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.
17 Description: llobdstat fix and enhancement
18 Details : add a counter to set a limit to how many samples will be returned
19 fix a wildcard in the path to limit to obdfilter stats only
23 Description: lfs check only allowed for root
24 Details : Code cleanups and added a regresion test of non-root use of
29 Description: MDS Oopses in ptlrpc_dispatch_difficult_reply()/__list_add()
30 Details : Fix hr->hr_index overload.
34 Description: Fix FIEMAP functionality.
38 Description: FC11 patchless client support added
42 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
43 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
47 Description: New acc-sm script and various test-framework enhancments.
51 Description: Previously landed patch was reverted.
55 Description: Update SLES10 kernel to 2.6.16.60-0.42.8 (patchless)
59 Description: Add ldiskfs maxdirsize mount option.
63 Description: Fix mkfs.lustre support for 16TB LUNs.
67 Description: Print mmp_check_interval
68 Details : print mmp_check_interval and make it possible to abort
69 mount operation in case it takes too long.
73 Description: Add cascading_rw to lustre mpi tests.
77 Description: Store SOM data in EA.
78 Details : Also several SOM related cleanups.
82 Description: Add refcount to md_open_data
86 Description: Use refcount on lov_request_set to prevent memory corruption.
90 Description: Statahead: use parent dir's i_mutex for synchronization.
94 Description: loadgen improvements.
98 Description: Enable peer health detection in ksocklnd and ko2iblnd.
102 Description: Control DCACHE_LUSTRE_INVALID flag with MDS_INODE_LOOKCUP lock
106 Description: Fix race on obd_devcie:md_stats
110 Description: Fix SLES 11 client builds.
111 Details : Resolved uninitalized pointer gcc warnings.
115 Description: Allow non-root access for "lfs check".
116 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
120 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
122 Severity : enhancement
124 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
125 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
129 Description: Sanity test_27p fails.
130 Details : OBD_FAIL_OST_ENOSPC has been moved to filter_statfs().
131 Sanity tests 27 were cleaned up and improved.
135 Description: Unify req format on client/servers
136 Details : Use new req_capsule API [almost] everywhere instead of old PTLRPC
137 buffers and swabbers approach..
142 Description: drop join file stuff.
144 Severity : enhancement
146 Description: Add 2.6.27 and SLES11 (patchless client) support
150 Description: Fix in ptlrpc_expire_one_request() to print the signed time
153 Severity : enhancement
155 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
158 Severity : enhancement
160 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
162 Severity : enhancement
164 Description: correctly handle big reply message.
165 Details : send LNet event if reply is bigger then buffer and adjust this buffer
170 Description: Drop unnecessary __GFP_NOMEMALLOC flag from filter_get_page()
172 Severity : enhancement
174 Description: Add support for OEL5.
178 Description: Conf-sanity.sh 50g test - deactivated OST should not cause a panic.
182 Description: client could not reconnect to OST because of an active request.
183 Details : The client now retries to reconnect to the same server, if a
184 connect request failed with EBUSY or -EAGAIN. Test case 35b
185 conf-sanity.sh has been created.
189 Description: Don't allow make backward step on assignin osc next id.
190 Details : race between allocation next id and ll_sync thread can be cause
191 of set wrong osc next id and can be kill valid ost objects.
195 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
199 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
203 Description: Error handling in osc_statfs_interpret() has been improved.
204 Details : Check in osc_statfs_interpret() for EBADR.
208 Description: Do not send statfs() requests to OSTs disabled by administrator.
209 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
211 Severity : enhancement
213 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
215 Severity : enhancement
217 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
219 Severity : enhancement
221 Description: Enhancement for lfs(1) command to use numeric uid/gid.
225 Description: Improved cleanup procedure for conf-sanity test_46a.
229 Description: Do not start bulk transfer, if client tried to reconnect.
230 Details : Do not start bulk transfer, if client tried to reconnect,
231 but failed due to active rpcs. Check for exp_abort_active_req
232 has been implemented.
236 Description: Prevent inconsistences between linux and lustre mount structures.
237 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
238 Make the sleep interruptible.
240 Severity : enhancement
242 Description: provide server to client comms path
243 Details : server to client communications path via new LDLM_SET_INFO rpc
247 Description: truncate starts GFP_FS allocation under transaction causing deadlock
248 Details : ldiskfs_truncate calls grab_cache_page which may start page
249 allocation under an open transaction. This may lead to
250 calling prune_icache with consequent lustre reentrance.
254 Description: send correctly lsm on open replay
255 Details : MDS is trust to LSM size on replay open, but client can set wrong size
260 Description: enable client interface failover
261 Details : When a child reconnects from another NID, properly update export
262 nid hash position and ldlm reverse import.
264 Severity : enhancement
266 Description: Communicate OST degraded/readonly state via statfs to MDS
267 Details : Flags in the statfs returned from OSTs indicate whether the
268 OST is in a degraded RAID state, or if the filesystem has
269 turned read-only after a filesystem error is detected.
274 Description: don't panic if EPROTO was hit when reading symlink
275 Details : correctly handling request reference in error cases.
277 Severity : enhancement
279 Description: Add LustreNetLink, a kernel-userspace communcation path.
281 Severity : enhancement
283 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
288 Description: access to llog context before init.
289 Details : move handling CATALOGS file at osc layer and forbid access to llog
293 Frequency : rare, only when client is writing to OST on local node
295 Description: Avoid deadlock for local client writes
296 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
297 memory freeing context. This allows OST threads to set the
298 PF_MEMALLOC flag on task structures in order to allocate memory
299 from reserved pools and complete IO.
300 Use GFP_HIGHUSER for OST allocations for non-local client writes,
301 so that the OST threads generate memory pressure and allow
302 inactive pages to be reclaimed.
304 Severity : enhancement
306 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
311 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
312 Details : move ldlm namespace creation in setup phase to avoid grab
313 _lprocfs_lock with cli_sem held.
317 Description: Temporarily disable grant shrink.
318 Details : Disable the feature for debugging.
321 Frequency : only when formatting test filesystems
323 Description: Unable to run several mkfs.lustre on loop devices at the same time
324 Details : mkfs.lustre returns error 256 on the concurrent loop devices
325 formatting. The solution is to proper handle the error.
327 Severity : enhancement
329 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
331 Severity : enhancement
333 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
338 Description: resolve race between obd_disconnect and class_disconnect_exports
339 Details : if obd_disconnect will be called to already disconnected export he
340 forget release one reference and osc module can't unloaded.
342 Severity : enhancement
344 Description: Allow tuning service thread via /proc
345 Details : For each service a new
346 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
347 created that can be used to set min/max thread counts, and get the
348 current number of running threads.
353 Descriptoin: don't return error if have partially created objects for file.
354 Details : lov_update_create_set uses set->set_success as index for created
355 objects, so if some requests will be failed, they will have hole
356 at end of array and qos_shrink_lsm for allocate correct lsm.
358 Severity : enhancement
360 Description: Update OFED support to 1.4.1
365 Description: Reduce small size read RPC
366 Details : Set read-ahead limit for every file and only do read-ahead when
367 available read-ahead pages are bigger than 1M to avoid small size
370 Severity : enhancement
372 Description: move AT tunable parameters for more consistent usage
373 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
375 Severity : enhancement
377 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
378 Details : allow skip disconnected ost for send statfs request and hide error
384 Description: fix racy locking of mballoc block bitmaps causing BUG
385 Details : The locking of the mballoc buddy bitmap and the in-memory
386 block bitmap was using two different spin locks in some
387 cases. This made it possible to incorrectly access the
388 mballoc bitmap while another process was modifying it,
389 causing a sanity assertion to fail. While no on-disk corruption
390 was reported, there was some risk of this happening.
393 Frequency : rare, on llog test 6
395 Descriptoin: don't allow connect to already connected import
396 Details : allowing connect to already connected import is hide connecting problem.
401 Descriptoin: allow kill process which wait statahead result
402 Details : for some reasons 'ls' can stick in waiting result from statahead,
403 in this case need way for kill this process.
405 Severity : enhancement
407 Description: Add state history info file, enhance import info file
408 Details : Track import connection state changes in a new osc/mdc proc file;
409 add overview-type data to the osc/mdc import proc file.
411 Severity : enhancement
413 Description: MDS create should not wait for statfs RPC while holding DLM lock.
415 Severity : enhancement
417 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
422 Descriptoin: OST grant too much space to client even there are not enough space.
423 Details : Client will shrink its grant cache to OST if there are no write
424 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
425 this grant cache if there are already not enough avaible space
426 (left_space < total_clients * 32M).
429 Frequency : start MDS on uncleanly shutdowned MDS device
431 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
432 Details : stay in waiting mds<>ost recovery finished produce random bugs
433 due race between two ll_sync thread for one lov target. send
434 ACTIVATE event only if connect realy finished and import have
438 Frequency : rare, connect and disconnect target at same time
440 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
441 Details : don't call obd_disconnect under lov_lock. this long time
442 operation and can block ptlrpcd which answer to connect request.
447 Descriptoin: don't lose wakeup for imp_recovery_waitq
448 Details : recover_import_no_retry or invalidate_import and import_close can
449 both sleep on imp_recovery_waitq, but we was send only one wakeup
453 Frequency : always with long access acl
455 Descriptoin: mds can't pack reply with long acl.
456 Details : mds don't control size of acl but they limited by reint/getattr
459 Severity : enhancement
461 Description: Update to SLES10 kernel-2.6.16.60-0.33.
463 Severity : enhancement
465 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
468 Frequency : start MDS on uncleanly shutdowned MDS device
470 Descriptoin: aborting recovery hang on MDS
471 Details : don't throttle destroy RPCs for the MDT.
474 Frequency : on remount
476 Description: external journal device not working after the remount
477 Details : clear dev_rdonly flag for external journal devices in
483 Description: shutdown vs evict race
484 Details : client_disconnect_export vs connect request race.
485 if client will evicted at this time - we start invalidate
486 thread without referece to import and import can be freed
490 Frequency : rare, need acl's on inode.
492 Description: client can't handle ost additional correctly
493 Details : if ost was added after client connected to mds client can have
494 hit lnet_try_match_md ... to big messages to wide striped files.
495 in this case need teach client to handle config events about add
496 lov target and update client max ea size at that event.
498 Severity : enhancement
500 Description: Changelogs
501 Details : Changelogs are a lightweight mechanism to track filesystem
502 metadata and namespace changes. The changelog is recorded
503 permanently on the MDTs, and is periodically "consumed" / purged
504 when records are no longer needed.
506 Severity : enhancement
508 Description: compact fld format with extents
509 Details : Store range of seq rather than every seq in FLD. Seq
510 controller update FLD rather than clients. In Case of CMD, mdt0
511 has FLD, all other metadata server act as non persistent proxy
512 for FLD queries and cache fld entries in fld cache.
517 Description: don't skip ost target if they assigned to file
518 Details : Drop slow OSCs if we can, but not for requested start idx.
519 This means "if OSC is slow and it is not the requested
520 start OST, then it can be skipped, otherwise skip it only
521 if it is inactive/recovering/out-of-space.
525 Description: more cleanup in mds_lov
526 Details : not send LOV EA under replay, we can't know about they size at this
527 time. Don't allow client connect to mds before any ost connected,
528 for avoid problems with LOV EA size and returning EIO to client.
530 Severity : enhancement
532 Description: Interoperability at server side (Disk interoperability)
534 Severity : enhancement
536 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
538 Severity : enhancement
540 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
542 Severity : enhancement
544 Description: New client IO stack (CLIO).
546 Severity : enhancement
548 Description: Commit on sharing. Eliminate inter-client dependencies between
549 uncommitted transactions by doing transaction commits.
550 Thereby clients may recovery independently.
553 Frequency : Create a symlink file with a very long name
555 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
556 Details : If there is no extra space in the request for early cancels,
557 ldlm_req_handles_avail() returns 0 instead of a negative value.
559 Severity : enhancement
561 Description: Add /proc entry for import status
562 Details : The mdc, osc, and mgc import directories now have
563 an import directory that contains useful import data for debugging
566 Severity : enhancement
568 Description: Re-disable certain /proc logging
569 Details : Enable and disable client's offset_stats, extents_stats and
570 extents_stats_per_process stats logging on the fly.
573 Frequency : Only on FC kernels 2.6.22+
575 Description: oops in statahead
576 Details : Do not drop reference count for the dentry from VFS when lookup,
577 VFS will do that by itself.
579 Severity : enhancement
581 Description: Generic /proc file permissions
582 Details : Set /Proc file permissions in a more generic way to enable non-
583 root users operate on some /proc files.
587 Description: Hitting mdc_commit_close() ASSERTION
588 Details : Properly handle request reference release in
589 ll_release_openhandle().
593 Description: quota recovery deadlock during mds failover
594 Details : This patch includes att18982, att18236, att18237 in bz14840.
596 1. fix osts hang when mds does failover with quotaon
597 2. prevent watchdog storm when osts threads wait for the
602 Frequency : only patchless client
603 Description: add workaround for race between add/remove dentry from hash
605 Severity : enhancement
607 Description: Allow OST glimpses to return PW locks
611 Description: LBUG when llog conf file is full
612 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
616 Description: Prevent import from entering FULL state when server in recovery
620 Description: service mount cannot take device name with ":"
621 Details : Only when device name contains ":/" will mount treat it as
627 Description: replace ptlrpcd with the statahead thread to interpret the async
628 statahead RPC callback
632 Frequency : on recovery
633 Description: I/O failures after umount during fail back
634 Details : if client reconnected to restarted server we need join to recovery
635 instead of find server handler is changed and process self
636 eviction with cancel all locks.
638 Severity : enhancement
640 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
642 Severity : enhancement
644 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
646 Severity : enhancement
648 Description: Upcall on Lustre log has been dumped
649 Details : Allow for a user mode script to be called once a Lustre log has
650 been dumped. It passes the filename of the dumped log to the
651 script, the location of the script can be specified via
652 /proc/sys/lnet/debug_log_upcall.
657 Description: avoid idr_remove called for id which is not allocated.
658 Details : Move assigment s_dev for clustered nfs to end of initialization,
659 to avoid problem with error handling.
664 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
665 Details : When connection is reused this not moved from CONN_UNUSED_HASH
666 into CONN_USED_HASH and this prodice warning when put connection
667 again in unused hash.
670 Severity : enhancement
672 Description: File striping can now be set to use an arbitrary pool of OSTs.
674 Severity : enhancement
676 Description: Export bytes_read/bytes_write count on OSC/OST.
680 Description: Early reply size mismatch, MGC loses connection
681 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
682 the connect flags are properly negotiated.
687 Description: quotas are not honored with O_DIRECT
688 Details : all writes with the flag O_DIRECT will use grants which leads to
689 this problem. Now using OBD_BRW_SYNC to guard this.
693 Description: add quota statistics
694 Details : 1. sort out quota proc entries and proc code.
695 2. add quota statistics
697 Severity : enhancement
699 Description: enable quota support for HEAD.
703 Description: Properly propagate oinfo flags from lov to osc for statfs
704 Details : restore missing copy oi_flags to lov requests.
706 Severity : enhancement
708 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
709 : l_getgroups(8), lst(8), routerstat(8)
710 Details : included man pages for llobdstat(8), llstat(8),
711 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
713 Severity : enhancement
715 Description: configure's --enable-quota should check the
716 : kernel .config for CONFIG_QUOTA
717 Details : configure is terminated if --enable-quota is passed but
718 : no quota support is in kernel
722 Description: Remove portals compatibility
723 Details : Remove portals compatibility, not interoperable with releases
728 Description: Resolve device initialization race
729 Details : Prevent proc handler from accessing devices added to the
730 obd_devs array but yet be intialized.
732 Severity : enhancement
734 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
736 Severity : enhancement
738 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
743 Description: Using wrong pointer in osc_brw_prep_request
744 Details : Access to array[-1] can produce panic if kernel compiled with
745 CONFIG_PAGE_ALLOC enabled
749 Description: Client runs out of low memory
750 Details : Consider only lowmem when counting initial number of llap pages
754 Description: *optional* service tags registration
755 Details : if the "service tags" package is installed on a Lustre node
756 When the filesystem is mounted, a local-node service tag will
757 be created. See http://inventory.sun.com/ for more information
758 about the Service Tags asset management system.
762 Description: Kernel BUG tries to release flock
763 Details : Lustre does not destroy flock lock before last reference goes
764 away. So always drop flock locks when client is evicted and
765 perform unlock regardless of successfulness of speaking to MDS.
769 Description: add refcount protection for osc callbacks, avoid panic on shutdown
773 Description: sanity test 65a fails if stripecount of -1 is set
774 Details : handle -1 striping on filesystem in ll_dirstripe_verify
779 Description: ASSERTION(CheckWriteback(page,cmd)) failed
780 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
783 Severity : enhancement
785 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
789 Description: do not process already freed flock
790 Details : flock can possibly be freed by another thread before it reaches
791 to ldlm_flock_completion_ast.
795 Description: LBUG during stress test
796 Details : Need properly lock accesses the flock deadlock detection list.
800 Description: oops in page fault handler
801 Details : kernel page fault handler can return two special 'pages' in error
802 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
806 Description: timeout with invalidate import.
807 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
808 should be handled by ptlrpcd. This produce long age waiting and
809 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
811 Severity : enhancement
813 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
817 Description: enable MGS and MDT services start separately
818 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
819 a co-located MGS without starting the MGS, which is a complement
820 to 'nosvc' mount option.
824 Frequency : after recovery
825 Description: precreate to many object's after del orphan.
826 Details : del orphan st in oscc last_id == next_id and this triger growing
827 count of precreated objects. Set flag LOW to skip increase count
828 of precreated objects.
832 Frequency : rare, on clear nid stats
833 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
834 Details : when clean nid stats sometimes try destroy live entry,
835 and this produce panic in free.
839 Description: Stack overflow during MDS log replay
840 ease stack pressure by using a thread dealing llog_process.
844 Description: wait until IO finished before start new when do lock cancel.
845 Details : VM protocol want old IO finished before start new, in this case
846 need wait until PG_writeback is cleared until check dirty flag and
847 call writepages in lock cancel callback.
849 Severity : enhancement
851 Description: using special macro for print time and cleanup in includes.
855 Description: mds_mfd_close() ASSERTION(rc == 0)
856 Details : In mds_mfd_close(), we need protect inode's writecount change
857 within its orphan write semaphore to prevent possible races.
861 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
865 Frequency : rare, on shutdown ost
866 Description: don't hit live lock with umount ost.
867 Details : shrink_dcache_parent can be in long loop with destroy dentries,
868 use shrink_dcache_sb instead.
872 Description: don't panic with use echo client
873 Details : echo client pass NULL as client nid pointer and this produce null
878 Description: fix build on ppc32
879 Details : compile code with -m64 flag produce wrong object file for ppc32.
883 Description: add message levels for liblustreapi
887 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
888 Details : Move the 'good_osts' check before the 'total_bavail' check. This
889 will result in an -EAGAIN and in the exit call path we call
890 alloc_rr() which will with increasing aggressiveness attempt to
891 aquire precreated objects on the minimum number of required OSCs.
895 Description: Use old size assignment to avoid deadlock
896 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
897 the scheduling while holding a spinlock. We do not need locking
898 for size in ll_update_inode() because size is only updated from
899 the MDS for directories or files without objects, so there is no
900 other place to do the update, and concurrent access to such inodes
901 are protected by the inode lock.
905 Description: resolve "_IOWR redefined" build error on SLES10
909 Description: dump the memory debugging after all modules are unloaded to
910 suppress false negative in conf_sanity test 39
912 Severity : enhancement
914 Description: build kernel-ib packages for OFED 1.3 in our release cycle
919 Description: fix SLES kernel versioning
920 Details : the kernel version for our SLES 10 kernel did not include a "-"
921 before the "smp" at the end. while this was not a problem in
922 general, it did mean that software trying to use the kernel
923 version to try to detect a vendor specific kernel would fail.
924 this was most evident by the OFED build scripts.
928 Description: Don't update lov_desc members until making sure they are valid
929 Details : When updating lov_desc members via proc fs, need fix their
930 validities before doing the real update.
934 Description: don't put request into delay list while invalidate in flight.
935 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
936 import in flight. this produce timeout for invalidate and sometimes
937 can cause stale data.
941 Frequency : on ppc only
942 Description: not convert ost objects for directory because it's not exist.
943 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
947 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
949 Severity : enhancement
951 Description: Update to support 2.6.22.14 vanilla kernel.
955 Frequency : rare, on recovery
956 Description: read procfs can produce deadlock in some situation
957 Details : Holding lprocfs lock which send rpc can produce block for destroy
958 obd objects and this also block reconnect with -EALREADY. This
959 isn't fix all lprocfs bugs - but make it rare.
961 Severity : enhancement
963 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
966 Frequency : frequent on X2 node
968 Description: mdc_set_open_replay_data LBUG
969 Details : Set replay data for requests that are eligible for replay.
973 Description: lustre_mgs: operation 101 on unconnected MGS
974 Details : When MGC is disconnected from MGS long enough, MGS will evict the
975 MGC, and late on MGC cannot successfully connect to MGS and a lot
976 of the error messages complaining that MGS is not connected.
980 Frequency : on network error
981 Description: panic with double free request if network error
982 Details : mdc_finish_enqueue is finish request if any network error ocuring,
983 but it's true only for synchronus enqueue, for async enqueue
984 (via ptlrpcd) this incorrect and ptlrpcd want finish request
987 Severity : enhancement
989 Description: client-side metadata stat-ahead during readdir(directory readahead)
990 Details : perform client-side metadata stat-ahead when the client detects
991 readdir and sequential stat of dir entries therein
994 Frequency : on start mds
996 Description: Implement get_info(last_id) in obdfilter.
999 Frequency : occasional
1001 Description: Correctly check stale fid, not start epoch if ost not support SOM
1002 Details : open with flag O_CREATE need set old fid in op_fid3 because
1003 op_fid2 was overwritten with new generated fid, but mds can answer
1004 with one of these two fids and both is not stale. Setattr
1005 incorrectly started an epoch and assume will be called
1006 done_writing, but without SOM done_writing ever being called.
1009 Frequency : rare, depends on device drivers and load
1011 Description: MDS or OSS nodes crash due to stack overflow
1012 Details : Code changes in 1.8.0 increased the stack usage of some functions.
1013 In some cases, in conjunction with device drivers that use a lot
1014 of stack the MDS (or possibly OSS) service threads could overflow
1015 the stack. One change which was identified to consume additional
1016 stack has been reworked to avoid the extra stack usage.
1019 Frequency : occasional
1021 Description: Do not fail import if osc_interpret_create gets -EAGAIN
1022 Details : If osc_interpret_create got -EAGAIN it immediately exits and
1023 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
1024 oscc_has_objects and see OSC has no objests and call
1025 oscc_internal_create to resend create request.
1027 Severity : enhancement
1029 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1031 Severity : enhancement
1033 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1036 Frequency : very rare
1038 Description: Fix replay if there is an un-replied request and open
1039 Details : In some cases, older replay request will revert the
1040 mcd->mcd_last_xid on MDS which is used to record the client's
1041 latest sent request.
1043 Severity : enhancement
1045 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1047 Severity : enhancement
1049 Description: Add rhel5 support to HEAD.
1051 Serverity : enhancement
1053 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1058 Description: Don't allow skipping OSTs if index has been specified.
1059 Details : Don't allow skipping OSTs if index has been specified, make
1060 locking in internal create lots better.
1064 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1065 Details : access to bitfield in structure is always rounded to long
1066 and this produce problem with not atomic change any bit.
1070 Description: Lustre make rpms failed.
1071 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1072 builds Lustre rpms from tarball.
1075 Frequency : rare on shutdown ost
1077 Description: If llog cancel was not send before clean_exports phase, this can
1078 produce deadlock in llog code.
1079 Details : If llog thread has last reference to obd and call class_import_put
1080 this produce deadlock because llog_cleanup_commit_master wait when
1081 last llog_commit_thread exited, but this never success because was
1082 called from llog_commit_thread.
1086 Description: allow userland application know is lost one of stripes.
1087 Details : fill lvb_blocks with error code on ost and return it to
1088 application if error flag found.
1092 Description: NULL lov_tgts causing MDS oops
1093 Details : more safe checks for NULL lov_tgts for avoid oops.
1095 Severity : enhancement
1097 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1101 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
1102 Details : If an OST is down the MDS will hang indefinitely in
1103 obd_statfs_rqset() waiting for the statfs data. While for
1104 MDS QOS usage of statfs, it should not stuck in waiting.
1106 Severity : enhancement
1108 Description: remote_acl support
1109 Details : Support ACL-based permission check for remote user.
1110 Support setfacl/getfacl for remote user with the utils
1111 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
1112 the system "{s,g}etfacl" utils.
1114 Severity : enhancement
1116 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
1118 Severity : enhancement
1120 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1124 Description: make mgs_setparam() handle fsname containing dash
1125 Details : fsname containing a dash does not work with lctl conf_param
1127 Severity : enhancement
1129 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1131 Severity : enhancement
1133 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1137 Frequency : rare, at shutdown
1138 Description: access already free / zero obd_namespace.
1139 Details : if client_disconnect_export was called without force flag set,
1140 and exist connect request in flight, this can produce access to
1141 NULL pointer (or already free pointer) when connect_interpret
1142 store ocd flags in obd_namespace.
1146 Frequency : only at startup
1147 Description: not alloc memory with spinlock held.
1148 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1149 if any spinlock held.
1151 Severity : enhancement
1153 Description: make lustre randomly fail allocating memory
1154 Details : Make lustre randomly failed allocating memory for testing purpose.
1156 Severity : enhancement
1158 Description: lost problems with lov objid file
1159 Details : Fixes some scability and access to not inited memory problems
1160 in work with lov objdid file.
1165 Description: lfs find does not continue on file error
1166 Details : Continue other files processing when a file/dir is absent.
1170 Description: Inconsistent usage of lustre_pack_reply()
1171 Details : Standardize the usage of lustre_pack_reply() such that it
1172 always generate a CERROR on failure.
1175 Frequency : occasional
1177 Description: MDS hang or stay in waiting lock
1178 Details : If client receive lock with CBPENDING flag ldlm need send lock
1179 cancel as separate rpc, to avoid situation when cancel request
1180 can't processed due all i/o threads stay in wait lock.
1184 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1188 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1192 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1196 Description: Update to RHEL4 latest kernel.
1198 Severity : enhancement
1200 Description: Build SLES10 patchless client fails
1201 Details : The configure was broken by run ./configure with
1202 --with-linux-obj=.... argument for patchless client. When the
1203 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1204 can't search header adequately. Use absolute path such as
1205 -I($LINUX)/include instead.
1209 Description: interrupt oig_wait produce painc on resend.
1210 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1211 private set, and this produce situation when rq_ptlrpcd_data not
1212 copyed to new allocated request and triggered LBUG on assert
1213 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1214 ptlrpcd set if request is changed and can be safety changed to use
1219 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1220 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1221 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1222 these macros and replace them with their difinitions instead.
1226 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1227 Details : Update which_patch & target file for SLES10 latest kernel.
1230 Description: Remove client patches from SLES 10 kernel.
1231 Details : This causes SLES 10 clients to behave as patchless clients
1232 even on a Lustre-patched (server) kernel.
1234 Severity : enhancement
1236 Description: self-adjustable client's lru lists
1237 Details : use adaptive algorithm for managing client cached locks lru
1238 lists according to current server load, other client's work
1239 pattern, memory activities, etc. Both, server and client
1240 side namespaces provide number of proc tunables for controlling
1243 Severity : enhancement
1245 Description: light-weight GSS support
1246 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1247 and reduce performance overhead.
1249 Severity : enhancement
1251 Description: Linux keyring support
1252 Details : Support using service of Linux keyring for Lustre GSS internal
1253 context refresh/cache mechanism.
1257 Description: Fix errors in lfs documentation
1258 Details : Fixes man pages
1262 Description: don't use GFP_* in generic Lustre code.
1263 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1267 Description: obdclass is limited by single OBD_ALLOC(idarray)
1268 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1272 Description: warnings with build patchless client with vanila 2.6.19 and up
1273 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1274 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1278 Frequency : Only for SLES
1279 Description: sanity_quota fail test_1
1280 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1281 file, which makes TSTID[2] inunique.
1285 Frequency : Only for SLES9
1286 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1287 Details : Update target/ChangeLog/which_patch .
1291 Frequency : Only for test_10 in sanity-quota.sh
1292 Description: a bug in quota test code
1293 Details : checking a test flag in a wrong place
1297 Frequency : Only for test_13 in sanity-quota.sh
1298 Description: a bug in quota test script
1299 Details : shouldn't check overfull allocated quota
1303 Description: jbd statistics
1304 Details : Port older jbd statistics patch for sles10
1306 Severity : enhancement
1308 Description: Add printing inode info into message about error in writepage.
1313 Description: reply_lock_interpret crash due to race with it and lock cancel.
1314 Details : Do not replay locks that are being cancelled. Do not reference
1315 locks by their address during replay, just by their handle.
1319 Frequency : When flocks are used.
1320 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1321 Details : Flock locks might destroy just granted lock if it could be merged
1322 with another existing flock, this is done in completion handler,
1323 so teach ldlm_cli_enquque_fini that this is a valid case for
1329 Description: Oops in read and write path when failing to allocate lock.
1330 Details : Check if lock allocation failed and return error back.
1334 Description: lstripe command fails for valid OST index
1335 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1336 instead of lov->desc.ld_active_tgt_count.
1340 Description: sanity.sh failed test 103
1341 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1346 Description: df doesn't work properly if diskfs blocksize != 4K
1347 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1351 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1352 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1354 Severity : enhancement
1356 Description: Adaptive timeouts
1357 Details : RPC timeouts adapt to changing server load and network
1358 conditions to reduce resend attempts and improve recovery time.
1362 Description: llapi_file_create() does not allow some changes
1363 Details : add llapi_file_open() that allows specifying the mode and
1364 open flags, and also returns an open file handle.
1368 Description: merge and cleanup kernel patches.
1369 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1373 Description: Add journal checksum support.(Kernel part)
1374 Details : The journal checksum feature adds two new flags i.e
1375 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1376 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1377 indicates that the commit block contains the checksum for
1378 the blocks described by the descriptor blocks. Now commit
1379 record can be sent to disk without waiting for descriptor
1380 blocks to be written to disk. This behavior is controlled
1381 using JBD2_FEATURE_ASYNC_COMMIT flag.
1385 Description: OSS needs mutliple precreate threads
1386 Details : Add ability to start more than one create thread per OSS.
1390 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1391 Details : execute lfs setstripe on client
1395 Description: mds_obd_create error creating tmp object
1396 Details : When the user sets quota on root, llog will be affected and can't
1397 create files and write files.
1400 Frequency : Always on ia64 patchless client, and possibly others.
1402 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1403 Details : This allows the patchless client to be loaded on architectures
1404 without this export.
1408 Description: RedHat Update kernel for RHEL5
1409 Details : Add the kernel config file for RHEL5.
1413 Description: RedHat Update kernel for RHEL5
1414 Details : Modify the kernel config file more closer RHEL5.
1418 Description: Build failure against Centos5 (RHEL5)
1419 Details : Define PAGE_SIZE when it isn't present.
1423 Description: test_7 on sanity-quota.sh can't be used on separate machines
1424 Details : add facet to handle it
1428 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1429 Details : with NFS, the anon dentry's parent was set to itself in
1430 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1431 corresponding dentry other than use rec->ur_name.
1433 Severity : enhancement
1435 Description: lfs setstripe enhancement
1436 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1440 Description: enable data checksumming by default
1441 Details : enable checksum by default, allow --disable-checksum
1442 configure option and "-o nochecksum" mount option. Checksums
1443 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1447 Description: lustre support for RHEL5
1448 Details : Add support for RHEL5.
1452 Description: Client eviction due to failover config
1453 Details : after a connection loss, the lustre client should attempt to
1454 reconnect to the last active server first before trying the
1455 other potential connections.
1459 Description: when mds and osts use different quota unit(32bit and 64bit),
1460 quota will be released repeatly.
1461 Details : void sending multiple quota reqs to mds, which will keep the
1462 status between the reqs.
1466 Description: osts not allocated evenly to files
1467 Details : change the condition to increase offset_idx
1470 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1472 Description: Data corruption for OSTs that are formatted larger than 2TB
1474 Details : When generating the bio request for lustre file writes the
1475 sector number would overflow a temporary variable before being
1476 used for the IO. The data reads correctly from Lustre (which
1477 will overflow in a similar manner) but other file data or
1478 filesystem metadata may be corrupted in some cases.
1482 Description: Tune the kernel for good SCSI performance.
1483 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1484 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1489 Description: rewrite ext2-derived code in obdclass/uuid.c
1490 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1491 from scratch preserving functionality.
1495 Description: "lfs find -obd UUID" prints directories
1496 Details : "lfs find -obd UUID" will return all directory names instead
1497 of just file names. It is incorrect because the directories
1498 do not reside on the OSTs.
1502 Description: use i_size_read and i_size_write in 2.6 port
1503 Details : replace inode->i_size access with i_size_read/write()
1506 Frequency : when removing large files
1508 Description: scheduling issue during removal of large Lustre files
1509 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1510 It causes scheduling issues when removing large files (17TB in the
1514 Frequency : only with liblustre clients on XT3
1516 Description: evictions taking too long
1517 Details : allow llrd to evict clients directly on OSTs
1520 Frequency : only on ppc
1522 Description: /proc/fs/lustre/devices broken on ppc
1523 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1524 We should make sure the type of 'index' is loff_t and then cast
1525 explicitly as needed below. Do not assign an explicitly cast
1529 Frequency : only for rhel5
1531 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1532 Details : Modify the target file & which_kernel.
1534 Severity : enhancement
1536 Description: omit set fsid for export NFS
1537 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1538 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1542 Frequency : Always, for kernels after 2.6.16
1543 Description: Fix warning idr_remove called for id=.. which is not allocated.
1544 Details : Last kernels save old s_dev before kill super and not allow
1545 to restore from callback - restore it before call kill_anon_super.
1549 Description: buffer overruns could theoretically occur
1550 Details : llapi_semantic_traverse() modifies the "path" argument by
1551 appending values to the end of the origin string, and a
1552 overrun may occur. Adding buffer overrun check in liblustreapi.
1556 Description: Fix error on 'ls .' at the top of the Lustre mount.
1557 Details : Don't revalidate dentry if it is a root dentry.
1561 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1562 Details : Modify vm-tunables-rhel4.patch.
1566 Description: Kernel config for 2.6.18-vanilla.
1567 Details : Modify targets/2.6-vanilla.target.in.
1568 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1569 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1570 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1571 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1575 Description: improve handling recoverable errors
1576 Details : if request processig with error which can be recoverable on server
1577 request should be resend, otherwise page released from cache and
1582 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1583 Details : Modify target file & which_patch.
1584 A flaw was found in the IA32 system call emulation provided
1585 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1586 value could be stored in the %RAX register, which could trigger an
1587 out-of-bounds system call table access. An untrusted local user
1588 could exploit this flaw to run code in the kernel
1589 (ie a root privilege escalation). (CVE-2007-4573).
1593 Description: O_DIRECT bypasses client statistics.
1594 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1595 read_ahead_stats were updated. copied the stats section from
1596 osc_send_oap_rpc() into async_internal().
1600 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1604 Description: change order of libsysio includes
1605 Details : '#include sysio.h' should always come before '#include xtio.h'
1609 Description: handle "serious error: objid * already exists" more gracefully
1610 Details : If LAST_ID value on disk is smaller than the objects existing in
1611 the O/0/d* directories, it indicates disk corruption and causes an
1612 LBUG(). If the object is 0-length, then we should use the existing
1613 object. This will help to avoid a full fsck in most cases.
1615 Severity : enhancement
1617 Description: adapt the lustre_config script to support the upgrade case
1618 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1619 server targets to 1.6 in parallel.
1623 Description: To avoid grant space > avaible space when the disk is almost
1624 full. Without this patch you might see the error "grant XXXX >
1625 available" or some LBUG about grant, when the disk is almost
1627 Details : In filter_check_grant, for non_grant cache write, we should
1628 check the left space by if (*left > ungranted + bytes), instead
1629 of (*left > ungranted), because only we are sure the left space
1630 is enough for another "bytes", then the ungrant space should be
1631 increase. In client, we should update cl_avail_grant only there
1632 is OBD_MD_FLGRANT in the reply.
1636 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1637 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1638 described in CVE-2007-4573. This problem could allow untrusted
1639 local users to gain root access.
1642 Frequency : when using O_DIRECT and quotas
1644 Description: Incorrect file ownership on O_DIRECT output files
1645 Details : block usage reported by 'lfs quota' does not take into account
1646 files that have been written with O_DIRECT.
1651 Description: touch file failed when fs is not full
1652 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1653 otherwise we can get ENOSP while fs is not full.
1657 Description: parallel lock callbacks
1658 Details : Instead of sending blocking and completion callbacks as separated
1659 requests, adding them to a set and sending in parallel.
1662 Frequency : only for Cray XT3
1663 Bugzilla : 12829/13455
1664 Description: Changing primary group doesn't change the group lustre assigns to
1666 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1667 overridden with the primary group provided by the group upcall,
1668 whereas the supplied fsgid can be trusted if it is in the list of
1669 supplementary groups returned by the group upcall.
1671 Severity : enhancement
1673 Description: Allow masking D_WARNING, D_ERROR messages from console
1674 Details : Console messages can now be disabled via lnet.printk.
1678 Description: User code with malformed file open parameter crashes client node
1679 Details : Before packing join_file req, all the related reference should be
1680 checked carefully in case some malformed flags cause fake
1681 join_file req on client.
1685 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1686 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1687 mode is equal to the granted mode. However, it should also check
1688 that LDLM_FL_CP_REQD is not set.
1692 Description: Heavy nfs access might result in deadlocks
1693 Details : After ELC code landed, it is now improper to enqueue any mds
1694 locks under och_sem, because enqueue might want to decide to
1695 cancel open locks for same inode we are holding och_sem for.
1699 Description: Client eviction while running blogbench
1700 Details : A lot of unlink operations with concurrent I/O can lead to a
1701 deadlock causing evictions. To address the problem, the number of
1702 oustanding OST_DESTROY requests is now throttled to
1703 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1704 callbacks are processed in priority.
1708 Description: enable ACLs on MDS by default
1709 Details : ACLs must be enabled on MDS by default.
1712 Frequency : PPC/PPC64 only
1714 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1715 Details : fix duplicated definitions between asm-ppc64/types.h and
1716 lustre_types.h on PPC.
1719 Frequency : PPC/PPC64 only
1721 Description: asm-ppc/segment.h does not exist
1722 Details : fix compile issue on PPC.
1726 Description: better handle error messages in extents code
1729 Frequency : RHEL4 only
1731 Description: mkfs is very slow on IA64/RHEL4
1732 Details : A performance regression has been discovered in the MPT Fusion
1733 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1734 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1735 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1737 Severity : enhancement
1739 Description: SNMP support enhancement
1740 Details : Adding total number of sampled request for an MDS node in snmp
1743 Severity : enhancement
1745 Description: Optimize ldlm waiting list processing for PR extent locks
1746 Details : When processing waiting list for read extent lock and meeting read
1747 lock that is same or wider to it that is not contended, skip
1748 processing rest of the list and immediatelly return current
1749 status of conflictness, since we are guaranteed there are no
1750 conflicting locks in the rest of the list.
1754 Description: Time out and refuse to reconnect
1755 Details : When the failover node is the primary node, it is possible
1756 to have two identical connections in imp_conn_list. We must
1757 compare not conn's pointers but NIDs, otherwise we can defeat
1758 connection throttling.
1762 Description: port llog fixes from b1_6 into HEAD
1763 Details : Port llog reference couting and some llog cleanups from b1_6
1764 (bug 10800) into HEAD, for protect from panic and access to already
1765 free llog structures.
1769 Description: Detect stride IO mode in read-ahead
1770 Details : When a client does stride read, read-ahead should detect that and
1771 read-ahead pages according to the detected stride pattern.
1775 Description: data checksumming impacts single node performance
1776 Details : add support for several checksum algorithm. Currently, only CRC32
1777 and Adler-32 are supported. The checksum type can be changed on
1778 the fly via /proc/fs/lustre/osc/*/checksum_type.
1782 Description: use adler32 for page checksums
1783 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1788 Description: build for x2 fails
1789 Details : fix compile issue on Cray systems.
1793 Description: Properly match for duplicate locks
1794 Details : Due to different lock order from skiplists code, we need to
1795 traverse entire list for now
1798 Frequency : only on PPC/SLES10
1800 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1801 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1802 be defined. Add a hack in mkfs_lustre.c to work around this compile
1807 Description: LASSERT on MDS when client holding flock lock dies
1808 Details : ldlm pool logic depends on number of granted locks equal to
1809 number of released locks which is not true for flock locks, so
1810 just exclude such locks from consideration.
1814 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1815 Details : Use fsfilt_sync() for both the whole filesystem sync and
1816 individual file sync to eliminate dangerous inode locking
1817 with I_LOCK that can lead to a deadlock.
1821 Description: performance in 1.6.3
1822 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1823 sure that 1MB requests are not fragmented by the block layer.
1825 Severity : enhancement
1827 Description: organize the server-side client stats on per-nid basis
1828 Details : Change the structure of stats under obdfilter and mds to
1836 The "uuid"s file would list the uuids of _active_ exports.
1837 And the clear entry is to clear all stats and stale nids.
1839 Severity : enhancement
1841 Description: eliminate client locks in face of contention
1842 Details : file contention detection and lockless i/o implementation
1843 for contended files.
1847 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1850 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1852 Description: MDS may lose file striping (and hence file data) in some cases
1853 Details : If there are additional extended attributes stored on the MDS,
1854 in particular ACLs, SELinux, or user attributes (if user_xattr
1855 is specified for the client mount options) then there is a risk
1856 of attribute loss. Additionally, the Lustre file striping
1857 needs to be larger than default (e.g. striped over all OSTs),
1858 and an additional attribute must be stored initially in the
1859 inode and then increase in size enough to be moved to the
1860 external attribute block (e.g. ACL growing in size) for file
1865 Description: skiplist implementation simplification
1866 Details : skiplists are used to group compatible locks on granted list
1867 that was implemented as tracking first and last lock of each
1868 lock group the patch changes that to using doubly linked lists
1872 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1873 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1874 because LASSERT on both the data supplied by a client, and the
1875 data on disk is dangerous and incorrect.
1877 Severity : enhancement
1879 Description: Slow truncate/writes to huge files at high offsets.
1880 Details : Directly associate cached pages to lock that protect those pages,
1881 this allows us to quickly find what pages to write and remove
1882 once lock callback is received.
1886 Description: more ldlm soft lockups
1887 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1888 starve other threads from the resource lock for a long time in
1889 case of long waiting queue, so change the debug level from
1890 D_OTHER to the less frequently used D_INFO.
1892 Severity : enhancement
1894 Description: add -gid, -group, -uid, -user options to lfs find
1898 Description: Hung threads in invalidate_inode_pages2_range
1899 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1900 once one is completed. As a result, some RPCs are stuck in the
1901 queue and are never sent.
1905 Description: filter threads hungs on waiting journal commit
1906 Details : Cleanup filter group llog code, then only filter group llog will
1907 be only created in the MDS/OST syncing process.
1911 Description: Procfs and llog threads access destoryed import sometimes.
1912 Details : Sync the import destoryed process with procfs and llog threads by
1913 the import refcount and semaphore.
1915 Severity : enhancement
1917 Description: openlock cache of b1_6 port to HEAD
1922 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1923 Details : If insertion of an extent fails, then discard the inode
1924 preallocation and free data blocks else it can lead to duplicate
1929 Description: don't always update ctime in ext3_xattr_set_handle()
1930 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1931 In some cases the ctime should not be updated, for example for
1932 2.0->1.8 compatibility it is necessary to delete an xattr and it
1933 should not update the ctime.
1937 Bugzilla : 15713/16362
1938 Description: Assertion in iopen_connect_dentry in 1.6.3
1939 Details : looking up an inode via iopen with the wrong generation number can
1940 populate the dcache with a disconneced dentry while the inode
1941 number is in the process of being reallocated. This causes an
1942 assertion failure in iopen since the inode's dentry list contains
1943 both a connected and disconnected dentry.
1947 Description: assertion failure in ldlm_handle2lock()
1948 Details : fix a race between class_handle_unhash() and class_handle2object()
1949 introduced in lustre 1.6.5 by bug 13622.
1954 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1955 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1960 Description: oops in sd_iostats_seq_show()
1961 Details : unloading/reloading the scsi low level driver triggers a kernel
1962 bug when trying to access the sd iostat file.
1967 Description: Kernel panics during QLogic driver reload
1968 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1969 patch, causing memory corruption.
1974 Description: journal_dev option does not work in b1_6
1975 Details : pass mount option during pre-mount.
1977 Severity : enhancement
1979 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1980 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1981 extent information of a file. It can be used to map logical blocks
1982 in a file to physical blocks in the block device.
1986 Description: LDLM soft lockups - improvement
1987 Details : It is be possible to send the lock handle along with each read
1988 or write request because the client is already doing a lock match
1989 itself so there isn't any reason the OST should have to re-do that
1993 Frequency : only on Cray X2
1995 Description: X2 build failures
1996 Details : fix build failures on Cray X2.
2000 Description: xid & resent requests
2001 Details : Initialize RPC XID from clock at startup (randomly if clock is
2004 Severity : enhancement
2006 Description: Add lustre_start utility to start or stop multiple Lustre servers
2011 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
2012 Details : In case of memory pressure, list_del() can be called twice on
2013 req->rq_history_list, causing a kernel oops.
2017 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
2018 Details : fix stack overflow in the distributed lock manager by defering
2019 export eviction after a failed AST to the elt thread instead of
2020 handling it in the dlm interpret routine.
2024 Description: Convert some comments to new format.
2025 Details : Update documenting comments to match doxygen conventions.
2029 Description: Grammar fixes.
2030 Details : A couple of trivial spelling fixes.
2034 Description: OSD_COUNTERS-mandatory
2035 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
2036 to simplify the code.
2040 Description: simplify cmm_device freeing logic.
2041 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
2045 Description: Add lockdep support to dt_object_operations locking interface.
2046 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
2047 parameter indicating lock ordering. Update mdd code to use new
2052 Description: Introduce failloc constants for lockless IO tests.
2053 Details : Add two new failloc constants to test lockless IO. Only one of
2054 them in implemented---another is checked in yet to be landed
2059 Description: Add lockdep support for inode mutex.
2060 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
2061 in the situations where Lustre has to lock more than one inode
2066 Description: Add optional invariants checking support.
2067 Details : Add new LINVRNT() macro, optional on new --enable-invariants
2068 configure switch. This macro is to be used for consistency and
2069 sanity checks that are too expensive to be left in `production'
2074 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
2075 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
2080 Description: Add lockdep support for ldlm_lock and ldlm_resource.
2081 Details : Use spin_lock_nested() in (the only) situation where more than
2082 one ldlm_lock is locked simultaneously. Also, fix possible
2083 dead-lock in ldlm_lock_change_resource() by enforcing particular
2088 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
2089 Details : Instead of specifying each ldlm_lock call-back through separate
2090 parameter, wrap them into struct ldlm_callback_suite.
2094 Description: Kill join_lru obd method and its callers.
2095 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
2096 in memory---a requirement implemented through ->o_join_lru() obd
2097 method in HEAD. Remove this method and its users.
2101 Description: Add asynchronous ldlm ENQUEUE completion handler.
2102 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
2103 case that stock ldlm_completion_ast() cannot handle as it waits
2104 until lock is granted. Introduce new ldlm_completion_ast_async()
2105 for this. Also comment ldlm_completion_ast().
2109 Description: ldlm_error <-> errno conversion.
2110 Details : Add functions to map (rather arbitrary) between LDLM error codes
2111 and standard errno values. CLIO needs this to prevent LDLM specific
2112 constants from escaping ldlm and osc.
2116 Description: Kill unused ldlm_handle2lock_ns() function.
2117 Details : Kill unused ldlm_handle2lock_ns() function.
2121 Description: Add lu_ref support to ldlm_lock
2122 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
2123 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
2124 and ldlm_resource. LDLM interface has to be changed, because code that
2125 releases a reference on a lock, has to "know" what reference this is.
2126 In the most frequent case
2128 lock = ldlm_handle2lock(handle);
2130 LDLM_LOCK_PUT(lock);
2132 no changes are required. When any other reference (received _not_ from
2133 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
2134 instead of LDLM_LOCK_PUT().
2136 Arguably, changes are pervasive, and interface requires some discipline
2137 for proper use. On the other hand, it was very instrumental in finding
2138 a few leaked lock references.
2142 Description: Add ldlm_lock_addref_try().
2143 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
2144 attempts to addref a lock that might be being canceled
2149 Description: Add ldlm_weigh_callback().
2150 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
2151 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
2152 hard-coded `number of pages' logic.
2156 Description: Add lockdep annotations to llog code.
2157 Details : Use appropriately tagged _nested() locking calls in the places
2158 where llog takes more than one ->lgh_lock lock.
2162 Description: Add loi_kms_set().
2163 Details : Wrap kms updates into a helper function.
2167 Description: Constify instances of struct lsm_operations.
2168 Details : Constify instances of struct lsm_operations.
2172 Description: lu_conf support.
2173 Details : On a server, a file system object is uniquely identified
2174 by a fid, which is sufficient to locate and load all object
2175 state (inode). On a client, on the other hand, more data are
2176 necessary instantiate an object. Change lu_object_find() and
2177 friends to take additional `lu_conf' argument describing object.
2178 Typically this includes layout information.
2182 Description: lu_context fixes.
2183 Details : Introduce new lu_context functions that are needed on the client
2184 side, where some system threads (ptlrpcd) are shared by multiple
2185 modules, and so cannot be stopped during module shutdown.
2189 Description: Add start and stop methods to lu_device_type_operations.
2190 Details : Introduce two new methods in lu_device_type_operations, that are
2191 invoked when first instance of a given type is created and last one
2192 is destroyed respectively. This is need by CLIO.
2196 Description: Add lu_ref support to struct lu_device.
2197 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2198 track leaked references.
2202 Description: Introduce lu_kmem_descr.
2203 Details : lu_kmem_descr and its companion interface allow to create
2204 and destroy a number of kmem caches at once.
2208 Description: Fix lu_object finalization race.
2209 Details : Fix a race between lu_object_find() finding an object and its
2210 concurrent finalization. This race is (most likely) not possible
2211 on the server, but might happen on the client.
2215 Description: Introduce lu_ref interface.
2216 Details : lu_ref is a debugging module allowing to track references to
2217 a given object. It is quite cpu expensive, and has to be
2218 explicitly enabled with --enable-lu_ref. See usage description
2223 Description: Factor lu_site procfs stats into a separate function.
2224 Details : Separate lu_site stats printing code into a separate function
2225 that can be reused on a client.
2229 Description: Constify instances of struct {lu,dt,md}_device_operations.
2230 Details : Constify instances of struct {lu,dt,md}_device_operations.
2234 Description: Introduce struct md_site and move meta-data specific parts of
2235 struct lu_site here.
2236 Details : Move md-specific fields out of struct lu_site into special struct
2237 md_site, so that lu_site can be used on a client.
2241 Description: Kill mdd_lov_destroy().
2242 Details : Remove unused mdd code.
2246 Description: Add st_block checking to multistat.c.
2247 Details : Add st_block checking to multistat.c.
2251 Description: Add lu_ref support to struct obd_device.
2252 Details : Add lu_ref tracking to obd_device.
2256 Description: Kill obd_set_fail_loc().
2257 Details : Remove unused code.
2261 Description: Add special type for ptlrpc_request interpret functions.
2262 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2263 there. Actual usage will be in CLIO.
2267 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2268 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2269 doesn't work with lockdep.
2273 Description: Add rwv.c test program.
2274 Details : New testing program exercising readv(2) and writev(2) (Qian).
2278 Description: Add sendfile.c test program.
2279 Details : New testing program exercising sendfile(2) (Jay).
2283 Description: Ratelimit a message that can be very frequent.
2284 Details : Ratelimit a memory allocation failure message that can
2289 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2290 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2294 Description: Liblustre build fixes.
2295 Details : Liblustre build fixes.
2299 Description: libcfs: add cfs_{need,cond}_resched() interface.
2300 Details : libcfs: add cfs_{need,cond}_resched() definition and
2301 implementations for Linux, NT, and liblustre.
2303 Severity : enhancement
2305 Description: More exported tunables for mballoc
2306 Details : Add support for tunable preallocation window and new tunables for
2307 large/small requests
2311 Description: Detect corruption of block bitmap and checking for preallocations
2312 Details : Checks validity of on-disk block bitmap. Also it does better
2313 checking of number of applied preallocations. When corruption is
2314 found, it turns filesystem readonly to prevent further corruptions.
2318 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2319 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2320 window to 0. If it does hit cache miss, and read-pattern is still
2321 stride-io mode, does not reset the stride window, but also does not
2322 increase the stride window length in this case.
2326 Frequency : only for big-endian servers
2327 Description: Check if system is big-endian while mounting fs with extents feature
2328 Details : Mounting a filesystem with extents feature will fail on big-endian
2329 systems since ext3-based ldiskfs is not supported on big-endian
2330 systems. This can be over-riden with "bigendian_extents" mount option.
2332 Severity : enhancement
2334 Description: The root squash functionality
2335 Details : A security feature, which is to prevent users from being able
2336 to mount lustre on their desktop, run as root, and delete
2337 all of the files in the filesystem. The goal is accomplished by
2338 remapping user id (UID) and group id (GID) of the root user to
2339 a UID and GID specified by the system administartor via Lustre
2340 configuration management server (MGS). The functionality also
2341 allows to specify sets of clients for which the remapping does
2346 Description: Excessive recovery window
2347 Details : With AT enabled, the recovery window can be excessively long (6000+
2348 seconds). To address this problem, we no longer use
2349 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2350 timeout no longer depends on the service time, it is set to
2351 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2352 time via pb_service_time.
2356 Description: Watchdog triggered on MDS failover
2357 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2358 the OSTs know that the MDS "UUID" can be reused for the same export
2359 from a different NID, so we do not need to wait for the export to be
2363 Frequency : rare, only if using MMP with Linux RAID
2365 Description: MMP doesn't work with Linux RAID
2366 Details : While using HA for Lustre servers with Linux RAID, it is possible
2367 that MMP will not detect multiple mounts. To make this work we
2368 need to unplug the device queue in RAID when the MMP block is being
2369 written. Also while reading the MMP block, we should read it from
2370 disk and not the cached one.
2372 Severity : enhancement
2374 Description: open file using fid
2375 Details : A file can be opened using just its fid, like
2376 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2379 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2382 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2383 Details : add proper locking for creation of jbd_4k slab cache
2387 Description: MMP check in ext3_remount() fails without displaying any error
2388 Details : When multiple mount protection fails during remount, proper error
2391 Severity : enhancement
2393 Description: Allow stripe size to be up to 4G-64k
2394 Details : Fix math logic to allow large stripe sizes.
2398 Description: add check for >8TB ldiskfs filesystems
2399 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2400 Don't allow >8TB ldiskfs filesystems to be mounted without
2401 force_over_8tb mount option
2406 Descriptoin: OSS complains "too many blocks" and crashes
2407 Details : In filter_commitrw_write():
2408 1. lquota_chkquota();
2409 2. obdo_from_inode();
2410 3. lquota_pending_commit();
2411 "2." has changed uid and gid in this case, then caused
2412 lquota_chkquota() and lquota_pending_commit() to use
2413 different uid and gid. That is the root of the bug.
2417 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2418 Details : Detect and print out the 'stripe count = -1' case properly.
2422 Description: Out or order replies might be lost on replay
2423 Details : In ptlrpc_retain_replayable_request if we cannot find retained
2424 request with tid smaller then one currently being added, add it
2425 to the start, not end of the list.
2427 --------------------------------------------------------------------------------
2429 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2431 * Support for kernels:
2432 2.6.5-7.283 (SLES 9),
2433 2.6.9-55.EL (RHEL 4),
2434 2.6.16.46-0.14 (SLES 10),
2435 2.6.18.8 vanilla (kernel.org)
2436 * Client support for unpatched kernels:
2437 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2438 2.6.16 - 2.6.22 vanilla (kernel.org)
2439 * Due to recently discovered recovery problems, we do not recommend
2440 using patchless RHEL 4 clients with this or any earlier release.
2441 * Recommended e2fsprogs version: 1.39.cfs8
2442 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2443 * Starting with this release, the ldiskfs backing filesystem required
2444 by Lustre is now in its own package, lustre-ldiskfs. This package
2445 should be installed. It is versioned separately from Lustre and
2446 may be released separately in future.
2450 Description: block reactivating mgc import until all deactivates complete
2451 Details : Fix race when failing back MDT/MGS to itself (testing)
2454 Frequency : at statup only
2456 Description: mds_lov_synchronize race leads to various problems
2457 Details : simultaneous MDT->OST connections at startup can cause the
2458 sync to abort, leaving the OSC in a bad state.
2460 Severity : enhancement
2462 Description: add optional extra BUILD_VERSION info
2463 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2464 to override the lustre version.
2467 Frequency : 2.6.18 servers only
2469 Description: ll_kern_mount() doesn't release the module reference
2470 Details : The ldiskfs module reference count never drops down to 0
2471 because ll_kern_mount() doesn't release the module reference.
2476 Description: server LBUG when using old ost_num_threads parameter
2477 Details : Accept the old ost_num_threads parameter but warn that it
2478 is deprecated, and fix an off-by-one error that caused an LBUG.
2483 Description: Transient SCSI error results in persistent IO issue
2484 Details : iobuf->dr_error is not reinitialized to 0 between two
2488 Frequency : sometimes when underlying device returns I/O errors
2490 Description: OSTs not going read-only during write failures
2491 Details : OSTs are not remounted read-only when the journal commit threads
2492 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2493 instead of the ext3 wrappers.
2497 Description: poor connect scaling with increasing client count
2498 Details : Don't run filter_grant_sanity_check for more than 100 exports
2499 to improve scaling for large numbers of clients.
2502 Frequency : SLES10 only
2504 Description: sanity-quota.sh quotacheck failed: rc = -22
2505 Details : Quotas cannot be enabled on SLES10.
2508 Frequency : liblustre clients only
2510 Description: getdirentries does not give error when run on compute nodes
2511 Details : getdirentries does not fail when the size specified as an argument
2512 is too small to contain at least one entry
2514 Severity : enhancement
2516 Description: Add LNET router traceability for debug purposes
2517 Details : If a checksum failure occurs with a router as part of the
2518 IO path, the NID of the last router that forwarded the bulk data
2519 is printed so it can be identified.
2524 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2525 Details : Due to a race condition, liblustre clients were occasionally
2526 evicted incorrectly.
2528 Severity : enhancement
2530 Description: lfs setstripe use optional parameters instead of postional
2533 Severity : enhancement
2535 Description: Nanosecond timestamp support for ldiskfs
2536 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2537 resolution timestamps. There is not yet support for this at
2538 the Lustre filesystem level.
2541 Frequency : during server recovery
2543 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2544 Details : request with rq_no_resend flag not awake l_wait_event if they get
2548 Frequency : nfs export on patchless client
2550 Description: connectathon hang when test nfs export over patchless client
2551 Details : Disconnected dentry cannot be found with lookup, so we do not need
2552 to unhash it or make it invalid
2555 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2556 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2557 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2562 Description: open req refcounting wrong on reconnect
2563 Details : If reconnect happened between getting open reply from server and
2564 call to mdc_set_replay_data in ll_file_open, we will schedule
2565 replay for unreferenced request that we are about to free.
2566 Subsequent close will crash in variety of ways.
2567 Check that request is still eligible for replay in
2568 mdc_set_replay_data().
2573 Description: disable writes to filesystem when reading health_check file
2574 Details : the default for reading the health_check proc file has changed
2575 to NOT do a journal transaction and write to disk, because this
2576 can cause reads of the /proc file to hang and block HA state
2577 checking on a healthy but otherwise heavily loaded system. It
2578 is possible to return to the previous behaviour during configure
2579 with --enable-health-write.
2581 Severity : enhancement
2583 Description: 64-bit inode version
2584 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2585 to the inode. This will be required for version-based recovery.
2590 Description: MDS fails to start if a duplicate client export is detected
2591 Details : in some rare cases it was possible for a client to connect to
2592 an MDS multiple times. Upon recovery the MDS would detect this
2593 and fail during startup. Handle this more gracefully.
2595 Severity : enhancement
2597 Description: Add -o localflock option to simulate old noflock
2599 Details : This will achieve local-only flock/fcntl locks
2605 Description: log_commit_thread vs filter_destroy race leads to crash
2606 Details : Take import reference before releasing llog record semaphore
2611 Description: Wrong request locking in request set processing
2612 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2613 imp_delayed_list, in this place should be used imp_lock.
2616 Frequency : when reconnection
2618 Description: Grant Leak when osc reconnect to OST
2619 Details : When osc reconnect ost, OST(filter) should check whether it
2620 should grant more space to client by comparing fed_grant and
2621 cl_avail_grant, and return the granted space to client instead
2622 of "new granted" space, because client will call osc_init_grant
2623 to update the client grant space info.
2626 Frequency : when client reconnect to OST
2628 Description: Grant Leak when osc do resend and replay bulk write
2629 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2630 bulk write request, because the grant info while be sync between
2631 OSC and OST when reconnect, and we should ignore the grant info
2632 these of resend/replay write req.
2637 Description: Grant space more than avaiable left space sometimes.
2638 Details : When then OST is about to be full, if two bulk writing from
2639 different clients came to OST. Accord the avaliable space of the
2640 OST, the first req should be permitted, and the second one
2641 should be denied by ENOSPC. But if the seconde arrived before
2642 the first one is commited. The OST might wrongly permit second
2643 writing, which will cause grant space > avaiable space.
2646 Frequency : when client is evicted
2648 Description: Grant might be wrongly erased when osc is evicted by OST
2649 Details : when the import is evicted by server, it will fork another
2650 thread ptlrpc_invalidate_import_thread to invalidate the
2651 import, where the grant will be set to 0. While the original
2652 thread will update the grant it got when connecting. So if
2653 the former happened latter, the grant will be wrongly errased
2654 because of this race.
2659 Description: Checking Stale with correct fid
2660 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2661 to check whether it is stale, because sometimes, we want the
2662 enqueue happened anyway, and op_data.fid2 will not be initialized.
2664 Severity : enhancement
2666 Description: update patchless client
2667 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2670 Frequency : only with 2.4 kernel
2672 Description: random memory corruption
2673 Details : size of struct ll_inode_info is to big for union inode.u and this
2674 can be cause of random memory corruption.
2679 Description: Memory leak in recovery
2680 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2681 It should also check obd_fail before fsfilt_start, otherwise if
2682 fsfilt_start return -EROFS,(failover mds during mds recovery).
2683 then the req will return with repmsg->transno = 0 and rc = EROFS.
2684 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2685 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2686 be freed no matter whether fsfilt_commit success or not.
2689 Frequency : only with huge count clients
2691 Description: Prevents from taking the superblock lock in llap_from_page for
2693 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2694 for llap_from_page call in ll_removepage prevents from taking the
2695 superblock lock for a soon died page.
2700 Description: Not check open intent error before release open handle
2701 Details : in some rare cases, the open intent error is not checked before
2702 release open handle, which may cause
2703 ASSERTION(open_req->rq_transno != 0), because it tries to release
2704 the failed open handle.
2709 Description: Set cat log bitmap only after create log success.
2710 Details : in some rare cases, the cat log bitmap is set too early. and it
2711 should be set only after create log success.
2715 Description: Accessing a block bevice can re-enable I/O when Lustre is
2716 tearing down a device.
2717 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2722 Description: service threads may hog cpus when there are a lot of requests
2723 Details : Insert cond_resched to give other threads a chance to use some CPU
2728 Description: the cat log was not initialized in recovery
2729 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2730 unlink log on mds will not be initialized until mds post
2731 recovery. And also in mds post recovery, the unlink log will
2732 initialization will be done asynchronausly, so there will be race
2733 between add unlink log and unlink log initialization.
2737 Description: brw_stats were being printed incorrectly
2738 Details : brw_stats were being printed as log2 but all of them were not
2739 recorded as log2. Also remove some code duplication arising from
2740 filter_tally_{read,write}.
2744 Frequency : rare, only in recovery.
2745 Description: ASSERTION(req->rq_type != LI_POISON) failed
2746 Details : imp_lock should be held while iterating over imp_sending_list for
2747 prevent destroy request after get timeout in ptlrpc_queue_wait.
2751 Description: replay-single.sh test 52 fails
2752 Details : A lock's skiplist need to be cleanup when it being unlinked
2753 from its resource list.
2757 Description: Short directio read returns full requested size rather than
2759 Details : Direct I/O operations should return actual amount of bytes
2760 transferred rather than requested size.
2762 Severity : enhancement
2764 Description: metadata RPC reduction (e.g. for rm performance)
2765 Details : decrease the amount of synchronous RPC between clients and servers
2766 by canceling conflicing lock before the operation on the client
2767 and packing thier handles into the main operation RPC to server.
2769 Severity : enhancement
2771 Description: Async OSC create to avoid the blocking unnecessarily.
2772 Details : If a OST has no remain object, system will block on the creating
2773 when need to create a new object on this OST. Now, ways use
2774 pre-created objects when available, instead of blocking on an
2775 empty osc while others are not empty. If we must block, we block
2776 for the shortest possible period of time.
2780 Description: improve handling recoverable errors
2781 Details : if request processig with error which can be recoverable on server
2782 request should be resend, otherwise page released from cache and
2785 Severity : enhancement
2787 Description: refine locking for avoid write wrong info into lov_objid file
2788 Details : fix possible races with add new target and write/update data in
2791 --------------------------------------------------------------------------------
2793 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2798 Frequency : on some architectures
2800 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2801 Details : Uninitialized flags sometimes cause configuration commands to
2805 Frequency : patchless clients only
2807 Description: missing __iget() symbol export
2808 Details : The __iget() symbol export is missing. To avoid the need for
2809 this on patchless clients the deathrow inode reaper is turned
2810 off, and we depend on the VM to clean up old inodes. This
2811 dependency was during via the fix for bug 12181.
2813 --------------------------------------------------------------------------------
2815 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2817 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2818 INTEROPERATE with older versions automatically. In many cases a
2819 special upgrade step is needed. Please read the
2820 user documentation before upgrading any part of a live system.
2821 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2822 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2823 * WARNING: Lustre configuration and startup changes are required with
2824 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2828 Severity : enhancement
2830 Description: MountConf
2831 Details : Lustre configuration is now managed via mkfs and mount
2832 commands instead of lmc and lconf. New obd types (MGS, MGC)
2833 are added for dynamic configuration management. See
2834 https://mail.clusterfs.com/wikis/lustre/MountConf for
2837 Severity : enhancement
2839 Description: dynamic OST addition
2840 Details : OSTs can now be added to a live filesystem
2842 Severity : enhancement
2844 Description: startup order invariance
2845 Details : MDTs and OSTs can be started in any order. Clients only
2846 require the MDT to complete startup.
2848 Severity : enhancement
2850 Description: parallel, asynchronous orphan cleanup
2851 Details : orphan cleanup is now performed in separate threads for each
2852 OST, allowing parallel non-blocking operation.
2854 Severity : enhancement
2856 Description: optimized stripe assignment
2857 Details : stripe assignments are now made based on ost space available,
2858 ost previous usage, and OSS previous usage, in order to try
2859 to optimize storage space and networking resources.
2861 Severity : enhancement
2863 Description: Permanently set tunables
2864 Details : All writable /proc/fs/lustre tunables can now be permanently
2865 set on a per-server basis, at mkfs time or on a live system.
2867 Severity : enhancement
2869 Description: Lustre message v2
2870 Details : Add lustre message format v2.
2872 Severity : enhancement
2874 Description: client OST exclusion list
2875 Details : Clients can be started with a list of OSTs that should be
2876 declared "inactive" for known non-responsive OSTs.
2880 Description: SPEC SFS validation failure on NFS v2 over lustre.
2881 Details : Changes the blocksize for regular files to be 2x RPC size,
2882 and not depend on stripe size.
2884 Severity : enhancement
2886 Description: Multiple MD RPCs in flight.
2887 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2888 To avoid overly-overloading MDS, introduce a limit on number of
2889 MDS RPCs in flight for a single client and add /proc controls
2890 to adjust this limit.
2892 Severity : enhancement
2894 Description: client read/write statistics
2895 Details : Add client read/write call usage stats for performance
2896 analysis of user processes.
2897 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2898 file access. extents_stats shows chunk size distribution.
2899 extents_stats_per_process show chunk size distribution per
2902 Severity : enhancement
2904 Description: mds statistics
2905 Details : Add detailed mds operations statistics in
2906 /proc/fs/lustre/mds/*/stats.
2910 Description: Failure of copying files with lustre special EAs.
2911 Details : Client side always return success for setxattr call for lustre
2912 special xattr (currently only "trusted.lov").
2917 Description: Refcount LNET uuids
2918 Details : The global LNET uuid list grew linearly with every startup;
2919 refcount repeated list entries instead of always adding to
2922 Severity : enhancement
2924 Description: Dynamic service threads
2925 Details : Within a small range, start extra service threads
2926 automatically when the request queue builds up.
2929 Frequency : mixed-endian client/server environments
2931 Description: mixed-endian crashes
2932 Details : The new msg_v2 system had some failures in mixed-endian
2935 Severity : enhancement
2937 Description: Easy OST removal
2938 Details : OSTs can be permanently deactivated with e.g. 'lctl
2939 conf_param lustre-OST0001.osc.active=0'
2941 Severity : enhancement
2943 Description: MGS proc entries
2944 Details : Added basic proc entries for the MGS showing what filesystems
2947 Severity : enhancement
2949 Description: provide MGS failover
2950 Details : Added config lock reacquisition after MGS server failover.
2952 Severity : enhancement
2954 Description: add Linux 2.4 support
2955 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2959 Description: a large application tries to do I/O to the same resource and dies
2960 in the middle of it.
2961 Details : Check the req->rq_arrival time after the call to
2962 ost_brw_lock_get(), but before we do anything about
2963 processing it & sending the BULK transfer request. This
2964 should help move old stale pending locks off the queue as
2965 quickly as obd_timeout.
2968 Frequency : when an incorrect nid is specified during startup
2970 Description: ptlrpc connect to non-existant node causes kernel crash
2971 Details : LNET can't be re-entered from an event callback, which
2972 happened when we expire a message after the export has been
2973 cleaned up. Instead, hand the zombie cleanup off to another
2976 Severity : enhancement
2978 Description: plain/inodebits lock performance improvement
2979 Details : Grouping plain/inodebits in granted list by their request modes
2980 and bits policy, thus improving the performance of search through
2984 Frequency : only if OST filesystem is corrupted
2986 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2987 Details : for a short time RPCs with bulk IO are in the replay list,
2988 but replay of bulk IOs is unimplemented. If the OST filesystem
2989 is corrupted due to disk cache incoherency and then replay is
2990 started it is possible to trip an assertion. Avoid putting
2991 committed RPCs into the replay list at all to avoid this issue.
2994 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2996 Description: System hang on startup
2997 Details : This bug allowed the liblustre (e.g. catamount) client to
2998 return to the app before handling all startup RPCs. This
2999 could leave the node unresponsive to lustre network traffic
3000 and manifested as a server ptllnd timeout.
3002 Severity : enhancement
3004 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3005 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
3006 boolean to control whether to print peer debug info when a
3007 client's RPC times out.
3010 Frequency : only for kernels with patches from Lustre below 1.4.3
3012 Description: Remove old rdonly API
3013 Details : Remove old rdonly API which unused from at least lustre 1.4.3
3016 Frequency : only for devices with external journals
3018 Description: Set external device read-only also
3019 Details : During a commanded failover stop, we set the disk device
3020 read-only while the server shuts down. We now also set any
3021 external journal device read-only at the same time.
3024 Frequency : when upgrading from 1.4 while trying to change parameters
3026 Description: The wrong (new) MDC name was used when setting parameters for
3027 upgraded MDT's. Also allows changing of OSC (and MDC)
3028 parameters if --writeconf is specified at tunefs upgrade time.
3031 Frequency : when setting specific ost indicies
3033 Description: QOS code breaks on skipped indicies
3034 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3035 created with --index need not be sequential.
3037 Severity : enhancement
3039 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3040 Details : The uninit_groups feature works in conjunction with the kernel
3041 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3042 up the pass1 processing of e2fsck. This is a read-only feature
3043 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3044 mount filesystems that have had this feature enabled.
3046 Severity : enhancement
3048 Description: Improve multi-block allocation algorithm to avoid fragmentation
3049 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3050 allocation locality and avoid filesystem fragmentation.
3052 ------------------------------------------------------------------------------
3054 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3056 * Support for kernels:
3057 2.4.21-47.0.1.EL (RHEL 3)
3058 2.6.5-7.283 (SLES 9)
3059 2.6.9-42.0.10.EL (RHEL 4)
3060 2.6.12.6 vanilla (kernel.org)
3061 2.6.16.27-0.9 (SLES 10)
3062 * Recommended e2fsprogs version: 1.39.cfs5
3064 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3068 Frequency : occasional, depends on client load and configuration
3069 Bugzilla : 12181, 12203
3070 Description: data loss for recently-modified files
3072 Details : In some cases it is possible that recently written or created
3073 files may not be written to disk in a timely manner (this should
3074 normally be within 30s unless client IO load is very high).
3075 The problem appears as zero-length files or files that are a
3076 multiple of 1MB in size after a client crash or client eviction
3077 that are missing data at the end of the file.
3079 This problem is more likely to be hit on clients where files are
3080 repeatedly created and unlinked in the same directory, clients
3081 have a large amount of RAM, have many CPUs, the filesystem has
3082 many OSTs, the clients are rebooted frequently, and/or the files
3083 are not accessed by other nodes after being written.
3085 The presence of the problem can be detected by looking at
3086 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
3087 smaller than the second (nr_unused) then dirty files will not
3088 be flushed automatically to disk. "sync; sleep 10" should be
3089 run several times on the node before unmounting it to update
3090 Lustre (this is also safe to run on nodes without this problem).
3092 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
3093 that can cause this same problem, so customers using that kernel
3094 also need to update the kernel in addition to Lustre. In order
3095 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3097 It is normal that files written just before a client crash (less
3098 than 30s) may not yet have been flushed to disk, even for local
3102 Frequency : frequent on thin XT3 nodes
3104 Description: UUID collision on thin XT3 Linux nodes
3105 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3106 randomly, since we relied on an insufficiently-seeded PRNG.
3111 Description: OSS hangs after "All ost request buffers busy"
3112 Details : A deadlock between quota and journal operations caused OSS
3113 hangs after printing "All ost request buffers busy."
3116 Frequency : always on liblustre builds
3118 Description: Cleanup compiler warnings on liblustre
3121 Frequency : always on liblustre builds on XT3
3123 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3127 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3128 > 32000 subdirectories
3129 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3130 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3131 subdirectory count crosses 32000. This will aid e2fsck to
3132 correctly handle more than 32000 subdirectories.
3135 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3137 Description: System hang on startup
3138 Details : This bug allowed the liblustre (e.g. catamount) client to
3139 return to the app before handling all startup RPCs. This
3140 could leave the node unresponsive to lustre network traffic
3141 and manifested as a server ptllnd timeout.
3143 Severity : enhancement
3145 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3146 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3147 boolean to control whether to print peer debug info when a
3148 client's RPC times out.
3153 Description: make O_SYNC working on 2.6 kernels
3154 Details : 2.6 kernels use different method for mark pages for write,
3155 so need add a code to lustre for O_SYNC work.
3160 Description: Failure to close file and release space on NFS
3161 Details : Put inode details into lock acquired in ll_intent_file_open.
3162 Use mdc_intent_lock in ll_intent_open to properly
3163 detect all kind of errors unhandled by mdc_enqueue.
3168 Description: proc file read during shutdown sometimes raced obd removal,
3170 Details : Add lock to prevent obd access after proc file removal.
3173 Frequency : Only for files larger than 4GB on 32-bit clients.
3175 Description: improperly doing page alignment of locks
3176 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3177 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3182 Description: RPCs being resent when they shouldn't be
3183 Details : Some RPCs that should not be resent are being resent. This
3184 can cause inconsistencies in the RPC state machine. Do not
3185 resend such requests.
3188 Frequency : rare, only with NFS export
3190 Description: Crash on NFS re-export node
3191 Details : under very unusual load conditions an assertion is hit in
3192 ll_intent_file_open()
3195 Frequency : only if OST filesystem is corrupted
3197 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3198 Details : for a short time RPCs with bulk IO are in the replay list,
3199 but replay of bulk IOs is unimplemented. If the OST filesystem
3200 is corrupted due to disk cache incoherency and then replay is
3201 started it is possible to trip an assertion. Avoid putting
3202 committed RPCs into the replay list at all to avoid this issue.
3207 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3208 Details : Large single O_DIRECT read and write calls can fail to allocate
3209 a sufficiently large buffer to process the request. In case of
3210 allocation failure the allocation is retried with a smaller
3211 buffer and broken into smaller requests.
3213 Severity : enhancement
3215 Description: Add -o localflock option to simulate old noflock behaviour.
3216 Details : This will achieve local-only flock/fcntl locks coherentness.
3221 Description: versioning check is incomplete
3222 Details : Checking the version difference of client vs. server, report
3223 error if the gap is too big.
3228 Description: add support PG_writeback bit
3229 Details : add support for PG_writeback bit for Lustre, for more carefull
3230 work with page cache in 2.6 kernel. This also fix some deadlocks
3231 and remove hack for work O_SYNC with 2.6 kernel.
3233 Severity : enhancement
3235 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3236 Details : The uninit_groups feature works in conjunction with the kernel
3237 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3238 up the pass1 processing of e2fsck. This is a read-only feature
3239 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3240 mount filesystems that have had this feature enabled.
3242 Severity : enhancement
3244 Description: Improve multi-block allocation algorithm to avoid fragmentation
3245 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3246 allocation locality and avoid filesystem fragmentation.
3248 ------------------------------------------------------------------------------
3250 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3252 * Support for kernels:
3253 2.6.9-42.0.3.EL (RHEL 4)
3254 2.6.5-7.276 (SLES 9)
3255 2.4.21-47.0.1.EL (RHEL 3)
3256 2.6.12.6 vanilla (kernel.org)
3257 2.6.16.21-0.8 (SLES10)
3258 * Recommended e2fsprogs version: 1.39.cfs2-0
3260 * The backwards-compatible /proc/sys/portals symlink has been removed
3261 in this release. Before upgrading, please ensure that you change
3262 any configuration scripts or /etc/sysctl.conf files that access
3263 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3264 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3265 in advance of the upgrade on any system running Lustre 1.4.6 or
3266 newer, since /proc/sys/lnet was added in that version.
3267 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3271 Frequency : only when quota is used
3273 Description: avoid scanning export list for quota master
3274 Details : Change the algorithms to avoid scanning export list in order
3275 to improve the efficiency.
3278 Frequency : MDS failover only, very rarely
3280 Description: "went back in time" messages on mds failover
3281 Details : The greatest transno may be lost when the current operation
3282 finishes with an error (transno==0) and the client's last_rcvd
3283 record is over-written. Save the greatest transno in the
3284 mds_last_transno for this case.
3287 Frequency : always for specific kernels and striping counts
3289 Description: client may get "Matching packet too big" without ACL support
3290 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3291 when trying to access files in certain configurations. The
3292 clients should in fact be denied when mounting because they do
3293 not understand ACLs.
3296 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3298 Description: many clients connecting with IO in progress causes connect timeouts
3299 Details : Avoid synchronous journal commits to avoid delays caused by many
3300 clients connecting/disconnecting when bulk IO is in progress.
3301 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3302 OST_IO_PORTAL to avoid delays behind potentially many pending
3306 Frequency : occasionally with multiple writers to a single file
3308 Description: shared writes to file may result in wrong size reported by stat()
3309 Details : Allow growing of kms when extent lock is cancelled
3312 Frequency : always with random mmap IO to multi-striped file
3314 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3315 Details : Only if the hole is at the end of OST object so that kms is too
3316 small. Fix is to increase kms accordingly in ll_nopage.
3319 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3321 Description: writes to a missing object would leak memory on the OST
3322 Details : If there is an inconsistency between the MDS and OST filesystems,
3323 such that the MDS references an object that doesn't exist, writes
3324 to that object will leak memory due to incorrect cleanup in the
3325 error handling path, eventually running out of memory on the OST.
3330 Description: Creating too long symlink causes lustre errors
3331 Details : Check symlink and name lengths before sending requests to MDS.
3334 Frequency : only if flock is enabled (not on by default)
3336 Description: posix locks not released on fd closure on 2.6.9+
3337 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3338 kernels, this caused such locks not to be released on fd close and
3339 then assertions on fs unmount about still used locks.
3342 Frequency : MDS failover only, very rarely
3344 Description: clients may get ASSERTION(granted_lock != NULL)
3345 Details : When request was taking a long time, and a client was resending
3346 a getattr by name lock request. The were multiple lock requests
3347 with the same client lock handle and
3348 mds_getattr_name->fixup_handle_for_resent_request found one of the
3349 lock handles but later failed with ASSERTION(granted_lock != NULL).
3354 Description: handle->h_buffer_credits > 0, assertion failure
3355 Details : h_buffer_credits is zero after truncate, causing assertion
3356 failure. This patch extends the transaction or creates a new
3360 Frequency : NFS re-export or patchless client
3361 Bugzilla : 11179, 10796
3362 Description: Crash on NFS re-export node (__d_move)
3363 Details : We do not want to hash the dentry if we don't have a lock.
3364 But if this dentry is later used in d_move, we'd hit uninitialised
3365 list head d_hash, so we just do this to init d_hash field but
3366 leave dentry unhashed.
3369 Frequency : NFS re-export or patchless client
3371 Description: NFS exports has problem with symbolic link
3372 Details : lustre client didn't properly install dentry when re-exported
3373 to NFS or running patchless client.
3376 Frequency : NFS re-export or patchless client
3378 Description: Various nfs/patchless fixes.
3379 Details : fixes reuse disconected alias for lookup process - this fixes
3380 warning "find_exported_dentry: npd != pd",
3381 fix permission error with open files at nfs.
3382 fix apply umask when do revalidate.
3385 Frequency : occasional
3387 Description: Crash on NFS re-export node
3388 Details : calling clear_page() on the wrong pointer triggered oops in
3389 generic_mapping_read().
3392 Frequency : rarely, using O_DIRECT IO
3394 Description: unaligned directio crashes client with LASSERT
3395 Details : check for unaligned buffers before trying any requests.
3398 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3400 Description: stale data returned from RAID cache
3401 Details : If only a small amount of IO is done to the RAID device before
3402 reading it again it is possible to get stale data from the RAID
3403 cache instead of reading it from disk.
3406 Frequency : always for sles10 kernel
3408 Description: sles10 support
3409 Details : ll_follow_link: compile fixes and using of nd_set_link
3410 under newer kernels.
3413 Frequency : depends on arch, kernel and compiler version, always on sles10
3416 Description: recursive or deep enough symlinks cause stack overflow
3417 Details : getting rid of large stack-allocated variable in
3421 Frequency : depends on hardware
3423 Description: lustre write performance loss in the SLES10 kernel
3424 Details : the performance loss is caused by using of write barriers in the
3425 ext3 code. The SLES10 kernel turns barrier support on by
3426 default. The fix is to undo that change for ldiskfs.
3428 ------------------------------------------------------------------------------
3430 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3432 * Support for kernels:
3433 2.6.9-42.0.3EL (RHEL 4)
3434 2.6.5-7.276 (SLES 9)
3435 2.4.21-47.0.1.EL (RHEL 3)
3436 2.6.12.6 vanilla (kernel.org)
3440 Frequency : quota enabled and large files being deleted
3442 Description: releasing more than 4GB of quota at once hangs OST
3443 Details : If a user deletes more than 4GB of files on a single OST it
3444 will cause the OST to spin in an infinite loop. Release
3445 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3450 Description: statfs data retrieved from /proc may be stale or zero
3451 Details : When reading per-device statfs data from /proc, in the
3452 {kbytes,files}_{total,free,avail} files, it may appear
3453 as zero or be out of date.
3456 Frequency : systems with MD RAID1 external journal devices
3458 Description: lconf's call to blkid is confused by RAID1 journal devices
3459 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3460 of returning all block devices that match the journal UUID.
3463 Frequency : always, for aggregate stripe size over 4GB
3465 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3466 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3467 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3470 Frequency : always if "lfs find" used on a local file/directory
3472 Description: "lfs find" segfaults if used on a local file/directory
3473 Details : The case where a directory component was not specified wasn't
3474 handled correctly. Handle this properly.
3477 Frequency : always on ppc64
3479 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3480 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3481 perform endianness conversion.
3484 Frequency : rarely (truncate to non-zero file size after write under load)
3485 Bugzilla : 10730, 10687
3486 Description: Files padded with zeros to next 4K multiple
3487 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3488 it is possible that files that are truncated to a non-zero size
3489 immediately after being written are filled with zero bytes beyond
3490 the truncated size. No file data is lost.
3492 Severity : enhancement
3494 Description: Allow recovery/failover for liblustre clients.
3495 Details : liblustre clients were unaware of failover configurations until
3498 Severity : enhancement
3500 Description: user file locks should fail when not mounting with flock option
3501 Details : Set up an error-returning stub in ll_file_operations.lock field
3502 to prevent incorrect behaviour when client is mounted without
3503 flock option. Also, set up properly f_op->flock field for
3507 Frequency : always on ia64
3509 Description: "lfs df" loops on printing out MDS statfs information
3510 Details : The obd_ioctl_data was not initialized and in some systems
3511 this caused a failure during the ioctl that did not return
3512 an error. Initialize the struct and return an error on failure.
3515 Frequency : SLES 9 only
3517 Description: Error of copying files with lustre special EAs as root
3518 Details : Client side always return success for setxattr call for lustre
3519 special xattr (currently only "trusted.lov").
3522 Frequency : rarely on clusters with both ia64+i386 clients
3524 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3525 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3526 clients are doing concurrent writes to the same file it is
3527 possible that the ia64 clients may LASSERT because the OST
3528 extent locks are not PAGE_SIZE aligned. Ensure that grown
3529 locks are always aligned on the request boundary.
3532 Frequency : specific use, occasional
3534 Description: Overwriting in use executable truncates on-disk binary image
3535 Details : If one node attempts to overwrite an executable in use by
3536 another node, we now correctly return ETXTBSY instead of
3537 truncating the file.
3539 Severity : enhancement
3541 Description: Async OSC create to avoid the blocking unnecessarily.
3542 Details : If a OST has no remain object, system will block on the creating
3543 when need to create a new object on this OST. Now, ways use
3544 pre-created objects when available, instead of blocking on an
3545 empty osc while others are not empty. If we must block, we block
3546 for the shortest possible period of time.
3551 Description: chmod on Lustre root is propagated to other clients
3552 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3553 invalid by the follow_mount time.
3556 Frequency : liblustre clients only
3558 Description: Race in 'instant cancel' lock handling could lead to such locks
3559 never to be granted in case of SMP MDS
3560 Details : Do not destroy not yet granted but cbpending locks in
3564 Frequency : replay/resend of open
3566 Description: non null lock assetion failure in mds_intent_policy
3567 Details : Trying to replay/resend lockless open requests resulted in
3568 mds_open() returning 0 with no lock. Now it sets a flag if
3569 a lock is going to be returned.
3571 Severity : enhancement
3573 Description: Checksum enhancements
3574 Details : New checksum enhancements allow for resending RPCs that failed
3577 Severity : enhancement
3579 Description: Tunables on number of dirty pages in cacche
3580 Details : Allow to set limit on number of dirty pages cached.
3585 Description: client crash on unmount - lock still has references
3586 Details : In some error handling cases it was possible to leak a lock
3587 reference on a client while accessing a file. This was not
3588 harmful to the client during operation, but would cause the
3589 client to crash when the filesystem is unmounted.
3592 Frequency : specific case, rare
3594 Description: ETXTBSY on mds though file not in use by client
3595 Details : ETXTBSY is no longer incorrectly returned when attempting to
3596 chmod or chown a directory that the user previously tried to
3597 execute or a currently-executing binary.
3600 Frequency : extremely rare except on liblustre-based clients
3602 Description: Lustre space not freed when files are deleted
3603 Details : Clean up open-unlinked files after client eviction. Previously
3604 the unlink was skipped and the files remained as orphans.
3609 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3610 Details : In some cases it was possible to send a blocking callback to
3611 a client doing a glimpse, even though that client didn't get
3612 a lock granted. When the glimpse lock is cancelled on the OST
3613 the freed lock is left on the waiting list and corrupted the list.
3616 Frequency : all core dumps
3618 Description: Broke core dumps to lustre
3619 Details : Negative dentry may be unhashed if parent does not have UPDATE
3620 lock, but some callers, e.g. do_coredump, expect dentry to be
3621 hashed after successful create, hash it in ll_create_it.
3623 ------------------------------------------------------------------------------
3625 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3627 * Support for kernels:
3628 2.6.9-42.0.2.EL (RHEL 4)
3629 2.6.5-7.276 (SLES 9)
3630 2.4.21-40.EL (RHEL 3)
3631 2.6.12.6 vanilla (kernel.org)
3635 Frequency : always on RHEL 3
3637 Description: Number of open files grows over time
3638 Details : The number of open files grows over time, whether or not
3639 Lustre is started. This was due to a filp leak introduced
3640 by one of our kernel patches.
3642 ------------------------------------------------------------------------------
3644 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3646 * Support for kernels:
3647 2.6.9-42.EL (RHEL 4)
3648 2.6.5-7.276 (SLES 9)
3649 2.4.21-40.EL (RHEL 3)
3650 2.6.12.6 vanilla (kernel.org)
3655 Bugzilla : 5719, 9635, 9792, 9684
3656 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3657 Details : If a server is under heavy load and cannot reply to new
3658 connection requests before the client resends the (re)connect,
3659 the connection handling code can behave badly if two service
3660 threads are concurrently handing separate (re)connections from
3661 the same client. Add better locking to the connection handling
3662 code, and ensure that only a single connection will be processed
3663 for a given client UUID, even if the lock is dropped.
3665 Severity : enhancement
3667 Description: add TCP zero-copy support to kernel
3668 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3669 sends if the hardware supports scatter-gather and checksumming.
3670 This allows socklnd to do client-write and server-read more
3671 efficiently and reduce CPU utilization from skbuf copying.
3674 Frequency : only if NFS exporting from client
3676 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3677 Details : If setting attributes on a file created under NFS that had
3678 never been opened it would be possible to oops the client
3679 if the file had no objects.
3682 Frequency : always for liblustre
3684 Description: liblustre client does MDS+OSTs setattr RPC for each write
3685 Details : When doing a write from a liblustre client, the client
3686 incorrectly issued an RPC to the MDS and each OST the file was
3687 striped over in order to update the timestamps. When writing
3688 with small chunks and many clients this could overwhelm the MDS
3689 with RPCs. In all cases it would slow down the write because
3690 these RPCs are unnecessary.
3692 Severity : enhancement
3694 Description: allow number of MDS service threads to be changed at module load
3695 Details : It is now possible to change the number of MDS service threads
3696 running. Adding "options mds mds_num_threads={N}" to the MDS's
3697 /etc/modprobe.conf will set the number of threads for the next
3698 time Lustre is restarted (assuming the "mds" module is also
3699 reloaded at that time). The default number of threads will
3700 stay the same, 32 for most systems.
3705 Description: OST crash if filesystem is unformatted or corrupt
3706 Details : If an OST is started on a device that has never been formatted
3707 or if the filesystem is corrupt and cannot even mount then the
3708 error handling cleanup routines would dereference a NULL pointer.
3713 Description: NULL pointer deref in llap_from_page.
3714 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3715 incorrect) mapping if the page was truncated/reclaimed while it was
3716 searched for. Check for this condition and skip such pages when
3717 doing readahead. Introduce extra check to llap_from_page() to
3718 verify page->mapping->host is non-NULL (so page is not anonymous).
3721 Frequency : Sometimes when using sys_sendfile
3723 Description: "page not covered by a lock" warnings from ll_readpage
3724 Details : sendfile called ll_readpage without right page locks present.
3725 Now we introduced ll_file_sendfile that does necessary locking
3726 around call to generic_file_sendfile() much like we do in
3730 Frequency : with certain MDS communication failures at client mount time
3732 Description: NULL pointer deref after failed client mount
3733 Details : a client connection request may delayed by the network layer
3734 and not be sent until after the PTLRPC layer has timed out the
3735 request. If the client fails the mount immediately it will try
3736 to clean up before the network times out the request. Add a
3737 reference from the request import to the obd device and delay
3738 the cleanup until the network drops the request.
3741 Frequency : occasionally during client (re)connect
3743 Description: assertion failure during client (re)connect
3744 Details : processing a client connection request may be delayed by the
3745 client or server longer than the client connect timeout. This
3746 causes the client to resend the connection request. If the
3747 original connection request is replied in this interval, the
3748 client may trip an assertion failure in ptlrpc_connect_interpret()
3749 which thought it would be the only running connect process.
3752 Frequency : only with obd_echo servers and clients that are rebooted
3754 Description: kernel BUG accessing uninitialized data structure
3755 Details : When running an obd_echo server it did not start the ping_evictor
3756 thread, and when a client was evicted an uninitialized data
3757 structure was accessed. Start the ping_evictor in the RPC
3758 service startup instead of the OBD startup.
3760 Severity : enhancement
3761 Bugzilla : 10193 (patchless)
3762 Description: Remove dependency on various unexported kernel interfaces.
3763 Details : No longer need reparent_to_init, exit_mm, exit_files,
3764 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3767 Frequency : rare (only users of deprecated and unsupported LDAP config)
3769 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3770 Details : LDAP apparently contains 'lustreName' attributes instead of
3771 'name'. A simple remapping of the name is sufficient.
3774 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3776 Description: waiting_locks_callback trips kernel BUG if client is evicted
3777 Details : Running with the dump_on_timeout debug flag turned on makes
3778 it possible that the waiting_locks_callback() can try to dump
3779 the Lustre kernel debug logs from an interrupt handler. Defer
3780 this log dumping to the expired_lock_main() thread.
3782 Severity : enhancement
3784 Description: Support NFS exporting on 2.6 kernels.
3785 Details : Implement non-rawops metadata methods for NFS server to use without
3786 changing NFS server code.
3789 Frequency : very rare (synthetic metadata workload only)
3791 Description: two racing renames might cause an MDS thread to deadlock
3792 Details : Running the "racer" program may cause one MDS thread to rename
3793 a file from being the source of a rename to being the target of
3794 a rename at exactly the same time that another thread is doing
3795 so, and the second thread has already enqueued these locks after
3796 doing a lookup of the target and is trying to relock them in
3797 order. Ensure that we don't try to re-lock the same resource.
3800 Frequency : only very large systems with liblustre clients
3802 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3803 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3804 all OSTs in parallel. This allows the eviction of stale liblustre
3805 clients to proceed much faster than if they were done in series,
3806 and also offers similar improvements for other set_info RPCs.
3811 Description: excessive CPU usage during initial read phase on client
3812 Details : During the initial read phase on a client, it would agressively
3813 retry readahead on the file, consuming too much CPU and impacting
3814 performance (since 1.4.5.8). Improve the readahead algorithm
3815 to avoid this, and also improve some other common cases (read
3816 of small files in particular, where "small" is files smaller than
3817 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3822 Description: MDS crash when receiving packet with unknown intent.
3823 Details : Do not LBUG in unknown intent case, just return -EFAULT
3825 Severity : enhancement
3826 Bugzilla : 9293, 9385
3827 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3828 Details : Do not serialize getattr (non-intent version) and statfs.
3831 Frequency : occasional, when OST network is overloaded/intermittent
3833 Description: client evicted by OST after bulk IO timeout
3834 Details : If a client sends a bulk IO request (read or write) the OST
3835 may evict the client if it is unresposive to its data GET/PUT
3836 request. This is incorrect if the network is overloaded (takes
3837 too long to transfer the RPC data) or dropped the OST GET/PUT
3838 request. There is no need to evict the client at all, since
3839 the pinger and/or lock callbacks will handle this, and the
3840 client can restart the bulk request.
3843 Frequency : Always when mmapping file with no objects
3845 Description: client crashes when mmapping file with no objects
3846 Details : Check that we actually have objects in a file before doing any
3847 operations on objects in ll_vm_open, ll_vm_close and
3853 Description: Request leak when working with deleted CWD
3854 Details : Introduce advanced request refcount tracking for requests
3855 referenced from lustre intent.
3857 Severity : Enhancement
3859 Description: Cache open file handles on client.
3860 Details : MDS now will return special lock along with openhandle, if
3861 requested and client is allowed to hold openhandle, even if unused,
3862 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3863 closing files for every read/write openration.
3865 Severity : Enhancement
3867 Description: Cache open negative dentries on client when possible.
3868 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3869 negative dentries on lock revocation.
3874 Description: Remounting a client read-only wasn't possible with a zconf mount
3875 Details : It wasn't possible to remount a client read-only with llmount.
3877 Severity : enhancement
3878 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3879 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3880 MPICH is included to add Lustre-specific ADIO interfaces.
3881 This is based closely on the UFS ADIO layer and only differs
3882 in file creation, in order to allow the OST striping to be set.
3883 This is user-contributed code and not supported by CFS.
3888 Description: extended inode attributes (immutable, append-only) work improperly
3889 when 2.4 and 2.6 kernels are used on client/server or vice versa
3890 Details : Introduce kernel-independent values for these flags.
3892 Severity : enhancement
3895 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3896 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3897 granularity. Now, Fractional MB granularity is be supported,
3898 this is very useful for low memory system.
3900 Severity : enhancement
3902 Description: Getattr by fid
3903 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3904 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3905 have locks and avoids getattr rpc requests for it on every stat.
3908 Frequency : Always, for filesystems larger than 2TB
3910 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3911 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3912 filesystems larger than 2TB to be mounted with mballoc enabled.
3915 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3917 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3918 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3919 filesystems larger than 2TB is mounted then the kernel will
3920 silently corrupt the start of the filesystem. CONFIG_LBD is
3921 enabled for all CFS-supported kernels, but the possibility of
3922 this happening with a modified kernel config exists.
3924 Severity : enhancement
3926 Description: add client O_DIRECT support for 2.6 kernels
3927 Details : It is now possible to do O_DIRECT reads and writes to files
3928 in the Lustre client mountpoint on 2.6 kernel clients.
3930 Severity : enhancement
3932 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3933 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3934 parallel, not waiting for response from every OST before sending
3935 a rpc to the next OST.
3940 Description: setattr vs write race when updating file timestamps
3941 Details : Client processes that update a file timestamp into the past
3942 right after writing to the file (e.g. tar) it is possible that
3943 the updated file modification time can be reset to the current
3944 time due to a race between processing the setattr and write RPC.
3946 Severity : enhancement
3948 Description: Bring 'lfs find' closer in line with regular Linux find.
3949 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3950 -print0 options and obtains all the needed info through the lustre
3953 Severity : enhancement
3955 Description: support up to 1024 configured devices on one node
3956 Details : change obd_dev array from statically allocated to dynamically
3957 allocated structs as they are first used to reduce memory usage
3962 Description: Flush dirty partially truncated pages during truncate
3963 Details : Immediatelly flush partially truncated pages in filter_setattr,
3964 this way we completely avoid having any pages in page cache on OST
3965 and can retire ugly workarounds during writes to flush such pages.
3970 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3971 Details : protect inode from truncation within vfs_unlink() context
3972 just take a reference before calling vfs_unlink() and release it
3973 when parent's i_sem is free.
3978 Description: last_id value checked outside lock on OST caused LASSERT failure
3979 Details : If there were multiple MDS->OST object precreate requests in
3980 flight, it was possible that the OST's last object id was checked
3981 outside a lock and incorrectly tripped an assertion. Move checks
3982 inside locks, and discard old precreate requests.
3985 Frequency : always, if extents are used on OSTs
3987 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3988 Details : OSTs using the extents format would not zero the high 16 bits of
3989 the index physical block number. This is not a problem for any
3990 OST filesystems smaller than 16TB, and no kernels support ext3
3991 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3992 new/modified files) and can be fixed for existing filesystems
3993 with e2fsprogs-1.39-cfs1.
3998 Description: import connection selection may be incorrect if timer wraps
3999 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
4000 import connections to be ignored if the 32-bit jiffies counter
4001 wraps. Use a 64-bit jiffies counter.
4004 Frequency : very large clusters immediately after boot
4006 Description: LNET request buffers exhausted under heavy short-term load
4007 Details : If a large number of client requests are generated on a service
4008 that has previously never seen so many requests it is possible
4009 that the request buffer growth cannot keep up with the spike in
4010 demand. Instead of dropping incoming requests, they are held in
4011 the LND until the RPC service can accept more requests.
4014 Frequency : Sometimes during replay
4016 Description: Assertion failure in ll_local_open after replay.
4017 Details : If replay happened on an open request reply before we were able
4018 to set replay handler, reply will become not swabbed tripping the
4019 assertion in ll_local_open. Now we set the handler right after
4020 recognising of open request
4023 Frequency : very rare
4025 Description: kernel reports "badness in vsnprintf"
4026 Details : Reading from the "recovery_status" /proc file in small chunks
4027 may cause a negative length in lprocfs_obd_rd_recovery_status()
4028 call to vsnprintf() (which is otherwise harmless). Exit early
4029 if there is no more space in the output buffer.
4031 Severity : enhancement
4033 Description: clear OBD RPC statistics by writing to them
4034 Details : It is now possible to clear the OBD RPC statistics by writing
4035 to the "stats" file.
4040 Description: Client mtime is not the same on different clients after utimes
4041 Details : In some cases, the client was using the utimes() syscall on
4042 a file cached on another node. The clients now validate the
4043 ctime from the MDS + OSTs to determine which one is right.
4048 Description: Inability to activate failout mode
4049 Details : lconf script incorrectly assumed that in pythong string's numeric
4050 value is used in comparisons.
4053 Frequency : always with multiple stripes per file
4055 Description: Inefficient object allocation for mutli-stripe files
4056 Details : When selecting which OSTs to stripe files over, for files with
4057 a stripe count that divides evenly into the number of OSTs,
4058 the MDS is always picking the same starting OST for each file.
4059 Return the OST selection heuristic to the original design.
4064 Description: mount failures may take full timeout to return an error
4065 Details : Under some heavy load conditions it is possible that a
4066 failed mount can wait for the full obd_timeout interval,
4067 possibly several minutes, before reporting an error.
4068 Instead return an error as soon as the status is known.
4070 Frequency : quota enabled and large files being deleted
4072 Description: releasing more than 4GB of quota at once hangs OST
4073 Details : If a user deletes more than 4GB of files on a single OST it
4074 will cause the OST to spin in an infinite loop. Release
4075 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4080 Description: statfs data retrieved from /proc may be stale or zero
4081 Details : When reading per-device statfs data from /proc, in the
4082 {kbytes,files}_{total,free,avail} files, it may appear
4083 as zero or be out of date.
4086 Frequency : systems with MD RAID1 external journal devices
4088 Description: lconf's call to blkid is confused by RAID1 journal devices
4089 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4090 of returning all block devices that match the journal UUID.
4093 Frequency : always, for aggregate stripe size over 4GB
4095 Description: assertion fails when trying to use 4GB stripe size
4096 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
4097 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
4100 Frequency : always on ppc64
4102 Description: the first write on an ext3 filesystem with mballoc got stuck
4103 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4104 perform endianness conversion.
4106 ------------------------------------------------------------------------------
4108 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
4110 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4111 INTEROPERATE with older versions automatically. Please read the
4112 user documentation before upgrading any part of a live system.
4113 * WARNING: Lustre networking configuration changes are required with
4114 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4117 * Support for kernels:
4118 2.6.9-22.0.2.EL (RHEL 4)
4119 2.6.5-7.244 (SLES 9)
4120 2.6.12.6 vanilla (kernel.org)
4123 Severity : enhancement
4124 Bugzilla : 7981/8208
4125 Description: Introduced Lustre Networking (LNET)
4126 Details : LNET is new networking infrastructure for Lustre, it includes
4127 a reorganized network configuration mode (see the user
4128 documentation for full details) as well as support for routing
4129 between different network fabrics. Lustre Networking Devices
4130 (LNDs) for the supported network fabrics have also been
4131 created for this new infrastructure.
4133 Severity : enhancement
4134 Description: Introduced Access control lists
4135 Details : clients can set ACLs on files and directories in order to have
4136 more fine-grained permissions than the standard Unix UGO+RWX.
4137 The MDS must be started with the "-o acl" mount option.
4139 Severity : enhancement
4140 Description: Introduced filesystem quotas
4141 Details : Administrators may now establish per-user quotas on the
4144 Severity : enhancement
4146 Description: Configuration change for the XT3
4147 The PTLLND is now used to run Lustre over Portals on the XT3
4148 The configure option(s) --with-cray-portals are no longer used.
4149 Rather --with-portals=<path-to-portals-includes> is used to
4150 enable building on the XT3. In addition to enable XT3 specific
4151 features the option --enable-cray-xt3 must be used.
4156 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4157 Details : A race between allocating a new llcd and re-getting the llcd_lock
4158 allowed another thread to grab newly-allocated llcd.
4160 Severity : enhancement
4162 Description: 2.6 OST async journal commit and locking fix to improve performance
4163 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4164 2.6 kernels are now async as they already were in 2.4 kernels so
4165 that they can commit concurrently with the network bulk transfer.
4166 For block-allocated files the filter allocation semaphore is held
4167 to avoid filesystem fragmentation during allocation. BKL lock
4168 removed for 2.6 xattr operations where it is no longer needed.
4173 Description: lconf incorrectly determined whether two IP networks could talk
4174 Details : In some more complicated routing and multiple-network
4175 configurations, lconf will avoid trying to make a network
4176 connection to a disjoint part of the IP space. It was doing the
4177 math incorrectly for one set of cases.
4182 Description: Fix for potential infinite loop processing records in an llog.
4183 Details : If an llog record is corrupted/zeroed, it is possible to loop
4184 forever in llog_process(). Validate the llog record length
4185 and skip the remainder of the block on error.
4188 Frequency : occasional (liblustre only)
4190 Description: liblustre could not open files whose last component is a symlink
4191 Details : sysio_path_walk() would incorrectly pass the open intent to
4192 intermediate path components.
4195 Frequency : rare (liblustre only with non-standard tuning)
4196 Bugzilla : 7201 (7350)
4197 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4198 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4199 after it has been released, only do this for referenced locks
4201 Severity : enhancement
4203 Description: specifying an (invalid) directory default stripe_size of -1
4204 would reset the directory default striping
4205 Details : stripe_size -1 was used internally to signal directory stripe
4206 removal, now use "all default" to signal dir stripe removal
4207 as a directory striping of "all default" is not useful
4210 Frequency : common for large clusters running liblustre clients
4212 Description: doing an ls when liblustre clients are running is slow
4213 Details : sending a glimpse AST to a liblustre client waits for every AST
4214 to time out, as liblustre clients will not respond. Since they
4215 cannot cache data we refresh the OST lock LVB from disk instead.
4217 Severity : enhancement
4219 Description: doing an ls at the same time as file IO can be slow
4220 Details : enqueue and other "small" requests can be blocked behind many
4221 large IO requests. Create a new OST IO portal for non-IO
4222 requests so they can be processed faster.
4225 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4227 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4228 created file with mode 000
4229 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4230 allow an ftruncate() as a non-root user to a file with mode 000.
4231 Lustre now respects this flag to disable mode checks when
4232 truncating a file owned by the user
4235 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4238 Description: Revoking locks from clients that went dead or catatonic might take
4240 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4241 cancellation of such locks instant on servers without waiting for
4242 any reply from clients. Clients drops these locks when cancel
4243 notification from server is received without replying.
4246 Frequency : liblustre-only, when liblustre client dies or becomes busy
4248 Description: Doing ls on Linux clients can take a long time with active
4250 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4251 granting such locks to it in the first place if possible. Locks
4252 are taken by proxy on the OST during the read or write and
4253 dropped immediately afterward. Add connect flags handling, do
4254 not grant locks to liblustre clients for glimpse ASTs.
4256 Severity : enhancement
4258 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4259 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4260 sequential reads and behaves badly with applications doing random
4261 reads. Improve it by reading ahead at least read region, and
4262 avoiding excessive large RPC for small reads.
4264 Severity : enhancement
4266 Description: Creating more than 1000 files for a single job may cause a load
4267 imbalance on the OSTs if there are also a large number of OSTs.
4268 Details : qos_prep_create() uses an OST index reseed value that is an
4269 even multiple of the number of available OSTs so that if the
4270 reseed happens in the middle of the object allocation it will
4271 still utilize the OSTs as uniformly as possible.
4276 Description: OST or MDS may oops in ping_evictor_main()
4277 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4278 but doesn't restart at beginning of obd_exports_timed list
4281 Severity : enhancement
4283 Description: improve by-nid export eviction on the MDS and OST
4284 Details : allow multiple exports with the same NID to be evicted at one
4285 time without re-searching the exports list.
4288 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4290 Description: MDS may oops in groups_free()
4291 Details : in rare race conditions a newly allocated group_info struct is
4292 freed again, and this can be NULL. The 2.4 compatibility code
4293 for groups_free() checked for a NULL pointer, but 2.6 did not.
4296 Frequency : common for liblustre clients doing little filesystem IO
4297 Bugzilla : 9352, 7313
4298 Description: server may evict liblustre clients accessing contended locks
4299 Details : if a client is granted a lock or receives a completion AST
4300 with a blocking AST already set it would not reply to the AST
4301 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4304 Frequency : lfs setstripe, only systems with more than 160 OSTs
4306 Description: unable to set striping with a starting offset beyond OST 160
4307 Details : llapi_create_file() incorrectly limited the starting stripe
4308 index to the maximum single-file stripe count.
4311 Frequency : LDAP users only
4313 Description: lconf did not handle in-kernel recovery with LDAP properly
4314 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4316 Severity : enhancement
4318 Description: bind OST threads to NUMA nodes to improve performance
4319 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4320 node and do their allocations there to localize memory access
4322 Severity : enhancement
4324 Description: llmount can determine client NID directly from Myrinet (GM)
4325 Details : the client NID code from gmnalnid was moved directly into
4326 llmount, removing the need to use this or specifying the
4327 client NID explicitly when mounting GM clients with zeroconf
4330 Frequency : if client is started with down MDS
4332 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4333 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4334 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4335 ptlrpc_import_delayed_req() because we didn't check if the
4336 request was interrupted, and we also didn't break out of the
4337 event loop if there was no timeout
4342 Description: data loss during non-page-aligned writes to a single file from
4343 both multiple nodes and multiple threads on one node at same time
4344 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4345 inconsistency led to false short-reads, that were cached and later
4346 used by ->prepare_write() to fill in partially written page,
4347 leading to data loss.
4350 Frequency : always, if lconf --abort_recovery used
4352 Description: lconf --abort_recovery fails with 'Operation not supported'
4353 Details : lconf was attempting to abort recovery on the MDT device and not
4356 Severity : enhancement
4358 Description: remove cleanup logs
4359 Details : replace lconf-generated cleanup logs with lustre internal
4360 cleanup routines. Eliminates the need for client-cleanup and
4363 Severity : enhancement
4365 Description: add support for EAs (user and system) on lustre filesystems
4366 Details : it is now possible to store extended attributes in the Lustre
4367 client filesystem, and with the user_xattr mount option it
4368 is possible to allow users to store EAs on their files also
4370 Severity : enhancement
4372 Description: Add possibility (config option) to show minimal available OST free
4374 Details : When compiled with --enable-mindf configure option, statfs(2)
4375 (and so, df) will return least minimal free space available from
4376 all OSTs as amount of free space on FS, instead of summary of
4377 free spaces of all OSTs.
4379 Severity : enhancement
4381 Description: do not expand extent locks acquired on OST-side
4382 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4383 by server: they are not cached anyway.
4386 Frequency : when mmap is used/binaries executed from Lustre
4388 Description: Unmmap pages before throwing them away from read cache.
4389 Details : llap_shrink cache now attempts to unmap pages before discarding
4390 them (if unmapping failed - do not discard). SLES9 kernel has
4391 extra checks that trigger if this unmapping is not done first.
4396 Description: lconf didn't resolve symlinks before checking to see whether a
4397 given mountpoint was already in use
4400 Frequency : when migrating failover services
4401 Bugzilla : 6395, 9514
4402 Description: When migrating a subset of services from a node (e.g. failback
4403 from a failover service node) the remaining services would
4404 time out and evict clients.
4405 Details : lconf --force (implied by --failover) sets the global obd_timeout
4406 to 5 seconds in order to quickly disconnect, but this caused
4407 other RPCs to time out too quickly. Do not change the global
4408 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4410 Severity : enhancement
4411 Frequency : if MDS is started with down OST
4412 Bugzilla : 9439,5706
4413 Description: Allow startup/shutdown of an MDS without depending on the
4414 availability of the OSTs.
4415 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4416 Add appropriate locking and lov-osc refcounts for safe
4417 cleaning. Add osc abort_inflight calls in case the
4418 synchronize never started.
4421 Frequency : occasional (Cray XT3 only)
4423 Description: root not authorized to access files in CRAY_PORTALS environment
4424 Details : The client process capabilities were not honoured on the MDS in
4425 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4426 been accessed by an authorized user then root was able to access
4427 the file on the local client also. The root user capabilities
4428 are now allowed on the MDS, as this environment has secure UID.
4431 Frequency : occasional
4433 Description: ldiskfs "too long searching" message happens too often
4434 Details : A debugging message (otherwise harmless) prints too often on
4435 the OST console. This has been reduced to only happen when
4436 there are fragmentation problems on the filesystem.
4441 Description: Division by zero in statfs when all OSCs are inactive
4442 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4443 lov_statfs divides by value returned by lov_get_stripecnt().
4447 Bugzilla : 9489, 3273
4448 Description: First write from each client to each OST was only 4kB in size,
4449 to initialize client writeback cache, which caused sub-optimal
4450 RPCs and poor layout on disk for the first writen file.
4451 Details : Clients now request an initial cache grant at (re)connect time
4452 and so that they can start streaming writes to the cache right
4453 away and always do full-sized RPCs if there is enough data.
4454 If the OST is rebooted the client also re-establishes its grant
4455 so that client cached writes will be honoured under the grant.
4460 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4461 Details : Now I/O RPCs go to different portal number and (presumably) fast
4462 lock requests (and glimses) and other RPCs get their own service
4463 threads pool that should be able to service those RPCs
4466 Severity : enhancement
4468 Description: Ability to exchange lustre version between client and servers and
4469 issue warnings at client side if client is too old. Also for
4470 liblustre clients there is ability to refuse connection of too old
4472 Details : New 'version' field is added to connect data structure that is
4473 filled with version info. That info is later checked by server and
4477 Frequency : rare, liblustre only.
4478 Bugzilla : 9296, 9581
4479 Description: Two simultaneous writes from liblustre at offset within same page
4480 might proceed at the same time overwriting eachother with stale
4482 Details : I/O lock withing llu_file_prwv was released too early, before data
4483 actually was hitting the wire. Extended lock-holding time until
4484 server acknowledges receiving data.
4487 Frequency : extremely rare. Never observed in practice.
4489 Description: avoid generating lustre_handle cookie of 0.
4490 Details : class_handle_hash() generates handle cookies by incrementing
4491 global counter, and can hit 0 occasionaly (this is unlikely, but
4492 not impossible, because initial value of cookie counter is
4493 selected randonly). Value of 0 is used as a sentinel meaning
4494 "unassigned handle" --- avoid it. Also coalesce two critical
4495 sections in this function into one.
4497 Severity : enhancement
4499 Description: allow liblustre clients to delegate truncate locking to OST
4500 Details : To avoid overhead of locking, liblustre client instructs OST to
4501 take extent lock in ost_punch() on client's behalf. New connection
4502 flag is added to handle backward compatibility.
4504 Severity : enhancement
4505 Bugzilla : 4928, 7341, 9758
4506 Description: allow number of OST service threads to be specified
4507 Details : a module parameter allows the number of OST service threads
4508 to be specified via "options ost ost_num_threads={N}" in the
4509 OSS's /etc/modules.conf or /etc/modprobe.conf.
4513 Bugzilla : 6146, 9635, 9895
4514 Description: servers crash with bad pointer in target_handle_connect()
4515 Details : In rare cases when a client is reconnecting it was possible that
4516 the connection request was the last reference for that export.
4517 We would temporarily drop the export reference and get a new
4518 one, but this may have been the last reference and the export
4519 was just destroyed. Get new reference before dropping old one.
4521 Severity : enhancement
4522 Frequency : if client is started with failover MDS
4524 Description: Allow multiple MDS hostnames in the mount command
4525 Details : Try to read the configuration from all specified MDS
4526 hostnames during a client mount in case the "primary"
4529 Severity : enhancement
4531 Description: Stop sending data to evicted clients as soon as possible.
4532 Details : Check if the client we are about to send or are sending data to
4533 was evicted already. (Check is done every second of waiting,
4534 for which l_wait_event interface was extended to allow checking
4535 of exit condition at specified intervals).
4538 Frequency : rare, normally only when NFS exporting is done from client
4540 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4542 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4545 Frequency : always, if srand() is called before liblustre initialization
4547 Description: Liblustre uses system PRNG disturbing its usage by user application
4548 Details : Introduce internal to lustre fast and high-quality PRNG for
4549 lustre usage and make liblustre and some other places in generic
4550 lustre code to use it.
4552 Severity : enhancement
4553 Bugzilla : 9477, 9557, 9870
4554 Description: Verify that the MDS configuration logs are updated when xml is
4555 Details : Check if the .xml configuration logs are newer than the config
4556 logs stored on the MDS and report an error if this is the case.
4557 Request --write-conf, or allow starting with --old_conf.
4559 Severity : enhancement
4561 Description: Handle symlinks in the path when checking if Lustre is mounted.
4562 Details : Resolve intermediate symlinks when checking if a client has
4563 mounted a filesystem to avoid duplicate client mounts.
4568 Description: lconf can hit an error exception but still return success.
4569 Details : The lconf command catches the Command error exception at the top
4570 level script context and will exit with the associated exit
4571 status, but doesn't ensure that this exit status is non-zero.
4576 Description: failure of ptlrpc thread startup can cause oops
4577 Details : Starting a ptlrpc service thread can fail if there are a large
4578 number of threads or the server memory is very fragmented.
4579 Handle this without oopsing.
4582 Frequency : always, only if liblustre and non-default acceptor port was used
4584 Description: liblustre cannot connect to servers with non-default acceptor port
4585 Details : tcpnal_set_default_params() was not called and was therefore
4586 ignoring the environment varaible TCPNAL_PORT, as well as other
4587 TCPNAL_ environment variables
4592 Description: two objects could be created on the same OST for a single file
4593 Details : If an OST is down, in some cases it was possible to create two
4594 objects on a single OST for a single file. No problems other
4595 than potential performance impact and spurious error messages.
4599 Bugzilla : 5681, 9562
4600 Description: Client may oops in ll_unhash_aliases
4601 Details : Client dcache may become inconsistent in race condition.
4602 In some cases "getcwd" can fail if the current directory is
4608 Description: Inode refcounting problems in NFS export code
4609 Details : link_raw functions used to call d_instantiate without obtaining
4610 extra inode reference first.
4614 Bugzilla : 9942, 9903
4615 Description: Referencing freed requests leading to crash, memleaks with NFS.
4616 Details : We used to require that call to ll_revalidate_it was always
4617 followed by ll_lookup_it. Also with revalidate_special() it is
4618 possible to call ll_revalidate_it() twice for the same dentry
4619 even if first occurence returned success. This fix changes semantic
4620 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4621 reference on a request referred from the intent.
4622 ll_intent_release() then releases such a request.
4625 Frequency : rare, normally benchmark loads only
4627 Description: unlinked inodes were kept in memory on the client
4628 Details : If a client is repeatedly creating and unlinking files it
4629 can accumulate a lot of stale inodes in the inode slab cache.
4630 If there is no other client load running this can cause the
4631 client node to run out of memory. Instead flush old inodes
4632 from client cache that have the same inode number as a new inode.
4635 Frequency : SLES9 2.6.5 kernel and long filenames only
4636 Bugzilla : 9969, 10379
4637 Description: utime reports stale NFS file handle
4638 Details : SLES9 uses out-of-dentry names in some cases, which confused
4639 the lustre dentry revalidation. Change it to always use the
4643 Frequency : rare, unless heavy write-truncate concurrency is continuous
4644 Bugzilla : 4180, 6984, 7171, 9963, 9331
4645 Description: OST becomes very slow and/or deadlocked during object unlink
4646 Details : filter_destroy() was holding onto the parent directory lock
4647 while truncating+unlinking objects. For very large objects this
4648 may block other threads for a long time and slow overall OST
4649 responsiveness. It may also be possible to get a lock ordering
4650 deadlock in this case, or run out of journal credits because of
4651 the combined truncate+unlink. Solution is to do object truncate
4652 first in one transaction without parent lock, and then do the
4653 final unlink in a new transaction with the parent lock. This
4654 reduces the lock hold time dramatically.
4657 Frequency : rare, 2.4 kernels only
4659 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4660 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4661 before unmounting their filesystem in order to avoid deadlock.
4662 The kernel_locked() function in 2.4 kernels only checks whether
4663 the kernel lock is held, not whether it is this process that is
4664 holding it as 2.6 kernels do.
4669 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4670 Details : The client ptlrpc code may be trying to reconnect to a down
4671 server before a previous connection attempt has timed out.
4672 Increase the reconnect interval to be longer than the connection
4673 timeout interval to avoid sending duplicate connections to
4677 Frequency : echo_client brw_test command
4679 Description: fix echo_client to work with OST preallocated code
4680 Details : OST preallocation code (5137) didn't take echo_client IO path
4681 into account: echo_client calls filter methods outside of any
4682 OST thread and, hence, there is no per-thread preallocated
4683 pages and buffers to use. Solution: hijack pga pages for IO. As
4684 a byproduct, this avoids unnecessary data copying.
4688 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4689 Description: Client can oops in mdc_commit_close() after open replay
4690 Details : It was possible for the MDS to return an open request with no
4691 transaction number in mds_finish_transno() if the client was
4692 evicted, but without actually returning an error. Clients
4693 would later try to replay that open and may trip an assertion
4694 Simplify the client close codepath, and always return an error
4695 from the MDS in case the open is not successful.
4698 Frequency : rare, 2.6 OSTs only
4700 Description: OST may deadlock under high load on fragmented files
4701 Details : If there was a heavy load and highly-fragmented OST filesystems
4702 it was possible to have all the OST threads deadlock waiting on
4703 allocation of biovecs, because the biovecs were not released
4704 until the entire RPC IO was completed. Instead, release biovecs
4705 as soon as they are complete to ensure forward IO progress.
4707 Severity : enhancement
4709 Description: Support for specifying external journal device at mount
4710 Details : If an OST or MDS device is formatted with an external journal
4711 device, this device major/minor is stored in the ext3 superblock
4712 and may not be valid for failover. Allow detecting and
4713 specifying the external journal at mount time.
4718 Description: Mounting an MDS with pending unlinked files may cause oops
4719 Details : target_finish_recovery() calls mds_postrecov() which returned
4720 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4721 considers this an error and immediately begins cleaning up the
4722 lov, just after starting the mds_lov process
4724 Severity : enhancement
4726 Description: Implement 'lfs df' to report actual free space on per-OST basis
4727 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4728 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4729 usage of objects; '-h' to report in human readable format.
4731 ------------------------------------------------------------------------------
4733 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4740 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4741 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4742 at startup during orphan unlinking. Instead, skip buddy update
4743 before setup, as it will be regenerated after recovery is complete.
4748 Description: If an OST is inactive, its locks might reference stale inodes.
4749 Details : lov_change_cbdata() must iterate over all namespaces, even if
4750 they are inactive to clear inode references from the lock.
4752 Severity : enhancement
4753 Frequency : occasional, if non-standard max_dirty_mb used
4755 Description: Client will block write RPCs if not enough grant
4756 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4757 then the client will block writes while waiting for another RPC
4758 to complete instead of consuming its dirty limit. With change
4759 we get improved performance when max_dirty_mb is small.
4761 Severity : enhancement
4762 Bugzilla : 3389, 6253
4763 Description: Add support for supplementary groups on the MDS.
4764 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4765 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4766 lookups for user supplementary groups into a cache.
4770 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4771 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4772 for liblustre O_EXCL is enforced by MDS.
4776 Description: readdir never returns NULL in liblustre.
4777 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4778 dentry in struct dirent.
4782 Frequency : liblustre only, depends on application IO pattern
4783 Description: liblustre clients evicted if not contacting servers
4784 Details : Don't put liblustre clients into the ping_evictor list, so
4785 they will not be evicted by the pinger ever.
4787 Severity : enhancement
4789 Description: Add ability to evict clients by NID from MDS.
4790 Details : By echoing "nid:$NID" string into
4791 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4792 $NID would be instantly evicted from this MDS and from all active
4793 OSTs connected to it.
4797 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4798 Details : lookup_it_finish() used to query file size from OSTs that was not
4803 Description: service threads change working directory to that of init
4804 Details : Starting lustre service threads may pin the working directory
4805 of the parent thread, making that filesystem busy. Threads
4806 now change to the working directory of init to avoid this.
4810 Frequency : during shutdown only
4811 Description: shutdown with a failed MDS or OST can cause unmount to hang
4812 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4817 Frequency : 2.6 only
4818 Description: chmod/chown may include an extra supplementary group
4819 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4820 supplementary group and if none is specified this is used.
4823 Bugzilla : 5479 (6816)
4825 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4826 Details : If lookup is in progress on a file that is unlinked we might try
4827 to revalidate the inode and fail in revalidate after lookup is
4828 complete and ll_file_open() enqueues the open again but
4829 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4832 Frequency : always, if lconf --abort_recovery used
4834 Description: lconf --abort_recovery fails with 'Operation not supported'
4835 Details : lconf was attempting to abort recovery on the MDT device and not
4838 ------------------------------------------------------------------------------
4840 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4845 Frequency : rare (only unsupported configurations with a node running as an
4847 Bugzilla : 6514, 5137
4848 Description: Mounting a Lustre file system on a node running as an OST could
4850 Details : OSTs now preallocates memory needed to write out data at
4851 startup, instead of when needed, to avoid having to
4852 allocate memory in possibly low memory situations.
4853 Specifically, if the file system is mounted on on OST,
4854 memory pressure could force it to try to write out data,
4855 which it needed to allocate memory to do. Due to the low
4856 memory, it would be unable to do so and the node would
4857 become unresponsive.
4859 Severity : enhancement
4861 Description: Addition of lconf --service command line option
4862 Details : lconf now accepts a '--service <arg>' option, which is
4863 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4865 Severity : enhancement
4867 Description: Failover mode is now the default for OSTs.
4868 Details : By default, OSTs will now run in failover mode. To return to
4869 the old behaviour, add '--failout' to the lmc line for OSTs.
4871 Severity : enhancement
4873 Description: Health checks are now provided for MDS and OSTs
4874 Details : Additional detailed health check information on MSD and OSTs
4875 is now provided through the procfs health_check value.
4878 Frequency : occasional, depends on IO load
4880 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4881 after numerous create/delete cycles
4882 Details : The ext3 inode allocation policy would not allocate new inodes
4883 very well on the OSTs because there are no new directories
4884 being created. Instead we look for groups with free space if
4885 the parent directories are nearly full.
4890 Description: Network or server problems during mount may cause partially
4891 mounted clients instead of returning an error.
4892 Details : The config llog parsing code may overwrite the error return
4893 code during mount error handling, returning success instead
4899 Description: MDS can fail to allocate large reply buffers
4900 Details : After long uptimes the MDS can fail to allocate large reply
4901 buffers (e.g. zconf client mount config records) due to memory
4902 fragmentation or consumption by the buffer cache. Preallocate
4903 some large reply buffers so that these replies can be sent even
4904 under memory pressure.
4908 Frequency : rare (liblustre)
4909 Description: fsx running with liblustre complained that using truncate() to
4910 extend the file doesn't work. This patch corrects that issue.
4911 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4912 ATTR_SIZE and lsm use in llu_setattr_raw.
4916 Frequency : rare, only 2.6 kernels
4917 Description: Unusual file access patterns on the MDS may result in inode
4918 data being lost in very rare circumstances.
4919 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4920 code in the 2.6 kernel caused the inode and/or EA data not to
4921 be read from disk, causing single-file corruption.
4925 Frequency : rare, only 2.6 filesystems using extents
4926 Description: Heavy concurrent write and delete load may cause data corruption.
4927 Details : It was possible under high-load situations to have an extent
4928 metadata block in the block device cache from a just-unlinked
4929 file overwrite a newly-allocated data block. We now unmap any
4930 metadata buffers that alias just-allocated data blocks.
4934 Frequency : filesystems with default stripe_count larger than 77
4935 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4936 Details : lconf specifies an inode size of 4096 bytes when the default
4937 stripe_count is larger than 77. This conflicts with the default
4938 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4939 this case to avoid pinning too much memory for large EAs.
4941 ------------------------------------------------------------------------------
4943 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4948 Frequency : rare (extremely heavy IO load with hundreds of clients)
4950 Description: Client is evicted, gets IO error writing to file
4951 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4952 caused clients to be evicted for AST timeouts. The fixes in
4953 bug 5192 mean we no longer need to have such short AST timeouts
4954 so ldlm_timeout has been increased.
4957 Frequency : occasional during --force or --failover shutdown under load
4958 Bugzilla : 5949, 4834
4959 Description: Server oops/LBUG if stopped with --force or --failover under load
4960 Details : a collection of import/export refcount and cleanup ordering
4961 issues fixed for safer force cleanup
4964 Frequency : only filesystems larger than 120 OSTs
4965 Bugzilla : 5990, 6223
4966 Description: lfs getstripe would oops on a very large filesystem
4967 Details : lov_getconfig used kfree on vmalloc'd memory
4970 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4971 Bugzilla : 6242, 6243
4972 Description: reading from files that had been truncated to a non-zero size
4973 but never opened returned no data
4974 Details : ll_file_read() reads zeros from no-object files to EOF
4979 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4980 Details : There's little harm in aborting MDS/OSS recovery and letting it
4981 try again, so I removed the LASSERT and return an error instead.
4983 Severity : enhancement
4985 Description: New debugging infrastructure for tracking down data corruption
4986 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4987 (b) cover more of the client-side code path, and (c) try to narrow
4988 down where problems occurred
4992 Bugzilla : 3819, 4364, 4397, 6313
4993 Description: Racing close and eviction MDS could cause assertion in mds_close
4994 Details : It was possible to get multiple mfd references during close and
4995 client eviction, leading to one thread referencing a freed mfd.
4997 Severity: : enhancement
4998 Bugzilla : 3262, 6359
4999 Description: Attempts to reconnect to servers are now more aggressive.
5000 Details : This builds on the enhanced upcall-less recovery that was added
5001 in 1.4.2. When trying to reconnect to servers, clients will
5002 now try each server in the failover group every 10 seconds. By
5003 default, clients would previously try one server every 25 seconds.
5008 Description: After recovery, certain operations trigger a failed
5009 assertion on a client.
5010 Details : Failing over an mds, using lconf -d --failover, while a
5011 client was doing a readdir() call would cause the client to
5012 LBUG after recovery completed and the readdir() was resent.
5014 Severity : enhancement
5016 Description: Default groups are now added by lconf
5017 Details : You can now run lconf --group <servicename> without having to
5018 manually add groups with lmc.
5021 Frequency : occasional
5023 Description: Nodes with an elan id of 0 trigger a failed assertion
5026 Frequency : always when accessing e.g. tty/console device nodes
5028 Description: tty and some other devices nodes cannot be used on lustre
5029 Details : file's private_data field is used by device data and lustre
5030 values in there got lost. New field was added to struct file to
5031 store fs-specific private data.
5034 Frequency : when exporting Lustre via NFS
5036 Description: NFSD failed occasionally when looking up a path component
5037 Details : NFSD is looking up ".." which was broken in ext3 directories
5038 that had grown large enough to become hashed.
5041 Frequency : Clusters with multiple interfaces not on the same subnet
5043 Description: Nodes will repeatedly try to reconnect to an interface which it
5044 cannot reach and report an error to the log.
5045 Details : Extra peer list entries will be created by lconf with some peers
5046 unreachable. lconf now validates the peer before adding it.
5049 Frequency : Only if a default stripe is set on the filesystem root.
5051 Description: Setting a default stripe on the filesystem root prevented the
5052 filesystem from being remounted.
5053 Details : The client was sending extra request flags in the root getattr
5054 request and did not allocate a reply buffer for the dir EA.
5057 Frequency : occasional, higher if lots of files are accessed by one client
5058 Bugzilla : 6159, 6097
5059 Description: Client trips assertion regarding lsm mismatch/magic
5060 Details : While revalidating inodes the VFS looks up inodes with ifind()
5061 and in rare cases can find an inode that is being freed.
5062 The ll_test_inode() code will free the lsm during ifind()
5063 when it finds an existing inode and then the VFS later attaches
5064 this free lsm to a new inode.
5068 Bugzilla : 6422, 7030
5069 Description: MDS deadlock between mkdir and client eviction
5070 Details : Creating a new file via mkdir or mknod (starting a transaction
5071 and getting the ns lock) can deadlock with client eviction
5072 (gets ns lock and trying to finish a synchronous transaction).
5075 Frequency : occasional
5076 Description: While starting a server, the fsfilt_ext3 module could not be
5078 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5079 kernels. Previously, lconf would still use the ext3 name
5080 when trying to load modules. Now, it will correctly use
5081 ext3 on 2.4 and ldiskfs on 2.6.
5083 Severity : enhancement
5084 Description: The default stripe count has been changed to 1
5085 Details : The interpretation of the default stripe count (0, to lfs
5086 or lmc) has been changed to mean striping across a single
5087 OST, rather than all available. For general usage we have
5088 found a stripe count of 1 or 2 works best.
5090 Severity : enhancement
5091 Description: Add support for compiling against Cray portals.
5092 Details : Conditional compiling for some areas that are different
5096 Frequency : occasional
5097 Bugzilla : 6409, 6834
5098 Description: Creating files with an explicit stripe count may lead to
5099 a failed assertion on the MDS
5100 Details : If some OSTs are full or unavailable, creating files may
5101 trigger a failed assertion on the MDS. Now, Lustre will
5102 try to use other servers or return an error to the
5106 Frequency : occasional
5108 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5109 caused lock timeout during readahead (since 1.4.2).
5110 Details : Processes doing readahead might match a lock that hasn't been
5111 granted yet if there are overlapping and conflicting lock
5112 requests. The readahead process waits on ungranted lock
5113 (original lock is CBPENDING), while OST waits for that process
5114 to cancel CBPENDING read lock and eventually evicts client.
5116 Severity : enhancement
5118 Description: Initial enabling of flock support for clients
5119 Details : Implements fcntl advisory locking and file status functions.
5120 This feature is provided as an optional mount flag (default
5121 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5122 locking are implemented yet, and those that are are not guaranteed
5123 to be completely correct in production environments.
5124 mount -t lustre -o [flock|noflock] ...
5127 Frequency : occasional
5129 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5130 assertion in the ext3 JBD (journaling) layer.
5131 Details : The b_committed_data struct is protected by the big kernel lock
5132 in 2.4 kernels, serializing journal_commit_transaction() and
5133 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5134 access to this struct. In 2.6 kernels there is finer grained
5135 locking to improve SMP performance of the JBD layer.
5139 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5144 Description: merge and cleanup kernel patches.
5147 -----------------------------------------------------------------------------
5149 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5151 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5152 versions of Lustre. Please update all servers and clients to
5153 version 1.4.2 or later at the same time. You must also run
5154 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5155 to update the configuration logs.
5157 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5158 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5159 - swab llog records of type '0' so we get proper header size/idx (5861)
5160 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5161 - fix rename of one directory over another leaking an inode (5953)
5162 - avoid SetPageDirty on 2.6 (5981)
5163 - don't re-add just-being-destroyed locks to the waiting list (5653)
5164 - when creating new directories, inherit the parent's custom
5165 striping settings if present parent (3048)
5166 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5167 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5168 - don't hold client locks on temporary worklist from l_lru (5666)
5169 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5170 - automatically evict dead clients (5921)
5171 - Update file size properly in create+truncate+fstat case (6196)
5172 - Do not unhash mountpoint dentries, do not allow removal of
5174 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5175 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5176 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5177 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5178 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5179 - Correctly handle reads of files with no objects (6243)
5180 - lctl recover will also mark a device active if deactivate used (5933)
5182 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5183 - allow --write-conf on an MDS with different nettype than client (5619)
5184 - don't write config llogs to MDS for mounts not from that MDS (5617)
5185 - lconf should create multiple TCP connections from a client (5201)
5186 - init scripts are now turned off by default; run chkconfig --on
5187 lustre and chkconfig --on lustrefs to use them
5188 - upcalls are no longer needed for clients to recover to failover
5190 - add --abort-recovery option to lconf to abort recovery on device
5192 - add support for an arbitrary number of OSTs (3026)
5193 - Quota support protocol changes.
5194 - forward compatibility changes to wire structs (6007)
5195 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5196 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5197 - improved "lustre status" script
5198 - initialize blocksize for non-regular files (6062)
5199 - added --disable-server and --disable-client configure options (5782)
5200 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5201 - Vanilla 2.4.29 support
5202 - increase maximum number of obd devices to 520 (6242)
5203 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5204 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5205 SLES 9 2.6.5 kernels
5206 - limit stripes per file to 160 (the maximum EA size) (6093)
5208 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5211 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5212 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5213 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5214 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5215 - small ext3 extents cleanups and fixes (5733)
5216 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5217 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5218 - avoid needless client->OST connect, fix handle mismatch (5317)
5219 - fix DLM error path that led to out-of-sync client, long delays (5779)
5220 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5221 - fix several locking issues related to i_size (5492,5624,5654,5672)
5222 - don't move pending lock onto export if it is already evicted (5683)
5223 - fix kernel oops when creating .foo in unlinked directory (5548)
5224 - fix deadlock in obdfilter statistics vs. object create (5811)
5225 - use time_{before,after} to avoid timer jiffies wrap (5882)
5226 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5227 - Do not leak request if server was not able to process it (5154)
5228 - If mds_open unable to find parent dir, make that negative lookup(5154)
5229 - don't create new directories with extent-mapping (5909, 5936)
5231 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5232 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5233 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5234 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5235 - support for RHEL4 kernel on i686 (5773)
5236 - provide error messages when incompatible logs are encountered (5898)
5238 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5239 * version 1.4.0.10 (1.4.1 release candidate 1)
5241 - don't keep a lock reference when lock is not granted (4238)
5242 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5243 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5244 - drop import inflight refcount on signal_completed_replay error (5255)
5245 - unlock page after async write error during send (3677)
5246 - handle missing objects in filter_preprw_read properly (5265)
5247 - no transno return for symlink open, don't save no-trasno open (3440)
5248 - don't try to complete elan receive that already failed (4012)
5249 - free RPC server reply state on error (5406)
5250 - clean up thread from ptlrpc_start_thread() on error (5160)
5251 - readahead could read extra page into cache that wasn't ejected (5388)
5252 - prevent races in class_attach/setup/cleanup/detach (5260)
5253 - don't dereference de->d_inode after l_dput of de (5458)
5254 - use "int" for stripe value returned from lock_to_stripe (5544)
5255 - mballoc allocation and error-checking fixes in 2.6 (5504)
5256 - block device patches to fix I/O request sizes in 2.6 (5482)
5257 - look up hostnames for IB nals (5602)
5258 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5259 - don't start multiple acceptors for the same port (5277)
5260 - fix incorrect LASSERT in mds_getattr_name (5635)
5261 - export a proc file for general "ping" checking (5628)
5262 - fix "lfs check" to not block when the MDS is down (5628)
5264 - service request history (4965)
5265 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5266 - create an "evict_client" /proc entry on OSTs, like the MDS has
5267 - fix mount usage message, return errors per mount(8) (5168)
5268 - change grep [] to grep "[]" in tests so they work in more UMLs
5269 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5270 - remove ancient LOV_MAGIC_V0 EA support (5047)
5271 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5272 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5273 - allow readahead allocations to fail when low on memory (5383)
5274 - mmap locking landed again, after considerable improvement (2828)
5275 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5277 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5280 - send OST transaction number in read/write reply to free req (4966)
5281 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5282 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5283 - the watchdog thread now runs as interruptible (5246)
5284 - flock/lockf fixes (but it's still disabled, pending 5135)
5285 - don't use EXT3 constants in llite code (5094)
5286 - memory shortage at startup could cause assertion (5176)
5288 - reorganization of lov code
5289 - single portals codebase
5291 - add extents/mballoc support (5025)
5292 - direct I/O reads in the obdfilter (4048)
5293 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5295 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5298 - send OST transaction number in read/write reply to free req (4966)
5299 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5300 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5301 - the watchdog thread now runs as interruptible (5246)
5302 - handle missing objects in filter_preprw_read properly (5265)
5303 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5304 - drop import inflight refcount on signal_completed_replay error (5255)
5305 - unlock page after async write error during send (3677)
5306 - return original error code on reconstructed replies (3761)
5307 - no transno return for symlink open, don't save no-trasno open (3440)
5309 - add pid to ldlm debugging output (4922)
5310 - bump the watchdog timeouts -- we can't handle 30sec yet
5311 - extra debugging for orphan dentry/inode bug (5259)
5313 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5316 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5317 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5318 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5319 - fix NULL dereference in /proc/sys/portals/routes (4827)
5320 - allow failed mdc_close() operations to be interrupted (4561)
5321 - stop precreate on OST before MDS would time out on it (4778)
5322 - don't send partial-page writes before EOF from client (4410)
5323 - discard client grant for sub-page writes on large-page clients (4520)
5324 - don't free dentries not owned by NFS code, check generation (4806)
5325 - fix lsm leak if mds_create_objects() fails (4801)
5326 - limit debug_daemon file size, always print CERROR messages (4789)
5327 - use transno after validating reply (3892)
5328 - process timed out requests if import state changes (3754)
5329 - update mtime on OST during writes, return in glimpse (4829)
5330 - add mkfsoptions to LDAP (4679)
5331 - use ->max_readahead method instead of zapping global ra (5039)
5332 - don't interrupt __l_wait_event() during strace
5334 - add software watchdogs to catch hung threads quickly (4941)
5335 - make lustrefs init script start after nfs is mounted
5336 - fix CWARN/ERROR duplication (4930)
5337 - return async write errors to application if possible (2248)
5338 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5339 - print NAL number in %x format (4645)
5340 - update barely-supported suse-2.4.21-171 series (4842)
5341 - support for sles 9 %post scripts
5342 - support for building 2.6 kernel-source packages
5343 - support for sles km_* packages
5345 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5348 - ignore -ENOENT errors in osc_destroy (3639)
5349 - notify osc create thread that OSC is being cleaned up (4600)
5350 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5351 - reconstruct ost_handle() like mds_handle() (4657)
5352 - create a new thread to do import eviction to avoid deadlock (3969)
5353 - let lconf resolve symlinked-to devices (4629)
5354 - don't unlink "objects" from directory with default EA (4554)
5355 - hold socknal file ref over connect in case target is down (4394)
5356 - allow more than 32000 subdirectories in a single directory (3244)
5357 - fix blocks count for O_DIRECT writes (3751)
5358 - OST returns ENOSPC from object create when no space left (4539)
5359 - don't send truncate RPC if file size isn't changing (4410)
5360 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5361 - bind to privileged port in socknal and tcpnal (3689)
5363 - rate limit CERROR/CWARN console message to avoid overload (4519)
5364 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5365 - basic mmap support (3918)
5366 - kernel patch series update from b1_4 (4711)
5368 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5371 - avoid crash during MDS cleanup with OST shut down (2775)
5372 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5373 - don't use bad inodes on the MDS (3744)
5374 - dynamic object preallocation to improve recovery speed (4236)
5375 - don't hold spinlock over lock dumping or change debug flags (4401)
5376 - don't zero obd_dev when it is force cleaned (3651)
5377 - print grants to console if they go negative (4431)
5378 - "lctl deactivate" will stop automatic recovery attempts (3406)
5379 - look for existing locks in ldlm_handle_enqueue() (3764)
5380 - don't resolve lock handle twice in recovery avoiding race (4401)
5381 - revalidate should check working dir is a directory (4134)
5383 - don't always mark "slow" obdfilter messages as errors (4418)
5385 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5388 - don't close LustreDB during write_conf until it is done (3860)
5389 - fix typo in lconf for_each_profile (3821)
5390 - allow dumping logs from multiple threads at one time (3820)
5391 - don't allow multiple threads in OSC recovery (3812)
5392 - fix debug_size parameters (3864)
5393 - fix mds_postrecov to initialize import for llog ctxt (3121)
5394 - replace config semaphore with spinlock (3306)
5395 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5396 - don't allow enqueue to complete on a destroyed export (3822)
5397 - down write_lock before checking llog header bitmap (3825)
5398 - recover from lock replay timeout (3764)
5399 - up llog sem before sending rpc (3652)
5400 - reduce ns lock hold times when setting kms (3267)
5401 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5402 - fix NULL deref and obd_dev leak on setup error (3312)
5403 - replace some LBUG about llog ops with error handling (3841)
5404 - don't match INVALID dentries from d_lookup and spin (3784)
5405 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5406 - fix invalid assertion in ptlrpc_set_wait (3880)
5408 - add libwrap support for the TCP acceptor (3996)
5409 - add /proc/sys/portals/routes for non-root route listing (3994)
5410 - allow setting MDS UUID in .xml (2580)
5411 - print the stack of a process that LBUGs (4228)
5413 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5416 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5417 - make sure to unset replay flag from failed open requests (3440)
5418 - if default stripe count is 0, use OST count for inode size (3636)
5419 - update parent mtime/ctime on client for create/unlink (2611)
5420 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5421 - free recovery state on server during a forced cleanup (3571)
5422 - unregister_reply for resent reqs (3063)
5423 - loop back devices mounting and status check on 2.6 (3563)
5424 - fix resource-creation race that can provoke i_size == 0 (3513)
5425 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5426 - more debugging for page-accounting assertion (3746)
5427 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5428 - avoid deadlock after precreation failure (3758)
5429 - fix race and lock order deadlock in orphan handling (3450, 3750)
5430 - add validity checks when grabbing inodes from l_ast_data (3599)
5432 - add /proc/.../recovery_status to obdfilter (3428)
5433 - lightweight CDEBUG infrastructure, debug daemon (3668)
5434 - change default OSC RPC parameters to be better on small clusters
5435 - turn off OST read cache for files smaller than 32MB
5436 - install man pages and include them in rpms (3100)
5437 - add new init script for (un)mounting lustre filesystems (2593)
5438 - run chkconfig in %post for init scripts (3701)
5439 - drop scimac NAL (unmaintained)
5441 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5444 - clean kiobufs before and after use (3485)
5445 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5446 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5447 - change init script to fail more clearly if not run as root (1528)
5448 - allow clients to reconnect during replay (1742)
5449 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5450 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5451 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5452 - fix cond_resched (3554)
5453 - only evict unfinished clients after recovery (3515)
5454 - allow bulk resend, prevent data loss (3570)
5455 - dynamic ptlrpc request buffer allocation (2102)
5456 - don't allow unlinking open directory if it isn't empty (2904)
5457 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5458 - remove extraneous obd dereference causing LASSERT failure (3334)
5459 - don't use get_cycles() when creating temp. files on the mds (3156)
5460 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5461 - handle EEXIST for set-stripe, set proper directory name (3336)
5463 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5464 - fix ksocknal_fmb_callback() error messages (2918)
5466 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5469 - don't copy lvb into (possibly NULL) reply on error (2983)
5470 - don't deref dentry after dput, don't free lvb on error (2922)
5471 - use the kms to determine writeback rpc length (2947)
5472 - increment oti_logcookies when osc is inactive (2948)
5473 - update client's i_blocks count via lvb messages (2543)
5474 - handle intent open/close of special files properly (1557)
5475 - mount MDS with errors=remount-ro, like obdfilter (2009)
5476 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5477 - don't use cancelling-locks' kms values (2947)
5478 - use highest lock extent for kms, not last one (2925)
5479 - don't dereference ERR_PTR() dentry in error handling path (3107)
5480 - fix thread race in portals_debug_dumplog() (3122)
5481 - create lprocfs device entries at setup instead of at attach (1519)
5482 - common AST error handler, don't evict client on completion race (3145)
5483 - zero nameidata in detach_mnt in 2.6 (3118)
5484 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5485 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5486 - initialize RPC timeout timer earlier for 2.6 (3219)
5487 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5488 - print nal/nid for unknown nid (3258)
5489 - additional checks for oscc recovery before doing precreate (3284)
5490 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5491 - don't crash in mdc_close for bad permissions on open (3285)
5492 - zero i_rdev for non-device files (3147)
5493 - clear page->private before handing to FS, better assertion (3119)
5494 - tune the read pipeline (3236)
5495 - fix incorrect decref of invalidated dentry (2350)
5496 - provide read-ahead stats and refine rpc in flight stats (3328)
5497 - don't hold journal transaction open across create RPC (3313)
5498 - update atime on MDS at close time (3265)
5499 - close LDAP connection when recovering to avoid server load (3315)
5500 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5501 - don't leak open file on MDS after open resend (3325)
5502 - serialize filter_precreate and filter_destroy_precreated (3329)
5503 - loop device shouldn't call sync_dev() for nul device (3092)
5504 - clear page cache after eviction (2766)
5505 - resynchronize MDS->OST in background (2824)
5506 - refuse to mount the same filesystem twice on same mountpoint (3394)
5507 - allow llmount to create routes for mounting behind routers (3320)
5508 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5509 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5510 - fix rare problem with rename on htree directories (3417)
5512 - allow default OST striping configuration per directory (1414)
5513 - fix compilation for qswnal for 2.6 kernels (3125)
5514 - increase maximum number of MDS request buffers for large systems
5515 - change liblustreapi to be useful for external progs like lfsck (3098)
5516 - increase local configuration timeout for slow disks (3353)
5517 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5519 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5522 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5523 - don't overwrite extent policy data in reply if lock was blocked (2901)
5524 - drop filter export grants atomically with removal from device (2663)
5525 - del obd_self_export from work_list in class_disconnect_exports (2908)
5526 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5527 - swab reply message in mdc_close, other PPC fixes (2464)
5528 - fix destroying of named logs (2325)
5529 - overwrite old logs when running lconf --write_conf (2264)
5530 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5531 - fix race in target_handle_connect (2898)
5532 - mds_reint_create() should take same inode create lock (2926)
5533 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5534 - don't close files for self_export to avoid uninitialized obd (2936)
5535 - allow MDS with the same name as client node (2939)
5536 - hold dentry reference for closed log files for unlink (2325)
5537 - reserve space for all logs during transactions (2059)
5538 - don't evict page beyond end of stripe extent (2925)
5539 - don't oops on a deleted current working directory (2399)
5540 - handle hard links to targets without a parent properly (2517)
5541 - don't dereference NULL lock when racing during eviction (2867)
5542 - don't grow lock extents when lots of conflicting locks (2919)
5544 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5547 - account for cache space usage on clients to avoid data loss (974)
5548 - lfsck support in lustre kernel code (2349)
5549 - reduce journal credits needed for BRW writes (2370)
5550 - orphan handling to avoid losing space on client/server crashes
5551 - ptlrpcd can be blocked, stopping ALL progress (2477)
5552 - use lock value blocks to assist in proper KMS, faster stat (1021)
5553 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5554 - recovery for initial connections (2355)
5555 - fixes for mds_cleanup_orphans (1934)
5556 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5557 - block all file creations until orphan recovery completes (1901)
5558 - client remove rq_connection from request struct (2423)
5559 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5560 - recovery timer race (2670)
5561 - mdc_close recovey bug (2532)
5562 - ptlrpc cleanup bug (2710)
5563 - mds timeout on local locks (2588)
5564 - namespace lock held during RPCs (2431)
5565 - handle interrupted sync write properly (2503)
5566 - don't try to handle a message that hasn't been replied to (2699)
5567 - client assert failure during cleanup after abort recovery (2701)
5568 - leak mdc device after failed mount (2712)
5569 - ptlrpc_check_set allows timedout requests to complete (2714)
5570 - wait for inflight reqs when ptlrpcd finishes (2710)
5571 - make sure unregistered services are removed from the srv_list
5572 - reset bulk XID's when resending them (caught by 1138 test)
5573 - unregister_bulk after timeout
5574 - fix lconf error (2694)
5575 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5576 - readahead locks pages, leaves pending causing memory pressure (2673)
5577 - increase OST request buffers to 4096 on large machines (2729)
5578 - fix up permission of existing directories in simple_mkdir (2661)
5579 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5580 - don't assign transno to errored transactions (2742)
5581 - don't delete objects on OST if given a bogus objid from MDS (2751)
5582 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5583 - if rq_no_resend, then timeout request after recovery (2432)
5584 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5585 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5586 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5587 - MDS recovery shouldn't delete objects that it has given out (2730)
5588 - if enqueue arrives after completion, don't clobber LVB (2819)
5589 - don't unlock pages twice when trigger_group_io returns error (2814)
5590 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5591 - don't write pages to disk if there was an error (1450)
5592 - don't ping imports that have recovery disabled (2676)
5593 - take buffered bytes into account when balancing socknal conn (2817)
5594 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5595 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5596 - remove little-endian swabbing of llog records (1987)
5597 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5598 - retry reposting request buffers if they fail (1191)
5599 - grow extent at grant time to avoid granting a revoked lock (2809)
5600 - lock revoke doesn't evict page if covered by a second lock (2765)
5601 - disable VM readahead to avoid reading outside lock extents (2805)
5603 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5604 - updated kernel patches for hp-2.4.20 kernel (2681)
5606 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5609 - fix truncated write corruption (2366)
5610 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5612 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5613 - ensure len(uuid) < 37 in lmc (1171)
5614 - fix ia64 OOPS in llog_test (2255)
5615 - zero end of page at obdfilter for partial page writes (2648)
5616 - don't leave stale dentries around after renames (bug 2428)
5617 - fix timeouts when evicting a client with a single lock held (2642)
5618 - set deadline for the initial HELLO message to drain (2634)
5619 - print out dotted-quad IP addresses in the socknal (2302)
5621 - additional debugging for MDS client eviction problem (2443)
5622 - fix mkfsoptions support for osts (2603, 2604)
5624 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5627 - add series for the vanilla 2.6.0 kernel
5628 - add series for the vanilla 2.4.24 kernel
5629 - add series for a cray x86/64 UL kernel drop
5630 - fix xattr patches for the vanilla 2.4.19 series
5632 - generate true UUIDs in lmc (1171)
5633 - have portals stack dumping break in UML (2466)
5634 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5635 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5636 - wake callback waiting threads on client eviction (2460)
5637 - Add --ptldebug and --subsystem to lmc (1719)
5638 - update assertion to allow safe interrupt allocation
5639 - set rq_no_resend for cancel requests (2432)
5640 - recalculate ptlrpcd timeout after resend (2494)
5641 - call vfs_rmdir when removing pending directories (2368)
5642 - fix renaming a file to itself (2429)
5643 - lmc creates a default one-stripe lov (2454)
5644 - expand procfs space to handle large clusters (2326)
5645 - increase UML stack to avoid overflow
5646 - update lconf's list of debug and subsystem masks
5647 - fix lfs find --obd (2510)
5648 - /proc tunable for disabling filter read caching (2591)
5649 - stop rpm packages from altering slapd.conf (2301)
5650 - disable nagle in the socknal under 0conf (2578)
5651 - choose mds inode size based on stripe count (2572)
5652 - fix kernel-source rpm problems (2516)
5654 - add --disable-doc to avoid pdf generation (2421)
5655 - update documentation, tests, type-os, comments
5656 - avoid format warnings on ia64
5657 - remove the TOE NAL
5658 - tiny code cleanups by removing unused fields
5660 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5663 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5664 - remove the most blatant lies from BUILDING file (2371)
5665 - change default debug level to reasonable production setting
5666 - reduce client side cache size to reduce cache flush time
5667 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5668 - make TCP zerocopy and pinger support enabled by default (2476)
5669 - sync writes completed after process exits caused crashes (2319)
5670 - maintain correct mount count on the MDS (2356)
5671 - backout 1557, because 2316 wasn't really fixed
5672 - better file I/O statistics gathering in /proc
5673 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5674 - another kernel patch to fix zero-copy TCP function export
5675 - don't take duplicate lock when processing re-sent getattr (2420)
5676 - lctl uses obd_self_export instead of creating new conn (2353)
5677 - MDS/OST recovery case which requires object creation asserted (2425)
5678 - move lfs from /usr/sbin to /usr/bin in packages
5679 - fix race between mds_client_add and mds_client_free (2417)
5680 - use kmalloc instead of slabs in portals (2430)
5681 - don't create duplicate records when a failover MDS is present (2442)
5682 - remove unnecessary mount age check (2332)
5683 - don't remove directory inodes from locks prematurely (2451)
5684 - don't break if MDS service name is the same as hostname (2103)
5685 - fix races in client write RPC generation when cache full (2482)
5687 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5690 - remove now-unused request->rq_obd (278)
5691 - if an allocation fails, print out how much memory we've used (1933)
5692 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5693 - add the "configurable stack size" patch to most series files (1256)
5694 - ability to write large log records, for 100+ OST configs (2306)
5695 - fix NULL deref when filter_prep fails (2314)
5696 - fix operator precedence error in filter_sync
5697 - dynamic allocation of socknal TX descriptors (2315)
5698 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5699 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5700 - ensure that CWARN messages in Portals always get to the syslog
5701 - __init/__exit are not for prototype decls (ldlm_init/exit)
5702 - x86-64 compile warning fixes
5703 - fix gateway LMC keyword conflict (2318)
5704 - fix MDS lock inversions in getattr/reint paths (1844)
5705 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5706 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5707 - DLM race condition prevented some lock evictions (2328)
5708 - ENOMEM detection and retry on socknal sends (2230)
5709 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5710 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5711 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5712 - better error messages when a client is rejected during recovery (1505)
5713 - avoid cancelling locks which were never granted, after failure (2330)
5714 - fix i_sem/journal inversion in mds_client_add (2333)
5715 - fix truncate/getattr lock cycle deadlock (2334)
5716 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5717 - fix two rare exit paths which could leak an l_lock() ref (2321)
5718 - fencepost error in MDS/OST orphan recovery (2226)
5719 - make log record alignment 8 bytes (1988)
5720 - lstripe now fails when requested offset > ost_count (2237)
5721 - ensure that all kernel series have a complete list.h (1607)
5722 - fix crashes in special-file operations (2316)
5723 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5725 - allow configurable automake binary, for testing new versions
5726 - small update to the lfs documentation
5728 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5730 * fix negative export reference count in fsfilt_sync (2312)
5732 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5733 * release candidate 0.9.1
5735 - orphans are moved into the PENDING directory for possible recovery
5736 - replayed opens now open by fid for orphan/rename safety (1042)
5737 - last close of an orphan inode generates a transno (683)
5738 - chdir() and mount() now pin the directory entry (1020)
5739 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5740 - discard very old requests without processing them (1502)
5741 - remove some common, well-understood CERRORs (1505)
5742 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5743 - clear "grant" flags in OST replies until OST grant code lands (1644)
5744 - fix read performance by not clobbering i_blksize on client (1598)
5745 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5746 - make LRU size a /proc tunable, clears locks when reduced (707)
5747 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5748 - server locks take references on exports now (1558)
5749 - build fixes for 2.4.20-rh trees (1663)
5750 - return an error from lov_create if all OSCs are inactive (1751)
5751 - fix import levels when a reconnect happens without a timeout (1597)
5752 - exit early from mds_open if we get a lookup error (1749)
5753 - partial page read at EOF wouldn't wait for disk before sending (1642)
5754 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5755 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5756 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5757 - write/truncate lock inversion (1639)
5758 - Don't auto-load obdclass, portals modules during cleanup (1495)
5759 - fix timestamps from jumping to "now" (1763)
5760 - extra journal assertions (1648)
5761 - add an extra multiunlink test (1771)
5762 - fix read_record/write_record API (1776)
5763 - fix leak of offset_extent, possible incorrect i_size later (1772)
5764 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5765 - Debugging for the kqswnal_get_idle_tx problems (1820)
5766 - Allow recovery to be attempted multiple times (1536)
5767 - Write out MDS last_rcvd file after it is first created (1600)
5768 - Fix tx_descriptor leak in failed transmit situations (1827)
5769 - ext3 journaling fixes for assertion failure after IO error (1871)
5770 - class_export_put() on freed export after completion AST error (1896)
5771 - Fix revalidate looping in VFS (1322)
5772 - Don't access a freed export during MDS_REINT timeout (1521)
5773 - Add open-unlink recovery support on the MDS (1673,1764)
5774 - Return an error if no MDS data was read from last_rcvd (1946)
5775 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5776 - Don't setup a disk device that doesn't match exported UUID (317)
5777 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5778 - avoid committing NULL handle in force close
5779 - local.sh is now a one-stripe LOV configuration
5780 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5781 - fix ext3 htree duplicate directory entry corruption (1516)
5782 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5783 - update to new LOV EA format (2097)
5784 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5786 2003-06-15 Phil Schwan <phil@clusterfs.com>
5789 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5790 - per-import/export recovery handling (958, 931, 959)
5791 - multiple last-rcvd slots, for serving multiple FSes (949)
5792 - connections are again shared between multiple imp/exports (963, 964)
5793 - "umount -f" would hang if any requests needed to be sent (393, 978)
5794 - avoid pinning large req buffer by copying for queued messages (989)
5795 - add "uuid" to "lctl device" command to help upcalls (991)
5796 - "open" RPCs with transnos would confuse recovery counters (1037)
5797 - do proper endian conversion of all wire messages (288, 340, 891)
5798 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5799 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5800 - fix ptlrpc_connection leak in target_handle_connect (1174)
5801 - fix import refcounting bug in OST and MDS cleanup (1134)
5802 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5803 - fix very unlikely obd_types race condition (501)
5804 - remove osc_open hack for echo_client (1187)
5805 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5806 - a failure in read_inode2 leads to deadlock (1139)
5807 - cancel ack-locks as soon as transaction is committed (1072)
5808 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5809 - make sure to commitrw after any preprw to avoid deadlock (1162)
5810 - failing to execute a file in a lustre FS would lock inode (1203)
5811 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5812 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5813 - obd_uuid2tgt would walk off the end of the list (1255)
5814 - on IA64 the osc would give portals incorrect bulk size (1258)
5815 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5816 - fix lock inversion caused by new llite matching code (1282)
5817 - limit the number of dirty pages on a client to 10MB (1286)
5818 - timed out locks were not being corrected cancelled (1289)
5819 - fix O_DIRECT above 4GB on IA-32 (1292)
5820 * major user-visible changes
5821 - fail out/fail over policy now controlled by the upcall (993)
5823 - add OBD_PING to check server availability and failure (954)
5824 - lustre messages are now sent in sending host order (288, 340, 891)
5825 - add eadatalen to MDS getattr reply (340)
5826 - OST read replies may contain second buffer, with per-page status (593)
5828 2003-03-11 Phil Schwan <phil@clusterfs.com>
5831 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5832 - failed open()s could cause deadlock; fixed (867, 869)
5833 - stop cancelling OST locks when files are closed (481)
5834 - overlapping XID spaces caused network corruption (851, 853)
5835 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5836 - setattr_raw added, to do single-RPC, server-side setattrs
5837 - lmc/lconf syntax change for OST UUIDs
5838 - fix crashy race condition between ptlrpc_free_req and osc_close
5839 - don't use request in mdc_enqueue if we hit a timeout (889)
5840 - don't set the inode i_size for regular files from the MDS (896)
5841 - handle out of order completion AST (842)
5842 - don't LBUG if a lock request times out after receiving AST (913)
5843 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5844 - if a bad lock AST arrives, send an error instead of dropping entirely
5845 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5846 - fix leak in bulk IO when only partially completed (899, 900, 926)
5847 - fix O_DIRECT for ia64 (55)
5848 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5849 - C-z after timeout could hang a process forever; fixed (977)
5851 - client-side I/O cache (678, 924, 929, 941, 970)
5853 - READPAGE and SETATTRs which don't take server-side locks get
5856 2003-02-11 Phil Schwan <phil@clusterfs.com>
5859 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5860 - Fix fsfilt_extN_readpage() to read a full page of directory
5861 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5862 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5863 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5864 - Most intent operations were removed, in favour of a new RPC mode
5865 that does a single RPC to the server and bypasses most of the VFS
5866 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5867 - Aggressively cancel local locks on DLM servers
5868 - mds_reint_unlink sends EA to the client if it's the last nlink.
5869 client uses that EA to unlink OST objects.
5870 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5871 - recursive symlinks were fixed (439)
5872 - fixed NULL deref in DEBUG_REQ
5873 - filter_update_lastobjid no longer calls sync, which annoyed extN
5874 - fixed multi-client small-writes to a single file problem (445)
5875 - fixed mtime updates during file writes (607)
5876 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5877 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5878 - fixed Linux OST size reporting problem (444, 656)
5879 - OST now updates object mtime with writes or setattr (607, 619)
5880 - client verifies file size before zeroing page past EOF (445)
5881 - OST now writes last allocated objid to disk with allocation (108)
5882 - LOV on echo now works (409)
5884 - mds_reint_unlink sends a new buffer, with the EA included. this
5885 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5886 set if a regular file was being unlinked, and it was the last link
5887 - use PtlGet from the target for bulk writes (315)
5888 - OST now updates object mtime with writes or setattr (607, 619)
5889 - LDLM now has a grant-time callback to revalidate locked items, if
5891 - Many MDS operations were reorganized to combat race conditions
5893 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5894 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5896 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5899 - Fully reactivate OST imports after reconnection (512, others)
5900 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5901 - More graceful error handling for truncating on dead OST (515)
5902 - Don't error out unless we're actually accessing dead stripes (474)
5903 - Fix garbage sizes when stripes are missing (410)
5904 - LRU counters were broken, causing constant lock purge (433, 432)
5905 - garbage on read from stripes with failed OSTs (441)
5906 - mark OSCs as active before reconnecting during recovery (438)
5907 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5908 - lfind did not preserve OST order in output (443)
5909 - symlinks cause hung clients, incorrect data (439)
5910 - stop dereferencing request after dropping refcount (457)
5911 - don't LASSERT(spin_is_locked) on non-SMP (455)
5912 - fixes for many rename() bugs
5913 - fstat didn't correctly synchronize attributes (399)
5914 - server must handle lock cancellation during blocking AST prep (487)
5915 - bulk descriptors were free()d too soon (511)
5916 - fix paths in lconf, which would load incorrect modules (451, 507)
5917 - fix confusing lconf 'host not found' error message (386)
5918 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5919 - fix race condition in mdc_blocking_ast() for inode access (526)
5920 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5921 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5922 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5923 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5924 - fix lstripe in several ways (kernel side) (527)
5925 - fix request leak in ldlm_cli_enqueue (262)
5926 - incorrect OSC was marked inactive after OST failure
5927 - call mds_fs_cleanup before unmounting filesystem (524)
5928 - fix races between taking ns_lock and ldlm_lock_change_resource
5929 - fix races updating LOV export open file list
5930 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5931 - fix recovery NULL deref in ldlm_cli_cancel_unused
5932 - fix some DLM races by using new hash table for lock handles (419)
5933 - permit the client to specify desired inodes, at replay
5934 - duplicate requests when we queue them for replay reintegration
5935 - fix last_rcvd offset calculation
5936 - sync after each recovered transaction, so we always make progress
5937 - never, not always, ERESTART requests without transnos
5938 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5940 - skip replay if the MDS says that the client is already connected
5941 - don't check for a recovery-enabled export to match lctl's UUID
5942 - don't INC_USE_COUNT for phantom exports
5943 - don't crash when cleaning up phantom exports (567)
5944 - don't double-finish or set replay data for errored mdc_open requests
5945 - abort requests when they time out, so we don't get old replies
5946 - send/receive replies for AST messages again
5947 - if the client says that it doesn't have the lock, cancel it on the
5949 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5950 mark it as destroyed, it will all work out when decref is called
5951 - fix module use counts (22, 581)
5953 - ASTs now expect a reply (server cancels lock on error reply)
5955 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5958 - fix many simultaneous client startup (392)
5959 - fix dentry->d_it clobbering
5960 - credentials weren't being shipped for readdir/getattr operations
5961 - remove invalid assertions triggered during some concurrent MD
5963 - proper Lustre versions added (336, 389)
5964 - fix memory leak for create error case (398)
5965 - fix LOV locking bug that would get cli/srv out of sync
5966 - fix echo client over LOV (409)
5967 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5968 - fix double-O_EXCL intent crash (424)
5969 - avoid sending multiple lock CANCELs (352)
5971 - MDS can do multi-client recovery (modulo bugs in new code)
5973 - many updates, edits, cleanups
5975 2002-11-18 Phil Schwan <phil@clusterfs.com>
5978 - fix null d_it dereference (346)
5979 - fix full OST/dbench hang (333)
5980 - fix permission problem with file removal (286)
5981 - fix removal of OSCs from LOV when they fail
5982 - fix NULL deref during bulk timeout (214)
5983 - fix problems related to multiple filesystems on one MDS (241)
5984 - fixed serious subtle metadata locking bugs
5985 - free locks on clients when inodes are removed due to memory
5987 - fix inode pointer in lock data (285)
5988 - partial support for multiple MDS on a single host (241)
5989 - data locks weren't cancelled at clear_inode time (290, 311)
5990 - intent locks could lead to unbounded lock growth (205)
5991 - added a maximum lock count, an LRU list, and a flusher
5992 - fix multiple rename (365)
5993 - properly abstracted the echo client
5994 - OSC locked 1 byte too many; fixed
5995 - rewrote brw callback code:
5996 - fixed recovery bugs related to LOVs (306)
5997 - fixed too-many-pages-in-one-write crash (191)
5998 - fixed (again) crash in sync_io_timeout (214)
5999 - probably fixed callback-related race (385)
6001 - Add capability to MDS protocol
6002 - LDLM cancellations and callbacks on different portals
6004 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
6007 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
6008 - defer open object destruction to close time (601981, 138)
6009 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
6010 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
6011 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
6012 - don't call obd_setattr in ll_file_release for destroyed objects
6013 * protocol change to lustre_msg: move |version| and add |flags|
6014 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
6015 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
6016 * added replay of create, unlink, link and rename operations during
6017 MDS failover; recovery should be much more robust now
6018 * remove failed OSCs from LOVs (only lov_create uses this so far)
6019 * the lustre-HOWTO was brought (more) up to date (582544)
6021 2002-10-23 Phil Schwan <phil@clusterfs.com>
6024 - in-use dentries weren't being reused properly (617851)
6025 - prevent multiple LDLM setup (599178)
6026 - fix LOV size calculations for truncate (617853)
6027 - fix client handling of MDS intent errors (POSIX)
6028 - fix permission bug in lovstripe.c test (624321)
6029 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6030 - truncate past end of file could corrupt data
6031 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6032 - a race in recovery could return ETIMEDOUT to apps (623947)
6033 - building outside the source directory was fixed
6034 * the lustre-HOWTO was brought (more) up to date (582544)
6035 * major progress was made on recovery functionality
6037 2002-10-10 Phil Schwan <phil@clusterfs.com>
6040 - recovery deadlock fix
6041 - rm -rf causes LBUG fix (617817)
6042 - file open by multiple tasks fix (618962)
6043 - directory permissions bugs (602707 and 620007)
6044 - journal_stop fixed with locking (611313)
6045 - O_APPEND failures resolved (618273, perhaps 614459)
6046 - lconf PATH fix (619770)
6047 - IA64 build fix (621450)
6048 - RPC buffer sizes scale with amount of memory
6050 2002-10-01 Phil Schwan <phil@clusterfs.com>
6053 - locks would be cancelled without throwing away data pages,
6054 resulting in inconsistent data (605627)
6055 - inode attributes were not always being refreshed (605627, 612449)
6056 - lconf now continues to cleanup after lctl reports an error
6057 - MDS now enforces user permissions (602707)
6058 - lprocfs cleanup fixed, but not yet enabled (614157)
6059 - fixed infinite server hang, should a client not respond to an AST
6060 - avoid going into recovery if user calls readlink() with a buffer
6061 that's too small (613941)
6062 - AST RPCs no longer require replies (614867) -- this may be changed
6063 - don't crash server if client sends an IOV that's too big (611336)
6064 - fixed lock conversion deadlock (611892)
6065 - fixed the following of symlinks (614622)
6066 * recovery: the server can remove locks from a client that dies, other
6067 clients can make progress
6068 * more extN patch fixes
6069 * compile-time configurable ptlrpc buffer allocations
6071 - collaborative read cache document
6072 - Lustre Lite Performance CDR document-in-progress
6074 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6077 - fix typo in patch-2.4.18
6079 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6082 - clear ptlrpc request each time in handle_incoming_request()
6083 - unlink of files now destroys the object on the OST
6085 2002-09-19 Peter Braam <braam@clusterfs.com>
6087 * add hard link support
6088 * change obdfile creation method
6089 * kernel patch changed
6091 2002-09-19 Peter Braam <braam@clusterfs.com>
6094 - stack overflow bug in extN fixed
6096 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6098 * documentation updates
6099 - add man pages for config tools
6100 - update tests/README to describe testing with new config tools
6101 - finish metadata API descriptions
6102 * bug fixes and cleanups
6103 - statfs workaround for 16TB limit
6104 - LOV stripe allocation improved, can stripe on subset of OSTs
6105 - LOV file size/IO offset was wrong for files > 4GB in size
6106 - object EA data was being dropped, caused files to be unreadable
6107 - memory overflow with non-LOV OST caused memory corruption
6108 - fixed regression tests to work with new config tools, obdfilter
6109 - fixed bug when directory size became larger than 1 block
6110 - fixed bug (for single client case) when PWD was deleted
6111 - invalidate local directory pages when doing intent-based ops
6112 - avoid LDLM oops when lock callback contained bad data
6114 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6116 * documentation updates
6117 * bug fixes and cleanups
6118 - configuration tools
6121 - 64-bit compile warnings
6122 - 64-bit internal statfs data
6124 * test_brw on persistent OST devices
6126 * lprocfs (disabled)
6128 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6130 * documentation updates
6131 * bug fixes and cleanups
6132 * configuration tools
6134 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6137 * many small fixes to 0_5_4
6138 * io/network handling
6139 * thinkos in MDS operations
6141 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6144 * crucial basic fixes to 0.5.3
6145 * IOR, Iozone work over Elan
6148 2002-08-07 Phil Schwan <phil@clusterfs.com>
6149 * version 0_5_3, our first alpha
6150 * we use the new Portals iovs
6151 * documentation updates
6152 * bug fixes and cleanups
6153 * small changes in the DLM wire protocol
6155 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6156 * version 0_5_1 with some initial stability,
6157 * locking on MD and file I/O.
6158 * documentation updates
6159 * several bug fixes since 0.5.0
6160 * small changes in wire protocol
6162 2002-07-18 Phil Schwan <phil@clusterfs.com>
6164 * delivered as Lustre Light Alpha
6165 * fixed a crash after handling invalid MDS requests
6166 * fixed directory pages for architectures with non-4k pages sizes
6168 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6169 * release version v0_4_4
6170 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6172 * Fixes a number of interruption problems with OST operations.
6173 * Update documentation for portals header changes
6174 * Move all wire protocol structs/defines to lustre_idl.h
6175 * Fixes symlink length bug.
6176 * Add tcpdump to repository.
6178 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6179 * release version v0_4_3
6180 * Fixes statfs for inodes on extN.
6181 * Fixes bug in runtests which would delete /etc/hosts.
6182 * Use 64-bit object IDs wherever possible (not into VFS though)
6183 Remove ost_get_info, which is unused by lustre, and out of date.
6185 2002-07-03 Peter Braam <braam@clusterfs.com>
6186 * release version v0_4_2 Fixes a lookup error (type not passed)
6187 * move forward to head of Portals
6188 * move forward to latest Lustre kernel
6190 2002-06-25 Peter Braam <braam@clusterfs.com>
6191 * release version v0_4_1. Hopefully stable on single node use.