1 2010-01-29 Sun Microsystems, Inc.
4 2.6.16.60-0.42.4 (SLES 10),
5 2.6.27.39-0.3.1 (SLES11),
6 2.6.18-164.6.1.el5 (RHEL 5)
7 2.6.18-164.6.1.0.1.el5 (OEL 5)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.16 - 2.6.30 vanilla (kernel.org)
11 * Recommended e2fsprogs version: 1.41.6.sun1
12 * The async journal commit feature (bug 19128) and the cancel
13 lock before replay feature (bug 16774) are disabled by default.
15 Severity : enhancement
16 Bugzilla : 21511/19848
17 Description: Update kernel to SLES11 2.6.27.39-0.3.1.
19 Severity : enhancement
21 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
22 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
24 Severity : enhancement
26 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
29 Severity : enhancement
31 Description: Vanilla kernel 2.6.30 patchless client support.
36 Description: bad entry in directory xxx: inode out of bounds
37 Details : fix locking issue in the rename path which could race with any
38 other operations updating the same directory.
40 Severity : enhancement
42 Description: Make watchdog timer messages to be more clear and descriptive.
46 Description: cp -p command does not preserve the dates and timestamp
47 Details : mtime could be spoiled by a write callback
51 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
55 Description: Allow non-root access for "lfs check".
56 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
58 Severity : enhancement
60 Description: quotacheck performance/scaling issues
61 Details : reduce quotacheck time on empty filesystem by skipping uninit
64 Severity : enhancement
66 Description: Enhancement for lfs(1) command to use numeric uid/gid.
68 Severity : enhancement
70 Description: Adjust locks' extents on their first enqueue, so that at the time
71 they get granted, there is no need for another pass through the
72 queues since they are already shaped into the proper forms.
76 Description: Fix mds_shrink_intent_reply()/mds_intent_policy() to pass correct
77 arguments and prevent LBUG() in lustre_shrink_reply_v2().
81 Description: Change tunefs.lustre and mkfs.lustre --mountfsoptions so that
82 exactly the specified mount options are used. Leaving off
83 any "mandatory" mount options is an error. Leaving off any
84 default mount options causes a warning, but is allowed.
85 Change errors=remount-ro from mandatory to default. Sanitize
86 the mount string before storing it. Update man pages accordingly.
90 Description: mds_getattr() should return 0, even if mds_fid2entry() fails with
91 -ENOENT. Also fix in ptlrpc_expire_one_request() to print signed
94 Severity : enhancement
96 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
98 Severity : enhancement
100 Description: Clients can replay thousands of unused locks during recovery
101 Details : Don't replay unused locks (only read locks for now) during
102 recovery. This feature is disabled by default and can be
103 enabled by running the following command on the clients:
104 lctl get_param ldlm.cancel_unused_locks_before_replay
108 Description: can't stat file in some situation.
109 Details : improve initialize osc date when target is added to mds and
110 ability to resend too big getattr request is client isn't have info
115 Description: Prevent inconsistences between linux and lustre mount structures.
116 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
117 Make the sleep interruptible.
119 Severity : enhancement
121 Description: Communicate OST degraded/readonly state via statfs to MDS
122 Details : Flags in the statfs returned from OSTs indicate whether the
123 OST is in a degraded RAID state, or if the filesystem has
124 turned read-only after a filesystem error is detected.
129 Description: don't panic if EPROTO was hit when reading symlink
130 Details : correctly handling request reference in error cases.
135 Description: open sometimes returns ENOENT instead of EACCES
136 Details : checking permission should be part of open part of mds_open, not
137 lookup part. so server should be set DISP_OPEN_OPEN disposition
138 before starting permission check.
139 Also not need revalidate dentry if client already have LOOKUP lock.
143 Description: enable client interface failover
144 Frequency : on servers with multiple network interfaces
145 Details : When a child reconnects from another NID, properly update export
146 nid hash position and ldlm reverse import.
148 Severity : enhancement
150 Description: implemented direct I/O with arbitrary (nonaligned) memory
151 addresses and file offsets.
153 Severity : enhancement
155 Description: added more recovery timeout options.
157 Severity : enhancement
159 Description: added llapi_file_open, llapi_file_create, llapi_file_get_stripe
164 Description: Avoid deadlock for local client writes
165 Frequency : only on systems with clients writing to an OST on the same node
166 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
167 memory freeing context. This allows OST threads to set the
168 PF_MEMALLOC flag on task structures in order to allocate memory
169 from reserved pools and complete IO.
170 Use GFP_HIGHUSER for OST allocations for non-local client writes,
171 so that the OST threads generate memory pressure and allow
172 inactive pages to be reclaimed.
177 Descriptoin: lock ordering violation between &cli->cl_sem and _lprocfs_lock
178 Details : move ldlm namespace creation in setup phase to avoid grab
179 _lprocfs_lock with cli_sem held.
183 Frequency : only during format of test systems
184 Description: Unable to run several mkfs.lustre on loop devices at the same time
185 Details : mkfs.lustre returns error 256 on the concurrent loop devices
186 formatting. The solution is to proper handle the error.
188 Severity : enhancement
190 Description: implement async create (obd_async_create) method for osc, to avoid
191 too long waiting new ost objects with holding ldlm lock.
195 Frequency : occasionally during network problems
196 Description: client not allowed to reconnect to OST because of active request
197 Details : abort bulk requests received by the OST once the client has timed
198 out since the client will resend the request anyway. The client
199 also now retries to reconnect to the same server if a connect
200 request failed with EBUSY or -EAGAIN.
203 Frequency : rare, if used wide striped file and one ost in down.
205 Descriptoin: don't return error if we created a subset of objects for file.
206 Details : lov_update_create_set() uses set->set_success as index for created
207 objects, so if some requests failed, they will have hole at end of
208 array and we can use qos_shrink_lsm for allocate correct lsm.
212 Description: Slow stale export processing during normal start up
213 Details : The global mgc lock prevents OST setup to be run in parallel.
214 Replace the global lock with a per-config_llog_data semaphore.
218 Description: Out or order replies might be lost on replay
219 Details : In ptlrpc_retain_replayable_request if we cannot find retained
220 request with tid smaller then one currently being added, add it
221 to the start, not end of the list.
225 Description: BUG: soft lockup - CPU#1 stuck for 10s! [ll_mdt_07:4523]
226 Details : add cond_resched() calls to avoid hogging the cpu for too long
227 in the hash code. Make also lustre_hash_for_each_empty() more
230 Severity : enhancement
232 Description: Performance improvements for debug messages with D_RPCTRACE,
233 D_LDLM, D_QUOTA options.
236 Frequency : only with NFS export
238 Description: (lov_merge.c:74:lov_merge_lvb())
239 ASSERTION(spin_is_locked(&lsm->lsm_lock)) failed (SR 71691004)
240 Details : Fix a race in the nfs export code by populating inode
241 info while the new inode is still locked
243 Severity : enhancement
245 Description: add a new file in procfs called force_lbug. Writting to this
246 file triggers a LBUG. Only for test purpose.
250 Description: OOM killer causes node hang
251 Details : really interrupt the sleep in osc_enter_cache on signals
255 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
256 Details : fix race during quota release on the slave.
258 Severity : enhancement
260 Description: smaller hash bucket sizes, cleanups
261 Details : increase hash table sizes and enabled rehashing for pools, quota
262 uuid, nid & per-nid stats.
264 Severity : enhancement
266 Description: Add ldiskfs maxdirsize mount option
267 Details : add max_dir size mount option
271 Description: panic in ll_statahead_thread
272 Details : prevent parent thread to be killed before its child
275 Frequency : only with 16TB device
277 Description: unable to perform "mount -t lustre" of 16TB OST device
278 Details : Mounting 16TB LUNs failed due to three bugs in mkfs.lustre.
282 Description: ASSERTION(atomic_read(&imp->imp_inflight) == 0) failed
283 Details : unregistering should be zero if no RPC inflight.
287 Description: hyperion: Oops during metabench
288 Details : Correct the refcount of lov_request_set
290 Severity : enhancement
292 Description: Add mptlinux and nxge drivers to Lustre builds
294 Severity : enhancement
296 Description: Fix watchdog timer message to be more clear
297 Details : Make watchdog timer messages more clear and descriptive.
301 Description: LNET soft lockups in socknal_cd thread
302 Details : don't hog CPU for active-connecting if another connd is
303 accepting connecting-requst from the same peer
307 Description: recovery-small test_17 hang
308 Details : Land several AT improvements & fixes.
312 Description: MDS panic and hanging client processes
313 Details : Replace exp_ops_stats with exp_nid_stats->nid_stats
317 Description: OSS stuck in recovery.
318 Details : fix race during recovery. class_unlink_export,
319 class_set_export_delayed and target_queue_last_replay_reply
320 may race while increasing/decreasing obd_recoverable_clients
321 and obd_delayed_clients, causing recovery to wait forever.
323 Severity : enhancement
325 Description: add cascading_rw.c to lustre/tests
329 Description: filter_last_id() NULL deref
330 Details : lprocfs_filter_rd_last_id() should check for the fully
331 setup obd device, before proceeding further.
333 Severity : enhancement
335 Description: Loadgen improvements
336 Details : stacksize and locking fixes for loadgen
339 Description: Quiet CERROR("dirty %d > system dirty_max %d\n"
340 Details : The atomic_read() allowing the atomic_inc() are not covered
341 by a lock. Thus they may safely race and trip this CERROR()
342 unless we add in a small fudge factor (+1).
344 Severity : enhancement
346 Description: shrink_slab: nr=-9223362083340912175
347 Details : fix spurious message from shrink_slab reporing negative nr
350 Description: Quiet bogus previously committed transno error
351 Details : suppress the "server went back in time" error message which
352 is always printed even in the common case after a client eviction
355 Description: Parallel statfs() calls result in client eviction
356 Details : cache statfs data for 1s.
359 Description: parallel-scale test_compilebench: @@@@@@ FAIL: compilebench
361 Details : fix serveral issues in pinger code causing clients not to ping
362 servers for too long, resulting in evictions.
365 Description: e2fsck should warn when MMP update interval is extended
366 Details : print mmp_check_interval and make it possible to abort mount
367 operation in case it takes too long.
369 -------------------------------------------------------------------------------
371 2009-10-16 Sun Microsystems, Inc.
373 * Support for kernels:
374 2.6.16.60-0.42.4 (SLES 10),
375 2.6.27.29-0.1 (SLES11, i686 & x84_64 only),
376 2.6.18-128.7.1.el5 (RHEL 5),
377 2.6.18-128.7.1.el5 (OEL 5),
378 * Client support for unpatched kernels:
379 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
380 2.6.16 - 2.6.27 vanilla (kernel.org)
381 * Recommended e2fsprogs version: 1.41.6.sun1
382 * ext4 support for RHEL5 is experimental and thus should not be
385 Severity : enhancement
387 Description: Add OEL5 support.
389 Severity : enhancement
391 Description: Update kernel to SLES11 2.6.27.29-0.1.
395 Description: File checksum failures with OST read cache on
396 Details : Disable page poisoning when the bulk transfer has to be aborted
397 because the client got evicted.
401 Description: Don't allow make backward step on assiging osc next id.
402 Details : race between allocation next id and ll_sync thread can be cause
403 of set wrong osc next id and can be kill valid ost objects.
405 Severity : enhancement
407 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
409 Severity : enhancement
411 Description: Update kernel to SLES10 SP2 2.6.16.60-0.42.4.
415 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
419 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
423 Description: Error handling in osc_statfs_interpret() has been improved.
424 Details : Check in osc_statfs_interpret() for EBADR.
428 Description: Do not send statfs() requests to OSTs disabled by administrator.
429 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
433 Description: Do not update ctime for the deleted inode.
434 Details : Check in mds_reint_unlink() before calling fsfilt_setattr().
438 Description: Increase of the size of the LDLM resource hash.
439 Details : Bump up RES_HASH_BITS=12.
443 Description: correctly send lsm on open replay
444 Details : MDS is trust to LSM size on replay open, but client can set wrong size
449 Description: Deadlock between filter_destroy() and filter_commitrw_write().
450 Details : filter_destroy() does not hold the DLM lock over the whole
451 operation. If the DLM lock is dropped, filter_commitrw() can go
452 through, causing the deadlock between page lock and i_mutex.
453 The i_alloc_sem should also be hold in filter_destroy() while
458 Description: truncate starts GFP_FS allocation under transaction causing deadlock
459 Details : ldiskfs_truncate calls grab_cache_page which may start page
460 allocation under an open transaction. This may lead to
461 calling prune_icache with consequent lustre reentrance.
464 Frequency : only when down/upgrading the MDS to 1.6/1.8 while 1.8 clients are
465 still up and when the OST pool feature is used
467 Description: interop testing got LBUG when run dd with OST pool
468 :LustreError: 30032:0:(llite_lib.c:1913:ll_replace_lsm()) LBUG
469 Details : down/upgrading the MDS to a version that doesn't/does support OST
470 pool can cause clients to crash because the lsm has changed
475 Description: missing tree_status on 1.8.1 RPM build
476 Details : make rpms failed due because the tree_status file is missing.
480 Description: continuing LustreError "mds adjust qunit failed!"
481 Details : don't print message on the console when ->adjust_qunit fails.
485 Description: don't increase ldlm timeout if previous client was evicted
486 Details : if a client doesn't respond to a blocking callback within the
487 adaptive ldlm enqueue timeout, don't adjust the adaptive estimate
488 when the lock is next granted.
492 Description: ost is being unmounted w/o all writes to last_rcvd landing on disk.
493 affects recovery negatively.
494 Details : make sure all exports have been properly destroyed by the zombie
495 thread processed before stopping the target.
499 Description: Performance degradation with O_DIRECT between 1.6 & 1.8.1 b190
500 Details : disable write barrier for ext4/SLES11.
504 Description: Kernel panic - not syncing: Out of memory and no killable
505 processes... on OSS when iozone
506 Details : fix memory leak in the journal checksum patch.
510 Description: group quota "too many blocks" OSS crashes
511 Details : we should keep the same uid/gid for lquota_chkquota() and
512 lquota_pending_commit()
516 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
517 Details : don't LBUG on release quota error. Just a workaround until the
518 problem is understood.
520 ------------------------------------------------------------------------------
521 2009-07-31 Sun Microsystems, Inc.
523 * Support for kernels:
524 2.6.16.60-0.39.3 (SLES 10),
525 2.6.27.23-0.1 (SLES11, i686 & x84_64 only),
526 2.6.18-128.1.14.el5 (RHEL 5),
527 * Client support for unpatched kernels:
528 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
529 2.6.16 - 2.6.27 vanilla (kernel.org)
530 * Recommended e2fsprogs version: 1.41.6.sun1
531 * File join has been disabled in this release, refer to Bugzilla 16929.
532 * NFS export disabled when stack size < 8192. Since the NFSv4 export
533 of Lustre filesystem with 4K stack may cause a stack overflow. For
534 more information, please refer to bugzilla 17630.
535 * ext4 support for RHEL5 is experimental and thus should not be
538 Severity : enhancement
540 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
543 Frequency : with 1.8 server and 1.6 clients
545 Description: correctly shrink reply for avoid send too big message to client.
546 Details : 1.8 mds is allocate to big buffer to LOV EA data and this produce
547 some problems with sending this reply to 1.6 client.
551 Description: Repeated atomic allocation failures.
552 Details : Use GFP_HIGHUSER | __GFP_NOMEMALLOC flags for memory allocations
553 to generate memory pressure and allow reclaiming of inactive pages.
554 At the same time, do not allow to exhaust emergency pools.
555 For local clients the use of GFP_NOFS will be introduced in 1.8.2
557 Severity : enhancement
558 Bugzilla : 19846, 18289
559 Description: Update kernel to RHEL5 2.6.18-128.1.14.el5.
561 Severity : enhancement
562 Bugzilla : 19625, 16893, 18668, 19848
563 Description: Add support for SLES11 2.6.27.23-0.1.
565 Severity : enhancement
567 Description: Update client support to vanila kernels up to 2.6.27.
569 Severity : enhancement
571 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
573 Severity : enhancement
575 Description: Compile with -Werror by default for i686 and x86_64.
579 Description: resolve race between obd_disconnect and class_disconnect_exports
580 Details : if obd_disconnect will be called to already disconnected export he
581 forget release one reference and osc module can't unloaded.
583 Severity : enhancement
585 Description: move AT tunable parameters for more consistent usage
586 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
590 Descriptoin: correctly skip time estimate if in recovery
591 Details : rq_send_state insn't bitmask so using bitwise ops is forbid.
595 Descriptoin: OSS DeadLock
596 Details : Use trylock to prevent deadlock when shrink icache.
598 Severity : enhancement
600 Description: Allow tuning service thread via /proc
601 Details : For each service a new
602 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
603 created that can be used to set min/max thread counts, and get the
604 current number of running threads.
606 Severity : enhancement
608 Description: Add state history info file, enhance import info file
609 Details : Track import connection state changes in a new osc/mdc proc file;
610 add overview-type data to the osc/mdc import proc file.
614 Description: Reduce small size read RPC
615 Details : Set read-ahead limite for every file and only do read-ahead when
616 available read-ahead pages are bigger than 1M to avoid small size
621 Description: free_entry erroneously used groups_free instead of put_group_info
623 Severity : enhancement
625 Description: Make read-ahead stripe size aligned.
627 Severity : enhancement
629 Description: MDS create should not wait for statfs RPC while holding DLM lock.
632 Frequency : rare, connect and disconnect target at same time
634 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
635 Details : don't call obd_disconnect under lov_lock. this long time
636 operation and can block ptlrpcd which answer to connect request.
639 Frequency : start MDS on uncleanly shutdowned MDS device
641 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
642 Details : stay in waiting mds<>ost recovery finished produce random bugs
643 due race between two ll_sync thread for one lov target. send
644 ACTIVATE event only if connect realy finished and import have
648 Frequency : start MDS on uncleanly shutdowned MDS device
650 Descriptoin: aborting recovery hang on MDS
651 Details : don't throttle destroy RPCs for the MDT.
655 Description: Slow reads beyond 8Tb offsets.
656 Details : Page index integer overflow in ll_read_ahead_page
660 Description: Soft lockup on OSS after MDS failover
661 Details : MSG_CONNECT_INITIAL is not set on the initial MDS->OST connect.
662 As a conseqence, the patch from bug 18224 is not operational
663 and the MDS export cannot be reused on the OSTs until it gets
667 Frequency : rare, only if using MMP with Linux RAID
669 Description: MMP doesn't work with Linux RAID
670 Details : While using HA for Lustre servers with Linux RAID, it is possible
671 that MMP will not detect multiple mounts. To make this work we
672 need to unplug the device queue in RAID when the MMP block is being
673 written. Also while reading the MMP block, we should read it from
674 disk and not the cached one.
677 Frequency : rare, during recovery
679 Description: Assertion failure in ldlm_lock_put
680 Details : Do not put cancelled locks into replay list, hold references on
685 Description: 1.6.5 mdsrate performance is slower than 1.4.11/12 (MDS is not cpu bound!)
686 Details : create_count always drops to the min value (=32) because grow_count
687 is being changed before the precreate RPC completes.
690 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
693 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
694 Details : add proper locking for creation of jbd_4k slab cache
698 Description: MMP check in ext3_remount() fails without displaying any error
699 Details : When multiple mount protection fails during remount, proper error
704 Description: Rare Client crash on resend if the file was deleted.
705 Details : When file is opened, but open reply is lost and file is
706 subsequently deleted before resend, resend processing logic
707 breaks trying to open the file again, should not try to open.
711 Description: add check for >8TB ldiskfs filesystems
712 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
713 Don't allow >8TB ldiskfs filesystems to be mounted without
714 force_over_8tb mount option
718 Description: Client locked up when running multiple instances of an app. on
719 multiple mount points
720 Details : ll_shrink_cache() can sleep while holding the ll_sb_lock.
721 Convert ll_sb_lock to a read/write semaphore to fix the problem.
725 Description: Cannot acces an NFS-mounted Lustre filesystem
726 Details : An NFS client cannot access the Lustre filesystem NFS-mounted
727 from a Lustre-client exporting the Lustre filesystem via NFS.
731 Description: panic in ll_statahead_thread
732 Details : grab dentry reference in parent process.
735 -------------------------------------------------------------------------------
737 tbd Sun Microsystems, Inc.
739 * Support for kernels:
740 2.6.16.60-0.31 (SLES 10),
741 2.6.18-128.1.6.el5 (RHEL 5),
742 2.6.22.14 vanilla (kernel.org)
743 * Client support for unpatched kernels:
744 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
745 2.6.16 - 2.6.22 vanilla (kernel.org)
746 * Recommended e2fsprogs version: 1.40.11-sun1
747 * File join has been disabled, refer to Bugzilla 16929.
748 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
749 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
750 Lustre filesystem with 4K stack may cause a stack overflow. For more
751 information, please refer to bugzilla 17630.
753 Severity : enhancement
755 Description: Update to RHEL5.3 kernel-2.6.18-128.1.6.el5.
757 Severity : enhancement
759 Description: Update OFED release to 1.4.1 RC4
761 Severity : major, only with big OST
763 Description: Very poor metadata performance on Infiniband lustre configuration
764 Details : OST object precreation becomes very slow on big OSTs. This is due
765 to the ialloc patch spending too much time scanning groups.
768 Frequency : during recovery
770 Description: don't mix llog inodes with normal.
771 Details : allocate inodes for log in last inode group
776 Description: fix lqs' reference which won't be put in some situations
777 Details : This patch fixes:
778 1. In quota_check_common(), this function will check quota
779 for user and group, but only send one return via "pending".
780 In most cases, the pendings should be same. But that is not
782 2. If quotaoff runs between lquota_chkquota() and
783 lquota_pending_commit(), the same thing will happen too.
784 That is why it comes:
785 - if (!ll_sb_any_quota_active(qctxt->lqc_sb))
788 -------------------------------------------------------------------------------
790 2008-12-31 Sun Microsystems, Inc.
792 * Support for kernels:
793 2.6.5-7.314 (SLES 9),
794 2.6.9-67.0.22.EL (RHEL 4),
795 2.6.16.60-0.31 (SLES 10),
796 2.6.18-92.1.17.el5 (RHEL 5),
797 2.6.22.14 vanilla (kernel.org)
798 * Client support for unpatched kernels:
799 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
800 2.6.16 - 2.6.22 vanilla (kernel.org)
801 * Client support for unpatched kernels:
802 we do not recommend using patchless RHEL4 clients with kernels
803 prior to 2.6.9-55EL (RHEL4U5).
804 * Recommended e2fsprogs version: 1.40.11-sun1
805 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
806 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
807 removed cwd "./" (refer to Bugzilla 14399).
808 * A new quota file format has been introduced in 1.6.5.
809 The format conversion from prior releases is handled transparently,
810 but releases older than 1.4.12/1.6.5 don't understand this new
811 format. The automatic format conversion can be avoided by running
812 the following command on the MDS:
813 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
814 For more information, please refer to bugzilla 13904.
815 * A new quota file format was introduced in 1.6.6/1.8.0 (kernels 2.6.16+).
816 The format conversion from prior releases is handled transparently,
817 but releases older than 1.6.6/1.8.0 don't understand this new
818 format. The automatic format conversion can be avoided by running
819 the following commands on the MDS and OSS servers (for
820 pre 1.4.12-1.6.5 quota files):
821 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV',
822 'tunefs.lustre --param="ost.quota_type=ug1" $MDTDEV'
823 or (for 1.4.12/1.6.5 quota files)
824 'tunefs.lustre --param="mdt.quota_type=ug2" $MDTDEV',
825 'tunefs.lustre --param="ost.quota_type=ug2" $MDTDEV'
826 For more information, please refer to bugzilla 13904.
827 * Output of lfs quota has been made less detailed by default,
828 old (verbose) output can be obtained by using -v option.
829 * File join has been disabled in this release, refer to Bugzilla 16929.
830 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
831 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
832 Lustre filesystem with 4K stack may cause a stack overflow. For more
833 information, please refer to bugzilla 17630.
835 Severity : enhancement
837 Description: Caching OSS
838 Details : introduce data caching on the OSS. The OSS now relies on the linux
839 kernel page cache to keep recently accessed data in memory.
840 It is worth noting that all write requests are still flushed
841 synchronously as in lustre 1.6.
843 Severity : enhancement
845 Description: Version based recovery
846 Details : introduce finer grained recovery able to detect transaction
847 dependencies and can deal with transaction gaps caused by clients
848 failing at the same time as the server.
850 Severity : enhancement
852 Description: Enable adaptive timeouts by default
853 Details : The Lustre timeout value in /proc/sys/lustre/timeout is now
854 managed dynamically based on server load and should not need
855 to be tuned manually based on cluster size. This allows Lustre
856 to work under a wider variety of system sizes and loads, without
857 unnecessarily causing lengthy recovery times.
859 Severity : enhancement
861 Description: Add OST Pools support
862 Details : File striping can now be set to use an arbitrary pool of OSTs
864 Severity : enhancement
866 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
867 Details : allow skip disconnected ost for send statfs request and hide error
871 Frequency : rare, on llog test 6
873 Descriptoin: don't allow connect to already connected import
874 Details : allowing connect to already connected import is hide connecting problem.
877 Frequency : rare, on failed llog setup
879 Descriptoin: don't leak obd reference on failed llog setup
880 Details : for failed llog setup - mgc forget call class_destroy_import for
881 client import, move destroy import to more generic place.
886 Descriptoin: allow kill process which wait statahead result
887 Details : for some reasons 'ls' can stick in waiting result from statahead,
888 in this case need way for kill this process.
891 Frequency : rare, at shutdown
893 Descriptoin: panic at umount
894 Details : llap_shrinker can be raced with killing super block from list and
895 this produce panic with access to already freeded pointer
900 Descriptoin: don't lose wakeup for imp_recovery_waitq
901 Details : recover_import_no_retry or invalidate_import and import_close can
902 both sleep on imp_recovery_waitq, but we was send only one wakeup
908 Descriptoin: panic in mds_open
909 Details : don't confuse mds_finish_transno() with PTR_ERR(-ENOENT)
914 Descriptoin: stuck in cache_remove_extent() or panic with accessing to already
916 Details : release lock refernce only after add page to pages list.
919 Frequency : always with long access acl
921 Descriptoin: mds can't pack reply with long acl.
922 Details : mds don't control size of acl but they limited by reint/getattr
927 Frequency : on remount
929 Description: external journal device not working after the remount
930 Details : clear dev_rdonly flag for external journal devices in
936 Description: shutdown vs evict race
937 Details : client_disconnect_export vs connect request race.
938 if client will evicted at this time - we start invalidate
939 thread without referece to import and import can be freed
945 Description: shrink LOV EAs before replying
946 Details : correctly adjust LOV EA buffer for reply.
951 Description: don't skip ost target if they assigned to file
952 Details : Drop slow OSCs if we can, but not for requested start idx.
953 This means "if OSC is slow and it is not the requested
954 start OST, then it can be skipped, otherwise skip it only
955 if it is inactive/recovering/out-of-space.
957 Severity : enhancement
959 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
962 Frequency : rare, need acl's on inode.
964 Description: client can't handle ost additional correctly
965 Details : if ost was added after client connected to mds client can have
966 hit lnet_try_match_md ... to big messages to wide striped files.
967 in this case need teach client to handle config events about add
968 lov target and update client max ea size at that event.
970 Severity : enhancement
972 Description: Update to sles9 kernel-2.6.5-7.314.
974 Severity : enhancement
976 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
979 Frequency : Create a symlink file with a very long name
981 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
982 Details : If there is no extra space in the request for early cancels,
983 ldlm_req_handles_avail() returns 0 instead of a negative value.
988 Description: mds is deadlocked
989 Details : in rare cases, inode in catalog can have i_no less than have parent
990 i_no, this produce wrong order for locking during open, and
991 parallel unlink can be lock open. this need teach mds_open to grab
992 locks in resource id order, not at parent -> child order.
994 Severity : enhancement
996 Description: Add /proc entry for import status
997 Details : The mdc, osc, and mgc import directories now have
998 an import directory that contains useful import data for debugging
1001 Severity : enhancement
1003 Description: Re-disable certain /proc logging
1004 Details : Enable and disable client's offset_stats, extents_stats and
1005 extents_stats_per_process stats logging on the fly.
1008 Frequency : Only on FC kernels 2.6.22+
1010 Description: oops in statahead
1011 Details : Do not drop reference count for the dentry from VFS when lookup,
1012 VFS will do that by itself.
1014 Severity : enhancement
1016 Description: Generic /proc file permissions
1017 Details : Set /Proc file permissions in a more generic way to enable non-
1018 root users operate on some /proc files.
1022 Description: Hitting mdc_commit_close() ASSERTION
1023 Details : Properly handle request reference release in
1024 ll_release_openhandle().
1028 Frequency : only patchless client
1029 Description: add workaround for race between add/remove dentry from hash
1031 Severity : enhancement
1033 Description: Allow OST glimpses to return PW locks
1037 Description: LBUG when llog conf file is full
1038 Details : When llog bitmap is full, ENOSPC should be returned for plain
1043 Description: Prevent import from entering FULL state when server in recovery
1047 Description: service mount cannot take device name with ":"
1048 Details : Only when device name contains ":/" will mount treat it as
1054 Description: replace ptlrpcd with the statahead thread to interpret the async
1055 statahead RPC callback
1059 Frequency : on recovery
1060 Description: I/O failures after umount during fail back
1061 Details : if client reconnected to restarted server we need join to recovery
1062 instead of find server handler is changed and process self eviction
1063 with cancel all locks.
1065 Severity : enhancement
1067 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
1071 Description: Kernel BUG tries to release flock
1072 Details : Lustre does not destroy flock lock before last reference goes
1073 away. So always drop flock locks when client is evicted and
1074 perform unlock regardless of successfulness of speaking to MDS.
1076 Severity : enhancement
1078 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
1080 Severity : enhancement
1082 Description: Upcall on Lustre log has been dumped
1083 Details : Allow for a user mode script to be called once a Lustre log has
1084 been dumped. It passes the filename of the dumped log to the
1085 script, the location of the script can be specified via
1086 /proc/sys/lnet/debug_log_upcall.
1091 Description: avoid messages about idr_remove called for id that is not allocated
1092 Details : Move assigment s_dev for clustered nfs to end of initialization,
1093 for avoid problem with error handling.
1098 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
1099 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
1100 CONN_USED_HASH and this prodice warning when put connection again
1106 Description: avoid ASSERTION(client_stat->nid_exp_ref_count == 0) failed
1107 Details : release reference to stats when client disconnected, not
1108 when export destroyed for avoid races when client destroyed
1109 after main ost export.
1113 Description: more cleanup in mds_lov
1114 Details : add workaround for get valid ost count for avoid warnings about
1115 drop too big messages, not init llog cat under semphore which
1116 can be blocked on reconnect and break normal replay, fix access
1119 Severity : enhancement
1121 Description: Export bytes_read/bytes_write count on OSC/OST.
1125 Description: Early reply size mismatch, MGC loses connection
1126 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
1127 the connect flags are properly negotiated.
1131 Description: Properly propagate oinfo flags from lov to osc for statfs
1132 Details : restore missing copy oi_flags to lov requests.
1136 Description: exports in /proc are broken
1137 Details : recreate /proc entries for clients when they reconnect.
1139 Severity : enhancement
1141 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
1142 : l_getgroups(8), lst(8), routerstat(8)
1143 Details : included man pages for llobdstat(8), llstat(8),
1144 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
1146 Severity : enhancement
1148 Description: Implement lustre ll_show_options method.
1150 Severity : enhancement
1152 Description: Update to SLES9 kernel-2.6.5-7.312.
1154 Severity : enhancement
1156 Description: Update to RHEL4 kernel-2.6.9-67.0.22.EL.
1160 Description: exports in /proc are broken
1161 Details : recreate /proc entries for clients when they reconnect.
1165 Description: don't fail open with -ERANGE
1166 Details : if client connected until mds will be know about real ost count
1167 get LOV EA can be fail because mds not allocate enougth buffer
1172 Description: Resolve device initialization race
1173 Details : Prevent proc handler from accessing devices added to the
1174 obd_devs array but yet be intialized.
1176 Severity : enhancement
1178 Description: configure's --enable-quota should check the
1179 : kernel .config for CONFIG_QUOTA
1180 Details : configure is terminated if --enable-quota is passed but
1181 : no quota support is in kernel
1183 Severity : enhancement
1185 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
1187 Severity : enhancement
1189 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
1193 Frequency : rare, on PPC clients
1194 Description: don't swab ost objects in response about directory, because
1196 Details : bug similar bug 14856, but in different function.
1198 Severity : enhancement
1200 Description: lfs quota tool enhancement
1201 Details : added units specifiers support for setquota, default to
1202 current uid/gid for quota report, short quota stats by
1203 default, nonpositional parameters for setquota, added
1204 llapi_quotactl manual page.
1206 Severity : enhancement
1208 Description: *optional* service tags registration
1209 Details : if the "service tags" package is installed on a Lustre node
1210 When the filesystem is mounted, a local-node service tag will
1211 be created. See http://inventory.sun.com/ for more information
1212 about the Service Tags asset management system.
1216 Description: Client runs out of low memory
1217 Details : Consider only lowmem when counting initial number of llap pages
1220 Frequency : occasional
1222 Description: add refcount for osc callbacks, so avoid panic on shutdown
1224 Severity : enhancement
1226 Description: Update to RHEL4 kernel-2.6.9-67.0.20.
1229 Frequency : testing only
1231 Description: sanity test 65a fails if stripecount of -1 is set
1232 Details : handle -1 striping on filesystem in ll_dirstripe_verify
1235 Frequency : only in unusual configurations
1237 Description: Kernel panic with find ost index.
1238 Details : lov_obd have panic if some OST's have sparse indexes.
1240 Severity : enhancement
1242 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
1245 Frequency : rarely, if filesystem is mounted with -o flock
1247 Description: do not process already freed flock
1248 Details : flock can possibly be freed by another thread before it reaches
1249 to ldlm_flock_completion_ast.
1252 Frequency : rarely, if filesystem is mounted with -o flock
1254 Description: LBUG during stress test
1255 Details : Need properly lock accesses the flock deadlock detection list.
1258 Frequency : rarely, if binaries are being run from Lustre
1260 Description: oops in page fault handler
1261 Details : kernel page fault handler can return two special 'pages' in
1262 error case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
1265 Frequency : rarely, during shutdown
1267 Description: timeout with invalidate import.
1268 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
1269 should be handled by ptlrpcd. This produce long age waiting and
1270 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
1276 Description: ASSERTION(CheckWriteback(page,cmd)) failed
1277 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
1281 Frequency : only with broken builds/installations
1283 Description: no LBUG if lquota.ko and fsfilt_ldiskfs.ko are different versions
1284 Details : just return an error to a user, put a console error message
1286 Severity : enhancement
1288 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
1290 Severity : enhancement
1292 Description: Update to RHEL4 kernel-2.6.9-67.0.15.
1294 Severity : enhancement
1296 Description: enable MGS and MDT services start separately
1297 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
1298 a co-located MGS without starting the MGS, which is a complement
1299 to 'nosvc' mount option.
1302 Frequency : always, on big-endian systems
1304 Description: cleanup in ptlrpc code, related to PPC platform
1305 Details : store magic in native order avoid panic's in recovery on PPC
1306 node and forbid from this error in future. Also fix posibily
1307 of twice swab data. Fix get lov striping to userpace.
1310 Frequency : rarely, if replay get lost on server
1312 Description: server incorrectly drop resent replays lead to recovery failure.
1313 Details : do not drop replay according to msg flags, instead we check the
1314 per-export recovery request queue for duplication of transno.
1317 Frequency : after recovery
1319 Description: precreate to many object's after del orphan.
1320 Details : del orphan st in oscc last_id == next_id and this triger growing
1321 count of precreated objects. Set flag LOW to skip increase count
1322 of precreated objects.
1325 Frequency : rare, on clear nid stats
1327 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
1328 Details : when clean nid stats sometimes try destroy live entry,
1329 and this produce panic in free.
1332 Frequency : occasionally since 1.6.4
1334 Description: Stack overflow during MDS log replay
1335 Details : ease stack pressure by using a thread dealing llog_process.
1338 Frequency : very rare
1340 Description: MDT cannot be unmounted, reporting "Mount still busy"
1341 Details : Mountpoint references were being leaked during open reply
1342 reconstruction after an MDS restart. Drop mountpoint reference
1343 in reconstruct_open() and free dentry reference also.
1348 Description: wait until IO finished before start new when do lock cancel.
1349 Details : VM protocol want old IO finished before start new, in this case
1350 need wait until PG_writeback is cleared until check dirty flag
1351 and call writepages in lock cancel callback.
1356 Description: mds_mfd_close() ASSERTION(rc == 0)
1357 Details : In mds_mfd_close(), we need protect inode's writecount change
1358 within its orphan write semaphore to prevent possible races.
1361 Frequency : rare, on shutdown ost
1363 Description: don't hit live lock with umount ost.
1364 Details : shrink_dcache_parent can be in long loop with destroy dentries,
1365 use shrink_dcache_sb instead.
1368 Frequency : only when echo_client is used
1370 Description: don't panic with use echo_client
1371 Details : echo client pass NULL as client nid pointer and this produce NULL
1372 pointer dereference.
1375 Frequency : Always on 32-bit PowerPC systems
1377 Description: fix build on PPC32
1378 Details : compile code with -m64 flag produce wrong object file for PPC32.
1383 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1384 Details : In reconstruct_* functions, LASSERTs on both the data supplied
1385 by a client, and the data on disk are dangerous and incorrect.
1386 Change them with client eviction.
1388 Severity : enhancement
1390 Description: skiplist implementation simplification
1391 Details : skiplists are used to group compatible locks on granted list
1392 that was implemented as tracking first and last lock of each lock
1393 group the patch changes that to using doubly linked lists
1397 Description: delete compatibility for 32bit qdata
1398 Details : as planned, when lustre is beyond b1_8, lquota won't support 32bit
1399 qunit. That means servers of b1_4 and servers of b1_8 can't be
1400 used together if users want to use quota.
1403 Frequency : only with administrator action
1405 Description: mount failure if config log has invalid conf_param setting
1406 Details : If administrator specified an incorrect configuration parameter
1407 with "lctl conf_param" this would cause an error during future
1408 client mounts. Instead, ignore the bad configuration parameter.
1411 Frequency : blocks per group < blocksize*8 and uninit_groups is enabled
1413 Description: ldiskfs error: XXX blocks in bitmap, YYY in gd
1414 Details : If blocks per group is less than blocksize*8, set rest of the
1418 Frequency : Application do stride read on lustre
1420 Description: The read performance will drop a lot if the application does
1422 Details : Because the stride_start_offset are missing in stride read-ahead,
1423 it will cause clients read a lot of unused pages in read-ahead,
1424 then the read-performance drops.
1428 Description: more ldlm soft lockups
1429 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1430 starve other threads from the resource lock for a long time in
1431 case of long waiting queue, so change the debug level from
1432 D_OTHER to the less frequently used D_INFO.
1434 Severity : enhancement
1436 Description: add -gid, -group, -uid, -user options to lfs find
1438 Severity : enhancement
1440 Description: ll_recover_lost_found_objs - recover objects in lost+found
1441 Details : OST corruption and subsequent e2fsck can leave objects in the
1442 lost+found directory. Using the "ll_recover_lost_found_objs"
1443 tool, these objects can be retrieved and data can be salvaged
1444 by using the object ID saved in the fid EA on each object.
1449 Description: this bug _only_ happens when inode quota limitation is very low
1450 (less than 12), so that inode quota unit is 1 at initialization.
1451 Details : if remaining quota equates 1, it is a sign to demonstate that quota
1452 is effective now. So least quota qunit should be 2.
1456 Description: Hung threads in invalidate_inode_pages2_range
1457 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1458 one is completed. As a result, some RPCs are stuck in the queue
1463 Description: Procfs and llog threads access destoryed import sometimes.
1464 Details : Sync the import destoryed process with procfs and llog threads by
1465 the import refcount and semaphore.
1469 Description: mds fails to respond, threads stuck in ldlm_completion_ast
1470 Details : Sort source/child resource pair after updating child resource.
1475 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1476 Details : If insertion of an extent fails, then discard the inode
1477 preallocation and free data blocks else it can lead to duplicate
1482 Description: don't always update ctime in ext3_xattr_set_handle()
1483 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle()
1484 In some cases the ctime should not be updated, for example for
1485 2.0->1.8 compatibility it is necessary to delete an xattr and it
1486 should not update the ctime.
1490 Description: add quota statistics
1491 Details : 1. sort out quota proc entries and proc code.
1492 2. add quota statistics
1497 Description: quotas are not honored with O_DIRECT
1498 Details : all writes with the flag O_DIRECT will use grants which leads to
1499 this problem. Now using OBD_BRW_SYNC to guard this.
1503 Bugzilla : 15713/16362
1504 Description: Assertion in iopen_connect_dentry in 1.6.3
1505 Details : looking up an inode via iopen with the wrong generation number can
1506 populate the dcache with a disconneced dentry while the inode
1507 number is in the process of being reallocated. This causes an
1508 assertion failure in iopen since the inode's dentry list contains
1509 both a connected and disconnected dentry.
1513 Description: assertion failure in ldlm_handle2lock()
1514 Details : fix a race between class_handle_unhash() and class_handle2object()
1515 introduced in lustre 1.6.5 by bug 13622.
1517 Severity : enhancement
1519 Description: superblock lock contention with many SMP cores on one client
1520 Details : several client filesystem locks were highly contended on SMP
1521 NUMA systems with 8 or more cores. Per-CPU datastructures
1522 and more efficient locking implemented to reduce contention.
1527 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1528 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1533 Description: oops in sd_iostats_seq_show()
1534 Details : unloading/reloading the scsi low level driver triggers a kernel
1535 bug when trying to access the sd iostat file.
1540 Description: Kernel panics during QLogic driver reload
1541 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1542 patch, causing memory corruption.
1547 Description: journal_dev option does not work in b1_6
1548 Details : pass mount option during pre-mount.
1550 Severity : enhancement
1552 Description: Add a FIEMAP(FIle Extent MAP) ioctl for ldiskfs
1553 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1554 extent information of a file. It can be used to map logical blocks
1555 in a file to physical blocks in the block device.
1558 Frequency : only with adaptive timeout enabled
1560 Description: DEBUG_REQ() bad paging request
1561 Details : ptlrpc_at_recv_early_reply() should not modify req->rq_repmsg
1562 because it can be accessed by reply_in_callback() without the
1566 Frequency : only on Cray X2
1568 Description: X2 build failures
1569 Details : fix build failures on Cray X2.
1573 Description: xid & resent requests
1574 Details : Initialize RPC XID from clock at startup (randomly if clock is
1579 Description: quota recovery deadlock during mds failover
1580 Details : This patch includes att18982, att18236, att18237 in bz14840.
1582 1. fix osts hang when mds does failover with quotaon
1583 2. prevent watchdog storm when osts threads wait for the
1588 Description: kernel panic on racer
1589 Details : Do not access dchild->d_inode when IS_ERR(dchild) is true.
1591 Severity : enhancement
1593 Description: Add lustre_start utility to start or stop multiple Lustre servers
1598 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1599 Details : In case of memory pressure, list_del() can be called twice on
1600 req->rq_history_list, causing a kernel oops.
1604 Description: (ptllnd_peer.c:557:kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1605 Details : fix stack overflow in the distributed lock manager by defering
1606 export eviction after a failed ast to the elt thread instead of
1607 handling it in the dlm interpret routine.
1609 Severity : enhancement
1611 Description: More exported tunables for mballoc
1612 Details : Add support for tunable preallocation window and new tunables for
1613 large/small requests
1617 Description: Detect corruption of block bitmap and checking for preallocations
1618 Details : Checks validity of on-disk block bitmap. Also it does better
1619 checking of number of applied preallocations. When corruption is
1620 found, it turns filesystem readonly to prevent further corruptions.
1624 Frequency : only for big-endian servers
1625 Description: Check for big-endian system while mounting fs with extents feature
1626 Details : Mounting a filesystem with extents feature will fail on big-endian
1627 systems since ext3-based ldiskfs is not supported on big-endian
1628 systems. Can be over-ridden with "bigendian_extents" mount option.
1632 Description: Excessive recovery window
1633 Details : With AT enabled, the recovery window can be excessively long (6000+
1634 seconds). To address this problem, we no longer use
1635 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
1636 timeout no longer depends on the service time, it is set to
1637 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
1638 time via pb_service_time.
1641 Descriptoin: Don't sync journal after every i/o
1642 Details : Implement write RPC replay to allow server replies for write RPCs
1643 before data is on disk. However, this feature is disabled by
1644 default since some issues leading to data corruptions have been
1645 found during recovery (e.g. bug 19128). This feature can be enabled
1646 by running the following command on the OSSs:
1647 lctl set_param obdfilter.*.sync_journal=0
1651 Description: Watchdog triggered on MDS failover
1652 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
1653 the OSTs know that the MDS "UUID" can be reused for the same export
1654 from a different NID, so we do not need to wait for the export to
1659 Descriptoin: Lustre detected file system corruption with inode out of bounds
1660 Details : don't update i_size on MDS_CLOSE for directories. This causes
1661 directory corruptions on the MDT.
1663 -------------------------------------------------------------------------------
1665 2008-05-26 Sun Microsystems, Inc.
1667 * Support for kernels:
1668 2.6.5-7.311 (SLES 9),
1669 2.6.9-67.0.7.EL (RHEL 4),
1670 2.6.16.54-0.2.5 (SLES 10),
1671 2.6.18-53.1.14.el5 (RHEL 5),
1672 2.6.22.14 vanilla (kernel.org)
1673 * Client support for unpatched kernels:
1674 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1675 2.6.16 - 2.6.22 vanilla (kernel.org)
1676 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
1677 we do not recommend using patchless RHEL4 clients with kernels
1678 prior to 2.6.9-55EL (RHEL4U5).
1679 * Recommended e2fsprogs version: 1.40.7-sun1
1680 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1681 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
1682 removed cwd "./" (refer to Bugzilla 14399).
1683 * A new quota file format has been introduced in 1.6.5.
1684 The format conversion from prior releases is handled transparently,
1685 but releases older than 1.4.12/1.6.5 will not understand this new
1686 format. The automatic format conversion can be avoided by running
1687 the following command on the MDS before upgrading:
1688 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
1689 For more information, please refer to bugzilla 13904.
1693 Description: quota performance fix
1694 Details : quota data is written in journalled mode instead of ordered to
1695 increase performance
1699 Description: lfs support for human-readable quota grace time strings
1700 Details : lfs setquota -t and lfs quota -t represent quota grace times
1701 in "XXwXXdXXhXXmXXs" format instead of large values in seconds
1704 Frequency : always with o2ib 1.3 and sles10
1706 Description: fix build with SLES10 and o2ib v3.
1707 Details : sles10 uses diffrent name for Module.symver file but configure
1708 assume this file has same name on RHEL/SLES/vanila kernels.
1711 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1713 Description: MDS may lose file striping (and hence file data) in some cases
1714 Details : If there are additional extended attributes stored on the MDS,
1715 in particular ACLs, SELinux, or user attributes (if user_xattr
1716 is specified for the client mount options) then there is a risk
1717 of attribute loss. Additionally, the Lustre file striping
1718 needs to be larger than default (e.g. striped over all OSTs),
1719 and an additional attribute must be stored initially in the
1720 inode and then increase in size enough to be moved to the
1721 external attribute block (e.g. ACL growing in size) for file
1724 Severity : enhancement
1726 Description: add message levels for liblustreapi
1731 Description: MDT cannot be unmounted, reporting "Mount still busy"
1732 Details : Mountpoint references were being leaked during open reply
1733 reconstruction after an MDS restart. Drop mountpoint reference
1734 in reconstruct_open() and free dentry reference also.
1739 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
1740 Details : Move the 'good_osts' check before the 'total_bavail' check. This
1741 will result in an -EAGAIN and in the exit call path we call
1742 alloc_rr() which will with increasing aggressiveness attempt to
1743 aquire precreated objects on the minimum number of required OSCs.
1747 Description: Use old size assignment to avoid deadlock
1748 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
1749 the scheduling while holding a spinlock. We do not need locking
1750 for size in ll_update_inode() because size is only updated from
1751 the MDS for directories or files without objects, so there is no
1752 other place to do the update, and concurrent access to such inodes
1753 are protected by the inode lock.
1757 Description: Use __u64 instead of int for valid bits
1761 Description: resolve "_IOWR redefined" build error on SLES10
1765 Description: dump the memory debugging after all modules are unloaded to
1766 suppress false negative in conf_sanity test 39
1770 Description: the recovery timer never expires
1771 Details : for new client connect request, the recovery timer should not be
1772 reset, otherwise recovery timer will never expired, if the old
1773 client never come. Only old client connect and first connection
1774 req should trigger recovery timer reset.
1778 Description: the min numbers of lproc stats are wrong
1779 Details : adding a new constant LC_MIN_INIT and use it for initialization
1783 Frequency : always with interactive lfs
1785 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1788 Frequency : with multiple concurrent readdir processes in same directory
1789 Bugzilla : 15406, 15169, 15175
1790 Description: misc fixes for directory readahead.
1791 Details : prevent previous statahead async RPC callback from processing the
1792 current "statahead_info", race condition between async RPC callback
1793 add dentry into dentry hash table and "ls" thread revalidate such
1794 dentry, statahead his/miss control for hidden items, and so on.
1796 Severity : enhancement
1798 Description: build kernel-ib packages for OFED 1.3 in our release cycle
1802 Description: incore types cleaning in quota code (with respect to 64-bit limits)
1803 Details : several u32 variables declarations are replaced with u64 declarations
1808 Description: fix SLES kernel versioning
1809 Details : the kernel version for our SLES 10 kernel did not include a "-"
1810 before the "smp" at the end. while this was not a problem in
1811 general, it did mean that software trying to use the kernel
1812 version to try to detect a vendor specific kernel would fail.
1813 this was most evident by the OFED build scripts.
1818 Description: Don't update lov_desc members until making sure they are valid
1819 Details : When updating lov_desc members via proc fs, need fix their
1820 validities before doing the real update.
1823 Frequency : very rare
1825 Description: don't put request into delay list while invalidate in flight.
1826 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
1827 import in flight. this produce timeout for invalidate and sometimes
1828 can cause stale data.
1830 Severity : enhancement
1832 Description: Update kernel to SLES9 2.6.5-7.311.
1834 Severity : enhancement
1836 Description: Update kernel to RHEL4 2.6.9-67.0.7.
1841 Frequency : on PPC only
1842 Description: not convert ost objects for directory because it's not exist.
1843 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
1845 Severity : enhancement
1847 Description: Fix warnings with compile liblustre at sles10/rhel5 which have
1848 __u64 as usingied long long type.
1851 Frequency : rare, on shutdown
1853 Description: race process ast vs remove callback
1854 Details : removing callback before disconnect import open race with
1855 processing callback.
1857 Severity : enhancement
1859 Description: Update kernel to SLES9 2.6.5-7.311.
1861 Severity : enhancement
1863 Description: Files open for execute are not marked busy on SLES10
1864 Details : Add FMODE_EXEC to SLES10 SP1 server kernel series.
1866 Severity : enhancement
1868 Description: Add server support for vanilla-2.6.22.14.
1871 Frequency : occasional
1873 Description: Avoid lov_create() getting stuck in obd_statfs_rqset()
1874 Details : If an OST is down the MDS will hang indefinitely in
1875 obd_statfs_rqset() waiting for the statfs data. While for
1876 MDS QOS usage of statfs, it should not stuck in waiting.
1878 Severity : enhancement
1880 Description: Disable adaptive timeouts by default
1883 Frequency : on network error
1885 Description: panic with double free request if network error
1886 Details : mdc_finish_enqueue is finish request if any network error occuring,
1887 but it's true only for synchronus enqueue, for async enqueue
1888 (via ptlrpcd) this incorrect and ptlrpcd want finish request
1892 Frequency : rare, on recovery
1894 Description: read procfs can produce deadlock in some situation
1895 Details : Holding lprocfs lock with send rpc can produce block for destroy
1896 obd objects and this also block reconnect with -EALREADY.
1897 This isn't fix all lprocfs bugs - but make it rare.
1899 Severity : enhancement
1901 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
1904 Frequency : frequent on X2 node
1906 Description: mdc_set_open_replay_data LBUG
1907 Details : Set replay data for requests that are eligible for replay.
1912 Description: lustre_mgs: operation 101 on unconnected MGS
1913 Details : When MGC is disconnected from MGS long enough, MGS will evict the
1914 MGC, and late on MGC cannot successfully connect to MGS and a lot
1915 of the error messages complaining that MGS is not connected.
1918 Frequency : rare, depends on device drivers and load
1920 Description: MDS or OSS nodes crash due to stack overflow
1921 Details : Code changes in 1.6.4 increased the stack usage of some functions.
1922 In some cases, in conjunction with device drivers that use a lot
1923 of stack, the MDS (or possibly OSS) service threads could overflow
1924 the stack. One change which was identified to consume additional
1925 stack has been reworked to avoid the extra stack usage.
1927 Severity : enhancement
1929 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1931 Severity : enhancement
1933 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1935 Severity : enhancement
1937 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1939 Serverity : enhancement
1941 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1944 Frequency : rare on shutdown OST
1946 Description: Don't allow skipping OSTs if index has been specified.
1947 Details : Don't allow skipping OSTs if index has been specified, make locking
1948 in internal create lots better.
1953 Description: ASSERTION(!PageDirty(page)) failed
1954 Details : Wrong check could lead to an assertion failure under specific
1960 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1961 Details : access to bitfield in structure is always rounded to long
1962 and this produce problem with not atomic change any bit.
1967 Description: Lustre make rpms failed.
1968 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1969 builds Lustre rpms from tarball.
1971 Severity : enhancement
1973 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
1976 Frequency : rare on shutdown OST
1978 Description: If llog cancel was not send before clean_exports phase, this can
1979 produce deadlock in llog code.
1980 Details : If llog thread has last reference to obd and call class_import_put
1981 this produce deadlock because llog_cleanup_commit_master wait when
1982 last llog_commit_thread exited, but this never success because was
1983 called from llog_commit_thread.
1986 Frequency : only if OST index is skipped
1988 Description: NULL lov_tgts causing MDS oops
1989 Details : more safe checks for NULL lov_tgts for avoid oops.
1991 Severity : enhancement
1993 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1995 Severity : enhancement
1997 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
2002 Description: make mgs_setparam() handle fsname containing dash
2003 Details : fsname containing a dash does not work with lctl conf_param
2005 Severity : enhancement
2007 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
2010 Frequency : rare, in recovery and (or) destroy lovobjid file.
2012 Description: rewrite lov objid code.
2013 Details : Cleanup for lov objid code, remove scability problems and wrong
2014 locking. Fix sending last_id into ost.
2016 Severity : enhancement
2018 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
2020 Severity : enhancement
2022 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
2023 Details : Use d_move_locked instead of __d_move.
2026 Frequency : rare, at shutdown
2028 Description: access already free / zero obd_namespace.
2029 Details : if client_disconnect_export was called without force flag set,
2030 and exist connect request in flight, this can produce access to
2031 NULL pointer (or already free pointer) when connect_interpret
2032 store ocd flags in obd_namespace.
2035 Frequency : only at startup
2037 Description: not alloc memory with spinlock held.
2038 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
2039 if any spinlock held.
2044 Description: lfs find does not continue on file error
2045 Details : Continue other files processing when a file/dir is absent.
2049 Description: Inconsistent usage of lustre_pack_reply()
2050 Details : Standardize the usage of lustre_pack_reply() such that it
2051 always generate a CERROR on failure.
2054 Frequency : very rare
2056 Description: Fix replay if there is an un-replied request and open
2057 Details : In some cases, older replay request will revert the
2058 mcd->mcd_last_xid on MDS which is used to record the client's
2059 latest sent request.
2061 Severity : enhancement
2063 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
2065 Severity : enhancement
2067 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
2069 Severity : enhancement
2071 Description: Update to SLES9 kernel-2.6.5-7.287.3.
2073 Severity : enhancement
2075 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
2077 Severity : enhancement
2079 Description: Build SLES10 patchless client fails
2080 Details : The configure was broken by run ./configure with
2081 --with-linux-obj=.... argument for patchless client. When the
2082 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
2083 can't search header adequately. Use absolute path such as
2084 -I($LINUX)/include instead.
2086 Severity : enhancement
2088 Description: Lustre Page Accounting
2089 Details : New macros for page alloc and free which enable accounting
2090 of page allocation of Lustre. Use percpu counters to store memory
2091 and page statistics.
2094 Frequency : only if debugging is disabled
2096 Description: LASSERT_{REQ,REP}SWAB macros are buggy
2097 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
2098 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
2099 these macros and replace them with their definitions instead.
2104 Description: interrupt oig_wait produce painc on resend.
2105 Details : brw_redo_request can be used for resend requests from ptlrpcd and
2106 private set, and this produce situation when rq_ptlrpcd_data not
2107 copyed to new allocated request and triggered LBUG on assert
2108 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
2109 ptlrpcd set if request is changed and can be safety changed to use
2112 Severity : enhancement
2114 Description: organize the server-side client stats on per-nid basis
2115 Details : Change the structure of stats under obdfilter and mds to
2123 The "uuid"s file would list the uuids of _active_ exports.
2124 And the clear entry is to clear all stats and stale nids.
2129 Description: Processes looping in ll_readdir() on Lustre clients finally causing
2130 a full node pseudo-hang
2131 Details : Concurrent access to the same directory from multiple clients with
2132 intensive file creation/removal can cause a client node to spin in
2133 ll_readdir(). i_version must be increased every time the lock is
2134 cancelled to ensure a revalidate is done.
2139 Description: touch file failed when fs is not full
2140 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
2141 otherwise we can get ENOSP while fs is not full.
2144 Frequency : only for Cray XT3
2145 Bugzilla : 12829/13455
2146 Description: Changing primary group doesn't change the group lustre assigns to
2148 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
2149 overridden with the primary group provided by the group upcall,
2150 whereas the supplied fsgid can be trusted if it is in the list of
2151 supplementary groups returned by the group upcall.
2153 Severity : enhancement
2155 Description: Root Squash Functionality
2156 Details : Implementation of NFS-like root squash capability. Specifically,
2157 don't allow someone with root access on a client node to be able
2158 to manipulate files owned by root on a server node.
2160 Severity : enhancement
2162 Description: Slow trucate/writes to huge files at high offsets.
2163 Details : Directly associate cached pages to lock that protect those pages,
2164 this allows us to quickly find what pages to write and remove
2165 once lock callback is received.
2170 Description: Too many locks accumulating on client during NFS usage
2171 Details : mds_open improperly used accmode to find out access mode to a
2172 file. Also mdc_intent_lock logic to find out if we already have
2173 lock similar to just received was flawed since introduction of
2174 skiplists - locks are now added to the front of the granted
2179 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
2181 Details : In obd setup failure handler, obd_stopping will not necessarily to
2182 be 1, and obd_set_up should also be checked to make sure whether
2183 obd is completely setup.
2185 Severity : enhancement
2187 Description: Allow masking D_WARNING, D_ERROR messages from console
2188 Details : Console messages can now be disabled via lnet.printk.
2193 Description: User code with malformed file open parameter crashes client node
2194 Details : Before packing join_file req, all the related reference should be
2195 checked carefully in case some malformed flags cause fake join_file
2201 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
2202 grain of quota doing harm to quota's accuracy
2203 Details : qunit size will be changed when quota limitation is too low/high;
2204 record the pending quota write in order to get more accureate
2205 quota; delete the patch for bug12588, which is unnecessary when
2206 this patch is landed. This bug also contains fixes for bug 14526,
2207 14299, 14601 and 13794.
2211 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
2212 Details : ldlm_completion_ast() assumes that a lock is granted when the req
2213 mode is equal to the granted mode. However, it should also check
2214 that LDLM_FL_CP_REQD is not set.
2218 Description: Heavy nfs access might result in deadlocks
2219 Details : After ELC code landed, it is now improper to enqueue any mds
2220 locks under och_sem, because enqueue might want to decide to
2221 cancel open locks for same inode we are holding och_sem for.
2225 Description: 35% write performance drop with ldiskfs2 when quotas are on
2226 Details : Enable ext3 journalled quota by default to improve performance
2227 when quotas are turned on.
2231 Description: Client eviction while running blogbench
2232 Details : A lot of unlink operations with concurrent I/O can lead to a
2233 deadlock causing evictions. To address the problem, the number of
2234 oustanding OST_DESTROY requests is now throttled to
2235 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
2236 callbacks are processed in priority.
2239 Frequency : RHEL4 only
2241 Description: mkfs is very slow on IA64/RHEL4
2242 Details : A performance regression has been discovered in the MPT Fusion
2243 driver between versions 3.02.73rh and 3.02.99.00rh. As a
2244 consequence, we have downgraded the MPT Fusion driver in the RHEL4
2245 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
2248 Frequency : PPC/PPC64 only
2250 Description: conflicts between asm-ppc64/types.h and lustre_types.h
2251 Details : fix duplicated definitions between asm-ppc64/types.h and
2252 lustre_types.h on PPC.
2255 Frequency : PPC/PPC64 only
2257 Description: asm-ppc/segment.h does not exist
2258 Details : fix compile issue on PPC.
2262 Description: data checksumming impacts single node performance
2263 Details : add support for several checksum algorithms. Currently, CRC32 and
2264 Adler-32 are supported. The checksum type can be changed on the fly
2265 through /proc/fs/lustre/osc/*/checksum_type.
2269 Description: use adler32 for page checksums
2270 Details : when available, use the Adler-32 algorithm instead of CRC32 for
2275 Description: better handle error messages in extents code
2277 Severity : enhancement
2279 Description: SNMP support enhancement
2280 Details : Adding total number of sampled request for an MDS node in snmp
2283 Severity : enhancement
2285 Description: Optimize ldlm waiting list processing for PR extent locks
2286 Details : When processing waiting list for read extent lock and meeting read
2287 lock that is same or wider to it that is not contended, skip
2288 processing rest of the list and immediatelly return current
2289 status of conflictness, since we are guaranteed there are no
2290 conflicting locks in the rest of the list.
2294 Description: Time out and refuse to reconnect
2295 Details : When the failover node is the primary node, it is possible
2296 to have two identical connections in imp_conn_list. We must
2297 compare not conn's pointers but NIDs, otherwise we
2298 can defeat connection throttling.
2302 Description: Client not clear own cache if answer to reconnect is lost.
2303 Details : Client gets evicted from server. Now client also thinks it is
2304 disconnected (or gets ENOTCONN on its operation) and decides to
2305 reconnect. Server receives reconnect message, but cannot find
2306 export. New export is created that is fully valid (new cookie!),
2307 but reply is lost and not reported to client. Client reconnects
2308 again and gets back a just-created connection, but it is not new
2309 so client thinks it was not evicted and does not do recovery.
2313 Description: Detect stride IO mode in read-ahead
2314 Details : When a client does stride read, read-ahead should detect that and
2315 read-ahead pages according to the detected stride pattern.
2319 Description: build for x2 fails
2320 Details : fix compile issue on Cray systems.
2322 Severity : enhancement
2324 Description: implement readv/writev APIs(aio_read/aio_writes in newer kernels)
2325 Details : This greatly improves speed of NFS writes on 2.6 kernels.
2328 Frequency : only on PPC/SLES10
2330 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
2331 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
2332 be defined. Add a hack in mkfs_lustre.c to work around this compile
2337 Description: LASSERT on MDS when client holding flock lock dies
2338 Details : ldlm pool logic depends on number of granted locks equal to
2339 number of released locks which is not true for flock locks, so
2340 just exclude such locks from consideration.
2344 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
2345 Details : Use fsfilt_sync() for both the whole filesystem sync and
2346 individual file sync to eliminate dangerous inode locking
2347 with I_LOCK that can lead to a deadlock.
2351 Description: Update an obsolete wirecheck.c generator
2352 Details : Update wirecheck.c/wirehdr.c and regenerate wiretest.c
2356 Description: Client can panic on open sometimes
2357 Details : It is possible that we try to free already freed request in
2358 ll_file_open in some error cases when we send request from
2363 Description: performance in 1.6.3
2364 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
2365 sure that 1MB requests are not fragmented by the block layer.
2369 Description: LDLM soft lockups - improvement
2370 Details : It is be possible to send the lock handle along with each read
2371 or write request because the client is already doing a lock match
2372 itself so there isn't any reason the OST should have to re-do that
2378 Description: lfs quota fails with deactivated OSTS
2379 Details : With this patch, three improvements are included:
2380 1. detete the softlimit in mds and osts when use "lfs quota".
2381 2. display the inaccurate data in the output of "lfs quota".
2382 3. try to get quota info when "lfs quota" is executed.
2387 Description: Extent locks not granted with no conflicts sometimes.
2388 Details : When race occurs in glimpse handler and nothing is returned,
2389 we do not reprocess the queue after lock cancel, and that leads
2390 to a stall until next activity on a resource
2393 Frequency : failover with quotaon
2395 Description: during mds failovers with quota on, OSTs got into deadlock state
2396 and causing dumpstack.
2397 Details : for every quota slave, at any time, there is only one quota req
2398 is sent to quota master for every uid/gid. Before that quota req
2399 returns, all the thread relative to the same uid/gid will wait.
2400 So if the quota req is lost because mds failovers or any other
2401 reasons, this bug will be hit. Now, dqacq_interpret() will handle
2402 quota reqs who time out.
2404 Severity : enhancement
2407 Description: when quota slave checks if quota is enough, there is an unnecessary
2409 Details : place this wait on necessary place instead of always waiting.
2411 --------------------------------------------------------------------------------
2413 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
2415 * Support for kernels:
2416 2.6.5-7.286 (SLES 9),
2417 2.6.9-55.0.9.EL (RHEL 4),
2418 2.6.16.53-0.8 (SLES 10),
2419 2.6.18-8.1.14.el5 (RHEL 5),
2420 2.6.18.8 vanilla (kernel.org)
2421 * Client support for unpatched kernels:
2422 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2423 2.6.16 - 2.6.22 vanilla (kernel.org)
2424 * Due to recently discovered recovery problems, we do not recommend
2425 using patchless RHEL 4 clients with this or any earlier release.
2426 * Recommended e2fsprogs version: 1.40.2-cfs1
2427 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2430 Frequency : occasional
2432 Description: MDS hang or stay in waiting lock
2433 Details : If client receive lock with CBPENDING flag ldlm need send lock
2434 cancel as separate rpc, to avoid situation when cancel request
2435 can't processed due all i/o threads stay in wait lock.
2438 Frequency : occasional
2440 Description: Do not fail import if osc_interpret_create gets -EAGAIN
2441 Details : If osc_interpret_create got -EAGAIN it immediately exits and
2442 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
2443 oscc_has_objects and see OSC has no objests and call
2444 oscc_internal_create to resend create request.
2446 Severity : enhancement
2448 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
2449 Details : Update which_patch & target file for SLES10 latest kernel.
2451 Severity : enhancement
2453 Description: add --type and --size parameters to lfs find
2454 Details : Enhance lfs find by adding filetype and filesize parameters. Also
2455 multiple OBDs can now be specified for the --obd option.
2457 Severity : enhancement
2459 Description: eliminate client locks in face of contention
2460 Details : file contention detection and lockless i/o implementation
2461 for contended files.
2463 Severity : enhancement
2465 Description: Remove client patches from SLES 10 kernel.
2466 Details : This causes SLES 10 clients to behave as patchless clients
2467 even on a Lustre-patched (server) kernel.
2469 Severity : enhancement
2471 Description: use i_size_read and i_size_write in 2.6 port
2472 Details : replace inode->i_size access with i_size_read/write()
2475 Frequency : when removing large files
2477 Description: scheduling issue during removal of large Lustre files
2478 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
2479 It causes scheduling issues when removing large files (17TB in the
2485 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
2486 Details : Impossible (EPROTO is returned) to access a directory that has a
2487 non-default striping and ACLs.
2490 Frequency : only on ppc
2492 Description: /proc/fs/lustre/devices broken on ppc
2493 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
2494 We should make sure the type of 'index' is loff_t and then cast
2495 explicitly as needed below. Do not assign an explicitly cast
2499 Frequency : only for rhel5
2501 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
2502 Details : Modify the target file & which_kernel.
2505 Frequency : if the uninit_groups feature is enabled on ldiskfs
2507 Description: e2fsck reports "invalid unused inodes count"
2508 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
2509 feature and only a single inode is created in a group then the
2510 "bg_unused_inodes" count is incorrectly updated. Creating a
2511 second inode in that group would update it correctly.
2516 Description: buffer overruns could theoretically occur
2517 Details : llapi_semantic_traverse() modifies the "path" argument by
2518 appending values to the end of the origin string, and a buffer
2519 overrun may occur. Adding buffer overrun check in liblustreapi.
2521 Severity : enhancement
2523 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
2526 Frequency : only if filesystem is inconsistent
2528 Description: handle "serious error: objid * already exists" more gracefully
2529 Details : If LAST_ID value on disk is smaller than the objects existing in
2530 the O/0/d* directories, it indicates disk corruption and causes an
2531 LBUG(). If the object is 0-length, then we should use the existing
2532 object. This will help to avoid a full fsck in most cases.
2534 Severity : enhancement
2536 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
2537 Details : Modify vm-tunables-rhel4.patch.
2539 Severity : enhancement
2541 Description: Kernel config for 2.6.18-vanilla.
2542 Details : Modify targets/2.6-vanilla.target.in.
2543 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
2544 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
2545 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
2546 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
2549 Frequency : occasional
2551 Description: improve handling recoverable errors
2552 Details : If request processed with error which can be recoverable on server
2553 request should be resend, otherwise page released from cache and
2559 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
2560 Details : Modify target file & which_patch.
2561 A flaw was found in the IA32 system call emulation provided
2562 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
2563 value could be stored in the %RAX register, which could trigger an
2564 out-of-bounds system call table access. An untrusted local user
2565 could exploit this flaw to run code in the kernel
2566 (ie a root privilege escalation). (CVE-2007-4573).
2570 Description: change order of libsysio includes
2571 Details : '#include sysio.h' should always come before '#include xtio.h'
2573 Severity : enhancement
2575 Description: adapt the lustre_config script to support the upgrade case
2576 Details : Add "-u" option for lustre_config script to support upgrading 1.4
2577 server targets to 1.6 in parallel.
2582 Description: To avoid grant space > avaible space when the disk is almost
2583 full. Without this patch you might see the error "grant XXXX >
2584 available" or some LBUG about grant, when the disk is almost
2586 Details : In filter_check_grant, for non_grant cache write, we should
2587 check the left space by if (*left > ungranted + bytes), instead
2588 of (*left > ungranted), because only we are sure the left space
2589 is enough for another "bytes", then the ungrant space should be
2590 increase. In client, we should update cl_avail_grant only there is
2591 OBD_MD_FLGRANT in the reply.
2596 Description: Update RHEL 4 kernel to fix local root privilege escalation.
2597 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
2598 described in CVE-2007-4573. This problem could allow untrusted
2599 local users to gain root access.
2602 Frequency : when using O_DIRECT and quotas
2604 Description: Incorrect file ownership on O_DIRECT output files
2605 Details : block usage reported by 'lfs quota' does not take into account
2606 files that have been written with O_DIRECT.
2610 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2611 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2612 window to 0. If it does hit cache miss, and read-pattern is still
2613 stride-io mode, does not reset the stride window, but also does not
2614 increase the stride window length in this case.
2616 --------------------------------------------------------------------------------
2618 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
2620 * Support for kernels:
2621 2.6.5-7.286 (SLES 9),
2622 2.6.9-55.0.2.EL (RHEL 4),
2623 2.6.16.46-0.14 (SLES 10),
2624 2.6.18-8.1.8.el5 (RHEL 5),
2625 2.6.18.8 vanilla (kernel.org)
2626 * Client support for unpatched kernels:
2627 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2628 2.6.16 - 2.6.21 vanilla (kernel.org)
2629 * Due to recently discovered recovery problems, we do not recommend
2630 using patchless RHEL 4 clients with this or any earlier release.
2631 * Recommended e2fsprogs version: 1.40.2-cfs1
2632 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2636 Description: Fix errors in lfs documentation
2637 Details : Fixes man pages
2639 Severity : enhancement
2641 Description: Adaptive timeouts
2642 Details : RPC timeouts adapt to changing server load and network
2643 conditions to reduce resend attempts and improve recovery time.
2645 Severity : enhancement
2647 Description: llapi_file_create() does not allow some changes
2648 Details : add llapi_file_open() that allows specifying the file creation
2649 mode and open flags, and also returns an open file handle.
2651 Severity : enhancement
2653 Description: df doesn't work properly if diskfs blocksize != 4K
2654 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
2657 Frequency : always on directories with default striping set
2659 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
2660 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
2662 Severity : enhancement
2664 Description: merge and cleanup kernel patches.
2665 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
2667 Severity : enhancement
2669 Description: RedHat Update kernel for RHEL5
2670 Details : Kernel config file for RHEL5.
2672 Severity : enhancement
2674 Description: OSS needs mutliple precreate threads
2675 Details : Add ability to start more than one create thread per OSS.
2678 Frequency : only with quota on the root user
2680 Description: mds_obd_create error creating tmp object
2681 Details : When the user sets quota on root, llog will be affected and can't
2682 create files and write files.
2685 Frequency : Always on ia64 patchless client, and possibly others.
2687 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
2688 Details : This allows the patchless client to be loaded on architectures
2689 without this export.
2694 Description: disorder of journal start and llog_add cause deadlock.
2695 Details : in llog_origin_connect, journal start should happen before
2696 llog_add keep the same order as other functions to avoid
2699 Severity : enhancement
2701 Description: RedHat Update kernel for RHEL5
2702 Details : Modify the kernel config file more closer RHEL5.
2704 Severity : enhancement
2706 Description: Build failure against Centos5 (RHEL5)
2707 Details : Define PAGE_SIZE when it isn't present.
2710 Frequency : occasionally when using NFS
2712 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
2713 Details : with NFS, the anon dentry's parent was set to itself in
2714 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
2715 corresponding dentry other than use rec->ur_name.
2717 Severity : enhancement
2719 Description: enable data checksumming by default
2720 Details : enable checksum by default, allow --disable-checksum
2721 configure option and "-o nochecksum" mount option. Checksums
2722 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
2725 Frequency : Occasionally with failover
2727 Description: Client eviction due to failover config
2728 Details : after a connection loss, the lustre client should attempt to
2729 reconnect to the last active server first before trying the
2730 other potential connections.
2732 Severity : enhancement
2734 Description: client-side metadata stat-ahead during readdir(directory readahead)
2735 Details : perform client-side metadata stat-ahead when the client detects
2736 readdir and sequential stat of dir entries therein
2741 Description: when mds and osts use different quota unit(32bit and 64bit),
2742 quota will be released repeatly.
2743 Details : void sending multiple quota reqs to mds, which will keep the
2744 status between the reqs.
2747 Frequency : only with liblustre clients on XT3
2749 Description: evictions taking too long
2750 Details : allow llrd to evict clients directly on OSTs
2754 Description: osts not allocated evenly to files
2755 Details : change the condition to increase offset_idx
2757 Severity : enhancement
2759 Description: self-adjustable client's lru lists
2760 Details : use adaptive algorithm for managing client cached locks lru
2761 lists according to current server load, other client's work
2762 pattern, memory activities, etc. Both, server and client
2763 side namespaces provide number of proc tunables for controlling
2768 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
2769 Details : rewrite inherited code (uuid parsing code from ext2 utils and
2770 readdir code from ext3) from scratch preserving functionality.
2774 Description: Only those disconnect error should be returned by rq_status.
2775 Details : In open/enqueue processs, Some errors, which will cause client
2776 disconnected, should be returned by rq_status, while other
2777 errors should still be returned by intent, then mdc or llite will
2780 Severity : enhancement
2782 Description: Tune the kernel for good SCSI performance.
2783 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
2784 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
2788 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
2789 Bugzilla : 13547, 13627
2790 Description: Data corruption for OSTs that are formatted larger than 2TB
2792 Details : When generating the bio request for lustre file writes the
2793 sector number would overflow a temporary variable before being
2794 used for the IO. The data reads correctly from Lustre (which
2795 will overflow in a similar manner) but other file data or
2796 filesystem metadata may be corrupted in some cases.
2800 Description: "lfs find -obd UUID" prints directories
2801 Details : "lfs find -obd UUID" will return all directory names instead
2802 of just file names. It is incorrect because the directories
2803 do not reside on the OSTs.
2806 Bugzilla : 13596, 13614
2807 Description: MDS hang after unclean shutdown of lots of clients
2808 Details : Never resend AST requests.
2812 Frequency : Always, for kernels after 2.6.16
2813 Description: Fix warning idr_remove called for id=.. which is not allocated.
2814 Details : Last kernels save old s_dev before kill super and not allow
2815 to restore from callback - restore it before call kill_anon_super.
2819 Description: Fix error on 'ls .' at the top of the Lustre mount.
2820 Details : Don't revalidate dentry if it is a root dentry.
2822 --------------------------------------------------------------------------------
2824 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
2826 * Support for kernels:
2827 2.6.5-7.286 (SLES 9),
2828 2.6.9-55.0.2.EL (RHEL 4),
2829 2.6.16.46-0.14 (SLES 10),
2830 2.6.18-8.1.8.el5 (RHEL 5),
2831 2.6.18.8 vanilla (kernel.org)
2832 * Client support for unpatched kernels:
2833 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2834 2.6.16 - 2.6.22 vanilla (kernel.org)
2835 * Due to recently discovered recovery problems, we do not recommend
2836 using patchless RHEL 4 clients with this or any earlier release.
2837 * Recommended e2fsprogs version: 1.39.cfs8
2838 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2843 Description: block reactivating mgc import until all deactivates complete
2844 Details : Fix race when failing back MDT/MGS to itself (testing)
2846 Severity : enhancement
2848 Description: lfs setstripe enhancement
2849 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
2852 Frequency : mds/oss recovery
2854 Description: llog ctxt is refrenced after it has been freed.
2855 Details : llog ctxt refcount was added to avoide the race between ctxt free
2856 and llog recovery process. Each llog user must hold ctxt refcount
2857 before it access the llog. And the llog ctxt can only be freed
2858 when its refcount is zero.
2860 Severity : enhancement
2862 Description: randomly memory allocation failure util
2863 Details : Make lustre randomly failed allocating memory for testing purpose.
2866 Frequency : only for SLES10
2868 Description: Update kernel patch for SLES10 SP1
2869 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
2871 Severity : enhancement
2873 Description: omit set fsid for export NFS
2874 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
2875 as FS_REQUIRES_DEV for avoid problems with generate fsid.
2878 Frequency : after network failures
2880 Description: Add sync option to mount_lustre.c
2881 Details : Client loses data written to lustre after a network interruption.
2883 Severity : enhancement
2885 Description: Error message improvement.
2886 Details : Merging of two LCONSOLE_ERROR_MSG into one.
2888 Severity : enhancement
2890 Description: don't use GFP_* in generic Lustre code.
2891 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
2893 Severity : enhancement
2895 Description: obdclass is limited by single OBD_ALLOC(idarray)
2896 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
2898 Severity : enhancement
2900 Description: updated patchess for new RHEL4 kernel
2901 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
2903 Severity : enhancement
2905 Description: warnings with build patchless client with vanila 2.6.19 and up
2906 Details : change old ctl_table style and replace ctl_table/ctl_table_header
2907 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
2909 Severity : enhancement
2911 Description: O_DIRECT bypasses client statistics.
2912 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
2913 read_ahead_stats were updated. Copied stats section from
2914 osc_send_oap_rpc() into async_internal().
2917 Frequency : only for Cray XT3
2919 Description: peer credits not enough on many OST per OSS systems.
2920 Details : Use new lnet way to add credits as we need those for pings and ASTs
2923 Frequency : only with liblustre
2925 Description: Liblustre is not releasing flock locks on file close.
2926 Details : Release flock locks on file close.
2929 Frequency : only for RHEL4
2931 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
2932 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
2933 Update target file and kernel config.
2937 Description: lustre support for RHEL5
2938 Details : Add support for RHEL5.
2943 Description: ASSERTION(export != NULL) failed in target_handle_connect
2944 Details : Assetion hit is result of rare race between disconnect and connet
2945 to same nid. target_handle_connect found old connect cockie and
2946 tried to reconnect, but can't find export for this cockie.
2951 Description: umount blocks forever on error
2952 Details : In result of wrong using obd_no_recov and obd_force flags client
2953 can hand if cancel or some other requests is lost.
2956 Frequency : Only for SLES
2958 Description: sanity_quota fail test_1
2959 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
2960 file, which makes TSTID[2] inunique.
2962 Severity : enhancement
2964 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
2965 Details : Update target/ChangeLog/which_patch .
2967 Severity : enhancement
2969 Description: jbd statistics
2970 Details : Port older jbd statistics patch for sles10
2975 Description: lvbo_init failed for resource with missing objects.
2976 Details : Fix returning error if we do stat for file with missing/corrupted
2977 objects and i_size set to all sum of size of all avaible objects.
2978 if we to truncate/write to missing object - it is recreated.
2983 Description: Oops in read and write path when failing to allocate lock.
2984 Details : Check if lock allocation failed and return error back.
2987 Frequency : When flocks are used.
2989 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
2990 Details : Flock locks might destroy just granted lock if it could be merged
2991 with another existing flock, this is done in completion handler,
2992 so teach ldlm_cli_enquque_fini that this is a valid case for
2998 Description: reply_lock_interpret crash due to race with it and lock cancel.
2999 Details : Do not replay locks that are being cancelled. Do not reference
3000 locks by their address during replay, just by their handle.
3003 Frequency : only with deactivated OSTs
3005 Description: lstripe command fails for valid OST index
3006 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
3007 instead of lov->desc.ld_active_tgt_count.
3009 Severity : enhancement
3011 Description: Build failure against Centos5 (RHEL5)
3012 Details : Use getpagesize() instead of PAGE_SIZE.
3014 --------------------------------------------------------------------------------
3016 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
3018 * Support for kernels:
3019 2.6.5-7.283 (SLES 9),
3020 2.6.9-55.EL (RHEL 4),
3021 2.6.16.46-0.14 (SLES 10),
3022 2.6.18.8 vanilla (kernel.org)
3023 * Client support for unpatched kernels:
3024 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3025 2.6.16 - 2.6.22 vanilla (kernel.org)
3026 * Due to recently discovered recovery problems, we do not recommend
3027 using patchless RHEL 4 clients with this or any earlier release.
3028 * Recommended e2fsprogs version: 1.39.cfs8
3029 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3030 * Starting with this release, the ldiskfs backing filesystem required
3031 by Lustre is now in its own package, lustre-ldiskfs. This package
3032 should be installed. It is versioned separately from Lustre and
3033 may be released separately in future.
3038 Description: server LBUG when shutting down
3039 Details : Block umount forever until the mount refcount is zero rather
3040 than giving up after an arbitrary timeout.
3042 Severity : enhancement
3044 Description: add optional extra BUILD_VERSION info
3045 Details : add a new environment variable (namely LUSTRE_VERS) which allows
3046 to override the lustre version.
3049 Frequency : 2.6.18 servers only
3051 Description: ll_kern_mount() doesn't release the module reference
3052 Details : The ldiskfs module reference count never drops down to 0
3053 because ll_kern_mount() doesn't release the module reference.
3058 Description: server LBUG when using old ost_num_threads parameter
3059 Details : Accept the old ost_num_threads parameter but warn that it
3060 is deprecated, and fix an off-by-one error that caused an LBUG.
3065 Description: Transient SCSI error results in persistent IO issue
3066 Details : iobuf->dr_error is not reinitialized to 0 between two
3070 Frequency : sometimes when underlying device returns I/O errors
3072 Description: OSTs not going read-only during write failures
3073 Details : OSTs are not remounted read-only when the journal commit threads
3074 get I/O errors because fsfilt_ext3 calls journal_start/stop()
3075 instead of the ext3 wrappers.
3078 Frequency : only with 10000 clients or more
3080 Description: poor connect scaling with increasing client count
3081 Details : Don't run filter_grant_sanity_check for more than 100 exports
3082 to improve scaling for large numbers of clients.
3085 Frequency : SLES10 only
3087 Description: sanity-quota.sh quotacheck failed: rc = -22
3088 Details : Quotas cannot be enabled on SLES10.
3091 Frequency : liblustre clients only
3093 Description: getdirentries does not give error when run on compute nodes
3094 Details : getdirentries does not fail when the size specified as an argument
3095 is too small to contain at least one entry
3097 Severity : enhancement
3099 Description: Add LNET router traceability for debug purposes
3100 Details : If a checksum failure occurs with a router as part of the
3101 IO path, the NID of the last router that forwarded the bulk data
3102 is printed so it can be identified.
3107 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
3108 Details : Due to a race condition, liblustre clients were occasionally
3109 evicted incorrectly.
3111 Severity : enhancement
3113 Description: lfs setstripe use optional parameters instead of postional
3116 Severity : enhancement
3118 Description: Nanosecond timestamp support for ldiskfs
3119 Details : The on-disk ldiskfs filesystem has added support for nanosecond
3120 resolution timestamps. There is not yet support for this at
3121 the Lustre filesystem level.
3124 Frequency : during server recovery
3126 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
3127 Details : request with rq_no_resend flag not awake l_wait_event if they get a
3131 Frequency : nfs export on patchless client
3133 Description: connectathon hang when test nfs export over patchless client
3134 Details : Disconnected dentry cannot be found with lookup, so we do not need
3135 to unhash it or make it invalid
3138 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
3139 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
3140 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
3145 Description: open req refcounting wrong on reconnect
3146 Details : If reconnect happened between getting open reply from server and
3147 call to mdc_set_replay_data in ll_file_open, we will schedule
3148 replay for unreferenced request that we are about to free.
3149 Subsequent close will crash in variety of ways.
3150 Check that request is still eligible for replay in
3151 mdc_set_replay_data().
3156 Description: disable writes to filesystem when reading health_check file
3157 Details : the default for reading the health_check proc file has changed
3158 to NOT do a journal transaction and write to disk, because this
3159 can cause reads of the /proc file to hang and block HA state
3160 checking on a healthy but otherwise heavily loaded system. It
3161 is possible to return to the previous behaviour during configure
3162 with --enable-health-write.
3164 Severity : enhancement
3166 Description: 64-bit inode version
3167 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
3168 to the inode. This will be required for version-based recovery.
3173 Description: MDS fails to start if a duplicate client export is detected
3174 Details : in some rare cases it was possible for a client to connect to
3175 an MDS multiple times. Upon recovery the MDS would detect this
3176 and fail during startup. Handle this more gracefully.
3178 Severity : enhancement
3180 Description: Add -o localflock option to simulate old noflock behaviour.
3181 Details : This will achieve local-only flock/fcntl locks coherentness.
3186 Description: log_commit_thread vs filter_destroy race leads to crash
3187 Details : Take import reference before releasing llog record semaphore
3192 Description: Wrong request locking in request set processing
3193 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
3194 imp_delayed_list, in this place should be used imp_lock.
3197 Frequency : when reconnecting
3199 Description: Grant leak when OSC reconnect to OST
3200 Details : When osc reconnect ost, OST(filter) should check whether it
3201 should grant more space to client by comparing fed_grant and
3202 cl_avail_grant, and return the granted space to client instead
3203 of "new granted" space, because client will call osc_init_grant
3204 to update the client grant space info.
3207 Frequency : when client reconnects to OST
3209 Description: Grant leak when OSC does a resend and replays bulk write
3210 Details : When osc reconnect to OST, OST(filter) should clear grant info of
3211 bulk write request, because the grant info while be sync between
3212 OSC and OST when reconnect, and we should ignore the grant info
3213 these of resend/replay write req.
3218 Description: Grant space more than avaiable space sometimes.
3219 Details : When then OST is about to be full, if two bulk writing from
3220 different clients came to OST. Accord the avaliable space of the
3221 OST, the first req should be permitted, and the second one
3222 should be denied by ENOSPC. But if the seconde arrived before
3223 the first one is commited. The OST might wrongly permit second
3224 writing, which will cause grant space > avaiable space.
3227 Frequency : when client is evicted
3229 Description: Grant might be wrongly erased when osc is evicted by OST
3230 Details : when the import is evicted by server, it will fork another
3231 thread ptlrpc_invalidate_import_thread to invalidate the
3232 import, where the grant will be set to 0. While the original
3233 thread will update the grant it got when connecting. So if
3234 the former happened latter, the grant will be wrongly errased
3235 because of this race.
3240 Description: Checking Stale with correct fid
3241 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
3242 to check whether it is stale, because sometimes, we want the
3243 enqueue happened anyway, and op_data.fid2 will not be initialized.
3245 Severity : enhancement
3247 Description: update patchless client
3248 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
3251 Frequency : only with 2.4 kernel
3253 Description: random memory corruption
3254 Details : size of struct ll_inode_info is to big for union inode.u and this
3255 can be cause of random memory corruption.
3260 Description: Memory leak in recovery
3261 Details : Lov_mds_md was not free in an error handler in mds_create_object.
3262 It should also check obd_fail before fsfilt_start, otherwise if
3263 fsfilt_start return -EROFS,(failover mds during mds recovery).
3264 then the req will return with repmsg->transno = 0 and rc = EROFS.
3265 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
3266 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
3267 be freed no matter whether fsfilt_commit success or not.
3270 Frequency : only with large numbers of cores on a single node
3272 Description: Prevents from taking the superblock lock in llap_from_page for
3274 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
3275 for llap_from_page call in ll_removepage() prevents from taking
3276 the superblock lock for a soon died page.
3281 Description: Not check open intent error before release open handle
3282 Details : in some rare cases, the open intent error is not checked before
3283 release open handle, which may cause
3284 ASSERTION(open_req->rq_transno != 0), because it tries to release
3285 the failed open handle.
3290 Description: Set cat log bitmap only after create log success.
3291 Details : in some rare cases, the cat log bitmap is set too early. and it
3292 should be set only after create log success.
3296 Description: Accessing a block bevice can re-enable I/O when Lustre is
3297 tearing down a device.
3298 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
3303 Description: service threads may hog cpus when there are a lot of requests
3304 Details : Insert cond_resched to give other threads a chance to use some CPU
3309 Description: the cat log was not initialized in recovery
3310 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
3311 unlink log on mds will not be initialized until mds post
3312 recovery. And also in mds post recovery, the unlink log will
3313 initialization will be done asynchronausly, so there will be race
3314 between add unlink log and unlink log initialization.
3318 Description: brw_stats were being printed incorrectly
3319 Details : brw_stats were being printed as log2 but all of them were not
3320 recorded as log2. Also remove some code duplication arising from
3321 filter_tally_{read,write}.
3325 Frequency : rare, only in recovery.
3326 Description: ASSERTION(req->rq_type != LI_POISON) failed
3327 Details : imp_lock should be held while iterating over imp_sending_list for
3328 prevent destroy request after get timeout in ptlrpc_queue_wait.
3332 Description: replay-single.sh test 52 fails
3333 Details : A lock's skiplist need to be cleanup when it being unlinked
3334 from its resource list.
3339 Description: Short directio read returns full requested size rather than
3341 Details : Direct I/O operations should return actual amount of bytes
3342 transferred rather than requested size.
3344 Severity : enhancement
3346 Description: metadata RPC reduction (e.g. for rm performance)
3347 Details : decrease the amount of synchronous RPC between clients and servers
3348 by canceling conflicing lock before the operation on the client
3349 and packing thier handles into the main operation RPC to server.
3351 Severity : enhancement
3353 Description: add #ifdef HAVE_KERNEL_CONFIG_H
3354 Details : kernels from 2.6.19 not need include linux/config.h, but add
3355 include linux/autoconf.h in commpiler command line.
3357 Severity : enhancement
3359 Description: patchless client support for 2.6.22 kernel
3360 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
3361 removed. In this case we need drop using os depended interface to
3362 kmem_cache and use cfs_mem_cache API.
3367 Description: fix mal-formatted messages
3368 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
3372 Frequency : always in liblustre
3373 Description: wrong IS_ERR implementation in liblustre.h
3374 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
3379 Description: Correct condition for output debug message.
3380 Details : inode i_nlink equal zero is not enough for output message about
3381 disk corruption, i_ctime and i_mode should be also checked.
3385 Frequency : always in patchless client
3386 Description: add configure check for truncate_complete_page
3387 Details : improve checks for exported symbols. This allow run check without
3388 sources, but with Module.symvers shipped with kernel distribution.
3389 add check for truncate_complete_page used by patchless client.
3393 Description: sanity.sh test_77h fails with "test_77h file compare failed"
3394 Details : test_77h uses a file which was messed by other test case.
3398 Description: Not Check whether lov_tgts is NULL in some lov functions
3399 Details : Checking whether lov_tgts is NULL in some functions.
3403 Description: replace obdo_alloc() with OBDO_ALLOC macro
3404 Details : nothing special is done in obdo_alloc() function, and for
3405 debugging purpose, it needs to be replaced with macros.
3409 Description: bad return value and errno from fcntl call
3410 Details : In liblustre API, errno should be a negative value if error
3415 Description: ptlrpc_check_set() LBUG
3416 Details : In case of positive reply from server and failed client bulk
3417 callback after bulk transfer shouldn't LBUG, but process this
3418 request as erroneous.
3420 Severity : enhancement
3422 Description: VFS operations stats tool.
3423 Details : Tool which collects stats by tracking value written in pid,
3424 ppid, gid and uses llstat to generate output to plot graph using
3426 Updated lustre/utils/Makefile.am
3427 Added lustre/utils/ltrack_stats.c
3429 Severity : enhancement
3431 Description: 2.6.18 server support (lustre 1.6.1)
3432 Details : Support for 2.6.18 kernels on the server side.
3437 Description: ASSERTION(imp->imp_conn_current) failed
3438 Details : an assertion failure is hit if a client node boots and attempts to
3439 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
3442 Frequency : only for i686
3444 Description: 1.4.11 RC1 build fails for RHEL 4, i686
3445 Details : Fixed config variable for build.
3450 Description: Updated patchess for new RHEL4 kernel
3451 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
3452 Updated patch jbd-stats-2.6.9.patch
3453 Updated patch qsnet-rhel4-2.6.patch
3454 Updated patch quota-deadlock-on-pagelock-core.patch
3455 Updated patch vfs_intent-2.6-rhel4.patch
3456 Updated patch vfs_races-2.6-rhel4.patch
3457 Updated series file 2.6-rhel4-titech.series
3458 Updated series file 2.6-rhel4.series
3459 Updated kernel config files
3464 Description: lquota slave complains LBUG when reconnecting with mds
3466 Details : quota slave depends qctxt->lqc_import to send its quota request.
3467 This pointer will be invalid if mds did failover or broke its
3468 connect to osts, which leads to LBUG.
3471 Frequency : when qunit size is too small(less than 20M)
3473 Description: write is stopped by improper -EDQUOT
3474 Details : If the master is busy and qunit size is small enough(let's say 1M),
3475 the slave can not get quota from master on time, which will lead
3476 slave to trigger a -EQUOTA to client.
3481 Description: Deadlock during metadata tests
3482 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
3483 called with the per-dentry lock held.
3486 Frequency : SLES9 only
3488 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
3489 Details : set CONFIG_NR_CPUS to 128 instead of 8.
3491 Severity : enhancement
3493 Description: remove fs_prep_san_write operation and related patches
3494 Details : remove the ext3-san-jdike patches which are no longer useful.
3499 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
3500 Details : a disk corruption can cause the mballoc code to assert on a
3501 double free or other extent corruptions. Handle these with
3502 ext3_error() instead of assertions.
3505 Frequency : only with mballoc3 code and deep extent trees
3507 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
3508 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
3509 can cause the filesystem to be remounted read-only.
3514 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
3515 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
3516 few external users of this ioctl number is lfsck's call to
3517 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
3518 Add the old ioctl number to the handler so both old and new
3523 Description: parallel lock callbacks
3524 Details : Instead of sending blocking and completion callbacks as separated
3525 requests, adding them to a set and sending in parallel.
3529 Description: Disable most debugging by default
3530 Details : To improve performance, disable most logging (for debug purposes)
3531 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
3532 default, and HA includes fewer messages.
3536 Frequency : only run on patchless client.
3537 Description: use do_facet on sanity.sh for test handling recoverables errors
3538 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
3542 Description: hash tables for lists of nids, connections and uuids
3543 Details : Hash tables noticeably help when a lot of clients connect to a
3544 server, to faster identify duplicate connections or reconnects,
3545 also to faster find export to evict in manual eviction case.
3549 Description: Sometimes, when the server evict a client, and the client will
3550 not be evicted as soon as possible.
3551 Details : In enqueue req, the error was returned by intent, instead of
3552 rq_status which make ptlrpc layer not detect this error, and
3553 does not evict the client. So enqueue error should be returned
3557 Frequency : only at startup
3559 Description: Delay client connections to MDT until fist MDT->OST connect
3560 Details : If a client tried to create a new file before the MDT had
3561 connected to any OSTs, the create would return EIO. Now
3562 the client will simply block until the MDT connects to the
3563 first OST and the create can succeed.
3566 Frequency : only for SLES9
3568 Description: Routinely utilize latest Quadrics drivers in CFS releases
3569 Details : Update patch qsnet-suse-2.6.patch.
3572 Frequency : only for sles10
3574 Description: Update patches for SLES 10 SP1 kernel.
3575 Details : Update the patch vfs_intent-2.6-sles10.patch.
3580 Description: Routinely utilize latest Quadrics drivers in CFS releases
3581 Details : Update patch qsnet-rhel4-2.6.patch.
3584 Frequency : at statup only
3586 Description: mds_lov_synchronize race leads to various problems
3587 Details : simultaneous MDT->OST connections at startup can cause the
3588 sync to abort, leaving the OSC in a bad state.
3593 Description: Using wrong pointer in osc_brw_prep_request
3594 Details : Access to array[-1] can produce panic if kernel compiled with
3595 CONFIG_PAGE_ALLOC enabled
3597 Severity : enhancement
3599 Description: Async OSC create to avoid the blocking unnecessarily.
3600 Details : If a OST has no remain object, system will block on the creating
3601 when need to create a new object on this OST. Now, ways use
3602 pre-created objects when available, instead of blocking on an
3603 empty osc while others are not empty. If we must block, we block
3604 for the shortest possible period of time.
3608 Frequency : only in recovery
3609 Description: Mark OST as early accessible if his start SYNC.
3610 Details : osc_precreate return flag early accessible if oscc marked as
3611 OSCC_FLAG_SYNC_IN_PROGRESS.
3616 Description: Sometimes precreate code can triger create object on wrong ost
3617 Details : Wrong protected or not not restored variables aftre precreate loop
3618 can produce creation object on wrong ost.
3621 Frequency : oss recovery
3623 Description: llog_commit_thread cleanup should sync with llog_commit_thread
3625 Details : llog_commit_thread_count should be synced between llog_commit
3626 start and cleanup, so new llog_commit thread should not be started
3627 when llog_commit threads being stopped to avoid accessing some
3630 Severity : enhancement
3632 Description: Add printing inode info into message about error in writepage.
3634 --------------------------------------------------------------------------------
3636 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
3641 Frequency : on some architectures
3643 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
3644 Details : Uninitialized flags sometimes cause configuration commands to
3648 Frequency : patchless clients only
3650 Description: missing __iget() symbol export
3651 Details : The __iget() symbol export is missing. To avoid the need for
3652 this on patchless clients the deathrow inode reaper is turned
3653 off, and we depend on the VM to clean up old inodes. This
3654 dependency was during via the fix for bug 12181.
3659 Description: sanity.sh fail: test_52b
3660 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
3661 incorrect inode's flags to client.
3663 --------------------------------------------------------------------------------
3665 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
3667 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
3668 INTEROPERATE with older versions automatically. In many cases a
3669 special upgrade step is needed. Please read the
3670 user documentation before upgrading any part of a 1.4.x system.
3671 * WARNING: Lustre configuration and startup changes are required with
3672 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
3674 * Support for kernels:
3675 2.4.21-47.0.1.EL (RHEL 3),
3676 2.6.5-7.283 (SLES 9),
3677 2.6.9-42.0.10.EL (RHEL 4),
3678 2.6.12.6 vanilla (kernel.org),
3679 2.6.16.27-0.9 (SLES10)
3680 * Client support for unpatched kernels:
3681 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
3682 2.6.16 - 2.6.19 vanilla (kernel.org),
3683 2.6.9-42.0.8EL (RHEL 4)
3684 * Recommended e2fsprogs version: 1.39.cfs6
3685 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3688 Severity : enhancement
3690 Description: Async OSC create to avoid the blocking unnecessarily.
3691 Details : If a OST has no remain object, system will block on the creating
3692 when need to create a new object on this OST. Now, ways use
3693 pre-created objects when available, instead of blocking on an
3694 empty osc while others are not empty. If we must block, we block
3695 for the shortest possible period of time.
3697 Severity : enhancement
3699 Description: MountConf
3700 Details : Lustre configuration is now managed via mkfs and mount
3701 commands instead of lmc and lconf. New obd types (MGS, MGC)
3702 are added for dynamic configuration management. See
3703 https://mail.clusterfs.com/wikis/lustre/MountConf for
3706 Severity : enhancement
3708 Description: dynamic OST addition
3709 Details : OSTs can now be added to a live filesystem
3711 Severity : enhancement
3713 Description: startup order invariance
3714 Details : MDTs and OSTs can be started in any order. Clients only
3715 require the MDT to complete startup.
3717 Severity : enhancement
3719 Description: parallel, asynchronous orphan cleanup
3720 Details : orphan cleanup is now performed in separate threads for each
3721 OST, allowing parallel non-blocking operation.
3723 Severity : enhancement
3725 Description: optimized stripe assignment
3726 Details : stripe assignments are now made based on ost space available,
3727 ost previous usage, and OSS previous usage, in order to try
3728 to optimize storage space and networking resources.
3730 Severity : enhancement
3732 Description: Permanently set tunables
3733 Details : All writable /proc/fs/lustre tunables can now be permanently
3734 set on a per-server basis, at mkfs time or on a live
3737 Severity : enhancement
3739 Description: Lustre message v2
3740 Details : Add lustre message format v2.
3742 Severity : enhancement
3744 Description: client OST exclusion list
3745 Details : Clients can be started with a list of OSTs that should be
3746 declared "inactive" for known non-responsive OSTs.
3750 Description: ENOENT returned for valid filehandle during dbench.
3751 Details : Check if a directory has children when invalidating dentries
3752 associated with an inode during lock cancellation. This fixes
3753 an incorrect ENOENT sometimes seen for valid filehandles during
3754 testing with dbench.
3757 Frequency : SFS test only (otherwise harmless)
3759 Description: SPEC SFS validation failure on NFS v2 over lustre.
3760 Details : Changes the blocksize for regular files to be 2x RPC size,
3761 and not depend on stripe size.
3763 Severity : enhancement
3765 Description: fine-grained SMP locking inside DLM
3766 Details : Improve DLM performance on SMP systems by removing the single
3767 per-namespace lock and replace it with per-resource locks.
3769 Severity : enhancement
3771 Description: don't hold multiple extent locks at one time
3772 Details : To avoid client eviction during large writes, locks are not
3773 held on multiple stripes at one time or for very large writes.
3774 Otherwise, clients can block waiting for a lock on a failed OST
3775 while holding locks on other OSTs and be evicted.
3777 Severity : enhancement
3779 Description: Multiple MD RPCs in flight.
3780 Details : Further unserialise some read-only MDT RPCs - learn about intents.
3781 To avoid overly-overloading MDT, introduce a limit on number of
3782 MDT RPCs in flight for a single client and add /proc controls
3783 to adjust this limit.
3785 Severity : enhancement
3787 Description: client read/write statistics
3788 Details : Add client read/write call usage stats for performance
3789 analysis of user processes.
3790 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
3791 file access. extents_stats shows chunk size distribution.
3792 extents_stats_per_process show chunk size distribution per
3795 Severity : enhancement
3797 Description: per-client statistics on server
3798 Details : Add ldlm and operations statistics for each client in
3799 /proc/fs/lustre/mds|obdfilter/*/exports/
3801 Severity : enhancement
3803 Description: improved MDT statistics
3804 Details : Add detailed MDT operations statistics in
3805 /proc/fs/lustre/mds/*/stats
3807 Severity : enhancement
3809 Description: VFS operations stats
3810 Details : Add client VFS call stats, trackable by pid, ppid, or gid
3811 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
3816 Description: Fix client-side osc byte counters
3817 Details : The osc read/write byte counters in
3818 /proc/fs/lustre/osc/*/stats are now working
3821 Frequency : always as root on SLES
3823 Description: Failure of copying files with lustre special EAs.
3824 Details : Client side always return success for setxattr call for lustre
3825 special xattr (currently only "trusted.lov").
3830 Description: Refcount LNET uuids
3831 Details : The global LNET uuid list grew linearly with every startup;
3832 refcount repeated list entries instead of always adding to
3835 Severity : enhancement
3837 Description: Dynamic service threads
3838 Details : Within a small range, start extra service threads
3839 automatically when the request queue builds up.
3842 Frequency : mixed-endian client/server environments
3844 Description: mixed-endian crashes
3845 Details : The new msg_v2 system had some failures in mixed-endian
3848 Severity : enhancement
3850 Description: Easy OST removal
3851 Details : OSTs can be permanently deactivated with e.g. 'lctl
3852 conf_param lustre-OST0001.osc.active=0'
3854 Severity : enhancement
3856 Description: MGS proc entries
3857 Details : Added basic proc entries for the MGS showing what filesystems
3860 Severity : enhancement
3862 Description: provide MGS failover
3863 Details : Added config lock reacquisition after MGS server failover.
3865 Severity : enhancement
3867 Description: add Linux 2.4 support
3868 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
3872 Description: a large application tries to do I/O to the same resource and dies
3873 in the middle of it.
3874 Details : Check the req->rq_arrival time after the call to
3875 ost_brw_lock_get(), but before we do anything about
3876 processing it & sending the BULK transfer request. This
3877 should help move old stale pending locks off the queue as
3878 quickly as obd_timeout.
3881 Frequency : when an incorrect nid is specified during startup
3883 Description: ptlrpc connect to non-existant node causes kernel crash
3884 Details : LNET can't be re-entered from an event callback, which
3885 happened when we expire a message after the export has been
3886 cleaned up. Instead, hand the zombie cleanup off to another
3889 Severity : enhancement
3891 Description: plain/inodebits lock performance improvement
3892 Details : Grouping plain/inodebits in granted list by their request modes
3893 and bits policy, thus improving the performance of search through
3897 Frequency : only if OST filesystem is corrupted
3899 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3900 Details : for a short time RPCs with bulk IO are in the replay list,
3901 but replay of bulk IOs is unimplemented. If the OST filesystem
3902 is corrupted due to disk cache incoherency and then replay is
3903 started it is possible to trip an assertion. Avoid putting
3904 committed RPCs into the replay list at all to avoid this issue.
3907 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3909 Description: System hang on startup
3910 Details : This bug allowed the liblustre (e.g. catamount) client to
3911 return to the app before handling all startup RPCs. This
3912 could leave the node unresponsive to lustre network traffic
3913 and manifested as a server ptllnd timeout.
3915 Severity : enhancement
3917 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3918 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
3919 boolean to control whether to print peer debug info when a
3920 client's RPC times out.
3923 Frequency : only for kernels with patches from Lustre below 1.4.3
3925 Description: Remove old rdonly API
3926 Details : Remove old rdonly API which unused from at least lustre 1.4.3
3929 Frequency : only for devices with external journals
3931 Description: Set external device read-only also
3932 Details : During a commanded failover stop, we set the disk device
3933 read-only while the server shuts down. We now also set any
3934 external journal device read-only at the same time.
3937 Frequency : when upgrading from 1.4 while trying to change parameters
3939 Description: The wrong (new) MDC name was used when setting parameters for
3940 upgraded MDT's. Also allows changing of OSC (and MDC)
3941 parameters if --writeconf is specified at tunefs upgrade time.
3944 Frequency : when setting specific ost indicies
3946 Description: QOS code breaks on skipped indicies
3947 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3948 created with --index need not be sequential.
3950 Severity : enhancement
3952 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3953 Details : The uninit_groups feature works in conjunction with the kernel
3954 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3955 up the pass1 processing of e2fsck. This is a read-only feature
3956 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3957 mount filesystems that have had this feature enabled.
3959 Severity : enhancement
3961 Description: Improve multi-block allocation algorithm to avoid fragmentation
3962 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3963 allocation locality and avoid filesystem fragmentation.
3965 ------------------------------------------------------------------------------
3967 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3969 * Support for kernels:
3970 2.4.21-47.0.1.EL (RHEL 3)
3971 2.6.5-7.283 (SLES 9)
3972 2.6.9-42.0.10.EL (RHEL 4)
3973 2.6.12.6 vanilla (kernel.org)
3974 2.6.16.27-0.9 (SLES 10)
3975 * Recommended e2fsprogs version: 1.39.cfs5
3977 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3981 Frequency : occasional, depends on client load and configuration
3982 Bugzilla : 12181, 12203
3983 Description: data loss for recently-modified files
3985 Details : In some cases it is possible that recently written or created
3986 files may not be written to disk in a timely manner (this should
3987 normally be within 30s unless client IO load is very high).
3988 The problem appears as zero-length files or files that are a
3989 multiple of 1MB in size after a client crash or client eviction
3990 that are missing data at the end of the file.
3992 This problem is more likely to be hit on clients where files are
3993 repeatedly created and unlinked in the same directory, clients
3994 have a large amount of RAM, have many CPUs, the filesystem has
3995 many OSTs, the clients are rebooted frequently, and/or the files
3996 are not accessed by other nodes after being written.
3998 The presence of the problem can be detected by looking at
3999 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
4000 smaller than the second (nr_unused) then dirty files will not
4001 be flushed automatically to disk. "sync; sleep 10" should be
4002 run several times on the node before unmounting it to update
4003 Lustre (this is also safe to run on nodes without this problem).
4005 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
4006 that can cause this same problem, so customers using that kernel
4007 also need to update the kernel in addition to Lustre. In order
4008 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
4010 It is normal that files written just before a client crash (less
4011 than 30s) may not yet have been flushed to disk, even for local
4015 Frequency : frequent on thin XT3 nodes
4017 Description: UUID collision on thin XT3 Linux nodes
4018 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
4019 randomly, since we relied on an insufficiently-seeded PRNG.
4024 Description: OSS hangs after "All ost request buffers busy"
4025 Details : A deadlock between quota and journal operations caused OSS
4026 hangs after printing "All ost request buffers busy."
4029 Frequency : always on liblustre builds
4031 Description: Cleanup compiler warnings on liblustre
4034 Frequency : always on liblustre builds on XT3
4036 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
4040 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
4041 > 32000 subdirectories
4042 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
4043 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
4044 subdirectory count crosses 32000. This will aid e2fsck to
4045 correctly handle more than 32000 subdirectories.
4048 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
4050 Description: System hang on startup
4051 Details : This bug allowed the liblustre (e.g. catamount) client to
4052 return to the app before handling all startup RPCs. This
4053 could leave the node unresponsive to lustre network traffic
4054 and manifested as a server ptllnd timeout.
4056 Severity : enhancement
4058 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
4059 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
4060 boolean to control whether to print peer debug info when a
4061 client's RPC times out.
4066 Description: make O_SYNC working on 2.6 kernels
4067 Details : 2.6 kernels use different method for mark pages for write,
4068 so need add a code to lustre for O_SYNC work.
4073 Description: Failure to close file and release space on NFS
4074 Details : Put inode details into lock acquired in ll_intent_file_open.
4075 Use mdc_intent_lock in ll_intent_open to properly
4076 detect all kind of errors unhandled by mdc_enqueue.
4081 Description: proc file read during shutdown sometimes raced obd removal,
4083 Details : Add lock to prevent obd access after proc file removal.
4086 Frequency : Only for files larger than 4GB on 32-bit clients.
4088 Description: improperly doing page alignment of locks
4089 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
4090 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
4095 Description: RPCs being resent when they shouldn't be
4096 Details : Some RPCs that should not be resent are being resent. This
4097 can cause inconsistencies in the RPC state machine. Do not
4098 resend such requests.
4101 Frequency : rare, only with NFS export
4103 Description: Crash on NFS re-export node
4104 Details : under very unusual load conditions an assertion is hit in
4105 ll_intent_file_open()
4108 Frequency : only if OST filesystem is corrupted
4110 Description: client incorrectly hits assertion in ptlrpc_replay_req()
4111 Details : for a short time RPCs with bulk IO are in the replay list,
4112 but replay of bulk IOs is unimplemented. If the OST filesystem
4113 is corrupted due to disk cache incoherency and then replay is
4114 started it is possible to trip an assertion. Avoid putting
4115 committed RPCs into the replay list at all to avoid this issue.
4120 Description: large O_DIRECT requests fail under memory pressure/fragmentation
4121 Details : Large single O_DIRECT read and write calls can fail to allocate
4122 a sufficiently large buffer to process the request. In case of
4123 allocation failure the allocation is retried with a smaller
4124 buffer and broken into smaller requests.
4126 Severity : enhancement
4128 Description: Add -o localflock option to simulate old noflock behaviour.
4129 Details : This will achieve local-only flock/fcntl locks coherentness.
4134 Description: versioning check is incomplete
4135 Details : Checking the version difference of client vs. server, report
4136 error if the gap is too big.
4141 Description: add support PG_writeback bit
4142 Details : add support for PG_writeback bit for Lustre, for more carefull
4143 work with page cache in 2.6 kernel. This also fix some deadlocks
4144 and remove hack for work O_SYNC with 2.6 kernel.
4146 Severity : enhancement
4148 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
4149 Details : The uninit_groups feature works in conjunction with the kernel
4150 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
4151 up the pass1 processing of e2fsck. This is a read-only feature
4152 in ldiskfs2 only, so older kernels and current ldiskfs cannot
4153 mount filesystems that have had this feature enabled.
4155 Severity : enhancement
4157 Description: Improve multi-block allocation algorithm to avoid fragmentation
4158 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
4159 allocation locality and avoid filesystem fragmentation.
4161 ------------------------------------------------------------------------------
4163 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
4165 * Support for kernels:
4166 2.6.9-42.0.3.EL (RHEL 4)
4167 2.6.5-7.276 (SLES 9)
4168 2.4.21-47.0.1.EL (RHEL 3)
4169 2.6.12.6 vanilla (kernel.org)
4170 2.6.16.21-0.8 (SLES10)
4171 * Recommended e2fsprogs version: 1.39.cfs2-0
4173 * The backwards-compatible /proc/sys/portals symlink has been removed
4174 in this release. Before upgrading, please ensure that you change
4175 any configuration scripts or /etc/sysctl.conf files that access
4176 /proc/sys/portals/* or sysctl portals.* to use the corresponding
4177 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
4178 in advance of the upgrade on any system running Lustre 1.4.6 or
4179 newer, since /proc/sys/lnet was added in that version.
4180 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4184 Frequency : only when quota is used
4186 Description: avoid scanning export list for quota master
4187 Details : Change the algorithms to avoid scanning export list in order
4188 to improve the efficiency.
4191 Frequency : MDS failover only, very rarely
4193 Description: "went back in time" messages on mds failover
4194 Details : The greatest transno may be lost when the current operation
4195 finishes with an error (transno==0) and the client's last_rcvd
4196 record is over-written. Save the greatest transno in the
4197 mds_last_transno for this case.
4200 Frequency : always for specific kernels and striping counts
4202 Description: client may get "Matching packet too big" without ACL support
4203 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
4204 when trying to access files in certain configurations. The
4205 clients should in fact be denied when mounting because they do
4206 not understand ACLs.
4209 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
4211 Description: many clients connecting with IO in progress causes connect timeouts
4212 Details : Avoid synchronous journal commits to avoid delays caused by many
4213 clients connecting/disconnecting when bulk IO is in progress.
4214 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
4215 OST_IO_PORTAL to avoid delays behind potentially many pending
4219 Frequency : occasionally with multiple writers to a single file
4221 Description: shared writes to file may result in wrong size reported by stat()
4222 Details : Allow growing of kms when extent lock is cancelled
4225 Frequency : always with random mmap IO to multi-striped file
4227 Description: mmap write might be lost if we are writing to a 'hole' in stripe
4228 Details : Only if the hole is at the end of OST object so that kms is too
4229 small. Fix is to increase kms accordingly in ll_nopage.
4232 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
4234 Description: writes to a missing object would leak memory on the OST
4235 Details : If there is an inconsistency between the MDS and OST filesystems,
4236 such that the MDS references an object that doesn't exist, writes
4237 to that object will leak memory due to incorrect cleanup in the
4238 error handling path, eventually running out of memory on the OST.
4243 Description: Creating too long symlink causes lustre errors
4244 Details : Check symlink and name lengths before sending requests to MDS.
4247 Frequency : only if flock is enabled (not on by default)
4249 Description: posix locks not released on fd closure on 2.6.9+
4250 Details : We failed to add posix locks to list of inode locks on 2.6.9+
4251 kernels, this caused such locks not to be released on fd close and
4252 then assertions on fs unmount about still used locks.
4255 Frequency : MDS failover only, very rarely
4257 Description: clients may get ASSERTION(granted_lock != NULL)
4258 Details : When request was taking a long time, and a client was resending
4259 a getattr by name lock request. The were multiple lock requests
4260 with the same client lock handle and
4261 mds_getattr_name->fixup_handle_for_resent_request found one of the
4262 lock handles but later failed with ASSERTION(granted_lock != NULL).
4267 Description: handle->h_buffer_credits > 0, assertion failure
4268 Details : h_buffer_credits is zero after truncate, causing assertion
4269 failure. This patch extends the transaction or creates a new
4273 Frequency : NFS re-export or patchless client
4274 Bugzilla : 11179, 10796
4275 Description: Crash on NFS re-export node (__d_move)
4276 Details : We do not want to hash the dentry if we don't have a lock.
4277 But if this dentry is later used in d_move, we'd hit uninitialised
4278 list head d_hash, so we just do this to init d_hash field but
4279 leave dentry unhashed.
4282 Frequency : NFS re-export or patchless client
4284 Description: NFS exports has problem with symbolic link
4285 Details : lustre client didn't properly install dentry when re-exported
4286 to NFS or running patchless client.
4289 Frequency : NFS re-export or patchless client
4291 Description: Various nfs/patchless fixes.
4292 Details : fixes reuse disconected alias for lookup process - this fixes
4293 warning "find_exported_dentry: npd != pd",
4294 fix permission error with open files at nfs.
4295 fix apply umask when do revalidate.
4298 Frequency : occasional
4300 Description: Crash on NFS re-export node
4301 Details : calling clear_page() on the wrong pointer triggered oops in
4302 generic_mapping_read().
4305 Frequency : rarely, using O_DIRECT IO
4307 Description: unaligned directio crashes client with LASSERT
4308 Details : check for unaligned buffers before trying any requests.
4311 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
4313 Description: stale data returned from RAID cache
4314 Details : If only a small amount of IO is done to the RAID device before
4315 reading it again it is possible to get stale data from the RAID
4316 cache instead of reading it from disk.
4319 Frequency : always for sles10 kernel
4321 Description: sles10 support
4322 Details : ll_follow_link: compile fixes and using of nd_set_link
4323 under newer kernels.
4326 Frequency : depends on arch, kernel and compiler version, always on sles10
4329 Description: recursive or deep enough symlinks cause stack overflow
4330 Details : getting rid of large stack-allocated variable in
4334 Frequency : depends on hardware
4336 Description: lustre write performance loss in the SLES10 kernel
4337 Details : the performance loss is caused by using of write barriers in the
4338 ext3 code. The SLES10 kernel turns barrier support on by
4339 default. The fix is to undo that change for ldiskfs.
4341 ------------------------------------------------------------------------------
4343 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
4345 * Support for kernels:
4346 2.6.9-42.0.3EL (RHEL 4)
4347 2.6.5-7.276 (SLES 9)
4348 2.4.21-47.0.1.EL (RHEL 3)
4349 2.6.12.6 vanilla (kernel.org)
4353 Frequency : quota enabled and large files being deleted
4355 Description: releasing more than 4GB of quota at once hangs OST
4356 Details : If a user deletes more than 4GB of files on a single OST it
4357 will cause the OST to spin in an infinite loop. Release
4358 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4363 Description: statfs data retrieved from /proc may be stale or zero
4364 Details : When reading per-device statfs data from /proc, in the
4365 {kbytes,files}_{total,free,avail} files, it may appear
4366 as zero or be out of date.
4369 Frequency : systems with MD RAID1 external journal devices
4371 Description: lconf's call to blkid is confused by RAID1 journal devices
4372 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4373 of returning all block devices that match the journal UUID.
4376 Frequency : always, for aggregate stripe size over 4GB
4378 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
4379 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
4380 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
4383 Frequency : always if "lfs find" used on a local file/directory
4385 Description: "lfs find" segfaults if used on a local file/directory
4386 Details : The case where a directory component was not specified wasn't
4387 handled correctly. Handle this properly.
4390 Frequency : always on ppc64
4392 Description: the write to an ext3 filesystem mounted with mballoc got stuck
4393 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4394 perform endianness conversion.
4397 Frequency : rarely (truncate to non-zero file size after write under load)
4398 Bugzilla : 10730, 10687
4399 Description: Files padded with zeros to next 4K multiple
4400 Details : With filesystems mounted using the "extents" option (2.6 kernels)
4401 it is possible that files that are truncated to a non-zero size
4402 immediately after being written are filled with zero bytes beyond
4403 the truncated size. No file data is lost.
4405 Severity : enhancement
4407 Description: Allow recovery/failover for liblustre clients.
4408 Details : liblustre clients were unaware of failover configurations until
4411 Severity : enhancement
4413 Description: user file locks should fail when not mounting with flock option
4414 Details : Set up an error-returning stub in ll_file_operations.lock field
4415 to prevent incorrect behaviour when client is mounted without
4416 flock option. Also, set up properly f_op->flock field for
4420 Frequency : always on ia64
4422 Description: "lfs df" loops on printing out MDS statfs information
4423 Details : The obd_ioctl_data was not initialized and in some systems
4424 this caused a failure during the ioctl that did not return
4425 an error. Initialize the struct and return an error on failure.
4428 Frequency : SLES 9 only
4430 Description: Error of copying files with lustre special EAs as root
4431 Details : Client side always return success for setxattr call for lustre
4432 special xattr (currently only "trusted.lov").
4435 Frequency : rarely on clusters with both ia64+i386 clients
4437 Description: ia64+i686 clients doing shared IO on the same file may LBUG
4438 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
4439 clients are doing concurrent writes to the same file it is
4440 possible that the ia64 clients may LASSERT because the OST
4441 extent locks are not PAGE_SIZE aligned. Ensure that grown
4442 locks are always aligned on the request boundary.
4445 Frequency : specific use, occasional
4447 Description: Overwriting in use executable truncates on-disk binary image
4448 Details : If one node attempts to overwrite an executable in use by
4449 another node, we now correctly return ETXTBSY instead of
4450 truncating the file.
4452 Severity : enhancement
4454 Description: Async OSC create to avoid the blocking unnecessarily.
4455 Details : If a OST has no remain object, system will block on the creating
4456 when need to create a new object on this OST. Now, ways use
4457 pre-created objects when available, instead of blocking on an
4458 empty osc while others are not empty. If we must block, we block
4459 for the shortest possible period of time.
4464 Description: chmod on Lustre root is propagated to other clients
4465 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
4466 invalid by the follow_mount time.
4469 Frequency : liblustre clients only
4471 Description: Race in 'instant cancel' lock handling could lead to such locks
4472 never to be granted in case of SMP MDS
4473 Details : Do not destroy not yet granted but cbpending locks in
4477 Frequency : replay/resend of open
4479 Description: non null lock assetion failure in mds_intent_policy
4480 Details : Trying to replay/resend lockless open requests resulted in
4481 mds_open() returning 0 with no lock. Now it sets a flag if
4482 a lock is going to be returned.
4484 Severity : enhancement
4486 Description: Checksum enhancements
4487 Details : New checksum enhancements allow for resending RPCs that failed
4490 Severity : enhancement
4492 Description: Tunables on number of dirty pages in cacche
4493 Details : Allow to set limit on number of dirty pages cached.
4498 Description: client crash on unmount - lock still has references
4499 Details : In some error handling cases it was possible to leak a lock
4500 reference on a client while accessing a file. This was not
4501 harmful to the client during operation, but would cause the
4502 client to crash when the filesystem is unmounted.
4505 Frequency : specific case, rare
4507 Description: ETXTBSY on mds though file not in use by client
4508 Details : ETXTBSY is no longer incorrectly returned when attempting to
4509 chmod or chown a directory that the user previously tried to
4510 execute or a currently-executing binary.
4513 Frequency : extremely rare except on liblustre-based clients
4515 Description: Lustre space not freed when files are deleted
4516 Details : Clean up open-unlinked files after client eviction. Previously
4517 the unlink was skipped and the files remained as orphans.
4522 Description: OST failure "would be an LBUG" in waiting_locks_callback()
4523 Details : In some cases it was possible to send a blocking callback to
4524 a client doing a glimpse, even though that client didn't get
4525 a lock granted. When the glimpse lock is cancelled on the OST
4526 the freed lock is left on the waiting list and corrupted the list.
4529 Frequency : all core dumps
4531 Description: Broke core dumps to lustre
4532 Details : Negative dentry may be unhashed if parent does not have UPDATE
4533 lock, but some callers, e.g. do_coredump, expect dentry to be
4534 hashed after successful create, hash it in ll_create_it.
4536 ------------------------------------------------------------------------------
4538 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
4540 * Support for kernels:
4541 2.6.9-42.0.2.EL (RHEL 4)
4542 2.6.5-7.276 (SLES 9)
4543 2.4.21-40.EL (RHEL 3)
4544 2.6.12.6 vanilla (kernel.org)
4548 Frequency : always on RHEL 3
4550 Description: Number of open files grows over time
4551 Details : The number of open files grows over time, whether or not
4552 Lustre is started. This was due to a filp leak introduced
4553 by one of our kernel patches.
4555 ------------------------------------------------------------------------------
4557 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
4559 * Support for kernels:
4560 2.6.9-42.EL (RHEL 4)
4561 2.6.5-7.267 (SLES 9)
4562 2.4.21-40.EL (RHEL 3)
4563 2.6.12.6 vanilla (kernel.org)
4568 Bugzilla : 5719, 9635, 9792, 9684
4569 Description: OST (or MDS) trips assertions in (re)connection under heavy load
4570 Details : If a server is under heavy load and cannot reply to new
4571 connection requests before the client resends the (re)connect,
4572 the connection handling code can behave badly if two service
4573 threads are concurrently handing separate (re)connections from
4574 the same client. Add better locking to the connection handling
4575 code, and ensure that only a single connection will be processed
4576 for a given client UUID, even if the lock is dropped.
4578 Severity : enhancement
4580 Description: add TCP zero-copy support to kernel
4581 Details : Add support to the kernel TCP stack to allow zero-copy bulk
4582 sends if the hardware supports scatter-gather and checksumming.
4583 This allows socklnd to do client-write and server-read more
4584 efficiently and reduce CPU utilization from skbuf copying.
4587 Frequency : only if NFS exporting from client
4589 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
4590 Details : If setting attributes on a file created under NFS that had
4591 never been opened it would be possible to oops the client
4592 if the file had no objects.
4596 Bugzilla : 9326, 10402, 10897
4597 Description: client crash in ptlrpcd_wake() thread when sending async RPC
4598 Details : It is possible that ptlrpcd_wake() dereferences a freed async
4599 RPC. In rare cases the ptlrpcd thread alread processed the RPC
4600 before ptlrpcd_wake() was called and the request was freed.
4603 Frequency : always for liblustre
4605 Description: liblustre client does MDS+OSTs setattr RPC for each write
4606 Details : When doing a write from a liblustre client, the client
4607 incorrectly issued an RPC to the MDS and each OST the file was
4608 striped over in order to update the timestamps. When writing
4609 with small chunks and many clients this could overwhelm the MDS
4610 with RPCs. In all cases it would slow down the write because
4611 these RPCs are unnecessary.
4613 Severity : enhancement
4615 Description: allow number of MDS service threads to be changed at module load
4616 Details : It is now possible to change the number of MDS service threads
4617 running. Adding "options mds mds_num_threads={N}" to the MDS's
4618 /etc/modprobe.conf will set the number of threads for the next
4619 time Lustre is restarted (assuming the "mds" module is also
4620 reloaded at that time). The default number of threads will
4621 stay the same, 32 for most systems.
4626 Description: OST crash if filesystem is unformatted or corrupt
4627 Details : If an OST is started on a device that has never been formatted
4628 or if the filesystem is corrupt and cannot even mount then the
4629 error handling cleanup routines would dereference a NULL pointer.
4634 Description: NULL pointer deref in llap_from_page.
4635 Details : get_cache_page_nowait can return a page with NULL (or otherwise
4636 incorrect) mapping if the page was truncated/reclaimed while it was
4637 searched for. Check for this condition and skip such pages when
4638 doing readahead. Introduce extra check to llap_from_page() to
4639 verify page->mapping->host is non-NULL (so page is not anonymous).
4642 Frequency : Sometimes when using sys_sendfile
4644 Description: "page not covered by a lock" warnings from ll_readpage
4645 Details : sendfile called ll_readpage without right page locks present.
4646 Now we introduced ll_file_sendfile that does necessary locking
4647 around call to generic_file_sendfile() much like we do in
4651 Frequency : with certain MDS communication failures at client mount time
4653 Description: NULL pointer deref after failed client mount
4654 Details : a client connection request may delayed by the network layer
4655 and not be sent until after the PTLRPC layer has timed out the
4656 request. If the client fails the mount immediately it will try
4657 to clean up before the network times out the request. Add a
4658 reference from the request import to the obd device and delay
4659 the cleanup until the network drops the request.
4662 Frequency : occasionally during client (re)connect
4664 Description: assertion failure during client (re)connect
4665 Details : processing a client connection request may be delayed by the
4666 client or server longer than the client connect timeout. This
4667 causes the client to resend the connection request. If the
4668 original connection request is replied in this interval, the
4669 client may trip an assertion failure in ptlrpc_connect_interpret()
4670 which thought it would be the only running connect process.
4673 Frequency : only with obd_echo servers and clients that are rebooted
4675 Description: kernel BUG accessing uninitialized data structure
4676 Details : When running an obd_echo server it did not start the ping_evictor
4677 thread, and when a client was evicted an uninitialized data
4678 structure was accessed. Start the ping_evictor in the RPC
4679 service startup instead of the OBD startup.
4681 Severity : enhancement
4682 Bugzilla : 10193 (patchless)
4683 Description: Remove dependency on various unexported kernel interfaces.
4684 Details : No longer need reparent_to_init, exit_mm, exit_files,
4685 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
4688 Frequency : rare (only users of deprecated and unsupported LDAP config)
4690 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
4691 Details : LDAP apparently contains 'lustreName' attributes instead of
4692 'name'. A simple remapping of the name is sufficient.
4695 Frequency : rare (only with non-default dump_on_timeout debug enabled)
4697 Description: waiting_locks_callback trips kernel BUG if client is evicted
4698 Details : Running with the dump_on_timeout debug flag turned on makes
4699 it possible that the waiting_locks_callback() can try to dump
4700 the Lustre kernel debug logs from an interrupt handler. Defer
4701 this log dumping to the expired_lock_main() thread.
4703 Severity : enhancement
4705 Description: Support NFS exporting on 2.6 kernels.
4706 Details : Implement non-rawops metadata methods for NFS server to use without
4707 changing NFS server code.
4710 Frequency : very rare (synthetic metadata workload only)
4712 Description: two racing renames might cause an MDS thread to deadlock
4713 Details : Running the "racer" program may cause one MDS thread to rename
4714 a file from being the source of a rename to being the target of
4715 a rename at exactly the same time that another thread is doing
4716 so, and the second thread has already enqueued these locks after
4717 doing a lookup of the target and is trying to relock them in
4718 order. Ensure that we don't try to re-lock the same resource.
4721 Frequency : only very large systems with liblustre clients
4723 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
4724 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
4725 all OSTs in parallel. This allows the eviction of stale liblustre
4726 clients to proceed much faster than if they were done in series,
4727 and also offers similar improvements for other set_info RPCs.
4732 Description: excessive CPU usage during initial read phase on client
4733 Details : During the initial read phase on a client, it would agressively
4734 retry readahead on the file, consuming too much CPU and impacting
4735 performance (since 1.4.5.8). Improve the readahead algorithm
4736 to avoid this, and also improve some other common cases (read
4737 of small files in particular, where "small" is files smaller than
4738 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
4743 Description: MDS crash when receiving packet with unknown intent.
4744 Details : Do not LBUG in unknown intent case, just return -EFAULT
4746 Severity : enhancement
4747 Bugzilla : 9293, 9385
4748 Description: MDS RPCs are serialised on client. This is unnecessary for some.
4749 Details : Do not serialize getattr (non-intent version) and statfs.
4752 Frequency : occasional, when OST network is overloaded/intermittent
4754 Description: client evicted by OST after bulk IO timeout
4755 Details : If a client sends a bulk IO request (read or write) the OST
4756 may evict the client if it is unresposive to its data GET/PUT
4757 request. This is incorrect if the network is overloaded (takes
4758 too long to transfer the RPC data) or dropped the OST GET/PUT
4759 request. There is no need to evict the client at all, since
4760 the pinger and/or lock callbacks will handle this, and the
4761 client can restart the bulk request.
4764 Frequency : Always when mmapping file with no objects
4766 Description: client crashes when mmapping file with no objects
4767 Details : Check that we actually have objects in a file before doing any
4768 operations on objects in ll_vm_open, ll_vm_close and
4774 Description: Request leak when working with deleted CWD
4775 Details : Introduce advanced request refcount tracking for requests
4776 referenced from lustre intent.
4778 Severity : Enhancement
4780 Description: Cache open file handles on client.
4781 Details : MDS now will return special lock along with openhandle, if
4782 requested and client is allowed to hold openhandle, even if unused,
4783 until such a lock is revoked. Helps NFS a lot, since NFS is opening
4784 closing files for every read/write openration.
4786 Severity : Enhancement
4788 Description: Cache open negative dentries on client when possible.
4789 Details : Guard negative dentries with UPDATE lock on parent dir, drop
4790 negative dentries on lock revocation.
4795 Description: Remounting a client read-only wasn't possible with a zconf mount
4796 Details : It wasn't possible to remount a client read-only with llmount.
4798 Severity : enhancement
4799 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
4800 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
4801 MPICH is included to add Lustre-specific ADIO interfaces.
4802 This is based closely on the UFS ADIO layer and only differs
4803 in file creation, in order to allow the OST striping to be set.
4804 This is user-contributed code and not supported by CFS.
4809 Description: extended inode attributes (immutable, append-only) work improperly
4810 when 2.4 and 2.6 kernels are used on client/server or vice versa
4811 Details : Introduce kernel-independent values for these flags.
4813 Severity : enhancement
4816 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
4817 Details : Many of the /proc/ tunables can only be tuned at a megabyte
4818 granularity. Now, Fractional MB granularity is be supported,
4819 this is very useful for low memory system.
4821 Severity : enhancement
4823 Description: Getattr by fid
4824 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
4825 locks, avoids extra getattr rpc requests to MDS, allows '/' to
4826 have locks and avoids getattr rpc requests for it on every stat.
4829 Frequency : Always, for filesystems larger than 2TB
4831 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
4832 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
4833 filesystems larger than 2TB to be mounted with mballoc enabled.
4836 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
4838 Description: filesystem corruption for non-standard kernels and very large OSTs
4839 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
4840 filesystems larger than 2TB is mounted then the kernel will
4841 silently corrupt the start of the filesystem. CONFIG_LBD is
4842 enabled for all CFS-supported kernels, but the possibility of
4843 this happening with a modified kernel config exists.
4845 Severity : enhancement
4847 Description: add client O_DIRECT support for 2.6 kernels
4848 Details : It is now possible to do O_DIRECT reads and writes to files
4849 in the Lustre client mountpoint on 2.6 kernel clients.
4851 Severity : enhancement
4853 Description: parallel glimpse, setattr, statfs, punch, destroy requests
4854 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
4855 parallel, not waiting for response from every OST before sending
4856 a rpc to the next OST.
4861 Description: setattr vs write race when updating file timestamps
4862 Details : Client processes that update a file timestamp into the past
4863 right after writing to the file (e.g. tar) it is possible that
4864 the updated file modification time can be reset to the current
4865 time due to a race between processing the setattr and write RPC.
4867 Severity : enhancement
4869 Description: Bring 'lfs find' closer in line with regular Linux find.
4870 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
4871 -print0 options and obtains all the needed info through the lustre
4874 Severity : enhancement
4876 Description: support up to 1024 configured devices on one node
4877 Details : change obd_dev array from statically allocated to dynamically
4878 allocated structs as they are first used to reduce memory usage
4883 Description: Flush dirty partially truncated pages during truncate
4884 Details : Immediatelly flush partially truncated pages in filter_setattr,
4885 this way we completely avoid having any pages in page cache on OST
4886 and can retire ugly workarounds during writes to flush such pages.
4891 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
4892 Details : protect inode from truncation within vfs_unlink() context
4893 just take a reference before calling vfs_unlink() and release it
4894 when parent's i_sem is free.
4897 Frequency : always, if extents are used on OSTs
4899 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
4900 Details : OSTs using the extents format would not zero the high 16 bits of
4901 the index physical block number. This is not a problem for any
4902 OST filesystems smaller than 16TB, and no kernels support ext3
4903 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
4904 new/modified files) and can be fixed for existing filesystems
4905 with e2fsprogs-1.39-cfs1.
4910 Description: import connection selection may be incorrect if timer wraps
4911 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
4912 import connections to be ignored if the 32-bit jiffies counter
4913 wraps. Use a 64-bit jiffies counter.
4916 Frequency : during server recovery
4918 Description: crash after server is denying duplicate export
4919 Details : If clients are resending connect requests to the server, the
4920 server refuses to allow a client to connect multiple times.
4921 Fixed a bug in the handling of this case.
4924 Frequency : very large clusters immediately after boot
4926 Description: LNET request buffers exhausted under heavy short-term load
4927 Details : If a large number of client requests are generated on a service
4928 that has previously never seen so many requests it is possible
4929 that the request buffer growth cannot keep up with the spike in
4930 demand. Instead of dropping incoming requests, they are held in
4931 the LND until the RPC service can accept more requests.
4934 Frequency : Sometimes during replay
4936 Description: Assertion failure in ll_local_open after replay.
4937 Details : If replay happened on an open request reply before we were able
4938 to set replay handler, reply will become not swabbed tripping the
4939 assertion in ll_local_open. Now we set the handler right after
4940 recognising of open request
4943 Frequency : very rare
4945 Description: kernel reports "badness in vsnprintf"
4946 Details : Reading from the "recovery_status" /proc file in small chunks
4947 may cause a negative length in lprocfs_obd_rd_recovery_status()
4948 call to vsnprintf() (which is otherwise harmless). Exit early
4949 if there is no more space in the output buffer.
4951 Severity : enhancement
4953 Description: clear OBD RPC statistics by writing to them
4954 Details : It is now possible to clear the OBD RPC statistics by writing
4955 to the "stats" file.
4960 Description: Client mtime is not the same on different clients after utimes
4961 Details : In some cases, the client was using the utimes() syscall on
4962 a file cached on another node. The clients now validate the
4963 ctime from the MDS + OSTs to determine which one is right.
4968 Description: Inability to activate failout mode
4969 Details : lconf script incorrectly assumed that in python string's numeric
4970 value is used in comparisons.
4973 Frequency : always with multiple stripes per file
4975 Description: Inefficient object allocation for mutli-stripe files
4976 Details : When selecting which OSTs to stripe files over, for files with
4977 a stripe count that divides evenly into the number of OSTs,
4978 the MDS is always picking the same starting OST for each file.
4979 Return the OST selection heuristic to the original design.
4984 Description: mount failures may take full timeout to return an error
4985 Details : Under some heavy load conditions it is possible that a
4986 failed mount can wait for the full obd_timeout interval,
4987 possibly several minutes, before reporting an error.
4988 Instead return an error as soon as the status is known.
4990 ------------------------------------------------------------------------------
4992 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
4994 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4995 INTEROPERATE with older versions automatically. Please read the
4996 user documentation before upgrading any part of a live system.
4997 * WARNING: Lustre networking configuration changes are required with
4998 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
5001 * Support for kernels:
5002 2.6.9-22.0.2.EL (RHEL 4)
5003 2.6.5-7.244 (SLES 9)
5004 2.6.12.6 vanilla (kernel.org)
5007 Severity : enhancement
5008 Bugzilla : 7981/8208
5009 Description: Introduced Lustre Networking (LNET)
5010 Details : LNET is new networking infrastructure for Lustre, it includes
5011 a reorganized network configuration mode (see the user
5012 documentation for full details) as well as support for routing
5013 between different network fabrics. Lustre Networking Devices
5014 (LNDs) for the supported network fabrics have also been
5015 created for this new infrastructure.
5017 Severity : enhancement
5018 Description: Introduced Access control lists
5019 Details : clients can set ACLs on files and directories in order to have
5020 more fine-grained permissions than the standard Unix UGO+RWX.
5021 The MDS must be started with the "-o acl" mount option.
5023 Severity : enhancement
5024 Description: Introduced filesystem quotas
5025 Details : Administrators may now establish per-user quotas on the
5028 Severity : enhancement
5030 Description: Configuration change for the XT3
5031 The PTLLND is now used to run Lustre over Portals on the XT3
5032 The configure option(s) --with-cray-portals are no longer used.
5033 Rather --with-portals=<path-to-portals-includes> is used to
5034 enable building on the XT3. In addition to enable XT3 specific
5035 features the option --enable-cray-xt3 must be used.
5040 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
5041 Details : A race between allocating a new llcd and re-getting the llcd_lock
5042 allowed another thread to grab newly-allocated llcd.
5044 Severity : enhancement
5046 Description: 2.6 OST async journal commit and locking fix to improve performance
5047 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
5048 2.6 kernels are now async as they already were in 2.4 kernels so
5049 that they can commit concurrently with the network bulk transfer.
5050 For block-allocated files the filter allocation semaphore is held
5051 to avoid filesystem fragmentation during allocation. BKL lock
5052 removed for 2.6 xattr operations where it is no longer needed.
5057 Description: lconf incorrectly determined whether two IP networks could talk
5058 Details : In some more complicated routing and multiple-network
5059 configurations, lconf will avoid trying to make a network
5060 connection to a disjoint part of the IP space. It was doing the
5061 math incorrectly for one set of cases.
5066 Description: Fix for potential infinite loop processing records in an llog.
5067 Details : If an llog record is corrupted/zeroed, it is possible to loop
5068 forever in llog_process(). Validate the llog record length
5069 and skip the remainder of the block on error.
5072 Frequency : occasional (liblustre only)
5074 Description: liblustre could not open files whose last component is a symlink
5075 Details : sysio_path_walk() would incorrectly pass the open intent to
5076 intermediate path components.
5079 Frequency : rare (liblustre only with non-standard tuning)
5080 Bugzilla : 7201 (7350)
5081 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
5082 Details : llu_lookup_finish_locks() tries to set lock data on a lock
5083 after it has been released, only do this for referenced locks
5085 Severity : enhancement
5087 Description: specifying an (invalid) directory default stripe_size of -1
5088 would reset the directory default striping
5089 Details : stripe_size -1 was used internally to signal directory stripe
5090 removal, now use "all default" to signal dir stripe removal
5091 as a directory striping of "all default" is not useful
5094 Frequency : common for large clusters running liblustre clients
5096 Description: doing an ls when liblustre clients are running is slow
5097 Details : sending a glimpse AST to a liblustre client waits for every AST
5098 to time out, as liblustre clients will not respond. Since they
5099 cannot cache data we refresh the OST lock LVB from disk instead.
5101 Severity : enhancement
5103 Description: doing an ls at the same time as file IO can be slow
5104 Details : enqueue and other "small" requests can be blocked behind many
5105 large IO requests. Create a new OST IO portal for non-IO
5106 requests so they can be processed faster.
5109 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
5111 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
5112 created file with mode 000
5113 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
5114 allow an ftruncate() as a non-root user to a file with mode 000.
5115 Lustre now respects this flag to disable mode checks when
5116 truncating a file owned by the user
5119 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
5122 Description: Revoking locks from clients that went dead or catatonic might take
5124 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
5125 cancellation of such locks instant on servers without waiting for
5126 any reply from clients. Clients drops these locks when cancel
5127 notification from server is received without replying.
5130 Frequency : liblustre-only, when liblustre client dies or becomes busy
5132 Description: Doing ls on Linux clients can take a long time with active
5134 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
5135 granting such locks to it in the first place if possible. Locks
5136 are taken by proxy on the OST during the read or write and
5137 dropped immediately afterward. Add connect flags handling, do
5138 not grant locks to liblustre clients for glimpse ASTs.
5140 Severity : enhancement
5142 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
5143 Details : Existing read-ahead algorithm is tuned for the case of streamlined
5144 sequential reads and behaves badly with applications doing random
5145 reads. Improve it by reading ahead at least read region, and
5146 avoiding excessive large RPC for small reads.
5148 Severity : enhancement
5150 Description: Creating more than 1000 files for a single job may cause a load
5151 imbalance on the OSTs if there are also a large number of OSTs.
5152 Details : qos_prep_create() uses an OST index reseed value that is an
5153 even multiple of the number of available OSTs so that if the
5154 reseed happens in the middle of the object allocation it will
5155 still utilize the OSTs as uniformly as possible.
5160 Description: OST or MDS may oops in ping_evictor_main()
5161 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
5162 but doesn't restart at beginning of obd_exports_timed list
5165 Severity : enhancement
5167 Description: improve by-nid export eviction on the MDS and OST
5168 Details : allow multiple exports with the same NID to be evicted at one
5169 time without re-searching the exports list.
5172 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
5174 Description: MDS may oops in groups_free()
5175 Details : in rare race conditions a newly allocated group_info struct is
5176 freed again, and this can be NULL. The 2.4 compatibility code
5177 for groups_free() checked for a NULL pointer, but 2.6 did not.
5180 Frequency : common for liblustre clients doing little filesystem IO
5181 Bugzilla : 9352, 7313
5182 Description: server may evict liblustre clients accessing contended locks
5183 Details : if a client is granted a lock or receives a completion AST
5184 with a blocking AST already set it would not reply to the AST
5185 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
5188 Frequency : lfs setstripe, only systems with more than 160 OSTs
5190 Description: unable to set striping with a starting offset beyond OST 160
5191 Details : llapi_create_file() incorrectly limited the starting stripe
5192 index to the maximum single-file stripe count.
5195 Frequency : LDAP users only
5197 Description: lconf did not handle in-kernel recovery with LDAP properly
5198 Details : lconf/LustreDB get_refs() is searching the wrong namespace
5200 Severity : enhancement
5202 Description: bind OST threads to NUMA nodes to improve performance
5203 Details : all OST threads are uniformly bound to CPUs on a single NUMA
5204 node and do their allocations there to localize memory access
5206 Severity : enhancement
5208 Description: llmount can determine client NID directly from Myrinet (GM)
5209 Details : the client NID code from gmnalnid was moved directly into
5210 llmount, removing the need to use this or specifying the
5211 client NID explicitly when mounting GM clients with zeroconf
5214 Frequency : if client is started with down MDS
5216 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
5217 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
5218 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
5219 ptlrpc_import_delayed_req() because we didn't check if the
5220 request was interrupted, and we also didn't break out of the
5221 event loop if there was no timeout
5226 Description: data loss during non-page-aligned writes to a single file from
5227 both multiple nodes and multiple threads on one node at same time
5228 Details : updates to KMS and lsm weren't protected by common lock. Resulting
5229 inconsistency led to false short-reads, that were cached and later
5230 used by ->prepare_write() to fill in partially written page,
5231 leading to data loss.
5234 Frequency : always, if lconf --abort_recovery used
5236 Description: lconf --abort_recovery fails with 'Operation not supported'
5237 Details : lconf was attempting to abort recovery on the MDT device and not
5240 Severity : enhancement
5242 Description: remove cleanup logs
5243 Details : replace lconf-generated cleanup logs with lustre internal
5244 cleanup routines. Eliminates the need for client-cleanup and
5247 Severity : enhancement
5249 Description: add support for EAs (user and system) on lustre filesystems
5250 Details : it is now possible to store extended attributes in the Lustre
5251 client filesystem, and with the user_xattr mount option it
5252 is possible to allow users to store EAs on their files also
5254 Severity : enhancement
5256 Description: Add possibility (config option) to show minimal available OST free
5258 Details : When compiled with --enable-mindf configure option, statfs(2)
5259 (and so, df) will return least minimal free space available from
5260 all OSTs as amount of free space on FS, instead of summary of
5261 free spaces of all OSTs.
5263 Severity : enhancement
5265 Description: do not expand extent locks acquired on OST-side
5266 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
5267 by server: they are not cached anyway.
5270 Frequency : when mmap is used/binaries executed from Lustre
5272 Description: Unmmap pages before throwing them away from read cache.
5273 Details : llap_shrink cache now attempts to unmap pages before discarding
5274 them (if unmapping failed - do not discard). SLES9 kernel has
5275 extra checks that trigger if this unmapping is not done first.
5280 Description: lconf didn't resolve symlinks before checking to see whether a
5281 given mountpoint was already in use
5284 Frequency : when migrating failover services
5285 Bugzilla : 6395, 9514
5286 Description: When migrating a subset of services from a node (e.g. failback
5287 from a failover service node) the remaining services would
5288 time out and evict clients.
5289 Details : lconf --force (implied by --failover) sets the global obd_timeout
5290 to 5 seconds in order to quickly disconnect, but this caused
5291 other RPCs to time out too quickly. Do not change the global
5292 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
5294 Severity : enhancement
5295 Frequency : if MDS is started with down OST
5296 Bugzilla : 9439,5706
5297 Description: Allow startup/shutdown of an MDS without depending on the
5298 availability of the OSTs.
5299 Details : Asynchronously call mds_lov_synchronize during MDS startup.
5300 Add appropriate locking and lov-osc refcounts for safe
5301 cleaning. Add osc abort_inflight calls in case the
5302 synchronize never started.
5305 Frequency : occasional (Cray XT3 only)
5307 Description: root not authorized to access files in CRAY_PORTALS environment
5308 Details : The client process capabilities were not honoured on the MDS in
5309 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
5310 been accessed by an authorized user then root was able to access
5311 the file on the local client also. The root user capabilities
5312 are now allowed on the MDS, as this environment has secure UID.
5315 Frequency : occasional
5317 Description: ldiskfs "too long searching" message happens too often
5318 Details : A debugging message (otherwise harmless) prints too often on
5319 the OST console. This has been reduced to only happen when
5320 there are fragmentation problems on the filesystem.
5325 Description: Division by zero in statfs when all OSCs are inactive
5326 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
5327 lov_statfs divides by value returned by lov_get_stripecnt().
5331 Bugzilla : 9489, 3273
5332 Description: First write from each client to each OST was only 4kB in size,
5333 to initialize client writeback cache, which caused sub-optimal
5334 RPCs and poor layout on disk for the first writen file.
5335 Details : Clients now request an initial cache grant at (re)connect time
5336 and so that they can start streaming writes to the cache right
5337 away and always do full-sized RPCs if there is enough data.
5338 If the OST is rebooted the client also re-establishes its grant
5339 so that client cached writes will be honoured under the grant.
5344 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
5345 Details : Now I/O RPCs go to different portal number and (presumably) fast
5346 lock requests (and glimses) and other RPCs get their own service
5347 threads pool that should be able to service those RPCs
5350 Severity : enhancement
5352 Description: Ability to exchange lustre version between client and servers and
5353 issue warnings at client side if client is too old. Also for
5354 liblustre clients there is ability to refuse connection of too old
5356 Details : New 'version' field is added to connect data structure that is
5357 filled with version info. That info is later checked by server and
5361 Frequency : rare, liblustre only.
5362 Bugzilla : 9296, 9581
5363 Description: Two simultaneous writes from liblustre at offset within same page
5364 might proceed at the same time overwriting eachother with stale
5366 Details : I/O lock withing llu_file_prwv was released too early, before data
5367 actually was hitting the wire. Extended lock-holding time until
5368 server acknowledges receiving data.
5371 Frequency : extremely rare. Never observed in practice.
5373 Description: avoid generating lustre_handle cookie of 0.
5374 Details : class_handle_hash() generates handle cookies by incrementing
5375 global counter, and can hit 0 occasionaly (this is unlikely, but
5376 not impossible, because initial value of cookie counter is
5377 selected randonly). Value of 0 is used as a sentinel meaning
5378 "unassigned handle" --- avoid it. Also coalesce two critical
5379 sections in this function into one.
5381 Severity : enhancement
5383 Description: allow liblustre clients to delegate truncate locking to OST
5384 Details : To avoid overhead of locking, liblustre client instructs OST to
5385 take extent lock in ost_punch() on client's behalf. New connection
5386 flag is added to handle backward compatibility.
5388 Severity : enhancement
5389 Bugzilla : 4928, 7341, 9758
5390 Description: allow number of OST service threads to be specified
5391 Details : a module parameter allows the number of OST service threads
5392 to be specified via "options ost ost_num_threads={N}" in the
5393 OSS's /etc/modules.conf or /etc/modprobe.conf.
5397 Bugzilla : 6146, 9635, 9895
5398 Description: servers crash with bad pointer in target_handle_connect()
5399 Details : In rare cases when a client is reconnecting it was possible that
5400 the connection request was the last reference for that export.
5401 We would temporarily drop the export reference and get a new
5402 one, but this may have been the last reference and the export
5403 was just destroyed. Get new reference before dropping old one.
5405 Severity : enhancement
5406 Frequency : if client is started with failover MDS
5408 Description: Allow multiple MDS hostnames in the mount command
5409 Details : Try to read the configuration from all specified MDS
5410 hostnames during a client mount in case the "primary"
5413 Severity : enhancement
5415 Description: Stop sending data to evicted clients as soon as possible.
5416 Details : Check if the client we are about to send or are sending data to
5417 was evicted already. (Check is done every second of waiting,
5418 for which l_wait_event interface was extended to allow checking
5419 of exit condition at specified intervals).
5422 Frequency : rare, normally only when NFS exporting is done from client
5424 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
5426 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
5429 Frequency : always, if srand() is called before liblustre initialization
5431 Description: Liblustre uses system PRNG disturbing its usage by user application
5432 Details : Introduce internal to lustre fast and high-quality PRNG for
5433 lustre usage and make liblustre and some other places in generic
5434 lustre code to use it.
5436 Severity : enhancement
5437 Bugzilla : 9477, 9557, 9870
5438 Description: Verify that the MDS configuration logs are updated when xml is
5439 Details : Check if the .xml configuration logs are newer than the config
5440 logs stored on the MDS and report an error if this is the case.
5441 Request --write-conf, or allow starting with --old_conf.
5443 Severity : enhancement
5445 Description: Handle symlinks in the path when checking if Lustre is mounted.
5446 Details : Resolve intermediate symlinks when checking if a client has
5447 mounted a filesystem to avoid duplicate client mounts.
5452 Description: lconf can hit an error exception but still return success.
5453 Details : The lconf command catches the Command error exception at the top
5454 level script context and will exit with the associated exit
5455 status, but doesn't ensure that this exit status is non-zero.
5460 Description: failure of ptlrpc thread startup can cause oops
5461 Details : Starting a ptlrpc service thread can fail if there are a large
5462 number of threads or the server memory is very fragmented.
5463 Handle this without oopsing.
5466 Frequency : always, only if liblustre and non-default acceptor port was used
5468 Description: liblustre cannot connect to servers with non-default acceptor port
5469 Details : tcpnal_set_default_params() was not called and was therefore
5470 ignoring the environment varaible TCPNAL_PORT, as well as other
5471 TCPNAL_ environment variables
5476 Description: two objects could be created on the same OST for a single file
5477 Details : If an OST is down, in some cases it was possible to create two
5478 objects on a single OST for a single file. No problems other
5479 than potential performance impact and spurious error messages.
5483 Bugzilla : 5681, 9562
5484 Description: Client may oops in ll_unhash_aliases
5485 Details : Client dcache may become inconsistent in race condition.
5486 In some cases "getcwd" can fail if the current directory is
5492 Description: Inode refcounting problems in NFS export code
5493 Details : link_raw functions used to call d_instantiate without obtaining
5494 extra inode reference first.
5498 Bugzilla : 9942, 9903
5499 Description: Referencing freed requests leading to crash, memleaks with NFS.
5500 Details : We used to require that call to ll_revalidate_it was always
5501 followed by ll_lookup_it. Also with revalidate_special() it is
5502 possible to call ll_revalidate_it() twice for the same dentry
5503 even if first occurence returned success. This fix changes semantic
5504 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
5505 reference on a request referred from the intent.
5506 ll_intent_release() then releases such a request.
5509 Frequency : rare, normally benchmark loads only
5511 Description: unlinked inodes were kept in memory on the client
5512 Details : If a client is repeatedly creating and unlinking files it
5513 can accumulate a lot of stale inodes in the inode slab cache.
5514 If there is no other client load running this can cause the
5515 client node to run out of memory. Instead flush old inodes
5516 from client cache that have the same inode number as a new inode.
5519 Frequency : SLES9 2.6.5 kernel and long filenames only
5520 Bugzilla : 9969, 10379
5521 Description: utime reports stale NFS file handle
5522 Details : SLES9 uses out-of-dentry names in some cases, which confused
5523 the lustre dentry revalidation. Change it to always use the
5527 Frequency : rare, unless heavy write-truncate concurrency is continuous
5528 Bugzilla : 4180, 6984, 7171, 9963, 9331
5529 Description: OST becomes very slow and/or deadlocked during object unlink
5530 Details : filter_destroy() was holding onto the parent directory lock
5531 while truncating+unlinking objects. For very large objects this
5532 may block other threads for a long time and slow overall OST
5533 responsiveness. It may also be possible to get a lock ordering
5534 deadlock in this case, or run out of journal credits because of
5535 the combined truncate+unlink. Solution is to do object truncate
5536 first in one transaction without parent lock, and then do the
5537 final unlink in a new transaction with the parent lock. This
5538 reduces the lock hold time dramatically.
5541 Frequency : rare, 2.4 kernels only
5543 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
5544 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
5545 before unmounting their filesystem in order to avoid deadlock.
5546 The kernel_locked() function in 2.4 kernels only checks whether
5547 the kernel lock is held, not whether it is this process that is
5548 holding it as 2.6 kernels do.
5553 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
5554 Details : The client ptlrpc code may be trying to reconnect to a down
5555 server before a previous connection attempt has timed out.
5556 Increase the reconnect interval to be longer than the connection
5557 timeout interval to avoid sending duplicate connections to
5561 Frequency : echo_client brw_test command
5563 Description: fix echo_client to work with OST preallocated code
5564 Details : OST preallocation code (5137) didn't take echo_client IO path
5565 into account: echo_client calls filter methods outside of any
5566 OST thread and, hence, there is no per-thread preallocated
5567 pages and buffers to use. Solution: hijack pga pages for IO. As
5568 a byproduct, this avoids unnecessary data copying.
5572 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
5573 Description: Client can oops in mdc_commit_close() after open replay
5574 Details : It was possible for the MDS to return an open request with no
5575 transaction number in mds_finish_transno() if the client was
5576 evicted, but without actually returning an error. Clients
5577 would later try to replay that open and may trip an assertion
5578 Simplify the client close codepath, and always return an error
5579 from the MDS in case the open is not successful.
5582 Frequency : rare, 2.6 OSTs only
5584 Description: OST may deadlock under high load on fragmented files
5585 Details : If there was a heavy load and highly-fragmented OST filesystems
5586 it was possible to have all the OST threads deadlock waiting on
5587 allocation of biovecs, because the biovecs were not released
5588 until the entire RPC IO was completed. Instead, release biovecs
5589 as soon as they are complete to ensure forward IO progress.
5591 Severity : enhancement
5593 Description: Support for specifying external journal device at mount
5594 Details : If an OST or MDS device is formatted with an external journal
5595 device, this device major/minor is stored in the ext3 superblock
5596 and may not be valid for failover. Allow detecting and
5597 specifying the external journal at mount time.
5602 Description: Mounting an MDS with pending unlinked files may cause oops
5603 Details : target_finish_recovery() calls mds_postrecov() which returned
5604 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
5605 considers this an error and immediately begins cleaning up the
5606 lov, just after starting the mds_lov process
5608 Severity : enhancement
5610 Description: Implement 'lfs df' to report actual free space on per-OST basis
5611 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
5612 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
5613 usage of objects; '-h' to report in human readable format.
5615 ------------------------------------------------------------------------------
5617 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
5624 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
5625 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
5626 at startup during orphan unlinking. Instead, skip buddy update
5627 before setup, as it will be regenerated after recovery is complete.
5632 Description: If an OST is inactive, its locks might reference stale inodes.
5633 Details : lov_change_cbdata() must iterate over all namespaces, even if
5634 they are inactive to clear inode references from the lock.
5636 Severity : enhancement
5637 Frequency : occasional, if non-standard max_dirty_mb used
5639 Description: Client will block write RPCs if not enough grant
5640 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
5641 then the client will block writes while waiting for another RPC
5642 to complete instead of consuming its dirty limit. With change
5643 we get improved performance when max_dirty_mb is small.
5645 Severity : enhancement
5646 Bugzilla : 3389, 6253
5647 Description: Add support for supplementary groups on the MDS.
5648 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
5649 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
5650 lookups for user supplementary groups into a cache.
5654 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
5655 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
5656 for liblustre O_EXCL is enforced by MDS.
5660 Description: readdir never returns NULL in liblustre.
5661 Details : Corrected llu_iop_getdirentries logic, to return offset of next
5662 dentry in struct dirent.
5666 Frequency : liblustre only, depends on application IO pattern
5667 Description: liblustre clients evicted if not contacting servers
5668 Details : Don't put liblustre clients into the ping_evictor list, so
5669 they will not be evicted by the pinger ever.
5671 Severity : enhancement
5673 Description: Add ability to evict clients by NID from MDS.
5674 Details : By echoing "nid:$NID" string into
5675 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
5676 $NID would be instantly evicted from this MDS and from all active
5677 OSTs connected to it.
5681 Description: Do not query file size twice, somewhat slowing stat(2) calls.
5682 Details : lookup_it_finish() used to query file size from OSTs that was not
5687 Description: service threads change working directory to that of init
5688 Details : Starting lustre service threads may pin the working directory
5689 of the parent thread, making that filesystem busy. Threads
5690 now change to the working directory of init to avoid this.
5694 Frequency : during shutdown only
5695 Description: shutdown with a failed MDS or OST can cause unmount to hang
5696 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
5701 Frequency : 2.6 only
5702 Description: chmod/chown may include an extra supplementary group
5703 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
5704 supplementary group and if none is specified this is used.
5707 Bugzilla : 5479 (6816)
5709 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
5710 Details : If lookup is in progress on a file that is unlinked we might try
5711 to revalidate the inode and fail in revalidate after lookup is
5712 complete and ll_file_open() enqueues the open again but
5713 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
5716 Frequency : always, if lconf --abort_recovery used
5718 Description: lconf --abort_recovery fails with 'Operation not supported'
5719 Details : lconf was attempting to abort recovery on the MDT device and not
5722 ------------------------------------------------------------------------------
5724 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
5729 Frequency : rare (only unsupported configurations with a node running as an
5731 Bugzilla : 6514, 5137
5732 Description: Mounting a Lustre file system on a node running as an OST could
5734 Details : OSTs now preallocates memory needed to write out data at
5735 startup, instead of when needed, to avoid having to
5736 allocate memory in possibly low memory situations.
5737 Specifically, if the file system is mounted on on OST,
5738 memory pressure could force it to try to write out data,
5739 which it needed to allocate memory to do. Due to the low
5740 memory, it would be unable to do so and the node would
5741 become unresponsive.
5743 Severity : enhancement
5745 Description: Addition of lconf --service command line option
5746 Details : lconf now accepts a '--service <arg>' option, which is
5747 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
5749 Severity : enhancement
5751 Description: Failover mode is now the default for OSTs.
5752 Details : By default, OSTs will now run in failover mode. To return to
5753 the old behaviour, add '--failout' to the lmc line for OSTs.
5755 Severity : enhancement
5757 Description: Health checks are now provided for MDS and OSTs
5758 Details : Additional detailed health check information on MSD and OSTs
5759 is now provided through the procfs health_check value.
5762 Frequency : occasional, depends on IO load
5764 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
5765 after numerous create/delete cycles
5766 Details : The ext3 inode allocation policy would not allocate new inodes
5767 very well on the OSTs because there are no new directories
5768 being created. Instead we look for groups with free space if
5769 the parent directories are nearly full.
5774 Description: Network or server problems during mount may cause partially
5775 mounted clients instead of returning an error.
5776 Details : The config llog parsing code may overwrite the error return
5777 code during mount error handling, returning success instead
5783 Description: MDS can fail to allocate large reply buffers
5784 Details : After long uptimes the MDS can fail to allocate large reply
5785 buffers (e.g. zconf client mount config records) due to memory
5786 fragmentation or consumption by the buffer cache. Preallocate
5787 some large reply buffers so that these replies can be sent even
5788 under memory pressure.
5792 Frequency : rare (liblustre)
5793 Description: fsx running with liblustre complained that using truncate() to
5794 extend the file doesn't work. This patch corrects that issue.
5795 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
5796 ATTR_SIZE and lsm use in llu_setattr_raw.
5800 Frequency : rare, only 2.6 kernels
5801 Description: Unusual file access patterns on the MDS may result in inode
5802 data being lost in very rare circumstances.
5803 Details : Bad interaction between the ea-in-inode patch and the "no-read"
5804 code in the 2.6 kernel caused the inode and/or EA data not to
5805 be read from disk, causing single-file corruption.
5809 Frequency : rare, only 2.6 filesystems using extents
5810 Description: Heavy concurrent write and delete load may cause data corruption.
5811 Details : It was possible under high-load situations to have an extent
5812 metadata block in the block device cache from a just-unlinked
5813 file overwrite a newly-allocated data block. We now unmap any
5814 metadata buffers that alias just-allocated data blocks.
5818 Frequency : filesystems with default stripe_count larger than 77
5819 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
5820 Details : lconf specifies an inode size of 4096 bytes when the default
5821 stripe_count is larger than 77. This conflicts with the default
5822 inode density of 1 per 4096 bytes. Allocate smaller inodes in
5823 this case to avoid pinning too much memory for large EAs.
5825 ------------------------------------------------------------------------------
5827 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
5832 Frequency : rare (extremely heavy IO load with hundreds of clients)
5834 Description: Client is evicted, gets IO error writing to file
5835 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
5836 caused clients to be evicted for AST timeouts. The fixes in
5837 bug 5192 mean we no longer need to have such short AST timeouts
5838 so ldlm_timeout has been increased.
5841 Frequency : occasional during --force or --failover shutdown under load
5842 Bugzilla : 5949, 4834
5843 Description: Server oops/LBUG if stopped with --force or --failover under load
5844 Details : a collection of import/export refcount and cleanup ordering
5845 issues fixed for safer force cleanup
5848 Frequency : only filesystems larger than 120 OSTs
5849 Bugzilla : 5990, 6223
5850 Description: lfs getstripe would oops on a very large filesystem
5851 Details : lov_getconfig used kfree on vmalloc'd memory
5854 Frequency : only filesystems exporting via NFS to Solaris 10 clients
5855 Bugzilla : 6242, 6243
5856 Description: reading from files that had been truncated to a non-zero size
5857 but never opened returned no data
5858 Details : ll_file_read() reads zeros from no-object files to EOF
5863 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
5864 Details : There's little harm in aborting MDS/OSS recovery and letting it
5865 try again, so I removed the LASSERT and return an error instead.
5867 Severity : enhancement
5869 Description: New debugging infrastructure for tracking down data corruption
5870 Details : The I/O checksum code was replaced to: (a) control it at runtime,
5871 (b) cover more of the client-side code path, and (c) try to narrow
5872 down where problems occurred
5876 Bugzilla : 3819, 4364, 4397, 6313
5877 Description: Racing close and eviction MDS could cause assertion in mds_close
5878 Details : It was possible to get multiple mfd references during close and
5879 client eviction, leading to one thread referencing a freed mfd.
5881 Severity: : enhancement
5882 Bugzilla : 3262, 6359
5883 Description: Attempts to reconnect to servers are now more aggressive.
5884 Details : This builds on the enhanced upcall-less recovery that was added
5885 in 1.4.2. When trying to reconnect to servers, clients will
5886 now try each server in the failover group every 10 seconds. By
5887 default, clients would previously try one server every 25 seconds.
5892 Description: After recovery, certain operations trigger a failed
5893 assertion on a client.
5894 Details : Failing over an mds, using lconf -d --failover, while a
5895 client was doing a readdir() call would cause the client to
5896 LBUG after recovery completed and the readdir() was resent.
5898 Severity : enhancement
5900 Description: Default groups are now added by lconf
5901 Details : You can now run lconf --group <servicename> without having to
5902 manually add groups with lmc.
5905 Frequency : occasional
5907 Description: Nodes with an elan id of 0 trigger a failed assertion
5910 Frequency : always when accessing e.g. tty/console device nodes
5912 Description: tty and some other devices nodes cannot be used on lustre
5913 Details : file's private_data field is used by device data and lustre
5914 values in there got lost. New field was added to struct file to
5915 store fs-specific private data.
5918 Frequency : when exporting Lustre via NFS
5920 Description: NFSD failed occasionally when looking up a path component
5921 Details : NFSD is looking up ".." which was broken in ext3 directories
5922 that had grown large enough to become hashed.
5925 Frequency : Clusters with multiple interfaces not on the same subnet
5927 Description: Nodes will repeatedly try to reconnect to an interface which it
5928 cannot reach and report an error to the log.
5929 Details : Extra peer list entries will be created by lconf with some peers
5930 unreachable. lconf now validates the peer before adding it.
5933 Frequency : Only if a default stripe is set on the filesystem root.
5935 Description: Setting a default stripe on the filesystem root prevented the
5936 filesystem from being remounted.
5937 Details : The client was sending extra request flags in the root getattr
5938 request and did not allocate a reply buffer for the dir EA.
5941 Frequency : occasional, higher if lots of files are accessed by one client
5942 Bugzilla : 6159, 6097
5943 Description: Client trips assertion regarding lsm mismatch/magic
5944 Details : While revalidating inodes the VFS looks up inodes with ifind()
5945 and in rare cases can find an inode that is being freed.
5946 The ll_test_inode() code will free the lsm during ifind()
5947 when it finds an existing inode and then the VFS later attaches
5948 this free lsm to a new inode.
5952 Bugzilla : 6422, 7030
5953 Description: MDS deadlock between mkdir and client eviction
5954 Details : Creating a new file via mkdir or mknod (starting a transaction
5955 and getting the ns lock) can deadlock with client eviction
5956 (gets ns lock and trying to finish a synchronous transaction).
5959 Frequency : occasional
5960 Description: While starting a server, the fsfilt_ext3 module could not be
5962 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5963 kernels. Previously, lconf would still use the ext3 name
5964 when trying to load modules. Now, it will correctly use
5965 ext3 on 2.4 and ldiskfs on 2.6.
5967 Severity : enhancement
5968 Description: The default stripe count has been changed to 1
5969 Details : The interpretation of the default stripe count (0, to lfs
5970 or lmc) has been changed to mean striping across a single
5971 OST, rather than all available. For general usage we have
5972 found a stripe count of 1 or 2 works best.
5974 Severity : enhancement
5975 Description: Add support for compiling against Cray portals.
5976 Details : Conditional compiling for some areas that are different
5980 Frequency : occasional
5981 Bugzilla : 6409, 6834
5982 Description: Creating files with an explicit stripe count may lead to
5983 a failed assertion on the MDS
5984 Details : If some OSTs are full or unavailable, creating files may
5985 trigger a failed assertion on the MDS. Now, Lustre will
5986 try to use other servers or return an error to the
5990 Frequency : occasional
5992 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5993 caused lock timeout during readahead (since 1.4.2).
5994 Details : Processes doing readahead might match a lock that hasn't been
5995 granted yet if there are overlapping and conflicting lock
5996 requests. The readahead process waits on ungranted lock
5997 (original lock is CBPENDING), while OST waits for that process
5998 to cancel CBPENDING read lock and eventually evicts client.
6000 Severity : enhancement
6002 Description: Initial enabling of flock support for clients
6003 Details : Implements fcntl advisory locking and file status functions.
6004 This feature is provided as an optional mount flag (default
6005 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
6006 locking are implemented yet, and those that are are not guaranteed
6007 to be completely correct in production environments.
6008 mount -t lustre -o [flock|noflock] ...
6011 Frequency : occasional
6013 Description: OSTs running 2.4 kernels but with extents enabled might trip an
6014 assertion in the ext3 JBD (journaling) layer.
6015 Details : The b_committed_data struct is protected by the big kernel lock
6016 in 2.4 kernels, serializing journal_commit_transaction() and
6017 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
6018 access to this struct. In 2.6 kernels there is finer grained
6019 locking to improve SMP performance of the JBD layer.
6023 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
6025 -----------------------------------------------------------------------------
6027 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
6029 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
6030 versions of Lustre. Please update all servers and clients to
6031 version 1.4.2 or later at the same time. You must also run
6032 "lconf --write-conf {config}.xml" on the MDS while it is stopped
6033 to update the configuration logs.
6035 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
6036 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
6037 - swab llog records of type '0' so we get proper header size/idx (5861)
6038 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
6039 - fix rename of one directory over another leaking an inode (5953)
6040 - avoid SetPageDirty on 2.6 (5981)
6041 - don't re-add just-being-destroyed locks to the waiting list (5653)
6042 - when creating new directories, inherit the parent's custom
6043 striping settings if present parent (3048)
6044 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
6045 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
6046 - don't hold client locks on temporary worklist from l_lru (5666)
6047 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
6048 - automatically evict dead clients (5921)
6049 - Update file size properly in create+truncate+fstat case (6196)
6050 - Do not unhash mountpoint dentries, do not allow removal of
6052 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
6053 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
6054 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
6055 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
6056 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
6057 - Correctly handle reads of files with no objects (6243)
6058 - lctl recover will also mark a device active if deactivate used (5933)
6060 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
6061 - allow --write-conf on an MDS with different nettype than client (5619)
6062 - don't write config llogs to MDS for mounts not from that MDS (5617)
6063 - lconf should create multiple TCP connections from a client (5201)
6064 - init scripts are now turned off by default; run chkconfig --on
6065 lustre and chkconfig --on lustrefs to use them
6066 - upcalls are no longer needed for clients to recover to failover
6068 - add --abort-recovery option to lconf to abort recovery on device
6070 - add support for an arbitrary number of OSTs (3026)
6071 - Quota support protocol changes.
6072 - forward compatibility changes to wire structs (6007)
6073 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
6074 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
6075 - improved "lustre status" script
6076 - initialize blocksize for non-regular files (6062)
6077 - added --disable-server and --disable-client configure options (5782)
6078 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
6079 - Vanilla 2.4.29 support
6080 - increase maximum number of obd devices to 520 (6242)
6081 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
6082 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
6083 SLES 9 2.6.5 kernels
6084 - limit stripes per file to 160 (the maximum EA size) (6093)
6086 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
6089 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
6090 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
6091 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
6092 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
6093 - small ext3 extents cleanups and fixes (5733)
6094 - improved mballoc code, several small races and bugs fixed (5733, 5638)
6095 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
6096 - avoid needless client->OST connect, fix handle mismatch (5317)
6097 - fix DLM error path that led to out-of-sync client, long delays (5779)
6098 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
6099 - fix several locking issues related to i_size (5492,5624,5654,5672)
6100 - don't move pending lock onto export if it is already evicted (5683)
6101 - fix kernel oops when creating .foo in unlinked directory (5548)
6102 - fix deadlock in obdfilter statistics vs. object create (5811)
6103 - use time_{before,after} to avoid timer jiffies wrap (5882)
6104 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
6105 - Do not leak request if server was not able to process it (5154)
6106 - If mds_open unable to find parent dir, make that negative lookup(5154)
6107 - don't create new directories with extent-mapping (5909, 5936)
6109 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
6110 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
6111 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
6112 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
6113 - support for RHEL4 kernel on i686 (5773)
6114 - provide error messages when incompatible logs are encountered (5898)
6116 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
6117 * version 1.4.0.10 (1.4.1 release candidate 1)
6119 - don't keep a lock reference when lock is not granted (4238)
6120 - unsafe list practices (rarely) led to infinite eviction loop (4908)
6121 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
6122 - drop import inflight refcount on signal_completed_replay error (5255)
6123 - unlock page after async write error during send (3677)
6124 - handle missing objects in filter_preprw_read properly (5265)
6125 - no transno return for symlink open, don't save no-trasno open (3440)
6126 - don't try to complete elan receive that already failed (4012)
6127 - free RPC server reply state on error (5406)
6128 - clean up thread from ptlrpc_start_thread() on error (5160)
6129 - readahead could read extra page into cache that wasn't ejected (5388)
6130 - prevent races in class_attach/setup/cleanup/detach (5260)
6131 - don't dereference de->d_inode after l_dput of de (5458)
6132 - use "int" for stripe value returned from lock_to_stripe (5544)
6133 - mballoc allocation and error-checking fixes in 2.6 (5504)
6134 - block device patches to fix I/O request sizes in 2.6 (5482)
6135 - look up hostnames for IB nals (5602)
6136 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
6137 - don't start multiple acceptors for the same port (5277)
6138 - fix incorrect LASSERT in mds_getattr_name (5635)
6139 - export a proc file for general "ping" checking (5628)
6140 - fix "lfs check" to not block when the MDS is down (5628)
6142 - service request history (4965)
6143 - put {ll,lov,osc}_async_page structs in a single slab (4699)
6144 - create an "evict_client" /proc entry on OSTs, like the MDS has
6145 - fix mount usage message, return errors per mount(8) (5168)
6146 - change grep [] to grep "[]" in tests so they work in more UMLs
6147 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
6148 - remove ancient LOV_MAGIC_V0 EA support (5047)
6149 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
6150 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
6151 - allow readahead allocations to fail when low on memory (5383)
6152 - mmap locking landed again, after considerable improvement (2828)
6153 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
6155 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
6158 - send OST transaction number in read/write reply to free req (4966)
6159 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
6160 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
6161 - the watchdog thread now runs as interruptible (5246)
6162 - flock/lockf fixes (but it's still disabled, pending 5135)
6163 - don't use EXT3 constants in llite code (5094)
6164 - memory shortage at startup could cause assertion (5176)
6166 - reorganization of lov code
6167 - single portals codebase
6169 - add extents/mballoc support (5025)
6170 - direct I/O reads in the obdfilter (4048)
6171 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
6173 tbd Cluster File Systems, Inc. <info@clusterfs.com>
6176 - send OST transaction number in read/write reply to free req (4966)
6177 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
6178 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
6179 - the watchdog thread now runs as interruptible (5246)
6180 - handle missing objects in filter_preprw_read properly (5265)
6181 - unsafe list practices (rarely) led to infinite eviction loop (4908)
6182 - drop import inflight refcount on signal_completed_replay error (5255)
6183 - unlock page after async write error during send (3677)
6184 - return original error code on reconstructed replies (3761)
6185 - no transno return for symlink open, don't save no-trasno open (3440)
6187 - add pid to ldlm debugging output (4922)
6188 - bump the watchdog timeouts -- we can't handle 30sec yet
6189 - extra debugging for orphan dentry/inode bug (5259)
6191 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
6194 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
6195 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
6196 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
6197 - fix NULL dereference in /proc/sys/portals/routes (4827)
6198 - allow failed mdc_close() operations to be interrupted (4561)
6199 - stop precreate on OST before MDS would time out on it (4778)
6200 - don't send partial-page writes before EOF from client (4410)
6201 - discard client grant for sub-page writes on large-page clients (4520)
6202 - don't free dentries not owned by NFS code, check generation (4806)
6203 - fix lsm leak if mds_create_objects() fails (4801)
6204 - limit debug_daemon file size, always print CERROR messages (4789)
6205 - use transno after validating reply (3892)
6206 - process timed out requests if import state changes (3754)
6207 - update mtime on OST during writes, return in glimpse (4829)
6208 - add mkfsoptions to LDAP (4679)
6209 - use ->max_readahead method instead of zapping global ra (5039)
6210 - don't interrupt __l_wait_event() during strace
6212 - add software watchdogs to catch hung threads quickly (4941)
6213 - make lustrefs init script start after nfs is mounted
6214 - fix CWARN/ERROR duplication (4930)
6215 - return async write errors to application if possible (2248)
6216 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
6217 - print NAL number in %x format (4645)
6218 - update barely-supported suse-2.4.21-171 series (4842)
6219 - support for sles 9 %post scripts
6220 - support for building 2.6 kernel-source packages
6221 - support for sles km_* packages
6223 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
6226 - ignore -ENOENT errors in osc_destroy (3639)
6227 - notify osc create thread that OSC is being cleaned up (4600)
6228 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
6229 - reconstruct ost_handle() like mds_handle() (4657)
6230 - create a new thread to do import eviction to avoid deadlock (3969)
6231 - let lconf resolve symlinked-to devices (4629)
6232 - don't unlink "objects" from directory with default EA (4554)
6233 - hold socknal file ref over connect in case target is down (4394)
6234 - allow more than 32000 subdirectories in a single directory (3244)
6235 - fix blocks count for O_DIRECT writes (3751)
6236 - OST returns ENOSPC from object create when no space left (4539)
6237 - don't send truncate RPC if file size isn't changing (4410)
6238 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
6239 - bind to privileged port in socknal and tcpnal (3689)
6241 - rate limit CERROR/CWARN console message to avoid overload (4519)
6242 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
6243 - basic mmap support (3918)
6244 - kernel patch series update from b1_4 (4711)
6246 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
6249 - avoid crash during MDS cleanup with OST shut down (2775)
6250 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
6251 - don't use bad inodes on the MDS (3744)
6252 - dynamic object preallocation to improve recovery speed (4236)
6253 - don't hold spinlock over lock dumping or change debug flags (4401)
6254 - don't zero obd_dev when it is force cleaned (3651)
6255 - print grants to console if they go negative (4431)
6256 - "lctl deactivate" will stop automatic recovery attempts (3406)
6257 - look for existing locks in ldlm_handle_enqueue() (3764)
6258 - don't resolve lock handle twice in recovery avoiding race (4401)
6259 - revalidate should check working dir is a directory (4134)
6261 - don't always mark "slow" obdfilter messages as errors (4418)
6263 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
6266 - don't close LustreDB during write_conf until it is done (3860)
6267 - fix typo in lconf for_each_profile (3821)
6268 - allow dumping logs from multiple threads at one time (3820)
6269 - don't allow multiple threads in OSC recovery (3812)
6270 - fix debug_size parameters (3864)
6271 - fix mds_postrecov to initialize import for llog ctxt (3121)
6272 - replace config semaphore with spinlock (3306)
6273 - be sure to send a reply for a CANCEL rpc with bad export (3863)
6274 - don't allow enqueue to complete on a destroyed export (3822)
6275 - down write_lock before checking llog header bitmap (3825)
6276 - recover from lock replay timeout (3764)
6277 - up llog sem before sending rpc (3652)
6278 - reduce ns lock hold times when setting kms (3267)
6279 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
6280 - fix NULL deref and obd_dev leak on setup error (3312)
6281 - replace some LBUG about llog ops with error handling (3841)
6282 - don't match INVALID dentries from d_lookup and spin (3784)
6283 - hold dcache_lock while marking dentries INVALID and hashing (4255)
6284 - fix invalid assertion in ptlrpc_set_wait (3880)
6286 - add libwrap support for the TCP acceptor (3996)
6287 - add /proc/sys/portals/routes for non-root route listing (3994)
6288 - allow setting MDS UUID in .xml (2580)
6289 - print the stack of a process that LBUGs (4228)
6291 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
6294 - don't cleanup request in ll_file_open() on failed MDS open (3430)
6295 - make sure to unset replay flag from failed open requests (3440)
6296 - if default stripe count is 0, use OST count for inode size (3636)
6297 - update parent mtime/ctime on client for create/unlink (2611)
6298 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
6299 - free recovery state on server during a forced cleanup (3571)
6300 - unregister_reply for resent reqs (3063)
6301 - loop back devices mounting and status check on 2.6 (3563)
6302 - fix resource-creation race that can provoke i_size == 0 (3513)
6303 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
6304 - more debugging for page-accounting assertion (3746)
6305 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
6306 - avoid deadlock after precreation failure (3758)
6307 - fix race and lock order deadlock in orphan handling (3450, 3750)
6308 - add validity checks when grabbing inodes from l_ast_data (3599)
6310 - add /proc/.../recovery_status to obdfilter (3428)
6311 - lightweight CDEBUG infrastructure, debug daemon (3668)
6312 - change default OSC RPC parameters to be better on small clusters
6313 - turn off OST read cache for files smaller than 32MB
6314 - install man pages and include them in rpms (3100)
6315 - add new init script for (un)mounting lustre filesystems (2593)
6316 - run chkconfig in %post for init scripts (3701)
6317 - drop scimac NAL (unmaintained)
6319 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
6322 - clean kiobufs before and after use (3485)
6323 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
6324 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
6325 - change init script to fail more clearly if not run as root (1528)
6326 - allow clients to reconnect during replay (1742)
6327 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
6328 - don't do DNS lookups on NIDs too small for IP addresses (3442)
6329 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
6330 - fix cond_resched (3554)
6331 - only evict unfinished clients after recovery (3515)
6332 - allow bulk resend, prevent data loss (3570)
6333 - dynamic ptlrpc request buffer allocation (2102)
6334 - don't allow unlinking open directory if it isn't empty (2904)
6335 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
6336 - remove extraneous obd dereference causing LASSERT failure (3334)
6337 - don't use get_cycles() when creating temp. files on the mds (3156)
6338 - hold i_sem when setting i_size in ll_extent_lock() (3564)
6339 - handle EEXIST for set-stripe, set proper directory name (3336)
6341 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
6342 - fix ksocknal_fmb_callback() error messages (2918)
6344 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
6347 - don't copy lvb into (possibly NULL) reply on error (2983)
6348 - don't deref dentry after dput, don't free lvb on error (2922)
6349 - use the kms to determine writeback rpc length (2947)
6350 - increment oti_logcookies when osc is inactive (2948)
6351 - update client's i_blocks count via lvb messages (2543)
6352 - handle intent open/close of special files properly (1557)
6353 - mount MDS with errors=remount-ro, like obdfilter (2009)
6354 - initialize lock handle to avoid ASSERT on error cleanup (3057)
6355 - don't use cancelling-locks' kms values (2947)
6356 - use highest lock extent for kms, not last one (2925)
6357 - don't dereference ERR_PTR() dentry in error handling path (3107)
6358 - fix thread race in portals_debug_dumplog() (3122)
6359 - create lprocfs device entries at setup instead of at attach (1519)
6360 - common AST error handler, don't evict client on completion race (3145)
6361 - zero nameidata in detach_mnt in 2.6 (3118)
6362 - verify d_inode after revalidate_special is valid in 2.6 (3116)
6363 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
6364 - initialize RPC timeout timer earlier for 2.6 (3219)
6365 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
6366 - print nal/nid for unknown nid (3258)
6367 - additional checks for oscc recovery before doing precreate (3284)
6368 - fix ll_extent_lock() error return code for 64-bit systems (3043)
6369 - don't crash in mdc_close for bad permissions on open (3285)
6370 - zero i_rdev for non-device files (3147)
6371 - clear page->private before handing to FS, better assertion (3119)
6372 - tune the read pipeline (3236)
6373 - fix incorrect decref of invalidated dentry (2350)
6374 - provide read-ahead stats and refine rpc in flight stats (3328)
6375 - don't hold journal transaction open across create RPC (3313)
6376 - update atime on MDS at close time (3265)
6377 - close LDAP connection when recovering to avoid server load (3315)
6378 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
6379 - don't leak open file on MDS after open resend (3325)
6380 - serialize filter_precreate and filter_destroy_precreated (3329)
6381 - loop device shouldn't call sync_dev() for nul device (3092)
6382 - clear page cache after eviction (2766)
6383 - resynchronize MDS->OST in background (2824)
6384 - refuse to mount the same filesystem twice on same mountpoint (3394)
6385 - allow llmount to create routes for mounting behind routers (3320)
6386 - push lock cancellation to blocking thread for glimpse ASTs (3409)
6387 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
6388 - fix rare problem with rename on htree directories (3417)
6390 - allow default OST striping configuration per directory (1414)
6391 - fix compilation for qswnal for 2.6 kernels (3125)
6392 - increase maximum number of MDS request buffers for large systems
6393 - change liblustreapi to be useful for external progs like lfsck (3098)
6394 - increase local configuration timeout for slow disks (3353)
6395 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
6397 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
6400 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
6401 - don't overwrite extent policy data in reply if lock was blocked (2901)
6402 - drop filter export grants atomically with removal from device (2663)
6403 - del obd_self_export from work_list in class_disconnect_exports (2908)
6404 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
6405 - swab reply message in mdc_close, other PPC fixes (2464)
6406 - fix destroying of named logs (2325)
6407 - overwrite old logs when running lconf --write_conf (2264)
6408 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
6409 - fix race in target_handle_connect (2898)
6410 - mds_reint_create() should take same inode create lock (2926)
6411 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
6412 - don't close files for self_export to avoid uninitialized obd (2936)
6413 - allow MDS with the same name as client node (2939)
6414 - hold dentry reference for closed log files for unlink (2325)
6415 - reserve space for all logs during transactions (2059)
6416 - don't evict page beyond end of stripe extent (2925)
6417 - don't oops on a deleted current working directory (2399)
6418 - handle hard links to targets without a parent properly (2517)
6419 - don't dereference NULL lock when racing during eviction (2867)
6420 - don't grow lock extents when lots of conflicting locks (2919)
6422 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
6425 - account for cache space usage on clients to avoid data loss (974)
6426 - lfsck support in lustre kernel code (2349)
6427 - reduce journal credits needed for BRW writes (2370)
6428 - orphan handling to avoid losing space on client/server crashes
6429 - ptlrpcd can be blocked, stopping ALL progress (2477)
6430 - use lock value blocks to assist in proper KMS, faster stat (1021)
6431 - takes i_sem instead of DLM locks internally on obdfilter (2720)
6432 - recovery for initial connections (2355)
6433 - fixes for mds_cleanup_orphans (1934)
6434 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
6435 - block all file creations until orphan recovery completes (1901)
6436 - client remove rq_connection from request struct (2423)
6437 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
6438 - recovery timer race (2670)
6439 - mdc_close recovey bug (2532)
6440 - ptlrpc cleanup bug (2710)
6441 - mds timeout on local locks (2588)
6442 - namespace lock held during RPCs (2431)
6443 - handle interrupted sync write properly (2503)
6444 - don't try to handle a message that hasn't been replied to (2699)
6445 - client assert failure during cleanup after abort recovery (2701)
6446 - leak mdc device after failed mount (2712)
6447 - ptlrpc_check_set allows timedout requests to complete (2714)
6448 - wait for inflight reqs when ptlrpcd finishes (2710)
6449 - make sure unregistered services are removed from the srv_list
6450 - reset bulk XID's when resending them (caught by 1138 test)
6451 - unregister_bulk after timeout
6452 - fix lconf error (2694)
6453 - handle write after unfinished setstripe, stripe-only getstripe (2388)
6454 - readahead locks pages, leaves pending causing memory pressure (2673)
6455 - increase OST request buffers to 4096 on large machines (2729)
6456 - fix up permission of existing directories in simple_mkdir (2661)
6457 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
6458 - don't assign transno to errored transactions (2742)
6459 - don't delete objects on OST if given a bogus objid from MDS (2751)
6460 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
6461 - if rq_no_resend, then timeout request after recovery (2432)
6462 - fix MDS llog_logid record size, 64-bit array alignment (2733)
6463 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
6464 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
6465 - MDS recovery shouldn't delete objects that it has given out (2730)
6466 - if enqueue arrives after completion, don't clobber LVB (2819)
6467 - don't unlock pages twice when trigger_group_io returns error (2814)
6468 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
6469 - don't write pages to disk if there was an error (1450)
6470 - don't ping imports that have recovery disabled (2676)
6471 - take buffered bytes into account when balancing socknal conn (2817)
6472 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
6473 - reconnect unlink llog connection after MDS reconnects to OST (2816)
6474 - remove little-endian swabbing of llog records (1987)
6475 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
6476 - retry reposting request buffers if they fail (1191)
6477 - grow extent at grant time to avoid granting a revoked lock (2809)
6478 - lock revoke doesn't evict page if covered by a second lock (2765)
6479 - disable VM readahead to avoid reading outside lock extents (2805)
6481 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
6482 - updated kernel patches for hp-2.4.20 kernel (2681)
6484 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
6487 - fix truncated write corruption (2366)
6488 - fix for failed assertion in iopen_connect_dentry (1792,2517)
6490 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
6491 - ensure len(uuid) < 37 in lmc (1171)
6492 - fix ia64 OOPS in llog_test (2255)
6493 - zero end of page at obdfilter for partial page writes (2648)
6494 - don't leave stale dentries around after renames (bug 2428)
6495 - fix timeouts when evicting a client with a single lock held (2642)
6496 - set deadline for the initial HELLO message to drain (2634)
6497 - print out dotted-quad IP addresses in the socknal (2302)
6499 - additional debugging for MDS client eviction problem (2443)
6500 - fix mkfsoptions support for osts (2603, 2604)
6502 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
6505 - add series for the vanilla 2.6.0 kernel
6506 - add series for the vanilla 2.4.24 kernel
6507 - add series for a cray x86/64 UL kernel drop
6508 - fix xattr patches for the vanilla 2.4.19 series
6510 - generate true UUIDs in lmc (1171)
6511 - have portals stack dumping break in UML (2466)
6512 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
6513 - allocate with _NOFS in ldlm to avoid deadlock (1933)
6514 - wake callback waiting threads on client eviction (2460)
6515 - Add --ptldebug and --subsystem to lmc (1719)
6516 - update assertion to allow safe interrupt allocation
6517 - set rq_no_resend for cancel requests (2432)
6518 - recalculate ptlrpcd timeout after resend (2494)
6519 - call vfs_rmdir when removing pending directories (2368)
6520 - fix renaming a file to itself (2429)
6521 - lmc creates a default one-stripe lov (2454)
6522 - expand procfs space to handle large clusters (2326)
6523 - increase UML stack to avoid overflow
6524 - update lconf's list of debug and subsystem masks
6525 - fix lfs find --obd (2510)
6526 - /proc tunable for disabling filter read caching (2591)
6527 - stop rpm packages from altering slapd.conf (2301)
6528 - disable nagle in the socknal under 0conf (2578)
6529 - choose mds inode size based on stripe count (2572)
6530 - fix kernel-source rpm problems (2516)
6532 - add --disable-doc to avoid pdf generation (2421)
6533 - update documentation, tests, type-os, comments
6534 - avoid format warnings on ia64
6535 - remove the TOE NAL
6536 - tiny code cleanups by removing unused fields
6538 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
6541 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
6542 - remove the most blatant lies from BUILDING file (2371)
6543 - change default debug level to reasonable production setting
6544 - reduce client side cache size to reduce cache flush time
6545 - reduce max RPCs in flight to avoid unnecessary file fragmentation
6546 - make TCP zerocopy and pinger support enabled by default (2476)
6547 - sync writes completed after process exits caused crashes (2319)
6548 - maintain correct mount count on the MDS (2356)
6549 - backout 1557, because 2316 wasn't really fixed
6550 - better file I/O statistics gathering in /proc
6551 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
6552 - another kernel patch to fix zero-copy TCP function export
6553 - don't take duplicate lock when processing re-sent getattr (2420)
6554 - lctl uses obd_self_export instead of creating new conn (2353)
6555 - MDS/OST recovery case which requires object creation asserted (2425)
6556 - move lfs from /usr/sbin to /usr/bin in packages
6557 - fix race between mds_client_add and mds_client_free (2417)
6558 - use kmalloc instead of slabs in portals (2430)
6559 - don't create duplicate records when a failover MDS is present (2442)
6560 - remove unnecessary mount age check (2332)
6561 - don't remove directory inodes from locks prematurely (2451)
6562 - don't break if MDS service name is the same as hostname (2103)
6563 - fix races in client write RPC generation when cache full (2482)
6565 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
6568 - remove now-unused request->rq_obd (278)
6569 - if an allocation fails, print out how much memory we've used (1933)
6570 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
6571 - add the "configurable stack size" patch to most series files (1256)
6572 - ability to write large log records, for 100+ OST configs (2306)
6573 - fix NULL deref when filter_prep fails (2314)
6574 - fix operator precedence error in filter_sync
6575 - dynamic allocation of socknal TX descriptors (2315)
6576 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
6577 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
6578 - ensure that CWARN messages in Portals always get to the syslog
6579 - __init/__exit are not for prototype decls (ldlm_init/exit)
6580 - x86-64 compile warning fixes
6581 - fix gateway LMC keyword conflict (2318)
6582 - fix MDS lock inversions in getattr/reint paths (1844)
6583 - fix a rare lock re-ordering bug, which caused deadlock (2322)
6584 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
6585 - DLM race condition prevented some lock evictions (2328)
6586 - ENOMEM detection and retry on socknal sends (2230)
6587 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
6588 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
6589 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
6590 - better error messages when a client is rejected during recovery (1505)
6591 - avoid cancelling locks which were never granted, after failure (2330)
6592 - fix i_sem/journal inversion in mds_client_add (2333)
6593 - fix truncate/getattr lock cycle deadlock (2334)
6594 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
6595 - fix two rare exit paths which could leak an l_lock() ref (2321)
6596 - fencepost error in MDS/OST orphan recovery (2226)
6597 - make log record alignment 8 bytes (1988)
6598 - lstripe now fails when requested offset > ost_count (2237)
6599 - ensure that all kernel series have a complete list.h (1607)
6600 - fix crashes in special-file operations (2316)
6601 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
6603 - allow configurable automake binary, for testing new versions
6604 - small update to the lfs documentation
6606 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
6608 * fix negative export reference count in fsfilt_sync (2312)
6610 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
6611 * release candidate 0.9.1
6613 - orphans are moved into the PENDING directory for possible recovery
6614 - replayed opens now open by fid for orphan/rename safety (1042)
6615 - last close of an orphan inode generates a transno (683)
6616 - chdir() and mount() now pin the directory entry (1020)
6617 - avoid CERROR in normal ll_setattr_raw() error case (1500)
6618 - discard very old requests without processing them (1502)
6619 - remove some common, well-understood CERRORs (1505)
6620 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
6621 - clear "grant" flags in OST replies until OST grant code lands (1644)
6622 - fix read performance by not clobbering i_blksize on client (1598)
6623 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
6624 - make LRU size a /proc tunable, clears locks when reduced (707)
6625 - fix some lprocfs rot that prevented ptlbd from loading (1732)
6626 - server locks take references on exports now (1558)
6627 - build fixes for 2.4.20-rh trees (1663)
6628 - return an error from lov_create if all OSCs are inactive (1751)
6629 - fix import levels when a reconnect happens without a timeout (1597)
6630 - exit early from mds_open if we get a lookup error (1749)
6631 - partial page read at EOF wouldn't wait for disk before sending (1642)
6632 - avoid NULL deref in obdfilter when reading page past EOF (1592)
6633 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
6634 - fix LBUG in ll_it_open_error with rc = -2 (1861)
6635 - write/truncate lock inversion (1639)
6636 - Don't auto-load obdclass, portals modules during cleanup (1495)
6637 - fix timestamps from jumping to "now" (1763)
6638 - extra journal assertions (1648)
6639 - add an extra multiunlink test (1771)
6640 - fix read_record/write_record API (1776)
6641 - fix leak of offset_extent, possible incorrect i_size later (1772)
6642 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
6643 - Debugging for the kqswnal_get_idle_tx problems (1820)
6644 - Allow recovery to be attempted multiple times (1536)
6645 - Write out MDS last_rcvd file after it is first created (1600)
6646 - Fix tx_descriptor leak in failed transmit situations (1827)
6647 - ext3 journaling fixes for assertion failure after IO error (1871)
6648 - class_export_put() on freed export after completion AST error (1896)
6649 - Fix revalidate looping in VFS (1322)
6650 - Don't access a freed export during MDS_REINT timeout (1521)
6651 - Add open-unlink recovery support on the MDS (1673,1764)
6652 - Return an error if no MDS data was read from last_rcvd (1946)
6653 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
6654 - Don't setup a disk device that doesn't match exported UUID (317)
6655 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
6656 - avoid committing NULL handle in force close
6657 - local.sh is now a one-stripe LOV configuration
6658 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
6659 - fix ext3 htree duplicate directory entry corruption (1516)
6660 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
6661 - update to new LOV EA format (2097)
6662 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
6664 2003-06-15 Phil Schwan <phil@clusterfs.com>
6667 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
6668 - per-import/export recovery handling (958, 931, 959)
6669 - multiple last-rcvd slots, for serving multiple FSes (949)
6670 - connections are again shared between multiple imp/exports (963, 964)
6671 - "umount -f" would hang if any requests needed to be sent (393, 978)
6672 - avoid pinning large req buffer by copying for queued messages (989)
6673 - add "uuid" to "lctl device" command to help upcalls (991)
6674 - "open" RPCs with transnos would confuse recovery counters (1037)
6675 - do proper endian conversion of all wire messages (288, 340, 891)
6676 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
6677 - call waiting locks callback from LDLM recovery thread (1127, 1151)
6678 - fix ptlrpc_connection leak in target_handle_connect (1174)
6679 - fix import refcounting bug in OST and MDS cleanup (1134)
6680 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
6681 - fix very unlikely obd_types race condition (501)
6682 - remove osc_open hack for echo_client (1187)
6683 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
6684 - a failure in read_inode2 leads to deadlock (1139)
6685 - cancel ack-locks as soon as transaction is committed (1072)
6686 - fix major leaks and crashes in the bulk I/O path (937, 1057)
6687 - make sure to commitrw after any preprw to avoid deadlock (1162)
6688 - failing to execute a file in a lustre FS would lock inode (1203)
6689 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
6690 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
6691 - obd_uuid2tgt would walk off the end of the list (1255)
6692 - on IA64 the osc would give portals incorrect bulk size (1258)
6693 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
6694 - fix lock inversion caused by new llite matching code (1282)
6695 - limit the number of dirty pages on a client to 10MB (1286)
6696 - timed out locks were not being corrected cancelled (1289)
6697 - fix O_DIRECT above 4GB on IA-32 (1292)
6698 * major user-visible changes
6699 - fail out/fail over policy now controlled by the upcall (993)
6701 - add OBD_PING to check server availability and failure (954)
6702 - lustre messages are now sent in sending host order (288, 340, 891)
6703 - add eadatalen to MDS getattr reply (340)
6704 - OST read replies may contain second buffer, with per-page status (593)
6706 2003-03-11 Phil Schwan <phil@clusterfs.com>
6709 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
6710 - failed open()s could cause deadlock; fixed (867, 869)
6711 - stop cancelling OST locks when files are closed (481)
6712 - overlapping XID spaces caused network corruption (851, 853)
6713 - fix unsafe fsfilt counter arithmetic; change to atomic_t
6714 - setattr_raw added, to do single-RPC, server-side setattrs
6715 - lmc/lconf syntax change for OST UUIDs
6716 - fix crashy race condition between ptlrpc_free_req and osc_close
6717 - don't use request in mdc_enqueue if we hit a timeout (889)
6718 - don't set the inode i_size for regular files from the MDS (896)
6719 - handle out of order completion AST (842)
6720 - don't LBUG if a lock request times out after receiving AST (913)
6721 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
6722 - if a bad lock AST arrives, send an error instead of dropping entirely
6723 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
6724 - fix leak in bulk IO when only partially completed (899, 900, 926)
6725 - fix O_DIRECT for ia64 (55)
6726 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
6727 - C-z after timeout could hang a process forever; fixed (977)
6729 - client-side I/O cache (678, 924, 929, 941, 970)
6731 - READPAGE and SETATTRs which don't take server-side locks get
6734 2003-02-11 Phil Schwan <phil@clusterfs.com>
6737 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
6738 - Fix fsfilt_extN_readpage() to read a full page of directory
6739 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
6740 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
6741 - Fix imbalanced LOV object allocation and out-of-bound access (469)
6742 - Most intent operations were removed, in favour of a new RPC mode
6743 that does a single RPC to the server and bypasses most of the VFS
6744 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
6745 - Aggressively cancel local locks on DLM servers
6746 - mds_reint_unlink sends EA to the client if it's the last nlink.
6747 client uses that EA to unlink OST objects.
6748 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
6749 - recursive symlinks were fixed (439)
6750 - fixed NULL deref in DEBUG_REQ
6751 - filter_update_lastobjid no longer calls sync, which annoyed extN
6752 - fixed multi-client small-writes to a single file problem (445)
6753 - fixed mtime updates during file writes (607)
6754 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
6755 - fixed bug in obd_brw_read/write() (under guise of testing 367)
6756 - fixed Linux OST size reporting problem (444, 656)
6757 - OST now updates object mtime with writes or setattr (607, 619)
6758 - client verifies file size before zeroing page past EOF (445)
6759 - OST now writes last allocated objid to disk with allocation (108)
6760 - LOV on echo now works (409)
6762 - mds_reint_unlink sends a new buffer, with the EA included. this
6763 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
6764 set if a regular file was being unlinked, and it was the last link
6765 - use PtlGet from the target for bulk writes (315)
6766 - OST now updates object mtime with writes or setattr (607, 619)
6767 - LDLM now has a grant-time callback to revalidate locked items, if
6769 - Many MDS operations were reorganized to combat race conditions
6771 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
6772 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
6774 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
6777 - Fully reactivate OST imports after reconnection (512, others)
6778 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
6779 - More graceful error handling for truncating on dead OST (515)
6780 - Don't error out unless we're actually accessing dead stripes (474)
6781 - Fix garbage sizes when stripes are missing (410)
6782 - LRU counters were broken, causing constant lock purge (433, 432)
6783 - garbage on read from stripes with failed OSTs (441)
6784 - mark OSCs as active before reconnecting during recovery (438)
6785 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
6786 - lfind did not preserve OST order in output (443)
6787 - symlinks cause hung clients, incorrect data (439)
6788 - stop dereferencing request after dropping refcount (457)
6789 - don't LASSERT(spin_is_locked) on non-SMP (455)
6790 - fixes for many rename() bugs
6791 - fstat didn't correctly synchronize attributes (399)
6792 - server must handle lock cancellation during blocking AST prep (487)
6793 - bulk descriptors were free()d too soon (511)
6794 - fix paths in lconf, which would load incorrect modules (451, 507)
6795 - fix confusing lconf 'host not found' error message (386)
6796 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
6797 - fix race condition in mdc_blocking_ast() for inode access (526)
6798 - fix lov_unpackmd() unpacking wrong number of stripes (537)
6799 - fix lov_set_osc_active() marking wrong OSC inactive (440)
6800 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
6801 - fix multiple writes of stripe MD to MDS (358, maybe 519)
6802 - fix lstripe in several ways (kernel side) (527)
6803 - fix request leak in ldlm_cli_enqueue (262)
6804 - incorrect OSC was marked inactive after OST failure
6805 - call mds_fs_cleanup before unmounting filesystem (524)
6806 - fix races between taking ns_lock and ldlm_lock_change_resource
6807 - fix races updating LOV export open file list
6808 - fix lov_enqueue error path, avoid decref-ing bad lock handle
6809 - fix recovery NULL deref in ldlm_cli_cancel_unused
6810 - fix some DLM races by using new hash table for lock handles (419)
6811 - permit the client to specify desired inodes, at replay
6812 - duplicate requests when we queue them for replay reintegration
6813 - fix last_rcvd offset calculation
6814 - sync after each recovered transaction, so we always make progress
6815 - never, not always, ERESTART requests without transnos
6816 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
6818 - skip replay if the MDS says that the client is already connected
6819 - don't check for a recovery-enabled export to match lctl's UUID
6820 - don't INC_USE_COUNT for phantom exports
6821 - don't crash when cleaning up phantom exports (567)
6822 - don't double-finish or set replay data for errored mdc_open requests
6823 - abort requests when they time out, so we don't get old replies
6824 - send/receive replies for AST messages again
6825 - if the client says that it doesn't have the lock, cancel it on the
6827 - if we timeout during I/O, don't try to cancel an in-use lock; instead
6828 mark it as destroyed, it will all work out when decref is called
6829 - fix module use counts (22, 581)
6831 - ASTs now expect a reply (server cancels lock on error reply)
6833 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
6836 - fix many simultaneous client startup (392)
6837 - fix dentry->d_it clobbering
6838 - credentials weren't being shipped for readdir/getattr operations
6839 - remove invalid assertions triggered during some concurrent MD
6841 - proper Lustre versions added (336, 389)
6842 - fix memory leak for create error case (398)
6843 - fix LOV locking bug that would get cli/srv out of sync
6844 - fix echo client over LOV (409)
6845 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
6846 - fix double-O_EXCL intent crash (424)
6847 - avoid sending multiple lock CANCELs (352)
6849 - MDS can do multi-client recovery (modulo bugs in new code)
6851 - many updates, edits, cleanups
6853 2002-11-18 Phil Schwan <phil@clusterfs.com>
6856 - fix null d_it dereference (346)
6857 - fix full OST/dbench hang (333)
6858 - fix permission problem with file removal (286)
6859 - fix removal of OSCs from LOV when they fail
6860 - fix NULL deref during bulk timeout (214)
6861 - fix problems related to multiple filesystems on one MDS (241)
6862 - fixed serious subtle metadata locking bugs
6863 - free locks on clients when inodes are removed due to memory
6865 - fix inode pointer in lock data (285)
6866 - partial support for multiple MDS on a single host (241)
6867 - data locks weren't cancelled at clear_inode time (290, 311)
6868 - intent locks could lead to unbounded lock growth (205)
6869 - added a maximum lock count, an LRU list, and a flusher
6870 - fix multiple rename (365)
6871 - properly abstracted the echo client
6872 - OSC locked 1 byte too many; fixed
6873 - rewrote brw callback code:
6874 - fixed recovery bugs related to LOVs (306)
6875 - fixed too-many-pages-in-one-write crash (191)
6876 - fixed (again) crash in sync_io_timeout (214)
6877 - probably fixed callback-related race (385)
6879 - Add capability to MDS protocol
6880 - LDLM cancellations and callbacks on different portals
6882 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
6885 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
6886 - defer open object destruction to close time (601981, 138)
6887 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
6888 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
6889 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
6890 - don't call obd_setattr in ll_file_release for destroyed objects
6891 * protocol change to lustre_msg: move |version| and add |flags|
6892 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
6893 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
6894 * added replay of create, unlink, link and rename operations during
6895 MDS failover; recovery should be much more robust now
6896 * remove failed OSCs from LOVs (only lov_create uses this so far)
6897 * the lustre-HOWTO was brought (more) up to date (582544)
6899 2002-10-23 Phil Schwan <phil@clusterfs.com>
6902 - in-use dentries weren't being reused properly (617851)
6903 - prevent multiple LDLM setup (599178)
6904 - fix LOV size calculations for truncate (617853)
6905 - fix client handling of MDS intent errors (POSIX)
6906 - fix permission bug in lovstripe.c test (624321)
6907 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6908 - truncate past end of file could corrupt data
6909 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6910 - a race in recovery could return ETIMEDOUT to apps (623947)
6911 - building outside the source directory was fixed
6912 * the lustre-HOWTO was brought (more) up to date (582544)
6913 * major progress was made on recovery functionality
6915 2002-10-10 Phil Schwan <phil@clusterfs.com>
6918 - recovery deadlock fix
6919 - rm -rf causes LBUG fix (617817)
6920 - file open by multiple tasks fix (618962)
6921 - directory permissions bugs (602707 and 620007)
6922 - journal_stop fixed with locking (611313)
6923 - O_APPEND failures resolved (618273, perhaps 614459)
6924 - lconf PATH fix (619770)
6925 - IA64 build fix (621450)
6926 - RPC buffer sizes scale with amount of memory
6928 2002-10-01 Phil Schwan <phil@clusterfs.com>
6931 - locks would be cancelled without throwing away data pages,
6932 resulting in inconsistent data (605627)
6933 - inode attributes were not always being refreshed (605627, 612449)
6934 - lconf now continues to cleanup after lctl reports an error
6935 - MDS now enforces user permissions (602707)
6936 - lprocfs cleanup fixed, but not yet enabled (614157)
6937 - fixed infinite server hang, should a client not respond to an AST
6938 - avoid going into recovery if user calls readlink() with a buffer
6939 that's too small (613941)
6940 - AST RPCs no longer require replies (614867) -- this may be changed
6941 - don't crash server if client sends an IOV that's too big (611336)
6942 - fixed lock conversion deadlock (611892)
6943 - fixed the following of symlinks (614622)
6944 * recovery: the server can remove locks from a client that dies, other
6945 clients can make progress
6946 * more extN patch fixes
6947 * compile-time configurable ptlrpc buffer allocations
6949 - collaborative read cache document
6950 - Lustre Lite Performance CDR document-in-progress
6952 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6955 - fix typo in patch-2.4.18
6957 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6960 - clear ptlrpc request each time in handle_incoming_request()
6961 - unlink of files now destroys the object on the OST
6963 2002-09-19 Peter Braam <braam@clusterfs.com>
6965 * add hard link support
6966 * change obdfile creation method
6967 * kernel patch changed
6969 2002-09-19 Peter Braam <braam@clusterfs.com>
6972 - stack overflow bug in extN fixed
6974 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6976 * documentation updates
6977 - add man pages for config tools
6978 - update tests/README to describe testing with new config tools
6979 - finish metadata API descriptions
6980 * bug fixes and cleanups
6981 - statfs workaround for 16TB limit
6982 - LOV stripe allocation improved, can stripe on subset of OSTs
6983 - LOV file size/IO offset was wrong for files > 4GB in size
6984 - object EA data was being dropped, caused files to be unreadable
6985 - memory overflow with non-LOV OST caused memory corruption
6986 - fixed regression tests to work with new config tools, obdfilter
6987 - fixed bug when directory size became larger than 1 block
6988 - fixed bug (for single client case) when PWD was deleted
6989 - invalidate local directory pages when doing intent-based ops
6990 - avoid LDLM oops when lock callback contained bad data
6992 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6994 * documentation updates
6995 * bug fixes and cleanups
6996 - configuration tools
6999 - 64-bit compile warnings
7000 - 64-bit internal statfs data
7002 * test_brw on persistent OST devices
7004 * lprocfs (disabled)
7006 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
7008 * documentation updates
7009 * bug fixes and cleanups
7010 * configuration tools
7012 2002-08-30 Peter J. Braam <braam@clusterfs.com>
7015 * many small fixes to 0_5_4
7016 * io/network handling
7017 * thinkos in MDS operations
7019 2002-08-24 Peter J. Braam <braam@clusterfs.com>
7022 * crucial basic fixes to 0.5.3
7023 * IOR, Iozone work over Elan
7026 2002-08-07 Phil Schwan <phil@clusterfs.com>
7027 * version 0_5_3, our first alpha
7028 * we use the new Portals iovs
7029 * documentation updates
7030 * bug fixes and cleanups
7031 * small changes in the DLM wire protocol
7033 2002-07-25 Peter J. Braam <braam@clusterfs.com>
7034 * version 0_5_1 with some initial stability,
7035 * locking on MD and file I/O.
7036 * documentation updates
7037 * several bug fixes since 0.5.0
7038 * small changes in wire protocol
7040 2002-07-18 Phil Schwan <phil@clusterfs.com>
7042 * delivered as Lustre Light Alpha
7043 * fixed a crash after handling invalid MDS requests
7044 * fixed directory pages for architectures with non-4k pages sizes
7046 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
7047 * release version v0_4_4
7048 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
7050 * Fixes a number of interruption problems with OST operations.
7051 * Update documentation for portals header changes
7052 * Move all wire protocol structs/defines to lustre_idl.h
7053 * Fixes symlink length bug.
7054 * Add tcpdump to repository.
7056 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
7057 * release version v0_4_3
7058 * Fixes statfs for inodes on extN.
7059 * Fixes bug in runtests which would delete /etc/hosts.
7060 * Use 64-bit object IDs wherever possible (not into VFS though)
7061 Remove ost_get_info, which is unused by lustre, and out of date.
7063 2002-07-03 Peter Braam <braam@clusterfs.com>
7064 * release version v0_4_2 Fixes a lookup error (type not passed)
7065 * move forward to head of Portals
7066 * move forward to latest Lustre kernel
7068 2002-06-25 Peter Braam <braam@clusterfs.com>
7069 * release version v0_4_1. Hopefully stable on single node use.