1 tbd Sun Microsystems, Inc.
3 * Server support for kernels:
4 2.6.18-164.11.1.el5 (RHEL 5)
5 2.6.18-164.11.1.0.1.el5 (OEL 5)
6 * Client support for unpatched kernels:
7 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
8 2.6.18-164.11.1.el5 (RHEL 5),
9 2.6.18-164.11.1.0.1.el5 (OEL 5)
10 2.6.16.60-0.42.8 (SLES 10),
12 * Recommended e2fsprogs version: 1.41.6.sun1
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
18 Description: lfs check only allowed for root
19 Details : Code cleanups and added a regresion test of non-root use of
24 Description: MDS Oopses in ptlrpc_dispatch_difficult_reply()/__list_add()
25 Details : Fix hr->hr_index overload.
29 Description: Fix FIEMAP functionality.
33 Description: FC11 patchless client support added
37 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
38 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
42 Description: New acc-sm script and various test-framework enhancments.
46 Description: Previously landed patch was reverted.
50 Description: Update SLES10 kernel to 2.6.16.60-0.42.8 (patchless)
54 Description: Add ldiskfs maxdirsize mount option.
58 Description: Fix mkfs.lustre support for 16TB LUNs.
62 Description: Print mmp_check_interval
63 Details : print mmp_check_interval and make it possible to abort
64 mount operation in case it takes too long.
68 Description: Add cascading_rw to lustre mpi tests.
72 Description: Store SOM data in EA.
73 Details : Also several SOM related cleanups.
77 Description: Add refcount to md_open_data
81 Description: Use refcount on lov_request_set to prevent memory corruption.
85 Description: Statahead: use parent dir's i_mutex for synchronization.
89 Description: loadgen improvements.
93 Description: Enable peer health detection in ksocklnd and ko2iblnd.
97 Description: Control DCACHE_LUSTRE_INVALID flag with MDS_INODE_LOOKCUP lock
101 Description: Fix race on obd_devcie:md_stats
105 Description: Fix SLES 11 client builds.
106 Details : Resolved uninitalized pointer gcc warnings.
110 Description: Allow non-root access for "lfs check".
111 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
115 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
117 Severity : enhancement
119 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
120 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
124 Description: Sanity test_27p fails.
125 Details : OBD_FAIL_OST_ENOSPC has been moved to filter_statfs().
126 Sanity tests 27 were cleaned up and improved.
130 Description: Unify req format on client/servers
131 Details : Use new req_capsule API [almost] everywhere instead of old PTLRPC
132 buffers and swabbers approach..
137 Description: drop join file stuff.
139 Severity : enhancement
141 Description: Add 2.6.27 and SLES11 (patchless client) support
145 Description: Fix in ptlrpc_expire_one_request() to print the signed time
148 Severity : enhancement
150 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
153 Severity : enhancement
155 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
157 Severity : enhancement
159 Description: correctly handle big reply message.
160 Details : send LNet event if reply is bigger then buffer and adjust this buffer
165 Description: Drop unnecessary __GFP_NOMEMALLOC flag from filter_get_page()
167 Severity : enhancement
169 Description: Add support for OEL5.
173 Description: Conf-sanity.sh 50g test - deactivated OST should not cause a panic.
177 Description: client could not reconnect to OST because of an active request.
178 Details : The client now retries to reconnect to the same server, if a
179 connect request failed with EBUSY or -EAGAIN. Test case 35b
180 conf-sanity.sh has been created.
184 Description: Don't allow make backward step on assignin osc next id.
185 Details : race between allocation next id and ll_sync thread can be cause
186 of set wrong osc next id and can be kill valid ost objects.
190 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
194 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
198 Description: Error handling in osc_statfs_interpret() has been improved.
199 Details : Check in osc_statfs_interpret() for EBADR.
203 Description: Do not send statfs() requests to OSTs disabled by administrator.
204 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
206 Severity : enhancement
208 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
210 Severity : enhancement
212 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
214 Severity : enhancement
216 Description: Enhancement for lfs(1) command to use numeric uid/gid.
220 Description: Improved cleanup procedure for conf-sanity test_46a.
224 Description: Do not start bulk transfer, if client tried to reconnect.
225 Details : Do not start bulk transfer, if client tried to reconnect,
226 but failed due to active rpcs. Check for exp_abort_active_req
227 has been implemented.
231 Description: Prevent inconsistences between linux and lustre mount structures.
232 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
233 Make the sleep interruptible.
235 Severity : enhancement
237 Description: provide server to client comms path
238 Details : server to client communications path via new LDLM_SET_INFO rpc
242 Description: truncate starts GFP_FS allocation under transaction causing deadlock
243 Details : ldiskfs_truncate calls grab_cache_page which may start page
244 allocation under an open transaction. This may lead to
245 calling prune_icache with consequent lustre reentrance.
249 Description: send correctly lsm on open replay
250 Details : MDS is trust to LSM size on replay open, but client can set wrong size
255 Description: enable client interface failover
256 Details : When a child reconnects from another NID, properly update export
257 nid hash position and ldlm reverse import.
259 Severity : enhancement
261 Description: Communicate OST degraded/readonly state via statfs to MDS
262 Details : Flags in the statfs returned from OSTs indicate whether the
263 OST is in a degraded RAID state, or if the filesystem has
264 turned read-only after a filesystem error is detected.
269 Description: don't panic if EPROTO was hit when reading symlink
270 Details : correctly handling request reference in error cases.
272 Severity : enhancement
274 Description: Add LustreNetLink, a kernel-userspace communcation path.
276 Severity : enhancement
278 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
283 Description: access to llog context before init.
284 Details : move handling CATALOGS file at osc layer and forbid access to llog
288 Frequency : rare, only when client is writing to OST on local node
290 Description: Avoid deadlock for local client writes
291 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
292 memory freeing context. This allows OST threads to set the
293 PF_MEMALLOC flag on task structures in order to allocate memory
294 from reserved pools and complete IO.
295 Use GFP_HIGHUSER for OST allocations for non-local client writes,
296 so that the OST threads generate memory pressure and allow
297 inactive pages to be reclaimed.
299 Severity : enhancement
301 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
306 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
307 Details : move ldlm namespace creation in setup phase to avoid grab
308 _lprocfs_lock with cli_sem held.
312 Description: Temporarily disable grant shrink.
313 Details : Disable the feature for debugging.
316 Frequency : only when formatting test filesystems
318 Description: Unable to run several mkfs.lustre on loop devices at the same time
319 Details : mkfs.lustre returns error 256 on the concurrent loop devices
320 formatting. The solution is to proper handle the error.
322 Severity : enhancement
324 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
326 Severity : enhancement
328 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
333 Description: resolve race between obd_disconnect and class_disconnect_exports
334 Details : if obd_disconnect will be called to already disconnected export he
335 forget release one reference and osc module can't unloaded.
337 Severity : enhancement
339 Description: Allow tuning service thread via /proc
340 Details : For each service a new
341 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
342 created that can be used to set min/max thread counts, and get the
343 current number of running threads.
348 Descriptoin: don't return error if have partially created objects for file.
349 Details : lov_update_create_set uses set->set_success as index for created
350 objects, so if some requests will be failed, they will have hole
351 at end of array and qos_shrink_lsm for allocate correct lsm.
353 Severity : enhancement
355 Description: Update OFED support to 1.4.1
360 Description: Reduce small size read RPC
361 Details : Set read-ahead limit for every file and only do read-ahead when
362 available read-ahead pages are bigger than 1M to avoid small size
365 Severity : enhancement
367 Description: move AT tunable parameters for more consistent usage
368 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
370 Severity : enhancement
372 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
373 Details : allow skip disconnected ost for send statfs request and hide error
379 Description: fix racy locking of mballoc block bitmaps causing BUG
380 Details : The locking of the mballoc buddy bitmap and the in-memory
381 block bitmap was using two different spin locks in some
382 cases. This made it possible to incorrectly access the
383 mballoc bitmap while another process was modifying it,
384 causing a sanity assertion to fail. While no on-disk corruption
385 was reported, there was some risk of this happening.
388 Frequency : rare, on llog test 6
390 Descriptoin: don't allow connect to already connected import
391 Details : allowing connect to already connected import is hide connecting problem.
396 Descriptoin: allow kill process which wait statahead result
397 Details : for some reasons 'ls' can stick in waiting result from statahead,
398 in this case need way for kill this process.
400 Severity : enhancement
402 Description: Add state history info file, enhance import info file
403 Details : Track import connection state changes in a new osc/mdc proc file;
404 add overview-type data to the osc/mdc import proc file.
406 Severity : enhancement
408 Description: MDS create should not wait for statfs RPC while holding DLM lock.
410 Severity : enhancement
412 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
417 Descriptoin: OST grant too much space to client even there are not enough space.
418 Details : Client will shrink its grant cache to OST if there are no write
419 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
420 this grant cache if there are already not enough avaible space
421 (left_space < total_clients * 32M).
424 Frequency : start MDS on uncleanly shutdowned MDS device
426 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
427 Details : stay in waiting mds<>ost recovery finished produce random bugs
428 due race between two ll_sync thread for one lov target. send
429 ACTIVATE event only if connect realy finished and import have
433 Frequency : rare, connect and disconnect target at same time
435 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
436 Details : don't call obd_disconnect under lov_lock. this long time
437 operation and can block ptlrpcd which answer to connect request.
442 Descriptoin: don't lose wakeup for imp_recovery_waitq
443 Details : recover_import_no_retry or invalidate_import and import_close can
444 both sleep on imp_recovery_waitq, but we was send only one wakeup
448 Frequency : always with long access acl
450 Descriptoin: mds can't pack reply with long acl.
451 Details : mds don't control size of acl but they limited by reint/getattr
454 Severity : enhancement
456 Description: Update to SLES10 kernel-2.6.16.60-0.33.
458 Severity : enhancement
460 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
463 Frequency : start MDS on uncleanly shutdowned MDS device
465 Descriptoin: aborting recovery hang on MDS
466 Details : don't throttle destroy RPCs for the MDT.
469 Frequency : on remount
471 Description: external journal device not working after the remount
472 Details : clear dev_rdonly flag for external journal devices in
478 Description: shutdown vs evict race
479 Details : client_disconnect_export vs connect request race.
480 if client will evicted at this time - we start invalidate
481 thread without referece to import and import can be freed
485 Frequency : rare, need acl's on inode.
487 Description: client can't handle ost additional correctly
488 Details : if ost was added after client connected to mds client can have
489 hit lnet_try_match_md ... to big messages to wide striped files.
490 in this case need teach client to handle config events about add
491 lov target and update client max ea size at that event.
493 Severity : enhancement
495 Description: Changelogs
496 Details : Changelogs are a lightweight mechanism to track filesystem
497 metadata and namespace changes. The changelog is recorded
498 permanently on the MDTs, and is periodically "consumed" / purged
499 when records are no longer needed.
501 Severity : enhancement
503 Description: compact fld format with extents
504 Details : Store range of seq rather than every seq in FLD. Seq
505 controller update FLD rather than clients. In Case of CMD, mdt0
506 has FLD, all other metadata server act as non persistent proxy
507 for FLD queries and cache fld entries in fld cache.
512 Description: don't skip ost target if they assigned to file
513 Details : Drop slow OSCs if we can, but not for requested start idx.
514 This means "if OSC is slow and it is not the requested
515 start OST, then it can be skipped, otherwise skip it only
516 if it is inactive/recovering/out-of-space.
520 Description: more cleanup in mds_lov
521 Details : not send LOV EA under replay, we can't know about they size at this
522 time. Don't allow client connect to mds before any ost connected,
523 for avoid problems with LOV EA size and returning EIO to client.
525 Severity : enhancement
527 Description: Interoperability at server side (Disk interoperability)
529 Severity : enhancement
531 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
533 Severity : enhancement
535 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
537 Severity : enhancement
539 Description: New client IO stack (CLIO).
541 Severity : enhancement
543 Description: Commit on sharing. Eliminate inter-client dependencies between
544 uncommitted transactions by doing transaction commits.
545 Thereby clients may recovery independently.
548 Frequency : Create a symlink file with a very long name
550 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
551 Details : If there is no extra space in the request for early cancels,
552 ldlm_req_handles_avail() returns 0 instead of a negative value.
554 Severity : enhancement
556 Description: Add /proc entry for import status
557 Details : The mdc, osc, and mgc import directories now have
558 an import directory that contains useful import data for debugging
561 Severity : enhancement
563 Description: Re-disable certain /proc logging
564 Details : Enable and disable client's offset_stats, extents_stats and
565 extents_stats_per_process stats logging on the fly.
568 Frequency : Only on FC kernels 2.6.22+
570 Description: oops in statahead
571 Details : Do not drop reference count for the dentry from VFS when lookup,
572 VFS will do that by itself.
574 Severity : enhancement
576 Description: Generic /proc file permissions
577 Details : Set /Proc file permissions in a more generic way to enable non-
578 root users operate on some /proc files.
582 Description: Hitting mdc_commit_close() ASSERTION
583 Details : Properly handle request reference release in
584 ll_release_openhandle().
588 Description: quota recovery deadlock during mds failover
589 Details : This patch includes att18982, att18236, att18237 in bz14840.
591 1. fix osts hang when mds does failover with quotaon
592 2. prevent watchdog storm when osts threads wait for the
597 Frequency : only patchless client
598 Description: add workaround for race between add/remove dentry from hash
600 Severity : enhancement
602 Description: Allow OST glimpses to return PW locks
606 Description: LBUG when llog conf file is full
607 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
611 Description: Prevent import from entering FULL state when server in recovery
615 Description: service mount cannot take device name with ":"
616 Details : Only when device name contains ":/" will mount treat it as
622 Description: replace ptlrpcd with the statahead thread to interpret the async
623 statahead RPC callback
627 Frequency : on recovery
628 Description: I/O failures after umount during fail back
629 Details : if client reconnected to restarted server we need join to recovery
630 instead of find server handler is changed and process self
631 eviction with cancel all locks.
633 Severity : enhancement
635 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
637 Severity : enhancement
639 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
641 Severity : enhancement
643 Description: Upcall on Lustre log has been dumped
644 Details : Allow for a user mode script to be called once a Lustre log has
645 been dumped. It passes the filename of the dumped log to the
646 script, the location of the script can be specified via
647 /proc/sys/lnet/debug_log_upcall.
652 Description: avoid idr_remove called for id which is not allocated.
653 Details : Move assigment s_dev for clustered nfs to end of initialization,
654 to avoid problem with error handling.
659 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
660 Details : When connection is reused this not moved from CONN_UNUSED_HASH
661 into CONN_USED_HASH and this prodice warning when put connection
662 again in unused hash.
665 Severity : enhancement
667 Description: File striping can now be set to use an arbitrary pool of OSTs.
669 Severity : enhancement
671 Description: Export bytes_read/bytes_write count on OSC/OST.
675 Description: Early reply size mismatch, MGC loses connection
676 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
677 the connect flags are properly negotiated.
682 Description: quotas are not honored with O_DIRECT
683 Details : all writes with the flag O_DIRECT will use grants which leads to
684 this problem. Now using OBD_BRW_SYNC to guard this.
688 Description: add quota statistics
689 Details : 1. sort out quota proc entries and proc code.
690 2. add quota statistics
692 Severity : enhancement
694 Description: enable quota support for HEAD.
698 Description: Properly propagate oinfo flags from lov to osc for statfs
699 Details : restore missing copy oi_flags to lov requests.
701 Severity : enhancement
703 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
704 : l_getgroups(8), lst(8), routerstat(8)
705 Details : included man pages for llobdstat(8), llstat(8),
706 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
708 Severity : enhancement
710 Description: configure's --enable-quota should check the
711 : kernel .config for CONFIG_QUOTA
712 Details : configure is terminated if --enable-quota is passed but
713 : no quota support is in kernel
717 Description: Remove portals compatibility
718 Details : Remove portals compatibility, not interoperable with releases
723 Description: Resolve device initialization race
724 Details : Prevent proc handler from accessing devices added to the
725 obd_devs array but yet be intialized.
727 Severity : enhancement
729 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
731 Severity : enhancement
733 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
738 Description: Using wrong pointer in osc_brw_prep_request
739 Details : Access to array[-1] can produce panic if kernel compiled with
740 CONFIG_PAGE_ALLOC enabled
744 Description: Client runs out of low memory
745 Details : Consider only lowmem when counting initial number of llap pages
749 Description: *optional* service tags registration
750 Details : if the "service tags" package is installed on a Lustre node
751 When the filesystem is mounted, a local-node service tag will
752 be created. See http://inventory.sun.com/ for more information
753 about the Service Tags asset management system.
757 Description: Kernel BUG tries to release flock
758 Details : Lustre does not destroy flock lock before last reference goes
759 away. So always drop flock locks when client is evicted and
760 perform unlock regardless of successfulness of speaking to MDS.
764 Description: add refcount protection for osc callbacks, avoid panic on shutdown
768 Description: sanity test 65a fails if stripecount of -1 is set
769 Details : handle -1 striping on filesystem in ll_dirstripe_verify
774 Description: ASSERTION(CheckWriteback(page,cmd)) failed
775 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
778 Severity : enhancement
780 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
784 Description: do not process already freed flock
785 Details : flock can possibly be freed by another thread before it reaches
786 to ldlm_flock_completion_ast.
790 Description: LBUG during stress test
791 Details : Need properly lock accesses the flock deadlock detection list.
795 Description: oops in page fault handler
796 Details : kernel page fault handler can return two special 'pages' in error
797 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
801 Description: timeout with invalidate import.
802 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
803 should be handled by ptlrpcd. This produce long age waiting and
804 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
806 Severity : enhancement
808 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
812 Description: enable MGS and MDT services start separately
813 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
814 a co-located MGS without starting the MGS, which is a complement
815 to 'nosvc' mount option.
819 Frequency : after recovery
820 Description: precreate to many object's after del orphan.
821 Details : del orphan st in oscc last_id == next_id and this triger growing
822 count of precreated objects. Set flag LOW to skip increase count
823 of precreated objects.
827 Frequency : rare, on clear nid stats
828 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
829 Details : when clean nid stats sometimes try destroy live entry,
830 and this produce panic in free.
834 Description: Stack overflow during MDS log replay
835 ease stack pressure by using a thread dealing llog_process.
839 Description: wait until IO finished before start new when do lock cancel.
840 Details : VM protocol want old IO finished before start new, in this case
841 need wait until PG_writeback is cleared until check dirty flag and
842 call writepages in lock cancel callback.
844 Severity : enhancement
846 Description: using special macro for print time and cleanup in includes.
850 Description: mds_mfd_close() ASSERTION(rc == 0)
851 Details : In mds_mfd_close(), we need protect inode's writecount change
852 within its orphan write semaphore to prevent possible races.
856 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
860 Frequency : rare, on shutdown ost
861 Description: don't hit live lock with umount ost.
862 Details : shrink_dcache_parent can be in long loop with destroy dentries,
863 use shrink_dcache_sb instead.
867 Description: don't panic with use echo client
868 Details : echo client pass NULL as client nid pointer and this produce null
873 Description: fix build on ppc32
874 Details : compile code with -m64 flag produce wrong object file for ppc32.
878 Description: add message levels for liblustreapi
882 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
883 Details : Move the 'good_osts' check before the 'total_bavail' check. This
884 will result in an -EAGAIN and in the exit call path we call
885 alloc_rr() which will with increasing aggressiveness attempt to
886 aquire precreated objects on the minimum number of required OSCs.
890 Description: Use old size assignment to avoid deadlock
891 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
892 the scheduling while holding a spinlock. We do not need locking
893 for size in ll_update_inode() because size is only updated from
894 the MDS for directories or files without objects, so there is no
895 other place to do the update, and concurrent access to such inodes
896 are protected by the inode lock.
900 Description: resolve "_IOWR redefined" build error on SLES10
904 Description: dump the memory debugging after all modules are unloaded to
905 suppress false negative in conf_sanity test 39
907 Severity : enhancement
909 Description: build kernel-ib packages for OFED 1.3 in our release cycle
914 Description: fix SLES kernel versioning
915 Details : the kernel version for our SLES 10 kernel did not include a "-"
916 before the "smp" at the end. while this was not a problem in
917 general, it did mean that software trying to use the kernel
918 version to try to detect a vendor specific kernel would fail.
919 this was most evident by the OFED build scripts.
923 Description: Don't update lov_desc members until making sure they are valid
924 Details : When updating lov_desc members via proc fs, need fix their
925 validities before doing the real update.
929 Description: don't put request into delay list while invalidate in flight.
930 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
931 import in flight. this produce timeout for invalidate and sometimes
932 can cause stale data.
936 Frequency : on ppc only
937 Description: not convert ost objects for directory because it's not exist.
938 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
942 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
944 Severity : enhancement
946 Description: Update to support 2.6.22.14 vanilla kernel.
950 Frequency : rare, on recovery
951 Description: read procfs can produce deadlock in some situation
952 Details : Holding lprocfs lock which send rpc can produce block for destroy
953 obd objects and this also block reconnect with -EALREADY. This
954 isn't fix all lprocfs bugs - but make it rare.
956 Severity : enhancement
958 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
961 Frequency : frequent on X2 node
963 Description: mdc_set_open_replay_data LBUG
964 Details : Set replay data for requests that are eligible for replay.
968 Description: lustre_mgs: operation 101 on unconnected MGS
969 Details : When MGC is disconnected from MGS long enough, MGS will evict the
970 MGC, and late on MGC cannot successfully connect to MGS and a lot
971 of the error messages complaining that MGS is not connected.
975 Frequency : on network error
976 Description: panic with double free request if network error
977 Details : mdc_finish_enqueue is finish request if any network error ocuring,
978 but it's true only for synchronus enqueue, for async enqueue
979 (via ptlrpcd) this incorrect and ptlrpcd want finish request
982 Severity : enhancement
984 Description: client-side metadata stat-ahead during readdir(directory readahead)
985 Details : perform client-side metadata stat-ahead when the client detects
986 readdir and sequential stat of dir entries therein
989 Frequency : on start mds
991 Description: Implement get_info(last_id) in obdfilter.
994 Frequency : occasional
996 Description: Correctly check stale fid, not start epoch if ost not support SOM
997 Details : open with flag O_CREATE need set old fid in op_fid3 because
998 op_fid2 was overwritten with new generated fid, but mds can answer
999 with one of these two fids and both is not stale. Setattr
1000 incorrectly started an epoch and assume will be called
1001 done_writing, but without SOM done_writing ever being called.
1004 Frequency : rare, depends on device drivers and load
1006 Description: MDS or OSS nodes crash due to stack overflow
1007 Details : Code changes in 1.8.0 increased the stack usage of some functions.
1008 In some cases, in conjunction with device drivers that use a lot
1009 of stack the MDS (or possibly OSS) service threads could overflow
1010 the stack. One change which was identified to consume additional
1011 stack has been reworked to avoid the extra stack usage.
1014 Frequency : occasional
1016 Description: Do not fail import if osc_interpret_create gets -EAGAIN
1017 Details : If osc_interpret_create got -EAGAIN it immediately exits and
1018 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
1019 oscc_has_objects and see OSC has no objests and call
1020 oscc_internal_create to resend create request.
1022 Severity : enhancement
1024 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1026 Severity : enhancement
1028 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1031 Frequency : very rare
1033 Description: Fix replay if there is an un-replied request and open
1034 Details : In some cases, older replay request will revert the
1035 mcd->mcd_last_xid on MDS which is used to record the client's
1036 latest sent request.
1038 Severity : enhancement
1040 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1042 Severity : enhancement
1044 Description: Add rhel5 support to HEAD.
1046 Serverity : enhancement
1048 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1053 Description: Don't allow skipping OSTs if index has been specified.
1054 Details : Don't allow skipping OSTs if index has been specified, make
1055 locking in internal create lots better.
1059 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1060 Details : access to bitfield in structure is always rounded to long
1061 and this produce problem with not atomic change any bit.
1065 Description: Lustre make rpms failed.
1066 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1067 builds Lustre rpms from tarball.
1070 Frequency : rare on shutdown ost
1072 Description: If llog cancel was not send before clean_exports phase, this can
1073 produce deadlock in llog code.
1074 Details : If llog thread has last reference to obd and call class_import_put
1075 this produce deadlock because llog_cleanup_commit_master wait when
1076 last llog_commit_thread exited, but this never success because was
1077 called from llog_commit_thread.
1081 Description: allow userland application know is lost one of stripes.
1082 Details : fill lvb_blocks with error code on ost and return it to
1083 application if error flag found.
1087 Description: NULL lov_tgts causing MDS oops
1088 Details : more safe checks for NULL lov_tgts for avoid oops.
1090 Severity : enhancement
1092 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1096 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
1097 Details : If an OST is down the MDS will hang indefinitely in
1098 obd_statfs_rqset() waiting for the statfs data. While for
1099 MDS QOS usage of statfs, it should not stuck in waiting.
1101 Severity : enhancement
1103 Description: remote_acl support
1104 Details : Support ACL-based permission check for remote user.
1105 Support setfacl/getfacl for remote user with the utils
1106 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
1107 the system "{s,g}etfacl" utils.
1109 Severity : enhancement
1111 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
1113 Severity : enhancement
1115 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1119 Description: make mgs_setparam() handle fsname containing dash
1120 Details : fsname containing a dash does not work with lctl conf_param
1122 Severity : enhancement
1124 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1126 Severity : enhancement
1128 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1132 Frequency : rare, at shutdown
1133 Description: access already free / zero obd_namespace.
1134 Details : if client_disconnect_export was called without force flag set,
1135 and exist connect request in flight, this can produce access to
1136 NULL pointer (or already free pointer) when connect_interpret
1137 store ocd flags in obd_namespace.
1141 Frequency : only at startup
1142 Description: not alloc memory with spinlock held.
1143 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1144 if any spinlock held.
1146 Severity : enhancement
1148 Description: make lustre randomly fail allocating memory
1149 Details : Make lustre randomly failed allocating memory for testing purpose.
1151 Severity : enhancement
1153 Description: lost problems with lov objid file
1154 Details : Fixes some scability and access to not inited memory problems
1155 in work with lov objdid file.
1160 Description: lfs find does not continue on file error
1161 Details : Continue other files processing when a file/dir is absent.
1165 Description: Inconsistent usage of lustre_pack_reply()
1166 Details : Standardize the usage of lustre_pack_reply() such that it
1167 always generate a CERROR on failure.
1170 Frequency : occasional
1172 Description: MDS hang or stay in waiting lock
1173 Details : If client receive lock with CBPENDING flag ldlm need send lock
1174 cancel as separate rpc, to avoid situation when cancel request
1175 can't processed due all i/o threads stay in wait lock.
1179 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1183 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1187 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1191 Description: Update to RHEL4 latest kernel.
1193 Severity : enhancement
1195 Description: Build SLES10 patchless client fails
1196 Details : The configure was broken by run ./configure with
1197 --with-linux-obj=.... argument for patchless client. When the
1198 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1199 can't search header adequately. Use absolute path such as
1200 -I($LINUX)/include instead.
1204 Description: interrupt oig_wait produce painc on resend.
1205 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1206 private set, and this produce situation when rq_ptlrpcd_data not
1207 copyed to new allocated request and triggered LBUG on assert
1208 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1209 ptlrpcd set if request is changed and can be safety changed to use
1214 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1215 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1216 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1217 these macros and replace them with their difinitions instead.
1221 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1222 Details : Update which_patch & target file for SLES10 latest kernel.
1225 Description: Remove client patches from SLES 10 kernel.
1226 Details : This causes SLES 10 clients to behave as patchless clients
1227 even on a Lustre-patched (server) kernel.
1229 Severity : enhancement
1231 Description: self-adjustable client's lru lists
1232 Details : use adaptive algorithm for managing client cached locks lru
1233 lists according to current server load, other client's work
1234 pattern, memory activities, etc. Both, server and client
1235 side namespaces provide number of proc tunables for controlling
1238 Severity : enhancement
1240 Description: light-weight GSS support
1241 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1242 and reduce performance overhead.
1244 Severity : enhancement
1246 Description: Linux keyring support
1247 Details : Support using service of Linux keyring for Lustre GSS internal
1248 context refresh/cache mechanism.
1252 Description: Fix errors in lfs documentation
1253 Details : Fixes man pages
1257 Description: don't use GFP_* in generic Lustre code.
1258 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1262 Description: obdclass is limited by single OBD_ALLOC(idarray)
1263 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1267 Description: warnings with build patchless client with vanila 2.6.19 and up
1268 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1269 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1273 Frequency : Only for SLES
1274 Description: sanity_quota fail test_1
1275 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1276 file, which makes TSTID[2] inunique.
1280 Frequency : Only for SLES9
1281 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1282 Details : Update target/ChangeLog/which_patch .
1286 Frequency : Only for test_10 in sanity-quota.sh
1287 Description: a bug in quota test code
1288 Details : checking a test flag in a wrong place
1292 Frequency : Only for test_13 in sanity-quota.sh
1293 Description: a bug in quota test script
1294 Details : shouldn't check overfull allocated quota
1298 Description: jbd statistics
1299 Details : Port older jbd statistics patch for sles10
1301 Severity : enhancement
1303 Description: Add printing inode info into message about error in writepage.
1308 Description: reply_lock_interpret crash due to race with it and lock cancel.
1309 Details : Do not replay locks that are being cancelled. Do not reference
1310 locks by their address during replay, just by their handle.
1314 Frequency : When flocks are used.
1315 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1316 Details : Flock locks might destroy just granted lock if it could be merged
1317 with another existing flock, this is done in completion handler,
1318 so teach ldlm_cli_enquque_fini that this is a valid case for
1324 Description: Oops in read and write path when failing to allocate lock.
1325 Details : Check if lock allocation failed and return error back.
1329 Description: lstripe command fails for valid OST index
1330 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1331 instead of lov->desc.ld_active_tgt_count.
1335 Description: sanity.sh failed test 103
1336 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1341 Description: df doesn't work properly if diskfs blocksize != 4K
1342 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1346 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1347 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1349 Severity : enhancement
1351 Description: Adaptive timeouts
1352 Details : RPC timeouts adapt to changing server load and network
1353 conditions to reduce resend attempts and improve recovery time.
1357 Description: llapi_file_create() does not allow some changes
1358 Details : add llapi_file_open() that allows specifying the mode and
1359 open flags, and also returns an open file handle.
1363 Description: merge and cleanup kernel patches.
1364 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1368 Description: Add journal checksum support.(Kernel part)
1369 Details : The journal checksum feature adds two new flags i.e
1370 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1371 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1372 indicates that the commit block contains the checksum for
1373 the blocks described by the descriptor blocks. Now commit
1374 record can be sent to disk without waiting for descriptor
1375 blocks to be written to disk. This behavior is controlled
1376 using JBD2_FEATURE_ASYNC_COMMIT flag.
1380 Description: OSS needs mutliple precreate threads
1381 Details : Add ability to start more than one create thread per OSS.
1385 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1386 Details : execute lfs setstripe on client
1390 Description: mds_obd_create error creating tmp object
1391 Details : When the user sets quota on root, llog will be affected and can't
1392 create files and write files.
1395 Frequency : Always on ia64 patchless client, and possibly others.
1397 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1398 Details : This allows the patchless client to be loaded on architectures
1399 without this export.
1403 Description: RedHat Update kernel for RHEL5
1404 Details : Add the kernel config file for RHEL5.
1408 Description: RedHat Update kernel for RHEL5
1409 Details : Modify the kernel config file more closer RHEL5.
1413 Description: Build failure against Centos5 (RHEL5)
1414 Details : Define PAGE_SIZE when it isn't present.
1418 Description: test_7 on sanity-quota.sh can't be used on separate machines
1419 Details : add facet to handle it
1423 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1424 Details : with NFS, the anon dentry's parent was set to itself in
1425 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1426 corresponding dentry other than use rec->ur_name.
1428 Severity : enhancement
1430 Description: lfs setstripe enhancement
1431 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1435 Description: enable data checksumming by default
1436 Details : enable checksum by default, allow --disable-checksum
1437 configure option and "-o nochecksum" mount option. Checksums
1438 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1442 Description: lustre support for RHEL5
1443 Details : Add support for RHEL5.
1447 Description: Client eviction due to failover config
1448 Details : after a connection loss, the lustre client should attempt to
1449 reconnect to the last active server first before trying the
1450 other potential connections.
1454 Description: when mds and osts use different quota unit(32bit and 64bit),
1455 quota will be released repeatly.
1456 Details : void sending multiple quota reqs to mds, which will keep the
1457 status between the reqs.
1461 Description: osts not allocated evenly to files
1462 Details : change the condition to increase offset_idx
1465 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1467 Description: Data corruption for OSTs that are formatted larger than 2TB
1469 Details : When generating the bio request for lustre file writes the
1470 sector number would overflow a temporary variable before being
1471 used for the IO. The data reads correctly from Lustre (which
1472 will overflow in a similar manner) but other file data or
1473 filesystem metadata may be corrupted in some cases.
1477 Description: Tune the kernel for good SCSI performance.
1478 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1479 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1484 Description: rewrite ext2-derived code in obdclass/uuid.c
1485 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1486 from scratch preserving functionality.
1490 Description: "lfs find -obd UUID" prints directories
1491 Details : "lfs find -obd UUID" will return all directory names instead
1492 of just file names. It is incorrect because the directories
1493 do not reside on the OSTs.
1497 Description: use i_size_read and i_size_write in 2.6 port
1498 Details : replace inode->i_size access with i_size_read/write()
1501 Frequency : when removing large files
1503 Description: scheduling issue during removal of large Lustre files
1504 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1505 It causes scheduling issues when removing large files (17TB in the
1509 Frequency : only with liblustre clients on XT3
1511 Description: evictions taking too long
1512 Details : allow llrd to evict clients directly on OSTs
1515 Frequency : only on ppc
1517 Description: /proc/fs/lustre/devices broken on ppc
1518 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1519 We should make sure the type of 'index' is loff_t and then cast
1520 explicitly as needed below. Do not assign an explicitly cast
1524 Frequency : only for rhel5
1526 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1527 Details : Modify the target file & which_kernel.
1529 Severity : enhancement
1531 Description: omit set fsid for export NFS
1532 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1533 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1537 Frequency : Always, for kernels after 2.6.16
1538 Description: Fix warning idr_remove called for id=.. which is not allocated.
1539 Details : Last kernels save old s_dev before kill super and not allow
1540 to restore from callback - restore it before call kill_anon_super.
1544 Description: buffer overruns could theoretically occur
1545 Details : llapi_semantic_traverse() modifies the "path" argument by
1546 appending values to the end of the origin string, and a
1547 overrun may occur. Adding buffer overrun check in liblustreapi.
1551 Description: Fix error on 'ls .' at the top of the Lustre mount.
1552 Details : Don't revalidate dentry if it is a root dentry.
1556 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1557 Details : Modify vm-tunables-rhel4.patch.
1561 Description: Kernel config for 2.6.18-vanilla.
1562 Details : Modify targets/2.6-vanilla.target.in.
1563 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1564 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1565 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1566 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1570 Description: improve handling recoverable errors
1571 Details : if request processig with error which can be recoverable on server
1572 request should be resend, otherwise page released from cache and
1577 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1578 Details : Modify target file & which_patch.
1579 A flaw was found in the IA32 system call emulation provided
1580 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1581 value could be stored in the %RAX register, which could trigger an
1582 out-of-bounds system call table access. An untrusted local user
1583 could exploit this flaw to run code in the kernel
1584 (ie a root privilege escalation). (CVE-2007-4573).
1588 Description: O_DIRECT bypasses client statistics.
1589 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1590 read_ahead_stats were updated. copied the stats section from
1591 osc_send_oap_rpc() into async_internal().
1595 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1599 Description: change order of libsysio includes
1600 Details : '#include sysio.h' should always come before '#include xtio.h'
1604 Description: handle "serious error: objid * already exists" more gracefully
1605 Details : If LAST_ID value on disk is smaller than the objects existing in
1606 the O/0/d* directories, it indicates disk corruption and causes an
1607 LBUG(). If the object is 0-length, then we should use the existing
1608 object. This will help to avoid a full fsck in most cases.
1610 Severity : enhancement
1612 Description: adapt the lustre_config script to support the upgrade case
1613 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1614 server targets to 1.6 in parallel.
1618 Description: To avoid grant space > avaible space when the disk is almost
1619 full. Without this patch you might see the error "grant XXXX >
1620 available" or some LBUG about grant, when the disk is almost
1622 Details : In filter_check_grant, for non_grant cache write, we should
1623 check the left space by if (*left > ungranted + bytes), instead
1624 of (*left > ungranted), because only we are sure the left space
1625 is enough for another "bytes", then the ungrant space should be
1626 increase. In client, we should update cl_avail_grant only there
1627 is OBD_MD_FLGRANT in the reply.
1631 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1632 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1633 described in CVE-2007-4573. This problem could allow untrusted
1634 local users to gain root access.
1637 Frequency : when using O_DIRECT and quotas
1639 Description: Incorrect file ownership on O_DIRECT output files
1640 Details : block usage reported by 'lfs quota' does not take into account
1641 files that have been written with O_DIRECT.
1646 Description: touch file failed when fs is not full
1647 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1648 otherwise we can get ENOSP while fs is not full.
1652 Description: parallel lock callbacks
1653 Details : Instead of sending blocking and completion callbacks as separated
1654 requests, adding them to a set and sending in parallel.
1657 Frequency : only for Cray XT3
1658 Bugzilla : 12829/13455
1659 Description: Changing primary group doesn't change the group lustre assigns to
1661 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1662 overridden with the primary group provided by the group upcall,
1663 whereas the supplied fsgid can be trusted if it is in the list of
1664 supplementary groups returned by the group upcall.
1666 Severity : enhancement
1668 Description: Allow masking D_WARNING, D_ERROR messages from console
1669 Details : Console messages can now be disabled via lnet.printk.
1673 Description: User code with malformed file open parameter crashes client node
1674 Details : Before packing join_file req, all the related reference should be
1675 checked carefully in case some malformed flags cause fake
1676 join_file req on client.
1680 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1681 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1682 mode is equal to the granted mode. However, it should also check
1683 that LDLM_FL_CP_REQD is not set.
1687 Description: Heavy nfs access might result in deadlocks
1688 Details : After ELC code landed, it is now improper to enqueue any mds
1689 locks under och_sem, because enqueue might want to decide to
1690 cancel open locks for same inode we are holding och_sem for.
1694 Description: Client eviction while running blogbench
1695 Details : A lot of unlink operations with concurrent I/O can lead to a
1696 deadlock causing evictions. To address the problem, the number of
1697 oustanding OST_DESTROY requests is now throttled to
1698 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1699 callbacks are processed in priority.
1703 Description: enable ACLs on MDS by default
1704 Details : ACLs must be enabled on MDS by default.
1707 Frequency : PPC/PPC64 only
1709 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1710 Details : fix duplicated definitions between asm-ppc64/types.h and
1711 lustre_types.h on PPC.
1714 Frequency : PPC/PPC64 only
1716 Description: asm-ppc/segment.h does not exist
1717 Details : fix compile issue on PPC.
1721 Description: better handle error messages in extents code
1724 Frequency : RHEL4 only
1726 Description: mkfs is very slow on IA64/RHEL4
1727 Details : A performance regression has been discovered in the MPT Fusion
1728 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1729 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1730 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1732 Severity : enhancement
1734 Description: SNMP support enhancement
1735 Details : Adding total number of sampled request for an MDS node in snmp
1738 Severity : enhancement
1740 Description: Optimize ldlm waiting list processing for PR extent locks
1741 Details : When processing waiting list for read extent lock and meeting read
1742 lock that is same or wider to it that is not contended, skip
1743 processing rest of the list and immediatelly return current
1744 status of conflictness, since we are guaranteed there are no
1745 conflicting locks in the rest of the list.
1749 Description: Time out and refuse to reconnect
1750 Details : When the failover node is the primary node, it is possible
1751 to have two identical connections in imp_conn_list. We must
1752 compare not conn's pointers but NIDs, otherwise we can defeat
1753 connection throttling.
1757 Description: port llog fixes from b1_6 into HEAD
1758 Details : Port llog reference couting and some llog cleanups from b1_6
1759 (bug 10800) into HEAD, for protect from panic and access to already
1760 free llog structures.
1764 Description: Detect stride IO mode in read-ahead
1765 Details : When a client does stride read, read-ahead should detect that and
1766 read-ahead pages according to the detected stride pattern.
1770 Description: data checksumming impacts single node performance
1771 Details : add support for several checksum algorithm. Currently, only CRC32
1772 and Adler-32 are supported. The checksum type can be changed on
1773 the fly via /proc/fs/lustre/osc/*/checksum_type.
1777 Description: use adler32 for page checksums
1778 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1783 Description: build for x2 fails
1784 Details : fix compile issue on Cray systems.
1788 Description: Properly match for duplicate locks
1789 Details : Due to different lock order from skiplists code, we need to
1790 traverse entire list for now
1793 Frequency : only on PPC/SLES10
1795 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1796 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1797 be defined. Add a hack in mkfs_lustre.c to work around this compile
1802 Description: LASSERT on MDS when client holding flock lock dies
1803 Details : ldlm pool logic depends on number of granted locks equal to
1804 number of released locks which is not true for flock locks, so
1805 just exclude such locks from consideration.
1809 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1810 Details : Use fsfilt_sync() for both the whole filesystem sync and
1811 individual file sync to eliminate dangerous inode locking
1812 with I_LOCK that can lead to a deadlock.
1816 Description: performance in 1.6.3
1817 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1818 sure that 1MB requests are not fragmented by the block layer.
1820 Severity : enhancement
1822 Description: organize the server-side client stats on per-nid basis
1823 Details : Change the structure of stats under obdfilter and mds to
1831 The "uuid"s file would list the uuids of _active_ exports.
1832 And the clear entry is to clear all stats and stale nids.
1834 Severity : enhancement
1836 Description: eliminate client locks in face of contention
1837 Details : file contention detection and lockless i/o implementation
1838 for contended files.
1842 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1845 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1847 Description: MDS may lose file striping (and hence file data) in some cases
1848 Details : If there are additional extended attributes stored on the MDS,
1849 in particular ACLs, SELinux, or user attributes (if user_xattr
1850 is specified for the client mount options) then there is a risk
1851 of attribute loss. Additionally, the Lustre file striping
1852 needs to be larger than default (e.g. striped over all OSTs),
1853 and an additional attribute must be stored initially in the
1854 inode and then increase in size enough to be moved to the
1855 external attribute block (e.g. ACL growing in size) for file
1860 Description: skiplist implementation simplification
1861 Details : skiplists are used to group compatible locks on granted list
1862 that was implemented as tracking first and last lock of each
1863 lock group the patch changes that to using doubly linked lists
1867 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1868 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1869 because LASSERT on both the data supplied by a client, and the
1870 data on disk is dangerous and incorrect.
1872 Severity : enhancement
1874 Description: Slow truncate/writes to huge files at high offsets.
1875 Details : Directly associate cached pages to lock that protect those pages,
1876 this allows us to quickly find what pages to write and remove
1877 once lock callback is received.
1881 Description: more ldlm soft lockups
1882 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1883 starve other threads from the resource lock for a long time in
1884 case of long waiting queue, so change the debug level from
1885 D_OTHER to the less frequently used D_INFO.
1887 Severity : enhancement
1889 Description: add -gid, -group, -uid, -user options to lfs find
1893 Description: Hung threads in invalidate_inode_pages2_range
1894 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1895 once one is completed. As a result, some RPCs are stuck in the
1896 queue and are never sent.
1900 Description: filter threads hungs on waiting journal commit
1901 Details : Cleanup filter group llog code, then only filter group llog will
1902 be only created in the MDS/OST syncing process.
1906 Description: Procfs and llog threads access destoryed import sometimes.
1907 Details : Sync the import destoryed process with procfs and llog threads by
1908 the import refcount and semaphore.
1910 Severity : enhancement
1912 Description: openlock cache of b1_6 port to HEAD
1917 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1918 Details : If insertion of an extent fails, then discard the inode
1919 preallocation and free data blocks else it can lead to duplicate
1924 Description: don't always update ctime in ext3_xattr_set_handle()
1925 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1926 In some cases the ctime should not be updated, for example for
1927 2.0->1.8 compatibility it is necessary to delete an xattr and it
1928 should not update the ctime.
1932 Bugzilla : 15713/16362
1933 Description: Assertion in iopen_connect_dentry in 1.6.3
1934 Details : looking up an inode via iopen with the wrong generation number can
1935 populate the dcache with a disconneced dentry while the inode
1936 number is in the process of being reallocated. This causes an
1937 assertion failure in iopen since the inode's dentry list contains
1938 both a connected and disconnected dentry.
1942 Description: assertion failure in ldlm_handle2lock()
1943 Details : fix a race between class_handle_unhash() and class_handle2object()
1944 introduced in lustre 1.6.5 by bug 13622.
1949 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1950 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1955 Description: oops in sd_iostats_seq_show()
1956 Details : unloading/reloading the scsi low level driver triggers a kernel
1957 bug when trying to access the sd iostat file.
1962 Description: Kernel panics during QLogic driver reload
1963 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1964 patch, causing memory corruption.
1969 Description: journal_dev option does not work in b1_6
1970 Details : pass mount option during pre-mount.
1972 Severity : enhancement
1974 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1975 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1976 extent information of a file. It can be used to map logical blocks
1977 in a file to physical blocks in the block device.
1981 Description: LDLM soft lockups - improvement
1982 Details : It is be possible to send the lock handle along with each read
1983 or write request because the client is already doing a lock match
1984 itself so there isn't any reason the OST should have to re-do that
1988 Frequency : only on Cray X2
1990 Description: X2 build failures
1991 Details : fix build failures on Cray X2.
1995 Description: xid & resent requests
1996 Details : Initialize RPC XID from clock at startup (randomly if clock is
1999 Severity : enhancement
2001 Description: Add lustre_start utility to start or stop multiple Lustre servers
2006 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
2007 Details : In case of memory pressure, list_del() can be called twice on
2008 req->rq_history_list, causing a kernel oops.
2012 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
2013 Details : fix stack overflow in the distributed lock manager by defering
2014 export eviction after a failed AST to the elt thread instead of
2015 handling it in the dlm interpret routine.
2019 Description: Convert some comments to new format.
2020 Details : Update documenting comments to match doxygen conventions.
2024 Description: Grammar fixes.
2025 Details : A couple of trivial spelling fixes.
2029 Description: OSD_COUNTERS-mandatory
2030 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
2031 to simplify the code.
2035 Description: simplify cmm_device freeing logic.
2036 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
2040 Description: Add lockdep support to dt_object_operations locking interface.
2041 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
2042 parameter indicating lock ordering. Update mdd code to use new
2047 Description: Introduce failloc constants for lockless IO tests.
2048 Details : Add two new failloc constants to test lockless IO. Only one of
2049 them in implemented---another is checked in yet to be landed
2054 Description: Add lockdep support for inode mutex.
2055 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
2056 in the situations where Lustre has to lock more than one inode
2061 Description: Add optional invariants checking support.
2062 Details : Add new LINVRNT() macro, optional on new --enable-invariants
2063 configure switch. This macro is to be used for consistency and
2064 sanity checks that are too expensive to be left in `production'
2069 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
2070 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
2075 Description: Add lockdep support for ldlm_lock and ldlm_resource.
2076 Details : Use spin_lock_nested() in (the only) situation where more than
2077 one ldlm_lock is locked simultaneously. Also, fix possible
2078 dead-lock in ldlm_lock_change_resource() by enforcing particular
2083 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
2084 Details : Instead of specifying each ldlm_lock call-back through separate
2085 parameter, wrap them into struct ldlm_callback_suite.
2089 Description: Kill join_lru obd method and its callers.
2090 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
2091 in memory---a requirement implemented through ->o_join_lru() obd
2092 method in HEAD. Remove this method and its users.
2096 Description: Add asynchronous ldlm ENQUEUE completion handler.
2097 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
2098 case that stock ldlm_completion_ast() cannot handle as it waits
2099 until lock is granted. Introduce new ldlm_completion_ast_async()
2100 for this. Also comment ldlm_completion_ast().
2104 Description: ldlm_error <-> errno conversion.
2105 Details : Add functions to map (rather arbitrary) between LDLM error codes
2106 and standard errno values. CLIO needs this to prevent LDLM specific
2107 constants from escaping ldlm and osc.
2111 Description: Kill unused ldlm_handle2lock_ns() function.
2112 Details : Kill unused ldlm_handle2lock_ns() function.
2116 Description: Add lu_ref support to ldlm_lock
2117 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
2118 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
2119 and ldlm_resource. LDLM interface has to be changed, because code that
2120 releases a reference on a lock, has to "know" what reference this is.
2121 In the most frequent case
2123 lock = ldlm_handle2lock(handle);
2125 LDLM_LOCK_PUT(lock);
2127 no changes are required. When any other reference (received _not_ from
2128 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
2129 instead of LDLM_LOCK_PUT().
2131 Arguably, changes are pervasive, and interface requires some discipline
2132 for proper use. On the other hand, it was very instrumental in finding
2133 a few leaked lock references.
2137 Description: Add ldlm_lock_addref_try().
2138 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
2139 attempts to addref a lock that might be being canceled
2144 Description: Add ldlm_weigh_callback().
2145 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
2146 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
2147 hard-coded `number of pages' logic.
2151 Description: Add lockdep annotations to llog code.
2152 Details : Use appropriately tagged _nested() locking calls in the places
2153 where llog takes more than one ->lgh_lock lock.
2157 Description: Add loi_kms_set().
2158 Details : Wrap kms updates into a helper function.
2162 Description: Constify instances of struct lsm_operations.
2163 Details : Constify instances of struct lsm_operations.
2167 Description: lu_conf support.
2168 Details : On a server, a file system object is uniquely identified
2169 by a fid, which is sufficient to locate and load all object
2170 state (inode). On a client, on the other hand, more data are
2171 necessary instantiate an object. Change lu_object_find() and
2172 friends to take additional `lu_conf' argument describing object.
2173 Typically this includes layout information.
2177 Description: lu_context fixes.
2178 Details : Introduce new lu_context functions that are needed on the client
2179 side, where some system threads (ptlrpcd) are shared by multiple
2180 modules, and so cannot be stopped during module shutdown.
2184 Description: Add start and stop methods to lu_device_type_operations.
2185 Details : Introduce two new methods in lu_device_type_operations, that are
2186 invoked when first instance of a given type is created and last one
2187 is destroyed respectively. This is need by CLIO.
2191 Description: Add lu_ref support to struct lu_device.
2192 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2193 track leaked references.
2197 Description: Introduce lu_kmem_descr.
2198 Details : lu_kmem_descr and its companion interface allow to create
2199 and destroy a number of kmem caches at once.
2203 Description: Fix lu_object finalization race.
2204 Details : Fix a race between lu_object_find() finding an object and its
2205 concurrent finalization. This race is (most likely) not possible
2206 on the server, but might happen on the client.
2210 Description: Introduce lu_ref interface.
2211 Details : lu_ref is a debugging module allowing to track references to
2212 a given object. It is quite cpu expensive, and has to be
2213 explicitly enabled with --enable-lu_ref. See usage description
2218 Description: Factor lu_site procfs stats into a separate function.
2219 Details : Separate lu_site stats printing code into a separate function
2220 that can be reused on a client.
2224 Description: Constify instances of struct {lu,dt,md}_device_operations.
2225 Details : Constify instances of struct {lu,dt,md}_device_operations.
2229 Description: Introduce struct md_site and move meta-data specific parts of
2230 struct lu_site here.
2231 Details : Move md-specific fields out of struct lu_site into special struct
2232 md_site, so that lu_site can be used on a client.
2236 Description: Kill mdd_lov_destroy().
2237 Details : Remove unused mdd code.
2241 Description: Add st_block checking to multistat.c.
2242 Details : Add st_block checking to multistat.c.
2246 Description: Add lu_ref support to struct obd_device.
2247 Details : Add lu_ref tracking to obd_device.
2251 Description: Kill obd_set_fail_loc().
2252 Details : Remove unused code.
2256 Description: Add special type for ptlrpc_request interpret functions.
2257 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2258 there. Actual usage will be in CLIO.
2262 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2263 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2264 doesn't work with lockdep.
2268 Description: Add rwv.c test program.
2269 Details : New testing program exercising readv(2) and writev(2) (Qian).
2273 Description: Add sendfile.c test program.
2274 Details : New testing program exercising sendfile(2) (Jay).
2278 Description: Ratelimit a message that can be very frequent.
2279 Details : Ratelimit a memory allocation failure message that can
2284 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2285 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2289 Description: Liblustre build fixes.
2290 Details : Liblustre build fixes.
2294 Description: libcfs: add cfs_{need,cond}_resched() interface.
2295 Details : libcfs: add cfs_{need,cond}_resched() definition and
2296 implementations for Linux, NT, and liblustre.
2298 Severity : enhancement
2300 Description: More exported tunables for mballoc
2301 Details : Add support for tunable preallocation window and new tunables for
2302 large/small requests
2306 Description: Detect corruption of block bitmap and checking for preallocations
2307 Details : Checks validity of on-disk block bitmap. Also it does better
2308 checking of number of applied preallocations. When corruption is
2309 found, it turns filesystem readonly to prevent further corruptions.
2313 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2314 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2315 window to 0. If it does hit cache miss, and read-pattern is still
2316 stride-io mode, does not reset the stride window, but also does not
2317 increase the stride window length in this case.
2321 Frequency : only for big-endian servers
2322 Description: Check if system is big-endian while mounting fs with extents feature
2323 Details : Mounting a filesystem with extents feature will fail on big-endian
2324 systems since ext3-based ldiskfs is not supported on big-endian
2325 systems. This can be over-riden with "bigendian_extents" mount option.
2327 Severity : enhancement
2329 Description: The root squash functionality
2330 Details : A security feature, which is to prevent users from being able
2331 to mount lustre on their desktop, run as root, and delete
2332 all of the files in the filesystem. The goal is accomplished by
2333 remapping user id (UID) and group id (GID) of the root user to
2334 a UID and GID specified by the system administartor via Lustre
2335 configuration management server (MGS). The functionality also
2336 allows to specify sets of clients for which the remapping does
2341 Description: Excessive recovery window
2342 Details : With AT enabled, the recovery window can be excessively long (6000+
2343 seconds). To address this problem, we no longer use
2344 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2345 timeout no longer depends on the service time, it is set to
2346 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2347 time via pb_service_time.
2351 Description: Watchdog triggered on MDS failover
2352 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2353 the OSTs know that the MDS "UUID" can be reused for the same export
2354 from a different NID, so we do not need to wait for the export to be
2358 Frequency : rare, only if using MMP with Linux RAID
2360 Description: MMP doesn't work with Linux RAID
2361 Details : While using HA for Lustre servers with Linux RAID, it is possible
2362 that MMP will not detect multiple mounts. To make this work we
2363 need to unplug the device queue in RAID when the MMP block is being
2364 written. Also while reading the MMP block, we should read it from
2365 disk and not the cached one.
2367 Severity : enhancement
2369 Description: open file using fid
2370 Details : A file can be opened using just its fid, like
2371 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2374 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2377 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2378 Details : add proper locking for creation of jbd_4k slab cache
2382 Description: MMP check in ext3_remount() fails without displaying any error
2383 Details : When multiple mount protection fails during remount, proper error
2386 Severity : enhancement
2388 Description: Allow stripe size to be up to 4G-64k
2389 Details : Fix math logic to allow large stripe sizes.
2393 Description: add check for >8TB ldiskfs filesystems
2394 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2395 Don't allow >8TB ldiskfs filesystems to be mounted without
2396 force_over_8tb mount option
2401 Descriptoin: OSS complains "too many blocks" and crashes
2402 Details : In filter_commitrw_write():
2403 1. lquota_chkquota();
2404 2. obdo_from_inode();
2405 3. lquota_pending_commit();
2406 "2." has changed uid and gid in this case, then caused
2407 lquota_chkquota() and lquota_pending_commit() to use
2408 different uid and gid. That is the root of the bug.
2412 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2413 Details : Detect and print out the 'stripe count = -1' case properly.
2417 Description: Out or order replies might be lost on replay
2418 Details : In ptlrpc_retain_replayable_request if we cannot find retained
2419 request with tid smaller then one currently being added, add it
2420 to the start, not end of the list.
2422 --------------------------------------------------------------------------------
2424 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2426 * Support for kernels:
2427 2.6.5-7.283 (SLES 9),
2428 2.6.9-55.EL (RHEL 4),
2429 2.6.16.46-0.14 (SLES 10),
2430 2.6.18.8 vanilla (kernel.org)
2431 * Client support for unpatched kernels:
2432 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2433 2.6.16 - 2.6.22 vanilla (kernel.org)
2434 * Due to recently discovered recovery problems, we do not recommend
2435 using patchless RHEL 4 clients with this or any earlier release.
2436 * Recommended e2fsprogs version: 1.39.cfs8
2437 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2438 * Starting with this release, the ldiskfs backing filesystem required
2439 by Lustre is now in its own package, lustre-ldiskfs. This package
2440 should be installed. It is versioned separately from Lustre and
2441 may be released separately in future.
2445 Description: block reactivating mgc import until all deactivates complete
2446 Details : Fix race when failing back MDT/MGS to itself (testing)
2449 Frequency : at statup only
2451 Description: mds_lov_synchronize race leads to various problems
2452 Details : simultaneous MDT->OST connections at startup can cause the
2453 sync to abort, leaving the OSC in a bad state.
2455 Severity : enhancement
2457 Description: add optional extra BUILD_VERSION info
2458 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2459 to override the lustre version.
2462 Frequency : 2.6.18 servers only
2464 Description: ll_kern_mount() doesn't release the module reference
2465 Details : The ldiskfs module reference count never drops down to 0
2466 because ll_kern_mount() doesn't release the module reference.
2471 Description: server LBUG when using old ost_num_threads parameter
2472 Details : Accept the old ost_num_threads parameter but warn that it
2473 is deprecated, and fix an off-by-one error that caused an LBUG.
2478 Description: Transient SCSI error results in persistent IO issue
2479 Details : iobuf->dr_error is not reinitialized to 0 between two
2483 Frequency : sometimes when underlying device returns I/O errors
2485 Description: OSTs not going read-only during write failures
2486 Details : OSTs are not remounted read-only when the journal commit threads
2487 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2488 instead of the ext3 wrappers.
2492 Description: poor connect scaling with increasing client count
2493 Details : Don't run filter_grant_sanity_check for more than 100 exports
2494 to improve scaling for large numbers of clients.
2497 Frequency : SLES10 only
2499 Description: sanity-quota.sh quotacheck failed: rc = -22
2500 Details : Quotas cannot be enabled on SLES10.
2503 Frequency : liblustre clients only
2505 Description: getdirentries does not give error when run on compute nodes
2506 Details : getdirentries does not fail when the size specified as an argument
2507 is too small to contain at least one entry
2509 Severity : enhancement
2511 Description: Add LNET router traceability for debug purposes
2512 Details : If a checksum failure occurs with a router as part of the
2513 IO path, the NID of the last router that forwarded the bulk data
2514 is printed so it can be identified.
2519 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2520 Details : Due to a race condition, liblustre clients were occasionally
2521 evicted incorrectly.
2523 Severity : enhancement
2525 Description: lfs setstripe use optional parameters instead of postional
2528 Severity : enhancement
2530 Description: Nanosecond timestamp support for ldiskfs
2531 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2532 resolution timestamps. There is not yet support for this at
2533 the Lustre filesystem level.
2536 Frequency : during server recovery
2538 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2539 Details : request with rq_no_resend flag not awake l_wait_event if they get
2543 Frequency : nfs export on patchless client
2545 Description: connectathon hang when test nfs export over patchless client
2546 Details : Disconnected dentry cannot be found with lookup, so we do not need
2547 to unhash it or make it invalid
2550 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2551 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2552 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2557 Description: open req refcounting wrong on reconnect
2558 Details : If reconnect happened between getting open reply from server and
2559 call to mdc_set_replay_data in ll_file_open, we will schedule
2560 replay for unreferenced request that we are about to free.
2561 Subsequent close will crash in variety of ways.
2562 Check that request is still eligible for replay in
2563 mdc_set_replay_data().
2568 Description: disable writes to filesystem when reading health_check file
2569 Details : the default for reading the health_check proc file has changed
2570 to NOT do a journal transaction and write to disk, because this
2571 can cause reads of the /proc file to hang and block HA state
2572 checking on a healthy but otherwise heavily loaded system. It
2573 is possible to return to the previous behaviour during configure
2574 with --enable-health-write.
2576 Severity : enhancement
2578 Description: 64-bit inode version
2579 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2580 to the inode. This will be required for version-based recovery.
2585 Description: MDS fails to start if a duplicate client export is detected
2586 Details : in some rare cases it was possible for a client to connect to
2587 an MDS multiple times. Upon recovery the MDS would detect this
2588 and fail during startup. Handle this more gracefully.
2590 Severity : enhancement
2592 Description: Add -o localflock option to simulate old noflock
2594 Details : This will achieve local-only flock/fcntl locks
2600 Description: log_commit_thread vs filter_destroy race leads to crash
2601 Details : Take import reference before releasing llog record semaphore
2606 Description: Wrong request locking in request set processing
2607 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2608 imp_delayed_list, in this place should be used imp_lock.
2611 Frequency : when reconnection
2613 Description: Grant Leak when osc reconnect to OST
2614 Details : When osc reconnect ost, OST(filter) should check whether it
2615 should grant more space to client by comparing fed_grant and
2616 cl_avail_grant, and return the granted space to client instead
2617 of "new granted" space, because client will call osc_init_grant
2618 to update the client grant space info.
2621 Frequency : when client reconnect to OST
2623 Description: Grant Leak when osc do resend and replay bulk write
2624 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2625 bulk write request, because the grant info while be sync between
2626 OSC and OST when reconnect, and we should ignore the grant info
2627 these of resend/replay write req.
2632 Description: Grant space more than avaiable left space sometimes.
2633 Details : When then OST is about to be full, if two bulk writing from
2634 different clients came to OST. Accord the avaliable space of the
2635 OST, the first req should be permitted, and the second one
2636 should be denied by ENOSPC. But if the seconde arrived before
2637 the first one is commited. The OST might wrongly permit second
2638 writing, which will cause grant space > avaiable space.
2641 Frequency : when client is evicted
2643 Description: Grant might be wrongly erased when osc is evicted by OST
2644 Details : when the import is evicted by server, it will fork another
2645 thread ptlrpc_invalidate_import_thread to invalidate the
2646 import, where the grant will be set to 0. While the original
2647 thread will update the grant it got when connecting. So if
2648 the former happened latter, the grant will be wrongly errased
2649 because of this race.
2654 Description: Checking Stale with correct fid
2655 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2656 to check whether it is stale, because sometimes, we want the
2657 enqueue happened anyway, and op_data.fid2 will not be initialized.
2659 Severity : enhancement
2661 Description: update patchless client
2662 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2665 Frequency : only with 2.4 kernel
2667 Description: random memory corruption
2668 Details : size of struct ll_inode_info is to big for union inode.u and this
2669 can be cause of random memory corruption.
2674 Description: Memory leak in recovery
2675 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2676 It should also check obd_fail before fsfilt_start, otherwise if
2677 fsfilt_start return -EROFS,(failover mds during mds recovery).
2678 then the req will return with repmsg->transno = 0 and rc = EROFS.
2679 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2680 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2681 be freed no matter whether fsfilt_commit success or not.
2684 Frequency : only with huge count clients
2686 Description: Prevents from taking the superblock lock in llap_from_page for
2688 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2689 for llap_from_page call in ll_removepage prevents from taking the
2690 superblock lock for a soon died page.
2695 Description: Not check open intent error before release open handle
2696 Details : in some rare cases, the open intent error is not checked before
2697 release open handle, which may cause
2698 ASSERTION(open_req->rq_transno != 0), because it tries to release
2699 the failed open handle.
2704 Description: Set cat log bitmap only after create log success.
2705 Details : in some rare cases, the cat log bitmap is set too early. and it
2706 should be set only after create log success.
2710 Description: Accessing a block bevice can re-enable I/O when Lustre is
2711 tearing down a device.
2712 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2717 Description: service threads may hog cpus when there are a lot of requests
2718 Details : Insert cond_resched to give other threads a chance to use some CPU
2723 Description: the cat log was not initialized in recovery
2724 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2725 unlink log on mds will not be initialized until mds post
2726 recovery. And also in mds post recovery, the unlink log will
2727 initialization will be done asynchronausly, so there will be race
2728 between add unlink log and unlink log initialization.
2732 Description: brw_stats were being printed incorrectly
2733 Details : brw_stats were being printed as log2 but all of them were not
2734 recorded as log2. Also remove some code duplication arising from
2735 filter_tally_{read,write}.
2739 Frequency : rare, only in recovery.
2740 Description: ASSERTION(req->rq_type != LI_POISON) failed
2741 Details : imp_lock should be held while iterating over imp_sending_list for
2742 prevent destroy request after get timeout in ptlrpc_queue_wait.
2746 Description: replay-single.sh test 52 fails
2747 Details : A lock's skiplist need to be cleanup when it being unlinked
2748 from its resource list.
2752 Description: Short directio read returns full requested size rather than
2754 Details : Direct I/O operations should return actual amount of bytes
2755 transferred rather than requested size.
2757 Severity : enhancement
2759 Description: metadata RPC reduction (e.g. for rm performance)
2760 Details : decrease the amount of synchronous RPC between clients and servers
2761 by canceling conflicing lock before the operation on the client
2762 and packing thier handles into the main operation RPC to server.
2764 Severity : enhancement
2766 Description: Async OSC create to avoid the blocking unnecessarily.
2767 Details : If a OST has no remain object, system will block on the creating
2768 when need to create a new object on this OST. Now, ways use
2769 pre-created objects when available, instead of blocking on an
2770 empty osc while others are not empty. If we must block, we block
2771 for the shortest possible period of time.
2775 Description: improve handling recoverable errors
2776 Details : if request processig with error which can be recoverable on server
2777 request should be resend, otherwise page released from cache and
2780 Severity : enhancement
2782 Description: refine locking for avoid write wrong info into lov_objid file
2783 Details : fix possible races with add new target and write/update data in
2786 --------------------------------------------------------------------------------
2788 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2793 Frequency : on some architectures
2795 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2796 Details : Uninitialized flags sometimes cause configuration commands to
2800 Frequency : patchless clients only
2802 Description: missing __iget() symbol export
2803 Details : The __iget() symbol export is missing. To avoid the need for
2804 this on patchless clients the deathrow inode reaper is turned
2805 off, and we depend on the VM to clean up old inodes. This
2806 dependency was during via the fix for bug 12181.
2808 --------------------------------------------------------------------------------
2810 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2812 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2813 INTEROPERATE with older versions automatically. In many cases a
2814 special upgrade step is needed. Please read the
2815 user documentation before upgrading any part of a live system.
2816 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2817 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2818 * WARNING: Lustre configuration and startup changes are required with
2819 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2823 Severity : enhancement
2825 Description: MountConf
2826 Details : Lustre configuration is now managed via mkfs and mount
2827 commands instead of lmc and lconf. New obd types (MGS, MGC)
2828 are added for dynamic configuration management. See
2829 https://mail.clusterfs.com/wikis/lustre/MountConf for
2832 Severity : enhancement
2834 Description: dynamic OST addition
2835 Details : OSTs can now be added to a live filesystem
2837 Severity : enhancement
2839 Description: startup order invariance
2840 Details : MDTs and OSTs can be started in any order. Clients only
2841 require the MDT to complete startup.
2843 Severity : enhancement
2845 Description: parallel, asynchronous orphan cleanup
2846 Details : orphan cleanup is now performed in separate threads for each
2847 OST, allowing parallel non-blocking operation.
2849 Severity : enhancement
2851 Description: optimized stripe assignment
2852 Details : stripe assignments are now made based on ost space available,
2853 ost previous usage, and OSS previous usage, in order to try
2854 to optimize storage space and networking resources.
2856 Severity : enhancement
2858 Description: Permanently set tunables
2859 Details : All writable /proc/fs/lustre tunables can now be permanently
2860 set on a per-server basis, at mkfs time or on a live system.
2862 Severity : enhancement
2864 Description: Lustre message v2
2865 Details : Add lustre message format v2.
2867 Severity : enhancement
2869 Description: client OST exclusion list
2870 Details : Clients can be started with a list of OSTs that should be
2871 declared "inactive" for known non-responsive OSTs.
2875 Description: SPEC SFS validation failure on NFS v2 over lustre.
2876 Details : Changes the blocksize for regular files to be 2x RPC size,
2877 and not depend on stripe size.
2879 Severity : enhancement
2881 Description: Multiple MD RPCs in flight.
2882 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2883 To avoid overly-overloading MDS, introduce a limit on number of
2884 MDS RPCs in flight for a single client and add /proc controls
2885 to adjust this limit.
2887 Severity : enhancement
2889 Description: client read/write statistics
2890 Details : Add client read/write call usage stats for performance
2891 analysis of user processes.
2892 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2893 file access. extents_stats shows chunk size distribution.
2894 extents_stats_per_process show chunk size distribution per
2897 Severity : enhancement
2899 Description: mds statistics
2900 Details : Add detailed mds operations statistics in
2901 /proc/fs/lustre/mds/*/stats.
2905 Description: Failure of copying files with lustre special EAs.
2906 Details : Client side always return success for setxattr call for lustre
2907 special xattr (currently only "trusted.lov").
2912 Description: Refcount LNET uuids
2913 Details : The global LNET uuid list grew linearly with every startup;
2914 refcount repeated list entries instead of always adding to
2917 Severity : enhancement
2919 Description: Dynamic service threads
2920 Details : Within a small range, start extra service threads
2921 automatically when the request queue builds up.
2924 Frequency : mixed-endian client/server environments
2926 Description: mixed-endian crashes
2927 Details : The new msg_v2 system had some failures in mixed-endian
2930 Severity : enhancement
2932 Description: Easy OST removal
2933 Details : OSTs can be permanently deactivated with e.g. 'lctl
2934 conf_param lustre-OST0001.osc.active=0'
2936 Severity : enhancement
2938 Description: MGS proc entries
2939 Details : Added basic proc entries for the MGS showing what filesystems
2942 Severity : enhancement
2944 Description: provide MGS failover
2945 Details : Added config lock reacquisition after MGS server failover.
2947 Severity : enhancement
2949 Description: add Linux 2.4 support
2950 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2954 Description: a large application tries to do I/O to the same resource and dies
2955 in the middle of it.
2956 Details : Check the req->rq_arrival time after the call to
2957 ost_brw_lock_get(), but before we do anything about
2958 processing it & sending the BULK transfer request. This
2959 should help move old stale pending locks off the queue as
2960 quickly as obd_timeout.
2963 Frequency : when an incorrect nid is specified during startup
2965 Description: ptlrpc connect to non-existant node causes kernel crash
2966 Details : LNET can't be re-entered from an event callback, which
2967 happened when we expire a message after the export has been
2968 cleaned up. Instead, hand the zombie cleanup off to another
2971 Severity : enhancement
2973 Description: plain/inodebits lock performance improvement
2974 Details : Grouping plain/inodebits in granted list by their request modes
2975 and bits policy, thus improving the performance of search through
2979 Frequency : only if OST filesystem is corrupted
2981 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2982 Details : for a short time RPCs with bulk IO are in the replay list,
2983 but replay of bulk IOs is unimplemented. If the OST filesystem
2984 is corrupted due to disk cache incoherency and then replay is
2985 started it is possible to trip an assertion. Avoid putting
2986 committed RPCs into the replay list at all to avoid this issue.
2989 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2991 Description: System hang on startup
2992 Details : This bug allowed the liblustre (e.g. catamount) client to
2993 return to the app before handling all startup RPCs. This
2994 could leave the node unresponsive to lustre network traffic
2995 and manifested as a server ptllnd timeout.
2997 Severity : enhancement
2999 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3000 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
3001 boolean to control whether to print peer debug info when a
3002 client's RPC times out.
3005 Frequency : only for kernels with patches from Lustre below 1.4.3
3007 Description: Remove old rdonly API
3008 Details : Remove old rdonly API which unused from at least lustre 1.4.3
3011 Frequency : only for devices with external journals
3013 Description: Set external device read-only also
3014 Details : During a commanded failover stop, we set the disk device
3015 read-only while the server shuts down. We now also set any
3016 external journal device read-only at the same time.
3019 Frequency : when upgrading from 1.4 while trying to change parameters
3021 Description: The wrong (new) MDC name was used when setting parameters for
3022 upgraded MDT's. Also allows changing of OSC (and MDC)
3023 parameters if --writeconf is specified at tunefs upgrade time.
3026 Frequency : when setting specific ost indicies
3028 Description: QOS code breaks on skipped indicies
3029 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3030 created with --index need not be sequential.
3032 Severity : enhancement
3034 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3035 Details : The uninit_groups feature works in conjunction with the kernel
3036 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3037 up the pass1 processing of e2fsck. This is a read-only feature
3038 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3039 mount filesystems that have had this feature enabled.
3041 Severity : enhancement
3043 Description: Improve multi-block allocation algorithm to avoid fragmentation
3044 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3045 allocation locality and avoid filesystem fragmentation.
3047 ------------------------------------------------------------------------------
3049 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3051 * Support for kernels:
3052 2.4.21-47.0.1.EL (RHEL 3)
3053 2.6.5-7.283 (SLES 9)
3054 2.6.9-42.0.10.EL (RHEL 4)
3055 2.6.12.6 vanilla (kernel.org)
3056 2.6.16.27-0.9 (SLES 10)
3057 * Recommended e2fsprogs version: 1.39.cfs5
3059 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3063 Frequency : occasional, depends on client load and configuration
3064 Bugzilla : 12181, 12203
3065 Description: data loss for recently-modified files
3067 Details : In some cases it is possible that recently written or created
3068 files may not be written to disk in a timely manner (this should
3069 normally be within 30s unless client IO load is very high).
3070 The problem appears as zero-length files or files that are a
3071 multiple of 1MB in size after a client crash or client eviction
3072 that are missing data at the end of the file.
3074 This problem is more likely to be hit on clients where files are
3075 repeatedly created and unlinked in the same directory, clients
3076 have a large amount of RAM, have many CPUs, the filesystem has
3077 many OSTs, the clients are rebooted frequently, and/or the files
3078 are not accessed by other nodes after being written.
3080 The presence of the problem can be detected by looking at
3081 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
3082 smaller than the second (nr_unused) then dirty files will not
3083 be flushed automatically to disk. "sync; sleep 10" should be
3084 run several times on the node before unmounting it to update
3085 Lustre (this is also safe to run on nodes without this problem).
3087 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
3088 that can cause this same problem, so customers using that kernel
3089 also need to update the kernel in addition to Lustre. In order
3090 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3092 It is normal that files written just before a client crash (less
3093 than 30s) may not yet have been flushed to disk, even for local
3097 Frequency : frequent on thin XT3 nodes
3099 Description: UUID collision on thin XT3 Linux nodes
3100 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3101 randomly, since we relied on an insufficiently-seeded PRNG.
3106 Description: OSS hangs after "All ost request buffers busy"
3107 Details : A deadlock between quota and journal operations caused OSS
3108 hangs after printing "All ost request buffers busy."
3111 Frequency : always on liblustre builds
3113 Description: Cleanup compiler warnings on liblustre
3116 Frequency : always on liblustre builds on XT3
3118 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3122 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3123 > 32000 subdirectories
3124 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3125 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3126 subdirectory count crosses 32000. This will aid e2fsck to
3127 correctly handle more than 32000 subdirectories.
3130 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3132 Description: System hang on startup
3133 Details : This bug allowed the liblustre (e.g. catamount) client to
3134 return to the app before handling all startup RPCs. This
3135 could leave the node unresponsive to lustre network traffic
3136 and manifested as a server ptllnd timeout.
3138 Severity : enhancement
3140 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3141 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3142 boolean to control whether to print peer debug info when a
3143 client's RPC times out.
3148 Description: make O_SYNC working on 2.6 kernels
3149 Details : 2.6 kernels use different method for mark pages for write,
3150 so need add a code to lustre for O_SYNC work.
3155 Description: Failure to close file and release space on NFS
3156 Details : Put inode details into lock acquired in ll_intent_file_open.
3157 Use mdc_intent_lock in ll_intent_open to properly
3158 detect all kind of errors unhandled by mdc_enqueue.
3163 Description: proc file read during shutdown sometimes raced obd removal,
3165 Details : Add lock to prevent obd access after proc file removal.
3168 Frequency : Only for files larger than 4GB on 32-bit clients.
3170 Description: improperly doing page alignment of locks
3171 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3172 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3177 Description: RPCs being resent when they shouldn't be
3178 Details : Some RPCs that should not be resent are being resent. This
3179 can cause inconsistencies in the RPC state machine. Do not
3180 resend such requests.
3183 Frequency : rare, only with NFS export
3185 Description: Crash on NFS re-export node
3186 Details : under very unusual load conditions an assertion is hit in
3187 ll_intent_file_open()
3190 Frequency : only if OST filesystem is corrupted
3192 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3193 Details : for a short time RPCs with bulk IO are in the replay list,
3194 but replay of bulk IOs is unimplemented. If the OST filesystem
3195 is corrupted due to disk cache incoherency and then replay is
3196 started it is possible to trip an assertion. Avoid putting
3197 committed RPCs into the replay list at all to avoid this issue.
3202 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3203 Details : Large single O_DIRECT read and write calls can fail to allocate
3204 a sufficiently large buffer to process the request. In case of
3205 allocation failure the allocation is retried with a smaller
3206 buffer and broken into smaller requests.
3208 Severity : enhancement
3210 Description: Add -o localflock option to simulate old noflock behaviour.
3211 Details : This will achieve local-only flock/fcntl locks coherentness.
3216 Description: versioning check is incomplete
3217 Details : Checking the version difference of client vs. server, report
3218 error if the gap is too big.
3223 Description: add support PG_writeback bit
3224 Details : add support for PG_writeback bit for Lustre, for more carefull
3225 work with page cache in 2.6 kernel. This also fix some deadlocks
3226 and remove hack for work O_SYNC with 2.6 kernel.
3228 Severity : enhancement
3230 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3231 Details : The uninit_groups feature works in conjunction with the kernel
3232 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3233 up the pass1 processing of e2fsck. This is a read-only feature
3234 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3235 mount filesystems that have had this feature enabled.
3237 Severity : enhancement
3239 Description: Improve multi-block allocation algorithm to avoid fragmentation
3240 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3241 allocation locality and avoid filesystem fragmentation.
3243 ------------------------------------------------------------------------------
3245 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3247 * Support for kernels:
3248 2.6.9-42.0.3.EL (RHEL 4)
3249 2.6.5-7.276 (SLES 9)
3250 2.4.21-47.0.1.EL (RHEL 3)
3251 2.6.12.6 vanilla (kernel.org)
3252 2.6.16.21-0.8 (SLES10)
3253 * Recommended e2fsprogs version: 1.39.cfs2-0
3255 * The backwards-compatible /proc/sys/portals symlink has been removed
3256 in this release. Before upgrading, please ensure that you change
3257 any configuration scripts or /etc/sysctl.conf files that access
3258 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3259 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3260 in advance of the upgrade on any system running Lustre 1.4.6 or
3261 newer, since /proc/sys/lnet was added in that version.
3262 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3266 Frequency : only when quota is used
3268 Description: avoid scanning export list for quota master
3269 Details : Change the algorithms to avoid scanning export list in order
3270 to improve the efficiency.
3273 Frequency : MDS failover only, very rarely
3275 Description: "went back in time" messages on mds failover
3276 Details : The greatest transno may be lost when the current operation
3277 finishes with an error (transno==0) and the client's last_rcvd
3278 record is over-written. Save the greatest transno in the
3279 mds_last_transno for this case.
3282 Frequency : always for specific kernels and striping counts
3284 Description: client may get "Matching packet too big" without ACL support
3285 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3286 when trying to access files in certain configurations. The
3287 clients should in fact be denied when mounting because they do
3288 not understand ACLs.
3291 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3293 Description: many clients connecting with IO in progress causes connect timeouts
3294 Details : Avoid synchronous journal commits to avoid delays caused by many
3295 clients connecting/disconnecting when bulk IO is in progress.
3296 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3297 OST_IO_PORTAL to avoid delays behind potentially many pending
3301 Frequency : occasionally with multiple writers to a single file
3303 Description: shared writes to file may result in wrong size reported by stat()
3304 Details : Allow growing of kms when extent lock is cancelled
3307 Frequency : always with random mmap IO to multi-striped file
3309 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3310 Details : Only if the hole is at the end of OST object so that kms is too
3311 small. Fix is to increase kms accordingly in ll_nopage.
3314 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3316 Description: writes to a missing object would leak memory on the OST
3317 Details : If there is an inconsistency between the MDS and OST filesystems,
3318 such that the MDS references an object that doesn't exist, writes
3319 to that object will leak memory due to incorrect cleanup in the
3320 error handling path, eventually running out of memory on the OST.
3325 Description: Creating too long symlink causes lustre errors
3326 Details : Check symlink and name lengths before sending requests to MDS.
3329 Frequency : only if flock is enabled (not on by default)
3331 Description: posix locks not released on fd closure on 2.6.9+
3332 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3333 kernels, this caused such locks not to be released on fd close and
3334 then assertions on fs unmount about still used locks.
3337 Frequency : MDS failover only, very rarely
3339 Description: clients may get ASSERTION(granted_lock != NULL)
3340 Details : When request was taking a long time, and a client was resending
3341 a getattr by name lock request. The were multiple lock requests
3342 with the same client lock handle and
3343 mds_getattr_name->fixup_handle_for_resent_request found one of the
3344 lock handles but later failed with ASSERTION(granted_lock != NULL).
3349 Description: handle->h_buffer_credits > 0, assertion failure
3350 Details : h_buffer_credits is zero after truncate, causing assertion
3351 failure. This patch extends the transaction or creates a new
3355 Frequency : NFS re-export or patchless client
3356 Bugzilla : 11179, 10796
3357 Description: Crash on NFS re-export node (__d_move)
3358 Details : We do not want to hash the dentry if we don't have a lock.
3359 But if this dentry is later used in d_move, we'd hit uninitialised
3360 list head d_hash, so we just do this to init d_hash field but
3361 leave dentry unhashed.
3364 Frequency : NFS re-export or patchless client
3366 Description: NFS exports has problem with symbolic link
3367 Details : lustre client didn't properly install dentry when re-exported
3368 to NFS or running patchless client.
3371 Frequency : NFS re-export or patchless client
3373 Description: Various nfs/patchless fixes.
3374 Details : fixes reuse disconected alias for lookup process - this fixes
3375 warning "find_exported_dentry: npd != pd",
3376 fix permission error with open files at nfs.
3377 fix apply umask when do revalidate.
3380 Frequency : occasional
3382 Description: Crash on NFS re-export node
3383 Details : calling clear_page() on the wrong pointer triggered oops in
3384 generic_mapping_read().
3387 Frequency : rarely, using O_DIRECT IO
3389 Description: unaligned directio crashes client with LASSERT
3390 Details : check for unaligned buffers before trying any requests.
3393 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3395 Description: stale data returned from RAID cache
3396 Details : If only a small amount of IO is done to the RAID device before
3397 reading it again it is possible to get stale data from the RAID
3398 cache instead of reading it from disk.
3401 Frequency : always for sles10 kernel
3403 Description: sles10 support
3404 Details : ll_follow_link: compile fixes and using of nd_set_link
3405 under newer kernels.
3408 Frequency : depends on arch, kernel and compiler version, always on sles10
3411 Description: recursive or deep enough symlinks cause stack overflow
3412 Details : getting rid of large stack-allocated variable in
3416 Frequency : depends on hardware
3418 Description: lustre write performance loss in the SLES10 kernel
3419 Details : the performance loss is caused by using of write barriers in the
3420 ext3 code. The SLES10 kernel turns barrier support on by
3421 default. The fix is to undo that change for ldiskfs.
3423 ------------------------------------------------------------------------------
3425 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3427 * Support for kernels:
3428 2.6.9-42.0.3EL (RHEL 4)
3429 2.6.5-7.276 (SLES 9)
3430 2.4.21-47.0.1.EL (RHEL 3)
3431 2.6.12.6 vanilla (kernel.org)
3435 Frequency : quota enabled and large files being deleted
3437 Description: releasing more than 4GB of quota at once hangs OST
3438 Details : If a user deletes more than 4GB of files on a single OST it
3439 will cause the OST to spin in an infinite loop. Release
3440 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3445 Description: statfs data retrieved from /proc may be stale or zero
3446 Details : When reading per-device statfs data from /proc, in the
3447 {kbytes,files}_{total,free,avail} files, it may appear
3448 as zero or be out of date.
3451 Frequency : systems with MD RAID1 external journal devices
3453 Description: lconf's call to blkid is confused by RAID1 journal devices
3454 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3455 of returning all block devices that match the journal UUID.
3458 Frequency : always, for aggregate stripe size over 4GB
3460 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3461 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3462 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3465 Frequency : always if "lfs find" used on a local file/directory
3467 Description: "lfs find" segfaults if used on a local file/directory
3468 Details : The case where a directory component was not specified wasn't
3469 handled correctly. Handle this properly.
3472 Frequency : always on ppc64
3474 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3475 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3476 perform endianness conversion.
3479 Frequency : rarely (truncate to non-zero file size after write under load)
3480 Bugzilla : 10730, 10687
3481 Description: Files padded with zeros to next 4K multiple
3482 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3483 it is possible that files that are truncated to a non-zero size
3484 immediately after being written are filled with zero bytes beyond
3485 the truncated size. No file data is lost.
3487 Severity : enhancement
3489 Description: Allow recovery/failover for liblustre clients.
3490 Details : liblustre clients were unaware of failover configurations until
3493 Severity : enhancement
3495 Description: user file locks should fail when not mounting with flock option
3496 Details : Set up an error-returning stub in ll_file_operations.lock field
3497 to prevent incorrect behaviour when client is mounted without
3498 flock option. Also, set up properly f_op->flock field for
3502 Frequency : always on ia64
3504 Description: "lfs df" loops on printing out MDS statfs information
3505 Details : The obd_ioctl_data was not initialized and in some systems
3506 this caused a failure during the ioctl that did not return
3507 an error. Initialize the struct and return an error on failure.
3510 Frequency : SLES 9 only
3512 Description: Error of copying files with lustre special EAs as root
3513 Details : Client side always return success for setxattr call for lustre
3514 special xattr (currently only "trusted.lov").
3517 Frequency : rarely on clusters with both ia64+i386 clients
3519 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3520 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3521 clients are doing concurrent writes to the same file it is
3522 possible that the ia64 clients may LASSERT because the OST
3523 extent locks are not PAGE_SIZE aligned. Ensure that grown
3524 locks are always aligned on the request boundary.
3527 Frequency : specific use, occasional
3529 Description: Overwriting in use executable truncates on-disk binary image
3530 Details : If one node attempts to overwrite an executable in use by
3531 another node, we now correctly return ETXTBSY instead of
3532 truncating the file.
3534 Severity : enhancement
3536 Description: Async OSC create to avoid the blocking unnecessarily.
3537 Details : If a OST has no remain object, system will block on the creating
3538 when need to create a new object on this OST. Now, ways use
3539 pre-created objects when available, instead of blocking on an
3540 empty osc while others are not empty. If we must block, we block
3541 for the shortest possible period of time.
3546 Description: chmod on Lustre root is propagated to other clients
3547 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3548 invalid by the follow_mount time.
3551 Frequency : liblustre clients only
3553 Description: Race in 'instant cancel' lock handling could lead to such locks
3554 never to be granted in case of SMP MDS
3555 Details : Do not destroy not yet granted but cbpending locks in
3559 Frequency : replay/resend of open
3561 Description: non null lock assetion failure in mds_intent_policy
3562 Details : Trying to replay/resend lockless open requests resulted in
3563 mds_open() returning 0 with no lock. Now it sets a flag if
3564 a lock is going to be returned.
3566 Severity : enhancement
3568 Description: Checksum enhancements
3569 Details : New checksum enhancements allow for resending RPCs that failed
3572 Severity : enhancement
3574 Description: Tunables on number of dirty pages in cacche
3575 Details : Allow to set limit on number of dirty pages cached.
3580 Description: client crash on unmount - lock still has references
3581 Details : In some error handling cases it was possible to leak a lock
3582 reference on a client while accessing a file. This was not
3583 harmful to the client during operation, but would cause the
3584 client to crash when the filesystem is unmounted.
3587 Frequency : specific case, rare
3589 Description: ETXTBSY on mds though file not in use by client
3590 Details : ETXTBSY is no longer incorrectly returned when attempting to
3591 chmod or chown a directory that the user previously tried to
3592 execute or a currently-executing binary.
3595 Frequency : extremely rare except on liblustre-based clients
3597 Description: Lustre space not freed when files are deleted
3598 Details : Clean up open-unlinked files after client eviction. Previously
3599 the unlink was skipped and the files remained as orphans.
3604 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3605 Details : In some cases it was possible to send a blocking callback to
3606 a client doing a glimpse, even though that client didn't get
3607 a lock granted. When the glimpse lock is cancelled on the OST
3608 the freed lock is left on the waiting list and corrupted the list.
3611 Frequency : all core dumps
3613 Description: Broke core dumps to lustre
3614 Details : Negative dentry may be unhashed if parent does not have UPDATE
3615 lock, but some callers, e.g. do_coredump, expect dentry to be
3616 hashed after successful create, hash it in ll_create_it.
3618 ------------------------------------------------------------------------------
3620 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3622 * Support for kernels:
3623 2.6.9-42.0.2.EL (RHEL 4)
3624 2.6.5-7.276 (SLES 9)
3625 2.4.21-40.EL (RHEL 3)
3626 2.6.12.6 vanilla (kernel.org)
3630 Frequency : always on RHEL 3
3632 Description: Number of open files grows over time
3633 Details : The number of open files grows over time, whether or not
3634 Lustre is started. This was due to a filp leak introduced
3635 by one of our kernel patches.
3637 ------------------------------------------------------------------------------
3639 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3641 * Support for kernels:
3642 2.6.9-42.EL (RHEL 4)
3643 2.6.5-7.276 (SLES 9)
3644 2.4.21-40.EL (RHEL 3)
3645 2.6.12.6 vanilla (kernel.org)
3650 Bugzilla : 5719, 9635, 9792, 9684
3651 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3652 Details : If a server is under heavy load and cannot reply to new
3653 connection requests before the client resends the (re)connect,
3654 the connection handling code can behave badly if two service
3655 threads are concurrently handing separate (re)connections from
3656 the same client. Add better locking to the connection handling
3657 code, and ensure that only a single connection will be processed
3658 for a given client UUID, even if the lock is dropped.
3660 Severity : enhancement
3662 Description: add TCP zero-copy support to kernel
3663 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3664 sends if the hardware supports scatter-gather and checksumming.
3665 This allows socklnd to do client-write and server-read more
3666 efficiently and reduce CPU utilization from skbuf copying.
3669 Frequency : only if NFS exporting from client
3671 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3672 Details : If setting attributes on a file created under NFS that had
3673 never been opened it would be possible to oops the client
3674 if the file had no objects.
3677 Frequency : always for liblustre
3679 Description: liblustre client does MDS+OSTs setattr RPC for each write
3680 Details : When doing a write from a liblustre client, the client
3681 incorrectly issued an RPC to the MDS and each OST the file was
3682 striped over in order to update the timestamps. When writing
3683 with small chunks and many clients this could overwhelm the MDS
3684 with RPCs. In all cases it would slow down the write because
3685 these RPCs are unnecessary.
3687 Severity : enhancement
3689 Description: allow number of MDS service threads to be changed at module load
3690 Details : It is now possible to change the number of MDS service threads
3691 running. Adding "options mds mds_num_threads={N}" to the MDS's
3692 /etc/modprobe.conf will set the number of threads for the next
3693 time Lustre is restarted (assuming the "mds" module is also
3694 reloaded at that time). The default number of threads will
3695 stay the same, 32 for most systems.
3700 Description: OST crash if filesystem is unformatted or corrupt
3701 Details : If an OST is started on a device that has never been formatted
3702 or if the filesystem is corrupt and cannot even mount then the
3703 error handling cleanup routines would dereference a NULL pointer.
3708 Description: NULL pointer deref in llap_from_page.
3709 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3710 incorrect) mapping if the page was truncated/reclaimed while it was
3711 searched for. Check for this condition and skip such pages when
3712 doing readahead. Introduce extra check to llap_from_page() to
3713 verify page->mapping->host is non-NULL (so page is not anonymous).
3716 Frequency : Sometimes when using sys_sendfile
3718 Description: "page not covered by a lock" warnings from ll_readpage
3719 Details : sendfile called ll_readpage without right page locks present.
3720 Now we introduced ll_file_sendfile that does necessary locking
3721 around call to generic_file_sendfile() much like we do in
3725 Frequency : with certain MDS communication failures at client mount time
3727 Description: NULL pointer deref after failed client mount
3728 Details : a client connection request may delayed by the network layer
3729 and not be sent until after the PTLRPC layer has timed out the
3730 request. If the client fails the mount immediately it will try
3731 to clean up before the network times out the request. Add a
3732 reference from the request import to the obd device and delay
3733 the cleanup until the network drops the request.
3736 Frequency : occasionally during client (re)connect
3738 Description: assertion failure during client (re)connect
3739 Details : processing a client connection request may be delayed by the
3740 client or server longer than the client connect timeout. This
3741 causes the client to resend the connection request. If the
3742 original connection request is replied in this interval, the
3743 client may trip an assertion failure in ptlrpc_connect_interpret()
3744 which thought it would be the only running connect process.
3747 Frequency : only with obd_echo servers and clients that are rebooted
3749 Description: kernel BUG accessing uninitialized data structure
3750 Details : When running an obd_echo server it did not start the ping_evictor
3751 thread, and when a client was evicted an uninitialized data
3752 structure was accessed. Start the ping_evictor in the RPC
3753 service startup instead of the OBD startup.
3755 Severity : enhancement
3756 Bugzilla : 10193 (patchless)
3757 Description: Remove dependency on various unexported kernel interfaces.
3758 Details : No longer need reparent_to_init, exit_mm, exit_files,
3759 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3762 Frequency : rare (only users of deprecated and unsupported LDAP config)
3764 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3765 Details : LDAP apparently contains 'lustreName' attributes instead of
3766 'name'. A simple remapping of the name is sufficient.
3769 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3771 Description: waiting_locks_callback trips kernel BUG if client is evicted
3772 Details : Running with the dump_on_timeout debug flag turned on makes
3773 it possible that the waiting_locks_callback() can try to dump
3774 the Lustre kernel debug logs from an interrupt handler. Defer
3775 this log dumping to the expired_lock_main() thread.
3777 Severity : enhancement
3779 Description: Support NFS exporting on 2.6 kernels.
3780 Details : Implement non-rawops metadata methods for NFS server to use without
3781 changing NFS server code.
3784 Frequency : very rare (synthetic metadata workload only)
3786 Description: two racing renames might cause an MDS thread to deadlock
3787 Details : Running the "racer" program may cause one MDS thread to rename
3788 a file from being the source of a rename to being the target of
3789 a rename at exactly the same time that another thread is doing
3790 so, and the second thread has already enqueued these locks after
3791 doing a lookup of the target and is trying to relock them in
3792 order. Ensure that we don't try to re-lock the same resource.
3795 Frequency : only very large systems with liblustre clients
3797 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3798 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3799 all OSTs in parallel. This allows the eviction of stale liblustre
3800 clients to proceed much faster than if they were done in series,
3801 and also offers similar improvements for other set_info RPCs.
3806 Description: excessive CPU usage during initial read phase on client
3807 Details : During the initial read phase on a client, it would agressively
3808 retry readahead on the file, consuming too much CPU and impacting
3809 performance (since 1.4.5.8). Improve the readahead algorithm
3810 to avoid this, and also improve some other common cases (read
3811 of small files in particular, where "small" is files smaller than
3812 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3817 Description: MDS crash when receiving packet with unknown intent.
3818 Details : Do not LBUG in unknown intent case, just return -EFAULT
3820 Severity : enhancement
3821 Bugzilla : 9293, 9385
3822 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3823 Details : Do not serialize getattr (non-intent version) and statfs.
3826 Frequency : occasional, when OST network is overloaded/intermittent
3828 Description: client evicted by OST after bulk IO timeout
3829 Details : If a client sends a bulk IO request (read or write) the OST
3830 may evict the client if it is unresposive to its data GET/PUT
3831 request. This is incorrect if the network is overloaded (takes
3832 too long to transfer the RPC data) or dropped the OST GET/PUT
3833 request. There is no need to evict the client at all, since
3834 the pinger and/or lock callbacks will handle this, and the
3835 client can restart the bulk request.
3838 Frequency : Always when mmapping file with no objects
3840 Description: client crashes when mmapping file with no objects
3841 Details : Check that we actually have objects in a file before doing any
3842 operations on objects in ll_vm_open, ll_vm_close and
3848 Description: Request leak when working with deleted CWD
3849 Details : Introduce advanced request refcount tracking for requests
3850 referenced from lustre intent.
3852 Severity : Enhancement
3854 Description: Cache open file handles on client.
3855 Details : MDS now will return special lock along with openhandle, if
3856 requested and client is allowed to hold openhandle, even if unused,
3857 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3858 closing files for every read/write openration.
3860 Severity : Enhancement
3862 Description: Cache open negative dentries on client when possible.
3863 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3864 negative dentries on lock revocation.
3869 Description: Remounting a client read-only wasn't possible with a zconf mount
3870 Details : It wasn't possible to remount a client read-only with llmount.
3872 Severity : enhancement
3873 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3874 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3875 MPICH is included to add Lustre-specific ADIO interfaces.
3876 This is based closely on the UFS ADIO layer and only differs
3877 in file creation, in order to allow the OST striping to be set.
3878 This is user-contributed code and not supported by CFS.
3883 Description: extended inode attributes (immutable, append-only) work improperly
3884 when 2.4 and 2.6 kernels are used on client/server or vice versa
3885 Details : Introduce kernel-independent values for these flags.
3887 Severity : enhancement
3890 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3891 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3892 granularity. Now, Fractional MB granularity is be supported,
3893 this is very useful for low memory system.
3895 Severity : enhancement
3897 Description: Getattr by fid
3898 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3899 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3900 have locks and avoids getattr rpc requests for it on every stat.
3903 Frequency : Always, for filesystems larger than 2TB
3905 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3906 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3907 filesystems larger than 2TB to be mounted with mballoc enabled.
3910 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3912 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3913 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3914 filesystems larger than 2TB is mounted then the kernel will
3915 silently corrupt the start of the filesystem. CONFIG_LBD is
3916 enabled for all CFS-supported kernels, but the possibility of
3917 this happening with a modified kernel config exists.
3919 Severity : enhancement
3921 Description: add client O_DIRECT support for 2.6 kernels
3922 Details : It is now possible to do O_DIRECT reads and writes to files
3923 in the Lustre client mountpoint on 2.6 kernel clients.
3925 Severity : enhancement
3927 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3928 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3929 parallel, not waiting for response from every OST before sending
3930 a rpc to the next OST.
3935 Description: setattr vs write race when updating file timestamps
3936 Details : Client processes that update a file timestamp into the past
3937 right after writing to the file (e.g. tar) it is possible that
3938 the updated file modification time can be reset to the current
3939 time due to a race between processing the setattr and write RPC.
3941 Severity : enhancement
3943 Description: Bring 'lfs find' closer in line with regular Linux find.
3944 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3945 -print0 options and obtains all the needed info through the lustre
3948 Severity : enhancement
3950 Description: support up to 1024 configured devices on one node
3951 Details : change obd_dev array from statically allocated to dynamically
3952 allocated structs as they are first used to reduce memory usage
3957 Description: Flush dirty partially truncated pages during truncate
3958 Details : Immediatelly flush partially truncated pages in filter_setattr,
3959 this way we completely avoid having any pages in page cache on OST
3960 and can retire ugly workarounds during writes to flush such pages.
3965 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3966 Details : protect inode from truncation within vfs_unlink() context
3967 just take a reference before calling vfs_unlink() and release it
3968 when parent's i_sem is free.
3973 Description: last_id value checked outside lock on OST caused LASSERT failure
3974 Details : If there were multiple MDS->OST object precreate requests in
3975 flight, it was possible that the OST's last object id was checked
3976 outside a lock and incorrectly tripped an assertion. Move checks
3977 inside locks, and discard old precreate requests.
3980 Frequency : always, if extents are used on OSTs
3982 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3983 Details : OSTs using the extents format would not zero the high 16 bits of
3984 the index physical block number. This is not a problem for any
3985 OST filesystems smaller than 16TB, and no kernels support ext3
3986 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3987 new/modified files) and can be fixed for existing filesystems
3988 with e2fsprogs-1.39-cfs1.
3993 Description: import connection selection may be incorrect if timer wraps
3994 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3995 import connections to be ignored if the 32-bit jiffies counter
3996 wraps. Use a 64-bit jiffies counter.
3999 Frequency : very large clusters immediately after boot
4001 Description: LNET request buffers exhausted under heavy short-term load
4002 Details : If a large number of client requests are generated on a service
4003 that has previously never seen so many requests it is possible
4004 that the request buffer growth cannot keep up with the spike in
4005 demand. Instead of dropping incoming requests, they are held in
4006 the LND until the RPC service can accept more requests.
4009 Frequency : Sometimes during replay
4011 Description: Assertion failure in ll_local_open after replay.
4012 Details : If replay happened on an open request reply before we were able
4013 to set replay handler, reply will become not swabbed tripping the
4014 assertion in ll_local_open. Now we set the handler right after
4015 recognising of open request
4018 Frequency : very rare
4020 Description: kernel reports "badness in vsnprintf"
4021 Details : Reading from the "recovery_status" /proc file in small chunks
4022 may cause a negative length in lprocfs_obd_rd_recovery_status()
4023 call to vsnprintf() (which is otherwise harmless). Exit early
4024 if there is no more space in the output buffer.
4026 Severity : enhancement
4028 Description: clear OBD RPC statistics by writing to them
4029 Details : It is now possible to clear the OBD RPC statistics by writing
4030 to the "stats" file.
4035 Description: Client mtime is not the same on different clients after utimes
4036 Details : In some cases, the client was using the utimes() syscall on
4037 a file cached on another node. The clients now validate the
4038 ctime from the MDS + OSTs to determine which one is right.
4043 Description: Inability to activate failout mode
4044 Details : lconf script incorrectly assumed that in pythong string's numeric
4045 value is used in comparisons.
4048 Frequency : always with multiple stripes per file
4050 Description: Inefficient object allocation for mutli-stripe files
4051 Details : When selecting which OSTs to stripe files over, for files with
4052 a stripe count that divides evenly into the number of OSTs,
4053 the MDS is always picking the same starting OST for each file.
4054 Return the OST selection heuristic to the original design.
4059 Description: mount failures may take full timeout to return an error
4060 Details : Under some heavy load conditions it is possible that a
4061 failed mount can wait for the full obd_timeout interval,
4062 possibly several minutes, before reporting an error.
4063 Instead return an error as soon as the status is known.
4065 Frequency : quota enabled and large files being deleted
4067 Description: releasing more than 4GB of quota at once hangs OST
4068 Details : If a user deletes more than 4GB of files on a single OST it
4069 will cause the OST to spin in an infinite loop. Release
4070 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4075 Description: statfs data retrieved from /proc may be stale or zero
4076 Details : When reading per-device statfs data from /proc, in the
4077 {kbytes,files}_{total,free,avail} files, it may appear
4078 as zero or be out of date.
4081 Frequency : systems with MD RAID1 external journal devices
4083 Description: lconf's call to blkid is confused by RAID1 journal devices
4084 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4085 of returning all block devices that match the journal UUID.
4088 Frequency : always, for aggregate stripe size over 4GB
4090 Description: assertion fails when trying to use 4GB stripe size
4091 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
4092 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
4095 Frequency : always on ppc64
4097 Description: the first write on an ext3 filesystem with mballoc got stuck
4098 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4099 perform endianness conversion.
4101 ------------------------------------------------------------------------------
4103 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
4105 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4106 INTEROPERATE with older versions automatically. Please read the
4107 user documentation before upgrading any part of a live system.
4108 * WARNING: Lustre networking configuration changes are required with
4109 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4112 * Support for kernels:
4113 2.6.9-22.0.2.EL (RHEL 4)
4114 2.6.5-7.244 (SLES 9)
4115 2.6.12.6 vanilla (kernel.org)
4118 Severity : enhancement
4119 Bugzilla : 7981/8208
4120 Description: Introduced Lustre Networking (LNET)
4121 Details : LNET is new networking infrastructure for Lustre, it includes
4122 a reorganized network configuration mode (see the user
4123 documentation for full details) as well as support for routing
4124 between different network fabrics. Lustre Networking Devices
4125 (LNDs) for the supported network fabrics have also been
4126 created for this new infrastructure.
4128 Severity : enhancement
4129 Description: Introduced Access control lists
4130 Details : clients can set ACLs on files and directories in order to have
4131 more fine-grained permissions than the standard Unix UGO+RWX.
4132 The MDS must be started with the "-o acl" mount option.
4134 Severity : enhancement
4135 Description: Introduced filesystem quotas
4136 Details : Administrators may now establish per-user quotas on the
4139 Severity : enhancement
4141 Description: Configuration change for the XT3
4142 The PTLLND is now used to run Lustre over Portals on the XT3
4143 The configure option(s) --with-cray-portals are no longer used.
4144 Rather --with-portals=<path-to-portals-includes> is used to
4145 enable building on the XT3. In addition to enable XT3 specific
4146 features the option --enable-cray-xt3 must be used.
4151 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4152 Details : A race between allocating a new llcd and re-getting the llcd_lock
4153 allowed another thread to grab newly-allocated llcd.
4155 Severity : enhancement
4157 Description: 2.6 OST async journal commit and locking fix to improve performance
4158 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4159 2.6 kernels are now async as they already were in 2.4 kernels so
4160 that they can commit concurrently with the network bulk transfer.
4161 For block-allocated files the filter allocation semaphore is held
4162 to avoid filesystem fragmentation during allocation. BKL lock
4163 removed for 2.6 xattr operations where it is no longer needed.
4168 Description: lconf incorrectly determined whether two IP networks could talk
4169 Details : In some more complicated routing and multiple-network
4170 configurations, lconf will avoid trying to make a network
4171 connection to a disjoint part of the IP space. It was doing the
4172 math incorrectly for one set of cases.
4177 Description: Fix for potential infinite loop processing records in an llog.
4178 Details : If an llog record is corrupted/zeroed, it is possible to loop
4179 forever in llog_process(). Validate the llog record length
4180 and skip the remainder of the block on error.
4183 Frequency : occasional (liblustre only)
4185 Description: liblustre could not open files whose last component is a symlink
4186 Details : sysio_path_walk() would incorrectly pass the open intent to
4187 intermediate path components.
4190 Frequency : rare (liblustre only with non-standard tuning)
4191 Bugzilla : 7201 (7350)
4192 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4193 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4194 after it has been released, only do this for referenced locks
4196 Severity : enhancement
4198 Description: specifying an (invalid) directory default stripe_size of -1
4199 would reset the directory default striping
4200 Details : stripe_size -1 was used internally to signal directory stripe
4201 removal, now use "all default" to signal dir stripe removal
4202 as a directory striping of "all default" is not useful
4205 Frequency : common for large clusters running liblustre clients
4207 Description: doing an ls when liblustre clients are running is slow
4208 Details : sending a glimpse AST to a liblustre client waits for every AST
4209 to time out, as liblustre clients will not respond. Since they
4210 cannot cache data we refresh the OST lock LVB from disk instead.
4212 Severity : enhancement
4214 Description: doing an ls at the same time as file IO can be slow
4215 Details : enqueue and other "small" requests can be blocked behind many
4216 large IO requests. Create a new OST IO portal for non-IO
4217 requests so they can be processed faster.
4220 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4222 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4223 created file with mode 000
4224 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4225 allow an ftruncate() as a non-root user to a file with mode 000.
4226 Lustre now respects this flag to disable mode checks when
4227 truncating a file owned by the user
4230 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4233 Description: Revoking locks from clients that went dead or catatonic might take
4235 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4236 cancellation of such locks instant on servers without waiting for
4237 any reply from clients. Clients drops these locks when cancel
4238 notification from server is received without replying.
4241 Frequency : liblustre-only, when liblustre client dies or becomes busy
4243 Description: Doing ls on Linux clients can take a long time with active
4245 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4246 granting such locks to it in the first place if possible. Locks
4247 are taken by proxy on the OST during the read or write and
4248 dropped immediately afterward. Add connect flags handling, do
4249 not grant locks to liblustre clients for glimpse ASTs.
4251 Severity : enhancement
4253 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4254 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4255 sequential reads and behaves badly with applications doing random
4256 reads. Improve it by reading ahead at least read region, and
4257 avoiding excessive large RPC for small reads.
4259 Severity : enhancement
4261 Description: Creating more than 1000 files for a single job may cause a load
4262 imbalance on the OSTs if there are also a large number of OSTs.
4263 Details : qos_prep_create() uses an OST index reseed value that is an
4264 even multiple of the number of available OSTs so that if the
4265 reseed happens in the middle of the object allocation it will
4266 still utilize the OSTs as uniformly as possible.
4271 Description: OST or MDS may oops in ping_evictor_main()
4272 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4273 but doesn't restart at beginning of obd_exports_timed list
4276 Severity : enhancement
4278 Description: improve by-nid export eviction on the MDS and OST
4279 Details : allow multiple exports with the same NID to be evicted at one
4280 time without re-searching the exports list.
4283 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4285 Description: MDS may oops in groups_free()
4286 Details : in rare race conditions a newly allocated group_info struct is
4287 freed again, and this can be NULL. The 2.4 compatibility code
4288 for groups_free() checked for a NULL pointer, but 2.6 did not.
4291 Frequency : common for liblustre clients doing little filesystem IO
4292 Bugzilla : 9352, 7313
4293 Description: server may evict liblustre clients accessing contended locks
4294 Details : if a client is granted a lock or receives a completion AST
4295 with a blocking AST already set it would not reply to the AST
4296 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4299 Frequency : lfs setstripe, only systems with more than 160 OSTs
4301 Description: unable to set striping with a starting offset beyond OST 160
4302 Details : llapi_create_file() incorrectly limited the starting stripe
4303 index to the maximum single-file stripe count.
4306 Frequency : LDAP users only
4308 Description: lconf did not handle in-kernel recovery with LDAP properly
4309 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4311 Severity : enhancement
4313 Description: bind OST threads to NUMA nodes to improve performance
4314 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4315 node and do their allocations there to localize memory access
4317 Severity : enhancement
4319 Description: llmount can determine client NID directly from Myrinet (GM)
4320 Details : the client NID code from gmnalnid was moved directly into
4321 llmount, removing the need to use this or specifying the
4322 client NID explicitly when mounting GM clients with zeroconf
4325 Frequency : if client is started with down MDS
4327 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4328 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4329 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4330 ptlrpc_import_delayed_req() because we didn't check if the
4331 request was interrupted, and we also didn't break out of the
4332 event loop if there was no timeout
4337 Description: data loss during non-page-aligned writes to a single file from
4338 both multiple nodes and multiple threads on one node at same time
4339 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4340 inconsistency led to false short-reads, that were cached and later
4341 used by ->prepare_write() to fill in partially written page,
4342 leading to data loss.
4345 Frequency : always, if lconf --abort_recovery used
4347 Description: lconf --abort_recovery fails with 'Operation not supported'
4348 Details : lconf was attempting to abort recovery on the MDT device and not
4351 Severity : enhancement
4353 Description: remove cleanup logs
4354 Details : replace lconf-generated cleanup logs with lustre internal
4355 cleanup routines. Eliminates the need for client-cleanup and
4358 Severity : enhancement
4360 Description: add support for EAs (user and system) on lustre filesystems
4361 Details : it is now possible to store extended attributes in the Lustre
4362 client filesystem, and with the user_xattr mount option it
4363 is possible to allow users to store EAs on their files also
4365 Severity : enhancement
4367 Description: Add possibility (config option) to show minimal available OST free
4369 Details : When compiled with --enable-mindf configure option, statfs(2)
4370 (and so, df) will return least minimal free space available from
4371 all OSTs as amount of free space on FS, instead of summary of
4372 free spaces of all OSTs.
4374 Severity : enhancement
4376 Description: do not expand extent locks acquired on OST-side
4377 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4378 by server: they are not cached anyway.
4381 Frequency : when mmap is used/binaries executed from Lustre
4383 Description: Unmmap pages before throwing them away from read cache.
4384 Details : llap_shrink cache now attempts to unmap pages before discarding
4385 them (if unmapping failed - do not discard). SLES9 kernel has
4386 extra checks that trigger if this unmapping is not done first.
4391 Description: lconf didn't resolve symlinks before checking to see whether a
4392 given mountpoint was already in use
4395 Frequency : when migrating failover services
4396 Bugzilla : 6395, 9514
4397 Description: When migrating a subset of services from a node (e.g. failback
4398 from a failover service node) the remaining services would
4399 time out and evict clients.
4400 Details : lconf --force (implied by --failover) sets the global obd_timeout
4401 to 5 seconds in order to quickly disconnect, but this caused
4402 other RPCs to time out too quickly. Do not change the global
4403 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4405 Severity : enhancement
4406 Frequency : if MDS is started with down OST
4407 Bugzilla : 9439,5706
4408 Description: Allow startup/shutdown of an MDS without depending on the
4409 availability of the OSTs.
4410 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4411 Add appropriate locking and lov-osc refcounts for safe
4412 cleaning. Add osc abort_inflight calls in case the
4413 synchronize never started.
4416 Frequency : occasional (Cray XT3 only)
4418 Description: root not authorized to access files in CRAY_PORTALS environment
4419 Details : The client process capabilities were not honoured on the MDS in
4420 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4421 been accessed by an authorized user then root was able to access
4422 the file on the local client also. The root user capabilities
4423 are now allowed on the MDS, as this environment has secure UID.
4426 Frequency : occasional
4428 Description: ldiskfs "too long searching" message happens too often
4429 Details : A debugging message (otherwise harmless) prints too often on
4430 the OST console. This has been reduced to only happen when
4431 there are fragmentation problems on the filesystem.
4436 Description: Division by zero in statfs when all OSCs are inactive
4437 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4438 lov_statfs divides by value returned by lov_get_stripecnt().
4442 Bugzilla : 9489, 3273
4443 Description: First write from each client to each OST was only 4kB in size,
4444 to initialize client writeback cache, which caused sub-optimal
4445 RPCs and poor layout on disk for the first writen file.
4446 Details : Clients now request an initial cache grant at (re)connect time
4447 and so that they can start streaming writes to the cache right
4448 away and always do full-sized RPCs if there is enough data.
4449 If the OST is rebooted the client also re-establishes its grant
4450 so that client cached writes will be honoured under the grant.
4455 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4456 Details : Now I/O RPCs go to different portal number and (presumably) fast
4457 lock requests (and glimses) and other RPCs get their own service
4458 threads pool that should be able to service those RPCs
4461 Severity : enhancement
4463 Description: Ability to exchange lustre version between client and servers and
4464 issue warnings at client side if client is too old. Also for
4465 liblustre clients there is ability to refuse connection of too old
4467 Details : New 'version' field is added to connect data structure that is
4468 filled with version info. That info is later checked by server and
4472 Frequency : rare, liblustre only.
4473 Bugzilla : 9296, 9581
4474 Description: Two simultaneous writes from liblustre at offset within same page
4475 might proceed at the same time overwriting eachother with stale
4477 Details : I/O lock withing llu_file_prwv was released too early, before data
4478 actually was hitting the wire. Extended lock-holding time until
4479 server acknowledges receiving data.
4482 Frequency : extremely rare. Never observed in practice.
4484 Description: avoid generating lustre_handle cookie of 0.
4485 Details : class_handle_hash() generates handle cookies by incrementing
4486 global counter, and can hit 0 occasionaly (this is unlikely, but
4487 not impossible, because initial value of cookie counter is
4488 selected randonly). Value of 0 is used as a sentinel meaning
4489 "unassigned handle" --- avoid it. Also coalesce two critical
4490 sections in this function into one.
4492 Severity : enhancement
4494 Description: allow liblustre clients to delegate truncate locking to OST
4495 Details : To avoid overhead of locking, liblustre client instructs OST to
4496 take extent lock in ost_punch() on client's behalf. New connection
4497 flag is added to handle backward compatibility.
4499 Severity : enhancement
4500 Bugzilla : 4928, 7341, 9758
4501 Description: allow number of OST service threads to be specified
4502 Details : a module parameter allows the number of OST service threads
4503 to be specified via "options ost ost_num_threads={N}" in the
4504 OSS's /etc/modules.conf or /etc/modprobe.conf.
4508 Bugzilla : 6146, 9635, 9895
4509 Description: servers crash with bad pointer in target_handle_connect()
4510 Details : In rare cases when a client is reconnecting it was possible that
4511 the connection request was the last reference for that export.
4512 We would temporarily drop the export reference and get a new
4513 one, but this may have been the last reference and the export
4514 was just destroyed. Get new reference before dropping old one.
4516 Severity : enhancement
4517 Frequency : if client is started with failover MDS
4519 Description: Allow multiple MDS hostnames in the mount command
4520 Details : Try to read the configuration from all specified MDS
4521 hostnames during a client mount in case the "primary"
4524 Severity : enhancement
4526 Description: Stop sending data to evicted clients as soon as possible.
4527 Details : Check if the client we are about to send or are sending data to
4528 was evicted already. (Check is done every second of waiting,
4529 for which l_wait_event interface was extended to allow checking
4530 of exit condition at specified intervals).
4533 Frequency : rare, normally only when NFS exporting is done from client
4535 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4537 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4540 Frequency : always, if srand() is called before liblustre initialization
4542 Description: Liblustre uses system PRNG disturbing its usage by user application
4543 Details : Introduce internal to lustre fast and high-quality PRNG for
4544 lustre usage and make liblustre and some other places in generic
4545 lustre code to use it.
4547 Severity : enhancement
4548 Bugzilla : 9477, 9557, 9870
4549 Description: Verify that the MDS configuration logs are updated when xml is
4550 Details : Check if the .xml configuration logs are newer than the config
4551 logs stored on the MDS and report an error if this is the case.
4552 Request --write-conf, or allow starting with --old_conf.
4554 Severity : enhancement
4556 Description: Handle symlinks in the path when checking if Lustre is mounted.
4557 Details : Resolve intermediate symlinks when checking if a client has
4558 mounted a filesystem to avoid duplicate client mounts.
4563 Description: lconf can hit an error exception but still return success.
4564 Details : The lconf command catches the Command error exception at the top
4565 level script context and will exit with the associated exit
4566 status, but doesn't ensure that this exit status is non-zero.
4571 Description: failure of ptlrpc thread startup can cause oops
4572 Details : Starting a ptlrpc service thread can fail if there are a large
4573 number of threads or the server memory is very fragmented.
4574 Handle this without oopsing.
4577 Frequency : always, only if liblustre and non-default acceptor port was used
4579 Description: liblustre cannot connect to servers with non-default acceptor port
4580 Details : tcpnal_set_default_params() was not called and was therefore
4581 ignoring the environment varaible TCPNAL_PORT, as well as other
4582 TCPNAL_ environment variables
4587 Description: two objects could be created on the same OST for a single file
4588 Details : If an OST is down, in some cases it was possible to create two
4589 objects on a single OST for a single file. No problems other
4590 than potential performance impact and spurious error messages.
4594 Bugzilla : 5681, 9562
4595 Description: Client may oops in ll_unhash_aliases
4596 Details : Client dcache may become inconsistent in race condition.
4597 In some cases "getcwd" can fail if the current directory is
4603 Description: Inode refcounting problems in NFS export code
4604 Details : link_raw functions used to call d_instantiate without obtaining
4605 extra inode reference first.
4609 Bugzilla : 9942, 9903
4610 Description: Referencing freed requests leading to crash, memleaks with NFS.
4611 Details : We used to require that call to ll_revalidate_it was always
4612 followed by ll_lookup_it. Also with revalidate_special() it is
4613 possible to call ll_revalidate_it() twice for the same dentry
4614 even if first occurence returned success. This fix changes semantic
4615 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4616 reference on a request referred from the intent.
4617 ll_intent_release() then releases such a request.
4620 Frequency : rare, normally benchmark loads only
4622 Description: unlinked inodes were kept in memory on the client
4623 Details : If a client is repeatedly creating and unlinking files it
4624 can accumulate a lot of stale inodes in the inode slab cache.
4625 If there is no other client load running this can cause the
4626 client node to run out of memory. Instead flush old inodes
4627 from client cache that have the same inode number as a new inode.
4630 Frequency : SLES9 2.6.5 kernel and long filenames only
4631 Bugzilla : 9969, 10379
4632 Description: utime reports stale NFS file handle
4633 Details : SLES9 uses out-of-dentry names in some cases, which confused
4634 the lustre dentry revalidation. Change it to always use the
4638 Frequency : rare, unless heavy write-truncate concurrency is continuous
4639 Bugzilla : 4180, 6984, 7171, 9963, 9331
4640 Description: OST becomes very slow and/or deadlocked during object unlink
4641 Details : filter_destroy() was holding onto the parent directory lock
4642 while truncating+unlinking objects. For very large objects this
4643 may block other threads for a long time and slow overall OST
4644 responsiveness. It may also be possible to get a lock ordering
4645 deadlock in this case, or run out of journal credits because of
4646 the combined truncate+unlink. Solution is to do object truncate
4647 first in one transaction without parent lock, and then do the
4648 final unlink in a new transaction with the parent lock. This
4649 reduces the lock hold time dramatically.
4652 Frequency : rare, 2.4 kernels only
4654 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4655 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4656 before unmounting their filesystem in order to avoid deadlock.
4657 The kernel_locked() function in 2.4 kernels only checks whether
4658 the kernel lock is held, not whether it is this process that is
4659 holding it as 2.6 kernels do.
4664 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4665 Details : The client ptlrpc code may be trying to reconnect to a down
4666 server before a previous connection attempt has timed out.
4667 Increase the reconnect interval to be longer than the connection
4668 timeout interval to avoid sending duplicate connections to
4672 Frequency : echo_client brw_test command
4674 Description: fix echo_client to work with OST preallocated code
4675 Details : OST preallocation code (5137) didn't take echo_client IO path
4676 into account: echo_client calls filter methods outside of any
4677 OST thread and, hence, there is no per-thread preallocated
4678 pages and buffers to use. Solution: hijack pga pages for IO. As
4679 a byproduct, this avoids unnecessary data copying.
4683 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4684 Description: Client can oops in mdc_commit_close() after open replay
4685 Details : It was possible for the MDS to return an open request with no
4686 transaction number in mds_finish_transno() if the client was
4687 evicted, but without actually returning an error. Clients
4688 would later try to replay that open and may trip an assertion
4689 Simplify the client close codepath, and always return an error
4690 from the MDS in case the open is not successful.
4693 Frequency : rare, 2.6 OSTs only
4695 Description: OST may deadlock under high load on fragmented files
4696 Details : If there was a heavy load and highly-fragmented OST filesystems
4697 it was possible to have all the OST threads deadlock waiting on
4698 allocation of biovecs, because the biovecs were not released
4699 until the entire RPC IO was completed. Instead, release biovecs
4700 as soon as they are complete to ensure forward IO progress.
4702 Severity : enhancement
4704 Description: Support for specifying external journal device at mount
4705 Details : If an OST or MDS device is formatted with an external journal
4706 device, this device major/minor is stored in the ext3 superblock
4707 and may not be valid for failover. Allow detecting and
4708 specifying the external journal at mount time.
4713 Description: Mounting an MDS with pending unlinked files may cause oops
4714 Details : target_finish_recovery() calls mds_postrecov() which returned
4715 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4716 considers this an error and immediately begins cleaning up the
4717 lov, just after starting the mds_lov process
4719 Severity : enhancement
4721 Description: Implement 'lfs df' to report actual free space on per-OST basis
4722 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4723 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4724 usage of objects; '-h' to report in human readable format.
4726 ------------------------------------------------------------------------------
4728 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4735 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4736 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4737 at startup during orphan unlinking. Instead, skip buddy update
4738 before setup, as it will be regenerated after recovery is complete.
4743 Description: If an OST is inactive, its locks might reference stale inodes.
4744 Details : lov_change_cbdata() must iterate over all namespaces, even if
4745 they are inactive to clear inode references from the lock.
4747 Severity : enhancement
4748 Frequency : occasional, if non-standard max_dirty_mb used
4750 Description: Client will block write RPCs if not enough grant
4751 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4752 then the client will block writes while waiting for another RPC
4753 to complete instead of consuming its dirty limit. With change
4754 we get improved performance when max_dirty_mb is small.
4756 Severity : enhancement
4757 Bugzilla : 3389, 6253
4758 Description: Add support for supplementary groups on the MDS.
4759 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4760 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4761 lookups for user supplementary groups into a cache.
4765 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4766 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4767 for liblustre O_EXCL is enforced by MDS.
4771 Description: readdir never returns NULL in liblustre.
4772 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4773 dentry in struct dirent.
4777 Frequency : liblustre only, depends on application IO pattern
4778 Description: liblustre clients evicted if not contacting servers
4779 Details : Don't put liblustre clients into the ping_evictor list, so
4780 they will not be evicted by the pinger ever.
4782 Severity : enhancement
4784 Description: Add ability to evict clients by NID from MDS.
4785 Details : By echoing "nid:$NID" string into
4786 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4787 $NID would be instantly evicted from this MDS and from all active
4788 OSTs connected to it.
4792 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4793 Details : lookup_it_finish() used to query file size from OSTs that was not
4798 Description: service threads change working directory to that of init
4799 Details : Starting lustre service threads may pin the working directory
4800 of the parent thread, making that filesystem busy. Threads
4801 now change to the working directory of init to avoid this.
4805 Frequency : during shutdown only
4806 Description: shutdown with a failed MDS or OST can cause unmount to hang
4807 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4812 Frequency : 2.6 only
4813 Description: chmod/chown may include an extra supplementary group
4814 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4815 supplementary group and if none is specified this is used.
4818 Bugzilla : 5479 (6816)
4820 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4821 Details : If lookup is in progress on a file that is unlinked we might try
4822 to revalidate the inode and fail in revalidate after lookup is
4823 complete and ll_file_open() enqueues the open again but
4824 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4827 Frequency : always, if lconf --abort_recovery used
4829 Description: lconf --abort_recovery fails with 'Operation not supported'
4830 Details : lconf was attempting to abort recovery on the MDT device and not
4833 ------------------------------------------------------------------------------
4835 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4840 Frequency : rare (only unsupported configurations with a node running as an
4842 Bugzilla : 6514, 5137
4843 Description: Mounting a Lustre file system on a node running as an OST could
4845 Details : OSTs now preallocates memory needed to write out data at
4846 startup, instead of when needed, to avoid having to
4847 allocate memory in possibly low memory situations.
4848 Specifically, if the file system is mounted on on OST,
4849 memory pressure could force it to try to write out data,
4850 which it needed to allocate memory to do. Due to the low
4851 memory, it would be unable to do so and the node would
4852 become unresponsive.
4854 Severity : enhancement
4856 Description: Addition of lconf --service command line option
4857 Details : lconf now accepts a '--service <arg>' option, which is
4858 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4860 Severity : enhancement
4862 Description: Failover mode is now the default for OSTs.
4863 Details : By default, OSTs will now run in failover mode. To return to
4864 the old behaviour, add '--failout' to the lmc line for OSTs.
4866 Severity : enhancement
4868 Description: Health checks are now provided for MDS and OSTs
4869 Details : Additional detailed health check information on MSD and OSTs
4870 is now provided through the procfs health_check value.
4873 Frequency : occasional, depends on IO load
4875 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4876 after numerous create/delete cycles
4877 Details : The ext3 inode allocation policy would not allocate new inodes
4878 very well on the OSTs because there are no new directories
4879 being created. Instead we look for groups with free space if
4880 the parent directories are nearly full.
4885 Description: Network or server problems during mount may cause partially
4886 mounted clients instead of returning an error.
4887 Details : The config llog parsing code may overwrite the error return
4888 code during mount error handling, returning success instead
4894 Description: MDS can fail to allocate large reply buffers
4895 Details : After long uptimes the MDS can fail to allocate large reply
4896 buffers (e.g. zconf client mount config records) due to memory
4897 fragmentation or consumption by the buffer cache. Preallocate
4898 some large reply buffers so that these replies can be sent even
4899 under memory pressure.
4903 Frequency : rare (liblustre)
4904 Description: fsx running with liblustre complained that using truncate() to
4905 extend the file doesn't work. This patch corrects that issue.
4906 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4907 ATTR_SIZE and lsm use in llu_setattr_raw.
4911 Frequency : rare, only 2.6 kernels
4912 Description: Unusual file access patterns on the MDS may result in inode
4913 data being lost in very rare circumstances.
4914 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4915 code in the 2.6 kernel caused the inode and/or EA data not to
4916 be read from disk, causing single-file corruption.
4920 Frequency : rare, only 2.6 filesystems using extents
4921 Description: Heavy concurrent write and delete load may cause data corruption.
4922 Details : It was possible under high-load situations to have an extent
4923 metadata block in the block device cache from a just-unlinked
4924 file overwrite a newly-allocated data block. We now unmap any
4925 metadata buffers that alias just-allocated data blocks.
4929 Frequency : filesystems with default stripe_count larger than 77
4930 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4931 Details : lconf specifies an inode size of 4096 bytes when the default
4932 stripe_count is larger than 77. This conflicts with the default
4933 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4934 this case to avoid pinning too much memory for large EAs.
4936 ------------------------------------------------------------------------------
4938 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4943 Frequency : rare (extremely heavy IO load with hundreds of clients)
4945 Description: Client is evicted, gets IO error writing to file
4946 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4947 caused clients to be evicted for AST timeouts. The fixes in
4948 bug 5192 mean we no longer need to have such short AST timeouts
4949 so ldlm_timeout has been increased.
4952 Frequency : occasional during --force or --failover shutdown under load
4953 Bugzilla : 5949, 4834
4954 Description: Server oops/LBUG if stopped with --force or --failover under load
4955 Details : a collection of import/export refcount and cleanup ordering
4956 issues fixed for safer force cleanup
4959 Frequency : only filesystems larger than 120 OSTs
4960 Bugzilla : 5990, 6223
4961 Description: lfs getstripe would oops on a very large filesystem
4962 Details : lov_getconfig used kfree on vmalloc'd memory
4965 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4966 Bugzilla : 6242, 6243
4967 Description: reading from files that had been truncated to a non-zero size
4968 but never opened returned no data
4969 Details : ll_file_read() reads zeros from no-object files to EOF
4974 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4975 Details : There's little harm in aborting MDS/OSS recovery and letting it
4976 try again, so I removed the LASSERT and return an error instead.
4978 Severity : enhancement
4980 Description: New debugging infrastructure for tracking down data corruption
4981 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4982 (b) cover more of the client-side code path, and (c) try to narrow
4983 down where problems occurred
4987 Bugzilla : 3819, 4364, 4397, 6313
4988 Description: Racing close and eviction MDS could cause assertion in mds_close
4989 Details : It was possible to get multiple mfd references during close and
4990 client eviction, leading to one thread referencing a freed mfd.
4992 Severity: : enhancement
4993 Bugzilla : 3262, 6359
4994 Description: Attempts to reconnect to servers are now more aggressive.
4995 Details : This builds on the enhanced upcall-less recovery that was added
4996 in 1.4.2. When trying to reconnect to servers, clients will
4997 now try each server in the failover group every 10 seconds. By
4998 default, clients would previously try one server every 25 seconds.
5003 Description: After recovery, certain operations trigger a failed
5004 assertion on a client.
5005 Details : Failing over an mds, using lconf -d --failover, while a
5006 client was doing a readdir() call would cause the client to
5007 LBUG after recovery completed and the readdir() was resent.
5009 Severity : enhancement
5011 Description: Default groups are now added by lconf
5012 Details : You can now run lconf --group <servicename> without having to
5013 manually add groups with lmc.
5016 Frequency : occasional
5018 Description: Nodes with an elan id of 0 trigger a failed assertion
5021 Frequency : always when accessing e.g. tty/console device nodes
5023 Description: tty and some other devices nodes cannot be used on lustre
5024 Details : file's private_data field is used by device data and lustre
5025 values in there got lost. New field was added to struct file to
5026 store fs-specific private data.
5029 Frequency : when exporting Lustre via NFS
5031 Description: NFSD failed occasionally when looking up a path component
5032 Details : NFSD is looking up ".." which was broken in ext3 directories
5033 that had grown large enough to become hashed.
5036 Frequency : Clusters with multiple interfaces not on the same subnet
5038 Description: Nodes will repeatedly try to reconnect to an interface which it
5039 cannot reach and report an error to the log.
5040 Details : Extra peer list entries will be created by lconf with some peers
5041 unreachable. lconf now validates the peer before adding it.
5044 Frequency : Only if a default stripe is set on the filesystem root.
5046 Description: Setting a default stripe on the filesystem root prevented the
5047 filesystem from being remounted.
5048 Details : The client was sending extra request flags in the root getattr
5049 request and did not allocate a reply buffer for the dir EA.
5052 Frequency : occasional, higher if lots of files are accessed by one client
5053 Bugzilla : 6159, 6097
5054 Description: Client trips assertion regarding lsm mismatch/magic
5055 Details : While revalidating inodes the VFS looks up inodes with ifind()
5056 and in rare cases can find an inode that is being freed.
5057 The ll_test_inode() code will free the lsm during ifind()
5058 when it finds an existing inode and then the VFS later attaches
5059 this free lsm to a new inode.
5063 Bugzilla : 6422, 7030
5064 Description: MDS deadlock between mkdir and client eviction
5065 Details : Creating a new file via mkdir or mknod (starting a transaction
5066 and getting the ns lock) can deadlock with client eviction
5067 (gets ns lock and trying to finish a synchronous transaction).
5070 Frequency : occasional
5071 Description: While starting a server, the fsfilt_ext3 module could not be
5073 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5074 kernels. Previously, lconf would still use the ext3 name
5075 when trying to load modules. Now, it will correctly use
5076 ext3 on 2.4 and ldiskfs on 2.6.
5078 Severity : enhancement
5079 Description: The default stripe count has been changed to 1
5080 Details : The interpretation of the default stripe count (0, to lfs
5081 or lmc) has been changed to mean striping across a single
5082 OST, rather than all available. For general usage we have
5083 found a stripe count of 1 or 2 works best.
5085 Severity : enhancement
5086 Description: Add support for compiling against Cray portals.
5087 Details : Conditional compiling for some areas that are different
5091 Frequency : occasional
5092 Bugzilla : 6409, 6834
5093 Description: Creating files with an explicit stripe count may lead to
5094 a failed assertion on the MDS
5095 Details : If some OSTs are full or unavailable, creating files may
5096 trigger a failed assertion on the MDS. Now, Lustre will
5097 try to use other servers or return an error to the
5101 Frequency : occasional
5103 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5104 caused lock timeout during readahead (since 1.4.2).
5105 Details : Processes doing readahead might match a lock that hasn't been
5106 granted yet if there are overlapping and conflicting lock
5107 requests. The readahead process waits on ungranted lock
5108 (original lock is CBPENDING), while OST waits for that process
5109 to cancel CBPENDING read lock and eventually evicts client.
5111 Severity : enhancement
5113 Description: Initial enabling of flock support for clients
5114 Details : Implements fcntl advisory locking and file status functions.
5115 This feature is provided as an optional mount flag (default
5116 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5117 locking are implemented yet, and those that are are not guaranteed
5118 to be completely correct in production environments.
5119 mount -t lustre -o [flock|noflock] ...
5122 Frequency : occasional
5124 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5125 assertion in the ext3 JBD (journaling) layer.
5126 Details : The b_committed_data struct is protected by the big kernel lock
5127 in 2.4 kernels, serializing journal_commit_transaction() and
5128 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5129 access to this struct. In 2.6 kernels there is finer grained
5130 locking to improve SMP performance of the JBD layer.
5134 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5139 Description: merge and cleanup kernel patches.
5142 -----------------------------------------------------------------------------
5144 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5146 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5147 versions of Lustre. Please update all servers and clients to
5148 version 1.4.2 or later at the same time. You must also run
5149 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5150 to update the configuration logs.
5152 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5153 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5154 - swab llog records of type '0' so we get proper header size/idx (5861)
5155 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5156 - fix rename of one directory over another leaking an inode (5953)
5157 - avoid SetPageDirty on 2.6 (5981)
5158 - don't re-add just-being-destroyed locks to the waiting list (5653)
5159 - when creating new directories, inherit the parent's custom
5160 striping settings if present parent (3048)
5161 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5162 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5163 - don't hold client locks on temporary worklist from l_lru (5666)
5164 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5165 - automatically evict dead clients (5921)
5166 - Update file size properly in create+truncate+fstat case (6196)
5167 - Do not unhash mountpoint dentries, do not allow removal of
5169 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5170 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5171 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5172 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5173 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5174 - Correctly handle reads of files with no objects (6243)
5175 - lctl recover will also mark a device active if deactivate used (5933)
5177 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5178 - allow --write-conf on an MDS with different nettype than client (5619)
5179 - don't write config llogs to MDS for mounts not from that MDS (5617)
5180 - lconf should create multiple TCP connections from a client (5201)
5181 - init scripts are now turned off by default; run chkconfig --on
5182 lustre and chkconfig --on lustrefs to use them
5183 - upcalls are no longer needed for clients to recover to failover
5185 - add --abort-recovery option to lconf to abort recovery on device
5187 - add support for an arbitrary number of OSTs (3026)
5188 - Quota support protocol changes.
5189 - forward compatibility changes to wire structs (6007)
5190 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5191 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5192 - improved "lustre status" script
5193 - initialize blocksize for non-regular files (6062)
5194 - added --disable-server and --disable-client configure options (5782)
5195 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5196 - Vanilla 2.4.29 support
5197 - increase maximum number of obd devices to 520 (6242)
5198 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5199 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5200 SLES 9 2.6.5 kernels
5201 - limit stripes per file to 160 (the maximum EA size) (6093)
5203 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5206 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5207 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5208 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5209 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5210 - small ext3 extents cleanups and fixes (5733)
5211 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5212 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5213 - avoid needless client->OST connect, fix handle mismatch (5317)
5214 - fix DLM error path that led to out-of-sync client, long delays (5779)
5215 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5216 - fix several locking issues related to i_size (5492,5624,5654,5672)
5217 - don't move pending lock onto export if it is already evicted (5683)
5218 - fix kernel oops when creating .foo in unlinked directory (5548)
5219 - fix deadlock in obdfilter statistics vs. object create (5811)
5220 - use time_{before,after} to avoid timer jiffies wrap (5882)
5221 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5222 - Do not leak request if server was not able to process it (5154)
5223 - If mds_open unable to find parent dir, make that negative lookup(5154)
5224 - don't create new directories with extent-mapping (5909, 5936)
5226 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5227 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5228 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5229 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5230 - support for RHEL4 kernel on i686 (5773)
5231 - provide error messages when incompatible logs are encountered (5898)
5233 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5234 * version 1.4.0.10 (1.4.1 release candidate 1)
5236 - don't keep a lock reference when lock is not granted (4238)
5237 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5238 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5239 - drop import inflight refcount on signal_completed_replay error (5255)
5240 - unlock page after async write error during send (3677)
5241 - handle missing objects in filter_preprw_read properly (5265)
5242 - no transno return for symlink open, don't save no-trasno open (3440)
5243 - don't try to complete elan receive that already failed (4012)
5244 - free RPC server reply state on error (5406)
5245 - clean up thread from ptlrpc_start_thread() on error (5160)
5246 - readahead could read extra page into cache that wasn't ejected (5388)
5247 - prevent races in class_attach/setup/cleanup/detach (5260)
5248 - don't dereference de->d_inode after l_dput of de (5458)
5249 - use "int" for stripe value returned from lock_to_stripe (5544)
5250 - mballoc allocation and error-checking fixes in 2.6 (5504)
5251 - block device patches to fix I/O request sizes in 2.6 (5482)
5252 - look up hostnames for IB nals (5602)
5253 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5254 - don't start multiple acceptors for the same port (5277)
5255 - fix incorrect LASSERT in mds_getattr_name (5635)
5256 - export a proc file for general "ping" checking (5628)
5257 - fix "lfs check" to not block when the MDS is down (5628)
5259 - service request history (4965)
5260 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5261 - create an "evict_client" /proc entry on OSTs, like the MDS has
5262 - fix mount usage message, return errors per mount(8) (5168)
5263 - change grep [] to grep "[]" in tests so they work in more UMLs
5264 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5265 - remove ancient LOV_MAGIC_V0 EA support (5047)
5266 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5267 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5268 - allow readahead allocations to fail when low on memory (5383)
5269 - mmap locking landed again, after considerable improvement (2828)
5270 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5272 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5275 - send OST transaction number in read/write reply to free req (4966)
5276 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5277 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5278 - the watchdog thread now runs as interruptible (5246)
5279 - flock/lockf fixes (but it's still disabled, pending 5135)
5280 - don't use EXT3 constants in llite code (5094)
5281 - memory shortage at startup could cause assertion (5176)
5283 - reorganization of lov code
5284 - single portals codebase
5286 - add extents/mballoc support (5025)
5287 - direct I/O reads in the obdfilter (4048)
5288 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5290 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5293 - send OST transaction number in read/write reply to free req (4966)
5294 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5295 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5296 - the watchdog thread now runs as interruptible (5246)
5297 - handle missing objects in filter_preprw_read properly (5265)
5298 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5299 - drop import inflight refcount on signal_completed_replay error (5255)
5300 - unlock page after async write error during send (3677)
5301 - return original error code on reconstructed replies (3761)
5302 - no transno return for symlink open, don't save no-trasno open (3440)
5304 - add pid to ldlm debugging output (4922)
5305 - bump the watchdog timeouts -- we can't handle 30sec yet
5306 - extra debugging for orphan dentry/inode bug (5259)
5308 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5311 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5312 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5313 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5314 - fix NULL dereference in /proc/sys/portals/routes (4827)
5315 - allow failed mdc_close() operations to be interrupted (4561)
5316 - stop precreate on OST before MDS would time out on it (4778)
5317 - don't send partial-page writes before EOF from client (4410)
5318 - discard client grant for sub-page writes on large-page clients (4520)
5319 - don't free dentries not owned by NFS code, check generation (4806)
5320 - fix lsm leak if mds_create_objects() fails (4801)
5321 - limit debug_daemon file size, always print CERROR messages (4789)
5322 - use transno after validating reply (3892)
5323 - process timed out requests if import state changes (3754)
5324 - update mtime on OST during writes, return in glimpse (4829)
5325 - add mkfsoptions to LDAP (4679)
5326 - use ->max_readahead method instead of zapping global ra (5039)
5327 - don't interrupt __l_wait_event() during strace
5329 - add software watchdogs to catch hung threads quickly (4941)
5330 - make lustrefs init script start after nfs is mounted
5331 - fix CWARN/ERROR duplication (4930)
5332 - return async write errors to application if possible (2248)
5333 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5334 - print NAL number in %x format (4645)
5335 - update barely-supported suse-2.4.21-171 series (4842)
5336 - support for sles 9 %post scripts
5337 - support for building 2.6 kernel-source packages
5338 - support for sles km_* packages
5340 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5343 - ignore -ENOENT errors in osc_destroy (3639)
5344 - notify osc create thread that OSC is being cleaned up (4600)
5345 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5346 - reconstruct ost_handle() like mds_handle() (4657)
5347 - create a new thread to do import eviction to avoid deadlock (3969)
5348 - let lconf resolve symlinked-to devices (4629)
5349 - don't unlink "objects" from directory with default EA (4554)
5350 - hold socknal file ref over connect in case target is down (4394)
5351 - allow more than 32000 subdirectories in a single directory (3244)
5352 - fix blocks count for O_DIRECT writes (3751)
5353 - OST returns ENOSPC from object create when no space left (4539)
5354 - don't send truncate RPC if file size isn't changing (4410)
5355 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5356 - bind to privileged port in socknal and tcpnal (3689)
5358 - rate limit CERROR/CWARN console message to avoid overload (4519)
5359 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5360 - basic mmap support (3918)
5361 - kernel patch series update from b1_4 (4711)
5363 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5366 - avoid crash during MDS cleanup with OST shut down (2775)
5367 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5368 - don't use bad inodes on the MDS (3744)
5369 - dynamic object preallocation to improve recovery speed (4236)
5370 - don't hold spinlock over lock dumping or change debug flags (4401)
5371 - don't zero obd_dev when it is force cleaned (3651)
5372 - print grants to console if they go negative (4431)
5373 - "lctl deactivate" will stop automatic recovery attempts (3406)
5374 - look for existing locks in ldlm_handle_enqueue() (3764)
5375 - don't resolve lock handle twice in recovery avoiding race (4401)
5376 - revalidate should check working dir is a directory (4134)
5378 - don't always mark "slow" obdfilter messages as errors (4418)
5380 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5383 - don't close LustreDB during write_conf until it is done (3860)
5384 - fix typo in lconf for_each_profile (3821)
5385 - allow dumping logs from multiple threads at one time (3820)
5386 - don't allow multiple threads in OSC recovery (3812)
5387 - fix debug_size parameters (3864)
5388 - fix mds_postrecov to initialize import for llog ctxt (3121)
5389 - replace config semaphore with spinlock (3306)
5390 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5391 - don't allow enqueue to complete on a destroyed export (3822)
5392 - down write_lock before checking llog header bitmap (3825)
5393 - recover from lock replay timeout (3764)
5394 - up llog sem before sending rpc (3652)
5395 - reduce ns lock hold times when setting kms (3267)
5396 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5397 - fix NULL deref and obd_dev leak on setup error (3312)
5398 - replace some LBUG about llog ops with error handling (3841)
5399 - don't match INVALID dentries from d_lookup and spin (3784)
5400 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5401 - fix invalid assertion in ptlrpc_set_wait (3880)
5403 - add libwrap support for the TCP acceptor (3996)
5404 - add /proc/sys/portals/routes for non-root route listing (3994)
5405 - allow setting MDS UUID in .xml (2580)
5406 - print the stack of a process that LBUGs (4228)
5408 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5411 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5412 - make sure to unset replay flag from failed open requests (3440)
5413 - if default stripe count is 0, use OST count for inode size (3636)
5414 - update parent mtime/ctime on client for create/unlink (2611)
5415 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5416 - free recovery state on server during a forced cleanup (3571)
5417 - unregister_reply for resent reqs (3063)
5418 - loop back devices mounting and status check on 2.6 (3563)
5419 - fix resource-creation race that can provoke i_size == 0 (3513)
5420 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5421 - more debugging for page-accounting assertion (3746)
5422 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5423 - avoid deadlock after precreation failure (3758)
5424 - fix race and lock order deadlock in orphan handling (3450, 3750)
5425 - add validity checks when grabbing inodes from l_ast_data (3599)
5427 - add /proc/.../recovery_status to obdfilter (3428)
5428 - lightweight CDEBUG infrastructure, debug daemon (3668)
5429 - change default OSC RPC parameters to be better on small clusters
5430 - turn off OST read cache for files smaller than 32MB
5431 - install man pages and include them in rpms (3100)
5432 - add new init script for (un)mounting lustre filesystems (2593)
5433 - run chkconfig in %post for init scripts (3701)
5434 - drop scimac NAL (unmaintained)
5436 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5439 - clean kiobufs before and after use (3485)
5440 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5441 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5442 - change init script to fail more clearly if not run as root (1528)
5443 - allow clients to reconnect during replay (1742)
5444 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5445 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5446 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5447 - fix cond_resched (3554)
5448 - only evict unfinished clients after recovery (3515)
5449 - allow bulk resend, prevent data loss (3570)
5450 - dynamic ptlrpc request buffer allocation (2102)
5451 - don't allow unlinking open directory if it isn't empty (2904)
5452 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5453 - remove extraneous obd dereference causing LASSERT failure (3334)
5454 - don't use get_cycles() when creating temp. files on the mds (3156)
5455 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5456 - handle EEXIST for set-stripe, set proper directory name (3336)
5458 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5459 - fix ksocknal_fmb_callback() error messages (2918)
5461 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5464 - don't copy lvb into (possibly NULL) reply on error (2983)
5465 - don't deref dentry after dput, don't free lvb on error (2922)
5466 - use the kms to determine writeback rpc length (2947)
5467 - increment oti_logcookies when osc is inactive (2948)
5468 - update client's i_blocks count via lvb messages (2543)
5469 - handle intent open/close of special files properly (1557)
5470 - mount MDS with errors=remount-ro, like obdfilter (2009)
5471 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5472 - don't use cancelling-locks' kms values (2947)
5473 - use highest lock extent for kms, not last one (2925)
5474 - don't dereference ERR_PTR() dentry in error handling path (3107)
5475 - fix thread race in portals_debug_dumplog() (3122)
5476 - create lprocfs device entries at setup instead of at attach (1519)
5477 - common AST error handler, don't evict client on completion race (3145)
5478 - zero nameidata in detach_mnt in 2.6 (3118)
5479 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5480 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5481 - initialize RPC timeout timer earlier for 2.6 (3219)
5482 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5483 - print nal/nid for unknown nid (3258)
5484 - additional checks for oscc recovery before doing precreate (3284)
5485 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5486 - don't crash in mdc_close for bad permissions on open (3285)
5487 - zero i_rdev for non-device files (3147)
5488 - clear page->private before handing to FS, better assertion (3119)
5489 - tune the read pipeline (3236)
5490 - fix incorrect decref of invalidated dentry (2350)
5491 - provide read-ahead stats and refine rpc in flight stats (3328)
5492 - don't hold journal transaction open across create RPC (3313)
5493 - update atime on MDS at close time (3265)
5494 - close LDAP connection when recovering to avoid server load (3315)
5495 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5496 - don't leak open file on MDS after open resend (3325)
5497 - serialize filter_precreate and filter_destroy_precreated (3329)
5498 - loop device shouldn't call sync_dev() for nul device (3092)
5499 - clear page cache after eviction (2766)
5500 - resynchronize MDS->OST in background (2824)
5501 - refuse to mount the same filesystem twice on same mountpoint (3394)
5502 - allow llmount to create routes for mounting behind routers (3320)
5503 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5504 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5505 - fix rare problem with rename on htree directories (3417)
5507 - allow default OST striping configuration per directory (1414)
5508 - fix compilation for qswnal for 2.6 kernels (3125)
5509 - increase maximum number of MDS request buffers for large systems
5510 - change liblustreapi to be useful for external progs like lfsck (3098)
5511 - increase local configuration timeout for slow disks (3353)
5512 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5514 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5517 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5518 - don't overwrite extent policy data in reply if lock was blocked (2901)
5519 - drop filter export grants atomically with removal from device (2663)
5520 - del obd_self_export from work_list in class_disconnect_exports (2908)
5521 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5522 - swab reply message in mdc_close, other PPC fixes (2464)
5523 - fix destroying of named logs (2325)
5524 - overwrite old logs when running lconf --write_conf (2264)
5525 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5526 - fix race in target_handle_connect (2898)
5527 - mds_reint_create() should take same inode create lock (2926)
5528 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5529 - don't close files for self_export to avoid uninitialized obd (2936)
5530 - allow MDS with the same name as client node (2939)
5531 - hold dentry reference for closed log files for unlink (2325)
5532 - reserve space for all logs during transactions (2059)
5533 - don't evict page beyond end of stripe extent (2925)
5534 - don't oops on a deleted current working directory (2399)
5535 - handle hard links to targets without a parent properly (2517)
5536 - don't dereference NULL lock when racing during eviction (2867)
5537 - don't grow lock extents when lots of conflicting locks (2919)
5539 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5542 - account for cache space usage on clients to avoid data loss (974)
5543 - lfsck support in lustre kernel code (2349)
5544 - reduce journal credits needed for BRW writes (2370)
5545 - orphan handling to avoid losing space on client/server crashes
5546 - ptlrpcd can be blocked, stopping ALL progress (2477)
5547 - use lock value blocks to assist in proper KMS, faster stat (1021)
5548 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5549 - recovery for initial connections (2355)
5550 - fixes for mds_cleanup_orphans (1934)
5551 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5552 - block all file creations until orphan recovery completes (1901)
5553 - client remove rq_connection from request struct (2423)
5554 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5555 - recovery timer race (2670)
5556 - mdc_close recovey bug (2532)
5557 - ptlrpc cleanup bug (2710)
5558 - mds timeout on local locks (2588)
5559 - namespace lock held during RPCs (2431)
5560 - handle interrupted sync write properly (2503)
5561 - don't try to handle a message that hasn't been replied to (2699)
5562 - client assert failure during cleanup after abort recovery (2701)
5563 - leak mdc device after failed mount (2712)
5564 - ptlrpc_check_set allows timedout requests to complete (2714)
5565 - wait for inflight reqs when ptlrpcd finishes (2710)
5566 - make sure unregistered services are removed from the srv_list
5567 - reset bulk XID's when resending them (caught by 1138 test)
5568 - unregister_bulk after timeout
5569 - fix lconf error (2694)
5570 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5571 - readahead locks pages, leaves pending causing memory pressure (2673)
5572 - increase OST request buffers to 4096 on large machines (2729)
5573 - fix up permission of existing directories in simple_mkdir (2661)
5574 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5575 - don't assign transno to errored transactions (2742)
5576 - don't delete objects on OST if given a bogus objid from MDS (2751)
5577 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5578 - if rq_no_resend, then timeout request after recovery (2432)
5579 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5580 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5581 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5582 - MDS recovery shouldn't delete objects that it has given out (2730)
5583 - if enqueue arrives after completion, don't clobber LVB (2819)
5584 - don't unlock pages twice when trigger_group_io returns error (2814)
5585 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5586 - don't write pages to disk if there was an error (1450)
5587 - don't ping imports that have recovery disabled (2676)
5588 - take buffered bytes into account when balancing socknal conn (2817)
5589 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5590 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5591 - remove little-endian swabbing of llog records (1987)
5592 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5593 - retry reposting request buffers if they fail (1191)
5594 - grow extent at grant time to avoid granting a revoked lock (2809)
5595 - lock revoke doesn't evict page if covered by a second lock (2765)
5596 - disable VM readahead to avoid reading outside lock extents (2805)
5598 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5599 - updated kernel patches for hp-2.4.20 kernel (2681)
5601 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5604 - fix truncated write corruption (2366)
5605 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5607 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5608 - ensure len(uuid) < 37 in lmc (1171)
5609 - fix ia64 OOPS in llog_test (2255)
5610 - zero end of page at obdfilter for partial page writes (2648)
5611 - don't leave stale dentries around after renames (bug 2428)
5612 - fix timeouts when evicting a client with a single lock held (2642)
5613 - set deadline for the initial HELLO message to drain (2634)
5614 - print out dotted-quad IP addresses in the socknal (2302)
5616 - additional debugging for MDS client eviction problem (2443)
5617 - fix mkfsoptions support for osts (2603, 2604)
5619 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5622 - add series for the vanilla 2.6.0 kernel
5623 - add series for the vanilla 2.4.24 kernel
5624 - add series for a cray x86/64 UL kernel drop
5625 - fix xattr patches for the vanilla 2.4.19 series
5627 - generate true UUIDs in lmc (1171)
5628 - have portals stack dumping break in UML (2466)
5629 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5630 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5631 - wake callback waiting threads on client eviction (2460)
5632 - Add --ptldebug and --subsystem to lmc (1719)
5633 - update assertion to allow safe interrupt allocation
5634 - set rq_no_resend for cancel requests (2432)
5635 - recalculate ptlrpcd timeout after resend (2494)
5636 - call vfs_rmdir when removing pending directories (2368)
5637 - fix renaming a file to itself (2429)
5638 - lmc creates a default one-stripe lov (2454)
5639 - expand procfs space to handle large clusters (2326)
5640 - increase UML stack to avoid overflow
5641 - update lconf's list of debug and subsystem masks
5642 - fix lfs find --obd (2510)
5643 - /proc tunable for disabling filter read caching (2591)
5644 - stop rpm packages from altering slapd.conf (2301)
5645 - disable nagle in the socknal under 0conf (2578)
5646 - choose mds inode size based on stripe count (2572)
5647 - fix kernel-source rpm problems (2516)
5649 - add --disable-doc to avoid pdf generation (2421)
5650 - update documentation, tests, type-os, comments
5651 - avoid format warnings on ia64
5652 - remove the TOE NAL
5653 - tiny code cleanups by removing unused fields
5655 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5658 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5659 - remove the most blatant lies from BUILDING file (2371)
5660 - change default debug level to reasonable production setting
5661 - reduce client side cache size to reduce cache flush time
5662 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5663 - make TCP zerocopy and pinger support enabled by default (2476)
5664 - sync writes completed after process exits caused crashes (2319)
5665 - maintain correct mount count on the MDS (2356)
5666 - backout 1557, because 2316 wasn't really fixed
5667 - better file I/O statistics gathering in /proc
5668 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5669 - another kernel patch to fix zero-copy TCP function export
5670 - don't take duplicate lock when processing re-sent getattr (2420)
5671 - lctl uses obd_self_export instead of creating new conn (2353)
5672 - MDS/OST recovery case which requires object creation asserted (2425)
5673 - move lfs from /usr/sbin to /usr/bin in packages
5674 - fix race between mds_client_add and mds_client_free (2417)
5675 - use kmalloc instead of slabs in portals (2430)
5676 - don't create duplicate records when a failover MDS is present (2442)
5677 - remove unnecessary mount age check (2332)
5678 - don't remove directory inodes from locks prematurely (2451)
5679 - don't break if MDS service name is the same as hostname (2103)
5680 - fix races in client write RPC generation when cache full (2482)
5682 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5685 - remove now-unused request->rq_obd (278)
5686 - if an allocation fails, print out how much memory we've used (1933)
5687 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5688 - add the "configurable stack size" patch to most series files (1256)
5689 - ability to write large log records, for 100+ OST configs (2306)
5690 - fix NULL deref when filter_prep fails (2314)
5691 - fix operator precedence error in filter_sync
5692 - dynamic allocation of socknal TX descriptors (2315)
5693 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5694 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5695 - ensure that CWARN messages in Portals always get to the syslog
5696 - __init/__exit are not for prototype decls (ldlm_init/exit)
5697 - x86-64 compile warning fixes
5698 - fix gateway LMC keyword conflict (2318)
5699 - fix MDS lock inversions in getattr/reint paths (1844)
5700 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5701 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5702 - DLM race condition prevented some lock evictions (2328)
5703 - ENOMEM detection and retry on socknal sends (2230)
5704 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5705 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5706 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5707 - better error messages when a client is rejected during recovery (1505)
5708 - avoid cancelling locks which were never granted, after failure (2330)
5709 - fix i_sem/journal inversion in mds_client_add (2333)
5710 - fix truncate/getattr lock cycle deadlock (2334)
5711 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5712 - fix two rare exit paths which could leak an l_lock() ref (2321)
5713 - fencepost error in MDS/OST orphan recovery (2226)
5714 - make log record alignment 8 bytes (1988)
5715 - lstripe now fails when requested offset > ost_count (2237)
5716 - ensure that all kernel series have a complete list.h (1607)
5717 - fix crashes in special-file operations (2316)
5718 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5720 - allow configurable automake binary, for testing new versions
5721 - small update to the lfs documentation
5723 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5725 * fix negative export reference count in fsfilt_sync (2312)
5727 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5728 * release candidate 0.9.1
5730 - orphans are moved into the PENDING directory for possible recovery
5731 - replayed opens now open by fid for orphan/rename safety (1042)
5732 - last close of an orphan inode generates a transno (683)
5733 - chdir() and mount() now pin the directory entry (1020)
5734 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5735 - discard very old requests without processing them (1502)
5736 - remove some common, well-understood CERRORs (1505)
5737 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5738 - clear "grant" flags in OST replies until OST grant code lands (1644)
5739 - fix read performance by not clobbering i_blksize on client (1598)
5740 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5741 - make LRU size a /proc tunable, clears locks when reduced (707)
5742 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5743 - server locks take references on exports now (1558)
5744 - build fixes for 2.4.20-rh trees (1663)
5745 - return an error from lov_create if all OSCs are inactive (1751)
5746 - fix import levels when a reconnect happens without a timeout (1597)
5747 - exit early from mds_open if we get a lookup error (1749)
5748 - partial page read at EOF wouldn't wait for disk before sending (1642)
5749 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5750 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5751 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5752 - write/truncate lock inversion (1639)
5753 - Don't auto-load obdclass, portals modules during cleanup (1495)
5754 - fix timestamps from jumping to "now" (1763)
5755 - extra journal assertions (1648)
5756 - add an extra multiunlink test (1771)
5757 - fix read_record/write_record API (1776)
5758 - fix leak of offset_extent, possible incorrect i_size later (1772)
5759 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5760 - Debugging for the kqswnal_get_idle_tx problems (1820)
5761 - Allow recovery to be attempted multiple times (1536)
5762 - Write out MDS last_rcvd file after it is first created (1600)
5763 - Fix tx_descriptor leak in failed transmit situations (1827)
5764 - ext3 journaling fixes for assertion failure after IO error (1871)
5765 - class_export_put() on freed export after completion AST error (1896)
5766 - Fix revalidate looping in VFS (1322)
5767 - Don't access a freed export during MDS_REINT timeout (1521)
5768 - Add open-unlink recovery support on the MDS (1673,1764)
5769 - Return an error if no MDS data was read from last_rcvd (1946)
5770 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5771 - Don't setup a disk device that doesn't match exported UUID (317)
5772 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5773 - avoid committing NULL handle in force close
5774 - local.sh is now a one-stripe LOV configuration
5775 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5776 - fix ext3 htree duplicate directory entry corruption (1516)
5777 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5778 - update to new LOV EA format (2097)
5779 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5781 2003-06-15 Phil Schwan <phil@clusterfs.com>
5784 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5785 - per-import/export recovery handling (958, 931, 959)
5786 - multiple last-rcvd slots, for serving multiple FSes (949)
5787 - connections are again shared between multiple imp/exports (963, 964)
5788 - "umount -f" would hang if any requests needed to be sent (393, 978)
5789 - avoid pinning large req buffer by copying for queued messages (989)
5790 - add "uuid" to "lctl device" command to help upcalls (991)
5791 - "open" RPCs with transnos would confuse recovery counters (1037)
5792 - do proper endian conversion of all wire messages (288, 340, 891)
5793 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5794 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5795 - fix ptlrpc_connection leak in target_handle_connect (1174)
5796 - fix import refcounting bug in OST and MDS cleanup (1134)
5797 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5798 - fix very unlikely obd_types race condition (501)
5799 - remove osc_open hack for echo_client (1187)
5800 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5801 - a failure in read_inode2 leads to deadlock (1139)
5802 - cancel ack-locks as soon as transaction is committed (1072)
5803 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5804 - make sure to commitrw after any preprw to avoid deadlock (1162)
5805 - failing to execute a file in a lustre FS would lock inode (1203)
5806 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5807 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5808 - obd_uuid2tgt would walk off the end of the list (1255)
5809 - on IA64 the osc would give portals incorrect bulk size (1258)
5810 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5811 - fix lock inversion caused by new llite matching code (1282)
5812 - limit the number of dirty pages on a client to 10MB (1286)
5813 - timed out locks were not being corrected cancelled (1289)
5814 - fix O_DIRECT above 4GB on IA-32 (1292)
5815 * major user-visible changes
5816 - fail out/fail over policy now controlled by the upcall (993)
5818 - add OBD_PING to check server availability and failure (954)
5819 - lustre messages are now sent in sending host order (288, 340, 891)
5820 - add eadatalen to MDS getattr reply (340)
5821 - OST read replies may contain second buffer, with per-page status (593)
5823 2003-03-11 Phil Schwan <phil@clusterfs.com>
5826 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5827 - failed open()s could cause deadlock; fixed (867, 869)
5828 - stop cancelling OST locks when files are closed (481)
5829 - overlapping XID spaces caused network corruption (851, 853)
5830 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5831 - setattr_raw added, to do single-RPC, server-side setattrs
5832 - lmc/lconf syntax change for OST UUIDs
5833 - fix crashy race condition between ptlrpc_free_req and osc_close
5834 - don't use request in mdc_enqueue if we hit a timeout (889)
5835 - don't set the inode i_size for regular files from the MDS (896)
5836 - handle out of order completion AST (842)
5837 - don't LBUG if a lock request times out after receiving AST (913)
5838 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5839 - if a bad lock AST arrives, send an error instead of dropping entirely
5840 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5841 - fix leak in bulk IO when only partially completed (899, 900, 926)
5842 - fix O_DIRECT for ia64 (55)
5843 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5844 - C-z after timeout could hang a process forever; fixed (977)
5846 - client-side I/O cache (678, 924, 929, 941, 970)
5848 - READPAGE and SETATTRs which don't take server-side locks get
5851 2003-02-11 Phil Schwan <phil@clusterfs.com>
5854 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5855 - Fix fsfilt_extN_readpage() to read a full page of directory
5856 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5857 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5858 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5859 - Most intent operations were removed, in favour of a new RPC mode
5860 that does a single RPC to the server and bypasses most of the VFS
5861 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5862 - Aggressively cancel local locks on DLM servers
5863 - mds_reint_unlink sends EA to the client if it's the last nlink.
5864 client uses that EA to unlink OST objects.
5865 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5866 - recursive symlinks were fixed (439)
5867 - fixed NULL deref in DEBUG_REQ
5868 - filter_update_lastobjid no longer calls sync, which annoyed extN
5869 - fixed multi-client small-writes to a single file problem (445)
5870 - fixed mtime updates during file writes (607)
5871 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5872 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5873 - fixed Linux OST size reporting problem (444, 656)
5874 - OST now updates object mtime with writes or setattr (607, 619)
5875 - client verifies file size before zeroing page past EOF (445)
5876 - OST now writes last allocated objid to disk with allocation (108)
5877 - LOV on echo now works (409)
5879 - mds_reint_unlink sends a new buffer, with the EA included. this
5880 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5881 set if a regular file was being unlinked, and it was the last link
5882 - use PtlGet from the target for bulk writes (315)
5883 - OST now updates object mtime with writes or setattr (607, 619)
5884 - LDLM now has a grant-time callback to revalidate locked items, if
5886 - Many MDS operations were reorganized to combat race conditions
5888 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5889 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5891 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5894 - Fully reactivate OST imports after reconnection (512, others)
5895 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5896 - More graceful error handling for truncating on dead OST (515)
5897 - Don't error out unless we're actually accessing dead stripes (474)
5898 - Fix garbage sizes when stripes are missing (410)
5899 - LRU counters were broken, causing constant lock purge (433, 432)
5900 - garbage on read from stripes with failed OSTs (441)
5901 - mark OSCs as active before reconnecting during recovery (438)
5902 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5903 - lfind did not preserve OST order in output (443)
5904 - symlinks cause hung clients, incorrect data (439)
5905 - stop dereferencing request after dropping refcount (457)
5906 - don't LASSERT(spin_is_locked) on non-SMP (455)
5907 - fixes for many rename() bugs
5908 - fstat didn't correctly synchronize attributes (399)
5909 - server must handle lock cancellation during blocking AST prep (487)
5910 - bulk descriptors were free()d too soon (511)
5911 - fix paths in lconf, which would load incorrect modules (451, 507)
5912 - fix confusing lconf 'host not found' error message (386)
5913 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5914 - fix race condition in mdc_blocking_ast() for inode access (526)
5915 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5916 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5917 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5918 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5919 - fix lstripe in several ways (kernel side) (527)
5920 - fix request leak in ldlm_cli_enqueue (262)
5921 - incorrect OSC was marked inactive after OST failure
5922 - call mds_fs_cleanup before unmounting filesystem (524)
5923 - fix races between taking ns_lock and ldlm_lock_change_resource
5924 - fix races updating LOV export open file list
5925 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5926 - fix recovery NULL deref in ldlm_cli_cancel_unused
5927 - fix some DLM races by using new hash table for lock handles (419)
5928 - permit the client to specify desired inodes, at replay
5929 - duplicate requests when we queue them for replay reintegration
5930 - fix last_rcvd offset calculation
5931 - sync after each recovered transaction, so we always make progress
5932 - never, not always, ERESTART requests without transnos
5933 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5935 - skip replay if the MDS says that the client is already connected
5936 - don't check for a recovery-enabled export to match lctl's UUID
5937 - don't INC_USE_COUNT for phantom exports
5938 - don't crash when cleaning up phantom exports (567)
5939 - don't double-finish or set replay data for errored mdc_open requests
5940 - abort requests when they time out, so we don't get old replies
5941 - send/receive replies for AST messages again
5942 - if the client says that it doesn't have the lock, cancel it on the
5944 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5945 mark it as destroyed, it will all work out when decref is called
5946 - fix module use counts (22, 581)
5948 - ASTs now expect a reply (server cancels lock on error reply)
5950 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5953 - fix many simultaneous client startup (392)
5954 - fix dentry->d_it clobbering
5955 - credentials weren't being shipped for readdir/getattr operations
5956 - remove invalid assertions triggered during some concurrent MD
5958 - proper Lustre versions added (336, 389)
5959 - fix memory leak for create error case (398)
5960 - fix LOV locking bug that would get cli/srv out of sync
5961 - fix echo client over LOV (409)
5962 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5963 - fix double-O_EXCL intent crash (424)
5964 - avoid sending multiple lock CANCELs (352)
5966 - MDS can do multi-client recovery (modulo bugs in new code)
5968 - many updates, edits, cleanups
5970 2002-11-18 Phil Schwan <phil@clusterfs.com>
5973 - fix null d_it dereference (346)
5974 - fix full OST/dbench hang (333)
5975 - fix permission problem with file removal (286)
5976 - fix removal of OSCs from LOV when they fail
5977 - fix NULL deref during bulk timeout (214)
5978 - fix problems related to multiple filesystems on one MDS (241)
5979 - fixed serious subtle metadata locking bugs
5980 - free locks on clients when inodes are removed due to memory
5982 - fix inode pointer in lock data (285)
5983 - partial support for multiple MDS on a single host (241)
5984 - data locks weren't cancelled at clear_inode time (290, 311)
5985 - intent locks could lead to unbounded lock growth (205)
5986 - added a maximum lock count, an LRU list, and a flusher
5987 - fix multiple rename (365)
5988 - properly abstracted the echo client
5989 - OSC locked 1 byte too many; fixed
5990 - rewrote brw callback code:
5991 - fixed recovery bugs related to LOVs (306)
5992 - fixed too-many-pages-in-one-write crash (191)
5993 - fixed (again) crash in sync_io_timeout (214)
5994 - probably fixed callback-related race (385)
5996 - Add capability to MDS protocol
5997 - LDLM cancellations and callbacks on different portals
5999 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
6002 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
6003 - defer open object destruction to close time (601981, 138)
6004 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
6005 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
6006 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
6007 - don't call obd_setattr in ll_file_release for destroyed objects
6008 * protocol change to lustre_msg: move |version| and add |flags|
6009 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
6010 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
6011 * added replay of create, unlink, link and rename operations during
6012 MDS failover; recovery should be much more robust now
6013 * remove failed OSCs from LOVs (only lov_create uses this so far)
6014 * the lustre-HOWTO was brought (more) up to date (582544)
6016 2002-10-23 Phil Schwan <phil@clusterfs.com>
6019 - in-use dentries weren't being reused properly (617851)
6020 - prevent multiple LDLM setup (599178)
6021 - fix LOV size calculations for truncate (617853)
6022 - fix client handling of MDS intent errors (POSIX)
6023 - fix permission bug in lovstripe.c test (624321)
6024 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6025 - truncate past end of file could corrupt data
6026 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6027 - a race in recovery could return ETIMEDOUT to apps (623947)
6028 - building outside the source directory was fixed
6029 * the lustre-HOWTO was brought (more) up to date (582544)
6030 * major progress was made on recovery functionality
6032 2002-10-10 Phil Schwan <phil@clusterfs.com>
6035 - recovery deadlock fix
6036 - rm -rf causes LBUG fix (617817)
6037 - file open by multiple tasks fix (618962)
6038 - directory permissions bugs (602707 and 620007)
6039 - journal_stop fixed with locking (611313)
6040 - O_APPEND failures resolved (618273, perhaps 614459)
6041 - lconf PATH fix (619770)
6042 - IA64 build fix (621450)
6043 - RPC buffer sizes scale with amount of memory
6045 2002-10-01 Phil Schwan <phil@clusterfs.com>
6048 - locks would be cancelled without throwing away data pages,
6049 resulting in inconsistent data (605627)
6050 - inode attributes were not always being refreshed (605627, 612449)
6051 - lconf now continues to cleanup after lctl reports an error
6052 - MDS now enforces user permissions (602707)
6053 - lprocfs cleanup fixed, but not yet enabled (614157)
6054 - fixed infinite server hang, should a client not respond to an AST
6055 - avoid going into recovery if user calls readlink() with a buffer
6056 that's too small (613941)
6057 - AST RPCs no longer require replies (614867) -- this may be changed
6058 - don't crash server if client sends an IOV that's too big (611336)
6059 - fixed lock conversion deadlock (611892)
6060 - fixed the following of symlinks (614622)
6061 * recovery: the server can remove locks from a client that dies, other
6062 clients can make progress
6063 * more extN patch fixes
6064 * compile-time configurable ptlrpc buffer allocations
6066 - collaborative read cache document
6067 - Lustre Lite Performance CDR document-in-progress
6069 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6072 - fix typo in patch-2.4.18
6074 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6077 - clear ptlrpc request each time in handle_incoming_request()
6078 - unlink of files now destroys the object on the OST
6080 2002-09-19 Peter Braam <braam@clusterfs.com>
6082 * add hard link support
6083 * change obdfile creation method
6084 * kernel patch changed
6086 2002-09-19 Peter Braam <braam@clusterfs.com>
6089 - stack overflow bug in extN fixed
6091 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6093 * documentation updates
6094 - add man pages for config tools
6095 - update tests/README to describe testing with new config tools
6096 - finish metadata API descriptions
6097 * bug fixes and cleanups
6098 - statfs workaround for 16TB limit
6099 - LOV stripe allocation improved, can stripe on subset of OSTs
6100 - LOV file size/IO offset was wrong for files > 4GB in size
6101 - object EA data was being dropped, caused files to be unreadable
6102 - memory overflow with non-LOV OST caused memory corruption
6103 - fixed regression tests to work with new config tools, obdfilter
6104 - fixed bug when directory size became larger than 1 block
6105 - fixed bug (for single client case) when PWD was deleted
6106 - invalidate local directory pages when doing intent-based ops
6107 - avoid LDLM oops when lock callback contained bad data
6109 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6111 * documentation updates
6112 * bug fixes and cleanups
6113 - configuration tools
6116 - 64-bit compile warnings
6117 - 64-bit internal statfs data
6119 * test_brw on persistent OST devices
6121 * lprocfs (disabled)
6123 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6125 * documentation updates
6126 * bug fixes and cleanups
6127 * configuration tools
6129 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6132 * many small fixes to 0_5_4
6133 * io/network handling
6134 * thinkos in MDS operations
6136 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6139 * crucial basic fixes to 0.5.3
6140 * IOR, Iozone work over Elan
6143 2002-08-07 Phil Schwan <phil@clusterfs.com>
6144 * version 0_5_3, our first alpha
6145 * we use the new Portals iovs
6146 * documentation updates
6147 * bug fixes and cleanups
6148 * small changes in the DLM wire protocol
6150 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6151 * version 0_5_1 with some initial stability,
6152 * locking on MD and file I/O.
6153 * documentation updates
6154 * several bug fixes since 0.5.0
6155 * small changes in wire protocol
6157 2002-07-18 Phil Schwan <phil@clusterfs.com>
6159 * delivered as Lustre Light Alpha
6160 * fixed a crash after handling invalid MDS requests
6161 * fixed directory pages for architectures with non-4k pages sizes
6163 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6164 * release version v0_4_4
6165 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6167 * Fixes a number of interruption problems with OST operations.
6168 * Update documentation for portals header changes
6169 * Move all wire protocol structs/defines to lustre_idl.h
6170 * Fixes symlink length bug.
6171 * Add tcpdump to repository.
6173 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6174 * release version v0_4_3
6175 * Fixes statfs for inodes on extN.
6176 * Fixes bug in runtests which would delete /etc/hosts.
6177 * Use 64-bit object IDs wherever possible (not into VFS though)
6178 Remove ost_get_info, which is unused by lustre, and out of date.
6180 2002-07-03 Peter Braam <braam@clusterfs.com>
6181 * release version v0_4_2 Fixes a lookup error (type not passed)
6182 * move forward to head of Portals
6183 * move forward to latest Lustre kernel
6185 2002-06-25 Peter Braam <braam@clusterfs.com>
6186 * release version v0_4_1. Hopefully stable on single node use.