4 2.6.16.60-0.42.8 (SLES 10),
5 2.6.27.39-0.3.1 (SLES11),
6 2.6.18-164.11.1.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
17 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
18 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
20 -------------------------------------------------------------------------------
22 2010-01-29 Sun Microsystems, Inc.
24 * Support for kernels:
25 2.6.16.60-0.42.8 (SLES 10),
26 2.6.27.39-0.3.1 (SLES11),
27 2.6.18-164.11.1.el5 (RHEL 5)
28 2.6.18-164.6.1.0.1.el5 (OEL 5)
29 * Client support for unpatched kernels:
30 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
31 2.6.16 - 2.6.30 vanilla (kernel.org)
32 * Recommended e2fsprogs version: 1.41.6.sun1
33 * The async journal commit feature (bug 19128) and the cancel
34 lock before replay feature (bug 16774) are disabled by default.
36 Severity : enhancement
38 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5.
40 Severity : enhancement
41 Bugzilla : 21511/19848
42 Description: Update SLES11 kernel to 2.6.27.39-0.3.1.
44 Severity : enhancement
46 Description: Update supported SLES10 kernel to 2.6.16.60-0.42.8.
48 Severity : enhancement
50 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
51 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
53 Severity : enhancement
55 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
58 Severity : enhancement
60 Description: Vanilla kernel 2.6.30 patchless client support.
65 Description: bad entry in directory xxx: inode out of bounds
66 Details : fix locking issue in the rename path which could race with any
67 other operations updating the same directory.
69 Severity : enhancement
71 Description: Make watchdog timer messages to be more clear and descriptive.
75 Description: cp -p command does not preserve the dates and timestamp
76 Details : mtime could be spoiled by a write callback
80 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
84 Description: Allow non-root access for "lfs check".
85 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
87 Severity : enhancement
89 Description: quotacheck performance/scaling issues
90 Details : reduce quotacheck time on empty filesystem by skipping uninit
93 Severity : enhancement
95 Description: Enhancement for lfs(1) command to use numeric uid/gid.
97 Severity : enhancement
99 Description: Adjust locks' extents on their first enqueue, so that at the time
100 they get granted, there is no need for another pass through the
101 queues since they are already shaped into the proper forms.
105 Description: Fix mds_shrink_intent_reply()/mds_intent_policy() to pass correct
106 arguments and prevent LBUG() in lustre_shrink_reply_v2().
110 Description: Change tunefs.lustre and mkfs.lustre --mountfsoptions so that
111 exactly the specified mount options are used. Leaving off
112 any "mandatory" mount options is an error. Leaving off any
113 default mount options causes a warning, but is allowed.
114 Change errors=remount-ro from mandatory to default. Sanitize
115 the mount string before storing it. Update man pages accordingly.
119 Description: mds_getattr() should return 0, even if mds_fid2entry() fails with
120 -ENOENT. Also fix in ptlrpc_expire_one_request() to print signed
123 Severity : enhancement
125 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
127 Severity : enhancement
129 Description: Clients can replay thousands of unused locks during recovery
130 Details : Don't replay unused locks (only read locks for now) during
131 recovery. This feature is disabled by default and can be
132 enabled by running the following command on the clients:
133 lctl get_param ldlm.cancel_unused_locks_before_replay
137 Description: can't stat file in some situation.
138 Details : improve initialize osc date when target is added to mds and
139 ability to resend too big getattr request is client isn't have info
144 Description: Prevent inconsistences between linux and lustre mount structures.
145 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
146 Make the sleep interruptible.
148 Severity : enhancement
150 Description: Communicate OST degraded/readonly state via statfs to MDS
151 Details : Flags in the statfs returned from OSTs indicate whether the
152 OST is in a degraded RAID state, or if the filesystem has
153 turned read-only after a filesystem error is detected.
158 Description: don't panic if EPROTO was hit when reading symlink
159 Details : correctly handling request reference in error cases.
164 Description: open sometimes returns ENOENT instead of EACCES
165 Details : checking permission should be part of open part of mds_open, not
166 lookup part. so server should be set DISP_OPEN_OPEN disposition
167 before starting permission check.
168 Also not need revalidate dentry if client already have LOOKUP lock.
172 Description: enable client interface failover
173 Frequency : on servers with multiple network interfaces
174 Details : When a child reconnects from another NID, properly update export
175 nid hash position and ldlm reverse import.
177 Severity : enhancement
179 Description: implemented direct I/O with arbitrary (nonaligned) memory
180 addresses and file offsets.
182 Severity : enhancement
184 Description: added more recovery timeout options.
186 Severity : enhancement
188 Description: added llapi_file_open, llapi_file_create, llapi_file_get_stripe
193 Description: Avoid deadlock for local client writes
194 Frequency : only on systems with clients writing to an OST on the same node
195 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
196 memory freeing context. This allows OST threads to set the
197 PF_MEMALLOC flag on task structures in order to allocate memory
198 from reserved pools and complete IO.
199 Use GFP_HIGHUSER for OST allocations for non-local client writes,
200 so that the OST threads generate memory pressure and allow
201 inactive pages to be reclaimed.
206 Descriptoin: lock ordering violation between &cli->cl_sem and _lprocfs_lock
207 Details : move ldlm namespace creation in setup phase to avoid grab
208 _lprocfs_lock with cli_sem held.
212 Frequency : only during format of test systems
213 Description: Unable to run several mkfs.lustre on loop devices at the same time
214 Details : mkfs.lustre returns error 256 on the concurrent loop devices
215 formatting. The solution is to proper handle the error.
217 Severity : enhancement
219 Description: implement async create (obd_async_create) method for osc, to avoid
220 too long waiting new ost objects with holding ldlm lock.
224 Frequency : occasionally during network problems
225 Description: client not allowed to reconnect to OST because of active request
226 Details : abort bulk requests received by the OST once the client has timed
227 out since the client will resend the request anyway. The client
228 also now retries to reconnect to the same server if a connect
229 request failed with EBUSY or -EAGAIN.
232 Frequency : rare, if used wide striped file and one ost in down.
234 Descriptoin: don't return error if we created a subset of objects for file.
235 Details : lov_update_create_set() uses set->set_success as index for created
236 objects, so if some requests failed, they will have hole at end of
237 array and we can use qos_shrink_lsm for allocate correct lsm.
241 Description: Slow stale export processing during normal start up
242 Details : The global mgc lock prevents OST setup to be run in parallel.
243 Replace the global lock with a per-config_llog_data semaphore.
247 Description: Out or order replies might be lost on replay
248 Details : In ptlrpc_retain_replayable_request if we cannot find retained
249 request with tid smaller then one currently being added, add it
250 to the start, not end of the list.
254 Description: BUG: soft lockup - CPU#1 stuck for 10s! [ll_mdt_07:4523]
255 Details : add cond_resched() calls to avoid hogging the cpu for too long
256 in the hash code. Make also lustre_hash_for_each_empty() more
259 Severity : enhancement
261 Description: Performance improvements for debug messages with D_RPCTRACE,
262 D_LDLM, D_QUOTA options.
265 Frequency : only with NFS export
267 Description: (lov_merge.c:74:lov_merge_lvb())
268 ASSERTION(spin_is_locked(&lsm->lsm_lock)) failed (SR 71691004)
269 Details : Fix a race in the nfs export code by populating inode
270 info while the new inode is still locked
272 Severity : enhancement
274 Description: add a new file in procfs called force_lbug. Writting to this
275 file triggers a LBUG. Only for test purpose.
279 Description: OOM killer causes node hang
280 Details : really interrupt the sleep in osc_enter_cache on signals
284 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
285 Details : fix race during quota release on the slave.
287 Severity : enhancement
289 Description: smaller hash bucket sizes, cleanups
290 Details : increase hash table sizes and enabled rehashing for pools, uuid,
293 Severity : enhancement
295 Description: Add ldiskfs maxdirsize mount option
296 Details : add max_dir size mount option
300 Description: panic in ll_statahead_thread
301 Details : prevent parent thread to be killed before its child
304 Frequency : only with 16TB device
306 Description: unable to perform "mount -t lustre" of 16TB OST device
307 Details : Mounting 16TB LUNs failed due to three bugs in mkfs.lustre.
311 Description: ASSERTION(atomic_read(&imp->imp_inflight) == 0) failed
312 Details : unregistering should be zero if no RPC inflight.
316 Description: hyperion: Oops during metabench
317 Details : Correct the refcount of lov_request_set
319 Severity : enhancement
321 Description: Add mptlinux and nxge drivers to Lustre builds
323 Severity : enhancement
325 Description: Fix watchdog timer message to be more clear
326 Details : Make watchdog timer messages more clear and descriptive.
330 Description: LNET soft lockups in socknal_cd thread
331 Details : don't hog CPU for active-connecting if another connd is
332 accepting connecting-requst from the same peer
336 Description: recovery-small test_17 hang
337 Details : Land several AT improvements & fixes.
341 Description: MDS panic and hanging client processes
342 Details : Replace exp_ops_stats with exp_nid_stats->nid_stats
346 Description: OSS stuck in recovery.
347 Details : fix race during recovery. class_unlink_export,
348 class_set_export_delayed and target_queue_last_replay_reply
349 may race while increasing/decreasing obd_recoverable_clients
350 and obd_delayed_clients, causing recovery to wait forever.
352 Severity : enhancement
354 Description: add cascading_rw.c to lustre/tests
358 Description: filter_last_id() NULL deref
359 Details : lprocfs_filter_rd_last_id() should check for the fully
360 setup obd device, before proceeding further.
362 Severity : enhancement
364 Description: Loadgen improvements
365 Details : stacksize and locking fixes for loadgen
369 Description: Quiet CERROR("dirty %d > system dirty_max %d\n"
370 Details : The atomic_read() allowing the atomic_inc() are not covered
371 by a lock. Thus they may safely race and trip this CERROR()
372 unless we add in a small fudge factor (+1).
374 Severity : enhancement
376 Description: shrink_slab: nr=-9223362083340912175
377 Details : fix spurious message from shrink_slab reporing negative nr
381 Description: Quiet bogus previously committed transno error
382 Details : suppress the "server went back in time" error message which
383 is always printed even in the common case after a client eviction
385 Severity : enhancement
387 Description: Parallel statfs() calls result in client eviction
388 Details : cache statfs data for 1s.
392 Description: parallel-scale test_compilebench: @@@@@@ FAIL: compilebench
394 Details : fix serveral issues in pinger code causing clients not to ping
395 servers for too long, resulting in evictions.
399 Description: e2fsck should warn when MMP update interval is extended
400 Details : print mmp_check_interval and make it possible to abort mount
401 operation in case it takes too long.
405 Description: mdsrate-create-large.sh, BUG: soft lockup - CPU#0 stuck for 10s!
406 Details : fix bug in the RHEL5's jbd2 callback patch.
410 Description: drop number of active requests when queued for recovery
411 Details : Now that we take a reference on the original request instead of
412 making a copy of it for recovery. We need to drop the number of
413 active requests or the queued requests will prevent all request
414 processing when they exceed (srv->srv_threads_running - 1).
416 Severity : enhancement
418 Description: refuse to invalidate operational quota files when they are in use
419 Details : an attempt to invalidate operational quota files on the quota
420 master is not actually permitted by VFS (returning -EPERM), but we
421 should not depend on that and should return the error earlier.
425 Description: Applications stuck in jbd2_log_wait_commit during exit
426 Details : fix deadlock between kjournald2 trying to acquire the page lock
427 owned by an ost_io thread waiting for journal commit.
429 -------------------------------------------------------------------------------
431 2009-10-16 Sun Microsystems, Inc.
433 * Support for kernels:
434 2.6.16.60-0.42.4 (SLES 10),
435 2.6.27.29-0.1 (SLES11, i686 & x84_64 only),
436 2.6.18-128.7.1.el5 (RHEL 5),
437 2.6.18-128.7.1.el5 (OEL 5),
438 * Client support for unpatched kernels:
439 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
440 2.6.16 - 2.6.27 vanilla (kernel.org)
441 * Recommended e2fsprogs version: 1.41.6.sun1
442 * ext4 support for RHEL5 is experimental and thus should not be
445 Severity : enhancement
447 Description: Add OEL5 support.
449 Severity : enhancement
451 Description: Update kernel to SLES11 2.6.27.29-0.1.
455 Description: File checksum failures with OST read cache on
456 Details : Disable page poisoning when the bulk transfer has to be aborted
457 because the client got evicted.
461 Description: Don't allow make backward step on assiging osc next id.
462 Details : race between allocation next id and ll_sync thread can be cause
463 of set wrong osc next id and can be kill valid ost objects.
465 Severity : enhancement
467 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
469 Severity : enhancement
471 Description: Update kernel to SLES10 SP2 2.6.16.60-0.42.4.
475 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
479 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
483 Description: Error handling in osc_statfs_interpret() has been improved.
484 Details : Check in osc_statfs_interpret() for EBADR.
488 Description: Do not send statfs() requests to OSTs disabled by administrator.
489 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
493 Description: Do not update ctime for the deleted inode.
494 Details : Check in mds_reint_unlink() before calling fsfilt_setattr().
498 Description: Increase of the size of the LDLM resource hash.
499 Details : Bump up RES_HASH_BITS=12.
503 Description: correctly send lsm on open replay
504 Details : MDS is trust to LSM size on replay open, but client can set wrong size
509 Description: Deadlock between filter_destroy() and filter_commitrw_write().
510 Details : filter_destroy() does not hold the DLM lock over the whole
511 operation. If the DLM lock is dropped, filter_commitrw() can go
512 through, causing the deadlock between page lock and i_mutex.
513 The i_alloc_sem should also be hold in filter_destroy() while
518 Description: truncate starts GFP_FS allocation under transaction causing deadlock
519 Details : ldiskfs_truncate calls grab_cache_page which may start page
520 allocation under an open transaction. This may lead to
521 calling prune_icache with consequent lustre reentrance.
524 Frequency : only when down/upgrading the MDS to 1.6/1.8 while 1.8 clients are
525 still up and when the OST pool feature is used
527 Description: interop testing got LBUG when run dd with OST pool
528 :LustreError: 30032:0:(llite_lib.c:1913:ll_replace_lsm()) LBUG
529 Details : down/upgrading the MDS to a version that doesn't/does support OST
530 pool can cause clients to crash because the lsm has changed
535 Description: missing tree_status on 1.8.1 RPM build
536 Details : make rpms failed due because the tree_status file is missing.
540 Description: continuing LustreError "mds adjust qunit failed!"
541 Details : don't print message on the console when ->adjust_qunit fails.
545 Description: don't increase ldlm timeout if previous client was evicted
546 Details : if a client doesn't respond to a blocking callback within the
547 adaptive ldlm enqueue timeout, don't adjust the adaptive estimate
548 when the lock is next granted.
552 Description: ost is being unmounted w/o all writes to last_rcvd landing on disk.
553 affects recovery negatively.
554 Details : make sure all exports have been properly destroyed by the zombie
555 thread processed before stopping the target.
559 Description: Performance degradation with O_DIRECT between 1.6 & 1.8.1 b190
560 Details : disable write barrier for ext4/SLES11.
564 Description: Kernel panic - not syncing: Out of memory and no killable
565 processes... on OSS when iozone
566 Details : fix memory leak in the journal checksum patch.
570 Description: group quota "too many blocks" OSS crashes
571 Details : we should keep the same uid/gid for lquota_chkquota() and
572 lquota_pending_commit()
576 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
577 Details : don't LBUG on release quota error. Just a workaround until the
578 problem is understood.
580 ------------------------------------------------------------------------------
581 2009-07-31 Sun Microsystems, Inc.
583 * Support for kernels:
584 2.6.16.60-0.39.3 (SLES 10),
585 2.6.27.23-0.1 (SLES11, i686 & x84_64 only),
586 2.6.18-128.1.14.el5 (RHEL 5),
587 * Client support for unpatched kernels:
588 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
589 2.6.16 - 2.6.27 vanilla (kernel.org)
590 * Recommended e2fsprogs version: 1.41.6.sun1
591 * File join has been disabled in this release, refer to Bugzilla 16929.
592 * NFS export disabled when stack size < 8192. Since the NFSv4 export
593 of Lustre filesystem with 4K stack may cause a stack overflow. For
594 more information, please refer to bugzilla 17630.
595 * ext4 support for RHEL5 is experimental and thus should not be
598 Severity : enhancement
600 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
603 Frequency : with 1.8 server and 1.6 clients
605 Description: correctly shrink reply for avoid send too big message to client.
606 Details : 1.8 mds is allocate to big buffer to LOV EA data and this produce
607 some problems with sending this reply to 1.6 client.
611 Description: Repeated atomic allocation failures.
612 Details : Use GFP_HIGHUSER | __GFP_NOMEMALLOC flags for memory allocations
613 to generate memory pressure and allow reclaiming of inactive pages.
614 At the same time, do not allow to exhaust emergency pools.
615 For local clients the use of GFP_NOFS will be introduced in 1.8.2
617 Severity : enhancement
618 Bugzilla : 19846, 18289
619 Description: Update kernel to RHEL5 2.6.18-128.1.14.el5.
621 Severity : enhancement
622 Bugzilla : 19625, 16893, 18668, 19848
623 Description: Add support for SLES11 2.6.27.23-0.1.
625 Severity : enhancement
627 Description: Update client support to vanila kernels up to 2.6.27.
629 Severity : enhancement
631 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
633 Severity : enhancement
635 Description: Compile with -Werror by default for i686 and x86_64.
639 Description: resolve race between obd_disconnect and class_disconnect_exports
640 Details : if obd_disconnect will be called to already disconnected export he
641 forget release one reference and osc module can't unloaded.
643 Severity : enhancement
645 Description: move AT tunable parameters for more consistent usage
646 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
650 Descriptoin: correctly skip time estimate if in recovery
651 Details : rq_send_state insn't bitmask so using bitwise ops is forbid.
655 Descriptoin: OSS DeadLock
656 Details : Use trylock to prevent deadlock when shrink icache.
658 Severity : enhancement
660 Description: Allow tuning service thread via /proc
661 Details : For each service a new
662 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
663 created that can be used to set min/max thread counts, and get the
664 current number of running threads.
666 Severity : enhancement
668 Description: Add state history info file, enhance import info file
669 Details : Track import connection state changes in a new osc/mdc proc file;
670 add overview-type data to the osc/mdc import proc file.
674 Description: Reduce small size read RPC
675 Details : Set read-ahead limite for every file and only do read-ahead when
676 available read-ahead pages are bigger than 1M to avoid small size
681 Description: free_entry erroneously used groups_free instead of put_group_info
683 Severity : enhancement
685 Description: Make read-ahead stripe size aligned.
687 Severity : enhancement
689 Description: MDS create should not wait for statfs RPC while holding DLM lock.
692 Frequency : rare, connect and disconnect target at same time
694 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
695 Details : don't call obd_disconnect under lov_lock. this long time
696 operation and can block ptlrpcd which answer to connect request.
699 Frequency : start MDS on uncleanly shutdowned MDS device
701 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
702 Details : stay in waiting mds<>ost recovery finished produce random bugs
703 due race between two ll_sync thread for one lov target. send
704 ACTIVATE event only if connect realy finished and import have
708 Frequency : start MDS on uncleanly shutdowned MDS device
710 Descriptoin: aborting recovery hang on MDS
711 Details : don't throttle destroy RPCs for the MDT.
715 Description: Slow reads beyond 8Tb offsets.
716 Details : Page index integer overflow in ll_read_ahead_page
720 Description: Soft lockup on OSS after MDS failover
721 Details : MSG_CONNECT_INITIAL is not set on the initial MDS->OST connect.
722 As a conseqence, the patch from bug 18224 is not operational
723 and the MDS export cannot be reused on the OSTs until it gets
727 Frequency : rare, only if using MMP with Linux RAID
729 Description: MMP doesn't work with Linux RAID
730 Details : While using HA for Lustre servers with Linux RAID, it is possible
731 that MMP will not detect multiple mounts. To make this work we
732 need to unplug the device queue in RAID when the MMP block is being
733 written. Also while reading the MMP block, we should read it from
734 disk and not the cached one.
737 Frequency : rare, during recovery
739 Description: Assertion failure in ldlm_lock_put
740 Details : Do not put cancelled locks into replay list, hold references on
745 Description: 1.6.5 mdsrate performance is slower than 1.4.11/12 (MDS is not cpu bound!)
746 Details : create_count always drops to the min value (=32) because grow_count
747 is being changed before the precreate RPC completes.
750 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
753 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
754 Details : add proper locking for creation of jbd_4k slab cache
758 Description: MMP check in ext3_remount() fails without displaying any error
759 Details : When multiple mount protection fails during remount, proper error
764 Description: Rare Client crash on resend if the file was deleted.
765 Details : When file is opened, but open reply is lost and file is
766 subsequently deleted before resend, resend processing logic
767 breaks trying to open the file again, should not try to open.
771 Description: add check for >8TB ldiskfs filesystems
772 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
773 Don't allow >8TB ldiskfs filesystems to be mounted without
774 force_over_8tb mount option
778 Description: Client locked up when running multiple instances of an app. on
779 multiple mount points
780 Details : ll_shrink_cache() can sleep while holding the ll_sb_lock.
781 Convert ll_sb_lock to a read/write semaphore to fix the problem.
785 Description: Cannot acces an NFS-mounted Lustre filesystem
786 Details : An NFS client cannot access the Lustre filesystem NFS-mounted
787 from a Lustre-client exporting the Lustre filesystem via NFS.
791 Description: panic in ll_statahead_thread
792 Details : grab dentry reference in parent process.
795 -------------------------------------------------------------------------------
797 tbd Sun Microsystems, Inc.
799 * Support for kernels:
800 2.6.16.60-0.31 (SLES 10),
801 2.6.18-128.1.6.el5 (RHEL 5),
802 2.6.22.14 vanilla (kernel.org)
803 * Client support for unpatched kernels:
804 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
805 2.6.16 - 2.6.22 vanilla (kernel.org)
806 * Recommended e2fsprogs version: 1.40.11-sun1
807 * File join has been disabled, refer to Bugzilla 16929.
808 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
809 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
810 Lustre filesystem with 4K stack may cause a stack overflow. For more
811 information, please refer to bugzilla 17630.
813 Severity : enhancement
815 Description: Update to RHEL5.3 kernel-2.6.18-128.1.6.el5.
817 Severity : enhancement
819 Description: Update OFED release to 1.4.1 RC4
821 Severity : major, only with big OST
823 Description: Very poor metadata performance on Infiniband lustre configuration
824 Details : OST object precreation becomes very slow on big OSTs. This is due
825 to the ialloc patch spending too much time scanning groups.
828 Frequency : during recovery
830 Description: don't mix llog inodes with normal.
831 Details : allocate inodes for log in last inode group
836 Description: fix lqs' reference which won't be put in some situations
837 Details : This patch fixes:
838 1. In quota_check_common(), this function will check quota
839 for user and group, but only send one return via "pending".
840 In most cases, the pendings should be same. But that is not
842 2. If quotaoff runs between lquota_chkquota() and
843 lquota_pending_commit(), the same thing will happen too.
844 That is why it comes:
845 - if (!ll_sb_any_quota_active(qctxt->lqc_sb))
848 -------------------------------------------------------------------------------
850 2008-12-31 Sun Microsystems, Inc.
852 * Support for kernels:
853 2.6.5-7.314 (SLES 9),
854 2.6.9-67.0.22.EL (RHEL 4),
855 2.6.16.60-0.31 (SLES 10),
856 2.6.18-92.1.17.el5 (RHEL 5),
857 2.6.22.14 vanilla (kernel.org)
858 * Client support for unpatched kernels:
859 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
860 2.6.16 - 2.6.22 vanilla (kernel.org)
861 * Client support for unpatched kernels:
862 we do not recommend using patchless RHEL4 clients with kernels
863 prior to 2.6.9-55EL (RHEL4U5).
864 * Recommended e2fsprogs version: 1.40.11-sun1
865 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
866 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
867 removed cwd "./" (refer to Bugzilla 14399).
868 * A new quota file format has been introduced in 1.6.5.
869 The format conversion from prior releases is handled transparently,
870 but releases older than 1.4.12/1.6.5 don't understand this new
871 format. The automatic format conversion can be avoided by running
872 the following command on the MDS:
873 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
874 For more information, please refer to bugzilla 13904.
875 * A new quota file format was introduced in 1.6.6/1.8.0 (kernels 2.6.16+).
876 The format conversion from prior releases is handled transparently,
877 but releases older than 1.6.6/1.8.0 don't understand this new
878 format. The automatic format conversion can be avoided by running
879 the following commands on the MDS and OSS servers (for
880 pre 1.4.12-1.6.5 quota files):
881 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV',
882 'tunefs.lustre --param="ost.quota_type=ug1" $MDTDEV'
883 or (for 1.4.12/1.6.5 quota files)
884 'tunefs.lustre --param="mdt.quota_type=ug2" $MDTDEV',
885 'tunefs.lustre --param="ost.quota_type=ug2" $MDTDEV'
886 For more information, please refer to bugzilla 13904.
887 * Output of lfs quota has been made less detailed by default,
888 old (verbose) output can be obtained by using -v option.
889 * File join has been disabled in this release, refer to Bugzilla 16929.
890 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
891 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
892 Lustre filesystem with 4K stack may cause a stack overflow. For more
893 information, please refer to bugzilla 17630.
895 Severity : enhancement
897 Description: Caching OSS
898 Details : introduce data caching on the OSS. The OSS now relies on the linux
899 kernel page cache to keep recently accessed data in memory.
900 It is worth noting that all write requests are still flushed
901 synchronously as in lustre 1.6.
903 Severity : enhancement
905 Description: Version based recovery
906 Details : introduce finer grained recovery able to detect transaction
907 dependencies and can deal with transaction gaps caused by clients
908 failing at the same time as the server.
910 Severity : enhancement
912 Description: Enable adaptive timeouts by default
913 Details : The Lustre timeout value in /proc/sys/lustre/timeout is now
914 managed dynamically based on server load and should not need
915 to be tuned manually based on cluster size. This allows Lustre
916 to work under a wider variety of system sizes and loads, without
917 unnecessarily causing lengthy recovery times.
919 Severity : enhancement
921 Description: Add OST Pools support
922 Details : File striping can now be set to use an arbitrary pool of OSTs
924 Severity : enhancement
926 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
927 Details : allow skip disconnected ost for send statfs request and hide error
931 Frequency : rare, on llog test 6
933 Descriptoin: don't allow connect to already connected import
934 Details : allowing connect to already connected import is hide connecting problem.
937 Frequency : rare, on failed llog setup
939 Descriptoin: don't leak obd reference on failed llog setup
940 Details : for failed llog setup - mgc forget call class_destroy_import for
941 client import, move destroy import to more generic place.
946 Descriptoin: allow kill process which wait statahead result
947 Details : for some reasons 'ls' can stick in waiting result from statahead,
948 in this case need way for kill this process.
951 Frequency : rare, at shutdown
953 Descriptoin: panic at umount
954 Details : llap_shrinker can be raced with killing super block from list and
955 this produce panic with access to already freeded pointer
960 Descriptoin: don't lose wakeup for imp_recovery_waitq
961 Details : recover_import_no_retry or invalidate_import and import_close can
962 both sleep on imp_recovery_waitq, but we was send only one wakeup
968 Descriptoin: panic in mds_open
969 Details : don't confuse mds_finish_transno() with PTR_ERR(-ENOENT)
974 Descriptoin: stuck in cache_remove_extent() or panic with accessing to already
976 Details : release lock refernce only after add page to pages list.
979 Frequency : always with long access acl
981 Descriptoin: mds can't pack reply with long acl.
982 Details : mds don't control size of acl but they limited by reint/getattr
987 Frequency : on remount
989 Description: external journal device not working after the remount
990 Details : clear dev_rdonly flag for external journal devices in
996 Description: shutdown vs evict race
997 Details : client_disconnect_export vs connect request race.
998 if client will evicted at this time - we start invalidate
999 thread without referece to import and import can be freed
1005 Description: shrink LOV EAs before replying
1006 Details : correctly adjust LOV EA buffer for reply.
1011 Description: don't skip ost target if they assigned to file
1012 Details : Drop slow OSCs if we can, but not for requested start idx.
1013 This means "if OSC is slow and it is not the requested
1014 start OST, then it can be skipped, otherwise skip it only
1015 if it is inactive/recovering/out-of-space.
1017 Severity : enhancement
1019 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
1022 Frequency : rare, need acl's on inode.
1024 Description: client can't handle ost additional correctly
1025 Details : if ost was added after client connected to mds client can have
1026 hit lnet_try_match_md ... to big messages to wide striped files.
1027 in this case need teach client to handle config events about add
1028 lov target and update client max ea size at that event.
1030 Severity : enhancement
1032 Description: Update to sles9 kernel-2.6.5-7.314.
1034 Severity : enhancement
1036 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
1039 Frequency : Create a symlink file with a very long name
1041 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
1042 Details : If there is no extra space in the request for early cancels,
1043 ldlm_req_handles_avail() returns 0 instead of a negative value.
1048 Description: mds is deadlocked
1049 Details : in rare cases, inode in catalog can have i_no less than have parent
1050 i_no, this produce wrong order for locking during open, and
1051 parallel unlink can be lock open. this need teach mds_open to grab
1052 locks in resource id order, not at parent -> child order.
1054 Severity : enhancement
1056 Description: Add /proc entry for import status
1057 Details : The mdc, osc, and mgc import directories now have
1058 an import directory that contains useful import data for debugging
1059 connection problems.
1061 Severity : enhancement
1063 Description: Re-disable certain /proc logging
1064 Details : Enable and disable client's offset_stats, extents_stats and
1065 extents_stats_per_process stats logging on the fly.
1068 Frequency : Only on FC kernels 2.6.22+
1070 Description: oops in statahead
1071 Details : Do not drop reference count for the dentry from VFS when lookup,
1072 VFS will do that by itself.
1074 Severity : enhancement
1076 Description: Generic /proc file permissions
1077 Details : Set /Proc file permissions in a more generic way to enable non-
1078 root users operate on some /proc files.
1082 Description: Hitting mdc_commit_close() ASSERTION
1083 Details : Properly handle request reference release in
1084 ll_release_openhandle().
1088 Frequency : only patchless client
1089 Description: add workaround for race between add/remove dentry from hash
1091 Severity : enhancement
1093 Description: Allow OST glimpses to return PW locks
1097 Description: LBUG when llog conf file is full
1098 Details : When llog bitmap is full, ENOSPC should be returned for plain
1103 Description: Prevent import from entering FULL state when server in recovery
1107 Description: service mount cannot take device name with ":"
1108 Details : Only when device name contains ":/" will mount treat it as
1114 Description: replace ptlrpcd with the statahead thread to interpret the async
1115 statahead RPC callback
1119 Frequency : on recovery
1120 Description: I/O failures after umount during fail back
1121 Details : if client reconnected to restarted server we need join to recovery
1122 instead of find server handler is changed and process self eviction
1123 with cancel all locks.
1125 Severity : enhancement
1127 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
1131 Description: Kernel BUG tries to release flock
1132 Details : Lustre does not destroy flock lock before last reference goes
1133 away. So always drop flock locks when client is evicted and
1134 perform unlock regardless of successfulness of speaking to MDS.
1136 Severity : enhancement
1138 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
1140 Severity : enhancement
1142 Description: Upcall on Lustre log has been dumped
1143 Details : Allow for a user mode script to be called once a Lustre log has
1144 been dumped. It passes the filename of the dumped log to the
1145 script, the location of the script can be specified via
1146 /proc/sys/lnet/debug_log_upcall.
1151 Description: avoid messages about idr_remove called for id that is not allocated
1152 Details : Move assigment s_dev for clustered nfs to end of initialization,
1153 for avoid problem with error handling.
1158 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
1159 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
1160 CONN_USED_HASH and this prodice warning when put connection again
1166 Description: avoid ASSERTION(client_stat->nid_exp_ref_count == 0) failed
1167 Details : release reference to stats when client disconnected, not
1168 when export destroyed for avoid races when client destroyed
1169 after main ost export.
1173 Description: more cleanup in mds_lov
1174 Details : add workaround for get valid ost count for avoid warnings about
1175 drop too big messages, not init llog cat under semphore which
1176 can be blocked on reconnect and break normal replay, fix access
1179 Severity : enhancement
1181 Description: Export bytes_read/bytes_write count on OSC/OST.
1185 Description: Early reply size mismatch, MGC loses connection
1186 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
1187 the connect flags are properly negotiated.
1191 Description: Properly propagate oinfo flags from lov to osc for statfs
1192 Details : restore missing copy oi_flags to lov requests.
1196 Description: exports in /proc are broken
1197 Details : recreate /proc entries for clients when they reconnect.
1199 Severity : enhancement
1201 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
1202 : l_getgroups(8), lst(8), routerstat(8)
1203 Details : included man pages for llobdstat(8), llstat(8),
1204 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
1206 Severity : enhancement
1208 Description: Implement lustre ll_show_options method.
1210 Severity : enhancement
1212 Description: Update to SLES9 kernel-2.6.5-7.312.
1214 Severity : enhancement
1216 Description: Update to RHEL4 kernel-2.6.9-67.0.22.EL.
1220 Description: exports in /proc are broken
1221 Details : recreate /proc entries for clients when they reconnect.
1225 Description: don't fail open with -ERANGE
1226 Details : if client connected until mds will be know about real ost count
1227 get LOV EA can be fail because mds not allocate enougth buffer
1232 Description: Resolve device initialization race
1233 Details : Prevent proc handler from accessing devices added to the
1234 obd_devs array but yet be intialized.
1236 Severity : enhancement
1238 Description: configure's --enable-quota should check the
1239 : kernel .config for CONFIG_QUOTA
1240 Details : configure is terminated if --enable-quota is passed but
1241 : no quota support is in kernel
1243 Severity : enhancement
1245 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
1247 Severity : enhancement
1249 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
1253 Frequency : rare, on PPC clients
1254 Description: don't swab ost objects in response about directory, because
1256 Details : bug similar bug 14856, but in different function.
1258 Severity : enhancement
1260 Description: lfs quota tool enhancement
1261 Details : added units specifiers support for setquota, default to
1262 current uid/gid for quota report, short quota stats by
1263 default, nonpositional parameters for setquota, added
1264 llapi_quotactl manual page.
1266 Severity : enhancement
1268 Description: *optional* service tags registration
1269 Details : if the "service tags" package is installed on a Lustre node
1270 When the filesystem is mounted, a local-node service tag will
1271 be created. See http://inventory.sun.com/ for more information
1272 about the Service Tags asset management system.
1276 Description: Client runs out of low memory
1277 Details : Consider only lowmem when counting initial number of llap pages
1280 Frequency : occasional
1282 Description: add refcount for osc callbacks, so avoid panic on shutdown
1284 Severity : enhancement
1286 Description: Update to RHEL4 kernel-2.6.9-67.0.20.
1289 Frequency : testing only
1291 Description: sanity test 65a fails if stripecount of -1 is set
1292 Details : handle -1 striping on filesystem in ll_dirstripe_verify
1295 Frequency : only in unusual configurations
1297 Description: Kernel panic with find ost index.
1298 Details : lov_obd have panic if some OST's have sparse indexes.
1300 Severity : enhancement
1302 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
1305 Frequency : rarely, if filesystem is mounted with -o flock
1307 Description: do not process already freed flock
1308 Details : flock can possibly be freed by another thread before it reaches
1309 to ldlm_flock_completion_ast.
1312 Frequency : rarely, if filesystem is mounted with -o flock
1314 Description: LBUG during stress test
1315 Details : Need properly lock accesses the flock deadlock detection list.
1318 Frequency : rarely, if binaries are being run from Lustre
1320 Description: oops in page fault handler
1321 Details : kernel page fault handler can return two special 'pages' in
1322 error case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
1325 Frequency : rarely, during shutdown
1327 Description: timeout with invalidate import.
1328 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
1329 should be handled by ptlrpcd. This produce long age waiting and
1330 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
1336 Description: ASSERTION(CheckWriteback(page,cmd)) failed
1337 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
1341 Frequency : only with broken builds/installations
1343 Description: no LBUG if lquota.ko and fsfilt_ldiskfs.ko are different versions
1344 Details : just return an error to a user, put a console error message
1346 Severity : enhancement
1348 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
1350 Severity : enhancement
1352 Description: Update to RHEL4 kernel-2.6.9-67.0.15.
1354 Severity : enhancement
1356 Description: enable MGS and MDT services start separately
1357 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
1358 a co-located MGS without starting the MGS, which is a complement
1359 to 'nosvc' mount option.
1362 Frequency : always, on big-endian systems
1364 Description: cleanup in ptlrpc code, related to PPC platform
1365 Details : store magic in native order avoid panic's in recovery on PPC
1366 node and forbid from this error in future. Also fix posibily
1367 of twice swab data. Fix get lov striping to userpace.
1370 Frequency : rarely, if replay get lost on server
1372 Description: server incorrectly drop resent replays lead to recovery failure.
1373 Details : do not drop replay according to msg flags, instead we check the
1374 per-export recovery request queue for duplication of transno.
1377 Frequency : after recovery
1379 Description: precreate to many object's after del orphan.
1380 Details : del orphan st in oscc last_id == next_id and this triger growing
1381 count of precreated objects. Set flag LOW to skip increase count
1382 of precreated objects.
1385 Frequency : rare, on clear nid stats
1387 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
1388 Details : when clean nid stats sometimes try destroy live entry,
1389 and this produce panic in free.
1392 Frequency : occasionally since 1.6.4
1394 Description: Stack overflow during MDS log replay
1395 Details : ease stack pressure by using a thread dealing llog_process.
1398 Frequency : very rare
1400 Description: MDT cannot be unmounted, reporting "Mount still busy"
1401 Details : Mountpoint references were being leaked during open reply
1402 reconstruction after an MDS restart. Drop mountpoint reference
1403 in reconstruct_open() and free dentry reference also.
1408 Description: wait until IO finished before start new when do lock cancel.
1409 Details : VM protocol want old IO finished before start new, in this case
1410 need wait until PG_writeback is cleared until check dirty flag
1411 and call writepages in lock cancel callback.
1416 Description: mds_mfd_close() ASSERTION(rc == 0)
1417 Details : In mds_mfd_close(), we need protect inode's writecount change
1418 within its orphan write semaphore to prevent possible races.
1421 Frequency : rare, on shutdown ost
1423 Description: don't hit live lock with umount ost.
1424 Details : shrink_dcache_parent can be in long loop with destroy dentries,
1425 use shrink_dcache_sb instead.
1428 Frequency : only when echo_client is used
1430 Description: don't panic with use echo_client
1431 Details : echo client pass NULL as client nid pointer and this produce NULL
1432 pointer dereference.
1435 Frequency : Always on 32-bit PowerPC systems
1437 Description: fix build on PPC32
1438 Details : compile code with -m64 flag produce wrong object file for PPC32.
1443 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1444 Details : In reconstruct_* functions, LASSERTs on both the data supplied
1445 by a client, and the data on disk are dangerous and incorrect.
1446 Change them with client eviction.
1448 Severity : enhancement
1450 Description: skiplist implementation simplification
1451 Details : skiplists are used to group compatible locks on granted list
1452 that was implemented as tracking first and last lock of each lock
1453 group the patch changes that to using doubly linked lists
1457 Description: delete compatibility for 32bit qdata
1458 Details : as planned, when lustre is beyond b1_8, lquota won't support 32bit
1459 qunit. That means servers of b1_4 and servers of b1_8 can't be
1460 used together if users want to use quota.
1463 Frequency : only with administrator action
1465 Description: mount failure if config log has invalid conf_param setting
1466 Details : If administrator specified an incorrect configuration parameter
1467 with "lctl conf_param" this would cause an error during future
1468 client mounts. Instead, ignore the bad configuration parameter.
1471 Frequency : blocks per group < blocksize*8 and uninit_groups is enabled
1473 Description: ldiskfs error: XXX blocks in bitmap, YYY in gd
1474 Details : If blocks per group is less than blocksize*8, set rest of the
1478 Frequency : Application do stride read on lustre
1480 Description: The read performance will drop a lot if the application does
1482 Details : Because the stride_start_offset are missing in stride read-ahead,
1483 it will cause clients read a lot of unused pages in read-ahead,
1484 then the read-performance drops.
1488 Description: more ldlm soft lockups
1489 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1490 starve other threads from the resource lock for a long time in
1491 case of long waiting queue, so change the debug level from
1492 D_OTHER to the less frequently used D_INFO.
1494 Severity : enhancement
1496 Description: add -gid, -group, -uid, -user options to lfs find
1498 Severity : enhancement
1500 Description: ll_recover_lost_found_objs - recover objects in lost+found
1501 Details : OST corruption and subsequent e2fsck can leave objects in the
1502 lost+found directory. Using the "ll_recover_lost_found_objs"
1503 tool, these objects can be retrieved and data can be salvaged
1504 by using the object ID saved in the fid EA on each object.
1509 Description: this bug _only_ happens when inode quota limitation is very low
1510 (less than 12), so that inode quota unit is 1 at initialization.
1511 Details : if remaining quota equates 1, it is a sign to demonstate that quota
1512 is effective now. So least quota qunit should be 2.
1516 Description: Hung threads in invalidate_inode_pages2_range
1517 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1518 one is completed. As a result, some RPCs are stuck in the queue
1523 Description: Procfs and llog threads access destoryed import sometimes.
1524 Details : Sync the import destoryed process with procfs and llog threads by
1525 the import refcount and semaphore.
1529 Description: mds fails to respond, threads stuck in ldlm_completion_ast
1530 Details : Sort source/child resource pair after updating child resource.
1535 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1536 Details : If insertion of an extent fails, then discard the inode
1537 preallocation and free data blocks else it can lead to duplicate
1542 Description: don't always update ctime in ext3_xattr_set_handle()
1543 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle()
1544 In some cases the ctime should not be updated, for example for
1545 2.0->1.8 compatibility it is necessary to delete an xattr and it
1546 should not update the ctime.
1550 Description: add quota statistics
1551 Details : 1. sort out quota proc entries and proc code.
1552 2. add quota statistics
1557 Description: quotas are not honored with O_DIRECT
1558 Details : all writes with the flag O_DIRECT will use grants which leads to
1559 this problem. Now using OBD_BRW_SYNC to guard this.
1563 Bugzilla : 15713/16362
1564 Description: Assertion in iopen_connect_dentry in 1.6.3
1565 Details : looking up an inode via iopen with the wrong generation number can
1566 populate the dcache with a disconneced dentry while the inode
1567 number is in the process of being reallocated. This causes an
1568 assertion failure in iopen since the inode's dentry list contains
1569 both a connected and disconnected dentry.
1573 Description: assertion failure in ldlm_handle2lock()
1574 Details : fix a race between class_handle_unhash() and class_handle2object()
1575 introduced in lustre 1.6.5 by bug 13622.
1577 Severity : enhancement
1579 Description: superblock lock contention with many SMP cores on one client
1580 Details : several client filesystem locks were highly contended on SMP
1581 NUMA systems with 8 or more cores. Per-CPU datastructures
1582 and more efficient locking implemented to reduce contention.
1587 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1588 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1593 Description: oops in sd_iostats_seq_show()
1594 Details : unloading/reloading the scsi low level driver triggers a kernel
1595 bug when trying to access the sd iostat file.
1600 Description: Kernel panics during QLogic driver reload
1601 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1602 patch, causing memory corruption.
1607 Description: journal_dev option does not work in b1_6
1608 Details : pass mount option during pre-mount.
1610 Severity : enhancement
1612 Description: Add a FIEMAP(FIle Extent MAP) ioctl for ldiskfs
1613 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1614 extent information of a file. It can be used to map logical blocks
1615 in a file to physical blocks in the block device.
1618 Frequency : only with adaptive timeout enabled
1620 Description: DEBUG_REQ() bad paging request
1621 Details : ptlrpc_at_recv_early_reply() should not modify req->rq_repmsg
1622 because it can be accessed by reply_in_callback() without the
1626 Frequency : only on Cray X2
1628 Description: X2 build failures
1629 Details : fix build failures on Cray X2.
1633 Description: xid & resent requests
1634 Details : Initialize RPC XID from clock at startup (randomly if clock is
1639 Description: quota recovery deadlock during mds failover
1640 Details : This patch includes att18982, att18236, att18237 in bz14840.
1642 1. fix osts hang when mds does failover with quotaon
1643 2. prevent watchdog storm when osts threads wait for the
1648 Description: kernel panic on racer
1649 Details : Do not access dchild->d_inode when IS_ERR(dchild) is true.
1651 Severity : enhancement
1653 Description: Add lustre_start utility to start or stop multiple Lustre servers
1658 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1659 Details : In case of memory pressure, list_del() can be called twice on
1660 req->rq_history_list, causing a kernel oops.
1664 Description: (ptllnd_peer.c:557:kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1665 Details : fix stack overflow in the distributed lock manager by defering
1666 export eviction after a failed ast to the elt thread instead of
1667 handling it in the dlm interpret routine.
1669 Severity : enhancement
1671 Description: More exported tunables for mballoc
1672 Details : Add support for tunable preallocation window and new tunables for
1673 large/small requests
1677 Description: Detect corruption of block bitmap and checking for preallocations
1678 Details : Checks validity of on-disk block bitmap. Also it does better
1679 checking of number of applied preallocations. When corruption is
1680 found, it turns filesystem readonly to prevent further corruptions.
1684 Frequency : only for big-endian servers
1685 Description: Check for big-endian system while mounting fs with extents feature
1686 Details : Mounting a filesystem with extents feature will fail on big-endian
1687 systems since ext3-based ldiskfs is not supported on big-endian
1688 systems. Can be over-ridden with "bigendian_extents" mount option.
1692 Description: Excessive recovery window
1693 Details : With AT enabled, the recovery window can be excessively long (6000+
1694 seconds). To address this problem, we no longer use
1695 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
1696 timeout no longer depends on the service time, it is set to
1697 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
1698 time via pb_service_time.
1701 Descriptoin: Don't sync journal after every i/o
1702 Details : Implement write RPC replay to allow server replies for write RPCs
1703 before data is on disk. However, this feature is disabled by
1704 default since some issues leading to data corruptions have been
1705 found during recovery (e.g. bug 19128). This feature can be enabled
1706 by running the following command on the OSSs:
1707 lctl set_param obdfilter.*.sync_journal=0
1711 Description: Watchdog triggered on MDS failover
1712 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
1713 the OSTs know that the MDS "UUID" can be reused for the same export
1714 from a different NID, so we do not need to wait for the export to
1719 Descriptoin: Lustre detected file system corruption with inode out of bounds
1720 Details : don't update i_size on MDS_CLOSE for directories. This causes
1721 directory corruptions on the MDT.
1723 -------------------------------------------------------------------------------
1725 2008-05-26 Sun Microsystems, Inc.
1727 * Support for kernels:
1728 2.6.5-7.311 (SLES 9),
1729 2.6.9-67.0.7.EL (RHEL 4),
1730 2.6.16.54-0.2.5 (SLES 10),
1731 2.6.18-53.1.14.el5 (RHEL 5),
1732 2.6.22.14 vanilla (kernel.org)
1733 * Client support for unpatched kernels:
1734 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1735 2.6.16 - 2.6.22 vanilla (kernel.org)
1736 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
1737 we do not recommend using patchless RHEL4 clients with kernels
1738 prior to 2.6.9-55EL (RHEL4U5).
1739 * Recommended e2fsprogs version: 1.40.7-sun1
1740 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1741 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
1742 removed cwd "./" (refer to Bugzilla 14399).
1743 * A new quota file format has been introduced in 1.6.5.
1744 The format conversion from prior releases is handled transparently,
1745 but releases older than 1.4.12/1.6.5 will not understand this new
1746 format. The automatic format conversion can be avoided by running
1747 the following command on the MDS before upgrading:
1748 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
1749 For more information, please refer to bugzilla 13904.
1753 Description: quota performance fix
1754 Details : quota data is written in journalled mode instead of ordered to
1755 increase performance
1759 Description: lfs support for human-readable quota grace time strings
1760 Details : lfs setquota -t and lfs quota -t represent quota grace times
1761 in "XXwXXdXXhXXmXXs" format instead of large values in seconds
1764 Frequency : always with o2ib 1.3 and sles10
1766 Description: fix build with SLES10 and o2ib v3.
1767 Details : sles10 uses diffrent name for Module.symver file but configure
1768 assume this file has same name on RHEL/SLES/vanila kernels.
1771 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1773 Description: MDS may lose file striping (and hence file data) in some cases
1774 Details : If there are additional extended attributes stored on the MDS,
1775 in particular ACLs, SELinux, or user attributes (if user_xattr
1776 is specified for the client mount options) then there is a risk
1777 of attribute loss. Additionally, the Lustre file striping
1778 needs to be larger than default (e.g. striped over all OSTs),
1779 and an additional attribute must be stored initially in the
1780 inode and then increase in size enough to be moved to the
1781 external attribute block (e.g. ACL growing in size) for file
1784 Severity : enhancement
1786 Description: add message levels for liblustreapi
1791 Description: MDT cannot be unmounted, reporting "Mount still busy"
1792 Details : Mountpoint references were being leaked during open reply
1793 reconstruction after an MDS restart. Drop mountpoint reference
1794 in reconstruct_open() and free dentry reference also.
1799 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
1800 Details : Move the 'good_osts' check before the 'total_bavail' check. This
1801 will result in an -EAGAIN and in the exit call path we call
1802 alloc_rr() which will with increasing aggressiveness attempt to
1803 aquire precreated objects on the minimum number of required OSCs.
1807 Description: Use old size assignment to avoid deadlock
1808 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
1809 the scheduling while holding a spinlock. We do not need locking
1810 for size in ll_update_inode() because size is only updated from
1811 the MDS for directories or files without objects, so there is no
1812 other place to do the update, and concurrent access to such inodes
1813 are protected by the inode lock.
1817 Description: Use __u64 instead of int for valid bits
1821 Description: resolve "_IOWR redefined" build error on SLES10
1825 Description: dump the memory debugging after all modules are unloaded to
1826 suppress false negative in conf_sanity test 39
1830 Description: the recovery timer never expires
1831 Details : for new client connect request, the recovery timer should not be
1832 reset, otherwise recovery timer will never expired, if the old
1833 client never come. Only old client connect and first connection
1834 req should trigger recovery timer reset.
1838 Description: the min numbers of lproc stats are wrong
1839 Details : adding a new constant LC_MIN_INIT and use it for initialization
1843 Frequency : always with interactive lfs
1845 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1848 Frequency : with multiple concurrent readdir processes in same directory
1849 Bugzilla : 15406, 15169, 15175
1850 Description: misc fixes for directory readahead.
1851 Details : prevent previous statahead async RPC callback from processing the
1852 current "statahead_info", race condition between async RPC callback
1853 add dentry into dentry hash table and "ls" thread revalidate such
1854 dentry, statahead his/miss control for hidden items, and so on.
1856 Severity : enhancement
1858 Description: build kernel-ib packages for OFED 1.3 in our release cycle
1862 Description: incore types cleaning in quota code (with respect to 64-bit limits)
1863 Details : several u32 variables declarations are replaced with u64 declarations
1868 Description: fix SLES kernel versioning
1869 Details : the kernel version for our SLES 10 kernel did not include a "-"
1870 before the "smp" at the end. while this was not a problem in
1871 general, it did mean that software trying to use the kernel
1872 version to try to detect a vendor specific kernel would fail.
1873 this was most evident by the OFED build scripts.
1878 Description: Don't update lov_desc members until making sure they are valid
1879 Details : When updating lov_desc members via proc fs, need fix their
1880 validities before doing the real update.
1883 Frequency : very rare
1885 Description: don't put request into delay list while invalidate in flight.
1886 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
1887 import in flight. this produce timeout for invalidate and sometimes
1888 can cause stale data.
1890 Severity : enhancement
1892 Description: Update kernel to SLES9 2.6.5-7.311.
1894 Severity : enhancement
1896 Description: Update kernel to RHEL4 2.6.9-67.0.7.
1901 Frequency : on PPC only
1902 Description: not convert ost objects for directory because it's not exist.
1903 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
1905 Severity : enhancement
1907 Description: Fix warnings with compile liblustre at sles10/rhel5 which have
1908 __u64 as usingied long long type.
1911 Frequency : rare, on shutdown
1913 Description: race process ast vs remove callback
1914 Details : removing callback before disconnect import open race with
1915 processing callback.
1917 Severity : enhancement
1919 Description: Update kernel to SLES9 2.6.5-7.311.
1921 Severity : enhancement
1923 Description: Files open for execute are not marked busy on SLES10
1924 Details : Add FMODE_EXEC to SLES10 SP1 server kernel series.
1926 Severity : enhancement
1928 Description: Add server support for vanilla-2.6.22.14.
1931 Frequency : occasional
1933 Description: Avoid lov_create() getting stuck in obd_statfs_rqset()
1934 Details : If an OST is down the MDS will hang indefinitely in
1935 obd_statfs_rqset() waiting for the statfs data. While for
1936 MDS QOS usage of statfs, it should not stuck in waiting.
1938 Severity : enhancement
1940 Description: Disable adaptive timeouts by default
1943 Frequency : on network error
1945 Description: panic with double free request if network error
1946 Details : mdc_finish_enqueue is finish request if any network error occuring,
1947 but it's true only for synchronus enqueue, for async enqueue
1948 (via ptlrpcd) this incorrect and ptlrpcd want finish request
1952 Frequency : rare, on recovery
1954 Description: read procfs can produce deadlock in some situation
1955 Details : Holding lprocfs lock with send rpc can produce block for destroy
1956 obd objects and this also block reconnect with -EALREADY.
1957 This isn't fix all lprocfs bugs - but make it rare.
1959 Severity : enhancement
1961 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
1964 Frequency : frequent on X2 node
1966 Description: mdc_set_open_replay_data LBUG
1967 Details : Set replay data for requests that are eligible for replay.
1972 Description: lustre_mgs: operation 101 on unconnected MGS
1973 Details : When MGC is disconnected from MGS long enough, MGS will evict the
1974 MGC, and late on MGC cannot successfully connect to MGS and a lot
1975 of the error messages complaining that MGS is not connected.
1978 Frequency : rare, depends on device drivers and load
1980 Description: MDS or OSS nodes crash due to stack overflow
1981 Details : Code changes in 1.6.4 increased the stack usage of some functions.
1982 In some cases, in conjunction with device drivers that use a lot
1983 of stack, the MDS (or possibly OSS) service threads could overflow
1984 the stack. One change which was identified to consume additional
1985 stack has been reworked to avoid the extra stack usage.
1987 Severity : enhancement
1989 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1991 Severity : enhancement
1993 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1995 Severity : enhancement
1997 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1999 Serverity : enhancement
2001 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
2004 Frequency : rare on shutdown OST
2006 Description: Don't allow skipping OSTs if index has been specified.
2007 Details : Don't allow skipping OSTs if index has been specified, make locking
2008 in internal create lots better.
2013 Description: ASSERTION(!PageDirty(page)) failed
2014 Details : Wrong check could lead to an assertion failure under specific
2020 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
2021 Details : access to bitfield in structure is always rounded to long
2022 and this produce problem with not atomic change any bit.
2027 Description: Lustre make rpms failed.
2028 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
2029 builds Lustre rpms from tarball.
2031 Severity : enhancement
2033 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
2036 Frequency : rare on shutdown OST
2038 Description: If llog cancel was not send before clean_exports phase, this can
2039 produce deadlock in llog code.
2040 Details : If llog thread has last reference to obd and call class_import_put
2041 this produce deadlock because llog_cleanup_commit_master wait when
2042 last llog_commit_thread exited, but this never success because was
2043 called from llog_commit_thread.
2046 Frequency : only if OST index is skipped
2048 Description: NULL lov_tgts causing MDS oops
2049 Details : more safe checks for NULL lov_tgts for avoid oops.
2051 Severity : enhancement
2053 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
2055 Severity : enhancement
2057 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
2062 Description: make mgs_setparam() handle fsname containing dash
2063 Details : fsname containing a dash does not work with lctl conf_param
2065 Severity : enhancement
2067 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
2070 Frequency : rare, in recovery and (or) destroy lovobjid file.
2072 Description: rewrite lov objid code.
2073 Details : Cleanup for lov objid code, remove scability problems and wrong
2074 locking. Fix sending last_id into ost.
2076 Severity : enhancement
2078 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
2080 Severity : enhancement
2082 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
2083 Details : Use d_move_locked instead of __d_move.
2086 Frequency : rare, at shutdown
2088 Description: access already free / zero obd_namespace.
2089 Details : if client_disconnect_export was called without force flag set,
2090 and exist connect request in flight, this can produce access to
2091 NULL pointer (or already free pointer) when connect_interpret
2092 store ocd flags in obd_namespace.
2095 Frequency : only at startup
2097 Description: not alloc memory with spinlock held.
2098 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
2099 if any spinlock held.
2104 Description: lfs find does not continue on file error
2105 Details : Continue other files processing when a file/dir is absent.
2109 Description: Inconsistent usage of lustre_pack_reply()
2110 Details : Standardize the usage of lustre_pack_reply() such that it
2111 always generate a CERROR on failure.
2114 Frequency : very rare
2116 Description: Fix replay if there is an un-replied request and open
2117 Details : In some cases, older replay request will revert the
2118 mcd->mcd_last_xid on MDS which is used to record the client's
2119 latest sent request.
2121 Severity : enhancement
2123 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
2125 Severity : enhancement
2127 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
2129 Severity : enhancement
2131 Description: Update to SLES9 kernel-2.6.5-7.287.3.
2133 Severity : enhancement
2135 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
2137 Severity : enhancement
2139 Description: Build SLES10 patchless client fails
2140 Details : The configure was broken by run ./configure with
2141 --with-linux-obj=.... argument for patchless client. When the
2142 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
2143 can't search header adequately. Use absolute path such as
2144 -I($LINUX)/include instead.
2146 Severity : enhancement
2148 Description: Lustre Page Accounting
2149 Details : New macros for page alloc and free which enable accounting
2150 of page allocation of Lustre. Use percpu counters to store memory
2151 and page statistics.
2154 Frequency : only if debugging is disabled
2156 Description: LASSERT_{REQ,REP}SWAB macros are buggy
2157 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
2158 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
2159 these macros and replace them with their definitions instead.
2164 Description: interrupt oig_wait produce painc on resend.
2165 Details : brw_redo_request can be used for resend requests from ptlrpcd and
2166 private set, and this produce situation when rq_ptlrpcd_data not
2167 copyed to new allocated request and triggered LBUG on assert
2168 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
2169 ptlrpcd set if request is changed and can be safety changed to use
2172 Severity : enhancement
2174 Description: organize the server-side client stats on per-nid basis
2175 Details : Change the structure of stats under obdfilter and mds to
2183 The "uuid"s file would list the uuids of _active_ exports.
2184 And the clear entry is to clear all stats and stale nids.
2189 Description: Processes looping in ll_readdir() on Lustre clients finally causing
2190 a full node pseudo-hang
2191 Details : Concurrent access to the same directory from multiple clients with
2192 intensive file creation/removal can cause a client node to spin in
2193 ll_readdir(). i_version must be increased every time the lock is
2194 cancelled to ensure a revalidate is done.
2199 Description: touch file failed when fs is not full
2200 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
2201 otherwise we can get ENOSP while fs is not full.
2204 Frequency : only for Cray XT3
2205 Bugzilla : 12829/13455
2206 Description: Changing primary group doesn't change the group lustre assigns to
2208 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
2209 overridden with the primary group provided by the group upcall,
2210 whereas the supplied fsgid can be trusted if it is in the list of
2211 supplementary groups returned by the group upcall.
2213 Severity : enhancement
2215 Description: Root Squash Functionality
2216 Details : Implementation of NFS-like root squash capability. Specifically,
2217 don't allow someone with root access on a client node to be able
2218 to manipulate files owned by root on a server node.
2220 Severity : enhancement
2222 Description: Slow trucate/writes to huge files at high offsets.
2223 Details : Directly associate cached pages to lock that protect those pages,
2224 this allows us to quickly find what pages to write and remove
2225 once lock callback is received.
2230 Description: Too many locks accumulating on client during NFS usage
2231 Details : mds_open improperly used accmode to find out access mode to a
2232 file. Also mdc_intent_lock logic to find out if we already have
2233 lock similar to just received was flawed since introduction of
2234 skiplists - locks are now added to the front of the granted
2239 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
2241 Details : In obd setup failure handler, obd_stopping will not necessarily to
2242 be 1, and obd_set_up should also be checked to make sure whether
2243 obd is completely setup.
2245 Severity : enhancement
2247 Description: Allow masking D_WARNING, D_ERROR messages from console
2248 Details : Console messages can now be disabled via lnet.printk.
2253 Description: User code with malformed file open parameter crashes client node
2254 Details : Before packing join_file req, all the related reference should be
2255 checked carefully in case some malformed flags cause fake join_file
2261 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
2262 grain of quota doing harm to quota's accuracy
2263 Details : qunit size will be changed when quota limitation is too low/high;
2264 record the pending quota write in order to get more accureate
2265 quota; delete the patch for bug12588, which is unnecessary when
2266 this patch is landed. This bug also contains fixes for bug 14526,
2267 14299, 14601 and 13794.
2271 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
2272 Details : ldlm_completion_ast() assumes that a lock is granted when the req
2273 mode is equal to the granted mode. However, it should also check
2274 that LDLM_FL_CP_REQD is not set.
2278 Description: Heavy nfs access might result in deadlocks
2279 Details : After ELC code landed, it is now improper to enqueue any mds
2280 locks under och_sem, because enqueue might want to decide to
2281 cancel open locks for same inode we are holding och_sem for.
2285 Description: 35% write performance drop with ldiskfs2 when quotas are on
2286 Details : Enable ext3 journalled quota by default to improve performance
2287 when quotas are turned on.
2291 Description: Client eviction while running blogbench
2292 Details : A lot of unlink operations with concurrent I/O can lead to a
2293 deadlock causing evictions. To address the problem, the number of
2294 oustanding OST_DESTROY requests is now throttled to
2295 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
2296 callbacks are processed in priority.
2299 Frequency : RHEL4 only
2301 Description: mkfs is very slow on IA64/RHEL4
2302 Details : A performance regression has been discovered in the MPT Fusion
2303 driver between versions 3.02.73rh and 3.02.99.00rh. As a
2304 consequence, we have downgraded the MPT Fusion driver in the RHEL4
2305 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
2308 Frequency : PPC/PPC64 only
2310 Description: conflicts between asm-ppc64/types.h and lustre_types.h
2311 Details : fix duplicated definitions between asm-ppc64/types.h and
2312 lustre_types.h on PPC.
2315 Frequency : PPC/PPC64 only
2317 Description: asm-ppc/segment.h does not exist
2318 Details : fix compile issue on PPC.
2322 Description: data checksumming impacts single node performance
2323 Details : add support for several checksum algorithms. Currently, CRC32 and
2324 Adler-32 are supported. The checksum type can be changed on the fly
2325 through /proc/fs/lustre/osc/*/checksum_type.
2329 Description: use adler32 for page checksums
2330 Details : when available, use the Adler-32 algorithm instead of CRC32 for
2335 Description: better handle error messages in extents code
2337 Severity : enhancement
2339 Description: SNMP support enhancement
2340 Details : Adding total number of sampled request for an MDS node in snmp
2343 Severity : enhancement
2345 Description: Optimize ldlm waiting list processing for PR extent locks
2346 Details : When processing waiting list for read extent lock and meeting read
2347 lock that is same or wider to it that is not contended, skip
2348 processing rest of the list and immediatelly return current
2349 status of conflictness, since we are guaranteed there are no
2350 conflicting locks in the rest of the list.
2354 Description: Time out and refuse to reconnect
2355 Details : When the failover node is the primary node, it is possible
2356 to have two identical connections in imp_conn_list. We must
2357 compare not conn's pointers but NIDs, otherwise we
2358 can defeat connection throttling.
2362 Description: Client not clear own cache if answer to reconnect is lost.
2363 Details : Client gets evicted from server. Now client also thinks it is
2364 disconnected (or gets ENOTCONN on its operation) and decides to
2365 reconnect. Server receives reconnect message, but cannot find
2366 export. New export is created that is fully valid (new cookie!),
2367 but reply is lost and not reported to client. Client reconnects
2368 again and gets back a just-created connection, but it is not new
2369 so client thinks it was not evicted and does not do recovery.
2373 Description: Detect stride IO mode in read-ahead
2374 Details : When a client does stride read, read-ahead should detect that and
2375 read-ahead pages according to the detected stride pattern.
2379 Description: build for x2 fails
2380 Details : fix compile issue on Cray systems.
2382 Severity : enhancement
2384 Description: implement readv/writev APIs(aio_read/aio_writes in newer kernels)
2385 Details : This greatly improves speed of NFS writes on 2.6 kernels.
2388 Frequency : only on PPC/SLES10
2390 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
2391 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
2392 be defined. Add a hack in mkfs_lustre.c to work around this compile
2397 Description: LASSERT on MDS when client holding flock lock dies
2398 Details : ldlm pool logic depends on number of granted locks equal to
2399 number of released locks which is not true for flock locks, so
2400 just exclude such locks from consideration.
2404 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
2405 Details : Use fsfilt_sync() for both the whole filesystem sync and
2406 individual file sync to eliminate dangerous inode locking
2407 with I_LOCK that can lead to a deadlock.
2411 Description: Update an obsolete wirecheck.c generator
2412 Details : Update wirecheck.c/wirehdr.c and regenerate wiretest.c
2416 Description: Client can panic on open sometimes
2417 Details : It is possible that we try to free already freed request in
2418 ll_file_open in some error cases when we send request from
2423 Description: performance in 1.6.3
2424 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
2425 sure that 1MB requests are not fragmented by the block layer.
2429 Description: LDLM soft lockups - improvement
2430 Details : It is be possible to send the lock handle along with each read
2431 or write request because the client is already doing a lock match
2432 itself so there isn't any reason the OST should have to re-do that
2438 Description: lfs quota fails with deactivated OSTS
2439 Details : With this patch, three improvements are included:
2440 1. detete the softlimit in mds and osts when use "lfs quota".
2441 2. display the inaccurate data in the output of "lfs quota".
2442 3. try to get quota info when "lfs quota" is executed.
2447 Description: Extent locks not granted with no conflicts sometimes.
2448 Details : When race occurs in glimpse handler and nothing is returned,
2449 we do not reprocess the queue after lock cancel, and that leads
2450 to a stall until next activity on a resource
2453 Frequency : failover with quotaon
2455 Description: during mds failovers with quota on, OSTs got into deadlock state
2456 and causing dumpstack.
2457 Details : for every quota slave, at any time, there is only one quota req
2458 is sent to quota master for every uid/gid. Before that quota req
2459 returns, all the thread relative to the same uid/gid will wait.
2460 So if the quota req is lost because mds failovers or any other
2461 reasons, this bug will be hit. Now, dqacq_interpret() will handle
2462 quota reqs who time out.
2464 Severity : enhancement
2467 Description: when quota slave checks if quota is enough, there is an unnecessary
2469 Details : place this wait on necessary place instead of always waiting.
2471 --------------------------------------------------------------------------------
2473 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
2475 * Support for kernels:
2476 2.6.5-7.286 (SLES 9),
2477 2.6.9-55.0.9.EL (RHEL 4),
2478 2.6.16.53-0.8 (SLES 10),
2479 2.6.18-8.1.14.el5 (RHEL 5),
2480 2.6.18.8 vanilla (kernel.org)
2481 * Client support for unpatched kernels:
2482 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2483 2.6.16 - 2.6.22 vanilla (kernel.org)
2484 * Due to recently discovered recovery problems, we do not recommend
2485 using patchless RHEL 4 clients with this or any earlier release.
2486 * Recommended e2fsprogs version: 1.40.2-cfs1
2487 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2490 Frequency : occasional
2492 Description: MDS hang or stay in waiting lock
2493 Details : If client receive lock with CBPENDING flag ldlm need send lock
2494 cancel as separate rpc, to avoid situation when cancel request
2495 can't processed due all i/o threads stay in wait lock.
2498 Frequency : occasional
2500 Description: Do not fail import if osc_interpret_create gets -EAGAIN
2501 Details : If osc_interpret_create got -EAGAIN it immediately exits and
2502 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
2503 oscc_has_objects and see OSC has no objests and call
2504 oscc_internal_create to resend create request.
2506 Severity : enhancement
2508 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
2509 Details : Update which_patch & target file for SLES10 latest kernel.
2511 Severity : enhancement
2513 Description: add --type and --size parameters to lfs find
2514 Details : Enhance lfs find by adding filetype and filesize parameters. Also
2515 multiple OBDs can now be specified for the --obd option.
2517 Severity : enhancement
2519 Description: eliminate client locks in face of contention
2520 Details : file contention detection and lockless i/o implementation
2521 for contended files.
2523 Severity : enhancement
2525 Description: Remove client patches from SLES 10 kernel.
2526 Details : This causes SLES 10 clients to behave as patchless clients
2527 even on a Lustre-patched (server) kernel.
2529 Severity : enhancement
2531 Description: use i_size_read and i_size_write in 2.6 port
2532 Details : replace inode->i_size access with i_size_read/write()
2535 Frequency : when removing large files
2537 Description: scheduling issue during removal of large Lustre files
2538 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
2539 It causes scheduling issues when removing large files (17TB in the
2545 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
2546 Details : Impossible (EPROTO is returned) to access a directory that has a
2547 non-default striping and ACLs.
2550 Frequency : only on ppc
2552 Description: /proc/fs/lustre/devices broken on ppc
2553 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
2554 We should make sure the type of 'index' is loff_t and then cast
2555 explicitly as needed below. Do not assign an explicitly cast
2559 Frequency : only for rhel5
2561 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
2562 Details : Modify the target file & which_kernel.
2565 Frequency : if the uninit_groups feature is enabled on ldiskfs
2567 Description: e2fsck reports "invalid unused inodes count"
2568 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
2569 feature and only a single inode is created in a group then the
2570 "bg_unused_inodes" count is incorrectly updated. Creating a
2571 second inode in that group would update it correctly.
2576 Description: buffer overruns could theoretically occur
2577 Details : llapi_semantic_traverse() modifies the "path" argument by
2578 appending values to the end of the origin string, and a buffer
2579 overrun may occur. Adding buffer overrun check in liblustreapi.
2581 Severity : enhancement
2583 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
2586 Frequency : only if filesystem is inconsistent
2588 Description: handle "serious error: objid * already exists" more gracefully
2589 Details : If LAST_ID value on disk is smaller than the objects existing in
2590 the O/0/d* directories, it indicates disk corruption and causes an
2591 LBUG(). If the object is 0-length, then we should use the existing
2592 object. This will help to avoid a full fsck in most cases.
2594 Severity : enhancement
2596 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
2597 Details : Modify vm-tunables-rhel4.patch.
2599 Severity : enhancement
2601 Description: Kernel config for 2.6.18-vanilla.
2602 Details : Modify targets/2.6-vanilla.target.in.
2603 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
2604 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
2605 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
2606 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
2609 Frequency : occasional
2611 Description: improve handling recoverable errors
2612 Details : If request processed with error which can be recoverable on server
2613 request should be resend, otherwise page released from cache and
2619 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
2620 Details : Modify target file & which_patch.
2621 A flaw was found in the IA32 system call emulation provided
2622 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
2623 value could be stored in the %RAX register, which could trigger an
2624 out-of-bounds system call table access. An untrusted local user
2625 could exploit this flaw to run code in the kernel
2626 (ie a root privilege escalation). (CVE-2007-4573).
2630 Description: change order of libsysio includes
2631 Details : '#include sysio.h' should always come before '#include xtio.h'
2633 Severity : enhancement
2635 Description: adapt the lustre_config script to support the upgrade case
2636 Details : Add "-u" option for lustre_config script to support upgrading 1.4
2637 server targets to 1.6 in parallel.
2642 Description: To avoid grant space > avaible space when the disk is almost
2643 full. Without this patch you might see the error "grant XXXX >
2644 available" or some LBUG about grant, when the disk is almost
2646 Details : In filter_check_grant, for non_grant cache write, we should
2647 check the left space by if (*left > ungranted + bytes), instead
2648 of (*left > ungranted), because only we are sure the left space
2649 is enough for another "bytes", then the ungrant space should be
2650 increase. In client, we should update cl_avail_grant only there is
2651 OBD_MD_FLGRANT in the reply.
2656 Description: Update RHEL 4 kernel to fix local root privilege escalation.
2657 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
2658 described in CVE-2007-4573. This problem could allow untrusted
2659 local users to gain root access.
2662 Frequency : when using O_DIRECT and quotas
2664 Description: Incorrect file ownership on O_DIRECT output files
2665 Details : block usage reported by 'lfs quota' does not take into account
2666 files that have been written with O_DIRECT.
2670 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2671 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2672 window to 0. If it does hit cache miss, and read-pattern is still
2673 stride-io mode, does not reset the stride window, but also does not
2674 increase the stride window length in this case.
2676 --------------------------------------------------------------------------------
2678 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
2680 * Support for kernels:
2681 2.6.5-7.286 (SLES 9),
2682 2.6.9-55.0.2.EL (RHEL 4),
2683 2.6.16.46-0.14 (SLES 10),
2684 2.6.18-8.1.8.el5 (RHEL 5),
2685 2.6.18.8 vanilla (kernel.org)
2686 * Client support for unpatched kernels:
2687 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2688 2.6.16 - 2.6.21 vanilla (kernel.org)
2689 * Due to recently discovered recovery problems, we do not recommend
2690 using patchless RHEL 4 clients with this or any earlier release.
2691 * Recommended e2fsprogs version: 1.40.2-cfs1
2692 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2696 Description: Fix errors in lfs documentation
2697 Details : Fixes man pages
2699 Severity : enhancement
2701 Description: Adaptive timeouts
2702 Details : RPC timeouts adapt to changing server load and network
2703 conditions to reduce resend attempts and improve recovery time.
2705 Severity : enhancement
2707 Description: llapi_file_create() does not allow some changes
2708 Details : add llapi_file_open() that allows specifying the file creation
2709 mode and open flags, and also returns an open file handle.
2711 Severity : enhancement
2713 Description: df doesn't work properly if diskfs blocksize != 4K
2714 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
2717 Frequency : always on directories with default striping set
2719 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
2720 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
2722 Severity : enhancement
2724 Description: merge and cleanup kernel patches.
2725 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
2727 Severity : enhancement
2729 Description: RedHat Update kernel for RHEL5
2730 Details : Kernel config file for RHEL5.
2732 Severity : enhancement
2734 Description: OSS needs mutliple precreate threads
2735 Details : Add ability to start more than one create thread per OSS.
2738 Frequency : only with quota on the root user
2740 Description: mds_obd_create error creating tmp object
2741 Details : When the user sets quota on root, llog will be affected and can't
2742 create files and write files.
2745 Frequency : Always on ia64 patchless client, and possibly others.
2747 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
2748 Details : This allows the patchless client to be loaded on architectures
2749 without this export.
2754 Description: disorder of journal start and llog_add cause deadlock.
2755 Details : in llog_origin_connect, journal start should happen before
2756 llog_add keep the same order as other functions to avoid
2759 Severity : enhancement
2761 Description: RedHat Update kernel for RHEL5
2762 Details : Modify the kernel config file more closer RHEL5.
2764 Severity : enhancement
2766 Description: Build failure against Centos5 (RHEL5)
2767 Details : Define PAGE_SIZE when it isn't present.
2770 Frequency : occasionally when using NFS
2772 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
2773 Details : with NFS, the anon dentry's parent was set to itself in
2774 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
2775 corresponding dentry other than use rec->ur_name.
2777 Severity : enhancement
2779 Description: enable data checksumming by default
2780 Details : enable checksum by default, allow --disable-checksum
2781 configure option and "-o nochecksum" mount option. Checksums
2782 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
2785 Frequency : Occasionally with failover
2787 Description: Client eviction due to failover config
2788 Details : after a connection loss, the lustre client should attempt to
2789 reconnect to the last active server first before trying the
2790 other potential connections.
2792 Severity : enhancement
2794 Description: client-side metadata stat-ahead during readdir(directory readahead)
2795 Details : perform client-side metadata stat-ahead when the client detects
2796 readdir and sequential stat of dir entries therein
2801 Description: when mds and osts use different quota unit(32bit and 64bit),
2802 quota will be released repeatly.
2803 Details : void sending multiple quota reqs to mds, which will keep the
2804 status between the reqs.
2807 Frequency : only with liblustre clients on XT3
2809 Description: evictions taking too long
2810 Details : allow llrd to evict clients directly on OSTs
2814 Description: osts not allocated evenly to files
2815 Details : change the condition to increase offset_idx
2817 Severity : enhancement
2819 Description: self-adjustable client's lru lists
2820 Details : use adaptive algorithm for managing client cached locks lru
2821 lists according to current server load, other client's work
2822 pattern, memory activities, etc. Both, server and client
2823 side namespaces provide number of proc tunables for controlling
2828 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
2829 Details : rewrite inherited code (uuid parsing code from ext2 utils and
2830 readdir code from ext3) from scratch preserving functionality.
2834 Description: Only those disconnect error should be returned by rq_status.
2835 Details : In open/enqueue processs, Some errors, which will cause client
2836 disconnected, should be returned by rq_status, while other
2837 errors should still be returned by intent, then mdc or llite will
2840 Severity : enhancement
2842 Description: Tune the kernel for good SCSI performance.
2843 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
2844 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
2848 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
2849 Bugzilla : 13547, 13627
2850 Description: Data corruption for OSTs that are formatted larger than 2TB
2852 Details : When generating the bio request for lustre file writes the
2853 sector number would overflow a temporary variable before being
2854 used for the IO. The data reads correctly from Lustre (which
2855 will overflow in a similar manner) but other file data or
2856 filesystem metadata may be corrupted in some cases.
2860 Description: "lfs find -obd UUID" prints directories
2861 Details : "lfs find -obd UUID" will return all directory names instead
2862 of just file names. It is incorrect because the directories
2863 do not reside on the OSTs.
2866 Bugzilla : 13596, 13614
2867 Description: MDS hang after unclean shutdown of lots of clients
2868 Details : Never resend AST requests.
2872 Frequency : Always, for kernels after 2.6.16
2873 Description: Fix warning idr_remove called for id=.. which is not allocated.
2874 Details : Last kernels save old s_dev before kill super and not allow
2875 to restore from callback - restore it before call kill_anon_super.
2879 Description: Fix error on 'ls .' at the top of the Lustre mount.
2880 Details : Don't revalidate dentry if it is a root dentry.
2882 --------------------------------------------------------------------------------
2884 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
2886 * Support for kernels:
2887 2.6.5-7.286 (SLES 9),
2888 2.6.9-55.0.2.EL (RHEL 4),
2889 2.6.16.46-0.14 (SLES 10),
2890 2.6.18-8.1.8.el5 (RHEL 5),
2891 2.6.18.8 vanilla (kernel.org)
2892 * Client support for unpatched kernels:
2893 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2894 2.6.16 - 2.6.22 vanilla (kernel.org)
2895 * Due to recently discovered recovery problems, we do not recommend
2896 using patchless RHEL 4 clients with this or any earlier release.
2897 * Recommended e2fsprogs version: 1.39.cfs8
2898 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2903 Description: block reactivating mgc import until all deactivates complete
2904 Details : Fix race when failing back MDT/MGS to itself (testing)
2906 Severity : enhancement
2908 Description: lfs setstripe enhancement
2909 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
2912 Frequency : mds/oss recovery
2914 Description: llog ctxt is refrenced after it has been freed.
2915 Details : llog ctxt refcount was added to avoide the race between ctxt free
2916 and llog recovery process. Each llog user must hold ctxt refcount
2917 before it access the llog. And the llog ctxt can only be freed
2918 when its refcount is zero.
2920 Severity : enhancement
2922 Description: randomly memory allocation failure util
2923 Details : Make lustre randomly failed allocating memory for testing purpose.
2926 Frequency : only for SLES10
2928 Description: Update kernel patch for SLES10 SP1
2929 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
2931 Severity : enhancement
2933 Description: omit set fsid for export NFS
2934 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
2935 as FS_REQUIRES_DEV for avoid problems with generate fsid.
2938 Frequency : after network failures
2940 Description: Add sync option to mount_lustre.c
2941 Details : Client loses data written to lustre after a network interruption.
2943 Severity : enhancement
2945 Description: Error message improvement.
2946 Details : Merging of two LCONSOLE_ERROR_MSG into one.
2948 Severity : enhancement
2950 Description: don't use GFP_* in generic Lustre code.
2951 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
2953 Severity : enhancement
2955 Description: obdclass is limited by single OBD_ALLOC(idarray)
2956 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
2958 Severity : enhancement
2960 Description: updated patchess for new RHEL4 kernel
2961 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
2963 Severity : enhancement
2965 Description: warnings with build patchless client with vanila 2.6.19 and up
2966 Details : change old ctl_table style and replace ctl_table/ctl_table_header
2967 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
2969 Severity : enhancement
2971 Description: O_DIRECT bypasses client statistics.
2972 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
2973 read_ahead_stats were updated. Copied stats section from
2974 osc_send_oap_rpc() into async_internal().
2977 Frequency : only for Cray XT3
2979 Description: peer credits not enough on many OST per OSS systems.
2980 Details : Use new lnet way to add credits as we need those for pings and ASTs
2983 Frequency : only with liblustre
2985 Description: Liblustre is not releasing flock locks on file close.
2986 Details : Release flock locks on file close.
2989 Frequency : only for RHEL4
2991 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
2992 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
2993 Update target file and kernel config.
2997 Description: lustre support for RHEL5
2998 Details : Add support for RHEL5.
3003 Description: ASSERTION(export != NULL) failed in target_handle_connect
3004 Details : Assetion hit is result of rare race between disconnect and connet
3005 to same nid. target_handle_connect found old connect cockie and
3006 tried to reconnect, but can't find export for this cockie.
3011 Description: umount blocks forever on error
3012 Details : In result of wrong using obd_no_recov and obd_force flags client
3013 can hand if cancel or some other requests is lost.
3016 Frequency : Only for SLES
3018 Description: sanity_quota fail test_1
3019 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
3020 file, which makes TSTID[2] inunique.
3022 Severity : enhancement
3024 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
3025 Details : Update target/ChangeLog/which_patch .
3027 Severity : enhancement
3029 Description: jbd statistics
3030 Details : Port older jbd statistics patch for sles10
3035 Description: lvbo_init failed for resource with missing objects.
3036 Details : Fix returning error if we do stat for file with missing/corrupted
3037 objects and i_size set to all sum of size of all avaible objects.
3038 if we to truncate/write to missing object - it is recreated.
3043 Description: Oops in read and write path when failing to allocate lock.
3044 Details : Check if lock allocation failed and return error back.
3047 Frequency : When flocks are used.
3049 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
3050 Details : Flock locks might destroy just granted lock if it could be merged
3051 with another existing flock, this is done in completion handler,
3052 so teach ldlm_cli_enquque_fini that this is a valid case for
3058 Description: reply_lock_interpret crash due to race with it and lock cancel.
3059 Details : Do not replay locks that are being cancelled. Do not reference
3060 locks by their address during replay, just by their handle.
3063 Frequency : only with deactivated OSTs
3065 Description: lstripe command fails for valid OST index
3066 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
3067 instead of lov->desc.ld_active_tgt_count.
3069 Severity : enhancement
3071 Description: Build failure against Centos5 (RHEL5)
3072 Details : Use getpagesize() instead of PAGE_SIZE.
3074 --------------------------------------------------------------------------------
3076 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
3078 * Support for kernels:
3079 2.6.5-7.283 (SLES 9),
3080 2.6.9-55.EL (RHEL 4),
3081 2.6.16.46-0.14 (SLES 10),
3082 2.6.18.8 vanilla (kernel.org)
3083 * Client support for unpatched kernels:
3084 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3085 2.6.16 - 2.6.22 vanilla (kernel.org)
3086 * Due to recently discovered recovery problems, we do not recommend
3087 using patchless RHEL 4 clients with this or any earlier release.
3088 * Recommended e2fsprogs version: 1.39.cfs8
3089 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3090 * Starting with this release, the ldiskfs backing filesystem required
3091 by Lustre is now in its own package, lustre-ldiskfs. This package
3092 should be installed. It is versioned separately from Lustre and
3093 may be released separately in future.
3098 Description: server LBUG when shutting down
3099 Details : Block umount forever until the mount refcount is zero rather
3100 than giving up after an arbitrary timeout.
3102 Severity : enhancement
3104 Description: add optional extra BUILD_VERSION info
3105 Details : add a new environment variable (namely LUSTRE_VERS) which allows
3106 to override the lustre version.
3109 Frequency : 2.6.18 servers only
3111 Description: ll_kern_mount() doesn't release the module reference
3112 Details : The ldiskfs module reference count never drops down to 0
3113 because ll_kern_mount() doesn't release the module reference.
3118 Description: server LBUG when using old ost_num_threads parameter
3119 Details : Accept the old ost_num_threads parameter but warn that it
3120 is deprecated, and fix an off-by-one error that caused an LBUG.
3125 Description: Transient SCSI error results in persistent IO issue
3126 Details : iobuf->dr_error is not reinitialized to 0 between two
3130 Frequency : sometimes when underlying device returns I/O errors
3132 Description: OSTs not going read-only during write failures
3133 Details : OSTs are not remounted read-only when the journal commit threads
3134 get I/O errors because fsfilt_ext3 calls journal_start/stop()
3135 instead of the ext3 wrappers.
3138 Frequency : only with 10000 clients or more
3140 Description: poor connect scaling with increasing client count
3141 Details : Don't run filter_grant_sanity_check for more than 100 exports
3142 to improve scaling for large numbers of clients.
3145 Frequency : SLES10 only
3147 Description: sanity-quota.sh quotacheck failed: rc = -22
3148 Details : Quotas cannot be enabled on SLES10.
3151 Frequency : liblustre clients only
3153 Description: getdirentries does not give error when run on compute nodes
3154 Details : getdirentries does not fail when the size specified as an argument
3155 is too small to contain at least one entry
3157 Severity : enhancement
3159 Description: Add LNET router traceability for debug purposes
3160 Details : If a checksum failure occurs with a router as part of the
3161 IO path, the NID of the last router that forwarded the bulk data
3162 is printed so it can be identified.
3167 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
3168 Details : Due to a race condition, liblustre clients were occasionally
3169 evicted incorrectly.
3171 Severity : enhancement
3173 Description: lfs setstripe use optional parameters instead of postional
3176 Severity : enhancement
3178 Description: Nanosecond timestamp support for ldiskfs
3179 Details : The on-disk ldiskfs filesystem has added support for nanosecond
3180 resolution timestamps. There is not yet support for this at
3181 the Lustre filesystem level.
3184 Frequency : during server recovery
3186 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
3187 Details : request with rq_no_resend flag not awake l_wait_event if they get a
3191 Frequency : nfs export on patchless client
3193 Description: connectathon hang when test nfs export over patchless client
3194 Details : Disconnected dentry cannot be found with lookup, so we do not need
3195 to unhash it or make it invalid
3198 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
3199 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
3200 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
3205 Description: open req refcounting wrong on reconnect
3206 Details : If reconnect happened between getting open reply from server and
3207 call to mdc_set_replay_data in ll_file_open, we will schedule
3208 replay for unreferenced request that we are about to free.
3209 Subsequent close will crash in variety of ways.
3210 Check that request is still eligible for replay in
3211 mdc_set_replay_data().
3216 Description: disable writes to filesystem when reading health_check file
3217 Details : the default for reading the health_check proc file has changed
3218 to NOT do a journal transaction and write to disk, because this
3219 can cause reads of the /proc file to hang and block HA state
3220 checking on a healthy but otherwise heavily loaded system. It
3221 is possible to return to the previous behaviour during configure
3222 with --enable-health-write.
3224 Severity : enhancement
3226 Description: 64-bit inode version
3227 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
3228 to the inode. This will be required for version-based recovery.
3233 Description: MDS fails to start if a duplicate client export is detected
3234 Details : in some rare cases it was possible for a client to connect to
3235 an MDS multiple times. Upon recovery the MDS would detect this
3236 and fail during startup. Handle this more gracefully.
3238 Severity : enhancement
3240 Description: Add -o localflock option to simulate old noflock behaviour.
3241 Details : This will achieve local-only flock/fcntl locks coherentness.
3246 Description: log_commit_thread vs filter_destroy race leads to crash
3247 Details : Take import reference before releasing llog record semaphore
3252 Description: Wrong request locking in request set processing
3253 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
3254 imp_delayed_list, in this place should be used imp_lock.
3257 Frequency : when reconnecting
3259 Description: Grant leak when OSC reconnect to OST
3260 Details : When osc reconnect ost, OST(filter) should check whether it
3261 should grant more space to client by comparing fed_grant and
3262 cl_avail_grant, and return the granted space to client instead
3263 of "new granted" space, because client will call osc_init_grant
3264 to update the client grant space info.
3267 Frequency : when client reconnects to OST
3269 Description: Grant leak when OSC does a resend and replays bulk write
3270 Details : When osc reconnect to OST, OST(filter) should clear grant info of
3271 bulk write request, because the grant info while be sync between
3272 OSC and OST when reconnect, and we should ignore the grant info
3273 these of resend/replay write req.
3278 Description: Grant space more than avaiable space sometimes.
3279 Details : When then OST is about to be full, if two bulk writing from
3280 different clients came to OST. Accord the avaliable space of the
3281 OST, the first req should be permitted, and the second one
3282 should be denied by ENOSPC. But if the seconde arrived before
3283 the first one is commited. The OST might wrongly permit second
3284 writing, which will cause grant space > avaiable space.
3287 Frequency : when client is evicted
3289 Description: Grant might be wrongly erased when osc is evicted by OST
3290 Details : when the import is evicted by server, it will fork another
3291 thread ptlrpc_invalidate_import_thread to invalidate the
3292 import, where the grant will be set to 0. While the original
3293 thread will update the grant it got when connecting. So if
3294 the former happened latter, the grant will be wrongly errased
3295 because of this race.
3300 Description: Checking Stale with correct fid
3301 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
3302 to check whether it is stale, because sometimes, we want the
3303 enqueue happened anyway, and op_data.fid2 will not be initialized.
3305 Severity : enhancement
3307 Description: update patchless client
3308 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
3311 Frequency : only with 2.4 kernel
3313 Description: random memory corruption
3314 Details : size of struct ll_inode_info is to big for union inode.u and this
3315 can be cause of random memory corruption.
3320 Description: Memory leak in recovery
3321 Details : Lov_mds_md was not free in an error handler in mds_create_object.
3322 It should also check obd_fail before fsfilt_start, otherwise if
3323 fsfilt_start return -EROFS,(failover mds during mds recovery).
3324 then the req will return with repmsg->transno = 0 and rc = EROFS.
3325 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
3326 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
3327 be freed no matter whether fsfilt_commit success or not.
3330 Frequency : only with large numbers of cores on a single node
3332 Description: Prevents from taking the superblock lock in llap_from_page for
3334 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
3335 for llap_from_page call in ll_removepage() prevents from taking
3336 the superblock lock for a soon died page.
3341 Description: Not check open intent error before release open handle
3342 Details : in some rare cases, the open intent error is not checked before
3343 release open handle, which may cause
3344 ASSERTION(open_req->rq_transno != 0), because it tries to release
3345 the failed open handle.
3350 Description: Set cat log bitmap only after create log success.
3351 Details : in some rare cases, the cat log bitmap is set too early. and it
3352 should be set only after create log success.
3356 Description: Accessing a block bevice can re-enable I/O when Lustre is
3357 tearing down a device.
3358 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
3363 Description: service threads may hog cpus when there are a lot of requests
3364 Details : Insert cond_resched to give other threads a chance to use some CPU
3369 Description: the cat log was not initialized in recovery
3370 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
3371 unlink log on mds will not be initialized until mds post
3372 recovery. And also in mds post recovery, the unlink log will
3373 initialization will be done asynchronausly, so there will be race
3374 between add unlink log and unlink log initialization.
3378 Description: brw_stats were being printed incorrectly
3379 Details : brw_stats were being printed as log2 but all of them were not
3380 recorded as log2. Also remove some code duplication arising from
3381 filter_tally_{read,write}.
3385 Frequency : rare, only in recovery.
3386 Description: ASSERTION(req->rq_type != LI_POISON) failed
3387 Details : imp_lock should be held while iterating over imp_sending_list for
3388 prevent destroy request after get timeout in ptlrpc_queue_wait.
3392 Description: replay-single.sh test 52 fails
3393 Details : A lock's skiplist need to be cleanup when it being unlinked
3394 from its resource list.
3399 Description: Short directio read returns full requested size rather than
3401 Details : Direct I/O operations should return actual amount of bytes
3402 transferred rather than requested size.
3404 Severity : enhancement
3406 Description: metadata RPC reduction (e.g. for rm performance)
3407 Details : decrease the amount of synchronous RPC between clients and servers
3408 by canceling conflicing lock before the operation on the client
3409 and packing thier handles into the main operation RPC to server.
3411 Severity : enhancement
3413 Description: add #ifdef HAVE_KERNEL_CONFIG_H
3414 Details : kernels from 2.6.19 not need include linux/config.h, but add
3415 include linux/autoconf.h in commpiler command line.
3417 Severity : enhancement
3419 Description: patchless client support for 2.6.22 kernel
3420 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
3421 removed. In this case we need drop using os depended interface to
3422 kmem_cache and use cfs_mem_cache API.
3427 Description: fix mal-formatted messages
3428 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
3432 Frequency : always in liblustre
3433 Description: wrong IS_ERR implementation in liblustre.h
3434 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
3439 Description: Correct condition for output debug message.
3440 Details : inode i_nlink equal zero is not enough for output message about
3441 disk corruption, i_ctime and i_mode should be also checked.
3445 Frequency : always in patchless client
3446 Description: add configure check for truncate_complete_page
3447 Details : improve checks for exported symbols. This allow run check without
3448 sources, but with Module.symvers shipped with kernel distribution.
3449 add check for truncate_complete_page used by patchless client.
3453 Description: sanity.sh test_77h fails with "test_77h file compare failed"
3454 Details : test_77h uses a file which was messed by other test case.
3458 Description: Not Check whether lov_tgts is NULL in some lov functions
3459 Details : Checking whether lov_tgts is NULL in some functions.
3463 Description: replace obdo_alloc() with OBDO_ALLOC macro
3464 Details : nothing special is done in obdo_alloc() function, and for
3465 debugging purpose, it needs to be replaced with macros.
3469 Description: bad return value and errno from fcntl call
3470 Details : In liblustre API, errno should be a negative value if error
3475 Description: ptlrpc_check_set() LBUG
3476 Details : In case of positive reply from server and failed client bulk
3477 callback after bulk transfer shouldn't LBUG, but process this
3478 request as erroneous.
3480 Severity : enhancement
3482 Description: VFS operations stats tool.
3483 Details : Tool which collects stats by tracking value written in pid,
3484 ppid, gid and uses llstat to generate output to plot graph using
3486 Updated lustre/utils/Makefile.am
3487 Added lustre/utils/ltrack_stats.c
3489 Severity : enhancement
3491 Description: 2.6.18 server support (lustre 1.6.1)
3492 Details : Support for 2.6.18 kernels on the server side.
3497 Description: ASSERTION(imp->imp_conn_current) failed
3498 Details : an assertion failure is hit if a client node boots and attempts to
3499 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
3502 Frequency : only for i686
3504 Description: 1.4.11 RC1 build fails for RHEL 4, i686
3505 Details : Fixed config variable for build.
3510 Description: Updated patchess for new RHEL4 kernel
3511 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
3512 Updated patch jbd-stats-2.6.9.patch
3513 Updated patch qsnet-rhel4-2.6.patch
3514 Updated patch quota-deadlock-on-pagelock-core.patch
3515 Updated patch vfs_intent-2.6-rhel4.patch
3516 Updated patch vfs_races-2.6-rhel4.patch
3517 Updated series file 2.6-rhel4-titech.series
3518 Updated series file 2.6-rhel4.series
3519 Updated kernel config files
3524 Description: lquota slave complains LBUG when reconnecting with mds
3526 Details : quota slave depends qctxt->lqc_import to send its quota request.
3527 This pointer will be invalid if mds did failover or broke its
3528 connect to osts, which leads to LBUG.
3531 Frequency : when qunit size is too small(less than 20M)
3533 Description: write is stopped by improper -EDQUOT
3534 Details : If the master is busy and qunit size is small enough(let's say 1M),
3535 the slave can not get quota from master on time, which will lead
3536 slave to trigger a -EQUOTA to client.
3541 Description: Deadlock during metadata tests
3542 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
3543 called with the per-dentry lock held.
3546 Frequency : SLES9 only
3548 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
3549 Details : set CONFIG_NR_CPUS to 128 instead of 8.
3551 Severity : enhancement
3553 Description: remove fs_prep_san_write operation and related patches
3554 Details : remove the ext3-san-jdike patches which are no longer useful.
3559 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
3560 Details : a disk corruption can cause the mballoc code to assert on a
3561 double free or other extent corruptions. Handle these with
3562 ext3_error() instead of assertions.
3565 Frequency : only with mballoc3 code and deep extent trees
3567 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
3568 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
3569 can cause the filesystem to be remounted read-only.
3574 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
3575 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
3576 few external users of this ioctl number is lfsck's call to
3577 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
3578 Add the old ioctl number to the handler so both old and new
3583 Description: parallel lock callbacks
3584 Details : Instead of sending blocking and completion callbacks as separated
3585 requests, adding them to a set and sending in parallel.
3589 Description: Disable most debugging by default
3590 Details : To improve performance, disable most logging (for debug purposes)
3591 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
3592 default, and HA includes fewer messages.
3596 Frequency : only run on patchless client.
3597 Description: use do_facet on sanity.sh for test handling recoverables errors
3598 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
3602 Description: hash tables for lists of nids, connections and uuids
3603 Details : Hash tables noticeably help when a lot of clients connect to a
3604 server, to faster identify duplicate connections or reconnects,
3605 also to faster find export to evict in manual eviction case.
3609 Description: Sometimes, when the server evict a client, and the client will
3610 not be evicted as soon as possible.
3611 Details : In enqueue req, the error was returned by intent, instead of
3612 rq_status which make ptlrpc layer not detect this error, and
3613 does not evict the client. So enqueue error should be returned
3617 Frequency : only at startup
3619 Description: Delay client connections to MDT until fist MDT->OST connect
3620 Details : If a client tried to create a new file before the MDT had
3621 connected to any OSTs, the create would return EIO. Now
3622 the client will simply block until the MDT connects to the
3623 first OST and the create can succeed.
3626 Frequency : only for SLES9
3628 Description: Routinely utilize latest Quadrics drivers in CFS releases
3629 Details : Update patch qsnet-suse-2.6.patch.
3632 Frequency : only for sles10
3634 Description: Update patches for SLES 10 SP1 kernel.
3635 Details : Update the patch vfs_intent-2.6-sles10.patch.
3640 Description: Routinely utilize latest Quadrics drivers in CFS releases
3641 Details : Update patch qsnet-rhel4-2.6.patch.
3644 Frequency : at statup only
3646 Description: mds_lov_synchronize race leads to various problems
3647 Details : simultaneous MDT->OST connections at startup can cause the
3648 sync to abort, leaving the OSC in a bad state.
3653 Description: Using wrong pointer in osc_brw_prep_request
3654 Details : Access to array[-1] can produce panic if kernel compiled with
3655 CONFIG_PAGE_ALLOC enabled
3657 Severity : enhancement
3659 Description: Async OSC create to avoid the blocking unnecessarily.
3660 Details : If a OST has no remain object, system will block on the creating
3661 when need to create a new object on this OST. Now, ways use
3662 pre-created objects when available, instead of blocking on an
3663 empty osc while others are not empty. If we must block, we block
3664 for the shortest possible period of time.
3668 Frequency : only in recovery
3669 Description: Mark OST as early accessible if his start SYNC.
3670 Details : osc_precreate return flag early accessible if oscc marked as
3671 OSCC_FLAG_SYNC_IN_PROGRESS.
3676 Description: Sometimes precreate code can triger create object on wrong ost
3677 Details : Wrong protected or not not restored variables aftre precreate loop
3678 can produce creation object on wrong ost.
3681 Frequency : oss recovery
3683 Description: llog_commit_thread cleanup should sync with llog_commit_thread
3685 Details : llog_commit_thread_count should be synced between llog_commit
3686 start and cleanup, so new llog_commit thread should not be started
3687 when llog_commit threads being stopped to avoid accessing some
3690 Severity : enhancement
3692 Description: Add printing inode info into message about error in writepage.
3694 --------------------------------------------------------------------------------
3696 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
3701 Frequency : on some architectures
3703 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
3704 Details : Uninitialized flags sometimes cause configuration commands to
3708 Frequency : patchless clients only
3710 Description: missing __iget() symbol export
3711 Details : The __iget() symbol export is missing. To avoid the need for
3712 this on patchless clients the deathrow inode reaper is turned
3713 off, and we depend on the VM to clean up old inodes. This
3714 dependency was during via the fix for bug 12181.
3719 Description: sanity.sh fail: test_52b
3720 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
3721 incorrect inode's flags to client.
3723 --------------------------------------------------------------------------------
3725 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
3727 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
3728 INTEROPERATE with older versions automatically. In many cases a
3729 special upgrade step is needed. Please read the
3730 user documentation before upgrading any part of a 1.4.x system.
3731 * WARNING: Lustre configuration and startup changes are required with
3732 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
3734 * Support for kernels:
3735 2.4.21-47.0.1.EL (RHEL 3),
3736 2.6.5-7.283 (SLES 9),
3737 2.6.9-42.0.10.EL (RHEL 4),
3738 2.6.12.6 vanilla (kernel.org),
3739 2.6.16.27-0.9 (SLES10)
3740 * Client support for unpatched kernels:
3741 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
3742 2.6.16 - 2.6.19 vanilla (kernel.org),
3743 2.6.9-42.0.8EL (RHEL 4)
3744 * Recommended e2fsprogs version: 1.39.cfs6
3745 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3748 Severity : enhancement
3750 Description: Async OSC create to avoid the blocking unnecessarily.
3751 Details : If a OST has no remain object, system will block on the creating
3752 when need to create a new object on this OST. Now, ways use
3753 pre-created objects when available, instead of blocking on an
3754 empty osc while others are not empty. If we must block, we block
3755 for the shortest possible period of time.
3757 Severity : enhancement
3759 Description: MountConf
3760 Details : Lustre configuration is now managed via mkfs and mount
3761 commands instead of lmc and lconf. New obd types (MGS, MGC)
3762 are added for dynamic configuration management. See
3763 https://mail.clusterfs.com/wikis/lustre/MountConf for
3766 Severity : enhancement
3768 Description: dynamic OST addition
3769 Details : OSTs can now be added to a live filesystem
3771 Severity : enhancement
3773 Description: startup order invariance
3774 Details : MDTs and OSTs can be started in any order. Clients only
3775 require the MDT to complete startup.
3777 Severity : enhancement
3779 Description: parallel, asynchronous orphan cleanup
3780 Details : orphan cleanup is now performed in separate threads for each
3781 OST, allowing parallel non-blocking operation.
3783 Severity : enhancement
3785 Description: optimized stripe assignment
3786 Details : stripe assignments are now made based on ost space available,
3787 ost previous usage, and OSS previous usage, in order to try
3788 to optimize storage space and networking resources.
3790 Severity : enhancement
3792 Description: Permanently set tunables
3793 Details : All writable /proc/fs/lustre tunables can now be permanently
3794 set on a per-server basis, at mkfs time or on a live
3797 Severity : enhancement
3799 Description: Lustre message v2
3800 Details : Add lustre message format v2.
3802 Severity : enhancement
3804 Description: client OST exclusion list
3805 Details : Clients can be started with a list of OSTs that should be
3806 declared "inactive" for known non-responsive OSTs.
3810 Description: ENOENT returned for valid filehandle during dbench.
3811 Details : Check if a directory has children when invalidating dentries
3812 associated with an inode during lock cancellation. This fixes
3813 an incorrect ENOENT sometimes seen for valid filehandles during
3814 testing with dbench.
3817 Frequency : SFS test only (otherwise harmless)
3819 Description: SPEC SFS validation failure on NFS v2 over lustre.
3820 Details : Changes the blocksize for regular files to be 2x RPC size,
3821 and not depend on stripe size.
3823 Severity : enhancement
3825 Description: fine-grained SMP locking inside DLM
3826 Details : Improve DLM performance on SMP systems by removing the single
3827 per-namespace lock and replace it with per-resource locks.
3829 Severity : enhancement
3831 Description: don't hold multiple extent locks at one time
3832 Details : To avoid client eviction during large writes, locks are not
3833 held on multiple stripes at one time or for very large writes.
3834 Otherwise, clients can block waiting for a lock on a failed OST
3835 while holding locks on other OSTs and be evicted.
3837 Severity : enhancement
3839 Description: Multiple MD RPCs in flight.
3840 Details : Further unserialise some read-only MDT RPCs - learn about intents.
3841 To avoid overly-overloading MDT, introduce a limit on number of
3842 MDT RPCs in flight for a single client and add /proc controls
3843 to adjust this limit.
3845 Severity : enhancement
3847 Description: client read/write statistics
3848 Details : Add client read/write call usage stats for performance
3849 analysis of user processes.
3850 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
3851 file access. extents_stats shows chunk size distribution.
3852 extents_stats_per_process show chunk size distribution per
3855 Severity : enhancement
3857 Description: per-client statistics on server
3858 Details : Add ldlm and operations statistics for each client in
3859 /proc/fs/lustre/mds|obdfilter/*/exports/
3861 Severity : enhancement
3863 Description: improved MDT statistics
3864 Details : Add detailed MDT operations statistics in
3865 /proc/fs/lustre/mds/*/stats
3867 Severity : enhancement
3869 Description: VFS operations stats
3870 Details : Add client VFS call stats, trackable by pid, ppid, or gid
3871 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
3876 Description: Fix client-side osc byte counters
3877 Details : The osc read/write byte counters in
3878 /proc/fs/lustre/osc/*/stats are now working
3881 Frequency : always as root on SLES
3883 Description: Failure of copying files with lustre special EAs.
3884 Details : Client side always return success for setxattr call for lustre
3885 special xattr (currently only "trusted.lov").
3890 Description: Refcount LNET uuids
3891 Details : The global LNET uuid list grew linearly with every startup;
3892 refcount repeated list entries instead of always adding to
3895 Severity : enhancement
3897 Description: Dynamic service threads
3898 Details : Within a small range, start extra service threads
3899 automatically when the request queue builds up.
3902 Frequency : mixed-endian client/server environments
3904 Description: mixed-endian crashes
3905 Details : The new msg_v2 system had some failures in mixed-endian
3908 Severity : enhancement
3910 Description: Easy OST removal
3911 Details : OSTs can be permanently deactivated with e.g. 'lctl
3912 conf_param lustre-OST0001.osc.active=0'
3914 Severity : enhancement
3916 Description: MGS proc entries
3917 Details : Added basic proc entries for the MGS showing what filesystems
3920 Severity : enhancement
3922 Description: provide MGS failover
3923 Details : Added config lock reacquisition after MGS server failover.
3925 Severity : enhancement
3927 Description: add Linux 2.4 support
3928 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
3932 Description: a large application tries to do I/O to the same resource and dies
3933 in the middle of it.
3934 Details : Check the req->rq_arrival time after the call to
3935 ost_brw_lock_get(), but before we do anything about
3936 processing it & sending the BULK transfer request. This
3937 should help move old stale pending locks off the queue as
3938 quickly as obd_timeout.
3941 Frequency : when an incorrect nid is specified during startup
3943 Description: ptlrpc connect to non-existant node causes kernel crash
3944 Details : LNET can't be re-entered from an event callback, which
3945 happened when we expire a message after the export has been
3946 cleaned up. Instead, hand the zombie cleanup off to another
3949 Severity : enhancement
3951 Description: plain/inodebits lock performance improvement
3952 Details : Grouping plain/inodebits in granted list by their request modes
3953 and bits policy, thus improving the performance of search through
3957 Frequency : only if OST filesystem is corrupted
3959 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3960 Details : for a short time RPCs with bulk IO are in the replay list,
3961 but replay of bulk IOs is unimplemented. If the OST filesystem
3962 is corrupted due to disk cache incoherency and then replay is
3963 started it is possible to trip an assertion. Avoid putting
3964 committed RPCs into the replay list at all to avoid this issue.
3967 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3969 Description: System hang on startup
3970 Details : This bug allowed the liblustre (e.g. catamount) client to
3971 return to the app before handling all startup RPCs. This
3972 could leave the node unresponsive to lustre network traffic
3973 and manifested as a server ptllnd timeout.
3975 Severity : enhancement
3977 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3978 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
3979 boolean to control whether to print peer debug info when a
3980 client's RPC times out.
3983 Frequency : only for kernels with patches from Lustre below 1.4.3
3985 Description: Remove old rdonly API
3986 Details : Remove old rdonly API which unused from at least lustre 1.4.3
3989 Frequency : only for devices with external journals
3991 Description: Set external device read-only also
3992 Details : During a commanded failover stop, we set the disk device
3993 read-only while the server shuts down. We now also set any
3994 external journal device read-only at the same time.
3997 Frequency : when upgrading from 1.4 while trying to change parameters
3999 Description: The wrong (new) MDC name was used when setting parameters for
4000 upgraded MDT's. Also allows changing of OSC (and MDC)
4001 parameters if --writeconf is specified at tunefs upgrade time.
4004 Frequency : when setting specific ost indicies
4006 Description: QOS code breaks on skipped indicies
4007 Details : Add checks for missing OST indicies in the QOS code, so OSTs
4008 created with --index need not be sequential.
4010 Severity : enhancement
4012 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
4013 Details : The uninit_groups feature works in conjunction with the kernel
4014 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
4015 up the pass1 processing of e2fsck. This is a read-only feature
4016 in ldiskfs2 only, so older kernels and current ldiskfs cannot
4017 mount filesystems that have had this feature enabled.
4019 Severity : enhancement
4021 Description: Improve multi-block allocation algorithm to avoid fragmentation
4022 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
4023 allocation locality and avoid filesystem fragmentation.
4025 ------------------------------------------------------------------------------
4027 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
4029 * Support for kernels:
4030 2.4.21-47.0.1.EL (RHEL 3)
4031 2.6.5-7.283 (SLES 9)
4032 2.6.9-42.0.10.EL (RHEL 4)
4033 2.6.12.6 vanilla (kernel.org)
4034 2.6.16.27-0.9 (SLES 10)
4035 * Recommended e2fsprogs version: 1.39.cfs5
4037 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4041 Frequency : occasional, depends on client load and configuration
4042 Bugzilla : 12181, 12203
4043 Description: data loss for recently-modified files
4045 Details : In some cases it is possible that recently written or created
4046 files may not be written to disk in a timely manner (this should
4047 normally be within 30s unless client IO load is very high).
4048 The problem appears as zero-length files or files that are a
4049 multiple of 1MB in size after a client crash or client eviction
4050 that are missing data at the end of the file.
4052 This problem is more likely to be hit on clients where files are
4053 repeatedly created and unlinked in the same directory, clients
4054 have a large amount of RAM, have many CPUs, the filesystem has
4055 many OSTs, the clients are rebooted frequently, and/or the files
4056 are not accessed by other nodes after being written.
4058 The presence of the problem can be detected by looking at
4059 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
4060 smaller than the second (nr_unused) then dirty files will not
4061 be flushed automatically to disk. "sync; sleep 10" should be
4062 run several times on the node before unmounting it to update
4063 Lustre (this is also safe to run on nodes without this problem).
4065 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
4066 that can cause this same problem, so customers using that kernel
4067 also need to update the kernel in addition to Lustre. In order
4068 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
4070 It is normal that files written just before a client crash (less
4071 than 30s) may not yet have been flushed to disk, even for local
4075 Frequency : frequent on thin XT3 nodes
4077 Description: UUID collision on thin XT3 Linux nodes
4078 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
4079 randomly, since we relied on an insufficiently-seeded PRNG.
4084 Description: OSS hangs after "All ost request buffers busy"
4085 Details : A deadlock between quota and journal operations caused OSS
4086 hangs after printing "All ost request buffers busy."
4089 Frequency : always on liblustre builds
4091 Description: Cleanup compiler warnings on liblustre
4094 Frequency : always on liblustre builds on XT3
4096 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
4100 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
4101 > 32000 subdirectories
4102 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
4103 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
4104 subdirectory count crosses 32000. This will aid e2fsck to
4105 correctly handle more than 32000 subdirectories.
4108 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
4110 Description: System hang on startup
4111 Details : This bug allowed the liblustre (e.g. catamount) client to
4112 return to the app before handling all startup RPCs. This
4113 could leave the node unresponsive to lustre network traffic
4114 and manifested as a server ptllnd timeout.
4116 Severity : enhancement
4118 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
4119 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
4120 boolean to control whether to print peer debug info when a
4121 client's RPC times out.
4126 Description: make O_SYNC working on 2.6 kernels
4127 Details : 2.6 kernels use different method for mark pages for write,
4128 so need add a code to lustre for O_SYNC work.
4133 Description: Failure to close file and release space on NFS
4134 Details : Put inode details into lock acquired in ll_intent_file_open.
4135 Use mdc_intent_lock in ll_intent_open to properly
4136 detect all kind of errors unhandled by mdc_enqueue.
4141 Description: proc file read during shutdown sometimes raced obd removal,
4143 Details : Add lock to prevent obd access after proc file removal.
4146 Frequency : Only for files larger than 4GB on 32-bit clients.
4148 Description: improperly doing page alignment of locks
4149 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
4150 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
4155 Description: RPCs being resent when they shouldn't be
4156 Details : Some RPCs that should not be resent are being resent. This
4157 can cause inconsistencies in the RPC state machine. Do not
4158 resend such requests.
4161 Frequency : rare, only with NFS export
4163 Description: Crash on NFS re-export node
4164 Details : under very unusual load conditions an assertion is hit in
4165 ll_intent_file_open()
4168 Frequency : only if OST filesystem is corrupted
4170 Description: client incorrectly hits assertion in ptlrpc_replay_req()
4171 Details : for a short time RPCs with bulk IO are in the replay list,
4172 but replay of bulk IOs is unimplemented. If the OST filesystem
4173 is corrupted due to disk cache incoherency and then replay is
4174 started it is possible to trip an assertion. Avoid putting
4175 committed RPCs into the replay list at all to avoid this issue.
4180 Description: large O_DIRECT requests fail under memory pressure/fragmentation
4181 Details : Large single O_DIRECT read and write calls can fail to allocate
4182 a sufficiently large buffer to process the request. In case of
4183 allocation failure the allocation is retried with a smaller
4184 buffer and broken into smaller requests.
4186 Severity : enhancement
4188 Description: Add -o localflock option to simulate old noflock behaviour.
4189 Details : This will achieve local-only flock/fcntl locks coherentness.
4194 Description: versioning check is incomplete
4195 Details : Checking the version difference of client vs. server, report
4196 error if the gap is too big.
4201 Description: add support PG_writeback bit
4202 Details : add support for PG_writeback bit for Lustre, for more carefull
4203 work with page cache in 2.6 kernel. This also fix some deadlocks
4204 and remove hack for work O_SYNC with 2.6 kernel.
4206 Severity : enhancement
4208 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
4209 Details : The uninit_groups feature works in conjunction with the kernel
4210 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
4211 up the pass1 processing of e2fsck. This is a read-only feature
4212 in ldiskfs2 only, so older kernels and current ldiskfs cannot
4213 mount filesystems that have had this feature enabled.
4215 Severity : enhancement
4217 Description: Improve multi-block allocation algorithm to avoid fragmentation
4218 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
4219 allocation locality and avoid filesystem fragmentation.
4221 ------------------------------------------------------------------------------
4223 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
4225 * Support for kernels:
4226 2.6.9-42.0.3.EL (RHEL 4)
4227 2.6.5-7.276 (SLES 9)
4228 2.4.21-47.0.1.EL (RHEL 3)
4229 2.6.12.6 vanilla (kernel.org)
4230 2.6.16.21-0.8 (SLES10)
4231 * Recommended e2fsprogs version: 1.39.cfs2-0
4233 * The backwards-compatible /proc/sys/portals symlink has been removed
4234 in this release. Before upgrading, please ensure that you change
4235 any configuration scripts or /etc/sysctl.conf files that access
4236 /proc/sys/portals/* or sysctl portals.* to use the corresponding
4237 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
4238 in advance of the upgrade on any system running Lustre 1.4.6 or
4239 newer, since /proc/sys/lnet was added in that version.
4240 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4244 Frequency : only when quota is used
4246 Description: avoid scanning export list for quota master
4247 Details : Change the algorithms to avoid scanning export list in order
4248 to improve the efficiency.
4251 Frequency : MDS failover only, very rarely
4253 Description: "went back in time" messages on mds failover
4254 Details : The greatest transno may be lost when the current operation
4255 finishes with an error (transno==0) and the client's last_rcvd
4256 record is over-written. Save the greatest transno in the
4257 mds_last_transno for this case.
4260 Frequency : always for specific kernels and striping counts
4262 Description: client may get "Matching packet too big" without ACL support
4263 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
4264 when trying to access files in certain configurations. The
4265 clients should in fact be denied when mounting because they do
4266 not understand ACLs.
4269 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
4271 Description: many clients connecting with IO in progress causes connect timeouts
4272 Details : Avoid synchronous journal commits to avoid delays caused by many
4273 clients connecting/disconnecting when bulk IO is in progress.
4274 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
4275 OST_IO_PORTAL to avoid delays behind potentially many pending
4279 Frequency : occasionally with multiple writers to a single file
4281 Description: shared writes to file may result in wrong size reported by stat()
4282 Details : Allow growing of kms when extent lock is cancelled
4285 Frequency : always with random mmap IO to multi-striped file
4287 Description: mmap write might be lost if we are writing to a 'hole' in stripe
4288 Details : Only if the hole is at the end of OST object so that kms is too
4289 small. Fix is to increase kms accordingly in ll_nopage.
4292 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
4294 Description: writes to a missing object would leak memory on the OST
4295 Details : If there is an inconsistency between the MDS and OST filesystems,
4296 such that the MDS references an object that doesn't exist, writes
4297 to that object will leak memory due to incorrect cleanup in the
4298 error handling path, eventually running out of memory on the OST.
4303 Description: Creating too long symlink causes lustre errors
4304 Details : Check symlink and name lengths before sending requests to MDS.
4307 Frequency : only if flock is enabled (not on by default)
4309 Description: posix locks not released on fd closure on 2.6.9+
4310 Details : We failed to add posix locks to list of inode locks on 2.6.9+
4311 kernels, this caused such locks not to be released on fd close and
4312 then assertions on fs unmount about still used locks.
4315 Frequency : MDS failover only, very rarely
4317 Description: clients may get ASSERTION(granted_lock != NULL)
4318 Details : When request was taking a long time, and a client was resending
4319 a getattr by name lock request. The were multiple lock requests
4320 with the same client lock handle and
4321 mds_getattr_name->fixup_handle_for_resent_request found one of the
4322 lock handles but later failed with ASSERTION(granted_lock != NULL).
4327 Description: handle->h_buffer_credits > 0, assertion failure
4328 Details : h_buffer_credits is zero after truncate, causing assertion
4329 failure. This patch extends the transaction or creates a new
4333 Frequency : NFS re-export or patchless client
4334 Bugzilla : 11179, 10796
4335 Description: Crash on NFS re-export node (__d_move)
4336 Details : We do not want to hash the dentry if we don't have a lock.
4337 But if this dentry is later used in d_move, we'd hit uninitialised
4338 list head d_hash, so we just do this to init d_hash field but
4339 leave dentry unhashed.
4342 Frequency : NFS re-export or patchless client
4344 Description: NFS exports has problem with symbolic link
4345 Details : lustre client didn't properly install dentry when re-exported
4346 to NFS or running patchless client.
4349 Frequency : NFS re-export or patchless client
4351 Description: Various nfs/patchless fixes.
4352 Details : fixes reuse disconected alias for lookup process - this fixes
4353 warning "find_exported_dentry: npd != pd",
4354 fix permission error with open files at nfs.
4355 fix apply umask when do revalidate.
4358 Frequency : occasional
4360 Description: Crash on NFS re-export node
4361 Details : calling clear_page() on the wrong pointer triggered oops in
4362 generic_mapping_read().
4365 Frequency : rarely, using O_DIRECT IO
4367 Description: unaligned directio crashes client with LASSERT
4368 Details : check for unaligned buffers before trying any requests.
4371 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
4373 Description: stale data returned from RAID cache
4374 Details : If only a small amount of IO is done to the RAID device before
4375 reading it again it is possible to get stale data from the RAID
4376 cache instead of reading it from disk.
4379 Frequency : always for sles10 kernel
4381 Description: sles10 support
4382 Details : ll_follow_link: compile fixes and using of nd_set_link
4383 under newer kernels.
4386 Frequency : depends on arch, kernel and compiler version, always on sles10
4389 Description: recursive or deep enough symlinks cause stack overflow
4390 Details : getting rid of large stack-allocated variable in
4394 Frequency : depends on hardware
4396 Description: lustre write performance loss in the SLES10 kernel
4397 Details : the performance loss is caused by using of write barriers in the
4398 ext3 code. The SLES10 kernel turns barrier support on by
4399 default. The fix is to undo that change for ldiskfs.
4401 ------------------------------------------------------------------------------
4403 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
4405 * Support for kernels:
4406 2.6.9-42.0.3EL (RHEL 4)
4407 2.6.5-7.276 (SLES 9)
4408 2.4.21-47.0.1.EL (RHEL 3)
4409 2.6.12.6 vanilla (kernel.org)
4413 Frequency : quota enabled and large files being deleted
4415 Description: releasing more than 4GB of quota at once hangs OST
4416 Details : If a user deletes more than 4GB of files on a single OST it
4417 will cause the OST to spin in an infinite loop. Release
4418 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4423 Description: statfs data retrieved from /proc may be stale or zero
4424 Details : When reading per-device statfs data from /proc, in the
4425 {kbytes,files}_{total,free,avail} files, it may appear
4426 as zero or be out of date.
4429 Frequency : systems with MD RAID1 external journal devices
4431 Description: lconf's call to blkid is confused by RAID1 journal devices
4432 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4433 of returning all block devices that match the journal UUID.
4436 Frequency : always, for aggregate stripe size over 4GB
4438 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
4439 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
4440 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
4443 Frequency : always if "lfs find" used on a local file/directory
4445 Description: "lfs find" segfaults if used on a local file/directory
4446 Details : The case where a directory component was not specified wasn't
4447 handled correctly. Handle this properly.
4450 Frequency : always on ppc64
4452 Description: the write to an ext3 filesystem mounted with mballoc got stuck
4453 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4454 perform endianness conversion.
4457 Frequency : rarely (truncate to non-zero file size after write under load)
4458 Bugzilla : 10730, 10687
4459 Description: Files padded with zeros to next 4K multiple
4460 Details : With filesystems mounted using the "extents" option (2.6 kernels)
4461 it is possible that files that are truncated to a non-zero size
4462 immediately after being written are filled with zero bytes beyond
4463 the truncated size. No file data is lost.
4465 Severity : enhancement
4467 Description: Allow recovery/failover for liblustre clients.
4468 Details : liblustre clients were unaware of failover configurations until
4471 Severity : enhancement
4473 Description: user file locks should fail when not mounting with flock option
4474 Details : Set up an error-returning stub in ll_file_operations.lock field
4475 to prevent incorrect behaviour when client is mounted without
4476 flock option. Also, set up properly f_op->flock field for
4480 Frequency : always on ia64
4482 Description: "lfs df" loops on printing out MDS statfs information
4483 Details : The obd_ioctl_data was not initialized and in some systems
4484 this caused a failure during the ioctl that did not return
4485 an error. Initialize the struct and return an error on failure.
4488 Frequency : SLES 9 only
4490 Description: Error of copying files with lustre special EAs as root
4491 Details : Client side always return success for setxattr call for lustre
4492 special xattr (currently only "trusted.lov").
4495 Frequency : rarely on clusters with both ia64+i386 clients
4497 Description: ia64+i686 clients doing shared IO on the same file may LBUG
4498 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
4499 clients are doing concurrent writes to the same file it is
4500 possible that the ia64 clients may LASSERT because the OST
4501 extent locks are not PAGE_SIZE aligned. Ensure that grown
4502 locks are always aligned on the request boundary.
4505 Frequency : specific use, occasional
4507 Description: Overwriting in use executable truncates on-disk binary image
4508 Details : If one node attempts to overwrite an executable in use by
4509 another node, we now correctly return ETXTBSY instead of
4510 truncating the file.
4512 Severity : enhancement
4514 Description: Async OSC create to avoid the blocking unnecessarily.
4515 Details : If a OST has no remain object, system will block on the creating
4516 when need to create a new object on this OST. Now, ways use
4517 pre-created objects when available, instead of blocking on an
4518 empty osc while others are not empty. If we must block, we block
4519 for the shortest possible period of time.
4524 Description: chmod on Lustre root is propagated to other clients
4525 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
4526 invalid by the follow_mount time.
4529 Frequency : liblustre clients only
4531 Description: Race in 'instant cancel' lock handling could lead to such locks
4532 never to be granted in case of SMP MDS
4533 Details : Do not destroy not yet granted but cbpending locks in
4537 Frequency : replay/resend of open
4539 Description: non null lock assetion failure in mds_intent_policy
4540 Details : Trying to replay/resend lockless open requests resulted in
4541 mds_open() returning 0 with no lock. Now it sets a flag if
4542 a lock is going to be returned.
4544 Severity : enhancement
4546 Description: Checksum enhancements
4547 Details : New checksum enhancements allow for resending RPCs that failed
4550 Severity : enhancement
4552 Description: Tunables on number of dirty pages in cacche
4553 Details : Allow to set limit on number of dirty pages cached.
4558 Description: client crash on unmount - lock still has references
4559 Details : In some error handling cases it was possible to leak a lock
4560 reference on a client while accessing a file. This was not
4561 harmful to the client during operation, but would cause the
4562 client to crash when the filesystem is unmounted.
4565 Frequency : specific case, rare
4567 Description: ETXTBSY on mds though file not in use by client
4568 Details : ETXTBSY is no longer incorrectly returned when attempting to
4569 chmod or chown a directory that the user previously tried to
4570 execute or a currently-executing binary.
4573 Frequency : extremely rare except on liblustre-based clients
4575 Description: Lustre space not freed when files are deleted
4576 Details : Clean up open-unlinked files after client eviction. Previously
4577 the unlink was skipped and the files remained as orphans.
4582 Description: OST failure "would be an LBUG" in waiting_locks_callback()
4583 Details : In some cases it was possible to send a blocking callback to
4584 a client doing a glimpse, even though that client didn't get
4585 a lock granted. When the glimpse lock is cancelled on the OST
4586 the freed lock is left on the waiting list and corrupted the list.
4589 Frequency : all core dumps
4591 Description: Broke core dumps to lustre
4592 Details : Negative dentry may be unhashed if parent does not have UPDATE
4593 lock, but some callers, e.g. do_coredump, expect dentry to be
4594 hashed after successful create, hash it in ll_create_it.
4596 ------------------------------------------------------------------------------
4598 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
4600 * Support for kernels:
4601 2.6.9-42.0.2.EL (RHEL 4)
4602 2.6.5-7.276 (SLES 9)
4603 2.4.21-40.EL (RHEL 3)
4604 2.6.12.6 vanilla (kernel.org)
4608 Frequency : always on RHEL 3
4610 Description: Number of open files grows over time
4611 Details : The number of open files grows over time, whether or not
4612 Lustre is started. This was due to a filp leak introduced
4613 by one of our kernel patches.
4615 ------------------------------------------------------------------------------
4617 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
4619 * Support for kernels:
4620 2.6.9-42.EL (RHEL 4)
4621 2.6.5-7.267 (SLES 9)
4622 2.4.21-40.EL (RHEL 3)
4623 2.6.12.6 vanilla (kernel.org)
4628 Bugzilla : 5719, 9635, 9792, 9684
4629 Description: OST (or MDS) trips assertions in (re)connection under heavy load
4630 Details : If a server is under heavy load and cannot reply to new
4631 connection requests before the client resends the (re)connect,
4632 the connection handling code can behave badly if two service
4633 threads are concurrently handing separate (re)connections from
4634 the same client. Add better locking to the connection handling
4635 code, and ensure that only a single connection will be processed
4636 for a given client UUID, even if the lock is dropped.
4638 Severity : enhancement
4640 Description: add TCP zero-copy support to kernel
4641 Details : Add support to the kernel TCP stack to allow zero-copy bulk
4642 sends if the hardware supports scatter-gather and checksumming.
4643 This allows socklnd to do client-write and server-read more
4644 efficiently and reduce CPU utilization from skbuf copying.
4647 Frequency : only if NFS exporting from client
4649 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
4650 Details : If setting attributes on a file created under NFS that had
4651 never been opened it would be possible to oops the client
4652 if the file had no objects.
4656 Bugzilla : 9326, 10402, 10897
4657 Description: client crash in ptlrpcd_wake() thread when sending async RPC
4658 Details : It is possible that ptlrpcd_wake() dereferences a freed async
4659 RPC. In rare cases the ptlrpcd thread alread processed the RPC
4660 before ptlrpcd_wake() was called and the request was freed.
4663 Frequency : always for liblustre
4665 Description: liblustre client does MDS+OSTs setattr RPC for each write
4666 Details : When doing a write from a liblustre client, the client
4667 incorrectly issued an RPC to the MDS and each OST the file was
4668 striped over in order to update the timestamps. When writing
4669 with small chunks and many clients this could overwhelm the MDS
4670 with RPCs. In all cases it would slow down the write because
4671 these RPCs are unnecessary.
4673 Severity : enhancement
4675 Description: allow number of MDS service threads to be changed at module load
4676 Details : It is now possible to change the number of MDS service threads
4677 running. Adding "options mds mds_num_threads={N}" to the MDS's
4678 /etc/modprobe.conf will set the number of threads for the next
4679 time Lustre is restarted (assuming the "mds" module is also
4680 reloaded at that time). The default number of threads will
4681 stay the same, 32 for most systems.
4686 Description: OST crash if filesystem is unformatted or corrupt
4687 Details : If an OST is started on a device that has never been formatted
4688 or if the filesystem is corrupt and cannot even mount then the
4689 error handling cleanup routines would dereference a NULL pointer.
4694 Description: NULL pointer deref in llap_from_page.
4695 Details : get_cache_page_nowait can return a page with NULL (or otherwise
4696 incorrect) mapping if the page was truncated/reclaimed while it was
4697 searched for. Check for this condition and skip such pages when
4698 doing readahead. Introduce extra check to llap_from_page() to
4699 verify page->mapping->host is non-NULL (so page is not anonymous).
4702 Frequency : Sometimes when using sys_sendfile
4704 Description: "page not covered by a lock" warnings from ll_readpage
4705 Details : sendfile called ll_readpage without right page locks present.
4706 Now we introduced ll_file_sendfile that does necessary locking
4707 around call to generic_file_sendfile() much like we do in
4711 Frequency : with certain MDS communication failures at client mount time
4713 Description: NULL pointer deref after failed client mount
4714 Details : a client connection request may delayed by the network layer
4715 and not be sent until after the PTLRPC layer has timed out the
4716 request. If the client fails the mount immediately it will try
4717 to clean up before the network times out the request. Add a
4718 reference from the request import to the obd device and delay
4719 the cleanup until the network drops the request.
4722 Frequency : occasionally during client (re)connect
4724 Description: assertion failure during client (re)connect
4725 Details : processing a client connection request may be delayed by the
4726 client or server longer than the client connect timeout. This
4727 causes the client to resend the connection request. If the
4728 original connection request is replied in this interval, the
4729 client may trip an assertion failure in ptlrpc_connect_interpret()
4730 which thought it would be the only running connect process.
4733 Frequency : only with obd_echo servers and clients that are rebooted
4735 Description: kernel BUG accessing uninitialized data structure
4736 Details : When running an obd_echo server it did not start the ping_evictor
4737 thread, and when a client was evicted an uninitialized data
4738 structure was accessed. Start the ping_evictor in the RPC
4739 service startup instead of the OBD startup.
4741 Severity : enhancement
4742 Bugzilla : 10193 (patchless)
4743 Description: Remove dependency on various unexported kernel interfaces.
4744 Details : No longer need reparent_to_init, exit_mm, exit_files,
4745 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
4748 Frequency : rare (only users of deprecated and unsupported LDAP config)
4750 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
4751 Details : LDAP apparently contains 'lustreName' attributes instead of
4752 'name'. A simple remapping of the name is sufficient.
4755 Frequency : rare (only with non-default dump_on_timeout debug enabled)
4757 Description: waiting_locks_callback trips kernel BUG if client is evicted
4758 Details : Running with the dump_on_timeout debug flag turned on makes
4759 it possible that the waiting_locks_callback() can try to dump
4760 the Lustre kernel debug logs from an interrupt handler. Defer
4761 this log dumping to the expired_lock_main() thread.
4763 Severity : enhancement
4765 Description: Support NFS exporting on 2.6 kernels.
4766 Details : Implement non-rawops metadata methods for NFS server to use without
4767 changing NFS server code.
4770 Frequency : very rare (synthetic metadata workload only)
4772 Description: two racing renames might cause an MDS thread to deadlock
4773 Details : Running the "racer" program may cause one MDS thread to rename
4774 a file from being the source of a rename to being the target of
4775 a rename at exactly the same time that another thread is doing
4776 so, and the second thread has already enqueued these locks after
4777 doing a lookup of the target and is trying to relock them in
4778 order. Ensure that we don't try to re-lock the same resource.
4781 Frequency : only very large systems with liblustre clients
4783 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
4784 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
4785 all OSTs in parallel. This allows the eviction of stale liblustre
4786 clients to proceed much faster than if they were done in series,
4787 and also offers similar improvements for other set_info RPCs.
4792 Description: excessive CPU usage during initial read phase on client
4793 Details : During the initial read phase on a client, it would agressively
4794 retry readahead on the file, consuming too much CPU and impacting
4795 performance (since 1.4.5.8). Improve the readahead algorithm
4796 to avoid this, and also improve some other common cases (read
4797 of small files in particular, where "small" is files smaller than
4798 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
4803 Description: MDS crash when receiving packet with unknown intent.
4804 Details : Do not LBUG in unknown intent case, just return -EFAULT
4806 Severity : enhancement
4807 Bugzilla : 9293, 9385
4808 Description: MDS RPCs are serialised on client. This is unnecessary for some.
4809 Details : Do not serialize getattr (non-intent version) and statfs.
4812 Frequency : occasional, when OST network is overloaded/intermittent
4814 Description: client evicted by OST after bulk IO timeout
4815 Details : If a client sends a bulk IO request (read or write) the OST
4816 may evict the client if it is unresposive to its data GET/PUT
4817 request. This is incorrect if the network is overloaded (takes
4818 too long to transfer the RPC data) or dropped the OST GET/PUT
4819 request. There is no need to evict the client at all, since
4820 the pinger and/or lock callbacks will handle this, and the
4821 client can restart the bulk request.
4824 Frequency : Always when mmapping file with no objects
4826 Description: client crashes when mmapping file with no objects
4827 Details : Check that we actually have objects in a file before doing any
4828 operations on objects in ll_vm_open, ll_vm_close and
4834 Description: Request leak when working with deleted CWD
4835 Details : Introduce advanced request refcount tracking for requests
4836 referenced from lustre intent.
4838 Severity : Enhancement
4840 Description: Cache open file handles on client.
4841 Details : MDS now will return special lock along with openhandle, if
4842 requested and client is allowed to hold openhandle, even if unused,
4843 until such a lock is revoked. Helps NFS a lot, since NFS is opening
4844 closing files for every read/write openration.
4846 Severity : Enhancement
4848 Description: Cache open negative dentries on client when possible.
4849 Details : Guard negative dentries with UPDATE lock on parent dir, drop
4850 negative dentries on lock revocation.
4855 Description: Remounting a client read-only wasn't possible with a zconf mount
4856 Details : It wasn't possible to remount a client read-only with llmount.
4858 Severity : enhancement
4859 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
4860 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
4861 MPICH is included to add Lustre-specific ADIO interfaces.
4862 This is based closely on the UFS ADIO layer and only differs
4863 in file creation, in order to allow the OST striping to be set.
4864 This is user-contributed code and not supported by CFS.
4869 Description: extended inode attributes (immutable, append-only) work improperly
4870 when 2.4 and 2.6 kernels are used on client/server or vice versa
4871 Details : Introduce kernel-independent values for these flags.
4873 Severity : enhancement
4876 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
4877 Details : Many of the /proc/ tunables can only be tuned at a megabyte
4878 granularity. Now, Fractional MB granularity is be supported,
4879 this is very useful for low memory system.
4881 Severity : enhancement
4883 Description: Getattr by fid
4884 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
4885 locks, avoids extra getattr rpc requests to MDS, allows '/' to
4886 have locks and avoids getattr rpc requests for it on every stat.
4889 Frequency : Always, for filesystems larger than 2TB
4891 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
4892 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
4893 filesystems larger than 2TB to be mounted with mballoc enabled.
4896 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
4898 Description: filesystem corruption for non-standard kernels and very large OSTs
4899 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
4900 filesystems larger than 2TB is mounted then the kernel will
4901 silently corrupt the start of the filesystem. CONFIG_LBD is
4902 enabled for all CFS-supported kernels, but the possibility of
4903 this happening with a modified kernel config exists.
4905 Severity : enhancement
4907 Description: add client O_DIRECT support for 2.6 kernels
4908 Details : It is now possible to do O_DIRECT reads and writes to files
4909 in the Lustre client mountpoint on 2.6 kernel clients.
4911 Severity : enhancement
4913 Description: parallel glimpse, setattr, statfs, punch, destroy requests
4914 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
4915 parallel, not waiting for response from every OST before sending
4916 a rpc to the next OST.
4921 Description: setattr vs write race when updating file timestamps
4922 Details : Client processes that update a file timestamp into the past
4923 right after writing to the file (e.g. tar) it is possible that
4924 the updated file modification time can be reset to the current
4925 time due to a race between processing the setattr and write RPC.
4927 Severity : enhancement
4929 Description: Bring 'lfs find' closer in line with regular Linux find.
4930 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
4931 -print0 options and obtains all the needed info through the lustre
4934 Severity : enhancement
4936 Description: support up to 1024 configured devices on one node
4937 Details : change obd_dev array from statically allocated to dynamically
4938 allocated structs as they are first used to reduce memory usage
4943 Description: Flush dirty partially truncated pages during truncate
4944 Details : Immediatelly flush partially truncated pages in filter_setattr,
4945 this way we completely avoid having any pages in page cache on OST
4946 and can retire ugly workarounds during writes to flush such pages.
4951 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
4952 Details : protect inode from truncation within vfs_unlink() context
4953 just take a reference before calling vfs_unlink() and release it
4954 when parent's i_sem is free.
4957 Frequency : always, if extents are used on OSTs
4959 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
4960 Details : OSTs using the extents format would not zero the high 16 bits of
4961 the index physical block number. This is not a problem for any
4962 OST filesystems smaller than 16TB, and no kernels support ext3
4963 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
4964 new/modified files) and can be fixed for existing filesystems
4965 with e2fsprogs-1.39-cfs1.
4970 Description: import connection selection may be incorrect if timer wraps
4971 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
4972 import connections to be ignored if the 32-bit jiffies counter
4973 wraps. Use a 64-bit jiffies counter.
4976 Frequency : during server recovery
4978 Description: crash after server is denying duplicate export
4979 Details : If clients are resending connect requests to the server, the
4980 server refuses to allow a client to connect multiple times.
4981 Fixed a bug in the handling of this case.
4984 Frequency : very large clusters immediately after boot
4986 Description: LNET request buffers exhausted under heavy short-term load
4987 Details : If a large number of client requests are generated on a service
4988 that has previously never seen so many requests it is possible
4989 that the request buffer growth cannot keep up with the spike in
4990 demand. Instead of dropping incoming requests, they are held in
4991 the LND until the RPC service can accept more requests.
4994 Frequency : Sometimes during replay
4996 Description: Assertion failure in ll_local_open after replay.
4997 Details : If replay happened on an open request reply before we were able
4998 to set replay handler, reply will become not swabbed tripping the
4999 assertion in ll_local_open. Now we set the handler right after
5000 recognising of open request
5003 Frequency : very rare
5005 Description: kernel reports "badness in vsnprintf"
5006 Details : Reading from the "recovery_status" /proc file in small chunks
5007 may cause a negative length in lprocfs_obd_rd_recovery_status()
5008 call to vsnprintf() (which is otherwise harmless). Exit early
5009 if there is no more space in the output buffer.
5011 Severity : enhancement
5013 Description: clear OBD RPC statistics by writing to them
5014 Details : It is now possible to clear the OBD RPC statistics by writing
5015 to the "stats" file.
5020 Description: Client mtime is not the same on different clients after utimes
5021 Details : In some cases, the client was using the utimes() syscall on
5022 a file cached on another node. The clients now validate the
5023 ctime from the MDS + OSTs to determine which one is right.
5028 Description: Inability to activate failout mode
5029 Details : lconf script incorrectly assumed that in python string's numeric
5030 value is used in comparisons.
5033 Frequency : always with multiple stripes per file
5035 Description: Inefficient object allocation for mutli-stripe files
5036 Details : When selecting which OSTs to stripe files over, for files with
5037 a stripe count that divides evenly into the number of OSTs,
5038 the MDS is always picking the same starting OST for each file.
5039 Return the OST selection heuristic to the original design.
5044 Description: mount failures may take full timeout to return an error
5045 Details : Under some heavy load conditions it is possible that a
5046 failed mount can wait for the full obd_timeout interval,
5047 possibly several minutes, before reporting an error.
5048 Instead return an error as soon as the status is known.
5050 ------------------------------------------------------------------------------
5052 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
5054 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
5055 INTEROPERATE with older versions automatically. Please read the
5056 user documentation before upgrading any part of a live system.
5057 * WARNING: Lustre networking configuration changes are required with
5058 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
5061 * Support for kernels:
5062 2.6.9-22.0.2.EL (RHEL 4)
5063 2.6.5-7.244 (SLES 9)
5064 2.6.12.6 vanilla (kernel.org)
5067 Severity : enhancement
5068 Bugzilla : 7981/8208
5069 Description: Introduced Lustre Networking (LNET)
5070 Details : LNET is new networking infrastructure for Lustre, it includes
5071 a reorganized network configuration mode (see the user
5072 documentation for full details) as well as support for routing
5073 between different network fabrics. Lustre Networking Devices
5074 (LNDs) for the supported network fabrics have also been
5075 created for this new infrastructure.
5077 Severity : enhancement
5078 Description: Introduced Access control lists
5079 Details : clients can set ACLs on files and directories in order to have
5080 more fine-grained permissions than the standard Unix UGO+RWX.
5081 The MDS must be started with the "-o acl" mount option.
5083 Severity : enhancement
5084 Description: Introduced filesystem quotas
5085 Details : Administrators may now establish per-user quotas on the
5088 Severity : enhancement
5090 Description: Configuration change for the XT3
5091 The PTLLND is now used to run Lustre over Portals on the XT3
5092 The configure option(s) --with-cray-portals are no longer used.
5093 Rather --with-portals=<path-to-portals-includes> is used to
5094 enable building on the XT3. In addition to enable XT3 specific
5095 features the option --enable-cray-xt3 must be used.
5100 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
5101 Details : A race between allocating a new llcd and re-getting the llcd_lock
5102 allowed another thread to grab newly-allocated llcd.
5104 Severity : enhancement
5106 Description: 2.6 OST async journal commit and locking fix to improve performance
5107 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
5108 2.6 kernels are now async as they already were in 2.4 kernels so
5109 that they can commit concurrently with the network bulk transfer.
5110 For block-allocated files the filter allocation semaphore is held
5111 to avoid filesystem fragmentation during allocation. BKL lock
5112 removed for 2.6 xattr operations where it is no longer needed.
5117 Description: lconf incorrectly determined whether two IP networks could talk
5118 Details : In some more complicated routing and multiple-network
5119 configurations, lconf will avoid trying to make a network
5120 connection to a disjoint part of the IP space. It was doing the
5121 math incorrectly for one set of cases.
5126 Description: Fix for potential infinite loop processing records in an llog.
5127 Details : If an llog record is corrupted/zeroed, it is possible to loop
5128 forever in llog_process(). Validate the llog record length
5129 and skip the remainder of the block on error.
5132 Frequency : occasional (liblustre only)
5134 Description: liblustre could not open files whose last component is a symlink
5135 Details : sysio_path_walk() would incorrectly pass the open intent to
5136 intermediate path components.
5139 Frequency : rare (liblustre only with non-standard tuning)
5140 Bugzilla : 7201 (7350)
5141 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
5142 Details : llu_lookup_finish_locks() tries to set lock data on a lock
5143 after it has been released, only do this for referenced locks
5145 Severity : enhancement
5147 Description: specifying an (invalid) directory default stripe_size of -1
5148 would reset the directory default striping
5149 Details : stripe_size -1 was used internally to signal directory stripe
5150 removal, now use "all default" to signal dir stripe removal
5151 as a directory striping of "all default" is not useful
5154 Frequency : common for large clusters running liblustre clients
5156 Description: doing an ls when liblustre clients are running is slow
5157 Details : sending a glimpse AST to a liblustre client waits for every AST
5158 to time out, as liblustre clients will not respond. Since they
5159 cannot cache data we refresh the OST lock LVB from disk instead.
5161 Severity : enhancement
5163 Description: doing an ls at the same time as file IO can be slow
5164 Details : enqueue and other "small" requests can be blocked behind many
5165 large IO requests. Create a new OST IO portal for non-IO
5166 requests so they can be processed faster.
5169 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
5171 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
5172 created file with mode 000
5173 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
5174 allow an ftruncate() as a non-root user to a file with mode 000.
5175 Lustre now respects this flag to disable mode checks when
5176 truncating a file owned by the user
5179 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
5182 Description: Revoking locks from clients that went dead or catatonic might take
5184 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
5185 cancellation of such locks instant on servers without waiting for
5186 any reply from clients. Clients drops these locks when cancel
5187 notification from server is received without replying.
5190 Frequency : liblustre-only, when liblustre client dies or becomes busy
5192 Description: Doing ls on Linux clients can take a long time with active
5194 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
5195 granting such locks to it in the first place if possible. Locks
5196 are taken by proxy on the OST during the read or write and
5197 dropped immediately afterward. Add connect flags handling, do
5198 not grant locks to liblustre clients for glimpse ASTs.
5200 Severity : enhancement
5202 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
5203 Details : Existing read-ahead algorithm is tuned for the case of streamlined
5204 sequential reads and behaves badly with applications doing random
5205 reads. Improve it by reading ahead at least read region, and
5206 avoiding excessive large RPC for small reads.
5208 Severity : enhancement
5210 Description: Creating more than 1000 files for a single job may cause a load
5211 imbalance on the OSTs if there are also a large number of OSTs.
5212 Details : qos_prep_create() uses an OST index reseed value that is an
5213 even multiple of the number of available OSTs so that if the
5214 reseed happens in the middle of the object allocation it will
5215 still utilize the OSTs as uniformly as possible.
5220 Description: OST or MDS may oops in ping_evictor_main()
5221 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
5222 but doesn't restart at beginning of obd_exports_timed list
5225 Severity : enhancement
5227 Description: improve by-nid export eviction on the MDS and OST
5228 Details : allow multiple exports with the same NID to be evicted at one
5229 time without re-searching the exports list.
5232 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
5234 Description: MDS may oops in groups_free()
5235 Details : in rare race conditions a newly allocated group_info struct is
5236 freed again, and this can be NULL. The 2.4 compatibility code
5237 for groups_free() checked for a NULL pointer, but 2.6 did not.
5240 Frequency : common for liblustre clients doing little filesystem IO
5241 Bugzilla : 9352, 7313
5242 Description: server may evict liblustre clients accessing contended locks
5243 Details : if a client is granted a lock or receives a completion AST
5244 with a blocking AST already set it would not reply to the AST
5245 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
5248 Frequency : lfs setstripe, only systems with more than 160 OSTs
5250 Description: unable to set striping with a starting offset beyond OST 160
5251 Details : llapi_create_file() incorrectly limited the starting stripe
5252 index to the maximum single-file stripe count.
5255 Frequency : LDAP users only
5257 Description: lconf did not handle in-kernel recovery with LDAP properly
5258 Details : lconf/LustreDB get_refs() is searching the wrong namespace
5260 Severity : enhancement
5262 Description: bind OST threads to NUMA nodes to improve performance
5263 Details : all OST threads are uniformly bound to CPUs on a single NUMA
5264 node and do their allocations there to localize memory access
5266 Severity : enhancement
5268 Description: llmount can determine client NID directly from Myrinet (GM)
5269 Details : the client NID code from gmnalnid was moved directly into
5270 llmount, removing the need to use this or specifying the
5271 client NID explicitly when mounting GM clients with zeroconf
5274 Frequency : if client is started with down MDS
5276 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
5277 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
5278 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
5279 ptlrpc_import_delayed_req() because we didn't check if the
5280 request was interrupted, and we also didn't break out of the
5281 event loop if there was no timeout
5286 Description: data loss during non-page-aligned writes to a single file from
5287 both multiple nodes and multiple threads on one node at same time
5288 Details : updates to KMS and lsm weren't protected by common lock. Resulting
5289 inconsistency led to false short-reads, that were cached and later
5290 used by ->prepare_write() to fill in partially written page,
5291 leading to data loss.
5294 Frequency : always, if lconf --abort_recovery used
5296 Description: lconf --abort_recovery fails with 'Operation not supported'
5297 Details : lconf was attempting to abort recovery on the MDT device and not
5300 Severity : enhancement
5302 Description: remove cleanup logs
5303 Details : replace lconf-generated cleanup logs with lustre internal
5304 cleanup routines. Eliminates the need for client-cleanup and
5307 Severity : enhancement
5309 Description: add support for EAs (user and system) on lustre filesystems
5310 Details : it is now possible to store extended attributes in the Lustre
5311 client filesystem, and with the user_xattr mount option it
5312 is possible to allow users to store EAs on their files also
5314 Severity : enhancement
5316 Description: Add possibility (config option) to show minimal available OST free
5318 Details : When compiled with --enable-mindf configure option, statfs(2)
5319 (and so, df) will return least minimal free space available from
5320 all OSTs as amount of free space on FS, instead of summary of
5321 free spaces of all OSTs.
5323 Severity : enhancement
5325 Description: do not expand extent locks acquired on OST-side
5326 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
5327 by server: they are not cached anyway.
5330 Frequency : when mmap is used/binaries executed from Lustre
5332 Description: Unmmap pages before throwing them away from read cache.
5333 Details : llap_shrink cache now attempts to unmap pages before discarding
5334 them (if unmapping failed - do not discard). SLES9 kernel has
5335 extra checks that trigger if this unmapping is not done first.
5340 Description: lconf didn't resolve symlinks before checking to see whether a
5341 given mountpoint was already in use
5344 Frequency : when migrating failover services
5345 Bugzilla : 6395, 9514
5346 Description: When migrating a subset of services from a node (e.g. failback
5347 from a failover service node) the remaining services would
5348 time out and evict clients.
5349 Details : lconf --force (implied by --failover) sets the global obd_timeout
5350 to 5 seconds in order to quickly disconnect, but this caused
5351 other RPCs to time out too quickly. Do not change the global
5352 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
5354 Severity : enhancement
5355 Frequency : if MDS is started with down OST
5356 Bugzilla : 9439,5706
5357 Description: Allow startup/shutdown of an MDS without depending on the
5358 availability of the OSTs.
5359 Details : Asynchronously call mds_lov_synchronize during MDS startup.
5360 Add appropriate locking and lov-osc refcounts for safe
5361 cleaning. Add osc abort_inflight calls in case the
5362 synchronize never started.
5365 Frequency : occasional (Cray XT3 only)
5367 Description: root not authorized to access files in CRAY_PORTALS environment
5368 Details : The client process capabilities were not honoured on the MDS in
5369 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
5370 been accessed by an authorized user then root was able to access
5371 the file on the local client also. The root user capabilities
5372 are now allowed on the MDS, as this environment has secure UID.
5375 Frequency : occasional
5377 Description: ldiskfs "too long searching" message happens too often
5378 Details : A debugging message (otherwise harmless) prints too often on
5379 the OST console. This has been reduced to only happen when
5380 there are fragmentation problems on the filesystem.
5385 Description: Division by zero in statfs when all OSCs are inactive
5386 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
5387 lov_statfs divides by value returned by lov_get_stripecnt().
5391 Bugzilla : 9489, 3273
5392 Description: First write from each client to each OST was only 4kB in size,
5393 to initialize client writeback cache, which caused sub-optimal
5394 RPCs and poor layout on disk for the first writen file.
5395 Details : Clients now request an initial cache grant at (re)connect time
5396 and so that they can start streaming writes to the cache right
5397 away and always do full-sized RPCs if there is enough data.
5398 If the OST is rebooted the client also re-establishes its grant
5399 so that client cached writes will be honoured under the grant.
5404 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
5405 Details : Now I/O RPCs go to different portal number and (presumably) fast
5406 lock requests (and glimses) and other RPCs get their own service
5407 threads pool that should be able to service those RPCs
5410 Severity : enhancement
5412 Description: Ability to exchange lustre version between client and servers and
5413 issue warnings at client side if client is too old. Also for
5414 liblustre clients there is ability to refuse connection of too old
5416 Details : New 'version' field is added to connect data structure that is
5417 filled with version info. That info is later checked by server and
5421 Frequency : rare, liblustre only.
5422 Bugzilla : 9296, 9581
5423 Description: Two simultaneous writes from liblustre at offset within same page
5424 might proceed at the same time overwriting eachother with stale
5426 Details : I/O lock withing llu_file_prwv was released too early, before data
5427 actually was hitting the wire. Extended lock-holding time until
5428 server acknowledges receiving data.
5431 Frequency : extremely rare. Never observed in practice.
5433 Description: avoid generating lustre_handle cookie of 0.
5434 Details : class_handle_hash() generates handle cookies by incrementing
5435 global counter, and can hit 0 occasionaly (this is unlikely, but
5436 not impossible, because initial value of cookie counter is
5437 selected randonly). Value of 0 is used as a sentinel meaning
5438 "unassigned handle" --- avoid it. Also coalesce two critical
5439 sections in this function into one.
5441 Severity : enhancement
5443 Description: allow liblustre clients to delegate truncate locking to OST
5444 Details : To avoid overhead of locking, liblustre client instructs OST to
5445 take extent lock in ost_punch() on client's behalf. New connection
5446 flag is added to handle backward compatibility.
5448 Severity : enhancement
5449 Bugzilla : 4928, 7341, 9758
5450 Description: allow number of OST service threads to be specified
5451 Details : a module parameter allows the number of OST service threads
5452 to be specified via "options ost ost_num_threads={N}" in the
5453 OSS's /etc/modules.conf or /etc/modprobe.conf.
5457 Bugzilla : 6146, 9635, 9895
5458 Description: servers crash with bad pointer in target_handle_connect()
5459 Details : In rare cases when a client is reconnecting it was possible that
5460 the connection request was the last reference for that export.
5461 We would temporarily drop the export reference and get a new
5462 one, but this may have been the last reference and the export
5463 was just destroyed. Get new reference before dropping old one.
5465 Severity : enhancement
5466 Frequency : if client is started with failover MDS
5468 Description: Allow multiple MDS hostnames in the mount command
5469 Details : Try to read the configuration from all specified MDS
5470 hostnames during a client mount in case the "primary"
5473 Severity : enhancement
5475 Description: Stop sending data to evicted clients as soon as possible.
5476 Details : Check if the client we are about to send or are sending data to
5477 was evicted already. (Check is done every second of waiting,
5478 for which l_wait_event interface was extended to allow checking
5479 of exit condition at specified intervals).
5482 Frequency : rare, normally only when NFS exporting is done from client
5484 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
5486 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
5489 Frequency : always, if srand() is called before liblustre initialization
5491 Description: Liblustre uses system PRNG disturbing its usage by user application
5492 Details : Introduce internal to lustre fast and high-quality PRNG for
5493 lustre usage and make liblustre and some other places in generic
5494 lustre code to use it.
5496 Severity : enhancement
5497 Bugzilla : 9477, 9557, 9870
5498 Description: Verify that the MDS configuration logs are updated when xml is
5499 Details : Check if the .xml configuration logs are newer than the config
5500 logs stored on the MDS and report an error if this is the case.
5501 Request --write-conf, or allow starting with --old_conf.
5503 Severity : enhancement
5505 Description: Handle symlinks in the path when checking if Lustre is mounted.
5506 Details : Resolve intermediate symlinks when checking if a client has
5507 mounted a filesystem to avoid duplicate client mounts.
5512 Description: lconf can hit an error exception but still return success.
5513 Details : The lconf command catches the Command error exception at the top
5514 level script context and will exit with the associated exit
5515 status, but doesn't ensure that this exit status is non-zero.
5520 Description: failure of ptlrpc thread startup can cause oops
5521 Details : Starting a ptlrpc service thread can fail if there are a large
5522 number of threads or the server memory is very fragmented.
5523 Handle this without oopsing.
5526 Frequency : always, only if liblustre and non-default acceptor port was used
5528 Description: liblustre cannot connect to servers with non-default acceptor port
5529 Details : tcpnal_set_default_params() was not called and was therefore
5530 ignoring the environment varaible TCPNAL_PORT, as well as other
5531 TCPNAL_ environment variables
5536 Description: two objects could be created on the same OST for a single file
5537 Details : If an OST is down, in some cases it was possible to create two
5538 objects on a single OST for a single file. No problems other
5539 than potential performance impact and spurious error messages.
5543 Bugzilla : 5681, 9562
5544 Description: Client may oops in ll_unhash_aliases
5545 Details : Client dcache may become inconsistent in race condition.
5546 In some cases "getcwd" can fail if the current directory is
5552 Description: Inode refcounting problems in NFS export code
5553 Details : link_raw functions used to call d_instantiate without obtaining
5554 extra inode reference first.
5558 Bugzilla : 9942, 9903
5559 Description: Referencing freed requests leading to crash, memleaks with NFS.
5560 Details : We used to require that call to ll_revalidate_it was always
5561 followed by ll_lookup_it. Also with revalidate_special() it is
5562 possible to call ll_revalidate_it() twice for the same dentry
5563 even if first occurence returned success. This fix changes semantic
5564 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
5565 reference on a request referred from the intent.
5566 ll_intent_release() then releases such a request.
5569 Frequency : rare, normally benchmark loads only
5571 Description: unlinked inodes were kept in memory on the client
5572 Details : If a client is repeatedly creating and unlinking files it
5573 can accumulate a lot of stale inodes in the inode slab cache.
5574 If there is no other client load running this can cause the
5575 client node to run out of memory. Instead flush old inodes
5576 from client cache that have the same inode number as a new inode.
5579 Frequency : SLES9 2.6.5 kernel and long filenames only
5580 Bugzilla : 9969, 10379
5581 Description: utime reports stale NFS file handle
5582 Details : SLES9 uses out-of-dentry names in some cases, which confused
5583 the lustre dentry revalidation. Change it to always use the
5587 Frequency : rare, unless heavy write-truncate concurrency is continuous
5588 Bugzilla : 4180, 6984, 7171, 9963, 9331
5589 Description: OST becomes very slow and/or deadlocked during object unlink
5590 Details : filter_destroy() was holding onto the parent directory lock
5591 while truncating+unlinking objects. For very large objects this
5592 may block other threads for a long time and slow overall OST
5593 responsiveness. It may also be possible to get a lock ordering
5594 deadlock in this case, or run out of journal credits because of
5595 the combined truncate+unlink. Solution is to do object truncate
5596 first in one transaction without parent lock, and then do the
5597 final unlink in a new transaction with the parent lock. This
5598 reduces the lock hold time dramatically.
5601 Frequency : rare, 2.4 kernels only
5603 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
5604 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
5605 before unmounting their filesystem in order to avoid deadlock.
5606 The kernel_locked() function in 2.4 kernels only checks whether
5607 the kernel lock is held, not whether it is this process that is
5608 holding it as 2.6 kernels do.
5613 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
5614 Details : The client ptlrpc code may be trying to reconnect to a down
5615 server before a previous connection attempt has timed out.
5616 Increase the reconnect interval to be longer than the connection
5617 timeout interval to avoid sending duplicate connections to
5621 Frequency : echo_client brw_test command
5623 Description: fix echo_client to work with OST preallocated code
5624 Details : OST preallocation code (5137) didn't take echo_client IO path
5625 into account: echo_client calls filter methods outside of any
5626 OST thread and, hence, there is no per-thread preallocated
5627 pages and buffers to use. Solution: hijack pga pages for IO. As
5628 a byproduct, this avoids unnecessary data copying.
5632 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
5633 Description: Client can oops in mdc_commit_close() after open replay
5634 Details : It was possible for the MDS to return an open request with no
5635 transaction number in mds_finish_transno() if the client was
5636 evicted, but without actually returning an error. Clients
5637 would later try to replay that open and may trip an assertion
5638 Simplify the client close codepath, and always return an error
5639 from the MDS in case the open is not successful.
5642 Frequency : rare, 2.6 OSTs only
5644 Description: OST may deadlock under high load on fragmented files
5645 Details : If there was a heavy load and highly-fragmented OST filesystems
5646 it was possible to have all the OST threads deadlock waiting on
5647 allocation of biovecs, because the biovecs were not released
5648 until the entire RPC IO was completed. Instead, release biovecs
5649 as soon as they are complete to ensure forward IO progress.
5651 Severity : enhancement
5653 Description: Support for specifying external journal device at mount
5654 Details : If an OST or MDS device is formatted with an external journal
5655 device, this device major/minor is stored in the ext3 superblock
5656 and may not be valid for failover. Allow detecting and
5657 specifying the external journal at mount time.
5662 Description: Mounting an MDS with pending unlinked files may cause oops
5663 Details : target_finish_recovery() calls mds_postrecov() which returned
5664 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
5665 considers this an error and immediately begins cleaning up the
5666 lov, just after starting the mds_lov process
5668 Severity : enhancement
5670 Description: Implement 'lfs df' to report actual free space on per-OST basis
5671 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
5672 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
5673 usage of objects; '-h' to report in human readable format.
5675 ------------------------------------------------------------------------------
5677 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
5684 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
5685 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
5686 at startup during orphan unlinking. Instead, skip buddy update
5687 before setup, as it will be regenerated after recovery is complete.
5692 Description: If an OST is inactive, its locks might reference stale inodes.
5693 Details : lov_change_cbdata() must iterate over all namespaces, even if
5694 they are inactive to clear inode references from the lock.
5696 Severity : enhancement
5697 Frequency : occasional, if non-standard max_dirty_mb used
5699 Description: Client will block write RPCs if not enough grant
5700 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
5701 then the client will block writes while waiting for another RPC
5702 to complete instead of consuming its dirty limit. With change
5703 we get improved performance when max_dirty_mb is small.
5705 Severity : enhancement
5706 Bugzilla : 3389, 6253
5707 Description: Add support for supplementary groups on the MDS.
5708 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
5709 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
5710 lookups for user supplementary groups into a cache.
5714 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
5715 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
5716 for liblustre O_EXCL is enforced by MDS.
5720 Description: readdir never returns NULL in liblustre.
5721 Details : Corrected llu_iop_getdirentries logic, to return offset of next
5722 dentry in struct dirent.
5726 Frequency : liblustre only, depends on application IO pattern
5727 Description: liblustre clients evicted if not contacting servers
5728 Details : Don't put liblustre clients into the ping_evictor list, so
5729 they will not be evicted by the pinger ever.
5731 Severity : enhancement
5733 Description: Add ability to evict clients by NID from MDS.
5734 Details : By echoing "nid:$NID" string into
5735 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
5736 $NID would be instantly evicted from this MDS and from all active
5737 OSTs connected to it.
5741 Description: Do not query file size twice, somewhat slowing stat(2) calls.
5742 Details : lookup_it_finish() used to query file size from OSTs that was not
5747 Description: service threads change working directory to that of init
5748 Details : Starting lustre service threads may pin the working directory
5749 of the parent thread, making that filesystem busy. Threads
5750 now change to the working directory of init to avoid this.
5754 Frequency : during shutdown only
5755 Description: shutdown with a failed MDS or OST can cause unmount to hang
5756 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
5761 Frequency : 2.6 only
5762 Description: chmod/chown may include an extra supplementary group
5763 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
5764 supplementary group and if none is specified this is used.
5767 Bugzilla : 5479 (6816)
5769 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
5770 Details : If lookup is in progress on a file that is unlinked we might try
5771 to revalidate the inode and fail in revalidate after lookup is
5772 complete and ll_file_open() enqueues the open again but
5773 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
5776 Frequency : always, if lconf --abort_recovery used
5778 Description: lconf --abort_recovery fails with 'Operation not supported'
5779 Details : lconf was attempting to abort recovery on the MDT device and not
5782 ------------------------------------------------------------------------------
5784 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
5789 Frequency : rare (only unsupported configurations with a node running as an
5791 Bugzilla : 6514, 5137
5792 Description: Mounting a Lustre file system on a node running as an OST could
5794 Details : OSTs now preallocates memory needed to write out data at
5795 startup, instead of when needed, to avoid having to
5796 allocate memory in possibly low memory situations.
5797 Specifically, if the file system is mounted on on OST,
5798 memory pressure could force it to try to write out data,
5799 which it needed to allocate memory to do. Due to the low
5800 memory, it would be unable to do so and the node would
5801 become unresponsive.
5803 Severity : enhancement
5805 Description: Addition of lconf --service command line option
5806 Details : lconf now accepts a '--service <arg>' option, which is
5807 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
5809 Severity : enhancement
5811 Description: Failover mode is now the default for OSTs.
5812 Details : By default, OSTs will now run in failover mode. To return to
5813 the old behaviour, add '--failout' to the lmc line for OSTs.
5815 Severity : enhancement
5817 Description: Health checks are now provided for MDS and OSTs
5818 Details : Additional detailed health check information on MSD and OSTs
5819 is now provided through the procfs health_check value.
5822 Frequency : occasional, depends on IO load
5824 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
5825 after numerous create/delete cycles
5826 Details : The ext3 inode allocation policy would not allocate new inodes
5827 very well on the OSTs because there are no new directories
5828 being created. Instead we look for groups with free space if
5829 the parent directories are nearly full.
5834 Description: Network or server problems during mount may cause partially
5835 mounted clients instead of returning an error.
5836 Details : The config llog parsing code may overwrite the error return
5837 code during mount error handling, returning success instead
5843 Description: MDS can fail to allocate large reply buffers
5844 Details : After long uptimes the MDS can fail to allocate large reply
5845 buffers (e.g. zconf client mount config records) due to memory
5846 fragmentation or consumption by the buffer cache. Preallocate
5847 some large reply buffers so that these replies can be sent even
5848 under memory pressure.
5852 Frequency : rare (liblustre)
5853 Description: fsx running with liblustre complained that using truncate() to
5854 extend the file doesn't work. This patch corrects that issue.
5855 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
5856 ATTR_SIZE and lsm use in llu_setattr_raw.
5860 Frequency : rare, only 2.6 kernels
5861 Description: Unusual file access patterns on the MDS may result in inode
5862 data being lost in very rare circumstances.
5863 Details : Bad interaction between the ea-in-inode patch and the "no-read"
5864 code in the 2.6 kernel caused the inode and/or EA data not to
5865 be read from disk, causing single-file corruption.
5869 Frequency : rare, only 2.6 filesystems using extents
5870 Description: Heavy concurrent write and delete load may cause data corruption.
5871 Details : It was possible under high-load situations to have an extent
5872 metadata block in the block device cache from a just-unlinked
5873 file overwrite a newly-allocated data block. We now unmap any
5874 metadata buffers that alias just-allocated data blocks.
5878 Frequency : filesystems with default stripe_count larger than 77
5879 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
5880 Details : lconf specifies an inode size of 4096 bytes when the default
5881 stripe_count is larger than 77. This conflicts with the default
5882 inode density of 1 per 4096 bytes. Allocate smaller inodes in
5883 this case to avoid pinning too much memory for large EAs.
5885 ------------------------------------------------------------------------------
5887 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
5892 Frequency : rare (extremely heavy IO load with hundreds of clients)
5894 Description: Client is evicted, gets IO error writing to file
5895 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
5896 caused clients to be evicted for AST timeouts. The fixes in
5897 bug 5192 mean we no longer need to have such short AST timeouts
5898 so ldlm_timeout has been increased.
5901 Frequency : occasional during --force or --failover shutdown under load
5902 Bugzilla : 5949, 4834
5903 Description: Server oops/LBUG if stopped with --force or --failover under load
5904 Details : a collection of import/export refcount and cleanup ordering
5905 issues fixed for safer force cleanup
5908 Frequency : only filesystems larger than 120 OSTs
5909 Bugzilla : 5990, 6223
5910 Description: lfs getstripe would oops on a very large filesystem
5911 Details : lov_getconfig used kfree on vmalloc'd memory
5914 Frequency : only filesystems exporting via NFS to Solaris 10 clients
5915 Bugzilla : 6242, 6243
5916 Description: reading from files that had been truncated to a non-zero size
5917 but never opened returned no data
5918 Details : ll_file_read() reads zeros from no-object files to EOF
5923 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
5924 Details : There's little harm in aborting MDS/OSS recovery and letting it
5925 try again, so I removed the LASSERT and return an error instead.
5927 Severity : enhancement
5929 Description: New debugging infrastructure for tracking down data corruption
5930 Details : The I/O checksum code was replaced to: (a) control it at runtime,
5931 (b) cover more of the client-side code path, and (c) try to narrow
5932 down where problems occurred
5936 Bugzilla : 3819, 4364, 4397, 6313
5937 Description: Racing close and eviction MDS could cause assertion in mds_close
5938 Details : It was possible to get multiple mfd references during close and
5939 client eviction, leading to one thread referencing a freed mfd.
5941 Severity: : enhancement
5942 Bugzilla : 3262, 6359
5943 Description: Attempts to reconnect to servers are now more aggressive.
5944 Details : This builds on the enhanced upcall-less recovery that was added
5945 in 1.4.2. When trying to reconnect to servers, clients will
5946 now try each server in the failover group every 10 seconds. By
5947 default, clients would previously try one server every 25 seconds.
5952 Description: After recovery, certain operations trigger a failed
5953 assertion on a client.
5954 Details : Failing over an mds, using lconf -d --failover, while a
5955 client was doing a readdir() call would cause the client to
5956 LBUG after recovery completed and the readdir() was resent.
5958 Severity : enhancement
5960 Description: Default groups are now added by lconf
5961 Details : You can now run lconf --group <servicename> without having to
5962 manually add groups with lmc.
5965 Frequency : occasional
5967 Description: Nodes with an elan id of 0 trigger a failed assertion
5970 Frequency : always when accessing e.g. tty/console device nodes
5972 Description: tty and some other devices nodes cannot be used on lustre
5973 Details : file's private_data field is used by device data and lustre
5974 values in there got lost. New field was added to struct file to
5975 store fs-specific private data.
5978 Frequency : when exporting Lustre via NFS
5980 Description: NFSD failed occasionally when looking up a path component
5981 Details : NFSD is looking up ".." which was broken in ext3 directories
5982 that had grown large enough to become hashed.
5985 Frequency : Clusters with multiple interfaces not on the same subnet
5987 Description: Nodes will repeatedly try to reconnect to an interface which it
5988 cannot reach and report an error to the log.
5989 Details : Extra peer list entries will be created by lconf with some peers
5990 unreachable. lconf now validates the peer before adding it.
5993 Frequency : Only if a default stripe is set on the filesystem root.
5995 Description: Setting a default stripe on the filesystem root prevented the
5996 filesystem from being remounted.
5997 Details : The client was sending extra request flags in the root getattr
5998 request and did not allocate a reply buffer for the dir EA.
6001 Frequency : occasional, higher if lots of files are accessed by one client
6002 Bugzilla : 6159, 6097
6003 Description: Client trips assertion regarding lsm mismatch/magic
6004 Details : While revalidating inodes the VFS looks up inodes with ifind()
6005 and in rare cases can find an inode that is being freed.
6006 The ll_test_inode() code will free the lsm during ifind()
6007 when it finds an existing inode and then the VFS later attaches
6008 this free lsm to a new inode.
6012 Bugzilla : 6422, 7030
6013 Description: MDS deadlock between mkdir and client eviction
6014 Details : Creating a new file via mkdir or mknod (starting a transaction
6015 and getting the ns lock) can deadlock with client eviction
6016 (gets ns lock and trying to finish a synchronous transaction).
6019 Frequency : occasional
6020 Description: While starting a server, the fsfilt_ext3 module could not be
6022 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
6023 kernels. Previously, lconf would still use the ext3 name
6024 when trying to load modules. Now, it will correctly use
6025 ext3 on 2.4 and ldiskfs on 2.6.
6027 Severity : enhancement
6028 Description: The default stripe count has been changed to 1
6029 Details : The interpretation of the default stripe count (0, to lfs
6030 or lmc) has been changed to mean striping across a single
6031 OST, rather than all available. For general usage we have
6032 found a stripe count of 1 or 2 works best.
6034 Severity : enhancement
6035 Description: Add support for compiling against Cray portals.
6036 Details : Conditional compiling for some areas that are different
6040 Frequency : occasional
6041 Bugzilla : 6409, 6834
6042 Description: Creating files with an explicit stripe count may lead to
6043 a failed assertion on the MDS
6044 Details : If some OSTs are full or unavailable, creating files may
6045 trigger a failed assertion on the MDS. Now, Lustre will
6046 try to use other servers or return an error to the
6050 Frequency : occasional
6052 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
6053 caused lock timeout during readahead (since 1.4.2).
6054 Details : Processes doing readahead might match a lock that hasn't been
6055 granted yet if there are overlapping and conflicting lock
6056 requests. The readahead process waits on ungranted lock
6057 (original lock is CBPENDING), while OST waits for that process
6058 to cancel CBPENDING read lock and eventually evicts client.
6060 Severity : enhancement
6062 Description: Initial enabling of flock support for clients
6063 Details : Implements fcntl advisory locking and file status functions.
6064 This feature is provided as an optional mount flag (default
6065 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
6066 locking are implemented yet, and those that are are not guaranteed
6067 to be completely correct in production environments.
6068 mount -t lustre -o [flock|noflock] ...
6071 Frequency : occasional
6073 Description: OSTs running 2.4 kernels but with extents enabled might trip an
6074 assertion in the ext3 JBD (journaling) layer.
6075 Details : The b_committed_data struct is protected by the big kernel lock
6076 in 2.4 kernels, serializing journal_commit_transaction() and
6077 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
6078 access to this struct. In 2.6 kernels there is finer grained
6079 locking to improve SMP performance of the JBD layer.
6083 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
6085 -----------------------------------------------------------------------------
6087 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
6089 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
6090 versions of Lustre. Please update all servers and clients to
6091 version 1.4.2 or later at the same time. You must also run
6092 "lconf --write-conf {config}.xml" on the MDS while it is stopped
6093 to update the configuration logs.
6095 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
6096 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
6097 - swab llog records of type '0' so we get proper header size/idx (5861)
6098 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
6099 - fix rename of one directory over another leaking an inode (5953)
6100 - avoid SetPageDirty on 2.6 (5981)
6101 - don't re-add just-being-destroyed locks to the waiting list (5653)
6102 - when creating new directories, inherit the parent's custom
6103 striping settings if present parent (3048)
6104 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
6105 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
6106 - don't hold client locks on temporary worklist from l_lru (5666)
6107 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
6108 - automatically evict dead clients (5921)
6109 - Update file size properly in create+truncate+fstat case (6196)
6110 - Do not unhash mountpoint dentries, do not allow removal of
6112 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
6113 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
6114 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
6115 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
6116 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
6117 - Correctly handle reads of files with no objects (6243)
6118 - lctl recover will also mark a device active if deactivate used (5933)
6120 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
6121 - allow --write-conf on an MDS with different nettype than client (5619)
6122 - don't write config llogs to MDS for mounts not from that MDS (5617)
6123 - lconf should create multiple TCP connections from a client (5201)
6124 - init scripts are now turned off by default; run chkconfig --on
6125 lustre and chkconfig --on lustrefs to use them
6126 - upcalls are no longer needed for clients to recover to failover
6128 - add --abort-recovery option to lconf to abort recovery on device
6130 - add support for an arbitrary number of OSTs (3026)
6131 - Quota support protocol changes.
6132 - forward compatibility changes to wire structs (6007)
6133 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
6134 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
6135 - improved "lustre status" script
6136 - initialize blocksize for non-regular files (6062)
6137 - added --disable-server and --disable-client configure options (5782)
6138 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
6139 - Vanilla 2.4.29 support
6140 - increase maximum number of obd devices to 520 (6242)
6141 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
6142 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
6143 SLES 9 2.6.5 kernels
6144 - limit stripes per file to 160 (the maximum EA size) (6093)
6146 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
6149 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
6150 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
6151 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
6152 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
6153 - small ext3 extents cleanups and fixes (5733)
6154 - improved mballoc code, several small races and bugs fixed (5733, 5638)
6155 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
6156 - avoid needless client->OST connect, fix handle mismatch (5317)
6157 - fix DLM error path that led to out-of-sync client, long delays (5779)
6158 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
6159 - fix several locking issues related to i_size (5492,5624,5654,5672)
6160 - don't move pending lock onto export if it is already evicted (5683)
6161 - fix kernel oops when creating .foo in unlinked directory (5548)
6162 - fix deadlock in obdfilter statistics vs. object create (5811)
6163 - use time_{before,after} to avoid timer jiffies wrap (5882)
6164 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
6165 - Do not leak request if server was not able to process it (5154)
6166 - If mds_open unable to find parent dir, make that negative lookup(5154)
6167 - don't create new directories with extent-mapping (5909, 5936)
6169 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
6170 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
6171 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
6172 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
6173 - support for RHEL4 kernel on i686 (5773)
6174 - provide error messages when incompatible logs are encountered (5898)
6176 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
6177 * version 1.4.0.10 (1.4.1 release candidate 1)
6179 - don't keep a lock reference when lock is not granted (4238)
6180 - unsafe list practices (rarely) led to infinite eviction loop (4908)
6181 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
6182 - drop import inflight refcount on signal_completed_replay error (5255)
6183 - unlock page after async write error during send (3677)
6184 - handle missing objects in filter_preprw_read properly (5265)
6185 - no transno return for symlink open, don't save no-trasno open (3440)
6186 - don't try to complete elan receive that already failed (4012)
6187 - free RPC server reply state on error (5406)
6188 - clean up thread from ptlrpc_start_thread() on error (5160)
6189 - readahead could read extra page into cache that wasn't ejected (5388)
6190 - prevent races in class_attach/setup/cleanup/detach (5260)
6191 - don't dereference de->d_inode after l_dput of de (5458)
6192 - use "int" for stripe value returned from lock_to_stripe (5544)
6193 - mballoc allocation and error-checking fixes in 2.6 (5504)
6194 - block device patches to fix I/O request sizes in 2.6 (5482)
6195 - look up hostnames for IB nals (5602)
6196 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
6197 - don't start multiple acceptors for the same port (5277)
6198 - fix incorrect LASSERT in mds_getattr_name (5635)
6199 - export a proc file for general "ping" checking (5628)
6200 - fix "lfs check" to not block when the MDS is down (5628)
6202 - service request history (4965)
6203 - put {ll,lov,osc}_async_page structs in a single slab (4699)
6204 - create an "evict_client" /proc entry on OSTs, like the MDS has
6205 - fix mount usage message, return errors per mount(8) (5168)
6206 - change grep [] to grep "[]" in tests so they work in more UMLs
6207 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
6208 - remove ancient LOV_MAGIC_V0 EA support (5047)
6209 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
6210 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
6211 - allow readahead allocations to fail when low on memory (5383)
6212 - mmap locking landed again, after considerable improvement (2828)
6213 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
6215 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
6218 - send OST transaction number in read/write reply to free req (4966)
6219 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
6220 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
6221 - the watchdog thread now runs as interruptible (5246)
6222 - flock/lockf fixes (but it's still disabled, pending 5135)
6223 - don't use EXT3 constants in llite code (5094)
6224 - memory shortage at startup could cause assertion (5176)
6226 - reorganization of lov code
6227 - single portals codebase
6229 - add extents/mballoc support (5025)
6230 - direct I/O reads in the obdfilter (4048)
6231 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
6233 tbd Cluster File Systems, Inc. <info@clusterfs.com>
6236 - send OST transaction number in read/write reply to free req (4966)
6237 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
6238 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
6239 - the watchdog thread now runs as interruptible (5246)
6240 - handle missing objects in filter_preprw_read properly (5265)
6241 - unsafe list practices (rarely) led to infinite eviction loop (4908)
6242 - drop import inflight refcount on signal_completed_replay error (5255)
6243 - unlock page after async write error during send (3677)
6244 - return original error code on reconstructed replies (3761)
6245 - no transno return for symlink open, don't save no-trasno open (3440)
6247 - add pid to ldlm debugging output (4922)
6248 - bump the watchdog timeouts -- we can't handle 30sec yet
6249 - extra debugging for orphan dentry/inode bug (5259)
6251 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
6254 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
6255 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
6256 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
6257 - fix NULL dereference in /proc/sys/portals/routes (4827)
6258 - allow failed mdc_close() operations to be interrupted (4561)
6259 - stop precreate on OST before MDS would time out on it (4778)
6260 - don't send partial-page writes before EOF from client (4410)
6261 - discard client grant for sub-page writes on large-page clients (4520)
6262 - don't free dentries not owned by NFS code, check generation (4806)
6263 - fix lsm leak if mds_create_objects() fails (4801)
6264 - limit debug_daemon file size, always print CERROR messages (4789)
6265 - use transno after validating reply (3892)
6266 - process timed out requests if import state changes (3754)
6267 - update mtime on OST during writes, return in glimpse (4829)
6268 - add mkfsoptions to LDAP (4679)
6269 - use ->max_readahead method instead of zapping global ra (5039)
6270 - don't interrupt __l_wait_event() during strace
6272 - add software watchdogs to catch hung threads quickly (4941)
6273 - make lustrefs init script start after nfs is mounted
6274 - fix CWARN/ERROR duplication (4930)
6275 - return async write errors to application if possible (2248)
6276 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
6277 - print NAL number in %x format (4645)
6278 - update barely-supported suse-2.4.21-171 series (4842)
6279 - support for sles 9 %post scripts
6280 - support for building 2.6 kernel-source packages
6281 - support for sles km_* packages
6283 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
6286 - ignore -ENOENT errors in osc_destroy (3639)
6287 - notify osc create thread that OSC is being cleaned up (4600)
6288 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
6289 - reconstruct ost_handle() like mds_handle() (4657)
6290 - create a new thread to do import eviction to avoid deadlock (3969)
6291 - let lconf resolve symlinked-to devices (4629)
6292 - don't unlink "objects" from directory with default EA (4554)
6293 - hold socknal file ref over connect in case target is down (4394)
6294 - allow more than 32000 subdirectories in a single directory (3244)
6295 - fix blocks count for O_DIRECT writes (3751)
6296 - OST returns ENOSPC from object create when no space left (4539)
6297 - don't send truncate RPC if file size isn't changing (4410)
6298 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
6299 - bind to privileged port in socknal and tcpnal (3689)
6301 - rate limit CERROR/CWARN console message to avoid overload (4519)
6302 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
6303 - basic mmap support (3918)
6304 - kernel patch series update from b1_4 (4711)
6306 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
6309 - avoid crash during MDS cleanup with OST shut down (2775)
6310 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
6311 - don't use bad inodes on the MDS (3744)
6312 - dynamic object preallocation to improve recovery speed (4236)
6313 - don't hold spinlock over lock dumping or change debug flags (4401)
6314 - don't zero obd_dev when it is force cleaned (3651)
6315 - print grants to console if they go negative (4431)
6316 - "lctl deactivate" will stop automatic recovery attempts (3406)
6317 - look for existing locks in ldlm_handle_enqueue() (3764)
6318 - don't resolve lock handle twice in recovery avoiding race (4401)
6319 - revalidate should check working dir is a directory (4134)
6321 - don't always mark "slow" obdfilter messages as errors (4418)
6323 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
6326 - don't close LustreDB during write_conf until it is done (3860)
6327 - fix typo in lconf for_each_profile (3821)
6328 - allow dumping logs from multiple threads at one time (3820)
6329 - don't allow multiple threads in OSC recovery (3812)
6330 - fix debug_size parameters (3864)
6331 - fix mds_postrecov to initialize import for llog ctxt (3121)
6332 - replace config semaphore with spinlock (3306)
6333 - be sure to send a reply for a CANCEL rpc with bad export (3863)
6334 - don't allow enqueue to complete on a destroyed export (3822)
6335 - down write_lock before checking llog header bitmap (3825)
6336 - recover from lock replay timeout (3764)
6337 - up llog sem before sending rpc (3652)
6338 - reduce ns lock hold times when setting kms (3267)
6339 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
6340 - fix NULL deref and obd_dev leak on setup error (3312)
6341 - replace some LBUG about llog ops with error handling (3841)
6342 - don't match INVALID dentries from d_lookup and spin (3784)
6343 - hold dcache_lock while marking dentries INVALID and hashing (4255)
6344 - fix invalid assertion in ptlrpc_set_wait (3880)
6346 - add libwrap support for the TCP acceptor (3996)
6347 - add /proc/sys/portals/routes for non-root route listing (3994)
6348 - allow setting MDS UUID in .xml (2580)
6349 - print the stack of a process that LBUGs (4228)
6351 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
6354 - don't cleanup request in ll_file_open() on failed MDS open (3430)
6355 - make sure to unset replay flag from failed open requests (3440)
6356 - if default stripe count is 0, use OST count for inode size (3636)
6357 - update parent mtime/ctime on client for create/unlink (2611)
6358 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
6359 - free recovery state on server during a forced cleanup (3571)
6360 - unregister_reply for resent reqs (3063)
6361 - loop back devices mounting and status check on 2.6 (3563)
6362 - fix resource-creation race that can provoke i_size == 0 (3513)
6363 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
6364 - more debugging for page-accounting assertion (3746)
6365 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
6366 - avoid deadlock after precreation failure (3758)
6367 - fix race and lock order deadlock in orphan handling (3450, 3750)
6368 - add validity checks when grabbing inodes from l_ast_data (3599)
6370 - add /proc/.../recovery_status to obdfilter (3428)
6371 - lightweight CDEBUG infrastructure, debug daemon (3668)
6372 - change default OSC RPC parameters to be better on small clusters
6373 - turn off OST read cache for files smaller than 32MB
6374 - install man pages and include them in rpms (3100)
6375 - add new init script for (un)mounting lustre filesystems (2593)
6376 - run chkconfig in %post for init scripts (3701)
6377 - drop scimac NAL (unmaintained)
6379 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
6382 - clean kiobufs before and after use (3485)
6383 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
6384 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
6385 - change init script to fail more clearly if not run as root (1528)
6386 - allow clients to reconnect during replay (1742)
6387 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
6388 - don't do DNS lookups on NIDs too small for IP addresses (3442)
6389 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
6390 - fix cond_resched (3554)
6391 - only evict unfinished clients after recovery (3515)
6392 - allow bulk resend, prevent data loss (3570)
6393 - dynamic ptlrpc request buffer allocation (2102)
6394 - don't allow unlinking open directory if it isn't empty (2904)
6395 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
6396 - remove extraneous obd dereference causing LASSERT failure (3334)
6397 - don't use get_cycles() when creating temp. files on the mds (3156)
6398 - hold i_sem when setting i_size in ll_extent_lock() (3564)
6399 - handle EEXIST for set-stripe, set proper directory name (3336)
6401 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
6402 - fix ksocknal_fmb_callback() error messages (2918)
6404 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
6407 - don't copy lvb into (possibly NULL) reply on error (2983)
6408 - don't deref dentry after dput, don't free lvb on error (2922)
6409 - use the kms to determine writeback rpc length (2947)
6410 - increment oti_logcookies when osc is inactive (2948)
6411 - update client's i_blocks count via lvb messages (2543)
6412 - handle intent open/close of special files properly (1557)
6413 - mount MDS with errors=remount-ro, like obdfilter (2009)
6414 - initialize lock handle to avoid ASSERT on error cleanup (3057)
6415 - don't use cancelling-locks' kms values (2947)
6416 - use highest lock extent for kms, not last one (2925)
6417 - don't dereference ERR_PTR() dentry in error handling path (3107)
6418 - fix thread race in portals_debug_dumplog() (3122)
6419 - create lprocfs device entries at setup instead of at attach (1519)
6420 - common AST error handler, don't evict client on completion race (3145)
6421 - zero nameidata in detach_mnt in 2.6 (3118)
6422 - verify d_inode after revalidate_special is valid in 2.6 (3116)
6423 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
6424 - initialize RPC timeout timer earlier for 2.6 (3219)
6425 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
6426 - print nal/nid for unknown nid (3258)
6427 - additional checks for oscc recovery before doing precreate (3284)
6428 - fix ll_extent_lock() error return code for 64-bit systems (3043)
6429 - don't crash in mdc_close for bad permissions on open (3285)
6430 - zero i_rdev for non-device files (3147)
6431 - clear page->private before handing to FS, better assertion (3119)
6432 - tune the read pipeline (3236)
6433 - fix incorrect decref of invalidated dentry (2350)
6434 - provide read-ahead stats and refine rpc in flight stats (3328)
6435 - don't hold journal transaction open across create RPC (3313)
6436 - update atime on MDS at close time (3265)
6437 - close LDAP connection when recovering to avoid server load (3315)
6438 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
6439 - don't leak open file on MDS after open resend (3325)
6440 - serialize filter_precreate and filter_destroy_precreated (3329)
6441 - loop device shouldn't call sync_dev() for nul device (3092)
6442 - clear page cache after eviction (2766)
6443 - resynchronize MDS->OST in background (2824)
6444 - refuse to mount the same filesystem twice on same mountpoint (3394)
6445 - allow llmount to create routes for mounting behind routers (3320)
6446 - push lock cancellation to blocking thread for glimpse ASTs (3409)
6447 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
6448 - fix rare problem with rename on htree directories (3417)
6450 - allow default OST striping configuration per directory (1414)
6451 - fix compilation for qswnal for 2.6 kernels (3125)
6452 - increase maximum number of MDS request buffers for large systems
6453 - change liblustreapi to be useful for external progs like lfsck (3098)
6454 - increase local configuration timeout for slow disks (3353)
6455 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
6457 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
6460 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
6461 - don't overwrite extent policy data in reply if lock was blocked (2901)
6462 - drop filter export grants atomically with removal from device (2663)
6463 - del obd_self_export from work_list in class_disconnect_exports (2908)
6464 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
6465 - swab reply message in mdc_close, other PPC fixes (2464)
6466 - fix destroying of named logs (2325)
6467 - overwrite old logs when running lconf --write_conf (2264)
6468 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
6469 - fix race in target_handle_connect (2898)
6470 - mds_reint_create() should take same inode create lock (2926)
6471 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
6472 - don't close files for self_export to avoid uninitialized obd (2936)
6473 - allow MDS with the same name as client node (2939)
6474 - hold dentry reference for closed log files for unlink (2325)
6475 - reserve space for all logs during transactions (2059)
6476 - don't evict page beyond end of stripe extent (2925)
6477 - don't oops on a deleted current working directory (2399)
6478 - handle hard links to targets without a parent properly (2517)
6479 - don't dereference NULL lock when racing during eviction (2867)
6480 - don't grow lock extents when lots of conflicting locks (2919)
6482 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
6485 - account for cache space usage on clients to avoid data loss (974)
6486 - lfsck support in lustre kernel code (2349)
6487 - reduce journal credits needed for BRW writes (2370)
6488 - orphan handling to avoid losing space on client/server crashes
6489 - ptlrpcd can be blocked, stopping ALL progress (2477)
6490 - use lock value blocks to assist in proper KMS, faster stat (1021)
6491 - takes i_sem instead of DLM locks internally on obdfilter (2720)
6492 - recovery for initial connections (2355)
6493 - fixes for mds_cleanup_orphans (1934)
6494 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
6495 - block all file creations until orphan recovery completes (1901)
6496 - client remove rq_connection from request struct (2423)
6497 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
6498 - recovery timer race (2670)
6499 - mdc_close recovey bug (2532)
6500 - ptlrpc cleanup bug (2710)
6501 - mds timeout on local locks (2588)
6502 - namespace lock held during RPCs (2431)
6503 - handle interrupted sync write properly (2503)
6504 - don't try to handle a message that hasn't been replied to (2699)
6505 - client assert failure during cleanup after abort recovery (2701)
6506 - leak mdc device after failed mount (2712)
6507 - ptlrpc_check_set allows timedout requests to complete (2714)
6508 - wait for inflight reqs when ptlrpcd finishes (2710)
6509 - make sure unregistered services are removed from the srv_list
6510 - reset bulk XID's when resending them (caught by 1138 test)
6511 - unregister_bulk after timeout
6512 - fix lconf error (2694)
6513 - handle write after unfinished setstripe, stripe-only getstripe (2388)
6514 - readahead locks pages, leaves pending causing memory pressure (2673)
6515 - increase OST request buffers to 4096 on large machines (2729)
6516 - fix up permission of existing directories in simple_mkdir (2661)
6517 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
6518 - don't assign transno to errored transactions (2742)
6519 - don't delete objects on OST if given a bogus objid from MDS (2751)
6520 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
6521 - if rq_no_resend, then timeout request after recovery (2432)
6522 - fix MDS llog_logid record size, 64-bit array alignment (2733)
6523 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
6524 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
6525 - MDS recovery shouldn't delete objects that it has given out (2730)
6526 - if enqueue arrives after completion, don't clobber LVB (2819)
6527 - don't unlock pages twice when trigger_group_io returns error (2814)
6528 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
6529 - don't write pages to disk if there was an error (1450)
6530 - don't ping imports that have recovery disabled (2676)
6531 - take buffered bytes into account when balancing socknal conn (2817)
6532 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
6533 - reconnect unlink llog connection after MDS reconnects to OST (2816)
6534 - remove little-endian swabbing of llog records (1987)
6535 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
6536 - retry reposting request buffers if they fail (1191)
6537 - grow extent at grant time to avoid granting a revoked lock (2809)
6538 - lock revoke doesn't evict page if covered by a second lock (2765)
6539 - disable VM readahead to avoid reading outside lock extents (2805)
6541 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
6542 - updated kernel patches for hp-2.4.20 kernel (2681)
6544 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
6547 - fix truncated write corruption (2366)
6548 - fix for failed assertion in iopen_connect_dentry (1792,2517)
6550 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
6551 - ensure len(uuid) < 37 in lmc (1171)
6552 - fix ia64 OOPS in llog_test (2255)
6553 - zero end of page at obdfilter for partial page writes (2648)
6554 - don't leave stale dentries around after renames (bug 2428)
6555 - fix timeouts when evicting a client with a single lock held (2642)
6556 - set deadline for the initial HELLO message to drain (2634)
6557 - print out dotted-quad IP addresses in the socknal (2302)
6559 - additional debugging for MDS client eviction problem (2443)
6560 - fix mkfsoptions support for osts (2603, 2604)
6562 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
6565 - add series for the vanilla 2.6.0 kernel
6566 - add series for the vanilla 2.4.24 kernel
6567 - add series for a cray x86/64 UL kernel drop
6568 - fix xattr patches for the vanilla 2.4.19 series
6570 - generate true UUIDs in lmc (1171)
6571 - have portals stack dumping break in UML (2466)
6572 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
6573 - allocate with _NOFS in ldlm to avoid deadlock (1933)
6574 - wake callback waiting threads on client eviction (2460)
6575 - Add --ptldebug and --subsystem to lmc (1719)
6576 - update assertion to allow safe interrupt allocation
6577 - set rq_no_resend for cancel requests (2432)
6578 - recalculate ptlrpcd timeout after resend (2494)
6579 - call vfs_rmdir when removing pending directories (2368)
6580 - fix renaming a file to itself (2429)
6581 - lmc creates a default one-stripe lov (2454)
6582 - expand procfs space to handle large clusters (2326)
6583 - increase UML stack to avoid overflow
6584 - update lconf's list of debug and subsystem masks
6585 - fix lfs find --obd (2510)
6586 - /proc tunable for disabling filter read caching (2591)
6587 - stop rpm packages from altering slapd.conf (2301)
6588 - disable nagle in the socknal under 0conf (2578)
6589 - choose mds inode size based on stripe count (2572)
6590 - fix kernel-source rpm problems (2516)
6592 - add --disable-doc to avoid pdf generation (2421)
6593 - update documentation, tests, type-os, comments
6594 - avoid format warnings on ia64
6595 - remove the TOE NAL
6596 - tiny code cleanups by removing unused fields
6598 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
6601 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
6602 - remove the most blatant lies from BUILDING file (2371)
6603 - change default debug level to reasonable production setting
6604 - reduce client side cache size to reduce cache flush time
6605 - reduce max RPCs in flight to avoid unnecessary file fragmentation
6606 - make TCP zerocopy and pinger support enabled by default (2476)
6607 - sync writes completed after process exits caused crashes (2319)
6608 - maintain correct mount count on the MDS (2356)
6609 - backout 1557, because 2316 wasn't really fixed
6610 - better file I/O statistics gathering in /proc
6611 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
6612 - another kernel patch to fix zero-copy TCP function export
6613 - don't take duplicate lock when processing re-sent getattr (2420)
6614 - lctl uses obd_self_export instead of creating new conn (2353)
6615 - MDS/OST recovery case which requires object creation asserted (2425)
6616 - move lfs from /usr/sbin to /usr/bin in packages
6617 - fix race between mds_client_add and mds_client_free (2417)
6618 - use kmalloc instead of slabs in portals (2430)
6619 - don't create duplicate records when a failover MDS is present (2442)
6620 - remove unnecessary mount age check (2332)
6621 - don't remove directory inodes from locks prematurely (2451)
6622 - don't break if MDS service name is the same as hostname (2103)
6623 - fix races in client write RPC generation when cache full (2482)
6625 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
6628 - remove now-unused request->rq_obd (278)
6629 - if an allocation fails, print out how much memory we've used (1933)
6630 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
6631 - add the "configurable stack size" patch to most series files (1256)
6632 - ability to write large log records, for 100+ OST configs (2306)
6633 - fix NULL deref when filter_prep fails (2314)
6634 - fix operator precedence error in filter_sync
6635 - dynamic allocation of socknal TX descriptors (2315)
6636 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
6637 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
6638 - ensure that CWARN messages in Portals always get to the syslog
6639 - __init/__exit are not for prototype decls (ldlm_init/exit)
6640 - x86-64 compile warning fixes
6641 - fix gateway LMC keyword conflict (2318)
6642 - fix MDS lock inversions in getattr/reint paths (1844)
6643 - fix a rare lock re-ordering bug, which caused deadlock (2322)
6644 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
6645 - DLM race condition prevented some lock evictions (2328)
6646 - ENOMEM detection and retry on socknal sends (2230)
6647 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
6648 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
6649 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
6650 - better error messages when a client is rejected during recovery (1505)
6651 - avoid cancelling locks which were never granted, after failure (2330)
6652 - fix i_sem/journal inversion in mds_client_add (2333)
6653 - fix truncate/getattr lock cycle deadlock (2334)
6654 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
6655 - fix two rare exit paths which could leak an l_lock() ref (2321)
6656 - fencepost error in MDS/OST orphan recovery (2226)
6657 - make log record alignment 8 bytes (1988)
6658 - lstripe now fails when requested offset > ost_count (2237)
6659 - ensure that all kernel series have a complete list.h (1607)
6660 - fix crashes in special-file operations (2316)
6661 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
6663 - allow configurable automake binary, for testing new versions
6664 - small update to the lfs documentation
6666 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
6668 * fix negative export reference count in fsfilt_sync (2312)
6670 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
6671 * release candidate 0.9.1
6673 - orphans are moved into the PENDING directory for possible recovery
6674 - replayed opens now open by fid for orphan/rename safety (1042)
6675 - last close of an orphan inode generates a transno (683)
6676 - chdir() and mount() now pin the directory entry (1020)
6677 - avoid CERROR in normal ll_setattr_raw() error case (1500)
6678 - discard very old requests without processing them (1502)
6679 - remove some common, well-understood CERRORs (1505)
6680 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
6681 - clear "grant" flags in OST replies until OST grant code lands (1644)
6682 - fix read performance by not clobbering i_blksize on client (1598)
6683 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
6684 - make LRU size a /proc tunable, clears locks when reduced (707)
6685 - fix some lprocfs rot that prevented ptlbd from loading (1732)
6686 - server locks take references on exports now (1558)
6687 - build fixes for 2.4.20-rh trees (1663)
6688 - return an error from lov_create if all OSCs are inactive (1751)
6689 - fix import levels when a reconnect happens without a timeout (1597)
6690 - exit early from mds_open if we get a lookup error (1749)
6691 - partial page read at EOF wouldn't wait for disk before sending (1642)
6692 - avoid NULL deref in obdfilter when reading page past EOF (1592)
6693 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
6694 - fix LBUG in ll_it_open_error with rc = -2 (1861)
6695 - write/truncate lock inversion (1639)
6696 - Don't auto-load obdclass, portals modules during cleanup (1495)
6697 - fix timestamps from jumping to "now" (1763)
6698 - extra journal assertions (1648)
6699 - add an extra multiunlink test (1771)
6700 - fix read_record/write_record API (1776)
6701 - fix leak of offset_extent, possible incorrect i_size later (1772)
6702 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
6703 - Debugging for the kqswnal_get_idle_tx problems (1820)
6704 - Allow recovery to be attempted multiple times (1536)
6705 - Write out MDS last_rcvd file after it is first created (1600)
6706 - Fix tx_descriptor leak in failed transmit situations (1827)
6707 - ext3 journaling fixes for assertion failure after IO error (1871)
6708 - class_export_put() on freed export after completion AST error (1896)
6709 - Fix revalidate looping in VFS (1322)
6710 - Don't access a freed export during MDS_REINT timeout (1521)
6711 - Add open-unlink recovery support on the MDS (1673,1764)
6712 - Return an error if no MDS data was read from last_rcvd (1946)
6713 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
6714 - Don't setup a disk device that doesn't match exported UUID (317)
6715 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
6716 - avoid committing NULL handle in force close
6717 - local.sh is now a one-stripe LOV configuration
6718 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
6719 - fix ext3 htree duplicate directory entry corruption (1516)
6720 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
6721 - update to new LOV EA format (2097)
6722 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
6724 2003-06-15 Phil Schwan <phil@clusterfs.com>
6727 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
6728 - per-import/export recovery handling (958, 931, 959)
6729 - multiple last-rcvd slots, for serving multiple FSes (949)
6730 - connections are again shared between multiple imp/exports (963, 964)
6731 - "umount -f" would hang if any requests needed to be sent (393, 978)
6732 - avoid pinning large req buffer by copying for queued messages (989)
6733 - add "uuid" to "lctl device" command to help upcalls (991)
6734 - "open" RPCs with transnos would confuse recovery counters (1037)
6735 - do proper endian conversion of all wire messages (288, 340, 891)
6736 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
6737 - call waiting locks callback from LDLM recovery thread (1127, 1151)
6738 - fix ptlrpc_connection leak in target_handle_connect (1174)
6739 - fix import refcounting bug in OST and MDS cleanup (1134)
6740 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
6741 - fix very unlikely obd_types race condition (501)
6742 - remove osc_open hack for echo_client (1187)
6743 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
6744 - a failure in read_inode2 leads to deadlock (1139)
6745 - cancel ack-locks as soon as transaction is committed (1072)
6746 - fix major leaks and crashes in the bulk I/O path (937, 1057)
6747 - make sure to commitrw after any preprw to avoid deadlock (1162)
6748 - failing to execute a file in a lustre FS would lock inode (1203)
6749 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
6750 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
6751 - obd_uuid2tgt would walk off the end of the list (1255)
6752 - on IA64 the osc would give portals incorrect bulk size (1258)
6753 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
6754 - fix lock inversion caused by new llite matching code (1282)
6755 - limit the number of dirty pages on a client to 10MB (1286)
6756 - timed out locks were not being corrected cancelled (1289)
6757 - fix O_DIRECT above 4GB on IA-32 (1292)
6758 * major user-visible changes
6759 - fail out/fail over policy now controlled by the upcall (993)
6761 - add OBD_PING to check server availability and failure (954)
6762 - lustre messages are now sent in sending host order (288, 340, 891)
6763 - add eadatalen to MDS getattr reply (340)
6764 - OST read replies may contain second buffer, with per-page status (593)
6766 2003-03-11 Phil Schwan <phil@clusterfs.com>
6769 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
6770 - failed open()s could cause deadlock; fixed (867, 869)
6771 - stop cancelling OST locks when files are closed (481)
6772 - overlapping XID spaces caused network corruption (851, 853)
6773 - fix unsafe fsfilt counter arithmetic; change to atomic_t
6774 - setattr_raw added, to do single-RPC, server-side setattrs
6775 - lmc/lconf syntax change for OST UUIDs
6776 - fix crashy race condition between ptlrpc_free_req and osc_close
6777 - don't use request in mdc_enqueue if we hit a timeout (889)
6778 - don't set the inode i_size for regular files from the MDS (896)
6779 - handle out of order completion AST (842)
6780 - don't LBUG if a lock request times out after receiving AST (913)
6781 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
6782 - if a bad lock AST arrives, send an error instead of dropping entirely
6783 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
6784 - fix leak in bulk IO when only partially completed (899, 900, 926)
6785 - fix O_DIRECT for ia64 (55)
6786 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
6787 - C-z after timeout could hang a process forever; fixed (977)
6789 - client-side I/O cache (678, 924, 929, 941, 970)
6791 - READPAGE and SETATTRs which don't take server-side locks get
6794 2003-02-11 Phil Schwan <phil@clusterfs.com>
6797 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
6798 - Fix fsfilt_extN_readpage() to read a full page of directory
6799 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
6800 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
6801 - Fix imbalanced LOV object allocation and out-of-bound access (469)
6802 - Most intent operations were removed, in favour of a new RPC mode
6803 that does a single RPC to the server and bypasses most of the VFS
6804 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
6805 - Aggressively cancel local locks on DLM servers
6806 - mds_reint_unlink sends EA to the client if it's the last nlink.
6807 client uses that EA to unlink OST objects.
6808 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
6809 - recursive symlinks were fixed (439)
6810 - fixed NULL deref in DEBUG_REQ
6811 - filter_update_lastobjid no longer calls sync, which annoyed extN
6812 - fixed multi-client small-writes to a single file problem (445)
6813 - fixed mtime updates during file writes (607)
6814 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
6815 - fixed bug in obd_brw_read/write() (under guise of testing 367)
6816 - fixed Linux OST size reporting problem (444, 656)
6817 - OST now updates object mtime with writes or setattr (607, 619)
6818 - client verifies file size before zeroing page past EOF (445)
6819 - OST now writes last allocated objid to disk with allocation (108)
6820 - LOV on echo now works (409)
6822 - mds_reint_unlink sends a new buffer, with the EA included. this
6823 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
6824 set if a regular file was being unlinked, and it was the last link
6825 - use PtlGet from the target for bulk writes (315)
6826 - OST now updates object mtime with writes or setattr (607, 619)
6827 - LDLM now has a grant-time callback to revalidate locked items, if
6829 - Many MDS operations were reorganized to combat race conditions
6831 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
6832 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
6834 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
6837 - Fully reactivate OST imports after reconnection (512, others)
6838 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
6839 - More graceful error handling for truncating on dead OST (515)
6840 - Don't error out unless we're actually accessing dead stripes (474)
6841 - Fix garbage sizes when stripes are missing (410)
6842 - LRU counters were broken, causing constant lock purge (433, 432)
6843 - garbage on read from stripes with failed OSTs (441)
6844 - mark OSCs as active before reconnecting during recovery (438)
6845 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
6846 - lfind did not preserve OST order in output (443)
6847 - symlinks cause hung clients, incorrect data (439)
6848 - stop dereferencing request after dropping refcount (457)
6849 - don't LASSERT(spin_is_locked) on non-SMP (455)
6850 - fixes for many rename() bugs
6851 - fstat didn't correctly synchronize attributes (399)
6852 - server must handle lock cancellation during blocking AST prep (487)
6853 - bulk descriptors were free()d too soon (511)
6854 - fix paths in lconf, which would load incorrect modules (451, 507)
6855 - fix confusing lconf 'host not found' error message (386)
6856 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
6857 - fix race condition in mdc_blocking_ast() for inode access (526)
6858 - fix lov_unpackmd() unpacking wrong number of stripes (537)
6859 - fix lov_set_osc_active() marking wrong OSC inactive (440)
6860 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
6861 - fix multiple writes of stripe MD to MDS (358, maybe 519)
6862 - fix lstripe in several ways (kernel side) (527)
6863 - fix request leak in ldlm_cli_enqueue (262)
6864 - incorrect OSC was marked inactive after OST failure
6865 - call mds_fs_cleanup before unmounting filesystem (524)
6866 - fix races between taking ns_lock and ldlm_lock_change_resource
6867 - fix races updating LOV export open file list
6868 - fix lov_enqueue error path, avoid decref-ing bad lock handle
6869 - fix recovery NULL deref in ldlm_cli_cancel_unused
6870 - fix some DLM races by using new hash table for lock handles (419)
6871 - permit the client to specify desired inodes, at replay
6872 - duplicate requests when we queue them for replay reintegration
6873 - fix last_rcvd offset calculation
6874 - sync after each recovered transaction, so we always make progress
6875 - never, not always, ERESTART requests without transnos
6876 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
6878 - skip replay if the MDS says that the client is already connected
6879 - don't check for a recovery-enabled export to match lctl's UUID
6880 - don't INC_USE_COUNT for phantom exports
6881 - don't crash when cleaning up phantom exports (567)
6882 - don't double-finish or set replay data for errored mdc_open requests
6883 - abort requests when they time out, so we don't get old replies
6884 - send/receive replies for AST messages again
6885 - if the client says that it doesn't have the lock, cancel it on the
6887 - if we timeout during I/O, don't try to cancel an in-use lock; instead
6888 mark it as destroyed, it will all work out when decref is called
6889 - fix module use counts (22, 581)
6891 - ASTs now expect a reply (server cancels lock on error reply)
6893 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
6896 - fix many simultaneous client startup (392)
6897 - fix dentry->d_it clobbering
6898 - credentials weren't being shipped for readdir/getattr operations
6899 - remove invalid assertions triggered during some concurrent MD
6901 - proper Lustre versions added (336, 389)
6902 - fix memory leak for create error case (398)
6903 - fix LOV locking bug that would get cli/srv out of sync
6904 - fix echo client over LOV (409)
6905 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
6906 - fix double-O_EXCL intent crash (424)
6907 - avoid sending multiple lock CANCELs (352)
6909 - MDS can do multi-client recovery (modulo bugs in new code)
6911 - many updates, edits, cleanups
6913 2002-11-18 Phil Schwan <phil@clusterfs.com>
6916 - fix null d_it dereference (346)
6917 - fix full OST/dbench hang (333)
6918 - fix permission problem with file removal (286)
6919 - fix removal of OSCs from LOV when they fail
6920 - fix NULL deref during bulk timeout (214)
6921 - fix problems related to multiple filesystems on one MDS (241)
6922 - fixed serious subtle metadata locking bugs
6923 - free locks on clients when inodes are removed due to memory
6925 - fix inode pointer in lock data (285)
6926 - partial support for multiple MDS on a single host (241)
6927 - data locks weren't cancelled at clear_inode time (290, 311)
6928 - intent locks could lead to unbounded lock growth (205)
6929 - added a maximum lock count, an LRU list, and a flusher
6930 - fix multiple rename (365)
6931 - properly abstracted the echo client
6932 - OSC locked 1 byte too many; fixed
6933 - rewrote brw callback code:
6934 - fixed recovery bugs related to LOVs (306)
6935 - fixed too-many-pages-in-one-write crash (191)
6936 - fixed (again) crash in sync_io_timeout (214)
6937 - probably fixed callback-related race (385)
6939 - Add capability to MDS protocol
6940 - LDLM cancellations and callbacks on different portals
6942 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
6945 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
6946 - defer open object destruction to close time (601981, 138)
6947 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
6948 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
6949 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
6950 - don't call obd_setattr in ll_file_release for destroyed objects
6951 * protocol change to lustre_msg: move |version| and add |flags|
6952 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
6953 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
6954 * added replay of create, unlink, link and rename operations during
6955 MDS failover; recovery should be much more robust now
6956 * remove failed OSCs from LOVs (only lov_create uses this so far)
6957 * the lustre-HOWTO was brought (more) up to date (582544)
6959 2002-10-23 Phil Schwan <phil@clusterfs.com>
6962 - in-use dentries weren't being reused properly (617851)
6963 - prevent multiple LDLM setup (599178)
6964 - fix LOV size calculations for truncate (617853)
6965 - fix client handling of MDS intent errors (POSIX)
6966 - fix permission bug in lovstripe.c test (624321)
6967 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6968 - truncate past end of file could corrupt data
6969 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6970 - a race in recovery could return ETIMEDOUT to apps (623947)
6971 - building outside the source directory was fixed
6972 * the lustre-HOWTO was brought (more) up to date (582544)
6973 * major progress was made on recovery functionality
6975 2002-10-10 Phil Schwan <phil@clusterfs.com>
6978 - recovery deadlock fix
6979 - rm -rf causes LBUG fix (617817)
6980 - file open by multiple tasks fix (618962)
6981 - directory permissions bugs (602707 and 620007)
6982 - journal_stop fixed with locking (611313)
6983 - O_APPEND failures resolved (618273, perhaps 614459)
6984 - lconf PATH fix (619770)
6985 - IA64 build fix (621450)
6986 - RPC buffer sizes scale with amount of memory
6988 2002-10-01 Phil Schwan <phil@clusterfs.com>
6991 - locks would be cancelled without throwing away data pages,
6992 resulting in inconsistent data (605627)
6993 - inode attributes were not always being refreshed (605627, 612449)
6994 - lconf now continues to cleanup after lctl reports an error
6995 - MDS now enforces user permissions (602707)
6996 - lprocfs cleanup fixed, but not yet enabled (614157)
6997 - fixed infinite server hang, should a client not respond to an AST
6998 - avoid going into recovery if user calls readlink() with a buffer
6999 that's too small (613941)
7000 - AST RPCs no longer require replies (614867) -- this may be changed
7001 - don't crash server if client sends an IOV that's too big (611336)
7002 - fixed lock conversion deadlock (611892)
7003 - fixed the following of symlinks (614622)
7004 * recovery: the server can remove locks from a client that dies, other
7005 clients can make progress
7006 * more extN patch fixes
7007 * compile-time configurable ptlrpc buffer allocations
7009 - collaborative read cache document
7010 - Lustre Lite Performance CDR document-in-progress
7012 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
7015 - fix typo in patch-2.4.18
7017 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
7020 - clear ptlrpc request each time in handle_incoming_request()
7021 - unlink of files now destroys the object on the OST
7023 2002-09-19 Peter Braam <braam@clusterfs.com>
7025 * add hard link support
7026 * change obdfile creation method
7027 * kernel patch changed
7029 2002-09-19 Peter Braam <braam@clusterfs.com>
7032 - stack overflow bug in extN fixed
7034 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
7036 * documentation updates
7037 - add man pages for config tools
7038 - update tests/README to describe testing with new config tools
7039 - finish metadata API descriptions
7040 * bug fixes and cleanups
7041 - statfs workaround for 16TB limit
7042 - LOV stripe allocation improved, can stripe on subset of OSTs
7043 - LOV file size/IO offset was wrong for files > 4GB in size
7044 - object EA data was being dropped, caused files to be unreadable
7045 - memory overflow with non-LOV OST caused memory corruption
7046 - fixed regression tests to work with new config tools, obdfilter
7047 - fixed bug when directory size became larger than 1 block
7048 - fixed bug (for single client case) when PWD was deleted
7049 - invalidate local directory pages when doing intent-based ops
7050 - avoid LDLM oops when lock callback contained bad data
7052 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
7054 * documentation updates
7055 * bug fixes and cleanups
7056 - configuration tools
7059 - 64-bit compile warnings
7060 - 64-bit internal statfs data
7062 * test_brw on persistent OST devices
7064 * lprocfs (disabled)
7066 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
7068 * documentation updates
7069 * bug fixes and cleanups
7070 * configuration tools
7072 2002-08-30 Peter J. Braam <braam@clusterfs.com>
7075 * many small fixes to 0_5_4
7076 * io/network handling
7077 * thinkos in MDS operations
7079 2002-08-24 Peter J. Braam <braam@clusterfs.com>
7082 * crucial basic fixes to 0.5.3
7083 * IOR, Iozone work over Elan
7086 2002-08-07 Phil Schwan <phil@clusterfs.com>
7087 * version 0_5_3, our first alpha
7088 * we use the new Portals iovs
7089 * documentation updates
7090 * bug fixes and cleanups
7091 * small changes in the DLM wire protocol
7093 2002-07-25 Peter J. Braam <braam@clusterfs.com>
7094 * version 0_5_1 with some initial stability,
7095 * locking on MD and file I/O.
7096 * documentation updates
7097 * several bug fixes since 0.5.0
7098 * small changes in wire protocol
7100 2002-07-18 Phil Schwan <phil@clusterfs.com>
7102 * delivered as Lustre Light Alpha
7103 * fixed a crash after handling invalid MDS requests
7104 * fixed directory pages for architectures with non-4k pages sizes
7106 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
7107 * release version v0_4_4
7108 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
7110 * Fixes a number of interruption problems with OST operations.
7111 * Update documentation for portals header changes
7112 * Move all wire protocol structs/defines to lustre_idl.h
7113 * Fixes symlink length bug.
7114 * Add tcpdump to repository.
7116 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
7117 * release version v0_4_3
7118 * Fixes statfs for inodes on extN.
7119 * Fixes bug in runtests which would delete /etc/hosts.
7120 * Use 64-bit object IDs wherever possible (not into VFS though)
7121 Remove ost_get_info, which is unused by lustre, and out of date.
7123 2002-07-03 Peter Braam <braam@clusterfs.com>
7124 * release version v0_4_2 Fixes a lookup error (type not passed)
7125 * move forward to head of Portals
7126 * move forward to latest Lustre kernel
7128 2002-06-25 Peter Braam <braam@clusterfs.com>
7129 * release version v0_4_1. Hopefully stable on single node use.