change obdfile creation method
6078 * kernel patch changed
6080 2002-09-19 Peter Braam <braam@clusterfs.com>
6083 - stack overflow bug in extN fixed
6085 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6087 * documentation updates
6088 - add man pages for config tools
6089 - update tests/README to describe testing with new config tools
6090 - finish metadata API descriptions
6091 * bug fixes and cleanups
6092 - statfs workaround for 16TB limit
6093 - LOV stripe allocation improved, can stripe on subset of OSTs
6094 - LOV file size/IO offset was wrong for files > 4GB in size
6095 - object EA data was being dropped, caused files to be unreadable
6096 - memory overflow with non-LOV OST caused memory corruption
6097 - fixed regression tests to work with new config tools, obdfilter
6098 - fixed bug when directory size became larger than 1 block
6099 - fixed bug (for single client case) when PWD was deleted
6100 - invalidate local directory pages when doing intent-based ops
6101 - avoid LDLM oops when lock callback contained bad data
6103 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6105 * documentation updates
6106 * bug fixes and cleanups
6107 - configuration tools
6110 - 64-bit compile warnings
6111 - 64-bit internal statfs data
6113 * test_brw on persistent OST devices
6115 * lprocfs (disabled)
6117 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6119 * documentation updates
6120 * bug fixes and cleanups
6121 * configuration tools
6123 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6126 * many small fixes to 0_5_4
6127 * io/network handling
6128 * thinkos in MDS operations
6130 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6133 * crucial basic fixes to 0.5.3
6134 * IOR, Iozone work over Elan
6137 2002-08-07 Phil Schwan <phil@clusterfs.com>
6138 * version 0_5_3, our first alpha
6139 * we use the new Portals iovs
6140 * documentation updates
6141 * bug fixes and cleanups
6142 * small changes in the DLM wire protocol
6144 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6145 * version 0_5_1 with some initial stability,
6146 * locking on MD and file I/O.
6147 * documentation updates
6148 * several bug fixes since 0.5.0
6149 * small changes in wire protocol
6151 2002-07-18 Phil Schwan <phil@clusterfs.com>
6153 * delivered as Lustre Light Alpha
6154 * fixed a crash after handling invalid MDS requests
6155 * fixed directory pages for architectures with non-4k pages sizes
6157 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6158 * release version v0_4_4
6159 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6161 * Fixes a number of interruption problems with OST operations.
6162 * Update documentation for portals header changes
6163 * Move all wire protocol structs/defines to lustre_idl.h
6164 * Fixes symlink length bug.
6165 * Add tcpdump to repository.
6167 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6168 * release version v0_4_3
6169 * Fixes statfs for inodes on extN.
6170 * Fixes bug in runtests which would delete /etc/hosts.
6171 * Use 64-bit object IDs wherever possible (not into VFS though)
6172 Remove ost_get_info, which is unused by lustre, and out of date.
6174 2002-07-03 Peter Braam <braam@clusterfs.com>
6175 * release version v0_4_2 Fixes a lookup error (type not passed)
6176 * move forward to head of Portals
6177 * move forward to latest Lustre kernel
6179 2002-06-25 Peter Braam <braam@clusterfs.com>
6180 * release version v0_4_1. Hopefully stable on single node use.