1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.42.4 (SLES 10),
5 2.6.27.29-0.1 (SLES11),
6 2.6.18-128.7.1.el5 (RHEL 5)
7 2.6.18-128.7.1.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 * File join has been disabled in this release, refer to Bugzilla 16929.
13 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
14 * NFS export disabled when stack size < 8192. Since the NFSv4 export
15 of Lustre filesystem with 4K stack may cause a stack overflow. For
16 more information, please refer to bugzilla 17630.
20 Description: can't stat file in some situation.
21 Details : improve initialize osc date when target is added to mds and
22 ability to resend too big getattr request is client isn't have info
25 Severity : enhancement
27 Description: Add OEL5 support.
29 Severity : enhancement
31 Description: Update kernel to SLES11 2.6.27.29-0.1.1.
33 Severity : enhancement
35 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
39 Description: Improved cleanup procedure for conf-sanity test_46a.
43 Description: Prevent inconsistences between linux and lustre mount structures.
44 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
45 Make the sleep interruptible.
47 Severity : enhancement
49 Description: Enhancement for lfs(1) command to use numeric uid/gid.
53 Description: Do not start bulk transfer, if client tried to reconnect.
54 Details : Do not start bulk transfer, if client tried to reconnect,
55 but failed due to active rpcs. Check for exp_abort_active_req
58 Severity : enhancement
60 Description: Communicate OST degraded/readonly state via statfs to MDS
61 Details : Flags in the statfs returned from OSTs indicate whether the
62 OST is in a degraded RAID state, or if the filesystem has
63 turned read-only after a filesystem error is detected.
65 Severity : enhancement
67 Description: Vanilla kernel 2.6.30 patchless support.
72 Description: don't panic if EPROTO was hit when reading symlink
73 Details : correctly handling request reference in error cases.
78 Description: open sometimes returns ENOENT instead of EACCES
79 Details : checking permission should be part of open part of mds_open, not
80 lookup part. so server should be set DISP_OPEN_OPEN disposition
81 before starting permission check.
82 Also not need revalidate dentry if client already have LOOKUP lock.
86 Description: enable client interface failover
87 Frequency : on servers with multiple network interfaces
88 Details : When a child reconnects from another NID, properly update export
89 nid hash position and ldlm reverse import.
91 Severity : enhancement
93 Description: implemented direct I/O with arbitrary (nonaligned) memory
94 addresses and file offsets.
96 Severity : enhancement
98 Description: added more recovery timeout options.
100 Severity : enhancement
102 Description: added llapi_file_open, llapi_file_create, llapi_file_get_stripe
107 Description: Avoid deadlock for local client writes
108 Frequency : only on systems with clients writing to an OST on the same node
109 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
110 memory freeing context. This allows OST threads to set the
111 PF_MEMALLOC flag on task structures in order to allocate memory
112 from reserved pools and complete IO.
113 Use GFP_HIGHUSER for OST allocations for non-local client writes,
114 so that the OST threads generate memory pressure and allow
115 inactive pages to be reclaimed.
120 Descriptoin: lock ordering violation between &cli->cl_sem and _lprocfs_lock
121 Details : move ldlm namespace creation in setup phase to avoid grab
122 _lprocfs_lock with cli_sem held.
126 Frequency : only during format of test systems
127 Description: Unable to run several mkfs.lustre on loop devices at the same time
128 Details : mkfs.lustre returns error 256 on the concurrent loop devices
129 formatting. The solution is to proper handle the error.
131 Severity : enhancement
133 Description: implement async create (obd_async_create) method for osc, to avoid
134 too long waiting new ost objects with holding ldlm lock.
136 Severity : enhancement
138 Descriptoin: OST grant too much space to client even there are not enough space.
139 Details : Client will shrink its grant cache to OST if there are no write
140 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
141 this grant cache if there are already not enough avaible space
142 (left_space < total_clients * 32M).
145 Frequency : rare, if clients are evicted due to failure to return a lock
147 Description: don't increase ldlm timeout if previous client was evicted
148 Details : if a client doesn't respond to a blocking callback within the
149 adaptive ldlm enqueue timeout, don't adjust the adaptive estimate
150 when the lock is next granted.
154 Frequency : occasionally during network problems
155 Description: client not allowed to reconnect to OST because of active request
156 Details : abort bulk requests received by the OST once the client has timed
157 out since the client will resend the request anyway. The client
158 also now retries to reconnect to the same server if a connect
159 request failed with EBUSY or -EAGAIN.
162 Frequency : rare, if used wide striped file and one ost in down.
164 Descriptoin: don't return error if we created a subset of objects for file.
165 Details : lov_update_create_set() uses set->set_success as index for created
166 objects, so if some requests failed, they will have hole at end of
167 array and we can use qos_shrink_lsm for allocate correct lsm.
170 Frequency : rare, only when quota is enabled
172 Descriptoin: OSS complains "too many blocks" and crashes
173 Details : In filter_commitrw_write():
174 1. lquota_chkquota();
175 2. obdo_from_inode();
176 3. lquota_pending_commit();
177 "2." has changed uid and gid in this case, then caused
178 lquota_chkquota() and lquota_pending_commit() to use
179 different uid and gid. That is the root of the bug.
182 ------------------------------------------------------------------------------
183 TBD Sun Microsystems, Inc.
185 * Support for kernels:
186 2.6.16.60-0.42.4 (SLES 10),
187 2.6.27.23-0.1 (SLES11, i686 & x84_64 only),
188 2.6.18-128.7.1.el5 (RHEL 5),
189 * Client support for unpatched kernels:
190 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
191 2.6.16 - 2.6.27 vanilla (kernel.org)
192 * Recommended e2fsprogs version: 1.41.6.sun1
193 * File join has been disabled in this release, refer to Bugzilla 16929.
194 * NFS export disabled when stack size < 8192. Since the NFSv4 export
195 of Lustre filesystem with 4K stack may cause a stack overflow. For
196 more information, please refer to bugzilla 17630.
197 * ext4 support for RHEL5 is experimental and thus should not be
202 Description: File checksum failures with OST read cache on
203 Details : Disable page poisoning when the bulk transfer has to be aborted
204 because the client got evicted.
208 Description: Don't allow make backward step on assign osc next id.
209 Details : race between allocation next id and ll_sync thread can be cause
210 of set wrong osc next id and can be kill valid ost objects.
212 Severity : enhancement
214 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
216 Severity : enhancement
218 Description: Update kernel to SLES10 SP2 2.6.16.60-0.42.4.
222 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
226 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
230 Description: Error handling in osc_statfs_interpret() has been improved.
231 Details : Check in osc_statfs_interpret() for EBADR.
235 Description: Do not send statfs() requests to OSTs disabled by administrator.
236 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
240 Description: Do not update ctime for the deleted inode.
241 Details : Check in mds_reint_unlink() before calling fsfilt_setattr().
245 Description: Increase of the size of the LDLM resource hash.
246 Details : Bump up RES_HASH_BITS=12.
250 Description: correctly send lsm on open replay
251 Details : MDS is trust to LSM size on replay open, but client can set wrong size
256 Description: Deadlock between filter_destroy() and filter_commitrw_write().
257 Details : filter_destroy() does not hold the DLM lock over the whole
258 operation. If the DLM lock is dropped, filter_commitrw() can go
259 through, causing the deadlock between page lock and i_mutex.
260 The i_alloc_sem should also be hold in filter_destroy() while
265 Description: truncate starts GFP_FS allocation under transaction causing deadlock
266 Details : ldiskfs_truncate calls grab_cache_page which may start page
267 allocation under an open transaction. This may lead to
268 calling prune_icache with consequent lustre reentrance.
271 Frequency : only when down/upgrading the MDS to 1.6/1.8 while 1.8 clients are
272 still up and when the OST pool feature is used
274 Description: interop testing got LBUG when run dd with OST pool
275 :LustreError: 30032:0:(llite_lib.c:1913:ll_replace_lsm()) LBUG
276 Details : down/upgrading the MDS to a version that doesn't/does support OST
277 pool can cause clients to crash because the lsm has changed
282 Description: missing tree_status on 1.8.1 RPM build
283 Details : make rpms failed due because the tree_status file is missing.
287 Description: continuing LustreError "mds adjust qunit failed!"
288 Details : don't print message on the console when ->adjust_qunit fails.
292 Description: OST is being unmounted w/o all writes to last_rcvd landing on disk.
293 Affects recovery negatively.
294 Details : Make sure all exports have been properly destroyed by the zombie
295 thread processed before stopping the target.
298 ------------------------------------------------------------------------------
299 2009-07-31 Sun Microsystems, Inc.
301 * Support for kernels:
302 2.6.16.60-0.39.3 (SLES 10),
303 2.6.27.23-0.1 (SLES11, i686 & x84_64 only),
304 2.6.18-128.1.14.el5 (RHEL 5),
305 * Client support for unpatched kernels:
306 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
307 2.6.16 - 2.6.27 vanilla (kernel.org)
308 * Recommended e2fsprogs version: 1.41.6.sun1
309 * File join has been disabled in this release, refer to Bugzilla 16929.
310 * NFS export disabled when stack size < 8192. Since the NFSv4 export
311 of Lustre filesystem with 4K stack may cause a stack overflow. For
312 more information, please refer to bugzilla 17630.
313 * ext4 support for RHEL5 is experimental and thus should not be
316 Severity : enhancement
318 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
321 Frequency : with 1.8 server and 1.6 clients
323 Description: correctly shrink reply for avoid send too big message to client.
324 Details : 1.8 mds is allocate to big buffer to LOV EA data and this produce
325 some problems with sending this reply to 1.6 client.
329 Description: Repeated atomic allocation failures.
330 Details : Use GFP_HIGHUSER | __GFP_NOMEMALLOC flags for memory allocations
331 to generate memory pressure and allow reclaiming of inactive pages.
332 At the same time, do not allow to exhaust emergency pools.
333 For local clients the use of GFP_NOFS will be introduced in 1.8.2
335 Severity : enhancement
336 Bugzilla : 19846, 18289
337 Description: Update kernel to RHEL5 2.6.18-128.1.14.el5.
339 Severity : enhancement
340 Bugzilla : 19625, 16893, 18668, 19848
341 Description: Add support for SLES11 2.6.27.23-0.1.
343 Severity : enhancement
345 Description: Update client support to vanila kernels up to 2.6.27.
347 Severity : enhancement
349 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
351 Severity : enhancement
353 Description: Compile with -Werror by default for i686 and x86_64.
357 Description: resolve race between obd_disconnect and class_disconnect_exports
358 Details : if obd_disconnect will be called to already disconnected export he
359 forget release one reference and osc module can't unloaded.
361 Severity : enhancement
363 Description: move AT tunable parameters for more consistent usage
364 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
368 Descriptoin: correctly skip time estimate if in recovery
369 Details : rq_send_state insn't bitmask so using bitwise ops is forbid.
373 Descriptoin: OSS DeadLock
374 Details : Use trylock to prevent deadlock when shrink icache.
376 Severity : enhancement
378 Description: Allow tuning service thread via /proc
379 Details : For each service a new
380 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
381 created that can be used to set min/max thread counts, and get the
382 current number of running threads.
384 Severity : enhancement
386 Description: Add state history info file, enhance import info file
387 Details : Track import connection state changes in a new osc/mdc proc file;
388 add overview-type data to the osc/mdc import proc file.
392 Description: Reduce small size read RPC
393 Details : Set read-ahead limite for every file and only do read-ahead when
394 available read-ahead pages are bigger than 1M to avoid small size
399 Description: free_entry erroneously used groups_free instead of put_group_info
401 Severity : enhancement
403 Description: Make read-ahead stripe size aligned.
405 Severity : enhancement
407 Description: MDS create should not wait for statfs RPC while holding DLM lock.
410 Frequency : rare, connect and disconnect target at same time
412 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
413 Details : don't call obd_disconnect under lov_lock. this long time
414 operation and can block ptlrpcd which answer to connect request.
417 Frequency : start MDS on uncleanly shutdowned MDS device
419 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
420 Details : stay in waiting mds<>ost recovery finished produce random bugs
421 due race between two ll_sync thread for one lov target. send
422 ACTIVATE event only if connect realy finished and import have
426 Frequency : start MDS on uncleanly shutdowned MDS device
428 Descriptoin: aborting recovery hang on MDS
429 Details : don't throttle destroy RPCs for the MDT.
433 Description: Slow reads beyond 8Tb offsets.
434 Details : Page index integer overflow in ll_read_ahead_page
438 Description: Soft lockup on OSS after MDS failover
439 Details : MSG_CONNECT_INITIAL is not set on the initial MDS->OST connect.
440 As a conseqence, the patch from bug 18224 is not operational
441 and the MDS export cannot be reused on the OSTs until it gets
445 Frequency : rare, only if using MMP with Linux RAID
447 Description: MMP doesn't work with Linux RAID
448 Details : While using HA for Lustre servers with Linux RAID, it is possible
449 that MMP will not detect multiple mounts. To make this work we
450 need to unplug the device queue in RAID when the MMP block is being
451 written. Also while reading the MMP block, we should read it from
452 disk and not the cached one.
455 Frequency : rare, during recovery
457 Description: Assertion failure in ldlm_lock_put
458 Details : Do not put cancelled locks into replay list, hold references on
463 Description: 1.6.5 mdsrate performance is slower than 1.4.11/12 (MDS is not cpu bound!)
464 Details : create_count always drops to the min value (=32) because grow_count
465 is being changed before the precreate RPC completes.
468 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
471 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
472 Details : add proper locking for creation of jbd_4k slab cache
476 Description: MMP check in ext3_remount() fails without displaying any error
477 Details : When multiple mount protection fails during remount, proper error
482 Description: Rare Client crash on resend if the file was deleted.
483 Details : When file is opened, but open reply is lost and file is
484 subsequently deleted before resend, resend processing logic
485 breaks trying to open the file again, should not try to open.
489 Description: add check for >8TB ldiskfs filesystems
490 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
491 Don't allow >8TB ldiskfs filesystems to be mounted without
492 force_over_8tb mount option
496 Description: Client locked up when running multiple instances of an app. on
497 multiple mount points
498 Details : ll_shrink_cache() can sleep while holding the ll_sb_lock.
499 Convert ll_sb_lock to a read/write semaphore to fix the problem.
503 Description: Cannot acces an NFS-mounted Lustre filesystem
504 Details : An NFS client cannot access the Lustre filesystem NFS-mounted
505 from a Lustre-client exporting the Lustre filesystem via NFS.
509 Description: panic in ll_statahead_thread
510 Details : grab dentry reference in parent process.
513 -------------------------------------------------------------------------------
515 tbd Sun Microsystems, Inc.
517 * Support for kernels:
518 2.6.16.60-0.31 (SLES 10),
519 2.6.18-128.1.6.el5 (RHEL 5),
520 2.6.22.14 vanilla (kernel.org)
521 * Client support for unpatched kernels:
522 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
523 2.6.16 - 2.6.22 vanilla (kernel.org)
524 * Recommended e2fsprogs version: 1.40.11-sun1
525 * File join has been disabled, refer to Bugzilla 16929.
526 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
527 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
528 Lustre filesystem with 4K stack may cause a stack overflow. For more
529 information, please refer to bugzilla 17630.
531 Severity : enhancement
533 Description: Update to RHEL5.3 kernel-2.6.18-128.1.6.el5.
535 Severity : enhancement
537 Description: Update OFED release to 1.4.1 RC4
539 Severity : major, only with big OST
541 Description: Very poor metadata performance on Infiniband lustre configuration
542 Details : OST object precreation becomes very slow on big OSTs. This is due
543 to the ialloc patch spending too much time scanning groups.
546 Frequency : during recovery
548 Description: don't mix llog inodes with normal.
549 Details : allocate inodes for log in last inode group
554 Description: fix lqs' reference which won't be put in some situations
555 Details : This patch fixes:
556 1. In quota_check_common(), this function will check quota
557 for user and group, but only send one return via "pending".
558 In most cases, the pendings should be same. But that is not
560 2. If quotaoff runs between lquota_chkquota() and
561 lquota_pending_commit(), the same thing will happen too.
562 That is why it comes:
563 - if (!ll_sb_any_quota_active(qctxt->lqc_sb))
566 -------------------------------------------------------------------------------
568 2008-12-31 Sun Microsystems, Inc.
570 * Support for kernels:
571 2.6.5-7.314 (SLES 9),
572 2.6.9-67.0.22.EL (RHEL 4),
573 2.6.16.60-0.31 (SLES 10),
574 2.6.18-92.1.17.el5 (RHEL 5),
575 2.6.22.14 vanilla (kernel.org)
576 * Client support for unpatched kernels:
577 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
578 2.6.16 - 2.6.22 vanilla (kernel.org)
579 * Client support for unpatched kernels:
580 we do not recommend using patchless RHEL4 clients with kernels
581 prior to 2.6.9-55EL (RHEL4U5).
582 * Recommended e2fsprogs version: 1.40.11-sun1
583 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
584 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
585 removed cwd "./" (refer to Bugzilla 14399).
586 * A new quota file format has been introduced in 1.6.5.
587 The format conversion from prior releases is handled transparently,
588 but releases older than 1.4.12/1.6.5 don't understand this new
589 format. The automatic format conversion can be avoided by running
590 the following command on the MDS:
591 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
592 For more information, please refer to bugzilla 13904.
593 * A new quota file format was introduced in 1.6.6/1.8.0 (kernels 2.6.16+).
594 The format conversion from prior releases is handled transparently,
595 but releases older than 1.6.6/1.8.0 don't understand this new
596 format. The automatic format conversion can be avoided by running
597 the following commands on the MDS and OSS servers (for
598 pre 1.4.12-1.6.5 quota files):
599 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV',
600 'tunefs.lustre --param="ost.quota_type=ug1" $MDTDEV'
601 or (for 1.4.12/1.6.5 quota files)
602 'tunefs.lustre --param="mdt.quota_type=ug2" $MDTDEV',
603 'tunefs.lustre --param="ost.quota_type=ug2" $MDTDEV'
604 For more information, please refer to bugzilla 13904.
605 * Output of lfs quota has been made less detailed by default,
606 old (verbose) output can be obtained by using -v option.
607 * File join has been disabled in this release, refer to Bugzilla 16929.
608 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
609 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
610 Lustre filesystem with 4K stack may cause a stack overflow. For more
611 information, please refer to bugzilla 17630.
613 Severity : enhancement
615 Description: Caching OSS
616 Details : introduce data caching on the OSS. The OSS now relies on the linux
617 kernel page cache to keep recently accessed data in memory.
618 It is worth noting that all write requests are still flushed
619 synchronously as in lustre 1.6.
621 Severity : enhancement
623 Description: Version based recovery
624 Details : introduce finer grained recovery able to detect transaction
625 dependencies and can deal with transaction gaps caused by clients
626 failing at the same time as the server.
628 Severity : enhancement
630 Description: Enable adaptive timeouts by default
631 Details : The Lustre timeout value in /proc/sys/lustre/timeout is now
632 managed dynamically based on server load and should not need
633 to be tuned manually based on cluster size. This allows Lustre
634 to work under a wider variety of system sizes and loads, without
635 unnecessarily causing lengthy recovery times.
637 Severity : enhancement
639 Description: Add OST Pools support
640 Details : File striping can now be set to use an arbitrary pool of OSTs
642 Severity : enhancement
644 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
645 Details : allow skip disconnected ost for send statfs request and hide error
649 Frequency : rare, on llog test 6
651 Descriptoin: don't allow connect to already connected import
652 Details : allowing connect to already connected import is hide connecting problem.
655 Frequency : rare, on failed llog setup
657 Descriptoin: don't leak obd reference on failed llog setup
658 Details : for failed llog setup - mgc forget call class_destroy_import for
659 client import, move destroy import to more generic place.
664 Descriptoin: allow kill process which wait statahead result
665 Details : for some reasons 'ls' can stick in waiting result from statahead,
666 in this case need way for kill this process.
669 Frequency : rare, at shutdown
671 Descriptoin: panic at umount
672 Details : llap_shrinker can be raced with killing super block from list and
673 this produce panic with access to already freeded pointer
678 Descriptoin: don't lose wakeup for imp_recovery_waitq
679 Details : recover_import_no_retry or invalidate_import and import_close can
680 both sleep on imp_recovery_waitq, but we was send only one wakeup
686 Descriptoin: panic in mds_open
687 Details : don't confuse mds_finish_transno() with PTR_ERR(-ENOENT)
692 Descriptoin: stuck in cache_remove_extent() or panic with accessing to already
694 Details : release lock refernce only after add page to pages list.
697 Frequency : always with long access acl
699 Descriptoin: mds can't pack reply with long acl.
700 Details : mds don't control size of acl but they limited by reint/getattr
705 Frequency : on remount
707 Description: external journal device not working after the remount
708 Details : clear dev_rdonly flag for external journal devices in
714 Description: shutdown vs evict race
715 Details : client_disconnect_export vs connect request race.
716 if client will evicted at this time - we start invalidate
717 thread without referece to import and import can be freed
723 Description: shrink LOV EAs before replying
724 Details : correctly adjust LOV EA buffer for reply.
729 Description: don't skip ost target if they assigned to file
730 Details : Drop slow OSCs if we can, but not for requested start idx.
731 This means "if OSC is slow and it is not the requested
732 start OST, then it can be skipped, otherwise skip it only
733 if it is inactive/recovering/out-of-space.
735 Severity : enhancement
737 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
740 Frequency : rare, need acl's on inode.
742 Description: client can't handle ost additional correctly
743 Details : if ost was added after client connected to mds client can have
744 hit lnet_try_match_md ... to big messages to wide striped files.
745 in this case need teach client to handle config events about add
746 lov target and update client max ea size at that event.
748 Severity : enhancement
750 Description: Update to sles9 kernel-2.6.5-7.314.
752 Severity : enhancement
754 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
757 Frequency : Create a symlink file with a very long name
759 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
760 Details : If there is no extra space in the request for early cancels,
761 ldlm_req_handles_avail() returns 0 instead of a negative value.
766 Description: mds is deadlocked
767 Details : in rare cases, inode in catalog can have i_no less than have parent
768 i_no, this produce wrong order for locking during open, and
769 parallel unlink can be lock open. this need teach mds_open to grab
770 locks in resource id order, not at parent -> child order.
772 Severity : enhancement
774 Description: Add /proc entry for import status
775 Details : The mdc, osc, and mgc import directories now have
776 an import directory that contains useful import data for debugging
779 Severity : enhancement
781 Description: Re-disable certain /proc logging
782 Details : Enable and disable client's offset_stats, extents_stats and
783 extents_stats_per_process stats logging on the fly.
786 Frequency : Only on FC kernels 2.6.22+
788 Description: oops in statahead
789 Details : Do not drop reference count for the dentry from VFS when lookup,
790 VFS will do that by itself.
792 Severity : enhancement
794 Description: Generic /proc file permissions
795 Details : Set /Proc file permissions in a more generic way to enable non-
796 root users operate on some /proc files.
800 Description: Hitting mdc_commit_close() ASSERTION
801 Details : Properly handle request reference release in
802 ll_release_openhandle().
806 Frequency : only patchless client
807 Description: add workaround for race between add/remove dentry from hash
809 Severity : enhancement
811 Description: Allow OST glimpses to return PW locks
815 Description: LBUG when llog conf file is full
816 Details : When llog bitmap is full, ENOSPC should be returned for plain
821 Description: Prevent import from entering FULL state when server in recovery
825 Description: service mount cannot take device name with ":"
826 Details : Only when device name contains ":/" will mount treat it as
832 Description: replace ptlrpcd with the statahead thread to interpret the async
833 statahead RPC callback
837 Frequency : on recovery
838 Description: I/O failures after umount during fail back
839 Details : if client reconnected to restarted server we need join to recovery
840 instead of find server handler is changed and process self eviction
841 with cancel all locks.
843 Severity : enhancement
845 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
849 Description: Kernel BUG tries to release flock
850 Details : Lustre does not destroy flock lock before last reference goes
851 away. So always drop flock locks when client is evicted and
852 perform unlock regardless of successfulness of speaking to MDS.
854 Severity : enhancement
856 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
858 Severity : enhancement
860 Description: Upcall on Lustre log has been dumped
861 Details : Allow for a user mode script to be called once a Lustre log has
862 been dumped. It passes the filename of the dumped log to the
863 script, the location of the script can be specified via
864 /proc/sys/lnet/debug_log_upcall.
869 Description: avoid messages about idr_remove called for id that is not allocated
870 Details : Move assigment s_dev for clustered nfs to end of initialization,
871 for avoid problem with error handling.
876 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
877 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
878 CONN_USED_HASH and this prodice warning when put connection again
884 Description: avoid ASSERTION(client_stat->nid_exp_ref_count == 0) failed
885 Details : release reference to stats when client disconnected, not
886 when export destroyed for avoid races when client destroyed
887 after main ost export.
891 Description: more cleanup in mds_lov
892 Details : add workaround for get valid ost count for avoid warnings about
893 drop too big messages, not init llog cat under semphore which
894 can be blocked on reconnect and break normal replay, fix access
897 Severity : enhancement
899 Description: Export bytes_read/bytes_write count on OSC/OST.
903 Description: Early reply size mismatch, MGC loses connection
904 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
905 the connect flags are properly negotiated.
909 Description: Properly propagate oinfo flags from lov to osc for statfs
910 Details : restore missing copy oi_flags to lov requests.
914 Description: exports in /proc are broken
915 Details : recreate /proc entries for clients when they reconnect.
917 Severity : enhancement
919 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
920 : l_getgroups(8), lst(8), routerstat(8)
921 Details : included man pages for llobdstat(8), llstat(8),
922 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
924 Severity : enhancement
926 Description: Implement lustre ll_show_options method.
928 Severity : enhancement
930 Description: Update to SLES9 kernel-2.6.5-7.312.
932 Severity : enhancement
934 Description: Update to RHEL4 kernel-2.6.9-67.0.22.EL.
938 Description: exports in /proc are broken
939 Details : recreate /proc entries for clients when they reconnect.
943 Description: don't fail open with -ERANGE
944 Details : if client connected until mds will be know about real ost count
945 get LOV EA can be fail because mds not allocate enougth buffer
950 Description: Resolve device initialization race
951 Details : Prevent proc handler from accessing devices added to the
952 obd_devs array but yet be intialized.
954 Severity : enhancement
956 Description: configure's --enable-quota should check the
957 : kernel .config for CONFIG_QUOTA
958 Details : configure is terminated if --enable-quota is passed but
959 : no quota support is in kernel
961 Severity : enhancement
963 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
965 Severity : enhancement
967 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
971 Frequency : rare, on PPC clients
972 Description: don't swab ost objects in response about directory, because
974 Details : bug similar bug 14856, but in different function.
976 Severity : enhancement
978 Description: lfs quota tool enhancement
979 Details : added units specifiers support for setquota, default to
980 current uid/gid for quota report, short quota stats by
981 default, nonpositional parameters for setquota, added
982 llapi_quotactl manual page.
984 Severity : enhancement
986 Description: *optional* service tags registration
987 Details : if the "service tags" package is installed on a Lustre node
988 When the filesystem is mounted, a local-node service tag will
989 be created. See http://inventory.sun.com/ for more information
990 about the Service Tags asset management system.
994 Description: Client runs out of low memory
995 Details : Consider only lowmem when counting initial number of llap pages
998 Frequency : occasional
1000 Description: add refcount for osc callbacks, so avoid panic on shutdown
1002 Severity : enhancement
1004 Description: Update to RHEL4 kernel-2.6.9-67.0.20.
1007 Frequency : testing only
1009 Description: sanity test 65a fails if stripecount of -1 is set
1010 Details : handle -1 striping on filesystem in ll_dirstripe_verify
1013 Frequency : only in unusual configurations
1015 Description: Kernel panic with find ost index.
1016 Details : lov_obd have panic if some OST's have sparse indexes.
1018 Severity : enhancement
1020 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
1023 Frequency : rarely, if filesystem is mounted with -o flock
1025 Description: do not process already freed flock
1026 Details : flock can possibly be freed by another thread before it reaches
1027 to ldlm_flock_completion_ast.
1030 Frequency : rarely, if filesystem is mounted with -o flock
1032 Description: LBUG during stress test
1033 Details : Need properly lock accesses the flock deadlock detection list.
1036 Frequency : rarely, if binaries are being run from Lustre
1038 Description: oops in page fault handler
1039 Details : kernel page fault handler can return two special 'pages' in
1040 error case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
1043 Frequency : rarely, during shutdown
1045 Description: timeout with invalidate import.
1046 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
1047 should be handled by ptlrpcd. This produce long age waiting and
1048 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
1054 Description: ASSERTION(CheckWriteback(page,cmd)) failed
1055 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
1059 Frequency : only with broken builds/installations
1061 Description: no LBUG if lquota.ko and fsfilt_ldiskfs.ko are different versions
1062 Details : just return an error to a user, put a console error message
1064 Severity : enhancement
1066 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
1068 Severity : enhancement
1070 Description: Update to RHEL4 kernel-2.6.9-67.0.15.
1072 Severity : enhancement
1074 Description: enable MGS and MDT services start separately
1075 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
1076 a co-located MGS without starting the MGS, which is a complement
1077 to 'nosvc' mount option.
1080 Frequency : always, on big-endian systems
1082 Description: cleanup in ptlrpc code, related to PPC platform
1083 Details : store magic in native order avoid panic's in recovery on PPC
1084 node and forbid from this error in future. Also fix posibily
1085 of twice swab data. Fix get lov striping to userpace.
1088 Frequency : rarely, if replay get lost on server
1090 Description: server incorrectly drop resent replays lead to recovery failure.
1091 Details : do not drop replay according to msg flags, instead we check the
1092 per-export recovery request queue for duplication of transno.
1095 Frequency : after recovery
1097 Description: precreate to many object's after del orphan.
1098 Details : del orphan st in oscc last_id == next_id and this triger growing
1099 count of precreated objects. Set flag LOW to skip increase count
1100 of precreated objects.
1103 Frequency : rare, on clear nid stats
1105 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
1106 Details : when clean nid stats sometimes try destroy live entry,
1107 and this produce panic in free.
1110 Frequency : occasionally since 1.6.4
1112 Description: Stack overflow during MDS log replay
1113 Details : ease stack pressure by using a thread dealing llog_process.
1116 Frequency : very rare
1118 Description: MDT cannot be unmounted, reporting "Mount still busy"
1119 Details : Mountpoint references were being leaked during open reply
1120 reconstruction after an MDS restart. Drop mountpoint reference
1121 in reconstruct_open() and free dentry reference also.
1126 Description: wait until IO finished before start new when do lock cancel.
1127 Details : VM protocol want old IO finished before start new, in this case
1128 need wait until PG_writeback is cleared until check dirty flag
1129 and call writepages in lock cancel callback.
1134 Description: mds_mfd_close() ASSERTION(rc == 0)
1135 Details : In mds_mfd_close(), we need protect inode's writecount change
1136 within its orphan write semaphore to prevent possible races.
1139 Frequency : rare, on shutdown ost
1141 Description: don't hit live lock with umount ost.
1142 Details : shrink_dcache_parent can be in long loop with destroy dentries,
1143 use shrink_dcache_sb instead.
1146 Frequency : only when echo_client is used
1148 Description: don't panic with use echo_client
1149 Details : echo client pass NULL as client nid pointer and this produce NULL
1150 pointer dereference.
1153 Frequency : Always on 32-bit PowerPC systems
1155 Description: fix build on PPC32
1156 Details : compile code with -m64 flag produce wrong object file for PPC32.
1161 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1162 Details : In reconstruct_* functions, LASSERTs on both the data supplied
1163 by a client, and the data on disk are dangerous and incorrect.
1164 Change them with client eviction.
1166 Severity : enhancement
1168 Description: skiplist implementation simplification
1169 Details : skiplists are used to group compatible locks on granted list
1170 that was implemented as tracking first and last lock of each lock
1171 group the patch changes that to using doubly linked lists
1175 Description: delete compatibility for 32bit qdata
1176 Details : as planned, when lustre is beyond b1_8, lquota won't support 32bit
1177 qunit. That means servers of b1_4 and servers of b1_8 can't be
1178 used together if users want to use quota.
1181 Frequency : only with administrator action
1183 Description: mount failure if config log has invalid conf_param setting
1184 Details : If administrator specified an incorrect configuration parameter
1185 with "lctl conf_param" this would cause an error during future
1186 client mounts. Instead, ignore the bad configuration parameter.
1189 Frequency : blocks per group < blocksize*8 and uninit_groups is enabled
1191 Description: ldiskfs error: XXX blocks in bitmap, YYY in gd
1192 Details : If blocks per group is less than blocksize*8, set rest of the
1196 Frequency : Application do stride read on lustre
1198 Description: The read performance will drop a lot if the application does
1200 Details : Because the stride_start_offset are missing in stride read-ahead,
1201 it will cause clients read a lot of unused pages in read-ahead,
1202 then the read-performance drops.
1206 Description: more ldlm soft lockups
1207 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1208 starve other threads from the resource lock for a long time in
1209 case of long waiting queue, so change the debug level from
1210 D_OTHER to the less frequently used D_INFO.
1212 Severity : enhancement
1214 Description: add -gid, -group, -uid, -user options to lfs find
1216 Severity : enhancement
1218 Description: ll_recover_lost_found_objs - recover objects in lost+found
1219 Details : OST corruption and subsequent e2fsck can leave objects in the
1220 lost+found directory. Using the "ll_recover_lost_found_objs"
1221 tool, these objects can be retrieved and data can be salvaged
1222 by using the object ID saved in the fid EA on each object.
1227 Description: this bug _only_ happens when inode quota limitation is very low
1228 (less than 12), so that inode quota unit is 1 at initialization.
1229 Details : if remaining quota equates 1, it is a sign to demonstate that quota
1230 is effective now. So least quota qunit should be 2.
1234 Description: Hung threads in invalidate_inode_pages2_range
1235 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1236 one is completed. As a result, some RPCs are stuck in the queue
1241 Description: Procfs and llog threads access destoryed import sometimes.
1242 Details : Sync the import destoryed process with procfs and llog threads by
1243 the import refcount and semaphore.
1247 Description: mds fails to respond, threads stuck in ldlm_completion_ast
1248 Details : Sort source/child resource pair after updating child resource.
1253 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1254 Details : If insertion of an extent fails, then discard the inode
1255 preallocation and free data blocks else it can lead to duplicate
1260 Description: don't always update ctime in ext3_xattr_set_handle()
1261 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle()
1262 In some cases the ctime should not be updated, for example for
1263 2.0->1.8 compatibility it is necessary to delete an xattr and it
1264 should not update the ctime.
1268 Description: add quota statistics
1269 Details : 1. sort out quota proc entries and proc code.
1270 2. add quota statistics
1275 Description: quotas are not honored with O_DIRECT
1276 Details : all writes with the flag O_DIRECT will use grants which leads to
1277 this problem. Now using OBD_BRW_SYNC to guard this.
1281 Bugzilla : 15713/16362
1282 Description: Assertion in iopen_connect_dentry in 1.6.3
1283 Details : looking up an inode via iopen with the wrong generation number can
1284 populate the dcache with a disconneced dentry while the inode
1285 number is in the process of being reallocated. This causes an
1286 assertion failure in iopen since the inode's dentry list contains
1287 both a connected and disconnected dentry.
1291 Description: assertion failure in ldlm_handle2lock()
1292 Details : fix a race between class_handle_unhash() and class_handle2object()
1293 introduced in lustre 1.6.5 by bug 13622.
1295 Severity : enhancement
1297 Description: superblock lock contention with many SMP cores on one client
1298 Details : several client filesystem locks were highly contended on SMP
1299 NUMA systems with 8 or more cores. Per-CPU datastructures
1300 and more efficient locking implemented to reduce contention.
1305 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1306 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1311 Description: oops in sd_iostats_seq_show()
1312 Details : unloading/reloading the scsi low level driver triggers a kernel
1313 bug when trying to access the sd iostat file.
1318 Description: Kernel panics during QLogic driver reload
1319 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1320 patch, causing memory corruption.
1325 Description: journal_dev option does not work in b1_6
1326 Details : pass mount option during pre-mount.
1328 Severity : enhancement
1330 Description: Add a FIEMAP(FIle Extent MAP) ioctl for ldiskfs
1331 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1332 extent information of a file. It can be used to map logical blocks
1333 in a file to physical blocks in the block device.
1336 Frequency : only with adaptive timeout enabled
1338 Description: DEBUG_REQ() bad paging request
1339 Details : ptlrpc_at_recv_early_reply() should not modify req->rq_repmsg
1340 because it can be accessed by reply_in_callback() without the
1344 Frequency : only on Cray X2
1346 Description: X2 build failures
1347 Details : fix build failures on Cray X2.
1351 Description: xid & resent requests
1352 Details : Initialize RPC XID from clock at startup (randomly if clock is
1357 Description: quota recovery deadlock during mds failover
1358 Details : This patch includes att18982, att18236, att18237 in bz14840.
1360 1. fix osts hang when mds does failover with quotaon
1361 2. prevent watchdog storm when osts threads wait for the
1366 Description: kernel panic on racer
1367 Details : Do not access dchild->d_inode when IS_ERR(dchild) is true.
1369 Severity : enhancement
1371 Description: Add lustre_start utility to start or stop multiple Lustre servers
1376 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1377 Details : In case of memory pressure, list_del() can be called twice on
1378 req->rq_history_list, causing a kernel oops.
1382 Description: (ptllnd_peer.c:557:kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1383 Details : fix stack overflow in the distributed lock manager by defering
1384 export eviction after a failed ast to the elt thread instead of
1385 handling it in the dlm interpret routine.
1387 Severity : enhancement
1389 Description: More exported tunables for mballoc
1390 Details : Add support for tunable preallocation window and new tunables for
1391 large/small requests
1395 Description: Detect corruption of block bitmap and checking for preallocations
1396 Details : Checks validity of on-disk block bitmap. Also it does better
1397 checking of number of applied preallocations. When corruption is
1398 found, it turns filesystem readonly to prevent further corruptions.
1402 Frequency : only for big-endian servers
1403 Description: Check for big-endian system while mounting fs with extents feature
1404 Details : Mounting a filesystem with extents feature will fail on big-endian
1405 systems since ext3-based ldiskfs is not supported on big-endian
1406 systems. Can be over-ridden with "bigendian_extents" mount option.
1410 Description: Excessive recovery window
1411 Details : With AT enabled, the recovery window can be excessively long (6000+
1412 seconds). To address this problem, we no longer use
1413 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
1414 timeout no longer depends on the service time, it is set to
1415 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
1416 time via pb_service_time.
1419 Descriptoin: Don't sync journal after every i/o
1420 Details : Implement write RPC replay to allow server replies for write RPCs
1421 before data is on disk. However, this feature is disabled by
1422 default since some issues leading to data corruptions have been
1423 found during recovery (e.g. bug 19128). This feature can be enabled
1424 by running the following command on the OSSs:
1425 lctl set_param obdfilter.*.sync_journal=0
1429 Description: Watchdog triggered on MDS failover
1430 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
1431 the OSTs know that the MDS "UUID" can be reused for the same export
1432 from a different NID, so we do not need to wait for the export to
1437 Descriptoin: Lustre detected file system corruption with inode out of bounds
1438 Details : don't update i_size on MDS_CLOSE for directories. This causes
1439 directory corruptions on the MDT.
1441 -------------------------------------------------------------------------------
1443 2008-05-26 Sun Microsystems, Inc.
1445 * Support for kernels:
1446 2.6.5-7.311 (SLES 9),
1447 2.6.9-67.0.7.EL (RHEL 4),
1448 2.6.16.54-0.2.5 (SLES 10),
1449 2.6.18-53.1.14.el5 (RHEL 5),
1450 2.6.22.14 vanilla (kernel.org)
1451 * Client support for unpatched kernels:
1452 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1453 2.6.16 - 2.6.22 vanilla (kernel.org)
1454 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
1455 we do not recommend using patchless RHEL4 clients with kernels
1456 prior to 2.6.9-55EL (RHEL4U5).
1457 * Recommended e2fsprogs version: 1.40.7-sun1
1458 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1459 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
1460 removed cwd "./" (refer to Bugzilla 14399).
1461 * A new quota file format has been introduced in 1.6.5.
1462 The format conversion from prior releases is handled transparently,
1463 but releases older than 1.4.12/1.6.5 will not understand this new
1464 format. The automatic format conversion can be avoided by running
1465 the following command on the MDS before upgrading:
1466 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
1467 For more information, please refer to bugzilla 13904.
1471 Description: quota performance fix
1472 Details : quota data is written in journalled mode instead of ordered to
1473 increase performance
1477 Description: lfs support for human-readable quota grace time strings
1478 Details : lfs setquota -t and lfs quota -t represent quota grace times
1479 in "XXwXXdXXhXXmXXs" format instead of large values in seconds
1482 Frequency : always with o2ib 1.3 and sles10
1484 Description: fix build with SLES10 and o2ib v3.
1485 Details : sles10 uses diffrent name for Module.symver file but configure
1486 assume this file has same name on RHEL/SLES/vanila kernels.
1489 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1491 Description: MDS may lose file striping (and hence file data) in some cases
1492 Details : If there are additional extended attributes stored on the MDS,
1493 in particular ACLs, SELinux, or user attributes (if user_xattr
1494 is specified for the client mount options) then there is a risk
1495 of attribute loss. Additionally, the Lustre file striping
1496 needs to be larger than default (e.g. striped over all OSTs),
1497 and an additional attribute must be stored initially in the
1498 inode and then increase in size enough to be moved to the
1499 external attribute block (e.g. ACL growing in size) for file
1502 Severity : enhancement
1504 Description: add message levels for liblustreapi
1509 Description: MDT cannot be unmounted, reporting "Mount still busy"
1510 Details : Mountpoint references were being leaked during open reply
1511 reconstruction after an MDS restart. Drop mountpoint reference
1512 in reconstruct_open() and free dentry reference also.
1517 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
1518 Details : Move the 'good_osts' check before the 'total_bavail' check. This
1519 will result in an -EAGAIN and in the exit call path we call
1520 alloc_rr() which will with increasing aggressiveness attempt to
1521 aquire precreated objects on the minimum number of required OSCs.
1525 Description: Use old size assignment to avoid deadlock
1526 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
1527 the scheduling while holding a spinlock. We do not need locking
1528 for size in ll_update_inode() because size is only updated from
1529 the MDS for directories or files without objects, so there is no
1530 other place to do the update, and concurrent access to such inodes
1531 are protected by the inode lock.
1535 Description: Use __u64 instead of int for valid bits
1539 Description: resolve "_IOWR redefined" build error on SLES10
1543 Description: dump the memory debugging after all modules are unloaded to
1544 suppress false negative in conf_sanity test 39
1548 Description: the recovery timer never expires
1549 Details : for new client connect request, the recovery timer should not be
1550 reset, otherwise recovery timer will never expired, if the old
1551 client never come. Only old client connect and first connection
1552 req should trigger recovery timer reset.
1556 Description: the min numbers of lproc stats are wrong
1557 Details : adding a new constant LC_MIN_INIT and use it for initialization
1561 Frequency : always with interactive lfs
1563 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1566 Frequency : with multiple concurrent readdir processes in same directory
1567 Bugzilla : 15406, 15169, 15175
1568 Description: misc fixes for directory readahead.
1569 Details : prevent previous statahead async RPC callback from processing the
1570 current "statahead_info", race condition between async RPC callback
1571 add dentry into dentry hash table and "ls" thread revalidate such
1572 dentry, statahead his/miss control for hidden items, and so on.
1574 Severity : enhancement
1576 Description: build kernel-ib packages for OFED 1.3 in our release cycle
1580 Description: incore types cleaning in quota code (with respect to 64-bit limits)
1581 Details : several u32 variables declarations are replaced with u64 declarations
1586 Description: fix SLES kernel versioning
1587 Details : the kernel version for our SLES 10 kernel did not include a "-"
1588 before the "smp" at the end. while this was not a problem in
1589 general, it did mean that software trying to use the kernel
1590 version to try to detect a vendor specific kernel would fail.
1591 this was most evident by the OFED build scripts.
1596 Description: Don't update lov_desc members until making sure they are valid
1597 Details : When updating lov_desc members via proc fs, need fix their
1598 validities before doing the real update.
1601 Frequency : very rare
1603 Description: don't put request into delay list while invalidate in flight.
1604 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
1605 import in flight. this produce timeout for invalidate and sometimes
1606 can cause stale data.
1608 Severity : enhancement
1610 Description: Update kernel to SLES9 2.6.5-7.311.
1612 Severity : enhancement
1614 Description: Update kernel to RHEL4 2.6.9-67.0.7.
1619 Frequency : on PPC only
1620 Description: not convert ost objects for directory because it's not exist.
1621 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
1623 Severity : enhancement
1625 Description: Fix warnings with compile liblustre at sles10/rhel5 which have
1626 __u64 as usingied long long type.
1629 Frequency : rare, on shutdown
1631 Description: race process ast vs remove callback
1632 Details : removing callback before disconnect import open race with
1633 processing callback.
1635 Severity : enhancement
1637 Description: Update kernel to SLES9 2.6.5-7.311.
1639 Severity : enhancement
1641 Description: Files open for execute are not marked busy on SLES10
1642 Details : Add FMODE_EXEC to SLES10 SP1 server kernel series.
1644 Severity : enhancement
1646 Description: Add server support for vanilla-2.6.22.14.
1649 Frequency : occasional
1651 Description: Avoid lov_create() getting stuck in obd_statfs_rqset()
1652 Details : If an OST is down the MDS will hang indefinitely in
1653 obd_statfs_rqset() waiting for the statfs data. While for
1654 MDS QOS usage of statfs, it should not stuck in waiting.
1656 Severity : enhancement
1658 Description: Disable adaptive timeouts by default
1661 Frequency : on network error
1663 Description: panic with double free request if network error
1664 Details : mdc_finish_enqueue is finish request if any network error occuring,
1665 but it's true only for synchronus enqueue, for async enqueue
1666 (via ptlrpcd) this incorrect and ptlrpcd want finish request
1670 Frequency : rare, on recovery
1672 Description: read procfs can produce deadlock in some situation
1673 Details : Holding lprocfs lock with send rpc can produce block for destroy
1674 obd objects and this also block reconnect with -EALREADY.
1675 This isn't fix all lprocfs bugs - but make it rare.
1677 Severity : enhancement
1679 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
1682 Frequency : frequent on X2 node
1684 Description: mdc_set_open_replay_data LBUG
1685 Details : Set replay data for requests that are eligible for replay.
1690 Description: lustre_mgs: operation 101 on unconnected MGS
1691 Details : When MGC is disconnected from MGS long enough, MGS will evict the
1692 MGC, and late on MGC cannot successfully connect to MGS and a lot
1693 of the error messages complaining that MGS is not connected.
1696 Frequency : rare, depends on device drivers and load
1698 Description: MDS or OSS nodes crash due to stack overflow
1699 Details : Code changes in 1.6.4 increased the stack usage of some functions.
1700 In some cases, in conjunction with device drivers that use a lot
1701 of stack, the MDS (or possibly OSS) service threads could overflow
1702 the stack. One change which was identified to consume additional
1703 stack has been reworked to avoid the extra stack usage.
1705 Severity : enhancement
1707 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
1709 Severity : enhancement
1711 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
1713 Severity : enhancement
1715 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
1717 Serverity : enhancement
1719 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
1722 Frequency : rare on shutdown OST
1724 Description: Don't allow skipping OSTs if index has been specified.
1725 Details : Don't allow skipping OSTs if index has been specified, make locking
1726 in internal create lots better.
1731 Description: ASSERTION(!PageDirty(page)) failed
1732 Details : Wrong check could lead to an assertion failure under specific
1738 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
1739 Details : access to bitfield in structure is always rounded to long
1740 and this produce problem with not atomic change any bit.
1745 Description: Lustre make rpms failed.
1746 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
1747 builds Lustre rpms from tarball.
1749 Severity : enhancement
1751 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
1754 Frequency : rare on shutdown OST
1756 Description: If llog cancel was not send before clean_exports phase, this can
1757 produce deadlock in llog code.
1758 Details : If llog thread has last reference to obd and call class_import_put
1759 this produce deadlock because llog_cleanup_commit_master wait when
1760 last llog_commit_thread exited, but this never success because was
1761 called from llog_commit_thread.
1764 Frequency : only if OST index is skipped
1766 Description: NULL lov_tgts causing MDS oops
1767 Details : more safe checks for NULL lov_tgts for avoid oops.
1769 Severity : enhancement
1771 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
1773 Severity : enhancement
1775 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
1780 Description: make mgs_setparam() handle fsname containing dash
1781 Details : fsname containing a dash does not work with lctl conf_param
1783 Severity : enhancement
1785 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
1788 Frequency : rare, in recovery and (or) destroy lovobjid file.
1790 Description: rewrite lov objid code.
1791 Details : Cleanup for lov objid code, remove scability problems and wrong
1792 locking. Fix sending last_id into ost.
1794 Severity : enhancement
1796 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
1798 Severity : enhancement
1800 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
1801 Details : Use d_move_locked instead of __d_move.
1804 Frequency : rare, at shutdown
1806 Description: access already free / zero obd_namespace.
1807 Details : if client_disconnect_export was called without force flag set,
1808 and exist connect request in flight, this can produce access to
1809 NULL pointer (or already free pointer) when connect_interpret
1810 store ocd flags in obd_namespace.
1813 Frequency : only at startup
1815 Description: not alloc memory with spinlock held.
1816 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1817 if any spinlock held.
1822 Description: lfs find does not continue on file error
1823 Details : Continue other files processing when a file/dir is absent.
1827 Description: Inconsistent usage of lustre_pack_reply()
1828 Details : Standardize the usage of lustre_pack_reply() such that it
1829 always generate a CERROR on failure.
1832 Frequency : very rare
1834 Description: Fix replay if there is an un-replied request and open
1835 Details : In some cases, older replay request will revert the
1836 mcd->mcd_last_xid on MDS which is used to record the client's
1837 latest sent request.
1839 Severity : enhancement
1841 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1843 Severity : enhancement
1845 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1847 Severity : enhancement
1849 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1851 Severity : enhancement
1853 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
1855 Severity : enhancement
1857 Description: Build SLES10 patchless client fails
1858 Details : The configure was broken by run ./configure with
1859 --with-linux-obj=.... argument for patchless client. When the
1860 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1861 can't search header adequately. Use absolute path such as
1862 -I($LINUX)/include instead.
1864 Severity : enhancement
1866 Description: Lustre Page Accounting
1867 Details : New macros for page alloc and free which enable accounting
1868 of page allocation of Lustre. Use percpu counters to store memory
1869 and page statistics.
1872 Frequency : only if debugging is disabled
1874 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1875 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1876 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1877 these macros and replace them with their definitions instead.
1882 Description: interrupt oig_wait produce painc on resend.
1883 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1884 private set, and this produce situation when rq_ptlrpcd_data not
1885 copyed to new allocated request and triggered LBUG on assert
1886 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1887 ptlrpcd set if request is changed and can be safety changed to use
1890 Severity : enhancement
1892 Description: organize the server-side client stats on per-nid basis
1893 Details : Change the structure of stats under obdfilter and mds to
1901 The "uuid"s file would list the uuids of _active_ exports.
1902 And the clear entry is to clear all stats and stale nids.
1907 Description: Processes looping in ll_readdir() on Lustre clients finally causing
1908 a full node pseudo-hang
1909 Details : Concurrent access to the same directory from multiple clients with
1910 intensive file creation/removal can cause a client node to spin in
1911 ll_readdir(). i_version must be increased every time the lock is
1912 cancelled to ensure a revalidate is done.
1917 Description: touch file failed when fs is not full
1918 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1919 otherwise we can get ENOSP while fs is not full.
1922 Frequency : only for Cray XT3
1923 Bugzilla : 12829/13455
1924 Description: Changing primary group doesn't change the group lustre assigns to
1926 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1927 overridden with the primary group provided by the group upcall,
1928 whereas the supplied fsgid can be trusted if it is in the list of
1929 supplementary groups returned by the group upcall.
1931 Severity : enhancement
1933 Description: Root Squash Functionality
1934 Details : Implementation of NFS-like root squash capability. Specifically,
1935 don't allow someone with root access on a client node to be able
1936 to manipulate files owned by root on a server node.
1938 Severity : enhancement
1940 Description: Slow trucate/writes to huge files at high offsets.
1941 Details : Directly associate cached pages to lock that protect those pages,
1942 this allows us to quickly find what pages to write and remove
1943 once lock callback is received.
1948 Description: Too many locks accumulating on client during NFS usage
1949 Details : mds_open improperly used accmode to find out access mode to a
1950 file. Also mdc_intent_lock logic to find out if we already have
1951 lock similar to just received was flawed since introduction of
1952 skiplists - locks are now added to the front of the granted
1957 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
1959 Details : In obd setup failure handler, obd_stopping will not necessarily to
1960 be 1, and obd_set_up should also be checked to make sure whether
1961 obd is completely setup.
1963 Severity : enhancement
1965 Description: Allow masking D_WARNING, D_ERROR messages from console
1966 Details : Console messages can now be disabled via lnet.printk.
1971 Description: User code with malformed file open parameter crashes client node
1972 Details : Before packing join_file req, all the related reference should be
1973 checked carefully in case some malformed flags cause fake join_file
1979 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
1980 grain of quota doing harm to quota's accuracy
1981 Details : qunit size will be changed when quota limitation is too low/high;
1982 record the pending quota write in order to get more accureate
1983 quota; delete the patch for bug12588, which is unnecessary when
1984 this patch is landed. This bug also contains fixes for bug 14526,
1985 14299, 14601 and 13794.
1989 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1990 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1991 mode is equal to the granted mode. However, it should also check
1992 that LDLM_FL_CP_REQD is not set.
1996 Description: Heavy nfs access might result in deadlocks
1997 Details : After ELC code landed, it is now improper to enqueue any mds
1998 locks under och_sem, because enqueue might want to decide to
1999 cancel open locks for same inode we are holding och_sem for.
2003 Description: 35% write performance drop with ldiskfs2 when quotas are on
2004 Details : Enable ext3 journalled quota by default to improve performance
2005 when quotas are turned on.
2009 Description: Client eviction while running blogbench
2010 Details : A lot of unlink operations with concurrent I/O can lead to a
2011 deadlock causing evictions. To address the problem, the number of
2012 oustanding OST_DESTROY requests is now throttled to
2013 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
2014 callbacks are processed in priority.
2017 Frequency : RHEL4 only
2019 Description: mkfs is very slow on IA64/RHEL4
2020 Details : A performance regression has been discovered in the MPT Fusion
2021 driver between versions 3.02.73rh and 3.02.99.00rh. As a
2022 consequence, we have downgraded the MPT Fusion driver in the RHEL4
2023 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
2026 Frequency : PPC/PPC64 only
2028 Description: conflicts between asm-ppc64/types.h and lustre_types.h
2029 Details : fix duplicated definitions between asm-ppc64/types.h and
2030 lustre_types.h on PPC.
2033 Frequency : PPC/PPC64 only
2035 Description: asm-ppc/segment.h does not exist
2036 Details : fix compile issue on PPC.
2040 Description: data checksumming impacts single node performance
2041 Details : add support for several checksum algorithms. Currently, CRC32 and
2042 Adler-32 are supported. The checksum type can be changed on the fly
2043 through /proc/fs/lustre/osc/*/checksum_type.
2047 Description: use adler32 for page checksums
2048 Details : when available, use the Adler-32 algorithm instead of CRC32 for
2053 Description: better handle error messages in extents code
2055 Severity : enhancement
2057 Description: SNMP support enhancement
2058 Details : Adding total number of sampled request for an MDS node in snmp
2061 Severity : enhancement
2063 Description: Optimize ldlm waiting list processing for PR extent locks
2064 Details : When processing waiting list for read extent lock and meeting read
2065 lock that is same or wider to it that is not contended, skip
2066 processing rest of the list and immediatelly return current
2067 status of conflictness, since we are guaranteed there are no
2068 conflicting locks in the rest of the list.
2072 Description: Time out and refuse to reconnect
2073 Details : When the failover node is the primary node, it is possible
2074 to have two identical connections in imp_conn_list. We must
2075 compare not conn's pointers but NIDs, otherwise we
2076 can defeat connection throttling.
2080 Description: Client not clear own cache if answer to reconnect is lost.
2081 Details : Client gets evicted from server. Now client also thinks it is
2082 disconnected (or gets ENOTCONN on its operation) and decides to
2083 reconnect. Server receives reconnect message, but cannot find
2084 export. New export is created that is fully valid (new cookie!),
2085 but reply is lost and not reported to client. Client reconnects
2086 again and gets back a just-created connection, but it is not new
2087 so client thinks it was not evicted and does not do recovery.
2091 Description: Detect stride IO mode in read-ahead
2092 Details : When a client does stride read, read-ahead should detect that and
2093 read-ahead pages according to the detected stride pattern.
2097 Description: build for x2 fails
2098 Details : fix compile issue on Cray systems.
2100 Severity : enhancement
2102 Description: implement readv/writev APIs(aio_read/aio_writes in newer kernels)
2103 Details : This greatly improves speed of NFS writes on 2.6 kernels.
2106 Frequency : only on PPC/SLES10
2108 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
2109 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
2110 be defined. Add a hack in mkfs_lustre.c to work around this compile
2115 Description: LASSERT on MDS when client holding flock lock dies
2116 Details : ldlm pool logic depends on number of granted locks equal to
2117 number of released locks which is not true for flock locks, so
2118 just exclude such locks from consideration.
2122 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
2123 Details : Use fsfilt_sync() for both the whole filesystem sync and
2124 individual file sync to eliminate dangerous inode locking
2125 with I_LOCK that can lead to a deadlock.
2129 Description: Update an obsolete wirecheck.c generator
2130 Details : Update wirecheck.c/wirehdr.c and regenerate wiretest.c
2134 Description: Client can panic on open sometimes
2135 Details : It is possible that we try to free already freed request in
2136 ll_file_open in some error cases when we send request from
2141 Description: performance in 1.6.3
2142 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
2143 sure that 1MB requests are not fragmented by the block layer.
2147 Description: LDLM soft lockups - improvement
2148 Details : It is be possible to send the lock handle along with each read
2149 or write request because the client is already doing a lock match
2150 itself so there isn't any reason the OST should have to re-do that
2156 Description: lfs quota fails with deactivated OSTS
2157 Details : With this patch, three improvements are included:
2158 1. detete the softlimit in mds and osts when use "lfs quota".
2159 2. display the inaccurate data in the output of "lfs quota".
2160 3. try to get quota info when "lfs quota" is executed.
2165 Description: Extent locks not granted with no conflicts sometimes.
2166 Details : When race occurs in glimpse handler and nothing is returned,
2167 we do not reprocess the queue after lock cancel, and that leads
2168 to a stall until next activity on a resource
2171 Frequency : failover with quotaon
2173 Description: during mds failovers with quota on, OSTs got into deadlock state
2174 and causing dumpstack.
2175 Details : for every quota slave, at any time, there is only one quota req
2176 is sent to quota master for every uid/gid. Before that quota req
2177 returns, all the thread relative to the same uid/gid will wait.
2178 So if the quota req is lost because mds failovers or any other
2179 reasons, this bug will be hit. Now, dqacq_interpret() will handle
2180 quota reqs who time out.
2182 Severity : enhancement
2185 Description: when quota slave checks if quota is enough, there is an unnecessary
2187 Details : place this wait on necessary place instead of always waiting.
2189 --------------------------------------------------------------------------------
2191 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
2193 * Support for kernels:
2194 2.6.5-7.286 (SLES 9),
2195 2.6.9-55.0.9.EL (RHEL 4),
2196 2.6.16.53-0.8 (SLES 10),
2197 2.6.18-8.1.14.el5 (RHEL 5),
2198 2.6.18.8 vanilla (kernel.org)
2199 * Client support for unpatched kernels:
2200 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2201 2.6.16 - 2.6.22 vanilla (kernel.org)
2202 * Due to recently discovered recovery problems, we do not recommend
2203 using patchless RHEL 4 clients with this or any earlier release.
2204 * Recommended e2fsprogs version: 1.40.2-cfs1
2205 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2208 Frequency : occasional
2210 Description: MDS hang or stay in waiting lock
2211 Details : If client receive lock with CBPENDING flag ldlm need send lock
2212 cancel as separate rpc, to avoid situation when cancel request
2213 can't processed due all i/o threads stay in wait lock.
2216 Frequency : occasional
2218 Description: Do not fail import if osc_interpret_create gets -EAGAIN
2219 Details : If osc_interpret_create got -EAGAIN it immediately exits and
2220 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
2221 oscc_has_objects and see OSC has no objests and call
2222 oscc_internal_create to resend create request.
2224 Severity : enhancement
2226 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
2227 Details : Update which_patch & target file for SLES10 latest kernel.
2229 Severity : enhancement
2231 Description: add --type and --size parameters to lfs find
2232 Details : Enhance lfs find by adding filetype and filesize parameters. Also
2233 multiple OBDs can now be specified for the --obd option.
2235 Severity : enhancement
2237 Description: eliminate client locks in face of contention
2238 Details : file contention detection and lockless i/o implementation
2239 for contended files.
2241 Severity : enhancement
2243 Description: Remove client patches from SLES 10 kernel.
2244 Details : This causes SLES 10 clients to behave as patchless clients
2245 even on a Lustre-patched (server) kernel.
2247 Severity : enhancement
2249 Description: use i_size_read and i_size_write in 2.6 port
2250 Details : replace inode->i_size access with i_size_read/write()
2253 Frequency : when removing large files
2255 Description: scheduling issue during removal of large Lustre files
2256 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
2257 It causes scheduling issues when removing large files (17TB in the
2263 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
2264 Details : Impossible (EPROTO is returned) to access a directory that has a
2265 non-default striping and ACLs.
2268 Frequency : only on ppc
2270 Description: /proc/fs/lustre/devices broken on ppc
2271 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
2272 We should make sure the type of 'index' is loff_t and then cast
2273 explicitly as needed below. Do not assign an explicitly cast
2277 Frequency : only for rhel5
2279 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
2280 Details : Modify the target file & which_kernel.
2283 Frequency : if the uninit_groups feature is enabled on ldiskfs
2285 Description: e2fsck reports "invalid unused inodes count"
2286 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
2287 feature and only a single inode is created in a group then the
2288 "bg_unused_inodes" count is incorrectly updated. Creating a
2289 second inode in that group would update it correctly.
2294 Description: buffer overruns could theoretically occur
2295 Details : llapi_semantic_traverse() modifies the "path" argument by
2296 appending values to the end of the origin string, and a buffer
2297 overrun may occur. Adding buffer overrun check in liblustreapi.
2299 Severity : enhancement
2301 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
2304 Frequency : only if filesystem is inconsistent
2306 Description: handle "serious error: objid * already exists" more gracefully
2307 Details : If LAST_ID value on disk is smaller than the objects existing in
2308 the O/0/d* directories, it indicates disk corruption and causes an
2309 LBUG(). If the object is 0-length, then we should use the existing
2310 object. This will help to avoid a full fsck in most cases.
2312 Severity : enhancement
2314 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
2315 Details : Modify vm-tunables-rhel4.patch.
2317 Severity : enhancement
2319 Description: Kernel config for 2.6.18-vanilla.
2320 Details : Modify targets/2.6-vanilla.target.in.
2321 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
2322 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
2323 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
2324 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
2327 Frequency : occasional
2329 Description: improve handling recoverable errors
2330 Details : If request processed with error which can be recoverable on server
2331 request should be resend, otherwise page released from cache and
2337 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
2338 Details : Modify target file & which_patch.
2339 A flaw was found in the IA32 system call emulation provided
2340 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
2341 value could be stored in the %RAX register, which could trigger an
2342 out-of-bounds system call table access. An untrusted local user
2343 could exploit this flaw to run code in the kernel
2344 (ie a root privilege escalation). (CVE-2007-4573).
2348 Description: change order of libsysio includes
2349 Details : '#include sysio.h' should always come before '#include xtio.h'
2351 Severity : enhancement
2353 Description: adapt the lustre_config script to support the upgrade case
2354 Details : Add "-u" option for lustre_config script to support upgrading 1.4
2355 server targets to 1.6 in parallel.
2360 Description: To avoid grant space > avaible space when the disk is almost
2361 full. Without this patch you might see the error "grant XXXX >
2362 available" or some LBUG about grant, when the disk is almost
2364 Details : In filter_check_grant, for non_grant cache write, we should
2365 check the left space by if (*left > ungranted + bytes), instead
2366 of (*left > ungranted), because only we are sure the left space
2367 is enough for another "bytes", then the ungrant space should be
2368 increase. In client, we should update cl_avail_grant only there is
2369 OBD_MD_FLGRANT in the reply.
2374 Description: Update RHEL 4 kernel to fix local root privilege escalation.
2375 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
2376 described in CVE-2007-4573. This problem could allow untrusted
2377 local users to gain root access.
2380 Frequency : when using O_DIRECT and quotas
2382 Description: Incorrect file ownership on O_DIRECT output files
2383 Details : block usage reported by 'lfs quota' does not take into account
2384 files that have been written with O_DIRECT.
2388 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2389 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2390 window to 0. If it does hit cache miss, and read-pattern is still
2391 stride-io mode, does not reset the stride window, but also does not
2392 increase the stride window length in this case.
2394 --------------------------------------------------------------------------------
2396 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
2398 * Support for kernels:
2399 2.6.5-7.286 (SLES 9),
2400 2.6.9-55.0.2.EL (RHEL 4),
2401 2.6.16.46-0.14 (SLES 10),
2402 2.6.18-8.1.8.el5 (RHEL 5),
2403 2.6.18.8 vanilla (kernel.org)
2404 * Client support for unpatched kernels:
2405 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2406 2.6.16 - 2.6.21 vanilla (kernel.org)
2407 * Due to recently discovered recovery problems, we do not recommend
2408 using patchless RHEL 4 clients with this or any earlier release.
2409 * Recommended e2fsprogs version: 1.40.2-cfs1
2410 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2414 Description: Fix errors in lfs documentation
2415 Details : Fixes man pages
2417 Severity : enhancement
2419 Description: Adaptive timeouts
2420 Details : RPC timeouts adapt to changing server load and network
2421 conditions to reduce resend attempts and improve recovery time.
2423 Severity : enhancement
2425 Description: llapi_file_create() does not allow some changes
2426 Details : add llapi_file_open() that allows specifying the file creation
2427 mode and open flags, and also returns an open file handle.
2429 Severity : enhancement
2431 Description: df doesn't work properly if diskfs blocksize != 4K
2432 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
2435 Frequency : always on directories with default striping set
2437 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
2438 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
2440 Severity : enhancement
2442 Description: merge and cleanup kernel patches.
2443 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
2445 Severity : enhancement
2447 Description: RedHat Update kernel for RHEL5
2448 Details : Kernel config file for RHEL5.
2450 Severity : enhancement
2452 Description: OSS needs mutliple precreate threads
2453 Details : Add ability to start more than one create thread per OSS.
2456 Frequency : only with quota on the root user
2458 Description: mds_obd_create error creating tmp object
2459 Details : When the user sets quota on root, llog will be affected and can't
2460 create files and write files.
2463 Frequency : Always on ia64 patchless client, and possibly others.
2465 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
2466 Details : This allows the patchless client to be loaded on architectures
2467 without this export.
2472 Description: disorder of journal start and llog_add cause deadlock.
2473 Details : in llog_origin_connect, journal start should happen before
2474 llog_add keep the same order as other functions to avoid
2477 Severity : enhancement
2479 Description: RedHat Update kernel for RHEL5
2480 Details : Modify the kernel config file more closer RHEL5.
2482 Severity : enhancement
2484 Description: Build failure against Centos5 (RHEL5)
2485 Details : Define PAGE_SIZE when it isn't present.
2488 Frequency : occasionally when using NFS
2490 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
2491 Details : with NFS, the anon dentry's parent was set to itself in
2492 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
2493 corresponding dentry other than use rec->ur_name.
2495 Severity : enhancement
2497 Description: enable data checksumming by default
2498 Details : enable checksum by default, allow --disable-checksum
2499 configure option and "-o nochecksum" mount option. Checksums
2500 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
2503 Frequency : Occasionally with failover
2505 Description: Client eviction due to failover config
2506 Details : after a connection loss, the lustre client should attempt to
2507 reconnect to the last active server first before trying the
2508 other potential connections.
2510 Severity : enhancement
2512 Description: client-side metadata stat-ahead during readdir(directory readahead)
2513 Details : perform client-side metadata stat-ahead when the client detects
2514 readdir and sequential stat of dir entries therein
2519 Description: when mds and osts use different quota unit(32bit and 64bit),
2520 quota will be released repeatly.
2521 Details : void sending multiple quota reqs to mds, which will keep the
2522 status between the reqs.
2525 Frequency : only with liblustre clients on XT3
2527 Description: evictions taking too long
2528 Details : allow llrd to evict clients directly on OSTs
2532 Description: osts not allocated evenly to files
2533 Details : change the condition to increase offset_idx
2535 Severity : enhancement
2537 Description: self-adjustable client's lru lists
2538 Details : use adaptive algorithm for managing client cached locks lru
2539 lists according to current server load, other client's work
2540 pattern, memory activities, etc. Both, server and client
2541 side namespaces provide number of proc tunables for controlling
2546 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
2547 Details : rewrite inherited code (uuid parsing code from ext2 utils and
2548 readdir code from ext3) from scratch preserving functionality.
2552 Description: Only those disconnect error should be returned by rq_status.
2553 Details : In open/enqueue processs, Some errors, which will cause client
2554 disconnected, should be returned by rq_status, while other
2555 errors should still be returned by intent, then mdc or llite will
2558 Severity : enhancement
2560 Description: Tune the kernel for good SCSI performance.
2561 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
2562 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
2566 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
2567 Bugzilla : 13547, 13627
2568 Description: Data corruption for OSTs that are formatted larger than 2TB
2570 Details : When generating the bio request for lustre file writes the
2571 sector number would overflow a temporary variable before being
2572 used for the IO. The data reads correctly from Lustre (which
2573 will overflow in a similar manner) but other file data or
2574 filesystem metadata may be corrupted in some cases.
2578 Description: "lfs find -obd UUID" prints directories
2579 Details : "lfs find -obd UUID" will return all directory names instead
2580 of just file names. It is incorrect because the directories
2581 do not reside on the OSTs.
2584 Bugzilla : 13596, 13614
2585 Description: MDS hang after unclean shutdown of lots of clients
2586 Details : Never resend AST requests.
2590 Frequency : Always, for kernels after 2.6.16
2591 Description: Fix warning idr_remove called for id=.. which is not allocated.
2592 Details : Last kernels save old s_dev before kill super and not allow
2593 to restore from callback - restore it before call kill_anon_super.
2597 Description: Fix error on 'ls .' at the top of the Lustre mount.
2598 Details : Don't revalidate dentry if it is a root dentry.
2600 --------------------------------------------------------------------------------
2602 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
2604 * Support for kernels:
2605 2.6.5-7.286 (SLES 9),
2606 2.6.9-55.0.2.EL (RHEL 4),
2607 2.6.16.46-0.14 (SLES 10),
2608 2.6.18-8.1.8.el5 (RHEL 5),
2609 2.6.18.8 vanilla (kernel.org)
2610 * Client support for unpatched kernels:
2611 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2612 2.6.16 - 2.6.22 vanilla (kernel.org)
2613 * Due to recently discovered recovery problems, we do not recommend
2614 using patchless RHEL 4 clients with this or any earlier release.
2615 * Recommended e2fsprogs version: 1.39.cfs8
2616 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2621 Description: block reactivating mgc import until all deactivates complete
2622 Details : Fix race when failing back MDT/MGS to itself (testing)
2624 Severity : enhancement
2626 Description: lfs setstripe enhancement
2627 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
2630 Frequency : mds/oss recovery
2632 Description: llog ctxt is refrenced after it has been freed.
2633 Details : llog ctxt refcount was added to avoide the race between ctxt free
2634 and llog recovery process. Each llog user must hold ctxt refcount
2635 before it access the llog. And the llog ctxt can only be freed
2636 when its refcount is zero.
2638 Severity : enhancement
2640 Description: randomly memory allocation failure util
2641 Details : Make lustre randomly failed allocating memory for testing purpose.
2644 Frequency : only for SLES10
2646 Description: Update kernel patch for SLES10 SP1
2647 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
2649 Severity : enhancement
2651 Description: omit set fsid for export NFS
2652 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
2653 as FS_REQUIRES_DEV for avoid problems with generate fsid.
2656 Frequency : after network failures
2658 Description: Add sync option to mount_lustre.c
2659 Details : Client loses data written to lustre after a network interruption.
2661 Severity : enhancement
2663 Description: Error message improvement.
2664 Details : Merging of two LCONSOLE_ERROR_MSG into one.
2666 Severity : enhancement
2668 Description: don't use GFP_* in generic Lustre code.
2669 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
2671 Severity : enhancement
2673 Description: obdclass is limited by single OBD_ALLOC(idarray)
2674 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
2676 Severity : enhancement
2678 Description: updated patchess for new RHEL4 kernel
2679 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
2681 Severity : enhancement
2683 Description: warnings with build patchless client with vanila 2.6.19 and up
2684 Details : change old ctl_table style and replace ctl_table/ctl_table_header
2685 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
2687 Severity : enhancement
2689 Description: O_DIRECT bypasses client statistics.
2690 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
2691 read_ahead_stats were updated. Copied stats section from
2692 osc_send_oap_rpc() into async_internal().
2695 Frequency : only for Cray XT3
2697 Description: peer credits not enough on many OST per OSS systems.
2698 Details : Use new lnet way to add credits as we need those for pings and ASTs
2701 Frequency : only with liblustre
2703 Description: Liblustre is not releasing flock locks on file close.
2704 Details : Release flock locks on file close.
2707 Frequency : only for RHEL4
2709 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
2710 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
2711 Update target file and kernel config.
2715 Description: lustre support for RHEL5
2716 Details : Add support for RHEL5.
2721 Description: ASSERTION(export != NULL) failed in target_handle_connect
2722 Details : Assetion hit is result of rare race between disconnect and connet
2723 to same nid. target_handle_connect found old connect cockie and
2724 tried to reconnect, but can't find export for this cockie.
2729 Description: umount blocks forever on error
2730 Details : In result of wrong using obd_no_recov and obd_force flags client
2731 can hand if cancel or some other requests is lost.
2734 Frequency : Only for SLES
2736 Description: sanity_quota fail test_1
2737 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
2738 file, which makes TSTID[2] inunique.
2740 Severity : enhancement
2742 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
2743 Details : Update target/ChangeLog/which_patch .
2745 Severity : enhancement
2747 Description: jbd statistics
2748 Details : Port older jbd statistics patch for sles10
2753 Description: lvbo_init failed for resource with missing objects.
2754 Details : Fix returning error if we do stat for file with missing/corrupted
2755 objects and i_size set to all sum of size of all avaible objects.
2756 if we to truncate/write to missing object - it is recreated.
2761 Description: Oops in read and write path when failing to allocate lock.
2762 Details : Check if lock allocation failed and return error back.
2765 Frequency : When flocks are used.
2767 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
2768 Details : Flock locks might destroy just granted lock if it could be merged
2769 with another existing flock, this is done in completion handler,
2770 so teach ldlm_cli_enquque_fini that this is a valid case for
2776 Description: reply_lock_interpret crash due to race with it and lock cancel.
2777 Details : Do not replay locks that are being cancelled. Do not reference
2778 locks by their address during replay, just by their handle.
2781 Frequency : only with deactivated OSTs
2783 Description: lstripe command fails for valid OST index
2784 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
2785 instead of lov->desc.ld_active_tgt_count.
2787 Severity : enhancement
2789 Description: Build failure against Centos5 (RHEL5)
2790 Details : Use getpagesize() instead of PAGE_SIZE.
2792 --------------------------------------------------------------------------------
2794 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
2796 * Support for kernels:
2797 2.6.5-7.283 (SLES 9),
2798 2.6.9-55.EL (RHEL 4),
2799 2.6.16.46-0.14 (SLES 10),
2800 2.6.18.8 vanilla (kernel.org)
2801 * Client support for unpatched kernels:
2802 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2803 2.6.16 - 2.6.22 vanilla (kernel.org)
2804 * Due to recently discovered recovery problems, we do not recommend
2805 using patchless RHEL 4 clients with this or any earlier release.
2806 * Recommended e2fsprogs version: 1.39.cfs8
2807 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2808 * Starting with this release, the ldiskfs backing filesystem required
2809 by Lustre is now in its own package, lustre-ldiskfs. This package
2810 should be installed. It is versioned separately from Lustre and
2811 may be released separately in future.
2816 Description: server LBUG when shutting down
2817 Details : Block umount forever until the mount refcount is zero rather
2818 than giving up after an arbitrary timeout.
2820 Severity : enhancement
2822 Description: add optional extra BUILD_VERSION info
2823 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2824 to override the lustre version.
2827 Frequency : 2.6.18 servers only
2829 Description: ll_kern_mount() doesn't release the module reference
2830 Details : The ldiskfs module reference count never drops down to 0
2831 because ll_kern_mount() doesn't release the module reference.
2836 Description: server LBUG when using old ost_num_threads parameter
2837 Details : Accept the old ost_num_threads parameter but warn that it
2838 is deprecated, and fix an off-by-one error that caused an LBUG.
2843 Description: Transient SCSI error results in persistent IO issue
2844 Details : iobuf->dr_error is not reinitialized to 0 between two
2848 Frequency : sometimes when underlying device returns I/O errors
2850 Description: OSTs not going read-only during write failures
2851 Details : OSTs are not remounted read-only when the journal commit threads
2852 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2853 instead of the ext3 wrappers.
2856 Frequency : only with 10000 clients or more
2858 Description: poor connect scaling with increasing client count
2859 Details : Don't run filter_grant_sanity_check for more than 100 exports
2860 to improve scaling for large numbers of clients.
2863 Frequency : SLES10 only
2865 Description: sanity-quota.sh quotacheck failed: rc = -22
2866 Details : Quotas cannot be enabled on SLES10.
2869 Frequency : liblustre clients only
2871 Description: getdirentries does not give error when run on compute nodes
2872 Details : getdirentries does not fail when the size specified as an argument
2873 is too small to contain at least one entry
2875 Severity : enhancement
2877 Description: Add LNET router traceability for debug purposes
2878 Details : If a checksum failure occurs with a router as part of the
2879 IO path, the NID of the last router that forwarded the bulk data
2880 is printed so it can be identified.
2885 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2886 Details : Due to a race condition, liblustre clients were occasionally
2887 evicted incorrectly.
2889 Severity : enhancement
2891 Description: lfs setstripe use optional parameters instead of postional
2894 Severity : enhancement
2896 Description: Nanosecond timestamp support for ldiskfs
2897 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2898 resolution timestamps. There is not yet support for this at
2899 the Lustre filesystem level.
2902 Frequency : during server recovery
2904 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2905 Details : request with rq_no_resend flag not awake l_wait_event if they get a
2909 Frequency : nfs export on patchless client
2911 Description: connectathon hang when test nfs export over patchless client
2912 Details : Disconnected dentry cannot be found with lookup, so we do not need
2913 to unhash it or make it invalid
2916 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2917 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2918 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2923 Description: open req refcounting wrong on reconnect
2924 Details : If reconnect happened between getting open reply from server and
2925 call to mdc_set_replay_data in ll_file_open, we will schedule
2926 replay for unreferenced request that we are about to free.
2927 Subsequent close will crash in variety of ways.
2928 Check that request is still eligible for replay in
2929 mdc_set_replay_data().
2934 Description: disable writes to filesystem when reading health_check file
2935 Details : the default for reading the health_check proc file has changed
2936 to NOT do a journal transaction and write to disk, because this
2937 can cause reads of the /proc file to hang and block HA state
2938 checking on a healthy but otherwise heavily loaded system. It
2939 is possible to return to the previous behaviour during configure
2940 with --enable-health-write.
2942 Severity : enhancement
2944 Description: 64-bit inode version
2945 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2946 to the inode. This will be required for version-based recovery.
2951 Description: MDS fails to start if a duplicate client export is detected
2952 Details : in some rare cases it was possible for a client to connect to
2953 an MDS multiple times. Upon recovery the MDS would detect this
2954 and fail during startup. Handle this more gracefully.
2956 Severity : enhancement
2958 Description: Add -o localflock option to simulate old noflock behaviour.
2959 Details : This will achieve local-only flock/fcntl locks coherentness.
2964 Description: log_commit_thread vs filter_destroy race leads to crash
2965 Details : Take import reference before releasing llog record semaphore
2970 Description: Wrong request locking in request set processing
2971 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2972 imp_delayed_list, in this place should be used imp_lock.
2975 Frequency : when reconnecting
2977 Description: Grant leak when OSC reconnect to OST
2978 Details : When osc reconnect ost, OST(filter) should check whether it
2979 should grant more space to client by comparing fed_grant and
2980 cl_avail_grant, and return the granted space to client instead
2981 of "new granted" space, because client will call osc_init_grant
2982 to update the client grant space info.
2985 Frequency : when client reconnects to OST
2987 Description: Grant leak when OSC does a resend and replays bulk write
2988 Details : When osc reconnect to OST, OST(filter) should clear grant info of
2989 bulk write request, because the grant info while be sync between
2990 OSC and OST when reconnect, and we should ignore the grant info
2991 these of resend/replay write req.
2996 Description: Grant space more than avaiable space sometimes.
2997 Details : When then OST is about to be full, if two bulk writing from
2998 different clients came to OST. Accord the avaliable space of the
2999 OST, the first req should be permitted, and the second one
3000 should be denied by ENOSPC. But if the seconde arrived before
3001 the first one is commited. The OST might wrongly permit second
3002 writing, which will cause grant space > avaiable space.
3005 Frequency : when client is evicted
3007 Description: Grant might be wrongly erased when osc is evicted by OST
3008 Details : when the import is evicted by server, it will fork another
3009 thread ptlrpc_invalidate_import_thread to invalidate the
3010 import, where the grant will be set to 0. While the original
3011 thread will update the grant it got when connecting. So if
3012 the former happened latter, the grant will be wrongly errased
3013 because of this race.
3018 Description: Checking Stale with correct fid
3019 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
3020 to check whether it is stale, because sometimes, we want the
3021 enqueue happened anyway, and op_data.fid2 will not be initialized.
3023 Severity : enhancement
3025 Description: update patchless client
3026 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
3029 Frequency : only with 2.4 kernel
3031 Description: random memory corruption
3032 Details : size of struct ll_inode_info is to big for union inode.u and this
3033 can be cause of random memory corruption.
3038 Description: Memory leak in recovery
3039 Details : Lov_mds_md was not free in an error handler in mds_create_object.
3040 It should also check obd_fail before fsfilt_start, otherwise if
3041 fsfilt_start return -EROFS,(failover mds during mds recovery).
3042 then the req will return with repmsg->transno = 0 and rc = EROFS.
3043 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
3044 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
3045 be freed no matter whether fsfilt_commit success or not.
3048 Frequency : only with large numbers of cores on a single node
3050 Description: Prevents from taking the superblock lock in llap_from_page for
3052 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
3053 for llap_from_page call in ll_removepage() prevents from taking
3054 the superblock lock for a soon died page.
3059 Description: Not check open intent error before release open handle
3060 Details : in some rare cases, the open intent error is not checked before
3061 release open handle, which may cause
3062 ASSERTION(open_req->rq_transno != 0), because it tries to release
3063 the failed open handle.
3068 Description: Set cat log bitmap only after create log success.
3069 Details : in some rare cases, the cat log bitmap is set too early. and it
3070 should be set only after create log success.
3074 Description: Accessing a block bevice can re-enable I/O when Lustre is
3075 tearing down a device.
3076 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
3081 Description: service threads may hog cpus when there are a lot of requests
3082 Details : Insert cond_resched to give other threads a chance to use some CPU
3087 Description: the cat log was not initialized in recovery
3088 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
3089 unlink log on mds will not be initialized until mds post
3090 recovery. And also in mds post recovery, the unlink log will
3091 initialization will be done asynchronausly, so there will be race
3092 between add unlink log and unlink log initialization.
3096 Description: brw_stats were being printed incorrectly
3097 Details : brw_stats were being printed as log2 but all of them were not
3098 recorded as log2. Also remove some code duplication arising from
3099 filter_tally_{read,write}.
3103 Frequency : rare, only in recovery.
3104 Description: ASSERTION(req->rq_type != LI_POISON) failed
3105 Details : imp_lock should be held while iterating over imp_sending_list for
3106 prevent destroy request after get timeout in ptlrpc_queue_wait.
3110 Description: replay-single.sh test 52 fails
3111 Details : A lock's skiplist need to be cleanup when it being unlinked
3112 from its resource list.
3117 Description: Short directio read returns full requested size rather than
3119 Details : Direct I/O operations should return actual amount of bytes
3120 transferred rather than requested size.
3122 Severity : enhancement
3124 Description: metadata RPC reduction (e.g. for rm performance)
3125 Details : decrease the amount of synchronous RPC between clients and servers
3126 by canceling conflicing lock before the operation on the client
3127 and packing thier handles into the main operation RPC to server.
3129 Severity : enhancement
3131 Description: add #ifdef HAVE_KERNEL_CONFIG_H
3132 Details : kernels from 2.6.19 not need include linux/config.h, but add
3133 include linux/autoconf.h in commpiler command line.
3135 Severity : enhancement
3137 Description: patchless client support for 2.6.22 kernel
3138 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
3139 removed. In this case we need drop using os depended interface to
3140 kmem_cache and use cfs_mem_cache API.
3145 Description: fix mal-formatted messages
3146 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
3150 Frequency : always in liblustre
3151 Description: wrong IS_ERR implementation in liblustre.h
3152 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
3157 Description: Correct condition for output debug message.
3158 Details : inode i_nlink equal zero is not enough for output message about
3159 disk corruption, i_ctime and i_mode should be also checked.
3163 Frequency : always in patchless client
3164 Description: add configure check for truncate_complete_page
3165 Details : improve checks for exported symbols. This allow run check without
3166 sources, but with Module.symvers shipped with kernel distribution.
3167 add check for truncate_complete_page used by patchless client.
3171 Description: sanity.sh test_77h fails with "test_77h file compare failed"
3172 Details : test_77h uses a file which was messed by other test case.
3176 Description: Not Check whether lov_tgts is NULL in some lov functions
3177 Details : Checking whether lov_tgts is NULL in some functions.
3181 Description: replace obdo_alloc() with OBDO_ALLOC macro
3182 Details : nothing special is done in obdo_alloc() function, and for
3183 debugging purpose, it needs to be replaced with macros.
3187 Description: bad return value and errno from fcntl call
3188 Details : In liblustre API, errno should be a negative value if error
3193 Description: ptlrpc_check_set() LBUG
3194 Details : In case of positive reply from server and failed client bulk
3195 callback after bulk transfer shouldn't LBUG, but process this
3196 request as erroneous.
3198 Severity : enhancement
3200 Description: VFS operations stats tool.
3201 Details : Tool which collects stats by tracking value written in pid,
3202 ppid, gid and uses llstat to generate output to plot graph using
3204 Updated lustre/utils/Makefile.am
3205 Added lustre/utils/ltrack_stats.c
3207 Severity : enhancement
3209 Description: 2.6.18 server support (lustre 1.6.1)
3210 Details : Support for 2.6.18 kernels on the server side.
3215 Description: ASSERTION(imp->imp_conn_current) failed
3216 Details : an assertion failure is hit if a client node boots and attempts to
3217 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
3220 Frequency : only for i686
3222 Description: 1.4.11 RC1 build fails for RHEL 4, i686
3223 Details : Fixed config variable for build.
3228 Description: Updated patchess for new RHEL4 kernel
3229 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
3230 Updated patch jbd-stats-2.6.9.patch
3231 Updated patch qsnet-rhel4-2.6.patch
3232 Updated patch quota-deadlock-on-pagelock-core.patch
3233 Updated patch vfs_intent-2.6-rhel4.patch
3234 Updated patch vfs_races-2.6-rhel4.patch
3235 Updated series file 2.6-rhel4-titech.series
3236 Updated series file 2.6-rhel4.series
3237 Updated kernel config files
3242 Description: lquota slave complains LBUG when reconnecting with mds
3244 Details : quota slave depends qctxt->lqc_import to send its quota request.
3245 This pointer will be invalid if mds did failover or broke its
3246 connect to osts, which leads to LBUG.
3249 Frequency : when qunit size is too small(less than 20M)
3251 Description: write is stopped by improper -EDQUOT
3252 Details : If the master is busy and qunit size is small enough(let's say 1M),
3253 the slave can not get quota from master on time, which will lead
3254 slave to trigger a -EQUOTA to client.
3259 Description: Deadlock during metadata tests
3260 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
3261 called with the per-dentry lock held.
3264 Frequency : SLES9 only
3266 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
3267 Details : set CONFIG_NR_CPUS to 128 instead of 8.
3269 Severity : enhancement
3271 Description: remove fs_prep_san_write operation and related patches
3272 Details : remove the ext3-san-jdike patches which are no longer useful.
3277 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
3278 Details : a disk corruption can cause the mballoc code to assert on a
3279 double free or other extent corruptions. Handle these with
3280 ext3_error() instead of assertions.
3283 Frequency : only with mballoc3 code and deep extent trees
3285 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
3286 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
3287 can cause the filesystem to be remounted read-only.
3292 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
3293 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
3294 few external users of this ioctl number is lfsck's call to
3295 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
3296 Add the old ioctl number to the handler so both old and new
3301 Description: parallel lock callbacks
3302 Details : Instead of sending blocking and completion callbacks as separated
3303 requests, adding them to a set and sending in parallel.
3307 Description: Disable most debugging by default
3308 Details : To improve performance, disable most logging (for debug purposes)
3309 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
3310 default, and HA includes fewer messages.
3314 Frequency : only run on patchless client.
3315 Description: use do_facet on sanity.sh for test handling recoverables errors
3316 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
3320 Description: hash tables for lists of nids, connections and uuids
3321 Details : Hash tables noticeably help when a lot of clients connect to a
3322 server, to faster identify duplicate connections or reconnects,
3323 also to faster find export to evict in manual eviction case.
3327 Description: Sometimes, when the server evict a client, and the client will
3328 not be evicted as soon as possible.
3329 Details : In enqueue req, the error was returned by intent, instead of
3330 rq_status which make ptlrpc layer not detect this error, and
3331 does not evict the client. So enqueue error should be returned
3335 Frequency : only at startup
3337 Description: Delay client connections to MDT until fist MDT->OST connect
3338 Details : If a client tried to create a new file before the MDT had
3339 connected to any OSTs, the create would return EIO. Now
3340 the client will simply block until the MDT connects to the
3341 first OST and the create can succeed.
3344 Frequency : only for SLES9
3346 Description: Routinely utilize latest Quadrics drivers in CFS releases
3347 Details : Update patch qsnet-suse-2.6.patch.
3350 Frequency : only for sles10
3352 Description: Update patches for SLES 10 SP1 kernel.
3353 Details : Update the patch vfs_intent-2.6-sles10.patch.
3358 Description: Routinely utilize latest Quadrics drivers in CFS releases
3359 Details : Update patch qsnet-rhel4-2.6.patch.
3362 Frequency : at statup only
3364 Description: mds_lov_synchronize race leads to various problems
3365 Details : simultaneous MDT->OST connections at startup can cause the
3366 sync to abort, leaving the OSC in a bad state.
3371 Description: Using wrong pointer in osc_brw_prep_request
3372 Details : Access to array[-1] can produce panic if kernel compiled with
3373 CONFIG_PAGE_ALLOC enabled
3375 Severity : enhancement
3377 Description: Async OSC create to avoid the blocking unnecessarily.
3378 Details : If a OST has no remain object, system will block on the creating
3379 when need to create a new object on this OST. Now, ways use
3380 pre-created objects when available, instead of blocking on an
3381 empty osc while others are not empty. If we must block, we block
3382 for the shortest possible period of time.
3386 Frequency : only in recovery
3387 Description: Mark OST as early accessible if his start SYNC.
3388 Details : osc_precreate return flag early accessible if oscc marked as
3389 OSCC_FLAG_SYNC_IN_PROGRESS.
3394 Description: Sometimes precreate code can triger create object on wrong ost
3395 Details : Wrong protected or not not restored variables aftre precreate loop
3396 can produce creation object on wrong ost.
3399 Frequency : oss recovery
3401 Description: llog_commit_thread cleanup should sync with llog_commit_thread
3403 Details : llog_commit_thread_count should be synced between llog_commit
3404 start and cleanup, so new llog_commit thread should not be started
3405 when llog_commit threads being stopped to avoid accessing some
3408 Severity : enhancement
3410 Description: Add printing inode info into message about error in writepage.
3412 --------------------------------------------------------------------------------
3414 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
3419 Frequency : on some architectures
3421 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
3422 Details : Uninitialized flags sometimes cause configuration commands to
3426 Frequency : patchless clients only
3428 Description: missing __iget() symbol export
3429 Details : The __iget() symbol export is missing. To avoid the need for
3430 this on patchless clients the deathrow inode reaper is turned
3431 off, and we depend on the VM to clean up old inodes. This
3432 dependency was during via the fix for bug 12181.
3437 Description: sanity.sh fail: test_52b
3438 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
3439 incorrect inode's flags to client.
3441 --------------------------------------------------------------------------------
3443 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
3445 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
3446 INTEROPERATE with older versions automatically. In many cases a
3447 special upgrade step is needed. Please read the
3448 user documentation before upgrading any part of a 1.4.x system.
3449 * WARNING: Lustre configuration and startup changes are required with
3450 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
3452 * Support for kernels:
3453 2.4.21-47.0.1.EL (RHEL 3),
3454 2.6.5-7.283 (SLES 9),
3455 2.6.9-42.0.10.EL (RHEL 4),
3456 2.6.12.6 vanilla (kernel.org),
3457 2.6.16.27-0.9 (SLES10)
3458 * Client support for unpatched kernels:
3459 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
3460 2.6.16 - 2.6.19 vanilla (kernel.org),
3461 2.6.9-42.0.8EL (RHEL 4)
3462 * Recommended e2fsprogs version: 1.39.cfs6
3463 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3466 Severity : enhancement
3468 Description: Async OSC create to avoid the blocking unnecessarily.
3469 Details : If a OST has no remain object, system will block on the creating
3470 when need to create a new object on this OST. Now, ways use
3471 pre-created objects when available, instead of blocking on an
3472 empty osc while others are not empty. If we must block, we block
3473 for the shortest possible period of time.
3475 Severity : enhancement
3477 Description: MountConf
3478 Details : Lustre configuration is now managed via mkfs and mount
3479 commands instead of lmc and lconf. New obd types (MGS, MGC)
3480 are added for dynamic configuration management. See
3481 https://mail.clusterfs.com/wikis/lustre/MountConf for
3484 Severity : enhancement
3486 Description: dynamic OST addition
3487 Details : OSTs can now be added to a live filesystem
3489 Severity : enhancement
3491 Description: startup order invariance
3492 Details : MDTs and OSTs can be started in any order. Clients only
3493 require the MDT to complete startup.
3495 Severity : enhancement
3497 Description: parallel, asynchronous orphan cleanup
3498 Details : orphan cleanup is now performed in separate threads for each
3499 OST, allowing parallel non-blocking operation.
3501 Severity : enhancement
3503 Description: optimized stripe assignment
3504 Details : stripe assignments are now made based on ost space available,
3505 ost previous usage, and OSS previous usage, in order to try
3506 to optimize storage space and networking resources.
3508 Severity : enhancement
3510 Description: Permanently set tunables
3511 Details : All writable /proc/fs/lustre tunables can now be permanently
3512 set on a per-server basis, at mkfs time or on a live
3515 Severity : enhancement
3517 Description: Lustre message v2
3518 Details : Add lustre message format v2.
3520 Severity : enhancement
3522 Description: client OST exclusion list
3523 Details : Clients can be started with a list of OSTs that should be
3524 declared "inactive" for known non-responsive OSTs.
3528 Description: ENOENT returned for valid filehandle during dbench.
3529 Details : Check if a directory has children when invalidating dentries
3530 associated with an inode during lock cancellation. This fixes
3531 an incorrect ENOENT sometimes seen for valid filehandles during
3532 testing with dbench.
3535 Frequency : SFS test only (otherwise harmless)
3537 Description: SPEC SFS validation failure on NFS v2 over lustre.
3538 Details : Changes the blocksize for regular files to be 2x RPC size,
3539 and not depend on stripe size.
3541 Severity : enhancement
3543 Description: fine-grained SMP locking inside DLM
3544 Details : Improve DLM performance on SMP systems by removing the single
3545 per-namespace lock and replace it with per-resource locks.
3547 Severity : enhancement
3549 Description: don't hold multiple extent locks at one time
3550 Details : To avoid client eviction during large writes, locks are not
3551 held on multiple stripes at one time or for very large writes.
3552 Otherwise, clients can block waiting for a lock on a failed OST
3553 while holding locks on other OSTs and be evicted.
3555 Severity : enhancement
3557 Description: Multiple MD RPCs in flight.
3558 Details : Further unserialise some read-only MDT RPCs - learn about intents.
3559 To avoid overly-overloading MDT, introduce a limit on number of
3560 MDT RPCs in flight for a single client and add /proc controls
3561 to adjust this limit.
3563 Severity : enhancement
3565 Description: client read/write statistics
3566 Details : Add client read/write call usage stats for performance
3567 analysis of user processes.
3568 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
3569 file access. extents_stats shows chunk size distribution.
3570 extents_stats_per_process show chunk size distribution per
3573 Severity : enhancement
3575 Description: per-client statistics on server
3576 Details : Add ldlm and operations statistics for each client in
3577 /proc/fs/lustre/mds|obdfilter/*/exports/
3579 Severity : enhancement
3581 Description: improved MDT statistics
3582 Details : Add detailed MDT operations statistics in
3583 /proc/fs/lustre/mds/*/stats
3585 Severity : enhancement
3587 Description: VFS operations stats
3588 Details : Add client VFS call stats, trackable by pid, ppid, or gid
3589 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
3594 Description: Fix client-side osc byte counters
3595 Details : The osc read/write byte counters in
3596 /proc/fs/lustre/osc/*/stats are now working
3599 Frequency : always as root on SLES
3601 Description: Failure of copying files with lustre special EAs.
3602 Details : Client side always return success for setxattr call for lustre
3603 special xattr (currently only "trusted.lov").
3608 Description: Refcount LNET uuids
3609 Details : The global LNET uuid list grew linearly with every startup;
3610 refcount repeated list entries instead of always adding to
3613 Severity : enhancement
3615 Description: Dynamic service threads
3616 Details : Within a small range, start extra service threads
3617 automatically when the request queue builds up.
3620 Frequency : mixed-endian client/server environments
3622 Description: mixed-endian crashes
3623 Details : The new msg_v2 system had some failures in mixed-endian
3626 Severity : enhancement
3628 Description: Easy OST removal
3629 Details : OSTs can be permanently deactivated with e.g. 'lctl
3630 conf_param lustre-OST0001.osc.active=0'
3632 Severity : enhancement
3634 Description: MGS proc entries
3635 Details : Added basic proc entries for the MGS showing what filesystems
3638 Severity : enhancement
3640 Description: provide MGS failover
3641 Details : Added config lock reacquisition after MGS server failover.
3643 Severity : enhancement
3645 Description: add Linux 2.4 support
3646 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
3650 Description: a large application tries to do I/O to the same resource and dies
3651 in the middle of it.
3652 Details : Check the req->rq_arrival time after the call to
3653 ost_brw_lock_get(), but before we do anything about
3654 processing it & sending the BULK transfer request. This
3655 should help move old stale pending locks off the queue as
3656 quickly as obd_timeout.
3659 Frequency : when an incorrect nid is specified during startup
3661 Description: ptlrpc connect to non-existant node causes kernel crash
3662 Details : LNET can't be re-entered from an event callback, which
3663 happened when we expire a message after the export has been
3664 cleaned up. Instead, hand the zombie cleanup off to another
3667 Severity : enhancement
3669 Description: plain/inodebits lock performance improvement
3670 Details : Grouping plain/inodebits in granted list by their request modes
3671 and bits policy, thus improving the performance of search through
3675 Frequency : only if OST filesystem is corrupted
3677 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3678 Details : for a short time RPCs with bulk IO are in the replay list,
3679 but replay of bulk IOs is unimplemented. If the OST filesystem
3680 is corrupted due to disk cache incoherency and then replay is
3681 started it is possible to trip an assertion. Avoid putting
3682 committed RPCs into the replay list at all to avoid this issue.
3685 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3687 Description: System hang on startup
3688 Details : This bug allowed the liblustre (e.g. catamount) client to
3689 return to the app before handling all startup RPCs. This
3690 could leave the node unresponsive to lustre network traffic
3691 and manifested as a server ptllnd timeout.
3693 Severity : enhancement
3695 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3696 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
3697 boolean to control whether to print peer debug info when a
3698 client's RPC times out.
3701 Frequency : only for kernels with patches from Lustre below 1.4.3
3703 Description: Remove old rdonly API
3704 Details : Remove old rdonly API which unused from at least lustre 1.4.3
3707 Frequency : only for devices with external journals
3709 Description: Set external device read-only also
3710 Details : During a commanded failover stop, we set the disk device
3711 read-only while the server shuts down. We now also set any
3712 external journal device read-only at the same time.
3715 Frequency : when upgrading from 1.4 while trying to change parameters
3717 Description: The wrong (new) MDC name was used when setting parameters for
3718 upgraded MDT's. Also allows changing of OSC (and MDC)
3719 parameters if --writeconf is specified at tunefs upgrade time.
3722 Frequency : when setting specific ost indicies
3724 Description: QOS code breaks on skipped indicies
3725 Details : Add checks for missing OST indicies in the QOS code, so OSTs
3726 created with --index need not be sequential.
3728 Severity : enhancement
3730 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3731 Details : The uninit_groups feature works in conjunction with the kernel
3732 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3733 up the pass1 processing of e2fsck. This is a read-only feature
3734 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3735 mount filesystems that have had this feature enabled.
3737 Severity : enhancement
3739 Description: Improve multi-block allocation algorithm to avoid fragmentation
3740 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3741 allocation locality and avoid filesystem fragmentation.
3743 ------------------------------------------------------------------------------
3745 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
3747 * Support for kernels:
3748 2.4.21-47.0.1.EL (RHEL 3)
3749 2.6.5-7.283 (SLES 9)
3750 2.6.9-42.0.10.EL (RHEL 4)
3751 2.6.12.6 vanilla (kernel.org)
3752 2.6.16.27-0.9 (SLES 10)
3753 * Recommended e2fsprogs version: 1.39.cfs5
3755 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3759 Frequency : occasional, depends on client load and configuration
3760 Bugzilla : 12181, 12203
3761 Description: data loss for recently-modified files
3763 Details : In some cases it is possible that recently written or created
3764 files may not be written to disk in a timely manner (this should
3765 normally be within 30s unless client IO load is very high).
3766 The problem appears as zero-length files or files that are a
3767 multiple of 1MB in size after a client crash or client eviction
3768 that are missing data at the end of the file.
3770 This problem is more likely to be hit on clients where files are
3771 repeatedly created and unlinked in the same directory, clients
3772 have a large amount of RAM, have many CPUs, the filesystem has
3773 many OSTs, the clients are rebooted frequently, and/or the files
3774 are not accessed by other nodes after being written.
3776 The presence of the problem can be detected by looking at
3777 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
3778 smaller than the second (nr_unused) then dirty files will not
3779 be flushed automatically to disk. "sync; sleep 10" should be
3780 run several times on the node before unmounting it to update
3781 Lustre (this is also safe to run on nodes without this problem).
3783 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
3784 that can cause this same problem, so customers using that kernel
3785 also need to update the kernel in addition to Lustre. In order
3786 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
3788 It is normal that files written just before a client crash (less
3789 than 30s) may not yet have been flushed to disk, even for local
3793 Frequency : frequent on thin XT3 nodes
3795 Description: UUID collision on thin XT3 Linux nodes
3796 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
3797 randomly, since we relied on an insufficiently-seeded PRNG.
3802 Description: OSS hangs after "All ost request buffers busy"
3803 Details : A deadlock between quota and journal operations caused OSS
3804 hangs after printing "All ost request buffers busy."
3807 Frequency : always on liblustre builds
3809 Description: Cleanup compiler warnings on liblustre
3812 Frequency : always on liblustre builds on XT3
3814 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
3818 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
3819 > 32000 subdirectories
3820 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
3821 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
3822 subdirectory count crosses 32000. This will aid e2fsck to
3823 correctly handle more than 32000 subdirectories.
3826 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
3828 Description: System hang on startup
3829 Details : This bug allowed the liblustre (e.g. catamount) client to
3830 return to the app before handling all startup RPCs. This
3831 could leave the node unresponsive to lustre network traffic
3832 and manifested as a server ptllnd timeout.
3834 Severity : enhancement
3836 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
3837 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
3838 boolean to control whether to print peer debug info when a
3839 client's RPC times out.
3844 Description: make O_SYNC working on 2.6 kernels
3845 Details : 2.6 kernels use different method for mark pages for write,
3846 so need add a code to lustre for O_SYNC work.
3851 Description: Failure to close file and release space on NFS
3852 Details : Put inode details into lock acquired in ll_intent_file_open.
3853 Use mdc_intent_lock in ll_intent_open to properly
3854 detect all kind of errors unhandled by mdc_enqueue.
3859 Description: proc file read during shutdown sometimes raced obd removal,
3861 Details : Add lock to prevent obd access after proc file removal.
3864 Frequency : Only for files larger than 4GB on 32-bit clients.
3866 Description: improperly doing page alignment of locks
3867 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3868 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3873 Description: RPCs being resent when they shouldn't be
3874 Details : Some RPCs that should not be resent are being resent. This
3875 can cause inconsistencies in the RPC state machine. Do not
3876 resend such requests.
3879 Frequency : rare, only with NFS export
3881 Description: Crash on NFS re-export node
3882 Details : under very unusual load conditions an assertion is hit in
3883 ll_intent_file_open()
3886 Frequency : only if OST filesystem is corrupted
3888 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3889 Details : for a short time RPCs with bulk IO are in the replay list,
3890 but replay of bulk IOs is unimplemented. If the OST filesystem
3891 is corrupted due to disk cache incoherency and then replay is
3892 started it is possible to trip an assertion. Avoid putting
3893 committed RPCs into the replay list at all to avoid this issue.
3898 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3899 Details : Large single O_DIRECT read and write calls can fail to allocate
3900 a sufficiently large buffer to process the request. In case of
3901 allocation failure the allocation is retried with a smaller
3902 buffer and broken into smaller requests.
3904 Severity : enhancement
3906 Description: Add -o localflock option to simulate old noflock behaviour.
3907 Details : This will achieve local-only flock/fcntl locks coherentness.
3912 Description: versioning check is incomplete
3913 Details : Checking the version difference of client vs. server, report
3914 error if the gap is too big.
3919 Description: add support PG_writeback bit
3920 Details : add support for PG_writeback bit for Lustre, for more carefull
3921 work with page cache in 2.6 kernel. This also fix some deadlocks
3922 and remove hack for work O_SYNC with 2.6 kernel.
3924 Severity : enhancement
3926 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3927 Details : The uninit_groups feature works in conjunction with the kernel
3928 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3929 up the pass1 processing of e2fsck. This is a read-only feature
3930 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3931 mount filesystems that have had this feature enabled.
3933 Severity : enhancement
3935 Description: Improve multi-block allocation algorithm to avoid fragmentation
3936 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3937 allocation locality and avoid filesystem fragmentation.
3939 ------------------------------------------------------------------------------
3941 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3943 * Support for kernels:
3944 2.6.9-42.0.3.EL (RHEL 4)
3945 2.6.5-7.276 (SLES 9)
3946 2.4.21-47.0.1.EL (RHEL 3)
3947 2.6.12.6 vanilla (kernel.org)
3948 2.6.16.21-0.8 (SLES10)
3949 * Recommended e2fsprogs version: 1.39.cfs2-0
3951 * The backwards-compatible /proc/sys/portals symlink has been removed
3952 in this release. Before upgrading, please ensure that you change
3953 any configuration scripts or /etc/sysctl.conf files that access
3954 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3955 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3956 in advance of the upgrade on any system running Lustre 1.4.6 or
3957 newer, since /proc/sys/lnet was added in that version.
3958 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3962 Frequency : only when quota is used
3964 Description: avoid scanning export list for quota master
3965 Details : Change the algorithms to avoid scanning export list in order
3966 to improve the efficiency.
3969 Frequency : MDS failover only, very rarely
3971 Description: "went back in time" messages on mds failover
3972 Details : The greatest transno may be lost when the current operation
3973 finishes with an error (transno==0) and the client's last_rcvd
3974 record is over-written. Save the greatest transno in the
3975 mds_last_transno for this case.
3978 Frequency : always for specific kernels and striping counts
3980 Description: client may get "Matching packet too big" without ACL support
3981 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3982 when trying to access files in certain configurations. The
3983 clients should in fact be denied when mounting because they do
3984 not understand ACLs.
3987 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3989 Description: many clients connecting with IO in progress causes connect timeouts
3990 Details : Avoid synchronous journal commits to avoid delays caused by many
3991 clients connecting/disconnecting when bulk IO is in progress.
3992 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3993 OST_IO_PORTAL to avoid delays behind potentially many pending
3997 Frequency : occasionally with multiple writers to a single file
3999 Description: shared writes to file may result in wrong size reported by stat()
4000 Details : Allow growing of kms when extent lock is cancelled
4003 Frequency : always with random mmap IO to multi-striped file
4005 Description: mmap write might be lost if we are writing to a 'hole' in stripe
4006 Details : Only if the hole is at the end of OST object so that kms is too
4007 small. Fix is to increase kms accordingly in ll_nopage.
4010 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
4012 Description: writes to a missing object would leak memory on the OST
4013 Details : If there is an inconsistency between the MDS and OST filesystems,
4014 such that the MDS references an object that doesn't exist, writes
4015 to that object will leak memory due to incorrect cleanup in the
4016 error handling path, eventually running out of memory on the OST.
4021 Description: Creating too long symlink causes lustre errors
4022 Details : Check symlink and name lengths before sending requests to MDS.
4025 Frequency : only if flock is enabled (not on by default)
4027 Description: posix locks not released on fd closure on 2.6.9+
4028 Details : We failed to add posix locks to list of inode locks on 2.6.9+
4029 kernels, this caused such locks not to be released on fd close and
4030 then assertions on fs unmount about still used locks.
4033 Frequency : MDS failover only, very rarely
4035 Description: clients may get ASSERTION(granted_lock != NULL)
4036 Details : When request was taking a long time, and a client was resending
4037 a getattr by name lock request. The were multiple lock requests
4038 with the same client lock handle and
4039 mds_getattr_name->fixup_handle_for_resent_request found one of the
4040 lock handles but later failed with ASSERTION(granted_lock != NULL).
4045 Description: handle->h_buffer_credits > 0, assertion failure
4046 Details : h_buffer_credits is zero after truncate, causing assertion
4047 failure. This patch extends the transaction or creates a new
4051 Frequency : NFS re-export or patchless client
4052 Bugzilla : 11179, 10796
4053 Description: Crash on NFS re-export node (__d_move)
4054 Details : We do not want to hash the dentry if we don't have a lock.
4055 But if this dentry is later used in d_move, we'd hit uninitialised
4056 list head d_hash, so we just do this to init d_hash field but
4057 leave dentry unhashed.
4060 Frequency : NFS re-export or patchless client
4062 Description: NFS exports has problem with symbolic link
4063 Details : lustre client didn't properly install dentry when re-exported
4064 to NFS or running patchless client.
4067 Frequency : NFS re-export or patchless client
4069 Description: Various nfs/patchless fixes.
4070 Details : fixes reuse disconected alias for lookup process - this fixes
4071 warning "find_exported_dentry: npd != pd",
4072 fix permission error with open files at nfs.
4073 fix apply umask when do revalidate.
4076 Frequency : occasional
4078 Description: Crash on NFS re-export node
4079 Details : calling clear_page() on the wrong pointer triggered oops in
4080 generic_mapping_read().
4083 Frequency : rarely, using O_DIRECT IO
4085 Description: unaligned directio crashes client with LASSERT
4086 Details : check for unaligned buffers before trying any requests.
4089 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
4091 Description: stale data returned from RAID cache
4092 Details : If only a small amount of IO is done to the RAID device before
4093 reading it again it is possible to get stale data from the RAID
4094 cache instead of reading it from disk.
4097 Frequency : always for sles10 kernel
4099 Description: sles10 support
4100 Details : ll_follow_link: compile fixes and using of nd_set_link
4101 under newer kernels.
4104 Frequency : depends on arch, kernel and compiler version, always on sles10
4107 Description: recursive or deep enough symlinks cause stack overflow
4108 Details : getting rid of large stack-allocated variable in
4112 Frequency : depends on hardware
4114 Description: lustre write performance loss in the SLES10 kernel
4115 Details : the performance loss is caused by using of write barriers in the
4116 ext3 code. The SLES10 kernel turns barrier support on by
4117 default. The fix is to undo that change for ldiskfs.
4119 ------------------------------------------------------------------------------
4121 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
4123 * Support for kernels:
4124 2.6.9-42.0.3EL (RHEL 4)
4125 2.6.5-7.276 (SLES 9)
4126 2.4.21-47.0.1.EL (RHEL 3)
4127 2.6.12.6 vanilla (kernel.org)
4131 Frequency : quota enabled and large files being deleted
4133 Description: releasing more than 4GB of quota at once hangs OST
4134 Details : If a user deletes more than 4GB of files on a single OST it
4135 will cause the OST to spin in an infinite loop. Release
4136 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4141 Description: statfs data retrieved from /proc may be stale or zero
4142 Details : When reading per-device statfs data from /proc, in the
4143 {kbytes,files}_{total,free,avail} files, it may appear
4144 as zero or be out of date.
4147 Frequency : systems with MD RAID1 external journal devices
4149 Description: lconf's call to blkid is confused by RAID1 journal devices
4150 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4151 of returning all block devices that match the journal UUID.
4154 Frequency : always, for aggregate stripe size over 4GB
4156 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
4157 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
4158 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
4161 Frequency : always if "lfs find" used on a local file/directory
4163 Description: "lfs find" segfaults if used on a local file/directory
4164 Details : The case where a directory component was not specified wasn't
4165 handled correctly. Handle this properly.
4168 Frequency : always on ppc64
4170 Description: the write to an ext3 filesystem mounted with mballoc got stuck
4171 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4172 perform endianness conversion.
4175 Frequency : rarely (truncate to non-zero file size after write under load)
4176 Bugzilla : 10730, 10687
4177 Description: Files padded with zeros to next 4K multiple
4178 Details : With filesystems mounted using the "extents" option (2.6 kernels)
4179 it is possible that files that are truncated to a non-zero size
4180 immediately after being written are filled with zero bytes beyond
4181 the truncated size. No file data is lost.
4183 Severity : enhancement
4185 Description: Allow recovery/failover for liblustre clients.
4186 Details : liblustre clients were unaware of failover configurations until
4189 Severity : enhancement
4191 Description: user file locks should fail when not mounting with flock option
4192 Details : Set up an error-returning stub in ll_file_operations.lock field
4193 to prevent incorrect behaviour when client is mounted without
4194 flock option. Also, set up properly f_op->flock field for
4198 Frequency : always on ia64
4200 Description: "lfs df" loops on printing out MDS statfs information
4201 Details : The obd_ioctl_data was not initialized and in some systems
4202 this caused a failure during the ioctl that did not return
4203 an error. Initialize the struct and return an error on failure.
4206 Frequency : SLES 9 only
4208 Description: Error of copying files with lustre special EAs as root
4209 Details : Client side always return success for setxattr call for lustre
4210 special xattr (currently only "trusted.lov").
4213 Frequency : rarely on clusters with both ia64+i386 clients
4215 Description: ia64+i686 clients doing shared IO on the same file may LBUG
4216 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
4217 clients are doing concurrent writes to the same file it is
4218 possible that the ia64 clients may LASSERT because the OST
4219 extent locks are not PAGE_SIZE aligned. Ensure that grown
4220 locks are always aligned on the request boundary.
4223 Frequency : specific use, occasional
4225 Description: Overwriting in use executable truncates on-disk binary image
4226 Details : If one node attempts to overwrite an executable in use by
4227 another node, we now correctly return ETXTBSY instead of
4228 truncating the file.
4230 Severity : enhancement
4232 Description: Async OSC create to avoid the blocking unnecessarily.
4233 Details : If a OST has no remain object, system will block on the creating
4234 when need to create a new object on this OST. Now, ways use
4235 pre-created objects when available, instead of blocking on an
4236 empty osc while others are not empty. If we must block, we block
4237 for the shortest possible period of time.
4242 Description: chmod on Lustre root is propagated to other clients
4243 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
4244 invalid by the follow_mount time.
4247 Frequency : liblustre clients only
4249 Description: Race in 'instant cancel' lock handling could lead to such locks
4250 never to be granted in case of SMP MDS
4251 Details : Do not destroy not yet granted but cbpending locks in
4255 Frequency : replay/resend of open
4257 Description: non null lock assetion failure in mds_intent_policy
4258 Details : Trying to replay/resend lockless open requests resulted in
4259 mds_open() returning 0 with no lock. Now it sets a flag if
4260 a lock is going to be returned.
4262 Severity : enhancement
4264 Description: Checksum enhancements
4265 Details : New checksum enhancements allow for resending RPCs that failed
4268 Severity : enhancement
4270 Description: Tunables on number of dirty pages in cacche
4271 Details : Allow to set limit on number of dirty pages cached.
4276 Description: client crash on unmount - lock still has references
4277 Details : In some error handling cases it was possible to leak a lock
4278 reference on a client while accessing a file. This was not
4279 harmful to the client during operation, but would cause the
4280 client to crash when the filesystem is unmounted.
4283 Frequency : specific case, rare
4285 Description: ETXTBSY on mds though file not in use by client
4286 Details : ETXTBSY is no longer incorrectly returned when attempting to
4287 chmod or chown a directory that the user previously tried to
4288 execute or a currently-executing binary.
4291 Frequency : extremely rare except on liblustre-based clients
4293 Description: Lustre space not freed when files are deleted
4294 Details : Clean up open-unlinked files after client eviction. Previously
4295 the unlink was skipped and the files remained as orphans.
4300 Description: OST failure "would be an LBUG" in waiting_locks_callback()
4301 Details : In some cases it was possible to send a blocking callback to
4302 a client doing a glimpse, even though that client didn't get
4303 a lock granted. When the glimpse lock is cancelled on the OST
4304 the freed lock is left on the waiting list and corrupted the list.
4307 Frequency : all core dumps
4309 Description: Broke core dumps to lustre
4310 Details : Negative dentry may be unhashed if parent does not have UPDATE
4311 lock, but some callers, e.g. do_coredump, expect dentry to be
4312 hashed after successful create, hash it in ll_create_it.
4314 ------------------------------------------------------------------------------
4316 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
4318 * Support for kernels:
4319 2.6.9-42.0.2.EL (RHEL 4)
4320 2.6.5-7.276 (SLES 9)
4321 2.4.21-40.EL (RHEL 3)
4322 2.6.12.6 vanilla (kernel.org)
4326 Frequency : always on RHEL 3
4328 Description: Number of open files grows over time
4329 Details : The number of open files grows over time, whether or not
4330 Lustre is started. This was due to a filp leak introduced
4331 by one of our kernel patches.
4333 ------------------------------------------------------------------------------
4335 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
4337 * Support for kernels:
4338 2.6.9-42.EL (RHEL 4)
4339 2.6.5-7.267 (SLES 9)
4340 2.4.21-40.EL (RHEL 3)
4341 2.6.12.6 vanilla (kernel.org)
4346 Bugzilla : 5719, 9635, 9792, 9684
4347 Description: OST (or MDS) trips assertions in (re)connection under heavy load
4348 Details : If a server is under heavy load and cannot reply to new
4349 connection requests before the client resends the (re)connect,
4350 the connection handling code can behave badly if two service
4351 threads are concurrently handing separate (re)connections from
4352 the same client. Add better locking to the connection handling
4353 code, and ensure that only a single connection will be processed
4354 for a given client UUID, even if the lock is dropped.
4356 Severity : enhancement
4358 Description: add TCP zero-copy support to kernel
4359 Details : Add support to the kernel TCP stack to allow zero-copy bulk
4360 sends if the hardware supports scatter-gather and checksumming.
4361 This allows socklnd to do client-write and server-read more
4362 efficiently and reduce CPU utilization from skbuf copying.
4365 Frequency : only if NFS exporting from client
4367 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
4368 Details : If setting attributes on a file created under NFS that had
4369 never been opened it would be possible to oops the client
4370 if the file had no objects.
4374 Bugzilla : 9326, 10402, 10897
4375 Description: client crash in ptlrpcd_wake() thread when sending async RPC
4376 Details : It is possible that ptlrpcd_wake() dereferences a freed async
4377 RPC. In rare cases the ptlrpcd thread alread processed the RPC
4378 before ptlrpcd_wake() was called and the request was freed.
4381 Frequency : always for liblustre
4383 Description: liblustre client does MDS+OSTs setattr RPC for each write
4384 Details : When doing a write from a liblustre client, the client
4385 incorrectly issued an RPC to the MDS and each OST the file was
4386 striped over in order to update the timestamps. When writing
4387 with small chunks and many clients this could overwhelm the MDS
4388 with RPCs. In all cases it would slow down the write because
4389 these RPCs are unnecessary.
4391 Severity : enhancement
4393 Description: allow number of MDS service threads to be changed at module load
4394 Details : It is now possible to change the number of MDS service threads
4395 running. Adding "options mds mds_num_threads={N}" to the MDS's
4396 /etc/modprobe.conf will set the number of threads for the next
4397 time Lustre is restarted (assuming the "mds" module is also
4398 reloaded at that time). The default number of threads will
4399 stay the same, 32 for most systems.
4404 Description: OST crash if filesystem is unformatted or corrupt
4405 Details : If an OST is started on a device that has never been formatted
4406 or if the filesystem is corrupt and cannot even mount then the
4407 error handling cleanup routines would dereference a NULL pointer.
4412 Description: NULL pointer deref in llap_from_page.
4413 Details : get_cache_page_nowait can return a page with NULL (or otherwise
4414 incorrect) mapping if the page was truncated/reclaimed while it was
4415 searched for. Check for this condition and skip such pages when
4416 doing readahead. Introduce extra check to llap_from_page() to
4417 verify page->mapping->host is non-NULL (so page is not anonymous).
4420 Frequency : Sometimes when using sys_sendfile
4422 Description: "page not covered by a lock" warnings from ll_readpage
4423 Details : sendfile called ll_readpage without right page locks present.
4424 Now we introduced ll_file_sendfile that does necessary locking
4425 around call to generic_file_sendfile() much like we do in
4429 Frequency : with certain MDS communication failures at client mount time
4431 Description: NULL pointer deref after failed client mount
4432 Details : a client connection request may delayed by the network layer
4433 and not be sent until after the PTLRPC layer has timed out the
4434 request. If the client fails the mount immediately it will try
4435 to clean up before the network times out the request. Add a
4436 reference from the request import to the obd device and delay
4437 the cleanup until the network drops the request.
4440 Frequency : occasionally during client (re)connect
4442 Description: assertion failure during client (re)connect
4443 Details : processing a client connection request may be delayed by the
4444 client or server longer than the client connect timeout. This
4445 causes the client to resend the connection request. If the
4446 original connection request is replied in this interval, the
4447 client may trip an assertion failure in ptlrpc_connect_interpret()
4448 which thought it would be the only running connect process.
4451 Frequency : only with obd_echo servers and clients that are rebooted
4453 Description: kernel BUG accessing uninitialized data structure
4454 Details : When running an obd_echo server it did not start the ping_evictor
4455 thread, and when a client was evicted an uninitialized data
4456 structure was accessed. Start the ping_evictor in the RPC
4457 service startup instead of the OBD startup.
4459 Severity : enhancement
4460 Bugzilla : 10193 (patchless)
4461 Description: Remove dependency on various unexported kernel interfaces.
4462 Details : No longer need reparent_to_init, exit_mm, exit_files,
4463 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
4466 Frequency : rare (only users of deprecated and unsupported LDAP config)
4468 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
4469 Details : LDAP apparently contains 'lustreName' attributes instead of
4470 'name'. A simple remapping of the name is sufficient.
4473 Frequency : rare (only with non-default dump_on_timeout debug enabled)
4475 Description: waiting_locks_callback trips kernel BUG if client is evicted
4476 Details : Running with the dump_on_timeout debug flag turned on makes
4477 it possible that the waiting_locks_callback() can try to dump
4478 the Lustre kernel debug logs from an interrupt handler. Defer
4479 this log dumping to the expired_lock_main() thread.
4481 Severity : enhancement
4483 Description: Support NFS exporting on 2.6 kernels.
4484 Details : Implement non-rawops metadata methods for NFS server to use without
4485 changing NFS server code.
4488 Frequency : very rare (synthetic metadata workload only)
4490 Description: two racing renames might cause an MDS thread to deadlock
4491 Details : Running the "racer" program may cause one MDS thread to rename
4492 a file from being the source of a rename to being the target of
4493 a rename at exactly the same time that another thread is doing
4494 so, and the second thread has already enqueued these locks after
4495 doing a lookup of the target and is trying to relock them in
4496 order. Ensure that we don't try to re-lock the same resource.
4499 Frequency : only very large systems with liblustre clients
4501 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
4502 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
4503 all OSTs in parallel. This allows the eviction of stale liblustre
4504 clients to proceed much faster than if they were done in series,
4505 and also offers similar improvements for other set_info RPCs.
4510 Description: excessive CPU usage during initial read phase on client
4511 Details : During the initial read phase on a client, it would agressively
4512 retry readahead on the file, consuming too much CPU and impacting
4513 performance (since 1.4.5.8). Improve the readahead algorithm
4514 to avoid this, and also improve some other common cases (read
4515 of small files in particular, where "small" is files smaller than
4516 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
4521 Description: MDS crash when receiving packet with unknown intent.
4522 Details : Do not LBUG in unknown intent case, just return -EFAULT
4524 Severity : enhancement
4525 Bugzilla : 9293, 9385
4526 Description: MDS RPCs are serialised on client. This is unnecessary for some.
4527 Details : Do not serialize getattr (non-intent version) and statfs.
4530 Frequency : occasional, when OST network is overloaded/intermittent
4532 Description: client evicted by OST after bulk IO timeout
4533 Details : If a client sends a bulk IO request (read or write) the OST
4534 may evict the client if it is unresposive to its data GET/PUT
4535 request. This is incorrect if the network is overloaded (takes
4536 too long to transfer the RPC data) or dropped the OST GET/PUT
4537 request. There is no need to evict the client at all, since
4538 the pinger and/or lock callbacks will handle this, and the
4539 client can restart the bulk request.
4542 Frequency : Always when mmapping file with no objects
4544 Description: client crashes when mmapping file with no objects
4545 Details : Check that we actually have objects in a file before doing any
4546 operations on objects in ll_vm_open, ll_vm_close and
4552 Description: Request leak when working with deleted CWD
4553 Details : Introduce advanced request refcount tracking for requests
4554 referenced from lustre intent.
4556 Severity : Enhancement
4558 Description: Cache open file handles on client.
4559 Details : MDS now will return special lock along with openhandle, if
4560 requested and client is allowed to hold openhandle, even if unused,
4561 until such a lock is revoked. Helps NFS a lot, since NFS is opening
4562 closing files for every read/write openration.
4564 Severity : Enhancement
4566 Description: Cache open negative dentries on client when possible.
4567 Details : Guard negative dentries with UPDATE lock on parent dir, drop
4568 negative dentries on lock revocation.
4573 Description: Remounting a client read-only wasn't possible with a zconf mount
4574 Details : It wasn't possible to remount a client read-only with llmount.
4576 Severity : enhancement
4577 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
4578 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
4579 MPICH is included to add Lustre-specific ADIO interfaces.
4580 This is based closely on the UFS ADIO layer and only differs
4581 in file creation, in order to allow the OST striping to be set.
4582 This is user-contributed code and not supported by CFS.
4587 Description: extended inode attributes (immutable, append-only) work improperly
4588 when 2.4 and 2.6 kernels are used on client/server or vice versa
4589 Details : Introduce kernel-independent values for these flags.
4591 Severity : enhancement
4594 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
4595 Details : Many of the /proc/ tunables can only be tuned at a megabyte
4596 granularity. Now, Fractional MB granularity is be supported,
4597 this is very useful for low memory system.
4599 Severity : enhancement
4601 Description: Getattr by fid
4602 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
4603 locks, avoids extra getattr rpc requests to MDS, allows '/' to
4604 have locks and avoids getattr rpc requests for it on every stat.
4607 Frequency : Always, for filesystems larger than 2TB
4609 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
4610 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
4611 filesystems larger than 2TB to be mounted with mballoc enabled.
4614 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
4616 Description: filesystem corruption for non-standard kernels and very large OSTs
4617 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
4618 filesystems larger than 2TB is mounted then the kernel will
4619 silently corrupt the start of the filesystem. CONFIG_LBD is
4620 enabled for all CFS-supported kernels, but the possibility of
4621 this happening with a modified kernel config exists.
4623 Severity : enhancement
4625 Description: add client O_DIRECT support for 2.6 kernels
4626 Details : It is now possible to do O_DIRECT reads and writes to files
4627 in the Lustre client mountpoint on 2.6 kernel clients.
4629 Severity : enhancement
4631 Description: parallel glimpse, setattr, statfs, punch, destroy requests
4632 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
4633 parallel, not waiting for response from every OST before sending
4634 a rpc to the next OST.
4639 Description: setattr vs write race when updating file timestamps
4640 Details : Client processes that update a file timestamp into the past
4641 right after writing to the file (e.g. tar) it is possible that
4642 the updated file modification time can be reset to the current
4643 time due to a race between processing the setattr and write RPC.
4645 Severity : enhancement
4647 Description: Bring 'lfs find' closer in line with regular Linux find.
4648 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
4649 -print0 options and obtains all the needed info through the lustre
4652 Severity : enhancement
4654 Description: support up to 1024 configured devices on one node
4655 Details : change obd_dev array from statically allocated to dynamically
4656 allocated structs as they are first used to reduce memory usage
4661 Description: Flush dirty partially truncated pages during truncate
4662 Details : Immediatelly flush partially truncated pages in filter_setattr,
4663 this way we completely avoid having any pages in page cache on OST
4664 and can retire ugly workarounds during writes to flush such pages.
4669 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
4670 Details : protect inode from truncation within vfs_unlink() context
4671 just take a reference before calling vfs_unlink() and release it
4672 when parent's i_sem is free.
4675 Frequency : always, if extents are used on OSTs
4677 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
4678 Details : OSTs using the extents format would not zero the high 16 bits of
4679 the index physical block number. This is not a problem for any
4680 OST filesystems smaller than 16TB, and no kernels support ext3
4681 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
4682 new/modified files) and can be fixed for existing filesystems
4683 with e2fsprogs-1.39-cfs1.
4688 Description: import connection selection may be incorrect if timer wraps
4689 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
4690 import connections to be ignored if the 32-bit jiffies counter
4691 wraps. Use a 64-bit jiffies counter.
4694 Frequency : during server recovery
4696 Description: crash after server is denying duplicate export
4697 Details : If clients are resending connect requests to the server, the
4698 server refuses to allow a client to connect multiple times.
4699 Fixed a bug in the handling of this case.
4702 Frequency : very large clusters immediately after boot
4704 Description: LNET request buffers exhausted under heavy short-term load
4705 Details : If a large number of client requests are generated on a service
4706 that has previously never seen so many requests it is possible
4707 that the request buffer growth cannot keep up with the spike in
4708 demand. Instead of dropping incoming requests, they are held in
4709 the LND until the RPC service can accept more requests.
4712 Frequency : Sometimes during replay
4714 Description: Assertion failure in ll_local_open after replay.
4715 Details : If replay happened on an open request reply before we were able
4716 to set replay handler, reply will become not swabbed tripping the
4717 assertion in ll_local_open. Now we set the handler right after
4718 recognising of open request
4721 Frequency : very rare
4723 Description: kernel reports "badness in vsnprintf"
4724 Details : Reading from the "recovery_status" /proc file in small chunks
4725 may cause a negative length in lprocfs_obd_rd_recovery_status()
4726 call to vsnprintf() (which is otherwise harmless). Exit early
4727 if there is no more space in the output buffer.
4729 Severity : enhancement
4731 Description: clear OBD RPC statistics by writing to them
4732 Details : It is now possible to clear the OBD RPC statistics by writing
4733 to the "stats" file.
4738 Description: Client mtime is not the same on different clients after utimes
4739 Details : In some cases, the client was using the utimes() syscall on
4740 a file cached on another node. The clients now validate the
4741 ctime from the MDS + OSTs to determine which one is right.
4746 Description: Inability to activate failout mode
4747 Details : lconf script incorrectly assumed that in python string's numeric
4748 value is used in comparisons.
4751 Frequency : always with multiple stripes per file
4753 Description: Inefficient object allocation for mutli-stripe files
4754 Details : When selecting which OSTs to stripe files over, for files with
4755 a stripe count that divides evenly into the number of OSTs,
4756 the MDS is always picking the same starting OST for each file.
4757 Return the OST selection heuristic to the original design.
4762 Description: mount failures may take full timeout to return an error
4763 Details : Under some heavy load conditions it is possible that a
4764 failed mount can wait for the full obd_timeout interval,
4765 possibly several minutes, before reporting an error.
4766 Instead return an error as soon as the status is known.
4768 ------------------------------------------------------------------------------
4770 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
4772 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
4773 INTEROPERATE with older versions automatically. Please read the
4774 user documentation before upgrading any part of a live system.
4775 * WARNING: Lustre networking configuration changes are required with
4776 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
4779 * Support for kernels:
4780 2.6.9-22.0.2.EL (RHEL 4)
4781 2.6.5-7.244 (SLES 9)
4782 2.6.12.6 vanilla (kernel.org)
4785 Severity : enhancement
4786 Bugzilla : 7981/8208
4787 Description: Introduced Lustre Networking (LNET)
4788 Details : LNET is new networking infrastructure for Lustre, it includes
4789 a reorganized network configuration mode (see the user
4790 documentation for full details) as well as support for routing
4791 between different network fabrics. Lustre Networking Devices
4792 (LNDs) for the supported network fabrics have also been
4793 created for this new infrastructure.
4795 Severity : enhancement
4796 Description: Introduced Access control lists
4797 Details : clients can set ACLs on files and directories in order to have
4798 more fine-grained permissions than the standard Unix UGO+RWX.
4799 The MDS must be started with the "-o acl" mount option.
4801 Severity : enhancement
4802 Description: Introduced filesystem quotas
4803 Details : Administrators may now establish per-user quotas on the
4806 Severity : enhancement
4808 Description: Configuration change for the XT3
4809 The PTLLND is now used to run Lustre over Portals on the XT3
4810 The configure option(s) --with-cray-portals are no longer used.
4811 Rather --with-portals=<path-to-portals-includes> is used to
4812 enable building on the XT3. In addition to enable XT3 specific
4813 features the option --enable-cray-xt3 must be used.
4818 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4819 Details : A race between allocating a new llcd and re-getting the llcd_lock
4820 allowed another thread to grab newly-allocated llcd.
4822 Severity : enhancement
4824 Description: 2.6 OST async journal commit and locking fix to improve performance
4825 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4826 2.6 kernels are now async as they already were in 2.4 kernels so
4827 that they can commit concurrently with the network bulk transfer.
4828 For block-allocated files the filter allocation semaphore is held
4829 to avoid filesystem fragmentation during allocation. BKL lock
4830 removed for 2.6 xattr operations where it is no longer needed.
4835 Description: lconf incorrectly determined whether two IP networks could talk
4836 Details : In some more complicated routing and multiple-network
4837 configurations, lconf will avoid trying to make a network
4838 connection to a disjoint part of the IP space. It was doing the
4839 math incorrectly for one set of cases.
4844 Description: Fix for potential infinite loop processing records in an llog.
4845 Details : If an llog record is corrupted/zeroed, it is possible to loop
4846 forever in llog_process(). Validate the llog record length
4847 and skip the remainder of the block on error.
4850 Frequency : occasional (liblustre only)
4852 Description: liblustre could not open files whose last component is a symlink
4853 Details : sysio_path_walk() would incorrectly pass the open intent to
4854 intermediate path components.
4857 Frequency : rare (liblustre only with non-standard tuning)
4858 Bugzilla : 7201 (7350)
4859 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4860 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4861 after it has been released, only do this for referenced locks
4863 Severity : enhancement
4865 Description: specifying an (invalid) directory default stripe_size of -1
4866 would reset the directory default striping
4867 Details : stripe_size -1 was used internally to signal directory stripe
4868 removal, now use "all default" to signal dir stripe removal
4869 as a directory striping of "all default" is not useful
4872 Frequency : common for large clusters running liblustre clients
4874 Description: doing an ls when liblustre clients are running is slow
4875 Details : sending a glimpse AST to a liblustre client waits for every AST
4876 to time out, as liblustre clients will not respond. Since they
4877 cannot cache data we refresh the OST lock LVB from disk instead.
4879 Severity : enhancement
4881 Description: doing an ls at the same time as file IO can be slow
4882 Details : enqueue and other "small" requests can be blocked behind many
4883 large IO requests. Create a new OST IO portal for non-IO
4884 requests so they can be processed faster.
4887 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4889 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4890 created file with mode 000
4891 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4892 allow an ftruncate() as a non-root user to a file with mode 000.
4893 Lustre now respects this flag to disable mode checks when
4894 truncating a file owned by the user
4897 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4900 Description: Revoking locks from clients that went dead or catatonic might take
4902 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4903 cancellation of such locks instant on servers without waiting for
4904 any reply from clients. Clients drops these locks when cancel
4905 notification from server is received without replying.
4908 Frequency : liblustre-only, when liblustre client dies or becomes busy
4910 Description: Doing ls on Linux clients can take a long time with active
4912 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4913 granting such locks to it in the first place if possible. Locks
4914 are taken by proxy on the OST during the read or write and
4915 dropped immediately afterward. Add connect flags handling, do
4916 not grant locks to liblustre clients for glimpse ASTs.
4918 Severity : enhancement
4920 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4921 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4922 sequential reads and behaves badly with applications doing random
4923 reads. Improve it by reading ahead at least read region, and
4924 avoiding excessive large RPC for small reads.
4926 Severity : enhancement
4928 Description: Creating more than 1000 files for a single job may cause a load
4929 imbalance on the OSTs if there are also a large number of OSTs.
4930 Details : qos_prep_create() uses an OST index reseed value that is an
4931 even multiple of the number of available OSTs so that if the
4932 reseed happens in the middle of the object allocation it will
4933 still utilize the OSTs as uniformly as possible.
4938 Description: OST or MDS may oops in ping_evictor_main()
4939 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4940 but doesn't restart at beginning of obd_exports_timed list
4943 Severity : enhancement
4945 Description: improve by-nid export eviction on the MDS and OST
4946 Details : allow multiple exports with the same NID to be evicted at one
4947 time without re-searching the exports list.
4950 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4952 Description: MDS may oops in groups_free()
4953 Details : in rare race conditions a newly allocated group_info struct is
4954 freed again, and this can be NULL. The 2.4 compatibility code
4955 for groups_free() checked for a NULL pointer, but 2.6 did not.
4958 Frequency : common for liblustre clients doing little filesystem IO
4959 Bugzilla : 9352, 7313
4960 Description: server may evict liblustre clients accessing contended locks
4961 Details : if a client is granted a lock or receives a completion AST
4962 with a blocking AST already set it would not reply to the AST
4963 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4966 Frequency : lfs setstripe, only systems with more than 160 OSTs
4968 Description: unable to set striping with a starting offset beyond OST 160
4969 Details : llapi_create_file() incorrectly limited the starting stripe
4970 index to the maximum single-file stripe count.
4973 Frequency : LDAP users only
4975 Description: lconf did not handle in-kernel recovery with LDAP properly
4976 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4978 Severity : enhancement
4980 Description: bind OST threads to NUMA nodes to improve performance
4981 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4982 node and do their allocations there to localize memory access
4984 Severity : enhancement
4986 Description: llmount can determine client NID directly from Myrinet (GM)
4987 Details : the client NID code from gmnalnid was moved directly into
4988 llmount, removing the need to use this or specifying the
4989 client NID explicitly when mounting GM clients with zeroconf
4992 Frequency : if client is started with down MDS
4994 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4995 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4996 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4997 ptlrpc_import_delayed_req() because we didn't check if the
4998 request was interrupted, and we also didn't break out of the
4999 event loop if there was no timeout
5004 Description: data loss during non-page-aligned writes to a single file from
5005 both multiple nodes and multiple threads on one node at same time
5006 Details : updates to KMS and lsm weren't protected by common lock. Resulting
5007 inconsistency led to false short-reads, that were cached and later
5008 used by ->prepare_write() to fill in partially written page,
5009 leading to data loss.
5012 Frequency : always, if lconf --abort_recovery used
5014 Description: lconf --abort_recovery fails with 'Operation not supported'
5015 Details : lconf was attempting to abort recovery on the MDT device and not
5018 Severity : enhancement
5020 Description: remove cleanup logs
5021 Details : replace lconf-generated cleanup logs with lustre internal
5022 cleanup routines. Eliminates the need for client-cleanup and
5025 Severity : enhancement
5027 Description: add support for EAs (user and system) on lustre filesystems
5028 Details : it is now possible to store extended attributes in the Lustre
5029 client filesystem, and with the user_xattr mount option it
5030 is possible to allow users to store EAs on their files also
5032 Severity : enhancement
5034 Description: Add possibility (config option) to show minimal available OST free
5036 Details : When compiled with --enable-mindf configure option, statfs(2)
5037 (and so, df) will return least minimal free space available from
5038 all OSTs as amount of free space on FS, instead of summary of
5039 free spaces of all OSTs.
5041 Severity : enhancement
5043 Description: do not expand extent locks acquired on OST-side
5044 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
5045 by server: they are not cached anyway.
5048 Frequency : when mmap is used/binaries executed from Lustre
5050 Description: Unmmap pages before throwing them away from read cache.
5051 Details : llap_shrink cache now attempts to unmap pages before discarding
5052 them (if unmapping failed - do not discard). SLES9 kernel has
5053 extra checks that trigger if this unmapping is not done first.
5058 Description: lconf didn't resolve symlinks before checking to see whether a
5059 given mountpoint was already in use
5062 Frequency : when migrating failover services
5063 Bugzilla : 6395, 9514
5064 Description: When migrating a subset of services from a node (e.g. failback
5065 from a failover service node) the remaining services would
5066 time out and evict clients.
5067 Details : lconf --force (implied by --failover) sets the global obd_timeout
5068 to 5 seconds in order to quickly disconnect, but this caused
5069 other RPCs to time out too quickly. Do not change the global
5070 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
5072 Severity : enhancement
5073 Frequency : if MDS is started with down OST
5074 Bugzilla : 9439,5706
5075 Description: Allow startup/shutdown of an MDS without depending on the
5076 availability of the OSTs.
5077 Details : Asynchronously call mds_lov_synchronize during MDS startup.
5078 Add appropriate locking and lov-osc refcounts for safe
5079 cleaning. Add osc abort_inflight calls in case the
5080 synchronize never started.
5083 Frequency : occasional (Cray XT3 only)
5085 Description: root not authorized to access files in CRAY_PORTALS environment
5086 Details : The client process capabilities were not honoured on the MDS in
5087 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
5088 been accessed by an authorized user then root was able to access
5089 the file on the local client also. The root user capabilities
5090 are now allowed on the MDS, as this environment has secure UID.
5093 Frequency : occasional
5095 Description: ldiskfs "too long searching" message happens too often
5096 Details : A debugging message (otherwise harmless) prints too often on
5097 the OST console. This has been reduced to only happen when
5098 there are fragmentation problems on the filesystem.
5103 Description: Division by zero in statfs when all OSCs are inactive
5104 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
5105 lov_statfs divides by value returned by lov_get_stripecnt().
5109 Bugzilla : 9489, 3273
5110 Description: First write from each client to each OST was only 4kB in size,
5111 to initialize client writeback cache, which caused sub-optimal
5112 RPCs and poor layout on disk for the first writen file.
5113 Details : Clients now request an initial cache grant at (re)connect time
5114 and so that they can start streaming writes to the cache right
5115 away and always do full-sized RPCs if there is enough data.
5116 If the OST is rebooted the client also re-establishes its grant
5117 so that client cached writes will be honoured under the grant.
5122 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
5123 Details : Now I/O RPCs go to different portal number and (presumably) fast
5124 lock requests (and glimses) and other RPCs get their own service
5125 threads pool that should be able to service those RPCs
5128 Severity : enhancement
5130 Description: Ability to exchange lustre version between client and servers and
5131 issue warnings at client side if client is too old. Also for
5132 liblustre clients there is ability to refuse connection of too old
5134 Details : New 'version' field is added to connect data structure that is
5135 filled with version info. That info is later checked by server and
5139 Frequency : rare, liblustre only.
5140 Bugzilla : 9296, 9581
5141 Description: Two simultaneous writes from liblustre at offset within same page
5142 might proceed at the same time overwriting eachother with stale
5144 Details : I/O lock withing llu_file_prwv was released too early, before data
5145 actually was hitting the wire. Extended lock-holding time until
5146 server acknowledges receiving data.
5149 Frequency : extremely rare. Never observed in practice.
5151 Description: avoid generating lustre_handle cookie of 0.
5152 Details : class_handle_hash() generates handle cookies by incrementing
5153 global counter, and can hit 0 occasionaly (this is unlikely, but
5154 not impossible, because initial value of cookie counter is
5155 selected randonly). Value of 0 is used as a sentinel meaning
5156 "unassigned handle" --- avoid it. Also coalesce two critical
5157 sections in this function into one.
5159 Severity : enhancement
5161 Description: allow liblustre clients to delegate truncate locking to OST
5162 Details : To avoid overhead of locking, liblustre client instructs OST to
5163 take extent lock in ost_punch() on client's behalf. New connection
5164 flag is added to handle backward compatibility.
5166 Severity : enhancement
5167 Bugzilla : 4928, 7341, 9758
5168 Description: allow number of OST service threads to be specified
5169 Details : a module parameter allows the number of OST service threads
5170 to be specified via "options ost ost_num_threads={N}" in the
5171 OSS's /etc/modules.conf or /etc/modprobe.conf.
5175 Bugzilla : 6146, 9635, 9895
5176 Description: servers crash with bad pointer in target_handle_connect()
5177 Details : In rare cases when a client is reconnecting it was possible that
5178 the connection request was the last reference for that export.
5179 We would temporarily drop the export reference and get a new
5180 one, but this may have been the last reference and the export
5181 was just destroyed. Get new reference before dropping old one.
5183 Severity : enhancement
5184 Frequency : if client is started with failover MDS
5186 Description: Allow multiple MDS hostnames in the mount command
5187 Details : Try to read the configuration from all specified MDS
5188 hostnames during a client mount in case the "primary"
5191 Severity : enhancement
5193 Description: Stop sending data to evicted clients as soon as possible.
5194 Details : Check if the client we are about to send or are sending data to
5195 was evicted already. (Check is done every second of waiting,
5196 for which l_wait_event interface was extended to allow checking
5197 of exit condition at specified intervals).
5200 Frequency : rare, normally only when NFS exporting is done from client
5202 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
5204 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
5207 Frequency : always, if srand() is called before liblustre initialization
5209 Description: Liblustre uses system PRNG disturbing its usage by user application
5210 Details : Introduce internal to lustre fast and high-quality PRNG for
5211 lustre usage and make liblustre and some other places in generic
5212 lustre code to use it.
5214 Severity : enhancement
5215 Bugzilla : 9477, 9557, 9870
5216 Description: Verify that the MDS configuration logs are updated when xml is
5217 Details : Check if the .xml configuration logs are newer than the config
5218 logs stored on the MDS and report an error if this is the case.
5219 Request --write-conf, or allow starting with --old_conf.
5221 Severity : enhancement
5223 Description: Handle symlinks in the path when checking if Lustre is mounted.
5224 Details : Resolve intermediate symlinks when checking if a client has
5225 mounted a filesystem to avoid duplicate client mounts.
5230 Description: lconf can hit an error exception but still return success.
5231 Details : The lconf command catches the Command error exception at the top
5232 level script context and will exit with the associated exit
5233 status, but doesn't ensure that this exit status is non-zero.
5238 Description: failure of ptlrpc thread startup can cause oops
5239 Details : Starting a ptlrpc service thread can fail if there are a large
5240 number of threads or the server memory is very fragmented.
5241 Handle this without oopsing.
5244 Frequency : always, only if liblustre and non-default acceptor port was used
5246 Description: liblustre cannot connect to servers with non-default acceptor port
5247 Details : tcpnal_set_default_params() was not called and was therefore
5248 ignoring the environment varaible TCPNAL_PORT, as well as other
5249 TCPNAL_ environment variables
5254 Description: two objects could be created on the same OST for a single file
5255 Details : If an OST is down, in some cases it was possible to create two
5256 objects on a single OST for a single file. No problems other
5257 than potential performance impact and spurious error messages.
5261 Bugzilla : 5681, 9562
5262 Description: Client may oops in ll_unhash_aliases
5263 Details : Client dcache may become inconsistent in race condition.
5264 In some cases "getcwd" can fail if the current directory is
5270 Description: Inode refcounting problems in NFS export code
5271 Details : link_raw functions used to call d_instantiate without obtaining
5272 extra inode reference first.
5276 Bugzilla : 9942, 9903
5277 Description: Referencing freed requests leading to crash, memleaks with NFS.
5278 Details : We used to require that call to ll_revalidate_it was always
5279 followed by ll_lookup_it. Also with revalidate_special() it is
5280 possible to call ll_revalidate_it() twice for the same dentry
5281 even if first occurence returned success. This fix changes semantic
5282 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
5283 reference on a request referred from the intent.
5284 ll_intent_release() then releases such a request.
5287 Frequency : rare, normally benchmark loads only
5289 Description: unlinked inodes were kept in memory on the client
5290 Details : If a client is repeatedly creating and unlinking files it
5291 can accumulate a lot of stale inodes in the inode slab cache.
5292 If there is no other client load running this can cause the
5293 client node to run out of memory. Instead flush old inodes
5294 from client cache that have the same inode number as a new inode.
5297 Frequency : SLES9 2.6.5 kernel and long filenames only
5298 Bugzilla : 9969, 10379
5299 Description: utime reports stale NFS file handle
5300 Details : SLES9 uses out-of-dentry names in some cases, which confused
5301 the lustre dentry revalidation. Change it to always use the
5305 Frequency : rare, unless heavy write-truncate concurrency is continuous
5306 Bugzilla : 4180, 6984, 7171, 9963, 9331
5307 Description: OST becomes very slow and/or deadlocked during object unlink
5308 Details : filter_destroy() was holding onto the parent directory lock
5309 while truncating+unlinking objects. For very large objects this
5310 may block other threads for a long time and slow overall OST
5311 responsiveness. It may also be possible to get a lock ordering
5312 deadlock in this case, or run out of journal credits because of
5313 the combined truncate+unlink. Solution is to do object truncate
5314 first in one transaction without parent lock, and then do the
5315 final unlink in a new transaction with the parent lock. This
5316 reduces the lock hold time dramatically.
5319 Frequency : rare, 2.4 kernels only
5321 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
5322 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
5323 before unmounting their filesystem in order to avoid deadlock.
5324 The kernel_locked() function in 2.4 kernels only checks whether
5325 the kernel lock is held, not whether it is this process that is
5326 holding it as 2.6 kernels do.
5331 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
5332 Details : The client ptlrpc code may be trying to reconnect to a down
5333 server before a previous connection attempt has timed out.
5334 Increase the reconnect interval to be longer than the connection
5335 timeout interval to avoid sending duplicate connections to
5339 Frequency : echo_client brw_test command
5341 Description: fix echo_client to work with OST preallocated code
5342 Details : OST preallocation code (5137) didn't take echo_client IO path
5343 into account: echo_client calls filter methods outside of any
5344 OST thread and, hence, there is no per-thread preallocated
5345 pages and buffers to use. Solution: hijack pga pages for IO. As
5346 a byproduct, this avoids unnecessary data copying.
5350 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
5351 Description: Client can oops in mdc_commit_close() after open replay
5352 Details : It was possible for the MDS to return an open request with no
5353 transaction number in mds_finish_transno() if the client was
5354 evicted, but without actually returning an error. Clients
5355 would later try to replay that open and may trip an assertion
5356 Simplify the client close codepath, and always return an error
5357 from the MDS in case the open is not successful.
5360 Frequency : rare, 2.6 OSTs only
5362 Description: OST may deadlock under high load on fragmented files
5363 Details : If there was a heavy load and highly-fragmented OST filesystems
5364 it was possible to have all the OST threads deadlock waiting on
5365 allocation of biovecs, because the biovecs were not released
5366 until the entire RPC IO was completed. Instead, release biovecs
5367 as soon as they are complete to ensure forward IO progress.
5369 Severity : enhancement
5371 Description: Support for specifying external journal device at mount
5372 Details : If an OST or MDS device is formatted with an external journal
5373 device, this device major/minor is stored in the ext3 superblock
5374 and may not be valid for failover. Allow detecting and
5375 specifying the external journal at mount time.
5380 Description: Mounting an MDS with pending unlinked files may cause oops
5381 Details : target_finish_recovery() calls mds_postrecov() which returned
5382 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
5383 considers this an error and immediately begins cleaning up the
5384 lov, just after starting the mds_lov process
5386 Severity : enhancement
5388 Description: Implement 'lfs df' to report actual free space on per-OST basis
5389 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
5390 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
5391 usage of objects; '-h' to report in human readable format.
5393 ------------------------------------------------------------------------------
5395 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
5402 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
5403 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
5404 at startup during orphan unlinking. Instead, skip buddy update
5405 before setup, as it will be regenerated after recovery is complete.
5410 Description: If an OST is inactive, its locks might reference stale inodes.
5411 Details : lov_change_cbdata() must iterate over all namespaces, even if
5412 they are inactive to clear inode references from the lock.
5414 Severity : enhancement
5415 Frequency : occasional, if non-standard max_dirty_mb used
5417 Description: Client will block write RPCs if not enough grant
5418 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
5419 then the client will block writes while waiting for another RPC
5420 to complete instead of consuming its dirty limit. With change
5421 we get improved performance when max_dirty_mb is small.
5423 Severity : enhancement
5424 Bugzilla : 3389, 6253
5425 Description: Add support for supplementary groups on the MDS.
5426 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
5427 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
5428 lookups for user supplementary groups into a cache.
5432 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
5433 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
5434 for liblustre O_EXCL is enforced by MDS.
5438 Description: readdir never returns NULL in liblustre.
5439 Details : Corrected llu_iop_getdirentries logic, to return offset of next
5440 dentry in struct dirent.
5444 Frequency : liblustre only, depends on application IO pattern
5445 Description: liblustre clients evicted if not contacting servers
5446 Details : Don't put liblustre clients into the ping_evictor list, so
5447 they will not be evicted by the pinger ever.
5449 Severity : enhancement
5451 Description: Add ability to evict clients by NID from MDS.
5452 Details : By echoing "nid:$NID" string into
5453 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
5454 $NID would be instantly evicted from this MDS and from all active
5455 OSTs connected to it.
5459 Description: Do not query file size twice, somewhat slowing stat(2) calls.
5460 Details : lookup_it_finish() used to query file size from OSTs that was not
5465 Description: service threads change working directory to that of init
5466 Details : Starting lustre service threads may pin the working directory
5467 of the parent thread, making that filesystem busy. Threads
5468 now change to the working directory of init to avoid this.
5472 Frequency : during shutdown only
5473 Description: shutdown with a failed MDS or OST can cause unmount to hang
5474 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
5479 Frequency : 2.6 only
5480 Description: chmod/chown may include an extra supplementary group
5481 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
5482 supplementary group and if none is specified this is used.
5485 Bugzilla : 5479 (6816)
5487 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
5488 Details : If lookup is in progress on a file that is unlinked we might try
5489 to revalidate the inode and fail in revalidate after lookup is
5490 complete and ll_file_open() enqueues the open again but
5491 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
5494 Frequency : always, if lconf --abort_recovery used
5496 Description: lconf --abort_recovery fails with 'Operation not supported'
5497 Details : lconf was attempting to abort recovery on the MDT device and not
5500 ------------------------------------------------------------------------------
5502 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
5507 Frequency : rare (only unsupported configurations with a node running as an
5509 Bugzilla : 6514, 5137
5510 Description: Mounting a Lustre file system on a node running as an OST could
5512 Details : OSTs now preallocates memory needed to write out data at
5513 startup, instead of when needed, to avoid having to
5514 allocate memory in possibly low memory situations.
5515 Specifically, if the file system is mounted on on OST,
5516 memory pressure could force it to try to write out data,
5517 which it needed to allocate memory to do. Due to the low
5518 memory, it would be unable to do so and the node would
5519 become unresponsive.
5521 Severity : enhancement
5523 Description: Addition of lconf --service command line option
5524 Details : lconf now accepts a '--service <arg>' option, which is
5525 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
5527 Severity : enhancement
5529 Description: Failover mode is now the default for OSTs.
5530 Details : By default, OSTs will now run in failover mode. To return to
5531 the old behaviour, add '--failout' to the lmc line for OSTs.
5533 Severity : enhancement
5535 Description: Health checks are now provided for MDS and OSTs
5536 Details : Additional detailed health check information on MSD and OSTs
5537 is now provided through the procfs health_check value.
5540 Frequency : occasional, depends on IO load
5542 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
5543 after numerous create/delete cycles
5544 Details : The ext3 inode allocation policy would not allocate new inodes
5545 very well on the OSTs because there are no new directories
5546 being created. Instead we look for groups with free space if
5547 the parent directories are nearly full.
5552 Description: Network or server problems during mount may cause partially
5553 mounted clients instead of returning an error.
5554 Details : The config llog parsing code may overwrite the error return
5555 code during mount error handling, returning success instead
5561 Description: MDS can fail to allocate large reply buffers
5562 Details : After long uptimes the MDS can fail to allocate large reply
5563 buffers (e.g. zconf client mount config records) due to memory
5564 fragmentation or consumption by the buffer cache. Preallocate
5565 some large reply buffers so that these replies can be sent even
5566 under memory pressure.
5570 Frequency : rare (liblustre)
5571 Description: fsx running with liblustre complained that using truncate() to
5572 extend the file doesn't work. This patch corrects that issue.
5573 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
5574 ATTR_SIZE and lsm use in llu_setattr_raw.
5578 Frequency : rare, only 2.6 kernels
5579 Description: Unusual file access patterns on the MDS may result in inode
5580 data being lost in very rare circumstances.
5581 Details : Bad interaction between the ea-in-inode patch and the "no-read"
5582 code in the 2.6 kernel caused the inode and/or EA data not to
5583 be read from disk, causing single-file corruption.
5587 Frequency : rare, only 2.6 filesystems using extents
5588 Description: Heavy concurrent write and delete load may cause data corruption.
5589 Details : It was possible under high-load situations to have an extent
5590 metadata block in the block device cache from a just-unlinked
5591 file overwrite a newly-allocated data block. We now unmap any
5592 metadata buffers that alias just-allocated data blocks.
5596 Frequency : filesystems with default stripe_count larger than 77
5597 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
5598 Details : lconf specifies an inode size of 4096 bytes when the default
5599 stripe_count is larger than 77. This conflicts with the default
5600 inode density of 1 per 4096 bytes. Allocate smaller inodes in
5601 this case to avoid pinning too much memory for large EAs.
5603 ------------------------------------------------------------------------------
5605 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
5610 Frequency : rare (extremely heavy IO load with hundreds of clients)
5612 Description: Client is evicted, gets IO error writing to file
5613 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
5614 caused clients to be evicted for AST timeouts. The fixes in
5615 bug 5192 mean we no longer need to have such short AST timeouts
5616 so ldlm_timeout has been increased.
5619 Frequency : occasional during --force or --failover shutdown under load
5620 Bugzilla : 5949, 4834
5621 Description: Server oops/LBUG if stopped with --force or --failover under load
5622 Details : a collection of import/export refcount and cleanup ordering
5623 issues fixed for safer force cleanup
5626 Frequency : only filesystems larger than 120 OSTs
5627 Bugzilla : 5990, 6223
5628 Description: lfs getstripe would oops on a very large filesystem
5629 Details : lov_getconfig used kfree on vmalloc'd memory
5632 Frequency : only filesystems exporting via NFS to Solaris 10 clients
5633 Bugzilla : 6242, 6243
5634 Description: reading from files that had been truncated to a non-zero size
5635 but never opened returned no data
5636 Details : ll_file_read() reads zeros from no-object files to EOF
5641 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
5642 Details : There's little harm in aborting MDS/OSS recovery and letting it
5643 try again, so I removed the LASSERT and return an error instead.
5645 Severity : enhancement
5647 Description: New debugging infrastructure for tracking down data corruption
5648 Details : The I/O checksum code was replaced to: (a) control it at runtime,
5649 (b) cover more of the client-side code path, and (c) try to narrow
5650 down where problems occurred
5654 Bugzilla : 3819, 4364, 4397, 6313
5655 Description: Racing close and eviction MDS could cause assertion in mds_close
5656 Details : It was possible to get multiple mfd references during close and
5657 client eviction, leading to one thread referencing a freed mfd.
5659 Severity: : enhancement
5660 Bugzilla : 3262, 6359
5661 Description: Attempts to reconnect to servers are now more aggressive.
5662 Details : This builds on the enhanced upcall-less recovery that was added
5663 in 1.4.2. When trying to reconnect to servers, clients will
5664 now try each server in the failover group every 10 seconds. By
5665 default, clients would previously try one server every 25 seconds.
5670 Description: After recovery, certain operations trigger a failed
5671 assertion on a client.
5672 Details : Failing over an mds, using lconf -d --failover, while a
5673 client was doing a readdir() call would cause the client to
5674 LBUG after recovery completed and the readdir() was resent.
5676 Severity : enhancement
5678 Description: Default groups are now added by lconf
5679 Details : You can now run lconf --group <servicename> without having to
5680 manually add groups with lmc.
5683 Frequency : occasional
5685 Description: Nodes with an elan id of 0 trigger a failed assertion
5688 Frequency : always when accessing e.g. tty/console device nodes
5690 Description: tty and some other devices nodes cannot be used on lustre
5691 Details : file's private_data field is used by device data and lustre
5692 values in there got lost. New field was added to struct file to
5693 store fs-specific private data.
5696 Frequency : when exporting Lustre via NFS
5698 Description: NFSD failed occasionally when looking up a path component
5699 Details : NFSD is looking up ".." which was broken in ext3 directories
5700 that had grown large enough to become hashed.
5703 Frequency : Clusters with multiple interfaces not on the same subnet
5705 Description: Nodes will repeatedly try to reconnect to an interface which it
5706 cannot reach and report an error to the log.
5707 Details : Extra peer list entries will be created by lconf with some peers
5708 unreachable. lconf now validates the peer before adding it.
5711 Frequency : Only if a default stripe is set on the filesystem root.
5713 Description: Setting a default stripe on the filesystem root prevented the
5714 filesystem from being remounted.
5715 Details : The client was sending extra request flags in the root getattr
5716 request and did not allocate a reply buffer for the dir EA.
5719 Frequency : occasional, higher if lots of files are accessed by one client
5720 Bugzilla : 6159, 6097
5721 Description: Client trips assertion regarding lsm mismatch/magic
5722 Details : While revalidating inodes the VFS looks up inodes with ifind()
5723 and in rare cases can find an inode that is being freed.
5724 The ll_test_inode() code will free the lsm during ifind()
5725 when it finds an existing inode and then the VFS later attaches
5726 this free lsm to a new inode.
5730 Bugzilla : 6422, 7030
5731 Description: MDS deadlock between mkdir and client eviction
5732 Details : Creating a new file via mkdir or mknod (starting a transaction
5733 and getting the ns lock) can deadlock with client eviction
5734 (gets ns lock and trying to finish a synchronous transaction).
5737 Frequency : occasional
5738 Description: While starting a server, the fsfilt_ext3 module could not be
5740 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
5741 kernels. Previously, lconf would still use the ext3 name
5742 when trying to load modules. Now, it will correctly use
5743 ext3 on 2.4 and ldiskfs on 2.6.
5745 Severity : enhancement
5746 Description: The default stripe count has been changed to 1
5747 Details : The interpretation of the default stripe count (0, to lfs
5748 or lmc) has been changed to mean striping across a single
5749 OST, rather than all available. For general usage we have
5750 found a stripe count of 1 or 2 works best.
5752 Severity : enhancement
5753 Description: Add support for compiling against Cray portals.
5754 Details : Conditional compiling for some areas that are different
5758 Frequency : occasional
5759 Bugzilla : 6409, 6834
5760 Description: Creating files with an explicit stripe count may lead to
5761 a failed assertion on the MDS
5762 Details : If some OSTs are full or unavailable, creating files may
5763 trigger a failed assertion on the MDS. Now, Lustre will
5764 try to use other servers or return an error to the
5768 Frequency : occasional
5770 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
5771 caused lock timeout during readahead (since 1.4.2).
5772 Details : Processes doing readahead might match a lock that hasn't been
5773 granted yet if there are overlapping and conflicting lock
5774 requests. The readahead process waits on ungranted lock
5775 (original lock is CBPENDING), while OST waits for that process
5776 to cancel CBPENDING read lock and eventually evicts client.
5778 Severity : enhancement
5780 Description: Initial enabling of flock support for clients
5781 Details : Implements fcntl advisory locking and file status functions.
5782 This feature is provided as an optional mount flag (default
5783 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
5784 locking are implemented yet, and those that are are not guaranteed
5785 to be completely correct in production environments.
5786 mount -t lustre -o [flock|noflock] ...
5789 Frequency : occasional
5791 Description: OSTs running 2.4 kernels but with extents enabled might trip an
5792 assertion in the ext3 JBD (journaling) layer.
5793 Details : The b_committed_data struct is protected by the big kernel lock
5794 in 2.4 kernels, serializing journal_commit_transaction() and
5795 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
5796 access to this struct. In 2.6 kernels there is finer grained
5797 locking to improve SMP performance of the JBD layer.
5801 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
5803 -----------------------------------------------------------------------------
5805 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
5807 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5808 versions of Lustre. Please update all servers and clients to
5809 version 1.4.2 or later at the same time. You must also run
5810 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5811 to update the configuration logs.
5813 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5814 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5815 - swab llog records of type '0' so we get proper header size/idx (5861)
5816 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5817 - fix rename of one directory over another leaking an inode (5953)
5818 - avoid SetPageDirty on 2.6 (5981)
5819 - don't re-add just-being-destroyed locks to the waiting list (5653)
5820 - when creating new directories, inherit the parent's custom
5821 striping settings if present parent (3048)
5822 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5823 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5824 - don't hold client locks on temporary worklist from l_lru (5666)
5825 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5826 - automatically evict dead clients (5921)
5827 - Update file size properly in create+truncate+fstat case (6196)
5828 - Do not unhash mountpoint dentries, do not allow removal of
5830 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5831 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5832 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5833 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5834 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5835 - Correctly handle reads of files with no objects (6243)
5836 - lctl recover will also mark a device active if deactivate used (5933)
5838 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5839 - allow --write-conf on an MDS with different nettype than client (5619)
5840 - don't write config llogs to MDS for mounts not from that MDS (5617)
5841 - lconf should create multiple TCP connections from a client (5201)
5842 - init scripts are now turned off by default; run chkconfig --on
5843 lustre and chkconfig --on lustrefs to use them
5844 - upcalls are no longer needed for clients to recover to failover
5846 - add --abort-recovery option to lconf to abort recovery on device
5848 - add support for an arbitrary number of OSTs (3026)
5849 - Quota support protocol changes.
5850 - forward compatibility changes to wire structs (6007)
5851 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5852 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5853 - improved "lustre status" script
5854 - initialize blocksize for non-regular files (6062)
5855 - added --disable-server and --disable-client configure options (5782)
5856 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5857 - Vanilla 2.4.29 support
5858 - increase maximum number of obd devices to 520 (6242)
5859 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5860 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5861 SLES 9 2.6.5 kernels
5862 - limit stripes per file to 160 (the maximum EA size) (6093)
5864 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5867 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5868 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5869 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5870 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5871 - small ext3 extents cleanups and fixes (5733)
5872 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5873 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5874 - avoid needless client->OST connect, fix handle mismatch (5317)
5875 - fix DLM error path that led to out-of-sync client, long delays (5779)
5876 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5877 - fix several locking issues related to i_size (5492,5624,5654,5672)
5878 - don't move pending lock onto export if it is already evicted (5683)
5879 - fix kernel oops when creating .foo in unlinked directory (5548)
5880 - fix deadlock in obdfilter statistics vs. object create (5811)
5881 - use time_{before,after} to avoid timer jiffies wrap (5882)
5882 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5883 - Do not leak request if server was not able to process it (5154)
5884 - If mds_open unable to find parent dir, make that negative lookup(5154)
5885 - don't create new directories with extent-mapping (5909, 5936)
5887 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5888 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5889 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5890 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5891 - support for RHEL4 kernel on i686 (5773)
5892 - provide error messages when incompatible logs are encountered (5898)
5894 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5895 * version 1.4.0.10 (1.4.1 release candidate 1)
5897 - don't keep a lock reference when lock is not granted (4238)
5898 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5899 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5900 - drop import inflight refcount on signal_completed_replay error (5255)
5901 - unlock page after async write error during send (3677)
5902 - handle missing objects in filter_preprw_read properly (5265)
5903 - no transno return for symlink open, don't save no-trasno open (3440)
5904 - don't try to complete elan receive that already failed (4012)
5905 - free RPC server reply state on error (5406)
5906 - clean up thread from ptlrpc_start_thread() on error (5160)
5907 - readahead could read extra page into cache that wasn't ejected (5388)
5908 - prevent races in class_attach/setup/cleanup/detach (5260)
5909 - don't dereference de->d_inode after l_dput of de (5458)
5910 - use "int" for stripe value returned from lock_to_stripe (5544)
5911 - mballoc allocation and error-checking fixes in 2.6 (5504)
5912 - block device patches to fix I/O request sizes in 2.6 (5482)
5913 - look up hostnames for IB nals (5602)
5914 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5915 - don't start multiple acceptors for the same port (5277)
5916 - fix incorrect LASSERT in mds_getattr_name (5635)
5917 - export a proc file for general "ping" checking (5628)
5918 - fix "lfs check" to not block when the MDS is down (5628)
5920 - service request history (4965)
5921 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5922 - create an "evict_client" /proc entry on OSTs, like the MDS has
5923 - fix mount usage message, return errors per mount(8) (5168)
5924 - change grep [] to grep "[]" in tests so they work in more UMLs
5925 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5926 - remove ancient LOV_MAGIC_V0 EA support (5047)
5927 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5928 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5929 - allow readahead allocations to fail when low on memory (5383)
5930 - mmap locking landed again, after considerable improvement (2828)
5931 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5933 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5936 - send OST transaction number in read/write reply to free req (4966)
5937 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5938 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5939 - the watchdog thread now runs as interruptible (5246)
5940 - flock/lockf fixes (but it's still disabled, pending 5135)
5941 - don't use EXT3 constants in llite code (5094)
5942 - memory shortage at startup could cause assertion (5176)
5944 - reorganization of lov code
5945 - single portals codebase
5947 - add extents/mballoc support (5025)
5948 - direct I/O reads in the obdfilter (4048)
5949 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5951 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5954 - send OST transaction number in read/write reply to free req (4966)
5955 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5956 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5957 - the watchdog thread now runs as interruptible (5246)
5958 - handle missing objects in filter_preprw_read properly (5265)
5959 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5960 - drop import inflight refcount on signal_completed_replay error (5255)
5961 - unlock page after async write error during send (3677)
5962 - return original error code on reconstructed replies (3761)
5963 - no transno return for symlink open, don't save no-trasno open (3440)
5965 - add pid to ldlm debugging output (4922)
5966 - bump the watchdog timeouts -- we can't handle 30sec yet
5967 - extra debugging for orphan dentry/inode bug (5259)
5969 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5972 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5973 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5974 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5975 - fix NULL dereference in /proc/sys/portals/routes (4827)
5976 - allow failed mdc_close() operations to be interrupted (4561)
5977 - stop precreate on OST before MDS would time out on it (4778)
5978 - don't send partial-page writes before EOF from client (4410)
5979 - discard client grant for sub-page writes on large-page clients (4520)
5980 - don't free dentries not owned by NFS code, check generation (4806)
5981 - fix lsm leak if mds_create_objects() fails (4801)
5982 - limit debug_daemon file size, always print CERROR messages (4789)
5983 - use transno after validating reply (3892)
5984 - process timed out requests if import state changes (3754)
5985 - update mtime on OST during writes, return in glimpse (4829)
5986 - add mkfsoptions to LDAP (4679)
5987 - use ->max_readahead method instead of zapping global ra (5039)
5988 - don't interrupt __l_wait_event() during strace
5990 - add software watchdogs to catch hung threads quickly (4941)
5991 - make lustrefs init script start after nfs is mounted
5992 - fix CWARN/ERROR duplication (4930)
5993 - return async write errors to application if possible (2248)
5994 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5995 - print NAL number in %x format (4645)
5996 - update barely-supported suse-2.4.21-171 series (4842)
5997 - support for sles 9 %post scripts
5998 - support for building 2.6 kernel-source packages
5999 - support for sles km_* packages
6001 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
6004 - ignore -ENOENT errors in osc_destroy (3639)
6005 - notify osc create thread that OSC is being cleaned up (4600)
6006 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
6007 - reconstruct ost_handle() like mds_handle() (4657)
6008 - create a new thread to do import eviction to avoid deadlock (3969)
6009 - let lconf resolve symlinked-to devices (4629)
6010 - don't unlink "objects" from directory with default EA (4554)
6011 - hold socknal file ref over connect in case target is down (4394)
6012 - allow more than 32000 subdirectories in a single directory (3244)
6013 - fix blocks count for O_DIRECT writes (3751)
6014 - OST returns ENOSPC from object create when no space left (4539)
6015 - don't send truncate RPC if file size isn't changing (4410)
6016 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
6017 - bind to privileged port in socknal and tcpnal (3689)
6019 - rate limit CERROR/CWARN console message to avoid overload (4519)
6020 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
6021 - basic mmap support (3918)
6022 - kernel patch series update from b1_4 (4711)
6024 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
6027 - avoid crash during MDS cleanup with OST shut down (2775)
6028 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
6029 - don't use bad inodes on the MDS (3744)
6030 - dynamic object preallocation to improve recovery speed (4236)
6031 - don't hold spinlock over lock dumping or change debug flags (4401)
6032 - don't zero obd_dev when it is force cleaned (3651)
6033 - print grants to console if they go negative (4431)
6034 - "lctl deactivate" will stop automatic recovery attempts (3406)
6035 - look for existing locks in ldlm_handle_enqueue() (3764)
6036 - don't resolve lock handle twice in recovery avoiding race (4401)
6037 - revalidate should check working dir is a directory (4134)
6039 - don't always mark "slow" obdfilter messages as errors (4418)
6041 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
6044 - don't close LustreDB during write_conf until it is done (3860)
6045 - fix typo in lconf for_each_profile (3821)
6046 - allow dumping logs from multiple threads at one time (3820)
6047 - don't allow multiple threads in OSC recovery (3812)
6048 - fix debug_size parameters (3864)
6049 - fix mds_postrecov to initialize import for llog ctxt (3121)
6050 - replace config semaphore with spinlock (3306)
6051 - be sure to send a reply for a CANCEL rpc with bad export (3863)
6052 - don't allow enqueue to complete on a destroyed export (3822)
6053 - down write_lock before checking llog header bitmap (3825)
6054 - recover from lock replay timeout (3764)
6055 - up llog sem before sending rpc (3652)
6056 - reduce ns lock hold times when setting kms (3267)
6057 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
6058 - fix NULL deref and obd_dev leak on setup error (3312)
6059 - replace some LBUG about llog ops with error handling (3841)
6060 - don't match INVALID dentries from d_lookup and spin (3784)
6061 - hold dcache_lock while marking dentries INVALID and hashing (4255)
6062 - fix invalid assertion in ptlrpc_set_wait (3880)
6064 - add libwrap support for the TCP acceptor (3996)
6065 - add /proc/sys/portals/routes for non-root route listing (3994)
6066 - allow setting MDS UUID in .xml (2580)
6067 - print the stack of a process that LBUGs (4228)
6069 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
6072 - don't cleanup request in ll_file_open() on failed MDS open (3430)
6073 - make sure to unset replay flag from failed open requests (3440)
6074 - if default stripe count is 0, use OST count for inode size (3636)
6075 - update parent mtime/ctime on client for create/unlink (2611)
6076 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
6077 - free recovery state on server during a forced cleanup (3571)
6078 - unregister_reply for resent reqs (3063)
6079 - loop back devices mounting and status check on 2.6 (3563)
6080 - fix resource-creation race that can provoke i_size == 0 (3513)
6081 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
6082 - more debugging for page-accounting assertion (3746)
6083 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
6084 - avoid deadlock after precreation failure (3758)
6085 - fix race and lock order deadlock in orphan handling (3450, 3750)
6086 - add validity checks when grabbing inodes from l_ast_data (3599)
6088 - add /proc/.../recovery_status to obdfilter (3428)
6089 - lightweight CDEBUG infrastructure, debug daemon (3668)
6090 - change default OSC RPC parameters to be better on small clusters
6091 - turn off OST read cache for files smaller than 32MB
6092 - install man pages and include them in rpms (3100)
6093 - add new init script for (un)mounting lustre filesystems (2593)
6094 - run chkconfig in %post for init scripts (3701)
6095 - drop scimac NAL (unmaintained)
6097 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
6100 - clean kiobufs before and after use (3485)
6101 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
6102 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
6103 - change init script to fail more clearly if not run as root (1528)
6104 - allow clients to reconnect during replay (1742)
6105 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
6106 - don't do DNS lookups on NIDs too small for IP addresses (3442)
6107 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
6108 - fix cond_resched (3554)
6109 - only evict unfinished clients after recovery (3515)
6110 - allow bulk resend, prevent data loss (3570)
6111 - dynamic ptlrpc request buffer allocation (2102)
6112 - don't allow unlinking open directory if it isn't empty (2904)
6113 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
6114 - remove extraneous obd dereference causing LASSERT failure (3334)
6115 - don't use get_cycles() when creating temp. files on the mds (3156)
6116 - hold i_sem when setting i_size in ll_extent_lock() (3564)
6117 - handle EEXIST for set-stripe, set proper directory name (3336)
6119 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
6120 - fix ksocknal_fmb_callback() error messages (2918)
6122 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
6125 - don't copy lvb into (possibly NULL) reply on error (2983)
6126 - don't deref dentry after dput, don't free lvb on error (2922)
6127 - use the kms to determine writeback rpc length (2947)
6128 - increment oti_logcookies when osc is inactive (2948)
6129 - update client's i_blocks count via lvb messages (2543)
6130 - handle intent open/close of special files properly (1557)
6131 - mount MDS with errors=remount-ro, like obdfilter (2009)
6132 - initialize lock handle to avoid ASSERT on error cleanup (3057)
6133 - don't use cancelling-locks' kms values (2947)
6134 - use highest lock extent for kms, not last one (2925)
6135 - don't dereference ERR_PTR() dentry in error handling path (3107)
6136 - fix thread race in portals_debug_dumplog() (3122)
6137 - create lprocfs device entries at setup instead of at attach (1519)
6138 - common AST error handler, don't evict client on completion race (3145)
6139 - zero nameidata in detach_mnt in 2.6 (3118)
6140 - verify d_inode after revalidate_special is valid in 2.6 (3116)
6141 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
6142 - initialize RPC timeout timer earlier for 2.6 (3219)
6143 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
6144 - print nal/nid for unknown nid (3258)
6145 - additional checks for oscc recovery before doing precreate (3284)
6146 - fix ll_extent_lock() error return code for 64-bit systems (3043)
6147 - don't crash in mdc_close for bad permissions on open (3285)
6148 - zero i_rdev for non-device files (3147)
6149 - clear page->private before handing to FS, better assertion (3119)
6150 - tune the read pipeline (3236)
6151 - fix incorrect decref of invalidated dentry (2350)
6152 - provide read-ahead stats and refine rpc in flight stats (3328)
6153 - don't hold journal transaction open across create RPC (3313)
6154 - update atime on MDS at close time (3265)
6155 - close LDAP connection when recovering to avoid server load (3315)
6156 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
6157 - don't leak open file on MDS after open resend (3325)
6158 - serialize filter_precreate and filter_destroy_precreated (3329)
6159 - loop device shouldn't call sync_dev() for nul device (3092)
6160 - clear page cache after eviction (2766)
6161 - resynchronize MDS->OST in background (2824)
6162 - refuse to mount the same filesystem twice on same mountpoint (3394)
6163 - allow llmount to create routes for mounting behind routers (3320)
6164 - push lock cancellation to blocking thread for glimpse ASTs (3409)
6165 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
6166 - fix rare problem with rename on htree directories (3417)
6168 - allow default OST striping configuration per directory (1414)
6169 - fix compilation for qswnal for 2.6 kernels (3125)
6170 - increase maximum number of MDS request buffers for large systems
6171 - change liblustreapi to be useful for external progs like lfsck (3098)
6172 - increase local configuration timeout for slow disks (3353)
6173 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
6175 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
6178 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
6179 - don't overwrite extent policy data in reply if lock was blocked (2901)
6180 - drop filter export grants atomically with removal from device (2663)
6181 - del obd_self_export from work_list in class_disconnect_exports (2908)
6182 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
6183 - swab reply message in mdc_close, other PPC fixes (2464)
6184 - fix destroying of named logs (2325)
6185 - overwrite old logs when running lconf --write_conf (2264)
6186 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
6187 - fix race in target_handle_connect (2898)
6188 - mds_reint_create() should take same inode create lock (2926)
6189 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
6190 - don't close files for self_export to avoid uninitialized obd (2936)
6191 - allow MDS with the same name as client node (2939)
6192 - hold dentry reference for closed log files for unlink (2325)
6193 - reserve space for all logs during transactions (2059)
6194 - don't evict page beyond end of stripe extent (2925)
6195 - don't oops on a deleted current working directory (2399)
6196 - handle hard links to targets without a parent properly (2517)
6197 - don't dereference NULL lock when racing during eviction (2867)
6198 - don't grow lock extents when lots of conflicting locks (2919)
6200 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
6203 - account for cache space usage on clients to avoid data loss (974)
6204 - lfsck support in lustre kernel code (2349)
6205 - reduce journal credits needed for BRW writes (2370)
6206 - orphan handling to avoid losing space on client/server crashes
6207 - ptlrpcd can be blocked, stopping ALL progress (2477)
6208 - use lock value blocks to assist in proper KMS, faster stat (1021)
6209 - takes i_sem instead of DLM locks internally on obdfilter (2720)
6210 - recovery for initial connections (2355)
6211 - fixes for mds_cleanup_orphans (1934)
6212 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
6213 - block all file creations until orphan recovery completes (1901)
6214 - client remove rq_connection from request struct (2423)
6215 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
6216 - recovery timer race (2670)
6217 - mdc_close recovey bug (2532)
6218 - ptlrpc cleanup bug (2710)
6219 - mds timeout on local locks (2588)
6220 - namespace lock held during RPCs (2431)
6221 - handle interrupted sync write properly (2503)
6222 - don't try to handle a message that hasn't been replied to (2699)
6223 - client assert failure during cleanup after abort recovery (2701)
6224 - leak mdc device after failed mount (2712)
6225 - ptlrpc_check_set allows timedout requests to complete (2714)
6226 - wait for inflight reqs when ptlrpcd finishes (2710)
6227 - make sure unregistered services are removed from the srv_list
6228 - reset bulk XID's when resending them (caught by 1138 test)
6229 - unregister_bulk after timeout
6230 - fix lconf error (2694)
6231 - handle write after unfinished setstripe, stripe-only getstripe (2388)
6232 - readahead locks pages, leaves pending causing memory pressure (2673)
6233 - increase OST request buffers to 4096 on large machines (2729)
6234 - fix up permission of existing directories in simple_mkdir (2661)
6235 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
6236 - don't assign transno to errored transactions (2742)
6237 - don't delete objects on OST if given a bogus objid from MDS (2751)
6238 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
6239 - if rq_no_resend, then timeout request after recovery (2432)
6240 - fix MDS llog_logid record size, 64-bit array alignment (2733)
6241 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
6242 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
6243 - MDS recovery shouldn't delete objects that it has given out (2730)
6244 - if enqueue arrives after completion, don't clobber LVB (2819)
6245 - don't unlock pages twice when trigger_group_io returns error (2814)
6246 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
6247 - don't write pages to disk if there was an error (1450)
6248 - don't ping imports that have recovery disabled (2676)
6249 - take buffered bytes into account when balancing socknal conn (2817)
6250 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
6251 - reconnect unlink llog connection after MDS reconnects to OST (2816)
6252 - remove little-endian swabbing of llog records (1987)
6253 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
6254 - retry reposting request buffers if they fail (1191)
6255 - grow extent at grant time to avoid granting a revoked lock (2809)
6256 - lock revoke doesn't evict page if covered by a second lock (2765)
6257 - disable VM readahead to avoid reading outside lock extents (2805)
6259 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
6260 - updated kernel patches for hp-2.4.20 kernel (2681)
6262 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
6265 - fix truncated write corruption (2366)
6266 - fix for failed assertion in iopen_connect_dentry (1792,2517)
6268 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
6269 - ensure len(uuid) < 37 in lmc (1171)
6270 - fix ia64 OOPS in llog_test (2255)
6271 - zero end of page at obdfilter for partial page writes (2648)
6272 - don't leave stale dentries around after renames (bug 2428)
6273 - fix timeouts when evicting a client with a single lock held (2642)
6274 - set deadline for the initial HELLO message to drain (2634)
6275 - print out dotted-quad IP addresses in the socknal (2302)
6277 - additional debugging for MDS client eviction problem (2443)
6278 - fix mkfsoptions support for osts (2603, 2604)
6280 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
6283 - add series for the vanilla 2.6.0 kernel
6284 - add series for the vanilla 2.4.24 kernel
6285 - add series for a cray x86/64 UL kernel drop
6286 - fix xattr patches for the vanilla 2.4.19 series
6288 - generate true UUIDs in lmc (1171)
6289 - have portals stack dumping break in UML (2466)
6290 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
6291 - allocate with _NOFS in ldlm to avoid deadlock (1933)
6292 - wake callback waiting threads on client eviction (2460)
6293 - Add --ptldebug and --subsystem to lmc (1719)
6294 - update assertion to allow safe interrupt allocation
6295 - set rq_no_resend for cancel requests (2432)
6296 - recalculate ptlrpcd timeout after resend (2494)
6297 - call vfs_rmdir when removing pending directories (2368)
6298 - fix renaming a file to itself (2429)
6299 - lmc creates a default one-stripe lov (2454)
6300 - expand procfs space to handle large clusters (2326)
6301 - increase UML stack to avoid overflow
6302 - update lconf's list of debug and subsystem masks
6303 - fix lfs find --obd (2510)
6304 - /proc tunable for disabling filter read caching (2591)
6305 - stop rpm packages from altering slapd.conf (2301)
6306 - disable nagle in the socknal under 0conf (2578)
6307 - choose mds inode size based on stripe count (2572)
6308 - fix kernel-source rpm problems (2516)
6310 - add --disable-doc to avoid pdf generation (2421)
6311 - update documentation, tests, type-os, comments
6312 - avoid format warnings on ia64
6313 - remove the TOE NAL
6314 - tiny code cleanups by removing unused fields
6316 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
6319 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
6320 - remove the most blatant lies from BUILDING file (2371)
6321 - change default debug level to reasonable production setting
6322 - reduce client side cache size to reduce cache flush time
6323 - reduce max RPCs in flight to avoid unnecessary file fragmentation
6324 - make TCP zerocopy and pinger support enabled by default (2476)
6325 - sync writes completed after process exits caused crashes (2319)
6326 - maintain correct mount count on the MDS (2356)
6327 - backout 1557, because 2316 wasn't really fixed
6328 - better file I/O statistics gathering in /proc
6329 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
6330 - another kernel patch to fix zero-copy TCP function export
6331 - don't take duplicate lock when processing re-sent getattr (2420)
6332 - lctl uses obd_self_export instead of creating new conn (2353)
6333 - MDS/OST recovery case which requires object creation asserted (2425)
6334 - move lfs from /usr/sbin to /usr/bin in packages
6335 - fix race between mds_client_add and mds_client_free (2417)
6336 - use kmalloc instead of slabs in portals (2430)
6337 - don't create duplicate records when a failover MDS is present (2442)
6338 - remove unnecessary mount age check (2332)
6339 - don't remove directory inodes from locks prematurely (2451)
6340 - don't break if MDS service name is the same as hostname (2103)
6341 - fix races in client write RPC generation when cache full (2482)
6343 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
6346 - remove now-unused request->rq_obd (278)
6347 - if an allocation fails, print out how much memory we've used (1933)
6348 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
6349 - add the "configurable stack size" patch to most series files (1256)
6350 - ability to write large log records, for 100+ OST configs (2306)
6351 - fix NULL deref when filter_prep fails (2314)
6352 - fix operator precedence error in filter_sync
6353 - dynamic allocation of socknal TX descriptors (2315)
6354 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
6355 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
6356 - ensure that CWARN messages in Portals always get to the syslog
6357 - __init/__exit are not for prototype decls (ldlm_init/exit)
6358 - x86-64 compile warning fixes
6359 - fix gateway LMC keyword conflict (2318)
6360 - fix MDS lock inversions in getattr/reint paths (1844)
6361 - fix a rare lock re-ordering bug, which caused deadlock (2322)
6362 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
6363 - DLM race condition prevented some lock evictions (2328)
6364 - ENOMEM detection and retry on socknal sends (2230)
6365 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
6366 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
6367 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
6368 - better error messages when a client is rejected during recovery (1505)
6369 - avoid cancelling locks which were never granted, after failure (2330)
6370 - fix i_sem/journal inversion in mds_client_add (2333)
6371 - fix truncate/getattr lock cycle deadlock (2334)
6372 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
6373 - fix two rare exit paths which could leak an l_lock() ref (2321)
6374 - fencepost error in MDS/OST orphan recovery (2226)
6375 - make log record alignment 8 bytes (1988)
6376 - lstripe now fails when requested offset > ost_count (2237)
6377 - ensure that all kernel series have a complete list.h (1607)
6378 - fix crashes in special-file operations (2316)
6379 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
6381 - allow configurable automake binary, for testing new versions
6382 - small update to the lfs documentation
6384 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
6386 * fix negative export reference count in fsfilt_sync (2312)
6388 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
6389 * release candidate 0.9.1
6391 - orphans are moved into the PENDING directory for possible recovery
6392 - replayed opens now open by fid for orphan/rename safety (1042)
6393 - last close of an orphan inode generates a transno (683)
6394 - chdir() and mount() now pin the directory entry (1020)
6395 - avoid CERROR in normal ll_setattr_raw() error case (1500)
6396 - discard very old requests without processing them (1502)
6397 - remove some common, well-understood CERRORs (1505)
6398 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
6399 - clear "grant" flags in OST replies until OST grant code lands (1644)
6400 - fix read performance by not clobbering i_blksize on client (1598)
6401 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
6402 - make LRU size a /proc tunable, clears locks when reduced (707)
6403 - fix some lprocfs rot that prevented ptlbd from loading (1732)
6404 - server locks take references on exports now (1558)
6405 - build fixes for 2.4.20-rh trees (1663)
6406 - return an error from lov_create if all OSCs are inactive (1751)
6407 - fix import levels when a reconnect happens without a timeout (1597)
6408 - exit early from mds_open if we get a lookup error (1749)
6409 - partial page read at EOF wouldn't wait for disk before sending (1642)
6410 - avoid NULL deref in obdfilter when reading page past EOF (1592)
6411 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
6412 - fix LBUG in ll_it_open_error with rc = -2 (1861)
6413 - write/truncate lock inversion (1639)
6414 - Don't auto-load obdclass, portals modules during cleanup (1495)
6415 - fix timestamps from jumping to "now" (1763)
6416 - extra journal assertions (1648)
6417 - add an extra multiunlink test (1771)
6418 - fix read_record/write_record API (1776)
6419 - fix leak of offset_extent, possible incorrect i_size later (1772)
6420 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
6421 - Debugging for the kqswnal_get_idle_tx problems (1820)
6422 - Allow recovery to be attempted multiple times (1536)
6423 - Write out MDS last_rcvd file after it is first created (1600)
6424 - Fix tx_descriptor leak in failed transmit situations (1827)
6425 - ext3 journaling fixes for assertion failure after IO error (1871)
6426 - class_export_put() on freed export after completion AST error (1896)
6427 - Fix revalidate looping in VFS (1322)
6428 - Don't access a freed export during MDS_REINT timeout (1521)
6429 - Add open-unlink recovery support on the MDS (1673,1764)
6430 - Return an error if no MDS data was read from last_rcvd (1946)
6431 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
6432 - Don't setup a disk device that doesn't match exported UUID (317)
6433 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
6434 - avoid committing NULL handle in force close
6435 - local.sh is now a one-stripe LOV configuration
6436 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
6437 - fix ext3 htree duplicate directory entry corruption (1516)
6438 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
6439 - update to new LOV EA format (2097)
6440 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
6442 2003-06-15 Phil Schwan <phil@clusterfs.com>
6445 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
6446 - per-import/export recovery handling (958, 931, 959)
6447 - multiple last-rcvd slots, for serving multiple FSes (949)
6448 - connections are again shared between multiple imp/exports (963, 964)
6449 - "umount -f" would hang if any requests needed to be sent (393, 978)
6450 - avoid pinning large req buffer by copying for queued messages (989)
6451 - add "uuid" to "lctl device" command to help upcalls (991)
6452 - "open" RPCs with transnos would confuse recovery counters (1037)
6453 - do proper endian conversion of all wire messages (288, 340, 891)
6454 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
6455 - call waiting locks callback from LDLM recovery thread (1127, 1151)
6456 - fix ptlrpc_connection leak in target_handle_connect (1174)
6457 - fix import refcounting bug in OST and MDS cleanup (1134)
6458 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
6459 - fix very unlikely obd_types race condition (501)
6460 - remove osc_open hack for echo_client (1187)
6461 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
6462 - a failure in read_inode2 leads to deadlock (1139)
6463 - cancel ack-locks as soon as transaction is committed (1072)
6464 - fix major leaks and crashes in the bulk I/O path (937, 1057)
6465 - make sure to commitrw after any preprw to avoid deadlock (1162)
6466 - failing to execute a file in a lustre FS would lock inode (1203)
6467 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
6468 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
6469 - obd_uuid2tgt would walk off the end of the list (1255)
6470 - on IA64 the osc would give portals incorrect bulk size (1258)
6471 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
6472 - fix lock inversion caused by new llite matching code (1282)
6473 - limit the number of dirty pages on a client to 10MB (1286)
6474 - timed out locks were not being corrected cancelled (1289)
6475 - fix O_DIRECT above 4GB on IA-32 (1292)
6476 * major user-visible changes
6477 - fail out/fail over policy now controlled by the upcall (993)
6479 - add OBD_PING to check server availability and failure (954)
6480 - lustre messages are now sent in sending host order (288, 340, 891)
6481 - add eadatalen to MDS getattr reply (340)
6482 - OST read replies may contain second buffer, with per-page status (593)
6484 2003-03-11 Phil Schwan <phil@clusterfs.com>
6487 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
6488 - failed open()s could cause deadlock; fixed (867, 869)
6489 - stop cancelling OST locks when files are closed (481)
6490 - overlapping XID spaces caused network corruption (851, 853)
6491 - fix unsafe fsfilt counter arithmetic; change to atomic_t
6492 - setattr_raw added, to do single-RPC, server-side setattrs
6493 - lmc/lconf syntax change for OST UUIDs
6494 - fix crashy race condition between ptlrpc_free_req and osc_close
6495 - don't use request in mdc_enqueue if we hit a timeout (889)
6496 - don't set the inode i_size for regular files from the MDS (896)
6497 - handle out of order completion AST (842)
6498 - don't LBUG if a lock request times out after receiving AST (913)
6499 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
6500 - if a bad lock AST arrives, send an error instead of dropping entirely
6501 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
6502 - fix leak in bulk IO when only partially completed (899, 900, 926)
6503 - fix O_DIRECT for ia64 (55)
6504 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
6505 - C-z after timeout could hang a process forever; fixed (977)
6507 - client-side I/O cache (678, 924, 929, 941, 970)
6509 - READPAGE and SETATTRs which don't take server-side locks get
6512 2003-02-11 Phil Schwan <phil@clusterfs.com>
6515 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
6516 - Fix fsfilt_extN_readpage() to read a full page of directory
6517 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
6518 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
6519 - Fix imbalanced LOV object allocation and out-of-bound access (469)
6520 - Most intent operations were removed, in favour of a new RPC mode
6521 that does a single RPC to the server and bypasses most of the VFS
6522 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
6523 - Aggressively cancel local locks on DLM servers
6524 - mds_reint_unlink sends EA to the client if it's the last nlink.
6525 client uses that EA to unlink OST objects.
6526 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
6527 - recursive symlinks were fixed (439)
6528 - fixed NULL deref in DEBUG_REQ
6529 - filter_update_lastobjid no longer calls sync, which annoyed extN
6530 - fixed multi-client small-writes to a single file problem (445)
6531 - fixed mtime updates during file writes (607)
6532 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
6533 - fixed bug in obd_brw_read/write() (under guise of testing 367)
6534 - fixed Linux OST size reporting problem (444, 656)
6535 - OST now updates object mtime with writes or setattr (607, 619)
6536 - client verifies file size before zeroing page past EOF (445)
6537 - OST now writes last allocated objid to disk with allocation (108)
6538 - LOV on echo now works (409)
6540 - mds_reint_unlink sends a new buffer, with the EA included. this
6541 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
6542 set if a regular file was being unlinked, and it was the last link
6543 - use PtlGet from the target for bulk writes (315)
6544 - OST now updates object mtime with writes or setattr (607, 619)
6545 - LDLM now has a grant-time callback to revalidate locked items, if
6547 - Many MDS operations were reorganized to combat race conditions
6549 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
6550 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
6552 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
6555 - Fully reactivate OST imports after reconnection (512, others)
6556 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
6557 - More graceful error handling for truncating on dead OST (515)
6558 - Don't error out unless we're actually accessing dead stripes (474)
6559 - Fix garbage sizes when stripes are missing (410)
6560 - LRU counters were broken, causing constant lock purge (433, 432)
6561 - garbage on read from stripes with failed OSTs (441)
6562 - mark OSCs as active before reconnecting during recovery (438)
6563 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
6564 - lfind did not preserve OST order in output (443)
6565 - symlinks cause hung clients, incorrect data (439)
6566 - stop dereferencing request after dropping refcount (457)
6567 - don't LASSERT(spin_is_locked) on non-SMP (455)
6568 - fixes for many rename() bugs
6569 - fstat didn't correctly synchronize attributes (399)
6570 - server must handle lock cancellation during blocking AST prep (487)
6571 - bulk descriptors were free()d too soon (511)
6572 - fix paths in lconf, which would load incorrect modules (451, 507)
6573 - fix confusing lconf 'host not found' error message (386)
6574 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
6575 - fix race condition in mdc_blocking_ast() for inode access (526)
6576 - fix lov_unpackmd() unpacking wrong number of stripes (537)
6577 - fix lov_set_osc_active() marking wrong OSC inactive (440)
6578 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
6579 - fix multiple writes of stripe MD to MDS (358, maybe 519)
6580 - fix lstripe in several ways (kernel side) (527)
6581 - fix request leak in ldlm_cli_enqueue (262)
6582 - incorrect OSC was marked inactive after OST failure
6583 - call mds_fs_cleanup before unmounting filesystem (524)
6584 - fix races between taking ns_lock and ldlm_lock_change_resource
6585 - fix races updating LOV export open file list
6586 - fix lov_enqueue error path, avoid decref-ing bad lock handle
6587 - fix recovery NULL deref in ldlm_cli_cancel_unused
6588 - fix some DLM races by using new hash table for lock handles (419)
6589 - permit the client to specify desired inodes, at replay
6590 - duplicate requests when we queue them for replay reintegration
6591 - fix last_rcvd offset calculation
6592 - sync after each recovered transaction, so we always make progress
6593 - never, not always, ERESTART requests without transnos
6594 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
6596 - skip replay if the MDS says that the client is already connected
6597 - don't check for a recovery-enabled export to match lctl's UUID
6598 - don't INC_USE_COUNT for phantom exports
6599 - don't crash when cleaning up phantom exports (567)
6600 - don't double-finish or set replay data for errored mdc_open requests
6601 - abort requests when they time out, so we don't get old replies
6602 - send/receive replies for AST messages again
6603 - if the client says that it doesn't have the lock, cancel it on the
6605 - if we timeout during I/O, don't try to cancel an in-use lock; instead
6606 mark it as destroyed, it will all work out when decref is called
6607 - fix module use counts (22, 581)
6609 - ASTs now expect a reply (server cancels lock on error reply)
6611 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
6614 - fix many simultaneous client startup (392)
6615 - fix dentry->d_it clobbering
6616 - credentials weren't being shipped for readdir/getattr operations
6617 - remove invalid assertions triggered during some concurrent MD
6619 - proper Lustre versions added (336, 389)
6620 - fix memory leak for create error case (398)
6621 - fix LOV locking bug that would get cli/srv out of sync
6622 - fix echo client over LOV (409)
6623 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
6624 - fix double-O_EXCL intent crash (424)
6625 - avoid sending multiple lock CANCELs (352)
6627 - MDS can do multi-client recovery (modulo bugs in new code)
6629 - many updates, edits, cleanups
6631 2002-11-18 Phil Schwan <phil@clusterfs.com>
6634 - fix null d_it dereference (346)
6635 - fix full OST/dbench hang (333)
6636 - fix permission problem with file removal (286)
6637 - fix removal of OSCs from LOV when they fail
6638 - fix NULL deref during bulk timeout (214)
6639 - fix problems related to multiple filesystems on one MDS (241)
6640 - fixed serious subtle metadata locking bugs
6641 - free locks on clients when inodes are removed due to memory
6643 - fix inode pointer in lock data (285)
6644 - partial support for multiple MDS on a single host (241)
6645 - data locks weren't cancelled at clear_inode time (290, 311)
6646 - intent locks could lead to unbounded lock growth (205)
6647 - added a maximum lock count, an LRU list, and a flusher
6648 - fix multiple rename (365)
6649 - properly abstracted the echo client
6650 - OSC locked 1 byte too many; fixed
6651 - rewrote brw callback code:
6652 - fixed recovery bugs related to LOVs (306)
6653 - fixed too-many-pages-in-one-write crash (191)
6654 - fixed (again) crash in sync_io_timeout (214)
6655 - probably fixed callback-related race (385)
6657 - Add capability to MDS protocol
6658 - LDLM cancellations and callbacks on different portals
6660 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
6663 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
6664 - defer open object destruction to close time (601981, 138)
6665 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
6666 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
6667 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
6668 - don't call obd_setattr in ll_file_release for destroyed objects
6669 * protocol change to lustre_msg: move |version| and add |flags|
6670 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
6671 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
6672 * added replay of create, unlink, link and rename operations during
6673 MDS failover; recovery should be much more robust now
6674 * remove failed OSCs from LOVs (only lov_create uses this so far)
6675 * the lustre-HOWTO was brought (more) up to date (582544)
6677 2002-10-23 Phil Schwan <phil@clusterfs.com>
6680 - in-use dentries weren't being reused properly (617851)
6681 - prevent multiple LDLM setup (599178)
6682 - fix LOV size calculations for truncate (617853)
6683 - fix client handling of MDS intent errors (POSIX)
6684 - fix permission bug in lovstripe.c test (624321)
6685 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
6686 - truncate past end of file could corrupt data
6687 - proper cleanup after timeouts, crashes, etc (592524, 550815)
6688 - a race in recovery could return ETIMEDOUT to apps (623947)
6689 - building outside the source directory was fixed
6690 * the lustre-HOWTO was brought (more) up to date (582544)
6691 * major progress was made on recovery functionality
6693 2002-10-10 Phil Schwan <phil@clusterfs.com>
6696 - recovery deadlock fix
6697 - rm -rf causes LBUG fix (617817)
6698 - file open by multiple tasks fix (618962)
6699 - directory permissions bugs (602707 and 620007)
6700 - journal_stop fixed with locking (611313)
6701 - O_APPEND failures resolved (618273, perhaps 614459)
6702 - lconf PATH fix (619770)
6703 - IA64 build fix (621450)
6704 - RPC buffer sizes scale with amount of memory
6706 2002-10-01 Phil Schwan <phil@clusterfs.com>
6709 - locks would be cancelled without throwing away data pages,
6710 resulting in inconsistent data (605627)
6711 - inode attributes were not always being refreshed (605627, 612449)
6712 - lconf now continues to cleanup after lctl reports an error
6713 - MDS now enforces user permissions (602707)
6714 - lprocfs cleanup fixed, but not yet enabled (614157)
6715 - fixed infinite server hang, should a client not respond to an AST
6716 - avoid going into recovery if user calls readlink() with a buffer
6717 that's too small (613941)
6718 - AST RPCs no longer require replies (614867) -- this may be changed
6719 - don't crash server if client sends an IOV that's too big (611336)
6720 - fixed lock conversion deadlock (611892)
6721 - fixed the following of symlinks (614622)
6722 * recovery: the server can remove locks from a client that dies, other
6723 clients can make progress
6724 * more extN patch fixes
6725 * compile-time configurable ptlrpc buffer allocations
6727 - collaborative read cache document
6728 - Lustre Lite Performance CDR document-in-progress
6730 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6733 - fix typo in patch-2.4.18
6735 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
6738 - clear ptlrpc request each time in handle_incoming_request()
6739 - unlink of files now destroys the object on the OST
6741 2002-09-19 Peter Braam <braam@clusterfs.com>
6743 * add hard link support
6744 * change obdfile creation method
6745 * kernel patch changed
6747 2002-09-19 Peter Braam <braam@clusterfs.com>
6750 - stack overflow bug in extN fixed
6752 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
6754 * documentation updates
6755 - add man pages for config tools
6756 - update tests/README to describe testing with new config tools
6757 - finish metadata API descriptions
6758 * bug fixes and cleanups
6759 - statfs workaround for 16TB limit
6760 - LOV stripe allocation improved, can stripe on subset of OSTs
6761 - LOV file size/IO offset was wrong for files > 4GB in size
6762 - object EA data was being dropped, caused files to be unreadable
6763 - memory overflow with non-LOV OST caused memory corruption
6764 - fixed regression tests to work with new config tools, obdfilter
6765 - fixed bug when directory size became larger than 1 block
6766 - fixed bug (for single client case) when PWD was deleted
6767 - invalidate local directory pages when doing intent-based ops
6768 - avoid LDLM oops when lock callback contained bad data
6770 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
6772 * documentation updates
6773 * bug fixes and cleanups
6774 - configuration tools
6777 - 64-bit compile warnings
6778 - 64-bit internal statfs data
6780 * test_brw on persistent OST devices
6782 * lprocfs (disabled)
6784 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
6786 * documentation updates
6787 * bug fixes and cleanups
6788 * configuration tools
6790 2002-08-30 Peter J. Braam <braam@clusterfs.com>
6793 * many small fixes to 0_5_4
6794 * io/network handling
6795 * thinkos in MDS operations
6797 2002-08-24 Peter J. Braam <braam@clusterfs.com>
6800 * crucial basic fixes to 0.5.3
6801 * IOR, Iozone work over Elan
6804 2002-08-07 Phil Schwan <phil@clusterfs.com>
6805 * version 0_5_3, our first alpha
6806 * we use the new Portals iovs
6807 * documentation updates
6808 * bug fixes and cleanups
6809 * small changes in the DLM wire protocol
6811 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6812 * version 0_5_1 with some initial stability,
6813 * locking on MD and file I/O.
6814 * documentation updates
6815 * several bug fixes since 0.5.0
6816 * small changes in wire protocol
6818 2002-07-18 Phil Schwan <phil@clusterfs.com>
6820 * delivered as Lustre Light Alpha
6821 * fixed a crash after handling invalid MDS requests
6822 * fixed directory pages for architectures with non-4k pages sizes
6824 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6825 * release version v0_4_4
6826 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6828 * Fixes a number of interruption problems with OST operations.
6829 * Update documentation for portals header changes
6830 * Move all wire protocol structs/defines to lustre_idl.h
6831 * Fixes symlink length bug.
6832 * Add tcpdump to repository.
6834 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6835 * release version v0_4_3
6836 * Fixes statfs for inodes on extN.
6837 * Fixes bug in runtests which would delete /etc/hosts.
6838 * Use 64-bit object IDs wherever possible (not into VFS though)
6839 Remove ost_get_info, which is unused by lustre, and out of date.
6841 2002-07-03 Peter Braam <braam@clusterfs.com>
6842 * release version v0_4_2 Fixes a lookup error (type not passed)
6843 * move forward to head of Portals
6844 * move forward to latest Lustre kernel
6846 2002-06-25 Peter Braam <braam@clusterfs.com>
6847 * release version v0_4_1. Hopefully stable on single node use.