1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.37 (SLES 10),
5 2.6.18-128.1.6.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org)
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.27 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.41.5.sun2
11 * File join has been disabled in this release, refer to Bugzilla 16929.
12 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
13 * NFS export disabled when stack size < 8192. Since the NFSv4 export
14 of Lustre filesystem with 4K stack may cause a stack overflow. For
15 more information, please refer to bugzilla 17630.
19 Description: abort bulk too early if client is reconnected
22 Frequency : rare, if used wide striped file and one ost in down.
24 Descriptoin: don't return error if we created a subset of objects for file.
25 Details : lov_update_create_set() uses set->set_success as index for created
26 objects, so if some requests failed, they will have hole at end of
27 array and we can use qos_shrink_lsm for allocate correct lsm.
29 Severity : enhancement
31 Description: implement async create (obd_async_create) method for osc, to avoid
32 too long waiting new ost objects with holding ldlm lock.
37 Description: fix lqs' reference which won't be put in some situations
38 Details : This patch fixes:
39 1. In quota_check_common(), this function will check quota
40 for user and group, but only send one return via "pending".
41 In most cases, the pendings should be same. But that is not
43 2. If quotaoff runs between lquota_chkquota() and
44 lquota_pending_commit(), the same thing will happen too.
46 - if (!ll_sb_any_quota_active(qctxt->lqc_sb))
49 ------------------------------------------------------------------------------
50 tbd Sun Microsystems, Inc.
52 * Support for kernels:
53 2.6.16.60-0.37 (SLES 10),
54 2.6.18-128.1.6.el5 (RHEL 5),
55 2.6.22.14 vanilla (kernel.org)
56 * Client support for unpatched kernels:
57 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
58 2.6.16 - 2.6.27 vanilla (kernel.org)
59 * Recommended e2fsprogs version: 1.41.5-sun2
60 * File join has been disabled in this release, refer to Bugzilla 16929.
61 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
62 * NFS export disabled when stack size < 8192. Since the NFSv4 export
63 of Lustre filesystem with 4K stack may cause a stack overflow. For
64 more information, please refer to bugzilla 17630.
68 Description: resolve race between obd_disconnect and class_disconnect_exports
69 Details : if obd_disconnect will be called to already disconnected export he
70 forget release one reference and osc module can't unloaded.
72 Severity : enhancement
74 Description: move AT tunable parameters for more consistent usage
75 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
77 Severity : enhancement
79 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
81 Severity : enhancement
83 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
85 Severity : enhancement
87 Description: Caching OSS
88 Details : introduce data caching on the OSS. The OSS now relies on the linux
89 kernel page cache to keep recently accessed data in memory.
90 It is worth noting that all write requests are still flushed
91 synchronously as in lustre 1.6.
93 Severity : enhancement
95 Description: version based recovery
96 Details : introduce finer grained recovery able to detect transaction
97 dependencies and can deal with transaction gaps.
99 Severity : enhancement
101 Description: Enable adaptive timeouts by default
105 Descriptoin: correctly skip time estimate if in recovery
106 Details : rq_send_state insn't bitmask so using bitwise ops is forbid.
110 Descriptoin: fix goal inodes
111 Details : Allocate inodes for llog in last inode group for avoid broken
116 Descriptoin: OSS DeadLock
117 Details : Use trylock to prevent deadlock when shrink icache.
119 Severity : enhancement
121 Description: Allow tuning service thread via /proc
122 Details : For each service a new
123 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
124 created that can be used to set min/max thread counts, and get the
125 current number of running threads.
127 Severity : enhancement
129 Description: Add state history info file, enhance import info file
130 Details : Track import connection state changes in a new osc/mdc proc file;
131 add overview-type data to the osc/mdc import proc file.
135 Description: Reduce small size read RPC
136 Details : Set read-ahead limite for every file and only do read-ahead when
137 available read-ahead pages are bigger than 1M to avoid small size
142 Description: free_entry erroneously used groups_free instead of put_group_info
144 Severity : enhancement
146 Description: Make read-ahead stripe size aligned.
148 Severity : enhancement
150 Description: MDS create should not wait for statfs RPC while holding DLM lock.
155 Descriptoin: OST grant too much space to client even there are not enough space.
156 Details : Client will shrink its grant cache to OST if there are no write
157 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
158 this grant cache if there are already not enough avaible space
159 (left_space < total_clients * 32M).
161 Severity : enhancement
163 Description: Update client support to vanilla kernels up to 2.6.27.
166 Frequency : rare, connect and disconnect target at same time
168 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
169 Details : Don't call obd_disconnect under lov_lock. this long time
170 operation and can block ptlrpcd which answer to connect request.
172 Severity : enhancement
174 Description: Update to sles9 kernel-2.6.5-7.315.
177 Frequency : start MDS on uncleanly shutdowned MDS device
179 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
180 Details : Stay in waiting mds<>ost recovery finished produce random bugs
181 due race between two ll_sync thread for one lov target. send
182 ACTIVATE event only if connect realy finished and import have
186 Frequency : start MDS on uncleanly shutdowned MDS device
188 Descriptoin: aborting recovery hang on MDS
189 Details : Don't throttle destroy RPCs for the MDT.
191 Severity : enhancement
193 Descriptoin: Don't sync journal after every i/o
194 Details : Implement write RPC replay to allow server replies for write RPCs
195 before data is on disk. However, this feature is disabled by
196 default since some issues leading to data corruptions have been
197 found during recovery (e.g. bug 19128). This feature can be
198 enabled by running the following command on the OSSs:
199 lctl set_param obdfilter.*.sync_journal=0
203 Description: Slow reads beyond 8Tb offsets.
204 Details : Page index integer overflow in ll_read_ahead_page
208 Description: Soft lockup on OSS after MDS failover
209 Details : MSG_CONNECT_INITIAL is not set on the initial MDS->OST connect.
210 As a conseqence, the patch from bug 18224 is not operational
211 and the MDS export cannot be reused on the OSTs until it gets
215 Frequency : rare, only if using MMP with Linux RAID
217 Description: MMP doesn't work with Linux RAID
218 Details : While using HA for Lustre servers with Linux RAID, it is possible
219 that MMP will not detect multiple mounts. To make this work we
220 need to unplug the device queue in RAID when the MMP block is being
221 written. Also while reading the MMP block, we should read it from
222 disk and not the cached one.
225 Frequency : rare, during recovery
227 Description: Assertion failure in ldlm_lock_put
228 Details : Do not put cancelled locks into replay list, hold references on
233 Description: 1.6.5 mdsrate performance is slower than 1.4.11 (MDS not cpu bound)
234 Details : create_count always drops to the min value (=32) because grow_count
235 is being changed before the precreate RPC completes.
238 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
241 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
242 Details : add proper locking for creation of jbd_4k slab cache
246 Description: MMP check in ext3_remount() fails without displaying any error
247 Details : When multiple mount protection fails during remount, proper error
250 -------------------------------------------------------------------------------
251 2008-12-31 Sun Microsystems, Inc.
253 * Support for kernels:
254 2.6.5-7.314 (SLES 9),
255 2.6.9-67.0.22.EL (RHEL 4),
256 2.6.16.60-0.31 (SLES 10),
257 2.6.18-92.1.17.el5 (RHEL 5),
258 2.6.22.14 vanilla (kernel.org)
259 * Client support for unpatched kernels:
260 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
261 2.6.16 - 2.6.22 vanilla (kernel.org)
262 * Client support for unpatched kernels:
263 we do not recommend using patchless RHEL4 clients with kernels
264 prior to 2.6.9-55EL (RHEL4U5).
265 * Recommended e2fsprogs version: 1.40.11-sun1
266 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
267 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
268 removed cwd "./" (refer to Bugzilla 14399).
269 * A new quota file format has been introduced in 1.6.5.
270 The format conversion from prior releases is handled transparently,
271 but releases older than 1.4.12/1.6.5 don't understand this new
272 format. The automatic format conversion can be avoided by running
273 the following command on the MDS:
274 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
275 For more information, please refer to bugzilla 13904.
276 * A new quota file format was introduced in 1.6.6/1.8.0 (kernel 2.6.16+)
277 The format conversion from prior releases is handled transparently,
278 but releases older than 1.6.6/1.8.0 don't understand this new
279 format. The automatic format conversion can be avoided by running
280 the following commands on the MDS and OSS servers (for
281 pre 1.4.12-1.6.5 quota files):
282 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV',
283 'tunefs.lustre --param="ost.quota_type=ug1" $MDTDEV'
284 or (for 1.4.12/1.6.5 quota files)
285 'tunefs.lustre --param="mdt.quota_type=ug2" $MDTDEV',
286 'tunefs.lustre --param="ost.quota_type=ug2" $MDTDEV'
287 For more information, please refer to bugzilla 13904.
288 * Output of lfs quota has been made less detailed by default,
289 old (verbose) output can be obtained by using -v option.
290 * File join has been disabled in this release, refer to Bugzilla 16929.
291 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
292 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
293 Lustre filesystem with 4K stack may cause a stack overflow. For more
294 information, please refer to bugzilla 17630.
296 Severity : enhancement
298 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
299 Details : Allow skip disconnected ost for send statfs request and hide error
303 Frequency : rare, on llog test 6
305 Descriptoin: don't allow connect to already connected import
306 Details : Connect to already connected import hides connection problem.
309 Frequency : rare, on failed llog setup
311 Descriptoin: don't leak obd reference on failed llog setup
312 Details : For failed llog setup - mgc forget call class_destroy_import for
313 client import, move destroy import to more generic place.
318 Descriptoin: allow kill process which wait statahead result
319 Details : For some reasons 'ls' can stick in waiting result from statahead,
320 in this case need way for kill this process.
323 Frequency : rare, at shutdown
325 Descriptoin: panic at umount
326 Details : llap_shrinker can be raced with killing super block from list and
327 this produce panic with access to already freeded pointer
332 Descriptoin: don't lose wakeup for imp_recovery_waitq
333 Details : Recover_import_no_retry or invalidate_import and import_close can
334 both sleep on imp_recovery_waitq, but we was send only one wakeup
340 Descriptoin: panic in mds_open
341 Details : don't confuse mds_finish_transno() with PTR_ERR(-ENOENT)
346 Descriptoin: stuck in cache_remove_extent() or panic with accessing to already
348 Details : Release lock refernce only after add page to pages list.
351 Frequency : always with long access acl
353 Descriptoin: mds can't pack reply with long acl.
354 Details : mds don't control size of acl but they limited by reint/getattr
359 Frequency : on remount
361 Description: external journal device not working after the remount
362 Details : clear dev_rdonly flag for external journal devices in
368 Description: shutdown vs evict race
369 Details : client_disconnect_export vs connect request race.
370 if client will evicted at this time - we start invalidate
371 thread without referece to import and import can be freed
377 Description: shrink LOV EAs before replying
378 Details : correctly adjust LOV EA buffer for reply.
383 Description: don't skip ost target if they assigned to file
384 Details : Drop slow OSCs if we can, but not for requested start idx.
385 This means "if OSC is slow and it is not the requested
386 start OST, then it can be skipped, otherwise skip it only
387 if it is inactive/recovering/out-of-space.
389 Severity : enhancement
391 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
394 Frequency : rare, need acl's on inode.
396 Description: client can't handle ost additional correctly
397 Details : if ost was added after client connected to mds client can have
398 hit lnet_try_match_md ... to big messages to wide striped files.
399 in this case need teach client to handle config events about add
400 lov target and update client max ea size at that event.
402 Severity : enhancement
404 Description: Update to sles9 kernel-2.6.5-7.314.
406 Severity : enhancement
408 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
411 Frequency : Create a symlink file with a very long name
413 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
414 Details : If there is no extra space in the request for early cancels,
415 ldlm_req_handles_avail() returns 0 instead of a negative value.
420 Description: mds is deadlocked
421 Details : In rare cases, inode in catalog can have i_no less than have parent
422 i_no, this produce wrong order for locking during open, and
423 parallel unlink can be lock open. this need teach mds_open to grab
424 locks in resource id order, not at parent -> child order.
426 Severity : enhancement
428 Description: Add /proc entry for import status
429 Details : The mdc, osc, and mgc import directories now have
430 an import directory that contains useful import data for debugging
433 Severity : enhancement
435 Description: Re-disable certain /proc logging
436 Details : Enable and disable client's offset_stats, extents_stats and
437 extents_stats_per_process stats logging on the fly.
440 Frequency : Only on FC kernels 2.6.22+
442 Description: oops in statahead
443 Details : Do not drop reference count for the dentry from VFS when lookup,
444 VFS will do that by itself.
446 Severity : enhancement
448 Description: Generic /proc file permissions
449 Details : Set /Proc file permissions in a more generic way to enable non-
450 root users operate on some /proc files.
454 Description: Hitting mdc_commit_close() ASSERTION
455 Details : Properly handle request reference release in
456 ll_release_openhandle().
460 Frequency : only patchless client
461 Description: add workaround for race between add/remove dentry from hash
463 Severity : enhancement
465 Description: Allow OST glimpses to return PW locks
469 Description: LBUG when llog conf file is full
470 Details : When llog bitmap is full, ENOSPC should be returned for plain
475 Description: Prevent import from entering FULL state when server in recovery
479 Description: service mount cannot take device name with ":"
480 Details : Only when device name contains ":/" will mount treat it as
486 Description: Replace ptlrpcd with the statahead thread to interpret the async
487 statahead RPC callback
491 Frequency : on recovery
492 Description: I/O failures after umount during fail back
493 Details : If client reconnected to restarted server we need join to recovery
494 instead of find server handler is changed and process self eviction
495 with cancel all locks.
497 Severity : enhancement
499 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
503 Description: Kernel BUG tries to release flock
504 Details : Lustre does not destroy flock lock before last reference goes
505 away. So always drop flock locks when client is evicted and
506 perform unlock regardless of successfulness of speaking to MDS.
508 Severity : enhancement
510 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
512 Severity : enhancement
514 Description: Upcall on Lustre log has been dumped
515 Details : Allow for a user mode script to be called once a Lustre log has
516 been dumped. It passes the filename of the dumped log to the
517 script, the location of the script can be specified via
518 /proc/sys/lnet/debug_log_upcall.
523 Description: avoid messages about idr_remove called for id that is not allocated
524 Details : Move assigment s_dev for clustered nfs to end of initialization,
525 for avoid problem with error handling.
530 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
531 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
532 CONN_USED_HASH and this prodice warning when put connection again
538 Description: avoid ASSERTION(client_stat->nid_exp_ref_count == 0) failed
539 Details : Release reference to stats when client disconnected, not
540 when export destroyed for avoid races when client destroyed
541 after main ost export.
545 Description: more cleanup in mds_lov
546 Details : Add workaround for get valid ost count for avoid warnings about
547 drop too big messages, not init llog cat under semphore which
548 can be blocked on reconnect and break normal replay, fix access
551 Severity : enhancement
553 Description: File striping can now be set to use an arbitrary pool of OSTs.
555 Severity : enhancement
557 Description: Export bytes_read/bytes_write count on OSC/OST.
561 Description: Early reply size mismatch, MGC loses connection
562 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
563 the connect flags are properly negotiated.
567 Description: Properly propagate oinfo flags from lov to osc for statfs
568 Details : Restore missing copy oi_flags to lov requests.
572 Description: exports in /proc are broken
573 Details : recreate /proc entries for clients when they reconnect.
575 Severity : enhancement
577 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
578 : l_getgroups(8), lst(8), routerstat(8)
579 Details : Included man pages for llobdstat(8), llstat(8),
580 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
582 Severity : enhancement
584 Description: Implement lustre ll_show_options method.
586 Severity : enhancement
588 Description: Update to SLES9 kernel-2.6.5-7.312.
590 Severity : enhancement
592 Description: Update to RHEL4 kernel-2.6.9-67.0.22.EL.
596 Description: exports in /proc are broken
597 Details : Recreate /proc entries for clients when they reconnect.
601 Description: don't fail open with -ERANGE
602 Details : If client connected until mds will be know about real ost count
603 get LOV EA can be fail because mds not allocate enougth buffer
608 Description: Resolve device initialization race
609 Details : Prevent proc handler from accessing devices added to the
610 obd_devs array but yet be intialized.
612 Severity : enhancement
614 Description: configure's --enable-quota should check the
615 kernel .config for CONFIG_QUOTA
616 Details : Configure is terminated if --enable-quota is passed but
617 no quota support is in kernel
619 Severity : enhancement
621 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
623 Severity : enhancement
625 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
629 Frequency : rare, on PPC clients
630 Description: don't swab ost objects in response about directory, because
632 Details : Bug similar bug 14856, but in different function.
634 Severity : enhancement
636 Description: lfs quota tool enhancement
637 Details : Added units specifiers support for setquota, default to
638 current uid/gid for quota report, short quota stats by
639 default, nonpositional parameters for setquota, added
640 llapi_quotactl manual page.
642 Severity : enhancement
644 Description: *optional* service tags registration
645 Details : If the "service tags" package is installed on a Lustre node
646 When the filesystem is mounted, a local-node service tag will
647 be created. See http://inventory.sun.com/ for more information
648 about the Service Tags asset management system.
652 Description: Client runs out of low memory
653 Details : Consider only lowmem when counting initial number of llap pages
656 Description: Kernel BUG tries to release flock
657 Details : Lustre does not destroy flock lock before last reference goes
658 away. So always drop flock locks when client is evicted and
659 perform unlock regardless of successfulness of speaking to MDS.
662 Frequency : occasional
664 Description: add refcount for osc callbacks, so avoid panic on shutdown
666 Severity : enhancement
668 Description: Update to RHEL4 kernel-2.6.9-67.0.20.
671 Frequency : testing only
673 Description: sanity test 65a fails if stripecount of -1 is set
674 Details : Handle -1 striping on filesystem in ll_dirstripe_verify
677 Frequency : only in unusual configurations
679 Description: Kernel panic with find ost index.
680 Details : lov_obd have panic if some OST's have sparse indexes.
682 Severity : enhancement
684 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
687 Frequency : rarely, if filesystem is mounted with -o flock
689 Description: do not process already freed flock
690 Details : Flock can possibly be freed by another thread before it reaches
691 to ldlm_flock_completion_ast.
694 Frequency : rarely, if filesystem is mounted with -o flock
696 Description: LBUG during stress test
697 Details : Need properly lock accesses the flock deadlock detection list.
700 Frequency : rarely, if binaries are being run from Lustre
702 Description: oops in page fault handler
703 Details : Kernel page fault handler can return two special 'pages' in
704 error case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
707 Frequency : rarely, during shutdown
709 Description: timeout with invalidate import.
710 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
711 should be handled by ptlrpcd. This produce long age waiting and
712 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
718 Description: ASSERTION(CheckWriteback(page,cmd)) failed
719 Details : Badly clear PG_Writeback bit in ll_ap_completion can produce false
723 Frequency : only with broken builds/installations
725 Description: no LBUG if lquota.ko and fsfilt_ldiskfs.ko are different versions
726 Details : Just return an error to a user, put a console error message.
728 Severity : enhancement
730 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
732 Severity : enhancement
734 Description: Update to RHEL4 kernel-2.6.9-67.0.15.
736 Severity : enhancement
738 Description: enable MGS and MDT services start separately
739 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
740 a co-located MGS without starting the MGS, which is a complement
741 to 'nosvc' mount option.
744 Frequency : always, on big-endian systems
746 Description: cleanup in ptlrpc code, related to PPC platform
747 Details : store magic in native order avoid panic's in recovery on PPC
748 node and forbid from this error in future. Also fix posibily
749 of twice swab data. Fix get lov striping to userpace.
752 Frequency : rarely, if replay get lost on server
754 Description: server incorrectly drop resent replays lead to recovery failure.
755 Details : do not drop replay according to msg flags, instead we check the
756 per-export recovery request queue for duplication of transno.
759 Frequency : after recovery
761 Description: precreate to many object's after del orphan.
762 Details : del orphan st in oscc last_id == next_id and this triger growing
763 count of precreated objects. Set flag LOW to skip increase count
764 of precreated objects.
767 Frequency : rare, on clear nid stats
769 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
770 Details : when clean nid stats sometimes try destroy live entry,
771 and this produce panic in free.
774 Frequency : occasionally since 1.6.4
776 Description: Stack overflow during MDS log replay
777 Details : ease stack pressure by using a thread dealing llog_process.
780 Frequency : very rare
782 Description: MDT cannot be unmounted, reporting "Mount still busy"
783 Details : Mountpoint references were being leaked during open reply
784 reconstruction after an MDS restart. Drop mountpoint reference
785 in reconstruct_open() and free dentry reference also.
790 Description: wait until IO finished before start new when do lock cancel.
791 Details : VM protocol want old IO finished before start new, in this case
792 need wait until PG_writeback is cleared until check dirty flag
793 and call writepages in lock cancel callback.
798 Description: mds_mfd_close() ASSERTION(rc == 0)
799 Details : In mds_mfd_close(), we need protect inode's writecount change
800 within its orphan write semaphore to prevent possible races.
803 Frequency : rare, on shutdown ost
805 Description: don't hit live lock with umount ost.
806 Details : shrink_dcache_parent can be in long loop with destroy dentries,
807 use shrink_dcache_sb instead.
810 Frequency : only when echo_client is used
812 Description: don't panic with use echo_client
813 Details : echo client pass NULL as client nid pointer and this produce NULL
817 Frequency : Always on 32-bit PowerPC systems
819 Description: fix build on PPC32
820 Details : compile code with -m64 flag produce wrong object file for PPC32.
825 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
826 Details : In reconstruct_* functions, LASSERTs on both the data supplied
827 by a client, and the data on disk are dangerous and incorrect.
828 Change them with client eviction.
830 Severity : enhancement
832 Description: skiplist implementation simplification
833 Details : skiplists are used to group compatible locks on granted list
834 that was implemented as tracking first and last lock of each lock
835 group the patch changes that to using doubly linked lists
839 Description: delete compatibility for 32bit qdata
840 Details : as planned, when lustre is beyond b1_8, lquota won't support 32bit
841 qunit. That means servers of b1_4 and servers of b1_8 can't be
842 used together if users want to use quota.
845 Frequency : only with administrator action
847 Description: mount failure if config log has invalid conf_param setting
848 Details : If administrator specified an incorrect configuration parameter
849 with "lctl conf_param" this would cause an error during future
850 client mounts. Instead, ignore the bad configuration parameter.
853 Frequency : blocks per group < blocksize*8 and uninit_groups is enabled
855 Description: ldiskfs error: XXX blocks in bitmap, YYY in gd
856 Details : If blocks per group is less than blocksize*8, set rest of the
860 Frequency : Application do stride read on lustre
862 Description: The read performance will drop a lot if the application does
864 Details : Because the stride_start_offset are missing in stride read-ahead,
865 it will cause clients read a lot of unused pages in read-ahead,
866 then the read-performance drops.
870 Description: more ldlm soft lockups
871 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
872 starve other threads from the resource lock for a long time in
873 case of long waiting queue, so change the debug level from
874 D_OTHER to the less frequently used D_INFO.
876 Severity : enhancement
878 Description: add -gid, -group, -uid, -user options to lfs find
880 Severity : enhancement
882 Description: ll_recover_lost_found_objs - recover objects in lost+found
883 Details : OST corruption and subsequent e2fsck can leave objects in the
884 lost+found directory. Using the "ll_recover_lost_found_objs"
885 tool, these objects can be retrieved and data can be salvaged
886 by using the object ID saved in the fid EA on each object.
891 Description: this bug _only_ happens when inode quota limitation is very low
892 (less than 12), so that inode quota unit is 1 at initialization.
893 Details : if remaining quota equates 1, it is a sign to demonstate that quota
894 is effective now. So least quota qunit should be 2.
898 Description: Hung threads in invalidate_inode_pages2_range
899 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
900 one is completed. As a result, some RPCs are stuck in the queue
905 Description: Procfs and llog threads access destoryed import sometimes.
906 Details : Sync the import destoryed process with procfs and llog threads by
907 the import refcount and semaphore.
911 Description: mds fails to respond, threads stuck in ldlm_completion_ast
912 Details : Sort source/child resource pair after updating child resource.
917 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
918 Details : If insertion of an extent fails, then discard the inode
919 preallocation and free data blocks else it can lead to duplicate
924 Description: don't always update ctime in ext3_xattr_set_handle()
925 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle()
926 In some cases the ctime should not be updated, for example for
927 2.0->1.8 compatibility it is necessary to delete an xattr and it
928 should not update the ctime.
932 Description: add quota statistics
933 Details : 1. sort out quota proc entries and proc code.
934 2. add quota statistics
939 Description: quotas are not honored with O_DIRECT
940 Details : all writes with the flag O_DIRECT will use grants which leads to
941 this problem. Now using OBD_BRW_SYNC to guard this.
945 Bugzilla : 15713/16362
946 Description: Assertion in iopen_connect_dentry in 1.6.3
947 Details : looking up an inode via iopen with the wrong generation number can
948 populate the dcache with a disconneced dentry while the inode
949 number is in the process of being reallocated. This causes an
950 assertion failure in iopen since the inode's dentry list contains
951 both a connected and disconnected dentry.
955 Description: assertion failure in ldlm_handle2lock()
956 Details : fix a race between class_handle_unhash() and class_handle2object()
957 introduced in lustre 1.6.5 by bug 13622.
959 Severity : enhancement
961 Description: superblock lock contention with many SMP cores on one client
962 Details : several client filesystem locks were highly contended on SMP
963 NUMA systems with 8 or more cores. Per-CPU datastructures
964 and more efficient locking implemented to reduce contention.
969 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
970 Details : remove the limit of 256 scsi disks in the sd_iostat patch
975 Description: oops in sd_iostats_seq_show()
976 Details : unloading/reloading the scsi low level driver triggers a kernel
977 bug when trying to access the sd iostat file.
982 Description: Kernel panics during QLogic driver reload
983 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
984 patch, causing memory corruption.
989 Description: journal_dev option does not work in b1_6
990 Details : pass mount option during pre-mount.
992 Severity : enhancement
994 Description: Add a FIEMAP(FIle Extent MAP) ioctl for ldiskfs
995 Details : FIEMAP ioctl will allow an application to efficiently fetch the
996 extent information of a file. It can be used to map logical blocks
997 in a file to physical blocks in the block device.
1000 Frequency : only with adaptive timeout enabled
1002 Description: DEBUG_REQ() bad paging request
1003 Details : ptlrpc_at_recv_early_reply() should not modify req->rq_repmsg
1004 because it can be accessed by reply_in_callback() without the
1008 Frequency : only on Cray X2
1010 Description: X2 build failures
1011 Details : fix build failures on Cray X2.
1015 Description: xid & resent requests
1016 Details : Initialize RPC XID from clock at startup (randomly if clock is
1021 Description: quota recovery deadlock during mds failover
1022 Details : This patch includes att18982, att18236, att18237 in bz14840.
1024 1. fix osts hang when mds does failover with quotaon
1025 2. prevent watchdog storm when osts threads wait for the
1030 Description: kernel panic on racer
1031 Details : Do not access dchild->d_inode when IS_ERR(dchild) is true.
1033 Severity : enhancement
1035 Description: Add lustre_start utility to start or stop multiple Lustre servers
1040 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1041 Details : In case of memory pressure, list_del() can be called twice on
1042 req->rq_history_list, causing a kernel oops.
1046 Description: (ptllnd_peer.c:557:kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1047 Details : fix stack overflow in the distributed lock manager by defering
1048 export eviction after a failed ast to the elt thread instead of
1049 handling it in the dlm interpret routine.
1051 Severity : enhancement
1053 Description: More exported tunables for mballoc
1054 Details : Add support for tunable preallocation window and new tunables for
1055 large/small requests
1059 Description: Detect corruption of block bitmap and checking for preallocations
1060 Details : Checks validity of on-disk block bitmap. Also it does better
1061 checking of number of applied preallocations. When corruption is
1062 found, it turns filesystem readonly to prevent further corruptions.
1066 Frequency : only for big-endian servers
1067 Description: Check for big-endian system while mounting fs with extents feature
1068 Details : Mounting a filesystem with extents feature will fail on big-endian
1069 systems since ext3-based ldiskfs is not supported on big-endian
1070 systems. Can be over-ridden with "bigendian_extents" mount option.
1074 Description: Excessive recovery window
1075 Details : With AT enabled, the recovery window can be excessively long (6000+
1076 seconds). To address this problem, we no longer use
1077 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
1078 timeout no longer depends on the service time, it is set to
1079 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
1080 time via pb_service_time.
1084 Description: Watchdog triggered on MDS failover
1085 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
1086 the OSTs know that the MDS "UUID" can be reused for the same export
1087 from a different NID, so we do not need to wait for the export to
1092 Descriptoin: Lustre detected file system corruption with inode out of bounds
1093 Details : don't update i_size on MDS_CLOSE for directories. This causes
1094 directory corruptions on the MDT.
1096 -------------------------------------------------------------------------------
1098 2008-05-26 Sun Microsystems, Inc.
1100 * Support for kernels:
1101 2.6.5-7.311 (SLES 9),
1102 2.6.9-67.0.7.EL (RHEL 4),
1103 2.6.16.54-0.2.5 (SLES 10),
1104 2.6.18-53.1.14.el5 (RHEL 5),
1105 2.6.22.14 vanilla (kernel.org)
1106 * Client support for unpatched kernels:
1107 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1108 2.6.16 - 2.6.22 vanilla (kernel.org)
1109 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
1110 we do not recommend using patchless RHEL4 clients with kernels
1111 prior to 2.6.9-55EL (RHEL4U5).
1112 * Recommended e2fsprogs version: 1.40.7-sun1
1113 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1114 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
1115 removed cwd "./" (refer to Bugzilla 14399).
1116 * A new quota file format has been introduced in 1.6.5.
1117 The format conversion from prior releases is handled transparently,
1118 but releases older than 1.4.12/1.6.5 will not understand this new
1119 format. The automatic format conversion can be avoided by running
1120 the following command on the MDS before upgrading:
1121 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
1122 For more information, please refer to bugzilla 13904.
1126 Description: quota performance fix
1127 Details : quota data is written in journalled mode instead of ordered to
1128 increase performance
1132 Description: lfs support for human-readable quota grace time strings
1133 Details : lfs setquota -t and lfs quota -t represent quota grace times
1134 in "XXwXXdXXhXXmXXs" format instead of large values in seconds
1137 Frequency : always with o2ib 1.3 and sles10
1139 Description: fix build with SLES10 and o2ib v3.
1140 Details : sles10 uses diffrent name for Module.symver file but configure
1141 assume this file has same name on RHEL/SLES/vanila kernels.
1144 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1146 Description: MDS may lose file striping (and hence file data) in some cases
1147 Details : If there are additional extended attributes stored on the MDS,
1148 in particular ACLs, SELinux, or user attributes (if user_xattr
1149 is specified for the client mount options) then there is a risk
1150 of attribute loss. Additionally, the Lustre file striping
1151 needs to be larger than default (e.g. striped over all OSTs),
1152 and an additional attribute must be stored initially in the
1153 inode and then increase in size enough to be moved to the
1154 external attribute block (e.g. ACL growing in size) for file
1157 Severity : enhancement
1159 Description: add message levels for liblustreapi
1164 Description: MDT cannot be unmounted, reporting "Mount still busy"
1165 Details : Mountpoint references were being leaked during open reply
1166 reconstruction after an MDS restart. Drop mountpoint reference
1167 in reconstruct_open() and free dentry reference also.
1172 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
1173 Details : Move the 'good_osts' check before the 'total_bavail' check. This
1174 will result in an -EAGAIN and in the exit call path we call
1175 alloc_rr() which will with increasing aggressiveness attempt to
1176 aquire precreated objects on the minimum number of required OSCs.
1180 Description: Use old size assignment to avoid deadlock
1181 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
1182 the scheduling while holding a spinlock. We do not need locking
1183 for size in ll_update_inode() because size is only updated from
1184 the MDS for directories or files without objects, so there is no
1185 other place to do the update, and concurrent access to such inodes
1186 are protected by the inode lock.
1190 Description: Use __u64 instead of int for valid bits
1194 Description: resolve "_IOWR redefined" build error on SLES10
1198 Description: dump the memory debugging after all modules are unloaded to
1199 suppress false negative in conf_sanity test 39
1203 Description: the recovery timer never expires
1204 Details : for new client connect request, the recovery timer should not be
1205 reset, otherwise recovery timer will never expired, if the old
1206 client never come. Only old client connect and first connection
1207 req should trigger recovery timer reset.
1211 Description: the min numbers of lproc stats are wrong
1212 Details : adding a new constant LC_MIN_INIT and use it for initialization
1216 Frequency : always with interactive lfs
1218 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1221 Frequency : with multiple concurrent readdir processes in same directory
1222 Bugzilla : 15406, 15169, 15175
1223 Description: misc fixes for directory readahead.
1224 Details : prevent previous statahead async RPC callback from processing the
1225 current "statahead_info", race condition between async RPC callback
1226 add dentry into dentry hash table and "ls" thread revalidate such
1227 dentry, statahead his/miss control for hidden items, and so on.
1229 Severity : enhancement
1231 Description: build kernel-ib packages for OFED 1.3 in our release cycle
1235 Description: incore types cleaning in quota code (with respect to 64-bit limits)
1236 Details : several u32 variables declarations are replaced with u64 declarations
1241 Description: fix SLES kernel versioning
1242 Details : the kernel version for our SLES 10 kernel did not include a "-"
1243 before the "smp" at the end. while this was not a problem in
1244 general, it did mean that software trying to use the kernel
1245 version to try to detect a vendor specific kernel would fail.
1246 this was most evident by the OFED build scripts.
1251 Description: Don't update lov_desc members until making sure they are valid
1252 Details : When updating lov_desc members via proc fs, need fix their
1253 validities before doing the real update.
1256 Frequency : very rare
1258 Description: don't put request into delay list while invalidate in flight.
1259 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
1260 import in flight. this produce timeout for invalidate and sometimes
1261 can cause stale data.
1263 Severity : enhancement
1265 Description: Update kernel to SLES9 2.6.5-7.311.
1267 Severity : enhancement
1269 Description: Update kernel to RHEL4 2.6.9-67.0.7.
1274 Frequency : on PPC only
1275 Description: not convert ost objects for directory because it's not exist.
1276 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
1278 Severity : enhancement
1280 Description: Fix warnings with compile liblustre at sles10/rhel5 which have
1281 __u64 as usingied long long type.
1284 Frequency : rare, on shutdown
1286 Description: race process ast vs remove callback
1287 Details : removing callback before disconnect import open race with
1288 processing callback.
1290 Severity : enhancement
1292 Description: Update kernel to SLES9 2.6.5-7.311.
1294 Severity : enhancement
1296 Description: Files open for execute are not marked busy on SLES10
1297 Details : Add FMODE_EXEC to SLES10 SP1 server kernel series.
1299 Severity : enhancement
1301 Description: Add server support for vanilla-2.6.22.14.
1304 Frequency : occasional
1306 Description: Avoid lov_create() getting stuck in obd_statfs_rqset()
1307 Details : If an OST is down the MDS will hang indefinitely in
1308 obd_statfs_rqset() waiting for the statfs data. While for
1309 MDS QOS usage of statfs, it should not stuck in waiting.
1311 Severity : enhancement
1313 Description: Disable adaptive timeouts by default
1316 Frequency : on network error
1318 Description: panic with double free request if network error
1319 Details : mdc_finish_enqueue is finish request if any network error occuring,
1320 but it's true only for synchronus enqueue, for async enqueue
1321 (via ptlrpcd) this incorrect and ptlrpcd want finish request
1325 Frequency : rare, on recovery
1327 Description: read procfs can produce deadlock in some situation
1328 Details : Holding lprocfs lock with send rpc can produce block for destroy
1329 obd objects and this also block reconnect with -EALREADY.
1330 This isn't fix all lprocfs bugs - but make it rare.
1332 Severity : enhancement
1334 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
1337 Frequency : frequent on X2 node
1339 Description: mdc_set_open_replay_data LBUG
1340 Details : Set replay data for requests that are eligible for replay.
1345 Description: lustre_mgs: operation 101 on unconnected MGS
1346 Details : When MGC is disconnected from MGS long enough, MGS will evict the
1347 MGC, and late on MGC cannot successfully connect to MGS and a lot
1348 of the error messages complaining that MGS is not connected.
1351 Frequency : rare, depends on device drivers and load
1353 Description: MDS or OSS nodes crash due to stack overflow
1354 Details : Code changes in 1.6.4 increased the stack usage of some functions.
1355 In some cases, in conjunction with device drivers that use a lot
1356 of stack, the MDS (or possibly OSS) service threads could overflow
1357 the stack. One change which was identified to consume additional
1358 stack has been reworked to avoid the extra stack usage.
1360 Severity : enhancement
1362 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1364 Severity : enhancement
1366 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1368 Severity : enhancement
1370 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1372 Serverity : enhancement
1374 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1377 Frequency : rare on shutdown OST
1379 Description: Don't allow skipping OSTs if index has been specified.
1380 Details : Don't allow skipping OSTs if index has been specified, make locking
1381 in internal create lots better.
1386 Description: ASSERTION(!PageDirty(page)) failed
1387 Details : Wrong check could lead to an assertion failure under specific
1393 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1394 Details : access to bitfield in structure is always rounded to long
1395 and this produce problem with not atomic change any bit.
1400 Description: Lustre make rpms failed.
1401 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1402 builds Lustre rpms from tarball.
1404 Severity : enhancement
1406 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
1409 Frequency : rare on shutdown OST
1411 Description: If llog cancel was not send before clean_exports phase, this can
1412 produce deadlock in llog code.
1413 Details : If llog thread has last reference to obd and call class_import_put
1414 this produce deadlock because llog_cleanup_commit_master wait when
1415 last llog_commit_thread exited, but this never success because was
1416 called from llog_commit_thread.
1419 Frequency : only if OST index is skipped
1421 Description: NULL lov_tgts causing MDS oops
1422 Details : more safe checks for NULL lov_tgts for avoid oops.
1424 Severity : enhancement
1426 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1428 Severity : enhancement
1430 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1435 Description: make mgs_setparam() handle fsname containing dash
1436 Details : fsname containing a dash does not work with lctl conf_param
1438 Severity : enhancement
1440 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
1443 Frequency : rare, in recovery and (or) destroy lovobjid file.
1445 Description: rewrite lov objid code.
1446 Details : Cleanup for lov objid code, remove scability problems and wrong
1447 locking. Fix sending last_id into ost.
1449 Severity : enhancement
1451 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1453 Severity : enhancement
1455 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1456 Details : Use d_move_locked instead of __d_move.
1459 Frequency : rare, at shutdown
1461 Description: access already free / zero obd_namespace.
1462 Details : if client_disconnect_export was called without force flag set,
1463 and exist connect request in flight, this can produce access to
1464 NULL pointer (or already free pointer) when connect_interpret
1465 store ocd flags in obd_namespace.
1468 Frequency : only at startup
1470 Description: not alloc memory with spinlock held.
1471 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1472 if any spinlock held.
1477 Description: lfs find does not continue on file error
1478 Details : Continue other files processing when a file/dir is absent.
1482 Description: Inconsistent usage of lustre_pack_reply()
1483 Details : Standardize the usage of lustre_pack_reply() such that it
1484 always generate a CERROR on failure.
1487 Frequency : very rare
1489 Description: Fix replay if there is an un-replied request and open
1490 Details : In some cases, older replay request will revert the
1491 mcd->mcd_last_xid on MDS which is used to record the client's
1492 latest sent request.
1494 Severity : enhancement
1496 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1498 Severity : enhancement
1500 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1502 Severity : enhancement
1504 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1506 Severity : enhancement
1508 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
1510 Severity : enhancement
1512 Description: Build SLES10 patchless client fails
1513 Details : The configure was broken by run ./configure with
1514 --with-linux-obj=.... argument for patchless client. When the
1515 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1516 can't search header adequately. Use absolute path such as
1517 -I($LINUX)/include instead.
1519 Severity : enhancement
1521 Description: Lustre Page Accounting
1522 Details : New macros for page alloc and free which enable accounting
1523 of page allocation of Lustre. Use percpu counters to store memory
1524 and page statistics.
1527 Frequency : only if debugging is disabled
1529 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1530 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1531 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1532 these macros and replace them with their definitions instead.
1537 Description: interrupt oig_wait produce painc on resend.
1538 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1539 private set, and this produce situation when rq_ptlrpcd_data not
1540 copyed to new allocated request and triggered LBUG on assert
1541 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1542 ptlrpcd set if request is changed and can be safety changed to use
1545 Severity : enhancement
1547 Description: organize the server-side client stats on per-nid basis
1548 Details : Change the structure of stats under obdfilter and mds to
1556 The "uuid"s file would list the uuids of _active_ exports.
1557 And the clear entry is to clear all stats and stale nids.
1562 Description: Processes looping in ll_readdir() on Lustre clients finally causing
1563 a full node pseudo-hang
1564 Details : Concurrent access to the same directory from multiple clients with
1565 intensive file creation/removal can cause a client node to spin in
1566 ll_readdir(). i_version must be increased every time the lock is
1567 cancelled to ensure a revalidate is done.
1572 Description: touch file failed when fs is not full
1573 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1574 otherwise we can get ENOSP while fs is not full.
1577 Frequency : only for Cray XT3
1578 Bugzilla : 12829/13455
1579 Description: Changing primary group doesn't change the group lustre assigns to
1581 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1582 overridden with the primary group provided by the group upcall,
1583 whereas the supplied fsgid can be trusted if it is in the list of
1584 supplementary groups returned by the group upcall.
1586 Severity : enhancement
1588 Description: Root Squash Functionality
1589 Details : Implementation of NFS-like root squash capability. Specifically,
1590 don't allow someone with root access on a client node to be able
1591 to manipulate files owned by root on a server node.
1593 Severity : enhancement
1595 Description: Slow trucate/writes to huge files at high offsets.
1596 Details : Directly associate cached pages to lock that protect those pages,
1597 this allows us to quickly find what pages to write and remove
1598 once lock callback is received.
1603 Description: Too many locks accumulating on client during NFS usage
1604 Details : mds_open improperly used accmode to find out access mode to a
1605 file. Also mdc_intent_lock logic to find out if we already have
1606 lock similar to just received was flawed since introduction of
1607 skiplists - locks are now added to the front of the granted
1612 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
1614 Details : In obd setup failure handler, obd_stopping will not necessarily to
1615 be 1, and obd_set_up should also be checked to make sure whether
1616 obd is completely setup.
1618 Severity : enhancement
1620 Description: Allow masking D_WARNING, D_ERROR messages from console
1621 Details : Console messages can now be disabled via lnet.printk.
1626 Description: User code with malformed file open parameter crashes client node
1627 Details : Before packing join_file req, all the related reference should be
1628 checked carefully in case some malformed flags cause fake join_file
1634 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
1635 grain of quota doing harm to quota's accuracy
1636 Details : qunit size will be changed when quota limitation is too low/high;
1637 record the pending quota write in order to get more accureate
1638 quota; delete the patch for bug12588, which is unnecessary when
1639 this patch is landed. This bug also contains fixes for bug 14526,
1640 14299, 14601 and 13794.
1644 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1645 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1646 mode is equal to the granted mode. However, it should also check
1647 that LDLM_FL_CP_REQD is not set.
1651 Description: Heavy nfs access might result in deadlocks
1652 Details : After ELC code landed, it is now improper to enqueue any mds
1653 locks under och_sem, because enqueue might want to decide to
1654 cancel open locks for same inode we are holding och_sem for.
1658 Description: 35% write performance drop with ldiskfs2 when quotas are on
1659 Details : Enable ext3 journalled quota by default to improve performance
1660 when quotas are turned on.
1664 Description: Client eviction while running blogbench
1665 Details : A lot of unlink operations with concurrent I/O can lead to a
1666 deadlock causing evictions. To address the problem, the number of
1667 oustanding OST_DESTROY requests is now throttled to
1668 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1669 callbacks are processed in priority.
1672 Frequency : RHEL4 only
1674 Description: mkfs is very slow on IA64/RHEL4
1675 Details : A performance regression has been discovered in the MPT Fusion
1676 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1677 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1678 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1681 Frequency : PPC/PPC64 only
1683 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1684 Details : fix duplicated definitions between asm-ppc64/types.h and
1685 lustre_types.h on PPC.
1688 Frequency : PPC/PPC64 only
1690 Description: asm-ppc/segment.h does not exist
1691 Details : fix compile issue on PPC.
1695 Description: data checksumming impacts single node performance
1696 Details : add support for several checksum algorithms. Currently, CRC32 and
1697 Adler-32 are supported. The checksum type can be changed on the fly
1698 through /proc/fs/lustre/osc/*/checksum_type.
1702 Description: use adler32 for page checksums
1703 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1708 Description: better handle error messages in extents code
1710 Severity : enhancement
1712 Description: SNMP support enhancement
1713 Details : Adding total number of sampled request for an MDS node in snmp
1716 Severity : enhancement
1718 Description: Optimize ldlm waiting list processing for PR extent locks
1719 Details : When processing waiting list for read extent lock and meeting read
1720 lock that is same or wider to it that is not contended, skip
1721 processing rest of the list and immediatelly return current
1722 status of conflictness, since we are guaranteed there are no
1723 conflicting locks in the rest of the list.
1727 Description: Time out and refuse to reconnect
1728 Details : When the failover node is the primary node, it is possible
1729 to have two identical connections in imp_conn_list. We must
1730 compare not conn's pointers but NIDs, otherwise we
1731 can defeat connection throttling.
1735 Description: Client not clear own cache if answer to reconnect is lost.
1736 Details : Client gets evicted from server. Now client also thinks it is
1737 disconnected (or gets ENOTCONN on its operation) and decides to
1738 reconnect. Server receives reconnect message, but cannot find
1739 export. New export is created that is fully valid (new cookie!),
1740 but reply is lost and not reported to client. Client reconnects
1741 again and gets back a just-created connection, but it is not new
1742 so client thinks it was not evicted and does not do recovery.
1746 Description: Detect stride IO mode in read-ahead
1747 Details : When a client does stride read, read-ahead should detect that and
1748 read-ahead pages according to the detected stride pattern.
1752 Description: build for x2 fails
1753 Details : fix compile issue on Cray systems.
1755 Severity : enhancement
1757 Description: implement readv/writev APIs(aio_read/aio_writes in newer kernels)
1758 Details : This greatly improves speed of NFS writes on 2.6 kernels.
1761 Frequency : only on PPC/SLES10
1763 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1764 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1765 be defined. Add a hack in mkfs_lustre.c to work around this compile
1770 Description: LASSERT on MDS when client holding flock lock dies
1771 Details : ldlm pool logic depends on number of granted locks equal to
1772 number of released locks which is not true for flock locks, so
1773 just exclude such locks from consideration.
1777 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1778 Details : Use fsfilt_sync() for both the whole filesystem sync and
1779 individual file sync to eliminate dangerous inode locking
1780 with I_LOCK that can lead to a deadlock.
1784 Description: Update an obsolete wirecheck.c generator
1785 Details : Update wirecheck.c/wirehdr.c and regenerate wiretest.c
1789 Description: Client can panic on open sometimes
1790 Details : It is possible that we try to free already freed request in
1791 ll_file_open in some error cases when we send request from
1796 Description: performance in 1.6.3
1797 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1798 sure that 1MB requests are not fragmented by the block layer.
1802 Description: LDLM soft lockups - improvement
1803 Details : It is be possible to send the lock handle along with each read
1804 or write request because the client is already doing a lock match
1805 itself so there isn't any reason the OST should have to re-do that
1811 Description: lfs quota fails with deactivated OSTS
1812 Details : With this patch, three improvements are included:
1813 1. detete the softlimit in mds and osts when use "lfs quota".
1814 2. display the inaccurate data in the output of "lfs quota".
1815 3. try to get quota info when "lfs quota" is executed.
1820 Description: Extent locks not granted with no conflicts sometimes.
1821 Details : When race occurs in glimpse handler and nothing is returned,
1822 we do not reprocess the queue after lock cancel, and that leads
1823 to a stall until next activity on a resource
1826 Frequency : failover with quotaon
1828 Description: during mds failovers with quota on, OSTs got into deadlock state
1829 and causing dumpstack.
1830 Details : for every quota slave, at any time, there is only one quota req
1831 is sent to quota master for every uid/gid. Before that quota req
1832 returns, all the thread relative to the same uid/gid will wait.
1833 So if the quota req is lost because mds failovers or any other
1834 reasons, this bug will be hit. Now, dqacq_interpret() will handle
1835 quota reqs who time out.
1837 Severity : enhancement
1840 Description: when quota slave checks if quota is enough, there is an unnecessary
1842 Details : place this wait on necessary place instead of always waiting.
1844 --------------------------------------------------------------------------------
1846 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
1848 * Support for kernels:
1849 2.6.5-7.286 (SLES 9),
1850 2.6.9-55.0.9.EL (RHEL 4),
1851 2.6.16.53-0.8 (SLES 10),
1852 2.6.18-8.1.14.el5 (RHEL 5),
1853 2.6.18.8 vanilla (kernel.org)
1854 * Client support for unpatched kernels:
1855 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1856 2.6.16 - 2.6.22 vanilla (kernel.org)
1857 * Due to recently discovered recovery problems, we do not recommend
1858 using patchless RHEL 4 clients with this or any earlier release.
1859 * Recommended e2fsprogs version: 1.40.2-cfs1
1860 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1863 Frequency : occasional
1865 Description: MDS hang or stay in waiting lock
1866 Details : If client receive lock with CBPENDING flag ldlm need send lock
1867 cancel as separate rpc, to avoid situation when cancel request
1868 can't processed due all i/o threads stay in wait lock.
1871 Frequency : occasional
1873 Description: Do not fail import if osc_interpret_create gets -EAGAIN
1874 Details : If osc_interpret_create got -EAGAIN it immediately exits and
1875 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
1876 oscc_has_objects and see OSC has no objests and call
1877 oscc_internal_create to resend create request.
1879 Severity : enhancement
1881 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1882 Details : Update which_patch & target file for SLES10 latest kernel.
1884 Severity : enhancement
1886 Description: add --type and --size parameters to lfs find
1887 Details : Enhance lfs find by adding filetype and filesize parameters. Also
1888 multiple OBDs can now be specified for the --obd option.
1890 Severity : enhancement
1892 Description: eliminate client locks in face of contention
1893 Details : file contention detection and lockless i/o implementation
1894 for contended files.
1896 Severity : enhancement
1898 Description: Remove client patches from SLES 10 kernel.
1899 Details : This causes SLES 10 clients to behave as patchless clients
1900 even on a Lustre-patched (server) kernel.
1902 Severity : enhancement
1904 Description: use i_size_read and i_size_write in 2.6 port
1905 Details : replace inode->i_size access with i_size_read/write()
1908 Frequency : when removing large files
1910 Description: scheduling issue during removal of large Lustre files
1911 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1912 It causes scheduling issues when removing large files (17TB in the
1918 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
1919 Details : Impossible (EPROTO is returned) to access a directory that has a
1920 non-default striping and ACLs.
1923 Frequency : only on ppc
1925 Description: /proc/fs/lustre/devices broken on ppc
1926 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1927 We should make sure the type of 'index' is loff_t and then cast
1928 explicitly as needed below. Do not assign an explicitly cast
1932 Frequency : only for rhel5
1934 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1935 Details : Modify the target file & which_kernel.
1938 Frequency : if the uninit_groups feature is enabled on ldiskfs
1940 Description: e2fsck reports "invalid unused inodes count"
1941 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
1942 feature and only a single inode is created in a group then the
1943 "bg_unused_inodes" count is incorrectly updated. Creating a
1944 second inode in that group would update it correctly.
1949 Description: buffer overruns could theoretically occur
1950 Details : llapi_semantic_traverse() modifies the "path" argument by
1951 appending values to the end of the origin string, and a buffer
1952 overrun may occur. Adding buffer overrun check in liblustreapi.
1954 Severity : enhancement
1956 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
1959 Frequency : only if filesystem is inconsistent
1961 Description: handle "serious error: objid * already exists" more gracefully
1962 Details : If LAST_ID value on disk is smaller than the objects existing in
1963 the O/0/d* directories, it indicates disk corruption and causes an
1964 LBUG(). If the object is 0-length, then we should use the existing
1965 object. This will help to avoid a full fsck in most cases.
1967 Severity : enhancement
1969 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1970 Details : Modify vm-tunables-rhel4.patch.
1972 Severity : enhancement
1974 Description: Kernel config for 2.6.18-vanilla.
1975 Details : Modify targets/2.6-vanilla.target.in.
1976 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1977 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1978 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1979 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1982 Frequency : occasional
1984 Description: improve handling recoverable errors
1985 Details : If request processed with error which can be recoverable on server
1986 request should be resend, otherwise page released from cache and
1992 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1993 Details : Modify target file & which_patch.
1994 A flaw was found in the IA32 system call emulation provided
1995 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1996 value could be stored in the %RAX register, which could trigger an
1997 out-of-bounds system call table access. An untrusted local user
1998 could exploit this flaw to run code in the kernel
1999 (ie a root privilege escalation). (CVE-2007-4573).
2003 Description: change order of libsysio includes
2004 Details : '#include sysio.h' should always come before '#include xtio.h'
2006 Severity : enhancement
2008 Description: adapt the lustre_config script to support the upgrade case
2009 Details : Add "-u" option for lustre_config script to support upgrading 1.4
2010 server targets to 1.6 in parallel.
2015 Description: To avoid grant space > avaible space when the disk is almost
2016 full. Without this patch you might see the error "grant XXXX >
2017 available" or some LBUG about grant, when the disk is almost
2019 Details : In filter_check_grant, for non_grant cache write, we should
2020 check the left space by if (*left > ungranted + bytes), instead
2021 of (*left > ungranted), because only we are sure the left space
2022 is enough for another "bytes", then the ungrant space should be
2023 increase. In client, we should update cl_avail_grant only there is
2024 OBD_MD_FLGRANT in the reply.
2029 Description: Update RHEL 4 kernel to fix local root privilege escalation.
2030 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
2031 described in CVE-2007-4573. This problem could allow untrusted
2032 local users to gain root access.
2035 Frequency : when using O_DIRECT and quotas
2037 Description: Incorrect file ownership on O_DIRECT output files
2038 Details : block usage reported by 'lfs quota' does not take into account
2039 files that have been written with O_DIRECT.
2043 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2044 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2045 window to 0. If it does hit cache miss, and read-pattern is still
2046 stride-io mode, does not reset the stride window, but also does not
2047 increase the stride window length in this case.
2049 --------------------------------------------------------------------------------
2051 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
2053 * Support for kernels:
2054 2.6.5-7.286 (SLES 9),
2055 2.6.9-55.0.2.EL (RHEL 4),
2056 2.6.16.46-0.14 (SLES 10),
2057 2.6.18-8.1.8.el5 (RHEL 5),
2058 2.6.18.8 vanilla (kernel.org)
2059 * Client support for unpatched kernels:
2060 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2061 2.6.16 - 2.6.21 vanilla (kernel.org)
2062 * Due to recently discovered recovery problems, we do not recommend
2063 using patchless RHEL 4 clients with this or any earlier release.
2064 * Recommended e2fsprogs version: 1.40.2-cfs1
2065 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2069 Description: Fix errors in lfs documentation
2070 Details : Fixes man pages
2072 Severity : enhancement
2074 Description: Adaptive timeouts
2075 Details : RPC timeouts adapt to changing server load and network
2076 conditions to reduce resend attempts and improve recovery time.
2078 Severity : enhancement
2080 Description: llapi_file_create() does not allow some changes
2081 Details : add llapi_file_open() that allows specifying the file creation
2082 mode and open flags, and also returns an open file handle.
2084 Severity : enhancement
2086 Description: df doesn't work properly if diskfs blocksize != 4K
2087 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
2090 Frequency : always on directories with default striping set
2092 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
2093 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
2095 Severity : enhancement
2097 Description: merge and cleanup kernel patches.
2098 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
2100 Severity : enhancement
2102 Description: RedHat Update kernel for RHEL5
2103 Details : Kernel config file for RHEL5.
2105 Severity : enhancement
2107 Description: OSS needs mutliple precreate threads
2108 Details : Add ability to start more than one create thread per OSS.
2111 Frequency : only with quota on the root user
2113 Description: mds_obd_create error creating tmp object
2114 Details : When the user sets quota on root, llog will be affected and can't
2115 create files and write files.
2118 Frequency : Always on ia64 patchless client, and possibly others.
2120 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
2121 Details : This allows the patchless client to be loaded on architectures
2122 without this export.
2127 Description: disorder of journal start and llog_add cause deadlock.
2128 Details : in llog_origin_connect, journal start should happen before
2129 llog_add keep the same order as other functions to avoid
2132 Severity : enhancement
2134 Description: RedHat Update kernel for RHEL5
2135 Details : Modify the kernel config file more closer RHEL5.
2137 Severity : enhancement
2139 Description: Build failure against Centos5 (RHEL5)
2140 Details : Define PAGE_SIZE when it isn't present.
2143 Frequency : occasionally when using NFS
2145 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
2146 Details : with NFS, the anon dentry's parent was set to itself in
2147 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
2148 corresponding dentry other than use rec->ur_name.
2150 Severity : enhancement
2152 Description: enable data checksumming by default
2153 Details : enable checksum by default, allow --disable-checksum
2154 configure option and "-o nochecksum" mount option. Checksums
2155 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
2158 Frequency : Occasionally with failover
2160 Description: Client eviction due to failover config
2161 Details : after a connection loss, the lustre client should attempt to
2162 reconnect to the last active server first before trying the
2163 other potential connections.
2165 Severity : enhancement
2167 Description: client-side metadata stat-ahead during readdir(directory readahead)
2168 Details : perform client-side metadata stat-ahead when the client detects
2169 readdir and sequential stat of dir entries therein
2174 Description: when mds and osts use different quota unit(32bit and 64bit),
2175 quota will be released repeatly.
2176 Details : void sending multiple quota reqs to mds, which will keep the
2177 status between the reqs.
2180 Frequency : only with liblustre clients on XT3
2182 Description: evictions taking too long
2183 Details : allow llrd to evict clients directly on OSTs
2187 Description: osts not allocated evenly to files
2188 Details : change the condition to increase offset_idx
2190 Severity : enhancement
2192 Description: self-adjustable client's lru lists
2193 Details : use adaptive algorithm for managing client cached locks lru
2194 lists according to current server load, other client's work
2195 pattern, memory activities, etc. Both, server and client
2196 side namespaces provide number of proc tunables for controlling
2201 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
2202 Details : rewrite inherited code (uuid parsing code from ext2 utils and
2203 readdir code from ext3) from scratch preserving functionality.
2207 Description: Only those disconnect error should be returned by rq_status.
2208 Details : In open/enqueue processs, Some errors, which will cause client
2209 disconnected, should be returned by rq_status, while other
2210 errors should still be returned by intent, then mdc or llite will
2213 Severity : enhancement
2215 Description: Tune the kernel for good SCSI performance.
2216 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
2217 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
2221 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
2222 Bugzilla : 13547, 13627
2223 Description: Data corruption for OSTs that are formatted larger than 2TB
2225 Details : When generating the bio request for lustre file writes the
2226 sector number would overflow a temporary variable before being
2227 used for the IO. The data reads correctly from Lustre (which
2228 will overflow in a similar manner) but other file data or
2229 filesystem metadata may be corrupted in some cases.
2233 Description: "lfs find -obd UUID" prints directories
2234 Details : "lfs find -obd UUID" will return all directory names instead
2235 of just file names. It is incorrect because the directories
2236 do not reside on the OSTs.
2239 Bugzilla : 13596, 13614
2240 Description: MDS hang after unclean shutdown of lots of clients
2241 Details : Never resend AST requests.
2245 Frequency : Always, for kernels after 2.6.16
2246 Description: Fix warning idr_remove called for id=.. which is not allocated.
2247 Details : Last kernels save old s_dev before kill super and not allow
2248 to restore from callback - restore it before call kill_anon_super.
2252 Description: Fix error on 'ls .' at the top of the Lustre mount.
2253 Details : Don't revalidate dentry if it is a root dentry.
2255 --------------------------------------------------------------------------------
2257 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
2259 * Support for kernels:
2260 2.6.5-7.286 (SLES 9),
2261 2.6.9-55.0.2.EL (RHEL 4),
2262 2.6.16.46-0.14 (SLES 10),
2263 2.6.18-8.1.8.el5 (RHEL 5),
2264 2.6.18.8 vanilla (kernel.org)
2265 * Client support for unpatched kernels:
2266 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2267 2.6.16 - 2.6.22 vanilla (kernel.org)
2268 * Due to recently discovered recovery problems, we do not recommend
2269 using patchless RHEL 4 clients with this or any earlier release.
2270 * Recommended e2fsprogs version: 1.39.cfs8
2271 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2276 Description: block reactivating mgc import until all deactivates complete
2277 Details : Fix race when failing back MDT/MGS to itself (testing)
2279 Severity : enhancement
2281 Description: lfs setstripe enhancement
2282 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
2285 Frequency : mds/oss recovery
2287 Description: llog ctxt is refrenced after it has been freed.
2288 Details : llog ctxt refcount was added to avoide the race between ctxt free
2289 and llog recovery process. Each llog user must hold ctxt refcount
2290 before it access the llog. And the llog ctxt can only be freed
2291 when its refcount is zero.
2293 Severity : enhancement
2295 Description: randomly memory allocation failure util
2296 Details : Make lustre randomly failed allocating memory for testing purpose.
2299 Frequency : only for SLES10
2301 Description: Update kernel patch for SLES10 SP1
2302 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
2304 Severity : enhancement
2306 Description: omit set fsid for export NFS
2307 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
2308 as FS_REQUIRES_DEV for avoid problems with generate fsid.
2311 Frequency : after network failures
2313 Description: Add sync option to mount_lustre.c
2314 Details : Client loses data written to lustre after a network interruption.
2316 Severity : enhancement
2318 Description: Error message improvement.
2319 Details : Merging of two LCONSOLE_ERROR_MSG into one.
2321 Severity : enhancement
2323 Description: don't use GFP_* in generic Lustre code.
2324 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
2326 Severity : enhancement
2328 Description: obdclass is limited by single OBD_ALLOC(idarray)
2329 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
2331 Severity : enhancement
2333 Description: updated patchess for new RHEL4 kernel
2334 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
2336 Severity : enhancement
2338 Description: warnings with build patchless client with vanila 2.6.19 and up
2339 Details : change old ctl_table style and replace ctl_table/ctl_table_header
2340 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
2342 Severity : enhancement
2344 Description: O_DIRECT bypasses client statistics.
2345 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
2346 read_ahead_stats were updated. Copied stats section from
2347 osc_send_oap_rpc() into async_internal().
2350 Frequency : only for Cray XT3
2352 Description: peer credits not enough on many OST per OSS systems.
2353 Details : Use new lnet way to add credits as we need those for pings and ASTs
2356 Frequency : only with liblustre
2358 Description: Liblustre is not releasing flock locks on file close.
2359 Details : Release flock locks on file close.
2362 Frequency : only for RHEL4
2364 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
2365 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
2366 Update target file and kernel config.
2370 Description: lustre support for RHEL5
2371 Details : Add support for RHEL5.
2376 Description: ASSERTION(export != NULL) failed in target_handle_connect
2377 Details : Assetion hit is result of rare race between disconnect and connet
2378 to same nid. target_handle_connect found old connect cockie and
2379 tried to reconnect, but can't find export for this cockie.
2384 Description: umount blocks forever on error
2385 Details : In result of wrong using obd_no_recov and obd_force flags client
2386 can hand if cancel or some other requests is lost.
2389 Frequency : Only for SLES
2391 Description: sanity_quota fail test_1
2392 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
2393 file, which makes TSTID[2] inunique.
2395 Severity : enhancement
2397 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
2398 Details : Update target/ChangeLog/which_patch .
2400 Severity : enhancement
2402 Description: jbd statistics
2403 Details : Port older jbd statistics patch for sles10
2408 Description: lvbo_init failed for resource with missing objects.
2409 Details : Fix returning error if we do stat for file with missing/corrupted
2410 objects and i_size set to all sum of size of all avaible objects.
2411 if we to truncate/write to missing object - it is recreated.
2416 Description: Oops in read and write path when failing to allocate lock.
2417 Details : Check if lock allocation failed and return error back.
2420 Frequency : When flocks are used.
2422 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
2423 Details : Flock locks might destroy just granted lock if it could be merged
2424 with another existing flock, this is done in completion handler,
2425 so teach ldlm_cli_enquque_fini that this is a valid case for
2431 Description: reply_lock_interpret crash due to race with it and lock cancel.
2432 Details : Do not replay locks that are being cancelled. Do not reference
2433 locks by their address during replay, just by their handle.
2436 Frequency : only with deactivated OSTs
2438 Description: lstripe command fails for valid OST index
2439 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
2440 instead of lov->desc.ld_active_tgt_count.
2442 Severity : enhancement
2444 Description: Build failure against Centos5 (RHEL5)
2445 Details : Use getpagesize() instead of PAGE_SIZE.
2447 --------------------------------------------------------------------------------
2449 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
2451 * Support for kernels:
2452 2.6.5-7.283 (SLES 9),
2453 2.6.9-55.EL (RHEL 4),
2454 2.6.16.46-0.14 (SLES 10),
2455 2.6.18.8 vanilla (kernel.org)
2456 * Client support for unpatched kernels:
2457 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2458 2.6.16 - 2.6.22 vanilla (kernel.org)
2459 * Due to recently discovered recovery problems, we do not recommend
2460 using patchless RHEL 4 clients with this or any earlier release.
2461 * Recommended e2fsprogs version: 1.39.cfs8
2462 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2463 * Starting with this release, the ldiskfs backing filesystem required
2464 by Lustre is now in its own package, lustre-ldiskfs. This package
2465 should be installed. It is versioned separately from Lustre and
2466 may be released separately in future.
2471 Description: server LBUG when shutting down
2472 Details : Block umount forever until the mount refcount is zero rather
2473 than giving up after an arbitrary timeout.
2475 Severity : enhancement
2477 Description: add optional extra BUILD_VERSION info
2478 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2479 to override the lustre version.
2482 Frequency : 2.6.18 servers only
2484 Description: ll_kern_mount() doesn't release the module reference
2485 Details : The ldiskfs module reference count never drops down to 0
2486 because ll_kern_mount() doesn't release the module reference.
2491 Description: server LBUG when using old ost_num_threads parameter
2492 Details : Accept the old ost_num_threads parameter but warn that it
2493 is deprecated, and fix an off-by-one error that caused an LBUG.
2498 Description: Transient SCSI error results in persistent IO issue
2499 Details : iobuf->dr_error is not reinitialized to 0 between two
2503 Frequency : sometimes when underlying device returns I/O errors
2505 Description: OSTs not going read-only during write failures
2506 Details : OSTs are not remounted read-only when the journal commit threads
2507 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2508 instead of the ext3 wrappers.
2511 Frequency : only with 10000 clients or more
2513 Description: poor connect scaling with increasing client count
2514 Details : Don't run filter_grant_sanity_check for more than 100 exports
2515 to improve scaling for large numbers of clients.
2518 Frequency : SLES10 only
2520 Description: sanity-quota.sh quotacheck failed: rc = -22
2521 Details : Quotas cannot be enabled on SLES10.
2524 Frequency : liblustre clients only
2526 Description: getdirentries does not give error when run on compute nodes
2527 Details : getdirentries does not fail when the size specified as an argument
2528 is too small to contain at least one entry
2530 Severity : enhancement
2532 Description: Add LNET router traceability for debug purposes
2533 Details : If a checksum failure occurs with a router as part of the
2534 IO path, the NID of the last router that forwarded the bulk data
2535 is printed so it can be identified.
2540 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2541 Details : Due to a race condition, liblustre clients were occasionally
2542 evicted incorrectly.
2544 Severity : enhancement
2546 Description: lfs setstripe use optional parameters instead of postional
2549 Severity : enhancement
2551 Description: Nanosecond timestamp support for ldiskfs
2552 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2553 resolution timestamps. There is not yet support for this at
2554 the Lustre filesystem level.
2557 Frequency : during server recovery
2559 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2560 Details : request with rq_no_resend flag not awake l_wait_event if they get a
2564 Frequency : nfs export on patchless client
2566 Description: connectathon hang when test nfs export over patchless client
2567 Details : Disconnected dentry cannot be found with lookup, so we do not need
2568 to unhash it or make it invalid
2571 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2572 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2573 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2578 Description: open req refcounting wrong on reconnect
2579 Details : If reconnect happened between getting open reply from server and
2580 call to mdc_set_replay_data in ll_file_open, we will schedule
2581 replay for unreferenced request that we are about to free.
2582 Subsequent close will crash in variety of ways.
2583 Check that request is still eligible for replay in
2584 mdc_set_replay_data().
2589 Description: disable writes to filesystem when reading health_check file
2590 Details : the default for reading the health_check proc file has changed
2591 to NOT do a journal transaction and write to disk, because this
2592 can cause reads of the /proc file to hang and block HA state
2593 checking on a healthy but otherwise heavily loaded system. It
2594 is possible to return to the previous behaviour during configure
2595 with --enable-health-write.
2597 Severity : enhancement
2599 Description: 64-bit inode version
2600 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2601 to the inode. This will be required for version-based recovery.
2606 Description: MDS fails to start if a duplicate client export is detected
2607 Details : in some rare cases it was possible for a client to connect to
2608 an MDS multiple times. Upon recovery the MDS would detect this
2609 and fail during startup. Handle this more gracefully.
2611 Severity : enhancement
2613 Description: Add -o localflock option to simulate old noflock behaviour.
2614 Details : This will achieve local-only flock/fcntl locks coherentness.
2619 Description: log_commit_thread vs filter_destroy race leads to crash
2620 Details : Take import reference before releasing llog record semaphore
2625 Description: Wrong request locking in request set processing
2626 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2627 imp_delayed_list, in this place should be used imp_lock.
2630 Frequency : when reconnecting
2632 Description: Grant leak when OSC reconnect to OST
2633 Details : When osc reconnect ost, OST(filter) should check whether it
2634 should grant more space to client by comparing fed_grant and
2635 cl_avail_grant, and return the granted space to client instead
2636 of "new granted" space, because client will call osc_init_grant
2637 to update the client grant space info.
2640 Frequency : when client reconnects to OST
2642 Description: Grant leak when OSC does a resend and replays bulk write
2643 Details : When osc reconnect to OST, OST(filter) should clear grant info of
2644 bulk write request, because the grant info while be sync between
2645 OSC and OST when reconnect, and we should ignore the grant info
2646 these of resend/replay write req.
2651 Description: Grant space more than avaiable space sometimes.
2652 Details : When then OST is about to be full, if two bulk writing from
2653 different clients came to OST. Accord the avaliable space of the
2654 OST, the first req should be permitted, and the second one
2655 should be denied by ENOSPC. But if the seconde arrived before
2656 the first one is commited. The OST might wrongly permit second
2657 writing, which will cause grant space > avaiable space.
2660 Frequency : when client is evicted
2662 Description: Grant might be wrongly erased when osc is evicted by OST
2663 Details : when the import is evicted by server, it will fork another
2664 thread ptlrpc_invalidate_import_thread to invalidate the
2665 import, where the grant will be set to 0. While the original
2666 thread will update the grant it got when connecting. So if
2667 the former happened latter, the grant will be wrongly errased
2668 because of this race.
2673 Description: Checking Stale with correct fid
2674 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2675 to check whether it is stale, because sometimes, we want the
2676 enqueue happened anyway, and op_data.fid2 will not be initialized.
2678 Severity : enhancement
2680 Description: update patchless client
2681 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2684 Frequency : only with 2.4 kernel
2686 Description: random memory corruption
2687 Details : size of struct ll_inode_info is to big for union inode.u and this
2688 can be cause of random memory corruption.
2693 Description: Memory leak in recovery
2694 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2695 It should also check obd_fail before fsfilt_start, otherwise if
2696 fsfilt_start return -EROFS,(failover mds during mds recovery).
2697 then the req will return with repmsg->transno = 0 and rc = EROFS.
2698 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2699 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2700 be freed no matter whether fsfilt_commit success or not.
2703 Frequency : only with large numbers of cores on a single node
2705 Description: Prevents from taking the superblock lock in llap_from_page for
2707 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2708 for llap_from_page call in ll_removepage() prevents from taking
2709 the superblock lock for a soon died page.
2714 Description: Not check open intent error before release open handle
2715 Details : in some rare cases, the open intent error is not checked before
2716 release open handle, which may cause
2717 ASSERTION(open_req->rq_transno != 0), because it tries to release
2718 the failed open handle.
2723 Description: Set cat log bitmap only after create log success.
2724 Details : in some rare cases, the cat log bitmap is set too early. and it
2725 should be set only after create log success.
2729 Description: Accessing a block bevice can re-enable I/O when Lustre is
2730 tearing down a device.
2731 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2736 Description: service threads may hog cpus when there are a lot of requests
2737 Details : Insert cond_resched to give other threads a chance to use some CPU
2742 Description: the cat log was not initialized in recovery
2743 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2744 unlink log on mds will not be initialized until mds post
2745 recovery. And also in mds post recovery, the unlink log will
2746 initialization will be done asynchronausly, so there will be race
2747 between add unlink log and unlink log initialization.
2751 Description: brw_stats were being printed incorrectly
2752 Details : brw_stats were being printed as log2 but all of them were not
2753 recorded as log2. Also remove some code duplication arising from
2754 filter_tally_{read,write}.
2758 Frequency : rare, only in recovery.
2759 Description: ASSERTION(req->rq_type != LI_POISON) failed
2760 Details : imp_lock should be held while iterating over imp_sending_list for
2761 prevent destroy request after get timeout in ptlrpc_queue_wait.
2765 Description: replay-single.sh test 52 fails
2766 Details : A lock's skiplist need to be cleanup when it being unlinked
2767 from its resource list.
2772 Description: Short directio read returns full requested size rather than
2774 Details : Direct I/O operations should return actual amount of bytes
2775 transferred rather than requested size.
2777 Severity : enhancement
2779 Description: metadata RPC reduction (e.g. for rm performance)
2780 Details : decrease the amount of synchronous RPC between clients and servers
2781 by canceling conflicing lock before the operation on the client
2782 and packing thier handles into the main operation RPC to server.
2784 Severity : enhancement
2786 Description: add #ifdef HAVE_KERNEL_CONFIG_H
2787 Details : kernels from 2.6.19 not need include linux/config.h, but add
2788 include linux/autoconf.h in commpiler command line.
2790 Severity : enhancement
2792 Description: patchless client support for 2.6.22 kernel
2793 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
2794 removed. In this case we need drop using os depended interface to
2795 kmem_cache and use cfs_mem_cache API.
2800 Description: fix mal-formatted messages
2801 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
2805 Frequency : always in liblustre
2806 Description: wrong IS_ERR implementation in liblustre.h
2807 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
2812 Description: Correct condition for output debug message.
2813 Details : inode i_nlink equal zero is not enough for output message about
2814 disk corruption, i_ctime and i_mode should be also checked.
2818 Frequency : always in patchless client
2819 Description: add configure check for truncate_complete_page
2820 Details : improve checks for exported symbols. This allow run check without
2821 sources, but with Module.symvers shipped with kernel distribution.
2822 add check for truncate_complete_page used by patchless client.
2826 Description: sanity.sh test_77h fails with "test_77h file compare failed"
2827 Details : test_77h uses a file which was messed by other test case.
2831 Description: Not Check whether lov_tgts is NULL in some lov functions
2832 Details : Checking whether lov_tgts is NULL in some functions.
2836 Description: replace obdo_alloc() with OBDO_ALLOC macro
2837 Details : nothing special is done in obdo_alloc() function, and for
2838 debugging purpose, it needs to be replaced with macros.
2842 Description: bad return value and errno from fcntl call
2843 Details : In liblustre API, errno should be a negative value if error
2848 Description: ptlrpc_check_set() LBUG
2849 Details : In case of positive reply from server and failed client bulk
2850 callback after bulk transfer shouldn't LBUG, but process this
2851 request as erroneous.
2853 Severity : enhancement
2855 Description: VFS operations stats tool.
2856 Details : Tool which collects stats by tracking value written in pid,
2857 ppid, gid and uses llstat to generate output to plot graph using
2859 Updated lustre/utils/Makefile.am
2860 Added lustre/utils/ltrack_stats.c
2862 Severity : enhancement
2864 Description: 2.6.18 server support (lustre 1.6.1)
2865 Details : Support for 2.6.18 kernels on the server side.
2870 Description: ASSERTION(imp->imp_conn_current) failed
2871 Details : an assertion failure is hit if a client node boots and attempts to
2872 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
2875 Frequency : only for i686
2877 Description: 1.4.11 RC1 build fails for RHEL 4, i686
2878 Details : Fixed config variable for build.
2883 Description: Updated patchess for new RHEL4 kernel
2884 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
2885 Updated patch jbd-stats-2.6.9.patch
2886 Updated patch qsnet-rhel4-2.6.patch
2887 Updated patch quota-deadlock-on-pagelock-core.patch
2888 Updated patch vfs_intent-2.6-rhel4.patch
2889 Updated patch vfs_races-2.6-rhel4.patch
2890 Updated series file 2.6-rhel4-titech.series
2891 Updated series file 2.6-rhel4.series
2892 Updated kernel config files
2897 Description: lquota slave complains LBUG when reconnecting with mds
2899 Details : quota slave depends qctxt->lqc_import to send its quota request.
2900 This pointer will be invalid if mds did failover or broke its
2901 connect to osts, which leads to LBUG.
2904 Frequency : when qunit size is too small(less than 20M)
2906 Description: write is stopped by improper -EDQUOT
2907 Details : If the master is busy and qunit size is small enough(let's say 1M),
2908 the slave can not get quota from master on time, which will lead
2909 slave to trigger a -EQUOTA to client.
2914 Description: Deadlock during metadata tests
2915 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
2916 called with the per-dentry lock held.
2919 Frequency : SLES9 only
2921 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
2922 Details : set CONFIG_NR_CPUS to 128 instead of 8.
2924 Severity : enhancement
2926 Description: remove fs_prep_san_write operation and related patches
2927 Details : remove the ext3-san-jdike patches which are no longer useful.
2932 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
2933 Details : a disk corruption can cause the mballoc code to assert on a
2934 double free or other extent corruptions. Handle these with
2935 ext3_error() instead of assertions.
2938 Frequency : only with mballoc3 code and deep extent trees
2940 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
2941 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
2942 can cause the filesystem to be remounted read-only.
2947 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
2948 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
2949 few external users of this ioctl number is lfsck's call to
2950 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
2951 Add the old ioctl number to the handler so both old and new
2956 Description: parallel lock callbacks
2957 Details : Instead of sending blocking and completion callbacks as separated
2958 requests, adding them to a set and sending in parallel.
2962 Description: Disable most debugging by default
2963 Details : To improve performance, disable most logging (for debug purposes)
2964 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
2965 default, and HA includes fewer messages.
2969 Frequency : only run on patchless client.
2970 Description: use do_facet on sanity.sh for test handling recoverables errors
2971 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
2975 Description: hash tables for lists of nids, connections and uuids
2976 Details : Hash tables noticeably help when a lot of clients connect to a
2977 server, to faster identify duplicate connections or reconnects,
2978 also to faster find export to evict in manual eviction case.
2982 Description: Sometimes, when the server evict a client, and the client will
2983 not be evicted as soon as possible.
2984 Details : In enqueue req, the error was returned by intent, instead of
2985 rq_status which make ptlrpc layer not detect this error, and
2986 does not evict the client. So enqueue error should be returned
2990 Frequency : only at startup
2992 Description: Delay client connections to MDT until fist MDT->OST connect
2993 Details : If a client tried to create a new file before the MDT had
2994 connected to any OSTs, the create would return EIO. Now
2995 the client will simply block until the MDT connects to the
2996 first OST and the create can succeed.
2999 Frequency : only for SLES9
3001 Description: Routinely utilize latest Quadrics drivers in CFS releases
3002 Details : Update patch qsnet-suse-2.6.patch.
3005 Frequency : only for sles10
3007 Description: Update patches for SLES 10 SP1 kernel.
3008 Details : Update the patch vfs_intent-2.6-sles10.patch.
3013 Description: Routinely utilize latest Quadrics drivers in CFS releases
3014 Details : Update patch qsnet-rhel4-2.6.patch.
3017 Frequency : at statup only
3019 Description: mds_lov_synchronize race leads to various problems
3020 Details : simultaneous MDT->OST connections at startup can cause the
3021 sync to abort, leaving the OSC in a bad state.
3026 Description: Using wrong pointer in osc_brw_prep_request
3027 Details : Access to array[-1] can produce panic if kernel compiled with
3028 CONFIG_PAGE_ALLOC enabled
3030 Severity : enhancement
3032 Description: Async OSC create to avoid the blocking unnecessarily.
3033 Details : If a OST has no remain object, system will block on the creating
3034 when need to create a new object on this OST. Now, ways use
3035 pre-created objects when available, instead of blocking on an
3036 empty osc while others are not empty. If we must block, we block
3037 for the shortest possible period of time.
3041 Frequency : only in recovery
3042 Description: Mark OST as early accessible if his start SYNC.
3043 Details : osc_precreate return flag early accessible if oscc marked as
3044 OSCC_FLAG_SYNC_IN_PROGRESS.
3049 Description: Sometimes precreate code can triger create object on wrong ost
3050 Details : Wrong protected or not not restored variables aftre precreate loop
3051 can produce creation object on wrong ost.
3054 Frequency : oss recovery
3056 Description: llog_commit_thread cleanup should sync with llog_commit_thread
3058 Details : llog_commit_thread_count should be synced between llog_commit
3059 start and cleanup, so new llog_commit thread should not be started
3060 when llog_commit threads being stopped to avoid accessing some
3063 Severity : enhancement
3065 Description: Add printing inode info into message about error in writepage.
3067 --------------------------------------------------------------------------------
3069 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
3074 Frequency : on some architectures
3076 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
3077 Details : Uninitialized flags sometimes cause configuration commands to
3081 Frequency : patchless clients only
3083 Description: missing __iget() symbol export
3084 Details : The __iget() symbol export is missing. To avoid the need for
3085 this on patchless clients the deathrow inode reaper is turned
3086 off, and we depend on the VM to clean up old inodes. This
3087 dependency was during via the fix for bug 12181.
3092 Description: sanity.sh fail: test_52b
3093 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
3094 incorrect inode's flags to client.
3096 --------------------------------------------------------------------------------
3098 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
3100 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
3101 INTEROPERATE with older versions automatically. In many cases a
3102 special upgrade step is needed. Please read the
3103 user documentation before upgrading any part of a 1.4.x system.
3104 * WARNING: Lustre configuration and startup changes are required with
3105 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
3107 * Support for kernels:
3108 2.4.21-47.0.1.EL (RHEL 3),
3109 2.6.5-7.283 (SLES 9),
3110 2.6.9-42.0.10.EL (RHEL 4),
3111 2.6.12.6 vanilla (kernel.org),
3112 2.6.16.27-0.9 (SLES10)
3113 * Client support for unpatched kernels:
3114 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
3115 2.6.16 - 2.6.19 vanilla (kernel.org),
3116 2.6.9-42.0.8EL (RHEL 4)
3117 * Recommended e2fsprogs version: 1.39.cfs6
3118 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3121 Severity : enhancement
3123 Description: Async OSC create to avoid the blocking unnecessarily.
3124 Details : If a OST has no remain object, system will block on the creating
3125 when need to create a new object on this OST. Now, ways use
3126 pre-created objects when available, instead of blocking on an
3127 empty osc while others are not empty. If we must block, we block
3128 for the shortest possible period of time.
3130 Severity : enhancement
3132 Description: MountConf
3133 Details : Lustre configuration is now managed via mkfs and mount
3134 commands instead of lmc and lconf. New obd types (MGS, MGC)
3135 are added for dynamic configuration management. See
3136 https://mail.clusterfs.com/wikis/lustre/MountConf for
3139 Severity : enhancement
3141 Description: dynamic OST addition
3142 Details : OSTs can now be added to a live filesystem
3144 Severity : enhancement
3146 Description: startup order invariance
3147 Details : MDTs and OSTs can be started in any order. Clients only
3148 require the MDT to complete startup.
3150 Severity : enhancement
3152 Description: parallel, asynchronous orphan cleanup
3153 Details : orphan cleanup is now performed in separate threads for each
3154 OST, allowing parallel non-blocking operation.
3156 Severity : enhancement
3158 Description: optimized stripe assignment
3159 Details : stripe assignments are now made based on ost space available,
3160 ost previous usage, and OSS previous usage, in order to try
3161 to optimize storage space and networking resources.
3163 Severity : enhancement
3165 Description: Permanently set tunables
3166 Details : All writable /proc/fs/lustre tunables can now be permanently
3167 set on a per-server basis, at mkfs time or on a live
3170 Severity : enhancement
3172 Description: Lustre message v2
3173 Details : Add lustre message format v2.
3175 Severity : enhancement
3177 Description: client OST exclusion list
3178 Details : Clients can be started with a list of OSTs that should be
3179 declared "inactive" for known non-responsive OSTs.
3183 Description: ENOENT returned for valid filehandle during dbench.
3184 Details : Check if a directory has children when invalidating dentries
3185 associated with an inode during lock cancellation. This fixes
3186 an incorrect ENOENT sometimes seen for valid filehandles during
3187 testing with dbench.
3190 Frequency : SFS test only (otherwise harmless)
3192 Description: SPEC SFS validation failure on NFS v2 over lustre.
3193 Details : Changes the blocksize for regular files to be 2x RPC size,
3194 and not depend on stripe size.
3196 Severity : enhancement
3198 Description: fine-grained SMP locking inside DLM
3199 Details : Improve DLM performance on SMP systems by removing the single
3200 per-namespace lock and replace it with per-resource locks.
3202 Severity : enhancement
3204 Description: don't hold multiple extent locks at one time
3205 Details : To avoid client eviction during large writes, locks are not
3206 held on multiple stripes at one time or for very large writes.
3207 Otherwise, clients can block waiting for a lock on a failed OST
3208 while holding locks on other OSTs and be evicted.
3210 Severity : enhancement
3212 Description: Multiple MD RPCs in flight.
3213 Details : Further unserialise some read-only MDT RPCs - learn about intents.
3214 To avoid overly-overloading MDT, introduce a limit on number of
3215 MDT RPCs in flight for a single client and add /proc controls
3216 to adjust this limit.
3218 Severity : enhancement
3220 Description: client read/write statistics
3221 Details : Add client read/write call usage stats for performance
3222 analysis of user processes.
3223 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
3224 file access. extents_stats shows chunk size distribution.
3225 extents_stats_per_process show chunk size distribution per
3228 Severity : enhancement
3230 Description: per-client statistics on server
3231 Details : Add ldlm and operations statistics for each client in
3232 /proc/fs/lustre/mds|obdfilter/*/exports/
3234 Severity : enhancement
3236 Description: improved MDT statistics
3237 Details : Add detailed MDT operations statistics in
3238 /proc/fs/lustre/mds/*/stats
3240 Severity : enhancement
3242 Description: VFS operations stats
3243 Details : Add client VFS call stats, trackable by pid, ppid, or gid
3244 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
3249 Description: Fix client-side osc byte counters
3250 Details : The osc read/write byte counters in
3251 /proc/fs/lustre/osc/*/stats are now working
3254 Frequency : always as root on SLES
3256 Description: Failure of copying files with lustre special EAs.
3257 Details : Client side always return success for setxattr call for lustre
3258 special xattr (currently only "trusted.lov").
3263 Description: Refcount LNET uuids
3264 Details : The global LNET uuid list grew linearly with every startup;
3265 refcount repeated list entries instead of always adding to
3268 Severity : enhancement
3270 Description: Dynamic service threads
3271 Details : Within a small range, start extra service threads
3272 automatically when the request queue builds up.
3275 Frequency : mixed-endian client/server environments
3277 Description: mixed-endian crashes
3278 Details : The new msg_v2 system had some failures in mixed-endian
3281 Severity : enhancement
3283 Description: Easy OST removal
3284 Details : OSTs can be permanently deactivated with e.g. 'lctl
3285 conf_param lustre-OST0001.osc.active=0'
3287 Severity : enhancement
3289 Description: MGS proc entries
3290 Details : Added basic proc entries for the MGS showing what filesystems
3293 Severity : enhancement
3295 Description: provide MGS failover
3296 Details : Added config lock reacquisition after MGS server failover.
3298 Severity : enhancement
3300 Description: add Linux 2.4 support
3301 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
3305 Description: a large application tries to do I/O to the same resource and dies
3306 in the middle of it.
3307 Details : Check the req->rq_arrival time after the call to
3308 ost_brw_lock_get(), but before we do anything about
3309 processing it & sending the BULK transfer request. This
3310 should help move old stale pending locks off the queue as
3311 quickly as obd_timeout.
3314 Frequency : when an incorrect nid is specified during startup
3316 Description: ptlrpc connect to non-existant node causes kernel crash
3317 Details : LNET can't be re-entered from an event callback, which
3318 happened when we expire a message after the export has been
3319 cleaned up. Instead, hand the zombie cleanup off to another
3322 Severity : enhancement
3324 Description: plain/inodebits lock performance improvement
3325 Details : Grouping plain/inodebits in granted list by their request modes
3326 and bits policy, thus improving the performance of search through
3330 Frequency : only if OST filesystem is corrupted
3332 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3333 Details : for a short time RPCs with bulk IO are in the replay list,
3334 but replay of bulk IOs is unimplemented. If the OST filesystem
3335 is corrupted due to disk cache incoherency and then replay is
3336 started it is possible to trip an assertion. Avoid putting
3337 committed RPCs into the replay list at all to avoid this issue.
3340 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3342 Description: System hang on startup
3343 Details : This bug allowed the liblustre (e.g. catamount) client to
3344 return to the app before handling all startup RPCs. This
3345 could leave the node unresponsive to lustre network traffic
3346 and manifested as a server ptllnd timeout.
3348 Severity : enhancement
3350 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3351 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
3352 boolean to control whether to print peer debug info when a
3353 client's RPC times out.
3356 Frequency : only for kernels with patches from Lustre below 1.4.3
3358 Description: Remove old rdonly API
3359 Details : Remove old rdonly API which unused from at least lustre 1.4.3
3362 Frequency : only for devices with external journals
3364 Description: Set external device read-only also
3365 Details : During a commanded failover stop, we set the disk device
3366 read-only while the server shuts down. We now also set any
3367 external journal device read-only at the same time.
3370 Frequency : when upgrading from 1.4 while trying to change parameters
3372 Description: The wrong (new) MDC name was used when setting parameters for
3373 upgraded MDT's. Also allows changing of OSC (and MDC)
3374 parameters if --writeconf is specified at tunefs upgrade time.
3377 Frequency : when setting specific ost indicies
3379 Description: QOS code breaks on skipped indicies
3380 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3381 created with --index need not be sequential.
3383 Severity : enhancement
3385 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3386 Details : The uninit_groups feature works in conjunction with the kernel
3387 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3388 up the pass1 processing of e2fsck. This is a read-only feature
3389 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3390 mount filesystems that have had this feature enabled.
3392 Severity : enhancement
3394 Description: Improve multi-block allocation algorithm to avoid fragmentation
3395 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3396 allocation locality and avoid filesystem fragmentation.
3398 ------------------------------------------------------------------------------
3400 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3402 * Support for kernels:
3403 2.4.21-47.0.1.EL (RHEL 3)
3404 2.6.5-7.283 (SLES 9)
3405 2.6.9-42.0.10.EL (RHEL 4)
3406 2.6.12.6 vanilla (kernel.org)
3407 2.6.16.27-0.9 (SLES 10)
3408 * Recommended e2fsprogs version: 1.39.cfs5
3410 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3414 Frequency : occasional, depends on client load and configuration
3415 Bugzilla : 12181, 12203
3416 Description: data loss for recently-modified files
3418 Details : In some cases it is possible that recently written or created
3419 files may not be written to disk in a timely manner (this should
3420 normally be within 30s unless client IO load is very high).
3421 The problem appears as zero-length files or files that are a
3422 multiple of 1MB in size after a client crash or client eviction
3423 that are missing data at the end of the file.
3425 This problem is more likely to be hit on clients where files are
3426 repeatedly created and unlinked in the same directory, clients
3427 have a large amount of RAM, have many CPUs, the filesystem has
3428 many OSTs, the clients are rebooted frequently, and/or the files
3429 are not accessed by other nodes after being written.
3431 The presence of the problem can be detected by looking at
3432 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
3433 smaller than the second (nr_unused) then dirty files will not
3434 be flushed automatically to disk. "sync; sleep 10" should be
3435 run several times on the node before unmounting it to update
3436 Lustre (this is also safe to run on nodes without this problem).
3438 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
3439 that can cause this same problem, so customers using that kernel
3440 also need to update the kernel in addition to Lustre. In order
3441 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3443 It is normal that files written just before a client crash (less
3444 than 30s) may not yet have been flushed to disk, even for local
3448 Frequency : frequent on thin XT3 nodes
3450 Description: UUID collision on thin XT3 Linux nodes
3451 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3452 randomly, since we relied on an insufficiently-seeded PRNG.
3457 Description: OSS hangs after "All ost request buffers busy"
3458 Details : A deadlock between quota and journal operations caused OSS
3459 hangs after printing "All ost request buffers busy."
3462 Frequency : always on liblustre builds
3464 Description: Cleanup compiler warnings on liblustre
3467 Frequency : always on liblustre builds on XT3
3469 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3473 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3474 > 32000 subdirectories
3475 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3476 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3477 subdirectory count crosses 32000. This will aid e2fsck to
3478 correctly handle more than 32000 subdirectories.
3481 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3483 Description: System hang on startup
3484 Details : This bug allowed the liblustre (e.g. catamount) client to
3485 return to the app before handling all startup RPCs. This
3486 could leave the node unresponsive to lustre network traffic
3487 and manifested as a server ptllnd timeout.
3489 Severity : enhancement
3491 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3492 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3493 boolean to control whether to print peer debug info when a
3494 client's RPC times out.
3499 Description: make O_SYNC working on 2.6 kernels
3500 Details : 2.6 kernels use different method for mark pages for write,
3501 so need add a code to lustre for O_SYNC work.
3506 Description: Failure to close file and release space on NFS
3507 Details : Put inode details into lock acquired in ll_intent_file_open.
3508 Use mdc_intent_lock in ll_intent_open to properly
3509 detect all kind of errors unhandled by mdc_enqueue.
3514 Description: proc file read during shutdown sometimes raced obd removal,
3516 Details : Add lock to prevent obd access after proc file removal.
3519 Frequency : Only for files larger than 4GB on 32-bit clients.
3521 Description: improperly doing page alignment of locks
3522 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3523 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3528 Description: RPCs being resent when they shouldn't be
3529 Details : Some RPCs that should not be resent are being resent. This
3530 can cause inconsistencies in the RPC state machine. Do not
3531 resend such requests.
3534 Frequency : rare, only with NFS export
3536 Description: Crash on NFS re-export node
3537 Details : under very unusual load conditions an assertion is hit in
3538 ll_intent_file_open()
3541 Frequency : only if OST filesystem is corrupted
3543 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3544 Details : for a short time RPCs with bulk IO are in the replay list,
3545 but replay of bulk IOs is unimplemented. If the OST filesystem
3546 is corrupted due to disk cache incoherency and then replay is
3547 started it is possible to trip an assertion. Avoid putting
3548 committed RPCs into the replay list at all to avoid this issue.
3553 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3554 Details : Large single O_DIRECT read and write calls can fail to allocate
3555 a sufficiently large buffer to process the request. In case of
3556 allocation failure the allocation is retried with a smaller
3557 buffer and broken into smaller requests.
3559 Severity : enhancement
3561 Description: Add -o localflock option to simulate old noflock behaviour.
3562 Details : This will achieve local-only flock/fcntl locks coherentness.
3567 Description: versioning check is incomplete
3568 Details : Checking the version difference of client vs. server, report
3569 error if the gap is too big.
3574 Description: add support PG_writeback bit
3575 Details : add support for PG_writeback bit for Lustre, for more carefull
3576 work with page cache in 2.6 kernel. This also fix some deadlocks
3577 and remove hack for work O_SYNC with 2.6 kernel.
3579 Severity : enhancement
3581 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3582 Details : The uninit_groups feature works in conjunction with the kernel
3583 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3584 up the pass1 processing of e2fsck. This is a read-only feature
3585 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3586 mount filesystems that have had this feature enabled.
3588 Severity : enhancement
3590 Description: Improve multi-block allocation algorithm to avoid fragmentation
3591 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3592 allocation locality and avoid filesystem fragmentation.
3594 ------------------------------------------------------------------------------
3596 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3598 * Support for kernels:
3599 2.6.9-42.0.3.EL (RHEL 4)
3600 2.6.5-7.276 (SLES 9)
3601 2.4.21-47.0.1.EL (RHEL 3)
3602 2.6.12.6 vanilla (kernel.org)
3603 2.6.16.21-0.8 (SLES10)
3604 * Recommended e2fsprogs version: 1.39.cfs2-0
3606 * The backwards-compatible /proc/sys/portals symlink has been removed
3607 in this release. Before upgrading, please ensure that you change
3608 any configuration scripts or /etc/sysctl.conf files that access
3609 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3610 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3611 in advance of the upgrade on any system running Lustre 1.4.6 or
3612 newer, since /proc/sys/lnet was added in that version.
3613 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3617 Frequency : only when quota is used
3619 Description: avoid scanning export list for quota master
3620 Details : Change the algorithms to avoid scanning export list in order
3621 to improve the efficiency.
3624 Frequency : MDS failover only, very rarely
3626 Description: "went back in time" messages on mds failover
3627 Details : The greatest transno may be lost when the current operation
3628 finishes with an error (transno==0) and the client's last_rcvd
3629 record is over-written. Save the greatest transno in the
3630 mds_last_transno for this case.
3633 Frequency : always for specific kernels and striping counts
3635 Description: client may get "Matching packet too big" without ACL support
3636 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3637 when trying to access files in certain configurations. The
3638 clients should in fact be denied when mounting because they do
3639 not understand ACLs.
3642 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3644 Description: many clients connecting with IO in progress causes connect timeouts
3645 Details : Avoid synchronous journal commits to avoid delays caused by many
3646 clients connecting/disconnecting when bulk IO is in progress.
3647 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3648 OST_IO_PORTAL to avoid delays behind potentially many pending
3652 Frequency : occasionally with multiple writers to a single file
3654 Description: shared writes to file may result in wrong size reported by stat()
3655 Details : Allow growing of kms when extent lock is cancelled
3658 Frequency : always with random mmap IO to multi-striped file
3660 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3661 Details : Only if the hole is at the end of OST object so that kms is too
3662 small. Fix is to increase kms accordingly in ll_nopage.
3665 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3667 Description: writes to a missing object would leak memory on the OST
3668 Details : If there is an inconsistency between the MDS and OST filesystems,
3669 such that the MDS references an object that doesn't exist, writes
3670 to that object will leak memory due to incorrect cleanup in the
3671 error handling path, eventually running out of memory on the OST.
3676 Description: Creating too long symlink causes lustre errors
3677 Details : Check symlink and name lengths before sending requests to MDS.
3680 Frequency : only if flock is enabled (not on by default)
3682 Description: posix locks not released on fd closure on 2.6.9+
3683 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3684 kernels, this caused such locks not to be released on fd close and
3685 then assertions on fs unmount about still used locks.
3688 Frequency : MDS failover only, very rarely
3690 Description: clients may get ASSERTION(granted_lock != NULL)
3691 Details : When request was taking a long time, and a client was resending
3692 a getattr by name lock request. The were multiple lock requests
3693 with the same client lock handle and
3694 mds_getattr_name->fixup_handle_for_resent_request found one of the
3695 lock handles but later failed with ASSERTION(granted_lock != NULL).
3700 Description: handle->h_buffer_credits > 0, assertion failure
3701 Details : h_buffer_credits is zero after truncate, causing assertion
3702 failure. This patch extends the transaction or creates a new
3706 Frequency : NFS re-export or patchless client
3707 Bugzilla : 11179, 10796
3708 Description: Crash on NFS re-export node (__d_move)
3709 Details : We do not want to hash the dentry if we don't have a lock.
3710 But if this dentry is later used in d_move, we'd hit uninitialised
3711 list head d_hash, so we just do this to init d_hash field but
3712 leave dentry unhashed.
3715 Frequency : NFS re-export or patchless client
3717 Description: NFS exports has problem with symbolic link
3718 Details : lustre client didn't properly install dentry when re-exported
3719 to NFS or running patchless client.
3722 Frequency : NFS re-export or patchless client
3724 Description: Various nfs/patchless fixes.
3725 Details : fixes reuse disconected alias for lookup process - this fixes
3726 warning "find_exported_dentry: npd != pd",
3727 fix permission error with open files at nfs.
3728 fix apply umask when do revalidate.
3731 Frequency : occasional
3733 Description: Crash on NFS re-export node
3734 Details : calling clear_page() on the wrong pointer triggered oops in
3735 generic_mapping_read().
3738 Frequency : rarely, using O_DIRECT IO
3740 Description: unaligned directio crashes client with LASSERT
3741 Details : check for unaligned buffers before trying any requests.
3744 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3746 Description: stale data returned from RAID cache
3747 Details : If only a small amount of IO is done to the RAID device before
3748 reading it again it is possible to get stale data from the RAID
3749 cache instead of reading it from disk.
3752 Frequency : always for sles10 kernel
3754 Description: sles10 support
3755 Details : ll_follow_link: compile fixes and using of nd_set_link
3756 under newer kernels.
3759 Frequency : depends on arch, kernel and compiler version, always on sles10
3762 Description: recursive or deep enough symlinks cause stack overflow
3763 Details : getting rid of large stack-allocated variable in
3767 Frequency : depends on hardware
3769 Description: lustre write performance loss in the SLES10 kernel
3770 Details : the performance loss is caused by using of write barriers in the
3771 ext3 code. The SLES10 kernel turns barrier support on by
3772 default. The fix is to undo that change for ldiskfs.
3774 ------------------------------------------------------------------------------
3776 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3778 * Support for kernels:
3779 2.6.9-42.0.3EL (RHEL 4)
3780 2.6.5-7.276 (SLES 9)
3781 2.4.21-47.0.1.EL (RHEL 3)
3782 2.6.12.6 vanilla (kernel.org)
3786 Frequency : quota enabled and large files being deleted
3788 Description: releasing more than 4GB of quota at once hangs OST
3789 Details : If a user deletes more than 4GB of files on a single OST it
3790 will cause the OST to spin in an infinite loop. Release
3791 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3796 Description: statfs data retrieved from /proc may be stale or zero
3797 Details : When reading per-device statfs data from /proc, in the
3798 {kbytes,files}_{total,free,avail} files, it may appear
3799 as zero or be out of date.
3802 Frequency : systems with MD RAID1 external journal devices
3804 Description: lconf's call to blkid is confused by RAID1 journal devices
3805 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3806 of returning all block devices that match the journal UUID.
3809 Frequency : always, for aggregate stripe size over 4GB
3811 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3812 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3813 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3816 Frequency : always if "lfs find" used on a local file/directory
3818 Description: "lfs find" segfaults if used on a local file/directory
3819 Details : The case where a directory component was not specified wasn't
3820 handled correctly. Handle this properly.
3823 Frequency : always on ppc64
3825 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3826 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3827 perform endianness conversion.
3830 Frequency : rarely (truncate to non-zero file size after write under load)
3831 Bugzilla : 10730, 10687
3832 Description: Files padded with zeros to next 4K multiple
3833 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3834 it is possible that files that are truncated to a non-zero size
3835 immediately after being written are filled with zero bytes beyond
3836 the truncated size. No file data is lost.
3838 Severity : enhancement
3840 Description: Allow recovery/failover for liblustre clients.
3841 Details : liblustre clients were unaware of failover configurations until
3844 Severity : enhancement
3846 Description: user file locks should fail when not mounting with flock option
3847 Details : Set up an error-returning stub in ll_file_operations.lock field
3848 to prevent incorrect behaviour when client is mounted without
3849 flock option. Also, set up properly f_op->flock field for
3853 Frequency : always on ia64
3855 Description: "lfs df" loops on printing out MDS statfs information
3856 Details : The obd_ioctl_data was not initialized and in some systems
3857 this caused a failure during the ioctl that did not return
3858 an error. Initialize the struct and return an error on failure.
3861 Frequency : SLES 9 only
3863 Description: Error of copying files with lustre special EAs as root
3864 Details : Client side always return success for setxattr call for lustre
3865 special xattr (currently only "trusted.lov").
3868 Frequency : rarely on clusters with both ia64+i386 clients
3870 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3871 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3872 clients are doing concurrent writes to the same file it is
3873 possible that the ia64 clients may LASSERT because the OST
3874 extent locks are not PAGE_SIZE aligned. Ensure that grown
3875 locks are always aligned on the request boundary.
3878 Frequency : specific use, occasional
3880 Description: Overwriting in use executable truncates on-disk binary image
3881 Details : If one node attempts to overwrite an executable in use by
3882 another node, we now correctly return ETXTBSY instead of
3883 truncating the file.
3885 Severity : enhancement
3887 Description: Async OSC create to avoid the blocking unnecessarily.
3888 Details : If a OST has no remain object, system will block on the creating
3889 when need to create a new object on this OST. Now, ways use
3890 pre-created objects when available, instead of blocking on an
3891 empty osc while others are not empty. If we must block, we block
3892 for the shortest possible period of time.
3897 Description: chmod on Lustre root is propagated to other clients
3898 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3899 invalid by the follow_mount time.
3902 Frequency : liblustre clients only
3904 Description: Race in 'instant cancel' lock handling could lead to such locks
3905 never to be granted in case of SMP MDS
3906 Details : Do not destroy not yet granted but cbpending locks in
3910 Frequency : replay/resend of open
3912 Description: non null lock assetion failure in mds_intent_policy
3913 Details : Trying to replay/resend lockless open requests resulted in
3914 mds_open() returning 0 with no lock. Now it sets a flag if
3915 a lock is going to be returned.
3917 Severity : enhancement
3919 Description: Checksum enhancements
3920 Details : New checksum enhancements allow for resending RPCs that failed
3923 Severity : enhancement
3925 Description: Tunables on number of dirty pages in cacche
3926 Details : Allow to set limit on number of dirty pages cached.
3931 Description: client crash on unmount - lock still has references
3932 Details : In some error handling cases it was possible to leak a lock
3933 reference on a client while accessing a file. This was not
3934 harmful to the client during operation, but would cause the
3935 client to crash when the filesystem is unmounted.
3938 Frequency : specific case, rare
3940 Description: ETXTBSY on mds though file not in use by client
3941 Details : ETXTBSY is no longer incorrectly returned when attempting to
3942 chmod or chown a directory that the user previously tried to
3943 execute or a currently-executing binary.
3946 Frequency : extremely rare except on liblustre-based clients
3948 Description: Lustre space not freed when files are deleted
3949 Details : Clean up open-unlinked files after client eviction. Previously
3950 the unlink was skipped and the files remained as orphans.
3955 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3956 Details : In some cases it was possible to send a blocking callback to
3957 a client doing a glimpse, even though that client didn't get
3958 a lock granted. When the glimpse lock is cancelled on the OST
3959 the freed lock is left on the waiting list and corrupted the list.
3962 Frequency : all core dumps
3964 Description: Broke core dumps to lustre
3965 Details : Negative dentry may be unhashed if parent does not have UPDATE
3966 lock, but some callers, e.g. do_coredump, expect dentry to be
3967 hashed after successful create, hash it in ll_create_it.
3969 ------------------------------------------------------------------------------
3971 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3973 * Support for kernels:
3974 2.6.9-42.0.2.EL (RHEL 4)
3975 2.6.5-7.276 (SLES 9)
3976 2.4.21-40.EL (RHEL 3)
3977 2.6.12.6 vanilla (kernel.org)
3981 Frequency : always on RHEL 3
3983 Description: Number of open files grows over time
3984 Details : The number of open files grows over time, whether or not
3985 Lustre is started. This was due to a filp leak introduced
3986 by one of our kernel patches.
3988 ------------------------------------------------------------------------------
3990 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
3992 * Support for kernels:
3993 2.6.9-42.EL (RHEL 4)
3994 2.6.5-7.267 (SLES 9)
3995 2.4.21-40.EL (RHEL 3)
3996 2.6.12.6 vanilla (kernel.org)
4001 Bugzilla : 5719, 9635, 9792, 9684
4002 Description: OST (or MDS) trips assertions in (re)connection under heavy load
4003 Details : If a server is under heavy load and cannot reply to new
4004 connection requests before the client resends the (re)connect,
4005 the connection handling code can behave badly if two service
4006 threads are concurrently handing separate (re)connections from
4007 the same client. Add better locking to the connection handling
4008 code, and ensure that only a single connection will be processed
4009 for a given client UUID, even if the lock is dropped.
4011 Severity : enhancement
4013 Description: add TCP zero-copy support to kernel
4014 Details : Add support to the kernel TCP stack to allow zero-copy bulk
4015 sends if the hardware supports scatter-gather and checksumming.
4016 This allows socklnd to do client-write and server-read more
4017 efficiently and reduce CPU utilization from skbuf copying.
4020 Frequency : only if NFS exporting from client
4022 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
4023 Details : If setting attributes on a file created under NFS that had
4024 never been opened it would be possible to oops the client
4025 if the file had no objects.
4029 Bugzilla : 9326, 10402, 10897
4030 Description: client crash in ptlrpcd_wake() thread when sending async RPC
4031 Details : It is possible that ptlrpcd_wake() dereferences a freed async
4032 RPC. In rare cases the ptlrpcd thread alread processed the RPC
4033 before ptlrpcd_wake() was called and the request was freed.
4036 Frequency : always for liblustre
4038 Description: liblustre client does MDS+OSTs setattr RPC for each write
4039 Details : When doing a write from a liblustre client, the client
4040 incorrectly issued an RPC to the MDS and each OST the file was
4041 striped over in order to update the timestamps. When writing
4042 with small chunks and many clients this could overwhelm the MDS
4043 with RPCs. In all cases it would slow down the write because
4044 these RPCs are unnecessary.
4046 Severity : enhancement
4048 Description: allow number of MDS service threads to be changed at module load
4049 Details : It is now possible to change the number of MDS service threads
4050 running. Adding "options mds mds_num_threads={N}" to the MDS's
4051 /etc/modprobe.conf will set the number of threads for the next
4052 time Lustre is restarted (assuming the "mds" module is also
4053 reloaded at that time). The default number of threads will
4054 stay the same, 32 for most systems.
4059 Description: OST crash if filesystem is unformatted or corrupt
4060 Details : If an OST is started on a device that has never been formatted
4061 or if the filesystem is corrupt and cannot even mount then the
4062 error handling cleanup routines would dereference a NULL pointer.
4067 Description: NULL pointer deref in llap_from_page.
4068 Details : get_cache_page_nowait can return a page with NULL (or otherwise
4069 incorrect) mapping if the page was truncated/reclaimed while it was
4070 searched for. Check for this condition and skip such pages when
4071 doing readahead. Introduce extra check to llap_from_page() to
4072 verify page->mapping->host is non-NULL (so page is not anonymous).
4075 Frequency : Sometimes when using sys_sendfile
4077 Description: "page not covered by a lock" warnings from ll_readpage
4078 Details : sendfile called ll_readpage without right page locks present.
4079 Now we introduced ll_file_sendfile that does necessary locking
4080 around call to generic_file_sendfile() much like we do in
4084 Frequency : with certain MDS communication failures at client mount time
4086 Description: NULL pointer deref after failed client mount
4087 Details : a client connection request may delayed by the network layer
4088 and not be sent until after the PTLRPC layer has timed out the
4089 request. If the client fails the mount immediately it will try
4090 to clean up before the network times out the request. Add a
4091 reference from the request import to the obd device and delay
4092 the cleanup until the network drops the request.
4095 Frequency : occasionally during client (re)connect
4097 Description: assertion failure during client (re)connect
4098 Details : processing a client connection request may be delayed by the
4099 client or server longer than the client connect timeout. This
4100 causes the client to resend the connection request. If the
4101 original connection request is replied in this interval, the
4102 client may trip an assertion failure in ptlrpc_connect_interpret()
4103 which thought it would be the only running connect process.
4106 Frequency : only with obd_echo servers and clients that are rebooted
4108 Description: kernel BUG accessing uninitialized data structure
4109 Details : When running an obd_echo server it did not start the ping_evictor
4110 thread, and when a client was evicted an uninitialized data
4111 structure was accessed. Start the ping_evictor in the RPC
4112 service startup instead of the OBD startup.
4114 Severity : enhancement
4115 Bugzilla : 10193 (patchless)
4116 Description: Remove dependency on various unexported kernel interfaces.
4117 Details : No longer need reparent_to_init, exit_mm, exit_files,
4118 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
4121 Frequency : rare (only users of deprecated and unsupported LDAP config)
4123 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
4124 Details : LDAP apparently contains 'lustreName' attributes instead of
4125 'name'. A simple remapping of the name is sufficient.
4128 Frequency : rare (only with non-default dump_on_timeout debug enabled)
4130 Description: waiting_locks_callback trips kernel BUG if client is evicted
4131 Details : Running with the dump_on_timeout debug flag turned on makes
4132 it possible that the waiting_locks_callback() can try to dump
4133 the Lustre kernel debug logs from an interrupt handler. Defer
4134 this log dumping to the expired_lock_main() thread.
4136 Severity : enhancement
4138 Description: Support NFS exporting on 2.6 kernels.
4139 Details : Implement non-rawops metadata methods for NFS server to use without
4140 changing NFS server code.
4143 Frequency : very rare (synthetic metadata workload only)
4145 Description: two racing renames might cause an MDS thread to deadlock
4146 Details : Running the "racer" program may cause one MDS thread to rename
4147 a file from being the source of a rename to being the target of
4148 a rename at exactly the same time that another thread is doing
4149 so, and the second thread has already enqueued these locks after
4150 doing a lookup of the target and is trying to relock them in
4151 order. Ensure that we don't try to re-lock the same resource.
4154 Frequency : only very large systems with liblustre clients
4156 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
4157 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
4158 all OSTs in parallel. This allows the eviction of stale liblustre
4159 clients to proceed much faster than if they were done in series,
4160 and also offers similar improvements for other set_info RPCs.
4165 Description: excessive CPU usage during initial read phase on client
4166 Details : During the initial read phase on a client, it would agressively
4167 retry readahead on the file, consuming too much CPU and impacting
4168 performance (since 1.4.5.8). Improve the readahead algorithm
4169 to avoid this, and also improve some other common cases (read
4170 of small files in particular, where "small" is files smaller than
4171 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
4176 Description: MDS crash when receiving packet with unknown intent.
4177 Details : Do not LBUG in unknown intent case, just return -EFAULT
4179 Severity : enhancement
4180 Bugzilla : 9293, 9385
4181 Description: MDS RPCs are serialised on client. This is unnecessary for some.
4182 Details : Do not serialize getattr (non-intent version) and statfs.
4185 Frequency : occasional, when OST network is overloaded/intermittent
4187 Description: client evicted by OST after bulk IO timeout
4188 Details : If a client sends a bulk IO request (read or write) the OST
4189 may evict the client if it is unresposive to its data GET/PUT
4190 request. This is incorrect if the network is overloaded (takes
4191 too long to transfer the RPC data) or dropped the OST GET/PUT
4192 request. There is no need to evict the client at all, since
4193 the pinger and/or lock callbacks will handle this, and the
4194 client can restart the bulk request.
4197 Frequency : Always when mmapping file with no objects
4199 Description: client crashes when mmapping file with no objects
4200 Details : Check that we actually have objects in a file before doing any
4201 operations on objects in ll_vm_open, ll_vm_close and
4207 Description: Request leak when working with deleted CWD
4208 Details : Introduce advanced request refcount tracking for requests
4209 referenced from lustre intent.
4211 Severity : Enhancement
4213 Description: Cache open file handles on client.
4214 Details : MDS now will return special lock along with openhandle, if
4215 requested and client is allowed to hold openhandle, even if unused,
4216 until such a lock is revoked. Helps NFS a lot, since NFS is opening
4217 closing files for every read/write openration.
4219 Severity : Enhancement
4221 Description: Cache open negative dentries on client when possible.
4222 Details : Guard negative dentries with UPDATE lock on parent dir, drop
4223 negative dentries on lock revocation.
4228 Description: Remounting a client read-only wasn't possible with a zconf mount
4229 Details : It wasn't possible to remount a client read-only with llmount.
4231 Severity : enhancement
4232 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
4233 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
4234 MPICH is included to add Lustre-specific ADIO interfaces.
4235 This is based closely on the UFS ADIO layer and only differs
4236 in file creation, in order to allow the OST striping to be set.
4237 This is user-contributed code and not supported by CFS.
4242 Description: extended inode attributes (immutable, append-only) work improperly
4243 when 2.4 and 2.6 kernels are used on client/server or vice versa
4244 Details : Introduce kernel-independent values for these flags.
4246 Severity : enhancement
4249 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
4250 Details : Many of the /proc/ tunables can only be tuned at a megabyte
4251 granularity. Now, Fractional MB granularity is be supported,
4252 this is very useful for low memory system.
4254 Severity : enhancement
4256 Description: Getattr by fid
4257 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
4258 locks, avoids extra getattr rpc requests to MDS, allows '/' to
4259 have locks and avoids getattr rpc requests for it on every stat.
4262 Frequency : Always, for filesystems larger than 2TB
4264 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
4265 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
4266 filesystems larger than 2TB to be mounted with mballoc enabled.
4269 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
4271 Description: filesystem corruption for non-standard kernels and very large OSTs
4272 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
4273 filesystems larger than 2TB is mounted then the kernel will
4274 silently corrupt the start of the filesystem. CONFIG_LBD is
4275 enabled for all CFS-supported kernels, but the possibility of
4276 this happening with a modified kernel config exists.
4278 Severity : enhancement
4280 Description: add client O_DIRECT support for 2.6 kernels
4281 Details : It is now possible to do O_DIRECT reads and writes to files
4282 in the Lustre client mountpoint on 2.6 kernel clients.
4284 Severity : enhancement
4286 Description: parallel glimpse, setattr, statfs, punch, destroy requests
4287 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
4288 parallel, not waiting for response from every OST before sending
4289 a rpc to the next OST.
4294 Description: setattr vs write race when updating file timestamps
4295 Details : Client processes that update a file timestamp into the past
4296 right after writing to the file (e.g. tar) it is possible that
4297 the updated file modification time can be reset to the current
4298 time due to a race between processing the setattr and write RPC.
4300 Severity : enhancement
4302 Description: Bring 'lfs find' closer in line with regular Linux find.
4303 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
4304 -print0 options and obtains all the needed info through the lustre
4307 Severity : enhancement
4309 Description: support up to 1024 configured devices on one node
4310 Details : change obd_dev array from statically allocated to dynamically
4311 allocated structs as they are first used to reduce memory usage
4316 Description: Flush dirty partially truncated pages during truncate
4317 Details : Immediatelly flush partially truncated pages in filter_setattr,
4318 this way we completely avoid having any pages in page cache on OST
4319 and can retire ugly workarounds during writes to flush such pages.
4324 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
4325 Details : protect inode from truncation within vfs_unlink() context
4326 just take a reference before calling vfs_unlink() and release it
4327 when parent's i_sem is free.
4330 Frequency : always, if extents are used on OSTs
4332 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
4333 Details : OSTs using the extents format would not zero the high 16 bits of
4334 the index physical block number. This is not a problem for any
4335 OST filesystems smaller than 16TB, and no kernels support ext3
4336 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
4337 new/modified files) and can be fixed for existing filesystems
4338 with e2fsprogs-1.39-cfs1.
4343 Description: import connection selection may be incorrect if timer wraps
4344 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
4345 import connections to be ignored if the 32-bit jiffies counter
4346 wraps. Use a 64-bit jiffies counter.
4349 Frequency : during server recovery
4351 Description: crash after server is denying duplicate export
4352 Details : If clients are resending connect requests to the server, the
4353 server refuses to allow a client to connect multiple times.
4354 Fixed a bug in the handling of this case.
4357 Frequency : very large clusters immediately after boot
4359 Description: LNET request buffers exhausted under heavy short-term load
4360 Details : If a large number of client requests are generated on a service
4361 that has previously never seen so many requests it is possible
4362 that the request buffer growth cannot keep up with the spike in
4363 demand. Instead of dropping incoming requests, they are held in
4364 the LND until the RPC service can accept more requests.
4367 Frequency : Sometimes during replay
4369 Description: Assertion failure in ll_local_open after replay.
4370 Details : If replay happened on an open request reply before we were able
4371 to set replay handler, reply will become not swabbed tripping the
4372 assertion in ll_local_open. Now we set the handler right after
4373 recognising of open request
4376 Frequency : very rare
4378 Description: kernel reports "badness in vsnprintf"
4379 Details : Reading from the "recovery_status" /proc file in small chunks
4380 may cause a negative length in lprocfs_obd_rd_recovery_status()
4381 call to vsnprintf() (which is otherwise harmless). Exit early
4382 if there is no more space in the output buffer.
4384 Severity : enhancement
4386 Description: clear OBD RPC statistics by writing to them
4387 Details : It is now possible to clear the OBD RPC statistics by writing
4388 to the "stats" file.
4393 Description: Client mtime is not the same on different clients after utimes
4394 Details : In some cases, the client was using the utimes() syscall on
4395 a file cached on another node. The clients now validate the
4396 ctime from the MDS + OSTs to determine which one is right.
4401 Description: Inability to activate failout mode
4402 Details : lconf script incorrectly assumed that in python string's numeric
4403 value is used in comparisons.
4406 Frequency : always with multiple stripes per file
4408 Description: Inefficient object allocation for mutli-stripe files
4409 Details : When selecting which OSTs to stripe files over, for files with
4410 a stripe count that divides evenly into the number of OSTs,
4411 the MDS is always picking the same starting OST for each file.
4412 Return the OST selection heuristic to the original design.
4417 Description: mount failures may take full timeout to return an error
4418 Details : Under some heavy load conditions it is possible that a
4419 failed mount can wait for the full obd_timeout interval,
4420 possibly several minutes, before reporting an error.
4421 Instead return an error as soon as the status is known.
4423 ------------------------------------------------------------------------------
4425 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
4427 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4428 INTEROPERATE with older versions automatically. Please read the
4429 user documentation before upgrading any part of a live system.
4430 * WARNING: Lustre networking configuration changes are required with
4431 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4434 * Support for kernels:
4435 2.6.9-22.0.2.EL (RHEL 4)
4436 2.6.5-7.244 (SLES 9)
4437 2.6.12.6 vanilla (kernel.org)
4440 Severity : enhancement
4441 Bugzilla : 7981/8208
4442 Description: Introduced Lustre Networking (LNET)
4443 Details : LNET is new networking infrastructure for Lustre, it includes
4444 a reorganized network configuration mode (see the user
4445 documentation for full details) as well as support for routing
4446 between different network fabrics. Lustre Networking Devices
4447 (LNDs) for the supported network fabrics have also been
4448 created for this new infrastructure.
4450 Severity : enhancement
4451 Description: Introduced Access control lists
4452 Details : clients can set ACLs on files and directories in order to have
4453 more fine-grained permissions than the standard Unix UGO+RWX.
4454 The MDS must be started with the "-o acl" mount option.
4456 Severity : enhancement
4457 Description: Introduced filesystem quotas
4458 Details : Administrators may now establish per-user quotas on the
4461 Severity : enhancement
4463 Description: Configuration change for the XT3
4464 The PTLLND is now used to run Lustre over Portals on the XT3
4465 The configure option(s) --with-cray-portals are no longer used.
4466 Rather --with-portals=<path-to-portals-includes> is used to
4467 enable building on the XT3. In addition to enable XT3 specific
4468 features the option --enable-cray-xt3 must be used.
4473 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4474 Details : A race between allocating a new llcd and re-getting the llcd_lock
4475 allowed another thread to grab newly-allocated llcd.
4477 Severity : enhancement
4479 Description: 2.6 OST async journal commit and locking fix to improve performance
4480 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4481 2.6 kernels are now async as they already were in 2.4 kernels so
4482 that they can commit concurrently with the network bulk transfer.
4483 For block-allocated files the filter allocation semaphore is held
4484 to avoid filesystem fragmentation during allocation. BKL lock
4485 removed for 2.6 xattr operations where it is no longer needed.
4490 Description: lconf incorrectly determined whether two IP networks could talk
4491 Details : In some more complicated routing and multiple-network
4492 configurations, lconf will avoid trying to make a network
4493 connection to a disjoint part of the IP space. It was doing the
4494 math incorrectly for one set of cases.
4499 Description: Fix for potential infinite loop processing records in an llog.
4500 Details : If an llog record is corrupted/zeroed, it is possible to loop
4501 forever in llog_process(). Validate the llog record length
4502 and skip the remainder of the block on error.
4505 Frequency : occasional (liblustre only)
4507 Description: liblustre could not open files whose last component is a symlink
4508 Details : sysio_path_walk() would incorrectly pass the open intent to
4509 intermediate path components.
4512 Frequency : rare (liblustre only with non-standard tuning)
4513 Bugzilla : 7201 (7350)
4514 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4515 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4516 after it has been released, only do this for referenced locks
4518 Severity : enhancement
4520 Description: specifying an (invalid) directory default stripe_size of -1
4521 would reset the directory default striping
4522 Details : stripe_size -1 was used internally to signal directory stripe
4523 removal, now use "all default" to signal dir stripe removal
4524 as a directory striping of "all default" is not useful
4527 Frequency : common for large clusters running liblustre clients
4529 Description: doing an ls when liblustre clients are running is slow
4530 Details : sending a glimpse AST to a liblustre client waits for every AST
4531 to time out, as liblustre clients will not respond. Since they
4532 cannot cache data we refresh the OST lock LVB from disk instead.
4534 Severity : enhancement
4536 Description: doing an ls at the same time as file IO can be slow
4537 Details : enqueue and other "small" requests can be blocked behind many
4538 large IO requests. Create a new OST IO portal for non-IO
4539 requests so they can be processed faster.
4542 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4544 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4545 created file with mode 000
4546 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4547 allow an ftruncate() as a non-root user to a file with mode 000.
4548 Lustre now respects this flag to disable mode checks when
4549 truncating a file owned by the user
4552 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4555 Description: Revoking locks from clients that went dead or catatonic might take
4557 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4558 cancellation of such locks instant on servers without waiting for
4559 any reply from clients. Clients drops these locks when cancel
4560 notification from server is received without replying.
4563 Frequency : liblustre-only, when liblustre client dies or becomes busy
4565 Description: Doing ls on Linux clients can take a long time with active
4567 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4568 granting such locks to it in the first place if possible. Locks
4569 are taken by proxy on the OST during the read or write and
4570 dropped immediately afterward. Add connect flags handling, do
4571 not grant locks to liblustre clients for glimpse ASTs.
4573 Severity : enhancement
4575 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4576 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4577 sequential reads and behaves badly with applications doing random
4578 reads. Improve it by reading ahead at least read region, and
4579 avoiding excessive large RPC for small reads.
4581 Severity : enhancement
4583 Description: Creating more than 1000 files for a single job may cause a load
4584 imbalance on the OSTs if there are also a large number of OSTs.
4585 Details : qos_prep_create() uses an OST index reseed value that is an
4586 even multiple of the number of available OSTs so that if the
4587 reseed happens in the middle of the object allocation it will
4588 still utilize the OSTs as uniformly as possible.
4593 Description: OST or MDS may oops in ping_evictor_main()
4594 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4595 but doesn't restart at beginning of obd_exports_timed list
4598 Severity : enhancement
4600 Description: improve by-nid export eviction on the MDS and OST
4601 Details : allow multiple exports with the same NID to be evicted at one
4602 time without re-searching the exports list.
4605 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4607 Description: MDS may oops in groups_free()
4608 Details : in rare race conditions a newly allocated group_info struct is
4609 freed again, and this can be NULL. The 2.4 compatibility code
4610 for groups_free() checked for a NULL pointer, but 2.6 did not.
4613 Frequency : common for liblustre clients doing little filesystem IO
4614 Bugzilla : 9352, 7313
4615 Description: server may evict liblustre clients accessing contended locks
4616 Details : if a client is granted a lock or receives a completion AST
4617 with a blocking AST already set it would not reply to the AST
4618 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4621 Frequency : lfs setstripe, only systems with more than 160 OSTs
4623 Description: unable to set striping with a starting offset beyond OST 160
4624 Details : llapi_create_file() incorrectly limited the starting stripe
4625 index to the maximum single-file stripe count.
4628 Frequency : LDAP users only
4630 Description: lconf did not handle in-kernel recovery with LDAP properly
4631 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4633 Severity : enhancement
4635 Description: bind OST threads to NUMA nodes to improve performance
4636 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4637 node and do their allocations there to localize memory access
4639 Severity : enhancement
4641 Description: llmount can determine client NID directly from Myrinet (GM)
4642 Details : the client NID code from gmnalnid was moved directly into
4643 llmount, removing the need to use this or specifying the
4644 client NID explicitly when mounting GM clients with zeroconf
4647 Frequency : if client is started with down MDS
4649 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4650 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4651 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4652 ptlrpc_import_delayed_req() because we didn't check if the
4653 request was interrupted, and we also didn't break out of the
4654 event loop if there was no timeout
4659 Description: data loss during non-page-aligned writes to a single file from
4660 both multiple nodes and multiple threads on one node at same time
4661 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4662 inconsistency led to false short-reads, that were cached and later
4663 used by ->prepare_write() to fill in partially written page,
4664 leading to data loss.
4667 Frequency : always, if lconf --abort_recovery used
4669 Description: lconf --abort_recovery fails with 'Operation not supported'
4670 Details : lconf was attempting to abort recovery on the MDT device and not
4673 Severity : enhancement
4675 Description: remove cleanup logs
4676 Details : replace lconf-generated cleanup logs with lustre internal
4677 cleanup routines. Eliminates the need for client-cleanup and
4680 Severity : enhancement
4682 Description: add support for EAs (user and system) on lustre filesystems
4683 Details : it is now possible to store extended attributes in the Lustre
4684 client filesystem, and with the user_xattr mount option it
4685 is possible to allow users to store EAs on their files also
4687 Severity : enhancement
4689 Description: Add possibility (config option) to show minimal available OST free
4691 Details : When compiled with --enable-mindf configure option, statfs(2)
4692 (and so, df) will return least minimal free space available from
4693 all OSTs as amount of free space on FS, instead of summary of
4694 free spaces of all OSTs.
4696 Severity : enhancement
4698 Description: do not expand extent locks acquired on OST-side
4699 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4700 by server: they are not cached anyway.
4703 Frequency : when mmap is used/binaries executed from Lustre
4705 Description: Unmmap pages before throwing them away from read cache.
4706 Details : llap_shrink cache now attempts to unmap pages before discarding
4707 them (if unmapping failed - do not discard). SLES9 kernel has
4708 extra checks that trigger if this unmapping is not done first.
4713 Description: lconf didn't resolve symlinks before checking to see whether a
4714 given mountpoint was already in use
4717 Frequency : when migrating failover services
4718 Bugzilla : 6395, 9514
4719 Description: When migrating a subset of services from a node (e.g. failback
4720 from a failover service node) the remaining services would
4721 time out and evict clients.
4722 Details : lconf --force (implied by --failover) sets the global obd_timeout
4723 to 5 seconds in order to quickly disconnect, but this caused
4724 other RPCs to time out too quickly. Do not change the global
4725 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4727 Severity : enhancement
4728 Frequency : if MDS is started with down OST
4729 Bugzilla : 9439,5706
4730 Description: Allow startup/shutdown of an MDS without depending on the
4731 availability of the OSTs.
4732 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4733 Add appropriate locking and lov-osc refcounts for safe
4734 cleaning. Add osc abort_inflight calls in case the
4735 synchronize never started.
4738 Frequency : occasional (Cray XT3 only)
4740 Description: root not authorized to access files in CRAY_PORTALS environment
4741 Details : The client process capabilities were not honoured on the MDS in
4742 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4743 been accessed by an authorized user then root was able to access
4744 the file on the local client also. The root user capabilities
4745 are now allowed on the MDS, as this environment has secure UID.
4748 Frequency : occasional
4750 Description: ldiskfs "too long searching" message happens too often
4751 Details : A debugging message (otherwise harmless) prints too often on
4752 the OST console. This has been reduced to only happen when
4753 there are fragmentation problems on the filesystem.
4758 Description: Division by zero in statfs when all OSCs are inactive
4759 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4760 lov_statfs divides by value returned by lov_get_stripecnt().
4764 Bugzilla : 9489, 3273
4765 Description: First write from each client to each OST was only 4kB in size,
4766 to initialize client writeback cache, which caused sub-optimal
4767 RPCs and poor layout on disk for the first writen file.
4768 Details : Clients now request an initial cache grant at (re)connect time
4769 and so that they can start streaming writes to the cache right
4770 away and always do full-sized RPCs if there is enough data.
4771 If the OST is rebooted the client also re-establishes its grant
4772 so that client cached writes will be honoured under the grant.
4777 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4778 Details : Now I/O RPCs go to different portal number and (presumably) fast
4779 lock requests (and glimses) and other RPCs get their own service
4780 threads pool that should be able to service those RPCs
4783 Severity : enhancement
4785 Description: Ability to exchange lustre version between client and servers and
4786 issue warnings at client side if client is too old. Also for
4787 liblustre clients there is ability to refuse connection of too old
4789 Details : New 'version' field is added to connect data structure that is
4790 filled with version info. That info is later checked by server and
4794 Frequency : rare, liblustre only.
4795 Bugzilla : 9296, 9581
4796 Description: Two simultaneous writes from liblustre at offset within same page
4797 might proceed at the same time overwriting eachother with stale
4799 Details : I/O lock withing llu_file_prwv was released too early, before data
4800 actually was hitting the wire. Extended lock-holding time until
4801 server acknowledges receiving data.
4804 Frequency : extremely rare. Never observed in practice.
4806 Description: avoid generating lustre_handle cookie of 0.
4807 Details : class_handle_hash() generates handle cookies by incrementing
4808 global counter, and can hit 0 occasionaly (this is unlikely, but
4809 not impossible, because initial value of cookie counter is
4810 selected randonly). Value of 0 is used as a sentinel meaning
4811 "unassigned handle" --- avoid it. Also coalesce two critical
4812 sections in this function into one.
4814 Severity : enhancement
4816 Description: allow liblustre clients to delegate truncate locking to OST
4817 Details : To avoid overhead of locking, liblustre client instructs OST to
4818 take extent lock in ost_punch() on client's behalf. New connection
4819 flag is added to handle backward compatibility.
4821 Severity : enhancement
4822 Bugzilla : 4928, 7341, 9758
4823 Description: allow number of OST service threads to be specified
4824 Details : a module parameter allows the number of OST service threads
4825 to be specified via "options ost ost_num_threads={N}" in the
4826 OSS's /etc/modules.conf or /etc/modprobe.conf.
4830 Bugzilla : 6146, 9635, 9895
4831 Description: servers crash with bad pointer in target_handle_connect()
4832 Details : In rare cases when a client is reconnecting it was possible that
4833 the connection request was the last reference for that export.
4834 We would temporarily drop the export reference and get a new
4835 one, but this may have been the last reference and the export
4836 was just destroyed. Get new reference before dropping old one.
4838 Severity : enhancement
4839 Frequency : if client is started with failover MDS
4841 Description: Allow multiple MDS hostnames in the mount command
4842 Details : Try to read the configuration from all specified MDS
4843 hostnames during a client mount in case the "primary"
4846 Severity : enhancement
4848 Description: Stop sending data to evicted clients as soon as possible.
4849 Details : Check if the client we are about to send or are sending data to
4850 was evicted already. (Check is done every second of waiting,
4851 for which l_wait_event interface was extended to allow checking
4852 of exit condition at specified intervals).
4855 Frequency : rare, normally only when NFS exporting is done from client
4857 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4859 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4862 Frequency : always, if srand() is called before liblustre initialization
4864 Description: Liblustre uses system PRNG disturbing its usage by user application
4865 Details : Introduce internal to lustre fast and high-quality PRNG for
4866 lustre usage and make liblustre and some other places in generic
4867 lustre code to use it.
4869 Severity : enhancement
4870 Bugzilla : 9477, 9557, 9870
4871 Description: Verify that the MDS configuration logs are updated when xml is
4872 Details : Check if the .xml configuration logs are newer than the config
4873 logs stored on the MDS and report an error if this is the case.
4874 Request --write-conf, or allow starting with --old_conf.
4876 Severity : enhancement
4878 Description: Handle symlinks in the path when checking if Lustre is mounted.
4879 Details : Resolve intermediate symlinks when checking if a client has
4880 mounted a filesystem to avoid duplicate client mounts.
4885 Description: lconf can hit an error exception but still return success.
4886 Details : The lconf command catches the Command error exception at the top
4887 level script context and will exit with the associated exit
4888 status, but doesn't ensure that this exit status is non-zero.
4893 Description: failure of ptlrpc thread startup can cause oops
4894 Details : Starting a ptlrpc service thread can fail if there are a large
4895 number of threads or the server memory is very fragmented.
4896 Handle this without oopsing.
4899 Frequency : always, only if liblustre and non-default acceptor port was used
4901 Description: liblustre cannot connect to servers with non-default acceptor port
4902 Details : tcpnal_set_default_params() was not called and was therefore
4903 ignoring the environment varaible TCPNAL_PORT, as well as other
4904 TCPNAL_ environment variables
4909 Description: two objects could be created on the same OST for a single file
4910 Details : If an OST is down, in some cases it was possible to create two
4911 objects on a single OST for a single file. No problems other
4912 than potential performance impact and spurious error messages.
4916 Bugzilla : 5681, 9562
4917 Description: Client may oops in ll_unhash_aliases
4918 Details : Client dcache may become inconsistent in race condition.
4919 In some cases "getcwd" can fail if the current directory is
4925 Description: Inode refcounting problems in NFS export code
4926 Details : link_raw functions used to call d_instantiate without obtaining
4927 extra inode reference first.
4931 Bugzilla : 9942, 9903
4932 Description: Referencing freed requests leading to crash, memleaks with NFS.
4933 Details : We used to require that call to ll_revalidate_it was always
4934 followed by ll_lookup_it. Also with revalidate_special() it is
4935 possible to call ll_revalidate_it() twice for the same dentry
4936 even if first occurence returned success. This fix changes semantic
4937 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4938 reference on a request referred from the intent.
4939 ll_intent_release() then releases such a request.
4942 Frequency : rare, normally benchmark loads only
4944 Description: unlinked inodes were kept in memory on the client
4945 Details : If a client is repeatedly creating and unlinking files it
4946 can accumulate a lot of stale inodes in the inode slab cache.
4947 If there is no other client load running this can cause the
4948 client node to run out of memory. Instead flush old inodes
4949 from client cache that have the same inode number as a new inode.
4952 Frequency : SLES9 2.6.5 kernel and long filenames only
4953 Bugzilla : 9969, 10379
4954 Description: utime reports stale NFS file handle
4955 Details : SLES9 uses out-of-dentry names in some cases, which confused
4956 the lustre dentry revalidation. Change it to always use the
4960 Frequency : rare, unless heavy write-truncate concurrency is continuous
4961 Bugzilla : 4180, 6984, 7171, 9963, 9331
4962 Description: OST becomes very slow and/or deadlocked during object unlink
4963 Details : filter_destroy() was holding onto the parent directory lock
4964 while truncating+unlinking objects. For very large objects this
4965 may block other threads for a long time and slow overall OST
4966 responsiveness. It may also be possible to get a lock ordering
4967 deadlock in this case, or run out of journal credits because of
4968 the combined truncate+unlink. Solution is to do object truncate
4969 first in one transaction without parent lock, and then do the
4970 final unlink in a new transaction with the parent lock. This
4971 reduces the lock hold time dramatically.
4974 Frequency : rare, 2.4 kernels only
4976 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4977 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4978 before unmounting their filesystem in order to avoid deadlock.
4979 The kernel_locked() function in 2.4 kernels only checks whether
4980 the kernel lock is held, not whether it is this process that is
4981 holding it as 2.6 kernels do.
4986 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4987 Details : The client ptlrpc code may be trying to reconnect to a down
4988 server before a previous connection attempt has timed out.
4989 Increase the reconnect interval to be longer than the connection
4990 timeout interval to avoid sending duplicate connections to
4994 Frequency : echo_client brw_test command
4996 Description: fix echo_client to work with OST preallocated code
4997 Details : OST preallocation code (5137) didn't take echo_client IO path
4998 into account: echo_client calls filter methods outside of any
4999 OST thread and, hence, there is no per-thread preallocated
5000 pages and buffers to use. Solution: hijack pga pages for IO. As
5001 a byproduct, this avoids unnecessary data copying.
5005 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
5006 Description: Client can oops in mdc_commit_close() after open replay
5007 Details : It was possible for the MDS to return an open request with no
5008 transaction number in mds_finish_transno() if the client was
5009 evicted, but without actually returning an error. Clients
5010 would later try to replay that open and may trip an assertion
5011 Simplify the client close codepath, and always return an error
5012 from the MDS in case the open is not successful.
5015 Frequency : rare, 2.6 OSTs only
5017 Description: OST may deadlock under high load on fragmented files
5018 Details : If there was a heavy load and highly-fragmented OST filesystems
5019 it was possible to have all the OST threads deadlock waiting on
5020 allocation of biovecs, because the biovecs were not released
5021 until the entire RPC IO was completed. Instead, release biovecs
5022 as soon as they are complete to ensure forward IO progress.
5024 Severity : enhancement
5026 Description: Support for specifying external journal device at mount
5027 Details : If an OST or MDS device is formatted with an external journal
5028 device, this device major/minor is stored in the ext3 superblock
5029 and may not be valid for failover. Allow detecting and
5030 specifying the external journal at mount time.
5035 Description: Mounting an MDS with pending unlinked files may cause oops
5036 Details : target_finish_recovery() calls mds_postrecov() which returned
5037 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
5038 considers this an error and immediately begins cleaning up the
5039 lov, just after starting the mds_lov process
5041 Severity : enhancement
5043 Description: Implement 'lfs df' to report actual free space on per-OST basis
5044 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
5045 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
5046 usage of objects; '-h' to report in human readable format.
5048 ------------------------------------------------------------------------------
5050 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
5057 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
5058 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
5059 at startup during orphan unlinking. Instead, skip buddy update
5060 before setup, as it will be regenerated after recovery is complete.
5065 Description: If an OST is inactive, its locks might reference stale inodes.
5066 Details : lov_change_cbdata() must iterate over all namespaces, even if
5067 they are inactive to clear inode references from the lock.
5069 Severity : enhancement
5070 Frequency : occasional, if non-standard max_dirty_mb used
5072 Description: Client will block write RPCs if not enough grant
5073 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
5074 then the client will block writes while waiting for another RPC
5075 to complete instead of consuming its dirty limit. With change
5076 we get improved performance when max_dirty_mb is small.
5078 Severity : enhancement
5079 Bugzilla : 3389, 6253
5080 Description: Add support for supplementary groups on the MDS.
5081 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
5082 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
5083 lookups for user supplementary groups into a cache.
5087 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
5088 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
5089 for liblustre O_EXCL is enforced by MDS.
5093 Description: readdir never returns NULL in liblustre.
5094 Details : Corrected llu_iop_getdirentries logic, to return offset of next
5095 dentry in struct dirent.
5099 Frequency : liblustre only, depends on application IO pattern
5100 Description: liblustre clients evicted if not contacting servers
5101 Details : Don't put liblustre clients into the ping_evictor list, so
5102 they will not be evicted by the pinger ever.
5104 Severity : enhancement
5106 Description: Add ability to evict clients by NID from MDS.
5107 Details : By echoing "nid:$NID" string into
5108 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
5109 $NID would be instantly evicted from this MDS and from all active
5110 OSTs connected to it.
5114 Description: Do not query file size twice, somewhat slowing stat(2) calls.
5115 Details : lookup_it_finish() used to query file size from OSTs that was not
5120 Description: service threads change working directory to that of init
5121 Details : Starting lustre service threads may pin the working directory
5122 of the parent thread, making that filesystem busy. Threads
5123 now change to the working directory of init to avoid this.
5127 Frequency : during shutdown only
5128 Description: shutdown with a failed MDS or OST can cause unmount to hang
5129 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
5134 Frequency : 2.6 only
5135 Description: chmod/chown may include an extra supplementary group
5136 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
5137 supplementary group and if none is specified this is used.
5140 Bugzilla : 5479 (6816)
5142 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
5143 Details : If lookup is in progress on a file that is unlinked we might try
5144 to revalidate the inode and fail in revalidate after lookup is
5145 complete and ll_file_open() enqueues the open again but
5146 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
5149 Frequency : always, if lconf --abort_recovery used
5151 Description: lconf --abort_recovery fails with 'Operation not supported'
5152 Details : lconf was attempting to abort recovery on the MDT device and not
5155 ------------------------------------------------------------------------------
5157 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
5162 Frequency : rare (only unsupported configurations with a node running as an
5164 Bugzilla : 6514, 5137
5165 Description: Mounting a Lustre file system on a node running as an OST could
5167 Details : OSTs now preallocates memory needed to write out data at
5168 startup, instead of when needed, to avoid having to
5169 allocate memory in possibly low memory situations.
5170 Specifically, if the file system is mounted on on OST,
5171 memory pressure could force it to try to write out data,
5172 which it needed to allocate memory to do. Due to the low
5173 memory, it would be unable to do so and the node would
5174 become unresponsive.
5176 Severity : enhancement
5178 Description: Addition of lconf --service command line option
5179 Details : lconf now accepts a '--service <arg>' option, which is
5180 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
5182 Severity : enhancement
5184 Description: Failover mode is now the default for OSTs.
5185 Details : By default, OSTs will now run in failover mode. To return to
5186 the old behaviour, add '--failout' to the lmc line for OSTs.
5188 Severity : enhancement
5190 Description: Health checks are now provided for MDS and OSTs
5191 Details : Additional detailed health check information on MSD and OSTs
5192 is now provided through the procfs health_check value.
5195 Frequency : occasional, depends on IO load
5197 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
5198 after numerous create/delete cycles
5199 Details : The ext3 inode allocation policy would not allocate new inodes
5200 very well on the OSTs because there are no new directories
5201 being created. Instead we look for groups with free space if
5202 the parent directories are nearly full.
5207 Description: Network or server problems during mount may cause partially
5208 mounted clients instead of returning an error.
5209 Details : The config llog parsing code may overwrite the error return
5210 code during mount error handling, returning success instead
5216 Description: MDS can fail to allocate large reply buffers
5217 Details : After long uptimes the MDS can fail to allocate large reply
5218 buffers (e.g. zconf client mount config records) due to memory
5219 fragmentation or consumption by the buffer cache. Preallocate
5220 some large reply buffers so that these replies can be sent even
5221 under memory pressure.
5225 Frequency : rare (liblustre)
5226 Description: fsx running with liblustre complained that using truncate() to
5227 extend the file doesn't work. This patch corrects that issue.
5228 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
5229 ATTR_SIZE and lsm use in llu_setattr_raw.
5233 Frequency : rare, only 2.6 kernels
5234 Description: Unusual file access patterns on the MDS may result in inode
5235 data being lost in very rare circumstances.
5236 Details : Bad interaction between the ea-in-inode patch and the "no-read"
5237 code in the 2.6 kernel caused the inode and/or EA data not to
5238 be read from disk, causing single-file corruption.
5242 Frequency : rare, only 2.6 filesystems using extents
5243 Description: Heavy concurrent write and delete load may cause data corruption.
5244 Details : It was possible under high-load situations to have an extent
5245 metadata block in the block device cache from a just-unlinked
5246 file overwrite a newly-allocated data block. We now unmap any
5247 metadata buffers that alias just-allocated data blocks.
5251 Frequency : filesystems with default stripe_count larger than 77
5252 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
5253 Details : lconf specifies an inode size of 4096 bytes when the default
5254 stripe_count is larger than 77. This conflicts with the default
5255 inode density of 1 per 4096 bytes. Allocate smaller inodes in
5256 this case to avoid pinning too much memory for large EAs.
5258 ------------------------------------------------------------------------------
5260 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
5265 Frequency : rare (extremely heavy IO load with hundreds of clients)
5267 Description: Client is evicted, gets IO error writing to file
5268 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
5269 caused clients to be evicted for AST timeouts. The fixes in
5270 bug 5192 mean we no longer need to have such short AST timeouts
5271 so ldlm_timeout has been increased.
5274 Frequency : occasional during --force or --failover shutdown under load
5275 Bugzilla : 5949, 4834
5276 Description: Server oops/LBUG if stopped with --force or --failover under load
5277 Details : a collection of import/export refcount and cleanup ordering
5278 issues fixed for safer force cleanup
5281 Frequency : only filesystems larger than 120 OSTs
5282 Bugzilla : 5990, 6223
5283 Description: lfs getstripe would oops on a very large filesystem
5284 Details : lov_getconfig used kfree on vmalloc'd memory
5287 Frequency : only filesystems exporting via NFS to Solaris 10 clients
5288 Bugzilla : 6242, 6243
5289 Description: reading from files that had been truncated to a non-zero size
5290 but never opened returned no data
5291 Details : ll_file_read() reads zeros from no-object files to EOF
5296 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
5297 Details : There's little harm in aborting MDS/OSS recovery and letting it
5298 try again, so I removed the LASSERT and return an error instead.
5300 Severity : enhancement
5302 Description: New debugging infrastructure for tracking down data corruption
5303 Details : The I/O checksum code was replaced to: (a) control it at runtime,
5304 (b) cover more of the client-side code path, and (c) try to narrow
5305 down where problems occurred
5309 Bugzilla : 3819, 4364, 4397, 6313
5310 Description: Racing close and eviction MDS could cause assertion in mds_close
5311 Details : It was possible to get multiple mfd references during close and
5312 client eviction, leading to one thread referencing a freed mfd.
5314 Severity: : enhancement
5315 Bugzilla : 3262, 6359
5316 Description: Attempts to reconnect to servers are now more aggressive.
5317 Details : This builds on the enhanced upcall-less recovery that was added
5318 in 1.4.2. When trying to reconnect to servers, clients will
5319 now try each server in the failover group every 10 seconds. By
5320 default, clients would previously try one server every 25 seconds.
5325 Description: After recovery, certain operations trigger a failed
5326 assertion on a client.
5327 Details : Failing over an mds, using lconf -d --failover, while a
5328 client was doing a readdir() call would cause the client to
5329 LBUG after recovery completed and the readdir() was resent.
5331 Severity : enhancement
5333 Description: Default groups are now added by lconf
5334 Details : You can now run lconf --group <servicename> without having to
5335 manually add groups with lmc.
5338 Frequency : occasional
5340 Description: Nodes with an elan id of 0 trigger a failed assertion
5343 Frequency : always when accessing e.g. tty/console device nodes
5345 Description: tty and some other devices nodes cannot be used on lustre
5346 Details : file's private_data field is used by device data and lustre
5347 values in there got lost. New field was added to struct file to
5348 store fs-specific private data.
5351 Frequency : when exporting Lustre via NFS
5353 Description: NFSD failed occasionally when looking up a path component
5354 Details : NFSD is looking up ".." which was broken in ext3 directories
5355 that had grown large enough to become hashed.
5358 Frequency : Clusters with multiple interfaces not on the same subnet
5360 Description: Nodes will repeatedly try to reconnect to an interface which it
5361 cannot reach and report an error to the log.
5362 Details : Extra peer list entries will be created by lconf with some peers
5363 unreachable. lconf now validates the peer before adding it.
5366 Frequency : Only if a default stripe is set on the filesystem root.
5368 Description: Setting a default stripe on the filesystem root prevented the
5369 filesystem from being remounted.
5370 Details : The client was sending extra request flags in the root getattr
5371 request and did not allocate a reply buffer for the dir EA.
5374 Frequency : occasional, higher if lots of files are accessed by one client
5375 Bugzilla : 6159, 6097
5376 Description: Client trips assertion regarding lsm mismatch/magic
5377 Details : While revalidating inodes the VFS looks up inodes with ifind()
5378 and in rare cases can find an inode that is being freed.
5379 The ll_test_inode() code will free the lsm during ifind()
5380 when it finds an existing inode and then the VFS later attaches
5381 this free lsm to a new inode.
5385 Bugzilla : 6422, 7030
5386 Description: MDS deadlock between mkdir and client eviction
5387 Details : Creating a new file via mkdir or mknod (starting a transaction
5388 and getting the ns lock) can deadlock with client eviction
5389 (gets ns lock and trying to finish a synchronous transaction).
5392 Frequency : occasional
5393 Description: While starting a server, the fsfilt_ext3 module could not be
5395 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5396 kernels. Previously, lconf would still use the ext3 name
5397 when trying to load modules. Now, it will correctly use
5398 ext3 on 2.4 and ldiskfs on 2.6.
5400 Severity : enhancement
5401 Description: The default stripe count has been changed to 1
5402 Details : The interpretation of the default stripe count (0, to lfs
5403 or lmc) has been changed to mean striping across a single
5404 OST, rather than all available. For general usage we have
5405 found a stripe count of 1 or 2 works best.
5407 Severity : enhancement
5408 Description: Add support for compiling against Cray portals.
5409 Details : Conditional compiling for some areas that are different
5413 Frequency : occasional
5414 Bugzilla : 6409, 6834
5415 Description: Creating files with an explicit stripe count may lead to
5416 a failed assertion on the MDS
5417 Details : If some OSTs are full or unavailable, creating files may
5418 trigger a failed assertion on the MDS. Now, Lustre will
5419 try to use other servers or return an error to the
5423 Frequency : occasional
5425 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5426 caused lock timeout during readahead (since 1.4.2).
5427 Details : Processes doing readahead might match a lock that hasn't been
5428 granted yet if there are overlapping and conflicting lock
5429 requests. The readahead process waits on ungranted lock
5430 (original lock is CBPENDING), while OST waits for that process
5431 to cancel CBPENDING read lock and eventually evicts client.
5433 Severity : enhancement
5435 Description: Initial enabling of flock support for clients
5436 Details : Implements fcntl advisory locking and file status functions.
5437 This feature is provided as an optional mount flag (default
5438 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5439 locking are implemented yet, and those that are are not guaranteed
5440 to be completely correct in production environments.
5441 mount -t lustre -o [flock|noflock] ...
5444 Frequency : occasional
5446 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5447 assertion in the ext3 JBD (journaling) layer.
5448 Details : The b_committed_data struct is protected by the big kernel lock
5449 in 2.4 kernels, serializing journal_commit_transaction() and
5450 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5451 access to this struct. In 2.6 kernels there is finer grained
5452 locking to improve SMP performance of the JBD layer.
5456 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5458 -----------------------------------------------------------------------------
5460 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5462 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5463 versions of Lustre. Please update all servers and clients to
5464 version 1.4.2 or later at the same time. You must also run
5465 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5466 to update the configuration logs.
5468 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5469 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5470 - swab llog records of type '0' so we get proper header size/idx (5861)
5471 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5472 - fix rename of one directory over another leaking an inode (5953)
5473 - avoid SetPageDirty on 2.6 (5981)
5474 - don't re-add just-being-destroyed locks to the waiting list (5653)
5475 - when creating new directories, inherit the parent's custom
5476 striping settings if present parent (3048)
5477 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5478 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5479 - don't hold client locks on temporary worklist from l_lru (5666)
5480 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5481 - automatically evict dead clients (5921)
5482 - Update file size properly in create+truncate+fstat case (6196)
5483 - Do not unhash mountpoint dentries, do not allow removal of
5485 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5486 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5487 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5488 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5489 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5490 - Correctly handle reads of files with no objects (6243)
5491 - lctl recover will also mark a device active if deactivate used (5933)
5493 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5494 - allow --write-conf on an MDS with different nettype than client (5619)
5495 - don't write config llogs to MDS for mounts not from that MDS (5617)
5496 - lconf should create multiple TCP connections from a client (5201)
5497 - init scripts are now turned off by default; run chkconfig --on
5498 lustre and chkconfig --on lustrefs to use them
5499 - upcalls are no longer needed for clients to recover to failover
5501 - add --abort-recovery option to lconf to abort recovery on device
5503 - add support for an arbitrary number of OSTs (3026)
5504 - Quota support protocol changes.
5505 - forward compatibility changes to wire structs (6007)
5506 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5507 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5508 - improved "lustre status" script
5509 - initialize blocksize for non-regular files (6062)
5510 - added --disable-server and --disable-client configure options (5782)
5511 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5512 - Vanilla 2.4.29 support
5513 - increase maximum number of obd devices to 520 (6242)
5514 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5515 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5516 SLES 9 2.6.5 kernels
5517 - limit stripes per file to 160 (the maximum EA size) (6093)
5519 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5522 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5523 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5524 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5525 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5526 - small ext3 extents cleanups and fixes (5733)
5527 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5528 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5529 - avoid needless client->OST connect, fix handle mismatch (5317)
5530 - fix DLM error path that led to out-of-sync client, long delays (5779)
5531 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5532 - fix several locking issues related to i_size (5492,5624,5654,5672)
5533 - don't move pending lock onto export if it is already evicted (5683)
5534 - fix kernel oops when creating .foo in unlinked directory (5548)
5535 - fix deadlock in obdfilter statistics vs. object create (5811)
5536 - use time_{before,after} to avoid timer jiffies wrap (5882)
5537 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5538 - Do not leak request if server was not able to process it (5154)
5539 - If mds_open unable to find parent dir, make that negative lookup(5154)
5540 - don't create new directories with extent-mapping (5909, 5936)
5542 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5543 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5544 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5545 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5546 - support for RHEL4 kernel on i686 (5773)
5547 - provide error messages when incompatible logs are encountered (5898)
5549 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5550 * version 1.4.0.10 (1.4.1 release candidate 1)
5552 - don't keep a lock reference when lock is not granted (4238)
5553 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5554 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5555 - drop import inflight refcount on signal_completed_replay error (5255)
5556 - unlock page after async write error during send (3677)
5557 - handle missing objects in filter_preprw_read properly (5265)
5558 - no transno return for symlink open, don't save no-trasno open (3440)
5559 - don't try to complete elan receive that already failed (4012)
5560 - free RPC server reply state on error (5406)
5561 - clean up thread from ptlrpc_start_thread() on error (5160)
5562 - readahead could read extra page into cache that wasn't ejected (5388)
5563 - prevent races in class_attach/setup/cleanup/detach (5260)
5564 - don't dereference de->d_inode after l_dput of de (5458)
5565 - use "int" for stripe value returned from lock_to_stripe (5544)
5566 - mballoc allocation and error-checking fixes in 2.6 (5504)
5567 - block device patches to fix I/O request sizes in 2.6 (5482)
5568 - look up hostnames for IB nals (5602)
5569 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5570 - don't start multiple acceptors for the same port (5277)
5571 - fix incorrect LASSERT in mds_getattr_name (5635)
5572 - export a proc file for general "ping" checking (5628)
5573 - fix "lfs check" to not block when the MDS is down (5628)
5575 - service request history (4965)
5576 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5577 - create an "evict_client" /proc entry on OSTs, like the MDS has
5578 - fix mount usage message, return errors per mount(8) (5168)
5579 - change grep [] to grep "[]" in tests so they work in more UMLs
5580 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5581 - remove ancient LOV_MAGIC_V0 EA support (5047)
5582 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5583 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5584 - allow readahead allocations to fail when low on memory (5383)
5585 - mmap locking landed again, after considerable improvement (2828)
5586 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5588 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5591 - send OST transaction number in read/write reply to free req (4966)
5592 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5593 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5594 - the watchdog thread now runs as interruptible (5246)
5595 - flock/lockf fixes (but it's still disabled, pending 5135)
5596 - don't use EXT3 constants in llite code (5094)
5597 - memory shortage at startup could cause assertion (5176)
5599 - reorganization of lov code
5600 - single portals codebase
5602 - add extents/mballoc support (5025)
5603 - direct I/O reads in the obdfilter (4048)
5604 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5606 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5609 - send OST transaction number in read/write reply to free req (4966)
5610 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5611 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5612 - the watchdog thread now runs as interruptible (5246)
5613 - handle missing objects in filter_preprw_read properly (5265)
5614 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5615 - drop import inflight refcount on signal_completed_replay error (5255)
5616 - unlock page after async write error during send (3677)
5617 - return original error code on reconstructed replies (3761)
5618 - no transno return for symlink open, don't save no-trasno open (3440)
5620 - add pid to ldlm debugging output (4922)
5621 - bump the watchdog timeouts -- we can't handle 30sec yet
5622 - extra debugging for orphan dentry/inode bug (5259)
5624 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5627 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5628 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5629 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5630 - fix NULL dereference in /proc/sys/portals/routes (4827)
5631 - allow failed mdc_close() operations to be interrupted (4561)
5632 - stop precreate on OST before MDS would time out on it (4778)
5633 - don't send partial-page writes before EOF from client (4410)
5634 - discard client grant for sub-page writes on large-page clients (4520)
5635 - don't free dentries not owned by NFS code, check generation (4806)
5636 - fix lsm leak if mds_create_objects() fails (4801)
5637 - limit debug_daemon file size, always print CERROR messages (4789)
5638 - use transno after validating reply (3892)
5639 - process timed out requests if import state changes (3754)
5640 - update mtime on OST during writes, return in glimpse (4829)
5641 - add mkfsoptions to LDAP (4679)
5642 - use ->max_readahead method instead of zapping global ra (5039)
5643 - don't interrupt __l_wait_event() during strace
5645 - add software watchdogs to catch hung threads quickly (4941)
5646 - make lustrefs init script start after nfs is mounted
5647 - fix CWARN/ERROR duplication (4930)
5648 - return async write errors to application if possible (2248)
5649 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5650 - print NAL number in %x format (4645)
5651 - update barely-supported suse-2.4.21-171 series (4842)
5652 - support for sles 9 %post scripts
5653 - support for building 2.6 kernel-source packages
5654 - support for sles km_* packages
5656 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5659 - ignore -ENOENT errors in osc_destroy (3639)
5660 - notify osc create thread that OSC is being cleaned up (4600)
5661 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5662 - reconstruct ost_handle() like mds_handle() (4657)
5663 - create a new thread to do import eviction to avoid deadlock (3969)
5664 - let lconf resolve symlinked-to devices (4629)
5665 - don't unlink "objects" from directory with default EA (4554)
5666 - hold socknal file ref over connect in case target is down (4394)
5667 - allow more than 32000 subdirectories in a single directory (3244)
5668 - fix blocks count for O_DIRECT writes (3751)
5669 - OST returns ENOSPC from object create when no space left (4539)
5670 - don't send truncate RPC if file size isn't changing (4410)
5671 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5672 - bind to privileged port in socknal and tcpnal (3689)
5674 - rate limit CERROR/CWARN console message to avoid overload (4519)
5675 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5676 - basic mmap support (3918)
5677 - kernel patch series update from b1_4 (4711)
5679 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5682 - avoid crash during MDS cleanup with OST shut down (2775)
5683 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5684 - don't use bad inodes on the MDS (3744)
5685 - dynamic object preallocation to improve recovery speed (4236)
5686 - don't hold spinlock over lock dumping or change debug flags (4401)
5687 - don't zero obd_dev when it is force cleaned (3651)
5688 - print grants to console if they go negative (4431)
5689 - "lctl deactivate" will stop automatic recovery attempts (3406)
5690 - look for existing locks in ldlm_handle_enqueue() (3764)
5691 - don't resolve lock handle twice in recovery avoiding race (4401)
5692 - revalidate should check working dir is a directory (4134)
5694 - don't always mark "slow" obdfilter messages as errors (4418)
5696 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5699 - don't close LustreDB during write_conf until it is done (3860)
5700 - fix typo in lconf for_each_profile (3821)
5701 - allow dumping logs from multiple threads at one time (3820)
5702 - don't allow multiple threads in OSC recovery (3812)
5703 - fix debug_size parameters (3864)
5704 - fix mds_postrecov to initialize import for llog ctxt (3121)
5705 - replace config semaphore with spinlock (3306)
5706 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5707 - don't allow enqueue to complete on a destroyed export (3822)
5708 - down write_lock before checking llog header bitmap (3825)
5709 - recover from lock replay timeout (3764)
5710 - up llog sem before sending rpc (3652)
5711 - reduce ns lock hold times when setting kms (3267)
5712 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5713 - fix NULL deref and obd_dev leak on setup error (3312)
5714 - replace some LBUG about llog ops with error handling (3841)
5715 - don't match INVALID dentries from d_lookup and spin (3784)
5716 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5717 - fix invalid assertion in ptlrpc_set_wait (3880)
5719 - add libwrap support for the TCP acceptor (3996)
5720 - add /proc/sys/portals/routes for non-root route listing (3994)
5721 - allow setting MDS UUID in .xml (2580)
5722 - print the stack of a process that LBUGs (4228)
5724 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5727 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5728 - make sure to unset replay flag from failed open requests (3440)
5729 - if default stripe count is 0, use OST count for inode size (3636)
5730 - update parent mtime/ctime on client for create/unlink (2611)
5731 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5732 - free recovery state on server during a forced cleanup (3571)
5733 - unregister_reply for resent reqs (3063)
5734 - loop back devices mounting and status check on 2.6 (3563)
5735 - fix resource-creation race that can provoke i_size == 0 (3513)
5736 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5737 - more debugging for page-accounting assertion (3746)
5738 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5739 - avoid deadlock after precreation failure (3758)
5740 - fix race and lock order deadlock in orphan handling (3450, 3750)
5741 - add validity checks when grabbing inodes from l_ast_data (3599)
5743 - add /proc/.../recovery_status to obdfilter (3428)
5744 - lightweight CDEBUG infrastructure, debug daemon (3668)
5745 - change default OSC RPC parameters to be better on small clusters
5746 - turn off OST read cache for files smaller than 32MB
5747 - install man pages and include them in rpms (3100)
5748 - add new init script for (un)mounting lustre filesystems (2593)
5749 - run chkconfig in %post for init scripts (3701)
5750 - drop scimac NAL (unmaintained)
5752 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5755 - clean kiobufs before and after use (3485)
5756 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5757 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5758 - change init script to fail more clearly if not run as root (1528)
5759 - allow clients to reconnect during replay (1742)
5760 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5761 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5762 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5763 - fix cond_resched (3554)
5764 - only evict unfinished clients after recovery (3515)
5765 - allow bulk resend, prevent data loss (3570)
5766 - dynamic ptlrpc request buffer allocation (2102)
5767 - don't allow unlinking open directory if it isn't empty (2904)
5768 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5769 - remove extraneous obd dereference causing LASSERT failure (3334)
5770 - don't use get_cycles() when creating temp. files on the mds (3156)
5771 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5772 - handle EEXIST for set-stripe, set proper directory name (3336)
5774 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5775 - fix ksocknal_fmb_callback() error messages (2918)
5777 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5780 - don't copy lvb into (possibly NULL) reply on error (2983)
5781 - don't deref dentry after dput, don't free lvb on error (2922)
5782 - use the kms to determine writeback rpc length (2947)
5783 - increment oti_logcookies when osc is inactive (2948)
5784 - update client's i_blocks count via lvb messages (2543)
5785 - handle intent open/close of special files properly (1557)
5786 - mount MDS with errors=remount-ro, like obdfilter (2009)
5787 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5788 - don't use cancelling-locks' kms values (2947)
5789 - use highest lock extent for kms, not last one (2925)
5790 - don't dereference ERR_PTR() dentry in error handling path (3107)
5791 - fix thread race in portals_debug_dumplog() (3122)
5792 - create lprocfs device entries at setup instead of at attach (1519)
5793 - common AST error handler, don't evict client on completion race (3145)
5794 - zero nameidata in detach_mnt in 2.6 (3118)
5795 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5796 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5797 - initialize RPC timeout timer earlier for 2.6 (3219)
5798 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5799 - print nal/nid for unknown nid (3258)
5800 - additional checks for oscc recovery before doing precreate (3284)
5801 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5802 - don't crash in mdc_close for bad permissions on open (3285)
5803 - zero i_rdev for non-device files (3147)
5804 - clear page->private before handing to FS, better assertion (3119)
5805 - tune the read pipeline (3236)
5806 - fix incorrect decref of invalidated dentry (2350)
5807 - provide read-ahead stats and refine rpc in flight stats (3328)
5808 - don't hold journal transaction open across create RPC (3313)
5809 - update atime on MDS at close time (3265)
5810 - close LDAP connection when recovering to avoid server load (3315)
5811 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5812 - don't leak open file on MDS after open resend (3325)
5813 - serialize filter_precreate and filter_destroy_precreated (3329)
5814 - loop device shouldn't call sync_dev() for nul device (3092)
5815 - clear page cache after eviction (2766)
5816 - resynchronize MDS->OST in background (2824)
5817 - refuse to mount the same filesystem twice on same mountpoint (3394)
5818 - allow llmount to create routes for mounting behind routers (3320)
5819 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5820 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5821 - fix rare problem with rename on htree directories (3417)
5823 - allow default OST striping configuration per directory (1414)
5824 - fix compilation for qswnal for 2.6 kernels (3125)
5825 - increase maximum number of MDS request buffers for large systems
5826 - change liblustreapi to be useful for external progs like lfsck (3098)
5827 - increase local configuration timeout for slow disks (3353)
5828 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5830 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5833 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5834 - don't overwrite extent policy data in reply if lock was blocked (2901)
5835 - drop filter export grants atomically with removal from device (2663)
5836 - del obd_self_export from work_list in class_disconnect_exports (2908)
5837 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5838 - swab reply message in mdc_close, other PPC fixes (2464)
5839 - fix destroying of named logs (2325)
5840 - overwrite old logs when running lconf --write_conf (2264)
5841 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5842 - fix race in target_handle_connect (2898)
5843 - mds_reint_create() should take same inode create lock (2926)
5844 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5845 - don't close files for self_export to avoid uninitialized obd (2936)
5846 - allow MDS with the same name as client node (2939)
5847 - hold dentry reference for closed log files for unlink (2325)
5848 - reserve space for all logs during transactions (2059)
5849 - don't evict page beyond end of stripe extent (2925)
5850 - don't oops on a deleted current working directory (2399)
5851 - handle hard links to targets without a parent properly (2517)
5852 - don't dereference NULL lock when racing during eviction (2867)
5853 - don't grow lock extents when lots of conflicting locks (2919)
5855 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5858 - account for cache space usage on clients to avoid data loss (974)
5859 - lfsck support in lustre kernel code (2349)
5860 - reduce journal credits needed for BRW writes (2370)
5861 - orphan handling to avoid losing space on client/server crashes
5862 - ptlrpcd can be blocked, stopping ALL progress (2477)
5863 - use lock value blocks to assist in proper KMS, faster stat (1021)
5864 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5865 - recovery for initial connections (2355)
5866 - fixes for mds_cleanup_orphans (1934)
5867 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5868 - block all file creations until orphan recovery completes (1901)
5869 - client remove rq_connection from request struct (2423)
5870 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5871 - recovery timer race (2670)
5872 - mdc_close recovey bug (2532)
5873 - ptlrpc cleanup bug (2710)
5874 - mds timeout on local locks (2588)
5875 - namespace lock held during RPCs (2431)
5876 - handle interrupted sync write properly (2503)
5877 - don't try to handle a message that hasn't been replied to (2699)
5878 - client assert failure during cleanup after abort recovery (2701)
5879 - leak mdc device after failed mount (2712)
5880 - ptlrpc_check_set allows timedout requests to complete (2714)
5881 - wait for inflight reqs when ptlrpcd finishes (2710)
5882 - make sure unregistered services are removed from the srv_list
5883 - reset bulk XID's when resending them (caught by 1138 test)
5884 - unregister_bulk after timeout
5885 - fix lconf error (2694)
5886 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5887 - readahead locks pages, leaves pending causing memory pressure (2673)
5888 - increase OST request buffers to 4096 on large machines (2729)
5889 - fix up permission of existing directories in simple_mkdir (2661)
5890 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5891 - don't assign transno to errored transactions (2742)
5892 - don't delete objects on OST if given a bogus objid from MDS (2751)
5893 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5894 - if rq_no_resend, then timeout request after recovery (2432)
5895 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5896 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5897 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5898 - MDS recovery shouldn't delete objects that it has given out (2730)
5899 - if enqueue arrives after completion, don't clobber LVB (2819)
5900 - don't unlock pages twice when trigger_group_io returns error (2814)
5901 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5902 - don't write pages to disk if there was an error (1450)
5903 - don't ping imports that have recovery disabled (2676)
5904 - take buffered bytes into account when balancing socknal conn (2817)
5905 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5906 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5907 - remove little-endian swabbing of llog records (1987)
5908 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5909 - retry reposting request buffers if they fail (1191)
5910 - grow extent at grant time to avoid granting a revoked lock (2809)
5911 - lock revoke doesn't evict page if covered by a second lock (2765)
5912 - disable VM readahead to avoid reading outside lock extents (2805)
5914 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5915 - updated kernel patches for hp-2.4.20 kernel (2681)
5917 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5920 - fix truncated write corruption (2366)
5921 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5923 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5924 - ensure len(uuid) < 37 in lmc (1171)
5925 - fix ia64 OOPS in llog_test (2255)
5926 - zero end of page at obdfilter for partial page writes (2648)
5927 - don't leave stale dentries around after renames (bug 2428)
5928 - fix timeouts when evicting a client with a single lock held (2642)
5929 - set deadline for the initial HELLO message to drain (2634)
5930 - print out dotted-quad IP addresses in the socknal (2302)
5932 - additional debugging for MDS client eviction problem (2443)
5933 - fix mkfsoptions support for osts (2603, 2604)
5935 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5938 - add series for the vanilla 2.6.0 kernel
5939 - add series for the vanilla 2.4.24 kernel
5940 - add series for a cray x86/64 UL kernel drop
5941 - fix xattr patches for the vanilla 2.4.19 series
5943 - generate true UUIDs in lmc (1171)
5944 - have portals stack dumping break in UML (2466)
5945 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5946 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5947 - wake callback waiting threads on client eviction (2460)
5948 - Add --ptldebug and --subsystem to lmc (1719)
5949 - update assertion to allow safe interrupt allocation
5950 - set rq_no_resend for cancel requests (2432)
5951 - recalculate ptlrpcd timeout after resend (2494)
5952 - call vfs_rmdir when removing pending directories (2368)
5953 - fix renaming a file to itself (2429)
5954 - lmc creates a default one-stripe lov (2454)
5955 - expand procfs space to handle large clusters (2326)
5956 - increase UML stack to avoid overflow
5957 - update lconf's list of debug and subsystem masks
5958 - fix lfs find --obd (2510)
5959 - /proc tunable for disabling filter read caching (2591)
5960 - stop rpm packages from altering slapd.conf (2301)
5961 - disable nagle in the socknal under 0conf (2578)
5962 - choose mds inode size based on stripe count (2572)
5963 - fix kernel-source rpm problems (2516)
5965 - add --disable-doc to avoid pdf generation (2421)
5966 - update documentation, tests, type-os, comments
5967 - avoid format warnings on ia64
5968 - remove the TOE NAL
5969 - tiny code cleanups by removing unused fields
5971 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5974 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5975 - remove the most blatant lies from BUILDING file (2371)
5976 - change default debug level to reasonable production setting
5977 - reduce client side cache size to reduce cache flush time
5978 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5979 - make TCP zerocopy and pinger support enabled by default (2476)
5980 - sync writes completed after process exits caused crashes (2319)
5981 - maintain correct mount count on the MDS (2356)
5982 - backout 1557, because 2316 wasn't really fixed
5983 - better file I/O statistics gathering in /proc
5984 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5985 - another kernel patch to fix zero-copy TCP function export
5986 - don't take duplicate lock when processing re-sent getattr (2420)
5987 - lctl uses obd_self_export instead of creating new conn (2353)
5988 - MDS/OST recovery case which requires object creation asserted (2425)
5989 - move lfs from /usr/sbin to /usr/bin in packages
5990 - fix race between mds_client_add and mds_client_free (2417)
5991 - use kmalloc instead of slabs in portals (2430)
5992 - don't create duplicate records when a failover MDS is present (2442)
5993 - remove unnecessary mount age check (2332)
5994 - don't remove directory inodes from locks prematurely (2451)
5995 - don't break if MDS service name is the same as hostname (2103)
5996 - fix races in client write RPC generation when cache full (2482)
5998 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
6001 - remove now-unused request->rq_obd (278)
6002 - if an allocation fails, print out how much memory we've used (1933)
6003 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
6004 - add the "configurable stack size" patch to most series files (1256)
6005 - ability to write large log records, for 100+ OST configs (2306)
6006 - fix NULL deref when filter_prep fails (2314)
6007 - fix operator precedence error in filter_sync
6008 - dynamic allocation of socknal TX descriptors (2315)
6009 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
6010 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
6011 - ensure that CWARN messages in Portals always get to the syslog
6012 - __init/__exit are not for prototype decls (ldlm_init/exit)
6013 - x86-64 compile warning fixes
6014 - fix gateway LMC keyword conflict (2318)
6015 - fix MDS lock inversions in getattr/reint paths (1844)
6016 - fix a rare lock re-ordering bug, which caused deadlock (2322)
6017 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
6018 - DLM race condition prevented some lock evictions (2328)
6019 - ENOMEM detection and retry on socknal sends (2230)
6020 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
6021 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
6022 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
6023 - better error messages when a client is rejected during recovery (1505)
6024 - avoid cancelling locks which were never granted, after failure (2330)
6025 - fix i_sem/journal inversion in mds_client_add (2333)
6026 - fix truncate/getattr lock cycle deadlock (2334)
6027 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
6028 - fix two rare exit paths which could leak an l_lock() ref (2321)
6029 - fencepost error in MDS/OST orphan recovery (2226)
6030 - make log record alignment 8 bytes (1988)
6031 - lstripe now fails when requested offset > ost_count (2237)
6032 - ensure that all kernel series have a complete list.h (1607)
6033 - fix crashes in special-file operations (2316)
6034 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
6036 - allow configurable automake binary, for testing new versions
6037 - small update to the lfs documentation
6039 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
6041 * fix negative export reference count in fsfilt_sync (2312)
6043 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
6044 * release candidate 0.9.1
6046 - orphans are moved into the PENDING directory for possible recovery
6047 - replayed opens now open by fid for orphan/rename safety (1042)
6048 - last close of an orphan inode generates a transno (683)
6049 - chdir() and mount() now pin the directory entry (1020)
6050 - avoid CERROR in normal ll_setattr_raw() error case (1500)
6051 - discard very old requests without processing them (1502)
6052 - remove some common, well-understood CERRORs (1505)
6053 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
6054 - clear "grant" flags in OST replies until OST grant code lands (1644)
6055 - fix read performance by not clobbering i_blksize on client (1598)
6056 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
6057 - make LRU size a /proc tunable, clears locks when reduced (707)
6058 - fix some lprocfs rot that prevented ptlbd from loading (1732)
6059 - server locks take references on exports now (1558)
6060 - build fixes for 2.4.20-rh trees (1663)
6061 - return an error from lov_create if all OSCs are inactive (1751)
6062 - fix import levels when a reconnect happens without a timeout (1597)
6063 - exit early from mds_open if we get a lookup error (1749)
6064 - partial page read at EOF wouldn't wait for disk before sending (1642)
6065 - avoid NULL deref in obdfilter when reading page past EOF (1592)
6066 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
6067 - fix LBUG in ll_it_open_error with rc = -2 (1861)
6068 - write/truncate lock inversion (1639)
6069 - Don't auto-load obdclass, portals modules during cleanup (1495)
6070 - fix timestamps from jumping to "now" (1763)
6071 - extra journal assertions (1648)
6072 - add an extra multiunlink test (1771)
6073 - fix read_record/write_record API (1776)
6074 - fix leak of offset_extent, possible incorrect i_size later (1772)
6075 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
6076 - Debugging for the kqswnal_get_idle_tx problems (1820)
6077 - Allow recovery to be attempted multiple times (1536)
6078 - Write out MDS last_rcvd file after it is first created (1600)
6079 - Fix tx_descriptor leak in failed transmit situations (1827)
6080 - ext3 journaling fixes for assertion failure after IO error (1871)
6081 - class_export_put() on freed export after completion AST error (1896)
6082 - Fix revalidate looping in VFS (1322)
6083 - Don't access a freed export during MDS_REINT timeout (1521)
6084 - Add open-unlink recovery support on the MDS (1673,1764)
6085 - Return an error if no MDS data was read from last_rcvd (1946)
6086 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
6087 - Don't setup a disk device that doesn't match exported UUID (317)
6088 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
6089 - avoid committing NULL handle in force close
6090 - local.sh is now a one-stripe LOV configuration
6091 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
6092 - fix ext3 htree duplicate directory entry corruption (1516)
6093 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
6094 - update to new LOV EA format (2097)
6095 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
6097 2003-06-15 Phil Schwan <phil@clusterfs.com>
6100 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
6101 - per-import/export recovery handling (958, 931, 959)
6102 - multiple last-rcvd slots, for serving multiple FSes (949)
6103 - connections are again shared between multiple imp/exports (963, 964)
6104 - "umount -f" would hang if any requests needed to be sent (393, 978)
6105 - avoid pinning large req buffer by copying for queued messages (989)
6106 - add "uuid" to "lctl device" command to help upcalls (991)
6107 - "open" RPCs with transnos would confuse recovery counters (1037)
6108 - do proper endian conversion of all wire messages (288, 340, 891)
6109 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
6110 - call waiting locks callback from LDLM recovery thread (1127, 1151)
6111 - fix ptlrpc_connection leak in target_handle_connect (1174)
6112 - fix import refcounting bug in OST and MDS cleanup (1134)
6113 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
6114 - fix very unlikely obd_types race condition (501)
6115 - remove osc_open hack for echo_client (1187)
6116 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
6117 - a failure in read_inode2 leads to deadlock (1139)
6118 - cancel ack-locks as soon as transaction is committed (1072)
6119 - fix major leaks and crashes in the bulk I/O path (937, 1057)
6120 - make sure to commitrw after any preprw to avoid deadlock (1162)
6121 - failing to execute a file in a lustre FS would lock inode (1203)
6122 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
6123 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
6124 - obd_uuid2tgt would walk off the end of the list (1255)
6125 - on IA64 the osc would give portals incorrect bulk size (1258)
6126 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
6127 - fix lock inversion caused by new llite matching code (1282)
6128 - limit the number of dirty pages on a client to 10MB (1286)
6129 - timed out locks were not being corrected cancelled (1289)
6130 - fix O_DIRECT above 4GB on IA-32 (1292)
6131 * major user-visible changes
6132 - fail out/fail over policy now controlled by the upcall (993)
6134 - add OBD_PING to check server availability and failure (954)
6135 - lustre messages are now sent in sending host order (288, 340, 891)
6136 - add eadatalen to MDS getattr reply (340)
6137 - OST read replies may contain second buffer, with per-page status (593)
6139 2003-03-11 Phil Schwan <phil@clusterfs.com>
6142 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
6143 - failed open()s could cause deadlock; fixed (867, 869)
6144 - stop cancelling OST locks when files are closed (481)
6145 - overlapping XID spaces caused network corruption (851, 853)
6146 - fix unsafe fsfilt counter arithmetic; change to atomic_t
6147 - setattr_raw added, to do single-RPC, server-side setattrs
6148 - lmc/lconf syntax change for OST UUIDs
6149 - fix crashy race condition between ptlrpc_free_req and osc_close
6150 - don't use request in mdc_enqueue if we hit a timeout (889)
6151 - don't set the inode i_size for regular files from the MDS (896)
6152 - handle out of order completion AST (842)
6153 - don't LBUG if a lock request times out after receiving AST (913)
6154 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
6155 - if a bad lock AST arrives, send an error instead of dropping entirely
6156 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
6157 - fix leak in bulk IO when only partially completed (899, 900, 926)
6158 - fix O_DIRECT for ia64 (55)
6159 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
6160 - C-z after timeout could hang a process forever; fixed (977)
6162 - client-side I/O cache (678, 924, 929, 941, 970)
6164 - READPAGE and SETATTRs which don't take server-side locks get
6167 2003-02-11 Phil Schwan <phil@clusterfs.com>
6170 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
6171 - Fix fsfilt_extN_readpage() to read a full page of directory
6172 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
6173 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
6174 - Fix imbalanced LOV object allocation and out-of-bound access (469)
6175 - Most intent operations were removed, in favour of a new RPC mode
6176 that does a single RPC to the server and bypasses most of the VFS
6177 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
6178 - Aggressively cancel local locks on DLM servers
6179 - mds_reint_unlink sends EA to the client if it's the last nlink.
6180 client uses that EA to unlink OST objects.
6181 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
6182 - recursive symlinks were fixed (439)
6183 - fixed NULL deref in DEBUG_REQ
6184 - filter_update_lastobjid no longer calls sync, which annoyed extN
6185 - fixed multi-client small-writes to a single file problem (445)
6186 - fixed mtime updates during file writes (607)
6187 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
6188 - fixed bug in obd_brw_read/write() (under guise of testing 367)
6189 - fixed Linux OST size reporting problem (444, 656)
6190 - OST now updates object mtime with writes or setattr (607, 619)
6191 - client verifies file size before zeroing page past EOF (445)
6192 - OST now writes last allocated objid to disk with allocation (108)
6193 - LOV on echo now works (409)
6195 - mds_reint_unlink sends a new buffer, with the EA included. this
6196 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
6197 set if a regular file was being unlinked, and it was the last link
6198 - use PtlGet from the target for bulk writes (315)
6199 - OST now updates object mtime with writes or setattr (607, 619)
6200 - LDLM now has a grant-time callback to revalidate locked items, if
6202 - Many MDS operations were reorganized to combat race conditions
6204 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
6205 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
6207 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
6210 - Fully reactivate OST imports after reconnection (512, others)
6211 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
6212 - More graceful error handling for truncating on dead OST (515)
6213 - Don't error out unless we're actually accessing dead stripes (474)
6214 - Fix garbage sizes when stripes are missing (410)
6215 - LRU counters were broken, causing constant lock purge (433, 432)
6216 - garbage on read from stripes with failed OSTs (441)
6217 - mark OSCs as active before reconnecting during recovery (438)
6218 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
6219 - lfind did not preserve OST order in output (443)
6220 - symlinks cause hung clients, incorrect data (439)
6221 - stop dereferencing request after dropping refcount (457)
6222 - don't LASSERT(spin_is_locked) on non-SMP (455)
6223 - fixes for many rename() bugs
6224 - fstat didn't correctly synchronize attributes (399)
6225 - server must handle lock cancellation during blocking AST prep (487)
6226 - bulk descriptors were free()d too soon (511)
6227 - fix paths in lconf, which would load incorrect modules (451, 507)
6228 - fix confusing lconf 'host not found' error message (386)
6229 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
6230 - fix race condition in mdc_blocking_ast() for inode access (526)
6231 - fix lov_unpackmd() unpacking wrong number of stripes (537)
6232 - fix lov_set_osc_active() marking wrong OSC inactive (440)
6233 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
6234 - fix multiple writes of stripe MD to MDS (358, maybe 519)
6235 - fix lstripe in several ways (kernel side) (527)
6236 - fix request leak in ldlm_cli_enqueue (262)
6237 - incorrect OSC was marked inactive after OST failure
6238 - call mds_fs_cleanup before unmounting filesystem (524)
6239 - fix races between taking ns_lock and ldlm_lock_change_resource
6240 - fix races updating LOV export open file list
6241 - fix lov_enqueue error path, avoid decref-ing bad lock handle
6242 - fix recovery NULL deref in ldlm_cli_cancel_unused
6243 - fix some DLM races by using new hash table for lock handles (419)
6244 - permit the client to specify desired inodes, at replay
6245 - duplicate requests when we queue them for replay reintegration
6246 - fix last_rcvd offset calculation
6247 - sync after each recovered transaction, so we always make progress
6248 - never, not always, ERESTART requests without transnos
6249 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
6251 - skip replay if the MDS says that the client is already connected
6252 - don't check for a recovery-enabled export to match lctl's UUID
6253 - don't INC_USE_COUNT for phantom exports
6254 - don't crash when cleaning up phantom exports (567)
6255 - don't double-finish or set replay data for errored mdc_open requests
6256 - abort requests when they time out, so we don't get old replies
6257 - send/receive replies for AST messages again
6258 - if the client says that it doesn't have the lock, cancel it on the
6260 - if we timeout during I/O, don't try to cancel an in-use lock; instead
6261 mark it as destroyed, it will all work out when decref is called
6262 - fix module use counts (22, 581)
6264 - ASTs now expect a reply (server cancels lock on error reply)
6266 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
6269 - fix many simultaneous client startup (392)
6270 - fix dentry->d_it clobbering
6271 - credentials weren't being shipped for readdir/getattr operations
6272 - remove invalid assertions triggered during some concurrent MD
6274 - proper Lustre versions added (336, 389)
6275 - fix memory leak for create error case (398)
6276 - fix LOV locking bug that would get cli/srv out of sync
6277 - fix echo client over LOV (409)
6278 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
6279 - fix double-O_EXCL intent crash (424)
6280 - avoid sending multiple lock CANCELs (352)
6282 - MDS can do multi-client recovery (modulo bugs in new code)
6284 - many updates, edits, cleanups
6286 2002-11-18 Phil Schwan <phil@clusterfs.com>
6289 - fix null d_it dereference (346)
6290 - fix full OST/dbench hang (333)
6291 - fix permission problem with file removal (286)
6292 - fix removal of OSCs from LOV when they fail
6293 - fix NULL deref during bulk timeout (214)
6294 - fix problems related to multiple filesystems on one MDS (241)
6295 - fixed serious subtle metadata locking bugs
6296 - free locks on clients when inodes are removed due to memory
6298 - fix inode pointer in lock data (285)
6299 - partial support for multiple MDS on a single host (241)
6300 - data locks weren't cancelled at clear_inode time (290, 311)
6301 - intent locks could lead to unbounded lock growth (205)
6302 - added a maximum lock count, an LRU list, and a flusher
6303 - fix multiple rename (365)
6304 - properly abstracted the echo client
6305 - OSC locked 1 byte too many; fixed
6306 - rewrote brw callback code:
6307 - fixed recovery bugs related to LOVs (306)
6308 - fixed too-many-pages-in-one-write crash (191)
6309 - fixed (again) crash in sync_io_timeout (214)
6310 - probably fixed callback-related race (385)
6312 - Add capability to MDS protocol
6313 - LDLM cancellations and callbacks on different portals
6315 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
6318 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
6319 - defer open object destruction to close time (601981, 138)
6320 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
6321 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
6322 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
6323 - don't call obd_setattr in ll_file_release for destroyed objects
6324 * protocol change to lustre_msg: move |version| and add |flags|
6325 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
6326 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
6327 * added replay of create, unlink, link and rename operations during
6328 MDS failover; recovery should be much more robust now
6329 * remove failed OSCs from LOVs (only lov_create uses this so far)
6330 * the lustre-HOWTO was brought (more) up to date (582544)
6332 2002-10-23 Phil Schwan <phil@clusterfs.com>
6335 - in-use dentries weren't being reused properly (617851)
6336 - prevent multiple LDLM setup (599178)
6337 - fix LOV size calculations for truncate (617853)
6338 - fix client handling of MDS intent errors (POSIX)
6339 - fix permission bug in lovstripe.c test (624321)
6340 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6341 - truncate past end of file could corrupt data
6342 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6343 - a race in recovery could return ETIMEDOUT to apps (623947)
6344 - building outside the source directory was fixed
6345 * the lustre-HOWTO was brought (more) up to date (582544)
6346 * major progress was made on recovery functionality
6348 2002-10-10 Phil Schwan <phil@clusterfs.com>
6351 - recovery deadlock fix
6352 - rm -rf causes LBUG fix (617817)
6353 - file open by multiple tasks fix (618962)
6354 - directory permissions bugs (602707 and 620007)
6355 - journal_stop fixed with locking (611313)
6356 - O_APPEND failures resolved (618273, perhaps 614459)
6357 - lconf PATH fix (619770)
6358 - IA64 build fix (621450)
6359 - RPC buffer sizes scale with amount of memory
6361 2002-10-01 Phil Schwan <phil@clusterfs.com>
6364 - locks would be cancelled without throwing away data pages,
6365 resulting in inconsistent data (605627)
6366 - inode attributes were not always being refreshed (605627, 612449)
6367 - lconf now continues to cleanup after lctl reports an error
6368 - MDS now enforces user permissions (602707)
6369 - lprocfs cleanup fixed, but not yet enabled (614157)
6370 - fixed infinite server hang, should a client not respond to an AST
6371 - avoid going into recovery if user calls readlink() with a buffer
6372 that's too small (613941)
6373 - AST RPCs no longer require replies (614867) -- this may be changed
6374 - don't crash server if client sends an IOV that's too big (611336)
6375 - fixed lock conversion deadlock (611892)
6376 - fixed the following of symlinks (614622)
6377 * recovery: the server can remove locks from a client that dies, other
6378 clients can make progress
6379 * more extN patch fixes
6380 * compile-time configurable ptlrpc buffer allocations
6382 - collaborative read cache document
6383 - Lustre Lite Performance CDR document-in-progress
6385 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6388 - fix typo in patch-2.4.18
6390 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6393 - clear ptlrpc request each time in handle_incoming_request()
6394 - unlink of files now destroys the object on the OST
6396 2002-09-19 Peter Braam <braam@clusterfs.com>
6398 * add hard link support
6399 * change obdfile creation method
6400 * kernel patch changed
6402 2002-09-19 Peter Braam <braam@clusterfs.com>
6405 - stack overflow bug in extN fixed
6407 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6409 * documentation updates
6410 - add man pages for config tools
6411 - update tests/README to describe testing with new config tools
6412 - finish metadata API descriptions
6413 * bug fixes and cleanups
6414 - statfs workaround for 16TB limit
6415 - LOV stripe allocation improved, can stripe on subset of OSTs
6416 - LOV file size/IO offset was wrong for files > 4GB in size
6417 - object EA data was being dropped, caused files to be unreadable
6418 - memory overflow with non-LOV OST caused memory corruption
6419 - fixed regression tests to work with new config tools, obdfilter
6420 - fixed bug when directory size became larger than 1 block
6421 - fixed bug (for single client case) when PWD was deleted
6422 - invalidate local directory pages when doing intent-based ops
6423 - avoid LDLM oops when lock callback contained bad data
6425 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6427 * documentation updates
6428 * bug fixes and cleanups
6429 - configuration tools
6432 - 64-bit compile warnings
6433 - 64-bit internal statfs data
6435 * test_brw on persistent OST devices
6437 * lprocfs (disabled)
6439 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6441 * documentation updates
6442 * bug fixes and cleanups
6443 * configuration tools
6445 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6448 * many small fixes to 0_5_4
6449 * io/network handling
6450 * thinkos in MDS operations
6452 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6455 * crucial basic fixes to 0.5.3
6456 * IOR, Iozone work over Elan
6459 2002-08-07 Phil Schwan <phil@clusterfs.com>
6460 * version 0_5_3, our first alpha
6461 * we use the new Portals iovs
6462 * documentation updates
6463 * bug fixes and cleanups
6464 * small changes in the DLM wire protocol
6466 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6467 * version 0_5_1 with some initial stability,
6468 * locking on MD and file I/O.
6469 * documentation updates
6470 * several bug fixes since 0.5.0
6471 * small changes in wire protocol
6473 2002-07-18 Phil Schwan <phil@clusterfs.com>
6475 * delivered as Lustre Light Alpha
6476 * fixed a crash after handling invalid MDS requests
6477 * fixed directory pages for architectures with non-4k pages sizes
6479 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6480 * release version v0_4_4
6481 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6483 * Fixes a number of interruption problems with OST operations.
6484 * Update documentation for portals header changes
6485 * Move all wire protocol structs/defines to lustre_idl.h
6486 * Fixes symlink length bug.
6487 * Add tcpdump to repository.
6489 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6490 * release version v0_4_3
6491 * Fixes statfs for inodes on extN.
6492 * Fixes bug in runtests which would delete /etc/hosts.
6493 * Use 64-bit object IDs wherever possible (not into VFS though)
6494 Remove ost_get_info, which is unused by lustre, and out of date.
6496 2002-07-03 Peter Braam <braam@clusterfs.com>
6497 * release version v0_4_2 Fixes a lookup error (type not passed)
6498 * move forward to head of Portals
6499 * move forward to latest Lustre kernel
6501 2002-06-25 Peter Braam <braam@clusterfs.com>
6502 * release version v0_4_1. Hopefully stable on single node use.