4 2.6.16.60-0.42.8 (SLES 10),
5 2.6.27.39-0.3.1 (SLES11),
6 2.6.18-164.11.1.el5 (RHEL 5)
7 2.6.18-164.11.1.0.1.el5 (OEL 5)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.16 - 2.6.30 vanilla (kernel.org)
11 * Recommended e2fsprogs version: 1.41.10-sun2
12 * The async journal commit feature (bug 19128) and the cancel
13 lock before replay feature (bug 16774) are disabled by default.
15 Severity : enhancement
17 Description: lfs quota output is not very convenient for awk/sed-parsing
18 Details : Some positions if lfs quota output table could be empty or
19 non-empty which made it hard to parse it with scripts, now
20 a dash is put instead of space where there is not supposed
25 Description: file operations after eviction have successful return values
26 Details : use vfs ->flush callback to return any pending async errors
31 Description: mdsrate fails to write after 1.3+M files opened
32 Details : decrease memory usage on clients by recycling dentries and
37 Description: obdfilter-survey gives unreasonably high numbers
38 Details : Wait for all threads to complete when running test_brw.
42 Description: do not set lustre read_only device when server umount and keep
43 client records for recoverable ones
47 Description: Prevent failover nids from registering with MGS first
51 Description: move sync_on_lock_cancel tunable to the obdfilter layer
52 Details : move the tunable to trigger a journal flush on lock cancel from
53 the ost layer to the obdfilter layer. This tunable is useful
54 when using the async journal commit feature.
56 -------------------------------------------------------------------------------
58 2010-04-30 Oracle, Inc.
60 * Support for kernels:
61 2.6.16.60-0.42.8 (SLES 10),
62 2.6.27.39-0.3.1 (SLES11),
63 2.6.18-164.11.1.el5 (RHEL 5)
64 2.6.18-164.11.1.0.1.el5 (OEL 5)
65 * Client support for unpatched kernels:
66 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
67 2.6.16 - 2.6.30 vanilla (kernel.org)
68 * Recommended e2fsprogs version: 1.41.10-sun2
69 * The async journal commit feature (bug 19128) and the cancel
70 lock before replay feature (bug 16774) are disabled by default.
74 Description: open-unlinked directories trigger MDS LBUG
75 Details : Fix regression introduced by the patch from bug 19640.
76 ext3_inc_count() can reset nlink to 1 when the directory
77 is indexed and inode->i_nlink == 2. Work around the problem
78 by incrementing nlink by 2 instead of 1.
82 Description: Reconnects are not throttled
83 Details : Don't wake up pinger on reconnect failures and rely on regular
84 pings to trigger the next reconnection. Please note that the
85 pinger already uses a smaller interval if the import is
89 Frequency : only with NFS export
91 Description: Console flooded with error message from ll_inode_from_lock()
93 Details : in mds_open, initialize the child_res_id before enqueuing
94 the OPEN lock for the child inode, then to avoid senting
95 wrong ldlm_res_id to client.
99 Description: allow multiple instances of the same nid in NID hash
100 Details : Case of multiple separate clients from the same NID (as
101 with liblustre) is legitimate and so we should allow
102 multiple instances of the same NID in nid hash.
106 Description: the readahead code can sleep on a semaphore while holding a
108 Details : in ras_update, "lov_get_info" could be called during increasing
109 readahead windows, which tries to get the mutex lock "lov_lock"
110 while holding the spin_lock "ras_lock", then causes system
115 Description: ASSERTION(cli->cl_avail_grant >= 0) failed
116 Details : fix assertion failure in the grant code.
120 Description: Use CNETERR (which is rate limited) in specific places in
121 the portal's LNET driver to avoid flooding the console.
125 Description: include last created object in precreate slow case
129 Description: don't do rep-ack if not created anything
130 Details : mds_open currently always put a lock into a rep-ack regardless
131 if something was created or not. This is pointless and only
132 creates needless contention. In fact the entire idea was to do
133 this for real creates as a recovery protection.
137 Description: Spurious error messages from smp_processor_id() on preemptible
139 Details : Disable a preemption by grabbing the lock in fs_trace_get_tcd()
140 first. The function fs_trace_get_tcd() was moved up.
144 Description: interval_erase() fix
145 Details : interval_erase() calls update_maxhigh() properly when child
150 Description: "lfs df" does not print stats for all mountpoints
151 Details : Print all mounted lustre filesystems with "lfs df"
155 Description: lfs setstripe -p not longer work with a relative pathname
156 Details : Use realpath() to provide absolute pathname.
160 Description: fix for truncated reply buffer
161 Details : reply buffer could be referred by reply_in_callback after released
166 Description: lustre.lov error when backing up symlinks with extended attributes
167 Details : Improved logic in ll_listxattr()
171 Description: properly handle null value for setattr -n lustre.lov
172 Details : Running "setfattr -n trusted.lov ." causes a NULL dereference
173 in ll_setxattr() due to no checking if "value" is NULL.
174 This command now resets to the default striping when executed
179 Description: stack overflow on lock cancellation due to fsync call
180 Details : sync_on_lock_cancel is needed for recovery when async journal
181 is enabled, but we actually just need to make sure that
182 metadata blocks have hit the journal, so doing a fs sync
183 should be enough and should consume less stack (just create an
184 empty handle and commmit it).
188 Description: using current->journal_info to store per-thread data leads
189 to problem under memory pressure
190 Details : disable the per-thread data (current->journal_info) containing
191 the lock info during I/O to work around the issue for short tem
195 Description: control DCACHE_LUSTRE_INVALID flag with MDS_INODELOCK_LOOKUP lock
196 Details : DCACHE_LUSTRE_INVALID is controlled by MDS_INODELOCK_LOOKUP
197 lock which is corresponding to "IT_LOOKUP", do not skip invalidate
202 Description: Cannot send after transport shutdown
203 Details : Clear imp_vbr_failed flag upon eviction
207 Description: soft lock in request set code during recovery
208 Details : during recovery, uses req->rq_set itself to replay the request
209 instead of ptlrpcd_recovery_pc
213 Description: Use CFS_ALLOC_IO instead of _STD in llap_from_page_with_lockh
214 Details : During an ll_readahead under ll_readpage, we have seen the the
215 OBD_SLAB_ALLOC hang under ldlm_pools_shrink when trying to lock
216 a page that is already locked by the readahead code.
221 Description: stop waitting for next replay transno if shutdown
222 Details : if the system is shutting down, wake up service thread blocked
223 to wait for next replay transno during recovery, then all the
224 references held by queued requests can be dropped and device
229 Description: lov_merge_lvb()) ASSERTION(spin_is_locked(&lsm->lsm_lock)) failed
230 Details : Protect lli->lli_smd pointer updates with lli->lli_lock.
234 Description: per-nid stats should not access lustre-hash internal structures
239 Description: mount.lustre fails to pass some options to mount()
244 Description: ext4 extent allocation is slower than in ext3
245 Details : Increase the default value of MB_DEFAULT_ORDER2_REQS to 8,
246 enlarge ext4 preallocation table for 2048 4K blocks extents
251 Description: incorrect triggering of synchronous IO
252 Details : The OSC can mistakenly fall back to synchronous IO when the
253 max_dirty_mb limit is reached and no write requests have yet
254 been issued. This can occur when the dirty pages are spread
255 over many files all of which are below the optimal request size.
259 Description: Optimize quota_ctl operations by sending requests in parallel
260 Details : Send MDS->OST quota_ctl requests in parallel, do not resend.
261 Compiled from two attachments in the ticket.
265 Description: ordering issue between transaction start & i_mutex
266 Details : start the transaction earlier in llog_lvfs_destroy to get
267 transaction start and inode mutex lock nested properly.
271 Description: lru resize SLV can get stuck
272 Details : calculate SLV with a greater precision to not lose small
273 changes due to interger math truncation; round up SLV only
274 if the amount of granted locks less than the limit to not
275 get stuck with this SLV
279 Description: avoid divide-by-zero in lprocfs_rd_import()
281 Severity : enhancement
283 Description: "weak-modules" support
284 Details : Implement "weak-modules" support which enables kernel modules
285 to be used with any kernel that implements the same kABI. In
286 order to achieve this modules are now installed in
287 /lib/modules/$(uname -r)/updates/kernel on all distributions.
291 Description: lfs quota failed when OSTs are down
292 Details : really return approximate block/inode usage when OSTs are down
296 Description: llobdstat fix and enhancement
297 Details : add a counter to set a limit to how many samples will be returned
298 fix a wildcard in the path to limit to obdfilter stats only
302 Description: recovery-small 51 hang - MDT can not stop, Mount still busy with
304 Details : abort lock enqueue processing sooner when we umount is in progress.
306 Severity : enhancement
308 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and OEL5.4 kernel to
309 2.6.18-164.11.1.0.1.el5.
311 Severity : enhancement
313 Description: error message improvements
314 Details : Use INFO/WARN instead of WARN/ERROR for the slow messages.
315 Simplify MDT/OST service start message.
316 Suppress "changing the import ..." warning.
319 Bugzilla : 21961/17914
320 Description: ignore trailing -mdc when determining index number
323 Frequency : only with SLES11
325 Description: fix for a race condition in linux quotas implementation
326 Details : dq_flags(struct dquot) access is not properly locked which could
327 lead to certain inconsistencies when accessing it using non-atomic
328 bit operations like __set_bit in do_set_dqblk.
332 Description: return any pending async errors in close(2) using flush callback
334 -------------------------------------------------------------------------------
336 2010-01-29 Sun Microsystems, Inc.
338 * Support for kernels:
339 2.6.16.60-0.42.8 (SLES 10),
340 2.6.27.39-0.3.1 (SLES11),
341 2.6.18-164.11.1.el5 (RHEL 5)
342 2.6.18-164.6.1.0.1.el5 (OEL 5)
343 * Client support for unpatched kernels:
344 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
345 2.6.16 - 2.6.30 vanilla (kernel.org)
346 * Recommended e2fsprogs version: 1.41.6.sun1
347 * The async journal commit feature (bug 19128) and the cancel
348 lock before replay feature (bug 16774) are disabled by default.
350 Severity : enhancement
352 Description: Update RHEL5.4 kernel to 2.6.18-164.11.1.el5 and
353 OEL5.4 kernel to 2.6.18-164.11.1.0.1.el5.
355 Severity : enhancement
356 Bugzilla : 21511/19848
357 Description: Update SLES11 kernel to 2.6.27.39-0.3.1.
359 Severity : enhancement
361 Description: Update supported SLES10 kernel to 2.6.16.60-0.42.8.
363 Severity : enhancement
365 Description: Update kernel to RHEL5.4 2.6.18-164.6.1.el5 and
366 OEL5 2.6.18-164.6.1.0.1.el5(Both in-kernel OFED enabled).
368 Severity : enhancement
370 Description: Build kernels (RHEL5, OEL5 and SLES10/11) using the vendor's own
373 Severity : enhancement
375 Description: Vanilla kernel 2.6.30 patchless client support.
380 Description: bad entry in directory xxx: inode out of bounds
381 Details : fix locking issue in the rename path which could race with any
382 other operations updating the same directory.
384 Severity : enhancement
386 Description: Make watchdog timer messages to be more clear and descriptive.
390 Description: cp -p command does not preserve the dates and timestamp
391 Details : mtime could be spoiled by a write callback
395 Description: Clear imp_force_reconnect correctly in ptlrpc_connect_interpret()
399 Description: Allow non-root access for "lfs check".
400 Details : Added a check in obd_class_ioctl() for OBD_IOC_PING_TARGET.
402 Severity : enhancement
404 Description: quotacheck performance/scaling issues
405 Details : reduce quotacheck time on empty filesystem by skipping uninit
408 Severity : enhancement
410 Description: Enhancement for lfs(1) command to use numeric uid/gid.
412 Severity : enhancement
414 Description: Adjust locks' extents on their first enqueue, so that at the time
415 they get granted, there is no need for another pass through the
416 queues since they are already shaped into the proper forms.
420 Description: Fix mds_shrink_intent_reply()/mds_intent_policy() to pass correct
421 arguments and prevent LBUG() in lustre_shrink_reply_v2().
425 Description: Change tunefs.lustre and mkfs.lustre --mountfsoptions so that
426 exactly the specified mount options are used. Leaving off
427 any "mandatory" mount options is an error. Leaving off any
428 default mount options causes a warning, but is allowed.
429 Change errors=remount-ro from mandatory to default. Sanitize
430 the mount string before storing it. Update man pages accordingly.
434 Description: mds_getattr() should return 0, even if mds_fid2entry() fails with
435 -ENOENT. Also fix in ptlrpc_expire_one_request() to print signed
438 Severity : enhancement
440 Description: Remove set_info(KEY_UNLINKED) from MDS/OSC
442 Severity : enhancement
444 Description: Clients can replay thousands of unused locks during recovery
445 Details : Don't replay unused locks (only read locks for now) during
446 recovery. This feature is disabled by default and can be
447 enabled by running the following command on the clients:
448 lctl get_param ldlm.cancel_unused_locks_before_replay
452 Description: can't stat file in some situation.
453 Details : improve initialize osc date when target is added to mds and
454 ability to resend too big getattr request is client isn't have info
459 Description: Prevent inconsistences between linux and lustre mount structures.
460 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
461 Make the sleep interruptible.
463 Severity : enhancement
465 Description: Communicate OST degraded/readonly state via statfs to MDS
466 Details : Flags in the statfs returned from OSTs indicate whether the
467 OST is in a degraded RAID state, or if the filesystem has
468 turned read-only after a filesystem error is detected.
473 Description: don't panic if EPROTO was hit when reading symlink
474 Details : correctly handling request reference in error cases.
479 Description: open sometimes returns ENOENT instead of EACCES
480 Details : checking permission should be part of open part of mds_open, not
481 lookup part. so server should be set DISP_OPEN_OPEN disposition
482 before starting permission check.
483 Also not need revalidate dentry if client already have LOOKUP lock.
487 Description: enable client interface failover
488 Frequency : on servers with multiple network interfaces
489 Details : When a child reconnects from another NID, properly update export
490 nid hash position and ldlm reverse import.
492 Severity : enhancement
494 Description: implemented direct I/O with arbitrary (nonaligned) memory
495 addresses and file offsets.
497 Severity : enhancement
499 Description: added more recovery timeout options.
501 Severity : enhancement
503 Description: added llapi_file_open, llapi_file_create, llapi_file_get_stripe
508 Description: Avoid deadlock for local client writes
509 Frequency : only on systems with clients writing to an OST on the same node
510 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
511 memory freeing context. This allows OST threads to set the
512 PF_MEMALLOC flag on task structures in order to allocate memory
513 from reserved pools and complete IO.
514 Use GFP_HIGHUSER for OST allocations for non-local client writes,
515 so that the OST threads generate memory pressure and allow
516 inactive pages to be reclaimed.
521 Descriptoin: lock ordering violation between &cli->cl_sem and _lprocfs_lock
522 Details : move ldlm namespace creation in setup phase to avoid grab
523 _lprocfs_lock with cli_sem held.
527 Frequency : only during format of test systems
528 Description: Unable to run several mkfs.lustre on loop devices at the same time
529 Details : mkfs.lustre returns error 256 on the concurrent loop devices
530 formatting. The solution is to proper handle the error.
532 Severity : enhancement
534 Description: implement async create (obd_async_create) method for osc, to avoid
535 too long waiting new ost objects with holding ldlm lock.
539 Frequency : occasionally during network problems
540 Description: client not allowed to reconnect to OST because of active request
541 Details : abort bulk requests received by the OST once the client has timed
542 out since the client will resend the request anyway. The client
543 also now retries to reconnect to the same server if a connect
544 request failed with EBUSY or -EAGAIN.
547 Frequency : rare, if used wide striped file and one ost in down.
549 Descriptoin: don't return error if we created a subset of objects for file.
550 Details : lov_update_create_set() uses set->set_success as index for created
551 objects, so if some requests failed, they will have hole at end of
552 array and we can use qos_shrink_lsm for allocate correct lsm.
556 Description: Slow stale export processing during normal start up
557 Details : The global mgc lock prevents OST setup to be run in parallel.
558 Replace the global lock with a per-config_llog_data semaphore.
562 Description: Out or order replies might be lost on replay
563 Details : In ptlrpc_retain_replayable_request if we cannot find retained
564 request with tid smaller then one currently being added, add it
565 to the start, not end of the list.
569 Description: BUG: soft lockup - CPU#1 stuck for 10s! [ll_mdt_07:4523]
570 Details : add cond_resched() calls to avoid hogging the cpu for too long
571 in the hash code. Make also lustre_hash_for_each_empty() more
574 Severity : enhancement
576 Description: Performance improvements for debug messages with D_RPCTRACE,
577 D_LDLM, D_QUOTA options.
580 Frequency : only with NFS export
582 Description: (lov_merge.c:74:lov_merge_lvb())
583 ASSERTION(spin_is_locked(&lsm->lsm_lock)) failed (SR 71691004)
584 Details : Fix a race in the nfs export code by populating inode
585 info while the new inode is still locked
587 Severity : enhancement
589 Description: add a new file in procfs called force_lbug. Writting to this
590 file triggers a LBUG. Only for test purpose.
594 Description: OOM killer causes node hang
595 Details : really interrupt the sleep in osc_enter_cache on signals
599 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
600 Details : fix race during quota release on the slave.
602 Severity : enhancement
604 Description: smaller hash bucket sizes, cleanups
605 Details : increase hash table sizes and enabled rehashing for pools, uuid,
608 Severity : enhancement
610 Description: Add ldiskfs maxdirsize mount option
611 Details : add max_dir size mount option
615 Description: panic in ll_statahead_thread
616 Details : prevent parent thread to be killed before its child
619 Frequency : only with 16TB device
621 Description: unable to perform "mount -t lustre" of 16TB OST device
622 Details : Mounting 16TB LUNs failed due to three bugs in mkfs.lustre.
626 Description: ASSERTION(atomic_read(&imp->imp_inflight) == 0) failed
627 Details : unregistering should be zero if no RPC inflight.
631 Description: hyperion: Oops during metabench
632 Details : Correct the refcount of lov_request_set
634 Severity : enhancement
636 Description: Add mptlinux and nxge drivers to Lustre builds
638 Severity : enhancement
640 Description: Fix watchdog timer message to be more clear
641 Details : Make watchdog timer messages more clear and descriptive.
645 Description: LNET soft lockups in socknal_cd thread
646 Details : don't hog CPU for active-connecting if another connd is
647 accepting connecting-requst from the same peer
651 Description: recovery-small test_17 hang
652 Details : Land several AT improvements & fixes.
656 Description: MDS panic and hanging client processes
657 Details : Replace exp_ops_stats with exp_nid_stats->nid_stats
661 Description: OSS stuck in recovery.
662 Details : fix race during recovery. class_unlink_export,
663 class_set_export_delayed and target_queue_last_replay_reply
664 may race while increasing/decreasing obd_recoverable_clients
665 and obd_delayed_clients, causing recovery to wait forever.
667 Severity : enhancement
669 Description: add cascading_rw.c to lustre/tests
673 Description: filter_last_id() NULL deref
674 Details : lprocfs_filter_rd_last_id() should check for the fully
675 setup obd device, before proceeding further.
677 Severity : enhancement
679 Description: Loadgen improvements
680 Details : stacksize and locking fixes for loadgen
684 Description: Quiet CERROR("dirty %d > system dirty_max %d\n"
685 Details : The atomic_read() allowing the atomic_inc() are not covered
686 by a lock. Thus they may safely race and trip this CERROR()
687 unless we add in a small fudge factor (+1).
689 Severity : enhancement
691 Description: shrink_slab: nr=-9223362083340912175
692 Details : fix spurious message from shrink_slab reporing negative nr
696 Description: Quiet bogus previously committed transno error
697 Details : suppress the "server went back in time" error message which
698 is always printed even in the common case after a client eviction
700 Severity : enhancement
702 Description: Parallel statfs() calls result in client eviction
703 Details : cache statfs data for 1s.
707 Description: parallel-scale test_compilebench: @@@@@@ FAIL: compilebench
709 Details : fix serveral issues in pinger code causing clients not to ping
710 servers for too long, resulting in evictions.
714 Description: e2fsck should warn when MMP update interval is extended
715 Details : print mmp_check_interval and make it possible to abort mount
716 operation in case it takes too long.
720 Description: mdsrate-create-large.sh, BUG: soft lockup - CPU#0 stuck for 10s!
721 Details : fix bug in the RHEL5's jbd2 callback patch.
725 Description: drop number of active requests when queued for recovery
726 Details : Now that we take a reference on the original request instead of
727 making a copy of it for recovery. We need to drop the number of
728 active requests or the queued requests will prevent all request
729 processing when they exceed (srv->srv_threads_running - 1).
731 Severity : enhancement
733 Description: refuse to invalidate operational quota files when they are in use
734 Details : an attempt to invalidate operational quota files on the quota
735 master is not actually permitted by VFS (returning -EPERM), but we
736 should not depend on that and should return the error earlier.
740 Description: Applications stuck in jbd2_log_wait_commit during exit
741 Details : fix deadlock between kjournald2 trying to acquire the page lock
742 owned by an ost_io thread waiting for journal commit.
744 -------------------------------------------------------------------------------
746 2009-10-16 Sun Microsystems, Inc.
748 * Support for kernels:
749 2.6.16.60-0.42.4 (SLES 10),
750 2.6.27.29-0.1 (SLES11, i686 & x84_64 only),
751 2.6.18-128.7.1.el5 (RHEL 5),
752 2.6.18-128.7.1.el5 (OEL 5),
753 * Client support for unpatched kernels:
754 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
755 2.6.16 - 2.6.27 vanilla (kernel.org)
756 * Recommended e2fsprogs version: 1.41.6.sun1
757 * ext4 support for RHEL5 is experimental and thus should not be
760 Severity : enhancement
762 Description: Add OEL5 support.
764 Severity : enhancement
766 Description: Update kernel to SLES11 2.6.27.29-0.1.
770 Description: File checksum failures with OST read cache on
771 Details : Disable page poisoning when the bulk transfer has to be aborted
772 because the client got evicted.
776 Description: Don't allow make backward step on assiging osc next id.
777 Details : race between allocation next id and ll_sync thread can be cause
778 of set wrong osc next id and can be kill valid ost objects.
780 Severity : enhancement
782 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
784 Severity : enhancement
786 Description: Update kernel to SLES10 SP2 2.6.16.60-0.42.4.
790 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
794 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
798 Description: Error handling in osc_statfs_interpret() has been improved.
799 Details : Check in osc_statfs_interpret() for EBADR.
803 Description: Do not send statfs() requests to OSTs disabled by administrator.
804 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
808 Description: Do not update ctime for the deleted inode.
809 Details : Check in mds_reint_unlink() before calling fsfilt_setattr().
813 Description: Increase of the size of the LDLM resource hash.
814 Details : Bump up RES_HASH_BITS=12.
818 Description: correctly send lsm on open replay
819 Details : MDS is trust to LSM size on replay open, but client can set wrong size
824 Description: Deadlock between filter_destroy() and filter_commitrw_write().
825 Details : filter_destroy() does not hold the DLM lock over the whole
826 operation. If the DLM lock is dropped, filter_commitrw() can go
827 through, causing the deadlock between page lock and i_mutex.
828 The i_alloc_sem should also be hold in filter_destroy() while
833 Description: truncate starts GFP_FS allocation under transaction causing deadlock
834 Details : ldiskfs_truncate calls grab_cache_page which may start page
835 allocation under an open transaction. This may lead to
836 calling prune_icache with consequent lustre reentrance.
839 Frequency : only when down/upgrading the MDS to 1.6/1.8 while 1.8 clients are
840 still up and when the OST pool feature is used
842 Description: interop testing got LBUG when run dd with OST pool
843 :LustreError: 30032:0:(llite_lib.c:1913:ll_replace_lsm()) LBUG
844 Details : down/upgrading the MDS to a version that doesn't/does support OST
845 pool can cause clients to crash because the lsm has changed
850 Description: missing tree_status on 1.8.1 RPM build
851 Details : make rpms failed due because the tree_status file is missing.
855 Description: continuing LustreError "mds adjust qunit failed!"
856 Details : don't print message on the console when ->adjust_qunit fails.
860 Description: don't increase ldlm timeout if previous client was evicted
861 Details : if a client doesn't respond to a blocking callback within the
862 adaptive ldlm enqueue timeout, don't adjust the adaptive estimate
863 when the lock is next granted.
867 Description: ost is being unmounted w/o all writes to last_rcvd landing on disk.
868 affects recovery negatively.
869 Details : make sure all exports have been properly destroyed by the zombie
870 thread processed before stopping the target.
874 Description: Performance degradation with O_DIRECT between 1.6 & 1.8.1 b190
875 Details : disable write barrier for ext4/SLES11.
879 Description: Kernel panic - not syncing: Out of memory and no killable
880 processes... on OSS when iozone
881 Details : fix memory leak in the journal checksum patch.
885 Description: group quota "too many blocks" OSS crashes
886 Details : we should keep the same uid/gid for lquota_chkquota() and
887 lquota_pending_commit()
891 Description: LustreError: 9153:0:(quota_context.c:622:dqacq_completion()) LBUG
892 Details : don't LBUG on release quota error. Just a workaround until the
893 problem is understood.
895 ------------------------------------------------------------------------------
896 2009-07-31 Sun Microsystems, Inc.
898 * Support for kernels:
899 2.6.16.60-0.39.3 (SLES 10),
900 2.6.27.23-0.1 (SLES11, i686 & x84_64 only),
901 2.6.18-128.1.14.el5 (RHEL 5),
902 * Client support for unpatched kernels:
903 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
904 2.6.16 - 2.6.27 vanilla (kernel.org)
905 * Recommended e2fsprogs version: 1.41.6.sun1
906 * File join has been disabled in this release, refer to Bugzilla 16929.
907 * NFS export disabled when stack size < 8192. Since the NFSv4 export
908 of Lustre filesystem with 4K stack may cause a stack overflow. For
909 more information, please refer to bugzilla 17630.
910 * ext4 support for RHEL5 is experimental and thus should not be
913 Severity : enhancement
915 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
918 Frequency : with 1.8 server and 1.6 clients
920 Description: correctly shrink reply for avoid send too big message to client.
921 Details : 1.8 mds is allocate to big buffer to LOV EA data and this produce
922 some problems with sending this reply to 1.6 client.
926 Description: Repeated atomic allocation failures.
927 Details : Use GFP_HIGHUSER | __GFP_NOMEMALLOC flags for memory allocations
928 to generate memory pressure and allow reclaiming of inactive pages.
929 At the same time, do not allow to exhaust emergency pools.
930 For local clients the use of GFP_NOFS will be introduced in 1.8.2
932 Severity : enhancement
933 Bugzilla : 19846, 18289
934 Description: Update kernel to RHEL5 2.6.18-128.1.14.el5.
936 Severity : enhancement
937 Bugzilla : 19625, 16893, 18668, 19848
938 Description: Add support for SLES11 2.6.27.23-0.1.
940 Severity : enhancement
942 Description: Update client support to vanila kernels up to 2.6.27.
944 Severity : enhancement
946 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
948 Severity : enhancement
950 Description: Compile with -Werror by default for i686 and x86_64.
954 Description: resolve race between obd_disconnect and class_disconnect_exports
955 Details : if obd_disconnect will be called to already disconnected export he
956 forget release one reference and osc module can't unloaded.
958 Severity : enhancement
960 Description: move AT tunable parameters for more consistent usage
961 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
965 Descriptoin: correctly skip time estimate if in recovery
966 Details : rq_send_state insn't bitmask so using bitwise ops is forbid.
970 Descriptoin: OSS DeadLock
971 Details : Use trylock to prevent deadlock when shrink icache.
973 Severity : enhancement
975 Description: Allow tuning service thread via /proc
976 Details : For each service a new
977 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
978 created that can be used to set min/max thread counts, and get the
979 current number of running threads.
981 Severity : enhancement
983 Description: Add state history info file, enhance import info file
984 Details : Track import connection state changes in a new osc/mdc proc file;
985 add overview-type data to the osc/mdc import proc file.
989 Description: Reduce small size read RPC
990 Details : Set read-ahead limite for every file and only do read-ahead when
991 available read-ahead pages are bigger than 1M to avoid small size
996 Description: free_entry erroneously used groups_free instead of put_group_info
998 Severity : enhancement
1000 Description: Make read-ahead stripe size aligned.
1002 Severity : enhancement
1004 Description: MDS create should not wait for statfs RPC while holding DLM lock.
1007 Frequency : rare, connect and disconnect target at same time
1009 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
1010 Details : don't call obd_disconnect under lov_lock. this long time
1011 operation and can block ptlrpcd which answer to connect request.
1014 Frequency : start MDS on uncleanly shutdowned MDS device
1016 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
1017 Details : stay in waiting mds<>ost recovery finished produce random bugs
1018 due race between two ll_sync thread for one lov target. send
1019 ACTIVATE event only if connect realy finished and import have
1023 Frequency : start MDS on uncleanly shutdowned MDS device
1025 Descriptoin: aborting recovery hang on MDS
1026 Details : don't throttle destroy RPCs for the MDT.
1030 Description: Slow reads beyond 8Tb offsets.
1031 Details : Page index integer overflow in ll_read_ahead_page
1035 Description: Soft lockup on OSS after MDS failover
1036 Details : MSG_CONNECT_INITIAL is not set on the initial MDS->OST connect.
1037 As a conseqence, the patch from bug 18224 is not operational
1038 and the MDS export cannot be reused on the OSTs until it gets
1042 Frequency : rare, only if using MMP with Linux RAID
1044 Description: MMP doesn't work with Linux RAID
1045 Details : While using HA for Lustre servers with Linux RAID, it is possible
1046 that MMP will not detect multiple mounts. To make this work we
1047 need to unplug the device queue in RAID when the MMP block is being
1048 written. Also while reading the MMP block, we should read it from
1049 disk and not the cached one.
1052 Frequency : rare, during recovery
1054 Description: Assertion failure in ldlm_lock_put
1055 Details : Do not put cancelled locks into replay list, hold references on
1056 locks in replay list
1060 Description: 1.6.5 mdsrate performance is slower than 1.4.11/12 (MDS is not cpu bound!)
1061 Details : create_count always drops to the min value (=32) because grow_count
1062 is being changed before the precreate RPC completes.
1065 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
1068 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
1069 Details : add proper locking for creation of jbd_4k slab cache
1073 Description: MMP check in ext3_remount() fails without displaying any error
1074 Details : When multiple mount protection fails during remount, proper error
1079 Description: Rare Client crash on resend if the file was deleted.
1080 Details : When file is opened, but open reply is lost and file is
1081 subsequently deleted before resend, resend processing logic
1082 breaks trying to open the file again, should not try to open.
1086 Description: add check for >8TB ldiskfs filesystems
1087 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
1088 Don't allow >8TB ldiskfs filesystems to be mounted without
1089 force_over_8tb mount option
1093 Description: Client locked up when running multiple instances of an app. on
1094 multiple mount points
1095 Details : ll_shrink_cache() can sleep while holding the ll_sb_lock.
1096 Convert ll_sb_lock to a read/write semaphore to fix the problem.
1100 Description: Cannot acces an NFS-mounted Lustre filesystem
1101 Details : An NFS client cannot access the Lustre filesystem NFS-mounted
1102 from a Lustre-client exporting the Lustre filesystem via NFS.
1106 Description: panic in ll_statahead_thread
1107 Details : grab dentry reference in parent process.
1110 -------------------------------------------------------------------------------
1112 tbd Sun Microsystems, Inc.
1114 * Support for kernels:
1115 2.6.16.60-0.31 (SLES 10),
1116 2.6.18-128.1.6.el5 (RHEL 5),
1117 2.6.22.14 vanilla (kernel.org)
1118 * Client support for unpatched kernels:
1119 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1120 2.6.16 - 2.6.22 vanilla (kernel.org)
1121 * Recommended e2fsprogs version: 1.40.11-sun1
1122 * File join has been disabled, refer to Bugzilla 16929.
1123 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
1124 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
1125 Lustre filesystem with 4K stack may cause a stack overflow. For more
1126 information, please refer to bugzilla 17630.
1128 Severity : enhancement
1130 Description: Update to RHEL5.3 kernel-2.6.18-128.1.6.el5.
1132 Severity : enhancement
1134 Description: Update OFED release to 1.4.1 RC4
1136 Severity : major, only with big OST
1138 Description: Very poor metadata performance on Infiniband lustre configuration
1139 Details : OST object precreation becomes very slow on big OSTs. This is due
1140 to the ialloc patch spending too much time scanning groups.
1143 Frequency : during recovery
1145 Description: don't mix llog inodes with normal.
1146 Details : allocate inodes for log in last inode group
1151 Description: fix lqs' reference which won't be put in some situations
1152 Details : This patch fixes:
1153 1. In quota_check_common(), this function will check quota
1154 for user and group, but only send one return via "pending".
1155 In most cases, the pendings should be same. But that is not
1157 2. If quotaoff runs between lquota_chkquota() and
1158 lquota_pending_commit(), the same thing will happen too.
1159 That is why it comes:
1160 - if (!ll_sb_any_quota_active(qctxt->lqc_sb))
1163 -------------------------------------------------------------------------------
1165 2008-12-31 Sun Microsystems, Inc.
1167 * Support for kernels:
1168 2.6.5-7.314 (SLES 9),
1169 2.6.9-67.0.22.EL (RHEL 4),
1170 2.6.16.60-0.31 (SLES 10),
1171 2.6.18-92.1.17.el5 (RHEL 5),
1172 2.6.22.14 vanilla (kernel.org)
1173 * Client support for unpatched kernels:
1174 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1175 2.6.16 - 2.6.22 vanilla (kernel.org)
1176 * Client support for unpatched kernels:
1177 we do not recommend using patchless RHEL4 clients with kernels
1178 prior to 2.6.9-55EL (RHEL4U5).
1179 * Recommended e2fsprogs version: 1.40.11-sun1
1180 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1181 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
1182 removed cwd "./" (refer to Bugzilla 14399).
1183 * A new quota file format has been introduced in 1.6.5.
1184 The format conversion from prior releases is handled transparently,
1185 but releases older than 1.4.12/1.6.5 don't understand this new
1186 format. The automatic format conversion can be avoided by running
1187 the following command on the MDS:
1188 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
1189 For more information, please refer to bugzilla 13904.
1190 * A new quota file format was introduced in 1.6.6/1.8.0 (kernels 2.6.16+).
1191 The format conversion from prior releases is handled transparently,
1192 but releases older than 1.6.6/1.8.0 don't understand this new
1193 format. The automatic format conversion can be avoided by running
1194 the following commands on the MDS and OSS servers (for
1195 pre 1.4.12-1.6.5 quota files):
1196 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV',
1197 'tunefs.lustre --param="ost.quota_type=ug1" $MDTDEV'
1198 or (for 1.4.12/1.6.5 quota files)
1199 'tunefs.lustre --param="mdt.quota_type=ug2" $MDTDEV',
1200 'tunefs.lustre --param="ost.quota_type=ug2" $MDTDEV'
1201 For more information, please refer to bugzilla 13904.
1202 * Output of lfs quota has been made less detailed by default,
1203 old (verbose) output can be obtained by using -v option.
1204 * File join has been disabled in this release, refer to Bugzilla 16929.
1205 * A new Lustre ADIO driver is available for MPICH2-1.0.7.
1206 * NFS export disabled when stack size < 8192. Since the NFSv4 export of
1207 Lustre filesystem with 4K stack may cause a stack overflow. For more
1208 information, please refer to bugzilla 17630.
1210 Severity : enhancement
1212 Description: Caching OSS
1213 Details : introduce data caching on the OSS. The OSS now relies on the linux
1214 kernel page cache to keep recently accessed data in memory.
1215 It is worth noting that all write requests are still flushed
1216 synchronously as in lustre 1.6.
1218 Severity : enhancement
1220 Description: Version based recovery
1221 Details : introduce finer grained recovery able to detect transaction
1222 dependencies and can deal with transaction gaps caused by clients
1223 failing at the same time as the server.
1225 Severity : enhancement
1227 Description: Enable adaptive timeouts by default
1228 Details : The Lustre timeout value in /proc/sys/lustre/timeout is now
1229 managed dynamically based on server load and should not need
1230 to be tuned manually based on cluster size. This allows Lustre
1231 to work under a wider variety of system sizes and loads, without
1232 unnecessarily causing lengthy recovery times.
1234 Severity : enhancement
1236 Description: Add OST Pools support
1237 Details : File striping can now be set to use an arbitrary pool of OSTs
1239 Severity : enhancement
1241 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
1242 Details : allow skip disconnected ost for send statfs request and hide error
1246 Frequency : rare, on llog test 6
1248 Descriptoin: don't allow connect to already connected import
1249 Details : allowing connect to already connected import is hide connecting problem.
1252 Frequency : rare, on failed llog setup
1254 Descriptoin: don't leak obd reference on failed llog setup
1255 Details : for failed llog setup - mgc forget call class_destroy_import for
1256 client import, move destroy import to more generic place.
1261 Descriptoin: allow kill process which wait statahead result
1262 Details : for some reasons 'ls' can stick in waiting result from statahead,
1263 in this case need way for kill this process.
1266 Frequency : rare, at shutdown
1268 Descriptoin: panic at umount
1269 Details : llap_shrinker can be raced with killing super block from list and
1270 this produce panic with access to already freeded pointer
1275 Descriptoin: don't lose wakeup for imp_recovery_waitq
1276 Details : recover_import_no_retry or invalidate_import and import_close can
1277 both sleep on imp_recovery_waitq, but we was send only one wakeup
1283 Descriptoin: panic in mds_open
1284 Details : don't confuse mds_finish_transno() with PTR_ERR(-ENOENT)
1289 Descriptoin: stuck in cache_remove_extent() or panic with accessing to already
1291 Details : release lock refernce only after add page to pages list.
1294 Frequency : always with long access acl
1296 Descriptoin: mds can't pack reply with long acl.
1297 Details : mds don't control size of acl but they limited by reint/getattr
1302 Frequency : on remount
1304 Description: external journal device not working after the remount
1305 Details : clear dev_rdonly flag for external journal devices in
1311 Description: shutdown vs evict race
1312 Details : client_disconnect_export vs connect request race.
1313 if client will evicted at this time - we start invalidate
1314 thread without referece to import and import can be freed
1320 Description: shrink LOV EAs before replying
1321 Details : correctly adjust LOV EA buffer for reply.
1326 Description: don't skip ost target if they assigned to file
1327 Details : Drop slow OSCs if we can, but not for requested start idx.
1328 This means "if OSC is slow and it is not the requested
1329 start OST, then it can be skipped, otherwise skip it only
1330 if it is inactive/recovering/out-of-space.
1332 Severity : enhancement
1334 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
1337 Frequency : rare, need acl's on inode.
1339 Description: client can't handle ost additional correctly
1340 Details : if ost was added after client connected to mds client can have
1341 hit lnet_try_match_md ... to big messages to wide striped files.
1342 in this case need teach client to handle config events about add
1343 lov target and update client max ea size at that event.
1345 Severity : enhancement
1347 Description: Update to sles9 kernel-2.6.5-7.314.
1349 Severity : enhancement
1351 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
1354 Frequency : Create a symlink file with a very long name
1356 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
1357 Details : If there is no extra space in the request for early cancels,
1358 ldlm_req_handles_avail() returns 0 instead of a negative value.
1363 Description: mds is deadlocked
1364 Details : in rare cases, inode in catalog can have i_no less than have parent
1365 i_no, this produce wrong order for locking during open, and
1366 parallel unlink can be lock open. this need teach mds_open to grab
1367 locks in resource id order, not at parent -> child order.
1369 Severity : enhancement
1371 Description: Add /proc entry for import status
1372 Details : The mdc, osc, and mgc import directories now have
1373 an import directory that contains useful import data for debugging
1374 connection problems.
1376 Severity : enhancement
1378 Description: Re-disable certain /proc logging
1379 Details : Enable and disable client's offset_stats, extents_stats and
1380 extents_stats_per_process stats logging on the fly.
1383 Frequency : Only on FC kernels 2.6.22+
1385 Description: oops in statahead
1386 Details : Do not drop reference count for the dentry from VFS when lookup,
1387 VFS will do that by itself.
1389 Severity : enhancement
1391 Description: Generic /proc file permissions
1392 Details : Set /Proc file permissions in a more generic way to enable non-
1393 root users operate on some /proc files.
1397 Description: Hitting mdc_commit_close() ASSERTION
1398 Details : Properly handle request reference release in
1399 ll_release_openhandle().
1403 Frequency : only patchless client
1404 Description: add workaround for race between add/remove dentry from hash
1406 Severity : enhancement
1408 Description: Allow OST glimpses to return PW locks
1412 Description: LBUG when llog conf file is full
1413 Details : When llog bitmap is full, ENOSPC should be returned for plain
1418 Description: Prevent import from entering FULL state when server in recovery
1422 Description: service mount cannot take device name with ":"
1423 Details : Only when device name contains ":/" will mount treat it as
1429 Description: replace ptlrpcd with the statahead thread to interpret the async
1430 statahead RPC callback
1434 Frequency : on recovery
1435 Description: I/O failures after umount during fail back
1436 Details : if client reconnected to restarted server we need join to recovery
1437 instead of find server handler is changed and process self eviction
1438 with cancel all locks.
1440 Severity : enhancement
1442 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
1446 Description: Kernel BUG tries to release flock
1447 Details : Lustre does not destroy flock lock before last reference goes
1448 away. So always drop flock locks when client is evicted and
1449 perform unlock regardless of successfulness of speaking to MDS.
1451 Severity : enhancement
1453 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
1455 Severity : enhancement
1457 Description: Upcall on Lustre log has been dumped
1458 Details : Allow for a user mode script to be called once a Lustre log has
1459 been dumped. It passes the filename of the dumped log to the
1460 script, the location of the script can be specified via
1461 /proc/sys/lnet/debug_log_upcall.
1466 Description: avoid messages about idr_remove called for id that is not allocated
1467 Details : Move assigment s_dev for clustered nfs to end of initialization,
1468 for avoid problem with error handling.
1473 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
1474 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
1475 CONN_USED_HASH and this prodice warning when put connection again
1481 Description: avoid ASSERTION(client_stat->nid_exp_ref_count == 0) failed
1482 Details : release reference to stats when client disconnected, not
1483 when export destroyed for avoid races when client destroyed
1484 after main ost export.
1488 Description: more cleanup in mds_lov
1489 Details : add workaround for get valid ost count for avoid warnings about
1490 drop too big messages, not init llog cat under semphore which
1491 can be blocked on reconnect and break normal replay, fix access
1494 Severity : enhancement
1496 Description: Export bytes_read/bytes_write count on OSC/OST.
1500 Description: Early reply size mismatch, MGC loses connection
1501 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
1502 the connect flags are properly negotiated.
1506 Description: Properly propagate oinfo flags from lov to osc for statfs
1507 Details : restore missing copy oi_flags to lov requests.
1511 Description: exports in /proc are broken
1512 Details : recreate /proc entries for clients when they reconnect.
1514 Severity : enhancement
1516 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
1517 : l_getgroups(8), lst(8), routerstat(8)
1518 Details : included man pages for llobdstat(8), llstat(8),
1519 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
1521 Severity : enhancement
1523 Description: Implement lustre ll_show_options method.
1525 Severity : enhancement
1527 Description: Update to SLES9 kernel-2.6.5-7.312.
1529 Severity : enhancement
1531 Description: Update to RHEL4 kernel-2.6.9-67.0.22.EL.
1535 Description: exports in /proc are broken
1536 Details : recreate /proc entries for clients when they reconnect.
1540 Description: don't fail open with -ERANGE
1541 Details : if client connected until mds will be know about real ost count
1542 get LOV EA can be fail because mds not allocate enougth buffer
1547 Description: Resolve device initialization race
1548 Details : Prevent proc handler from accessing devices added to the
1549 obd_devs array but yet be intialized.
1551 Severity : enhancement
1553 Description: configure's --enable-quota should check the
1554 : kernel .config for CONFIG_QUOTA
1555 Details : configure is terminated if --enable-quota is passed but
1556 : no quota support is in kernel
1558 Severity : enhancement
1560 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
1562 Severity : enhancement
1564 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
1568 Frequency : rare, on PPC clients
1569 Description: don't swab ost objects in response about directory, because
1571 Details : bug similar bug 14856, but in different function.
1573 Severity : enhancement
1575 Description: lfs quota tool enhancement
1576 Details : added units specifiers support for setquota, default to
1577 current uid/gid for quota report, short quota stats by
1578 default, nonpositional parameters for setquota, added
1579 llapi_quotactl manual page.
1581 Severity : enhancement
1583 Description: *optional* service tags registration
1584 Details : if the "service tags" package is installed on a Lustre node
1585 When the filesystem is mounted, a local-node service tag will
1586 be created. See http://inventory.sun.com/ for more information
1587 about the Service Tags asset management system.
1591 Description: Client runs out of low memory
1592 Details : Consider only lowmem when counting initial number of llap pages
1595 Frequency : occasional
1597 Description: add refcount for osc callbacks, so avoid panic on shutdown
1599 Severity : enhancement
1601 Description: Update to RHEL4 kernel-2.6.9-67.0.20.
1604 Frequency : testing only
1606 Description: sanity test 65a fails if stripecount of -1 is set
1607 Details : handle -1 striping on filesystem in ll_dirstripe_verify
1610 Frequency : only in unusual configurations
1612 Description: Kernel panic with find ost index.
1613 Details : lov_obd have panic if some OST's have sparse indexes.
1615 Severity : enhancement
1617 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
1620 Frequency : rarely, if filesystem is mounted with -o flock
1622 Description: do not process already freed flock
1623 Details : flock can possibly be freed by another thread before it reaches
1624 to ldlm_flock_completion_ast.
1627 Frequency : rarely, if filesystem is mounted with -o flock
1629 Description: LBUG during stress test
1630 Details : Need properly lock accesses the flock deadlock detection list.
1633 Frequency : rarely, if binaries are being run from Lustre
1635 Description: oops in page fault handler
1636 Details : kernel page fault handler can return two special 'pages' in
1637 error case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
1640 Frequency : rarely, during shutdown
1642 Description: timeout with invalidate import.
1643 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
1644 should be handled by ptlrpcd. This produce long age waiting and
1645 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
1651 Description: ASSERTION(CheckWriteback(page,cmd)) failed
1652 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
1656 Frequency : only with broken builds/installations
1658 Description: no LBUG if lquota.ko and fsfilt_ldiskfs.ko are different versions
1659 Details : just return an error to a user, put a console error message
1661 Severity : enhancement
1663 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
1665 Severity : enhancement
1667 Description: Update to RHEL4 kernel-2.6.9-67.0.15.
1669 Severity : enhancement
1671 Description: enable MGS and MDT services start separately
1672 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
1673 a co-located MGS without starting the MGS, which is a complement
1674 to 'nosvc' mount option.
1677 Frequency : always, on big-endian systems
1679 Description: cleanup in ptlrpc code, related to PPC platform
1680 Details : store magic in native order avoid panic's in recovery on PPC
1681 node and forbid from this error in future. Also fix posibily
1682 of twice swab data. Fix get lov striping to userpace.
1685 Frequency : rarely, if replay get lost on server
1687 Description: server incorrectly drop resent replays lead to recovery failure.
1688 Details : do not drop replay according to msg flags, instead we check the
1689 per-export recovery request queue for duplication of transno.
1692 Frequency : after recovery
1694 Description: precreate to many object's after del orphan.
1695 Details : del orphan st in oscc last_id == next_id and this triger growing
1696 count of precreated objects. Set flag LOW to skip increase count
1697 of precreated objects.
1700 Frequency : rare, on clear nid stats
1702 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
1703 Details : when clean nid stats sometimes try destroy live entry,
1704 and this produce panic in free.
1707 Frequency : occasionally since 1.6.4
1709 Description: Stack overflow during MDS log replay
1710 Details : ease stack pressure by using a thread dealing llog_process.
1713 Frequency : very rare
1715 Description: MDT cannot be unmounted, reporting "Mount still busy"
1716 Details : Mountpoint references were being leaked during open reply
1717 reconstruction after an MDS restart. Drop mountpoint reference
1718 in reconstruct_open() and free dentry reference also.
1723 Description: wait until IO finished before start new when do lock cancel.
1724 Details : VM protocol want old IO finished before start new, in this case
1725 need wait until PG_writeback is cleared until check dirty flag
1726 and call writepages in lock cancel callback.
1731 Description: mds_mfd_close() ASSERTION(rc == 0)
1732 Details : In mds_mfd_close(), we need protect inode's writecount change
1733 within its orphan write semaphore to prevent possible races.
1736 Frequency : rare, on shutdown ost
1738 Description: don't hit live lock with umount ost.
1739 Details : shrink_dcache_parent can be in long loop with destroy dentries,
1740 use shrink_dcache_sb instead.
1743 Frequency : only when echo_client is used
1745 Description: don't panic with use echo_client
1746 Details : echo client pass NULL as client nid pointer and this produce NULL
1747 pointer dereference.
1750 Frequency : Always on 32-bit PowerPC systems
1752 Description: fix build on PPC32
1753 Details : compile code with -m64 flag produce wrong object file for PPC32.
1758 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1759 Details : In reconstruct_* functions, LASSERTs on both the data supplied
1760 by a client, and the data on disk are dangerous and incorrect.
1761 Change them with client eviction.
1763 Severity : enhancement
1765 Description: skiplist implementation simplification
1766 Details : skiplists are used to group compatible locks on granted list
1767 that was implemented as tracking first and last lock of each lock
1768 group the patch changes that to using doubly linked lists
1772 Description: delete compatibility for 32bit qdata
1773 Details : as planned, when lustre is beyond b1_8, lquota won't support 32bit
1774 qunit. That means servers of b1_4 and servers of b1_8 can't be
1775 used together if users want to use quota.
1778 Frequency : only with administrator action
1780 Description: mount failure if config log has invalid conf_param setting
1781 Details : If administrator specified an incorrect configuration parameter
1782 with "lctl conf_param" this would cause an error during future
1783 client mounts. Instead, ignore the bad configuration parameter.
1786 Frequency : blocks per group < blocksize*8 and uninit_groups is enabled
1788 Description: ldiskfs error: XXX blocks in bitmap, YYY in gd
1789 Details : If blocks per group is less than blocksize*8, set rest of the
1793 Frequency : Application do stride read on lustre
1795 Description: The read performance will drop a lot if the application does
1797 Details : Because the stride_start_offset are missing in stride read-ahead,
1798 it will cause clients read a lot of unused pages in read-ahead,
1799 then the read-performance drops.
1803 Description: more ldlm soft lockups
1804 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1805 starve other threads from the resource lock for a long time in
1806 case of long waiting queue, so change the debug level from
1807 D_OTHER to the less frequently used D_INFO.
1809 Severity : enhancement
1811 Description: add -gid, -group, -uid, -user options to lfs find
1813 Severity : enhancement
1815 Description: ll_recover_lost_found_objs - recover objects in lost+found
1816 Details : OST corruption and subsequent e2fsck can leave objects in the
1817 lost+found directory. Using the "ll_recover_lost_found_objs"
1818 tool, these objects can be retrieved and data can be salvaged
1819 by using the object ID saved in the fid EA on each object.
1824 Description: this bug _only_ happens when inode quota limitation is very low
1825 (less than 12), so that inode quota unit is 1 at initialization.
1826 Details : if remaining quota equates 1, it is a sign to demonstate that quota
1827 is effective now. So least quota qunit should be 2.
1831 Description: Hung threads in invalidate_inode_pages2_range
1832 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1833 one is completed. As a result, some RPCs are stuck in the queue
1838 Description: Procfs and llog threads access destoryed import sometimes.
1839 Details : Sync the import destoryed process with procfs and llog threads by
1840 the import refcount and semaphore.
1844 Description: mds fails to respond, threads stuck in ldlm_completion_ast
1845 Details : Sort source/child resource pair after updating child resource.
1850 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1851 Details : If insertion of an extent fails, then discard the inode
1852 preallocation and free data blocks else it can lead to duplicate
1857 Description: don't always update ctime in ext3_xattr_set_handle()
1858 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle()
1859 In some cases the ctime should not be updated, for example for
1860 2.0->1.8 compatibility it is necessary to delete an xattr and it
1861 should not update the ctime.
1865 Description: add quota statistics
1866 Details : 1. sort out quota proc entries and proc code.
1867 2. add quota statistics
1872 Description: quotas are not honored with O_DIRECT
1873 Details : all writes with the flag O_DIRECT will use grants which leads to
1874 this problem. Now using OBD_BRW_SYNC to guard this.
1878 Bugzilla : 15713/16362
1879 Description: Assertion in iopen_connect_dentry in 1.6.3
1880 Details : looking up an inode via iopen with the wrong generation number can
1881 populate the dcache with a disconneced dentry while the inode
1882 number is in the process of being reallocated. This causes an
1883 assertion failure in iopen since the inode's dentry list contains
1884 both a connected and disconnected dentry.
1888 Description: assertion failure in ldlm_handle2lock()
1889 Details : fix a race between class_handle_unhash() and class_handle2object()
1890 introduced in lustre 1.6.5 by bug 13622.
1892 Severity : enhancement
1894 Description: superblock lock contention with many SMP cores on one client
1895 Details : several client filesystem locks were highly contended on SMP
1896 NUMA systems with 8 or more cores. Per-CPU datastructures
1897 and more efficient locking implemented to reduce contention.
1902 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1903 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1908 Description: oops in sd_iostats_seq_show()
1909 Details : unloading/reloading the scsi low level driver triggers a kernel
1910 bug when trying to access the sd iostat file.
1915 Description: Kernel panics during QLogic driver reload
1916 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1917 patch, causing memory corruption.
1922 Description: journal_dev option does not work in b1_6
1923 Details : pass mount option during pre-mount.
1925 Severity : enhancement
1927 Description: Add a FIEMAP(FIle Extent MAP) ioctl for ldiskfs
1928 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1929 extent information of a file. It can be used to map logical blocks
1930 in a file to physical blocks in the block device.
1933 Frequency : only with adaptive timeout enabled
1935 Description: DEBUG_REQ() bad paging request
1936 Details : ptlrpc_at_recv_early_reply() should not modify req->rq_repmsg
1937 because it can be accessed by reply_in_callback() without the
1941 Frequency : only on Cray X2
1943 Description: X2 build failures
1944 Details : fix build failures on Cray X2.
1948 Description: xid & resent requests
1949 Details : Initialize RPC XID from clock at startup (randomly if clock is
1954 Description: quota recovery deadlock during mds failover
1955 Details : This patch includes att18982, att18236, att18237 in bz14840.
1957 1. fix osts hang when mds does failover with quotaon
1958 2. prevent watchdog storm when osts threads wait for the
1963 Description: kernel panic on racer
1964 Details : Do not access dchild->d_inode when IS_ERR(dchild) is true.
1966 Severity : enhancement
1968 Description: Add lustre_start utility to start or stop multiple Lustre servers
1973 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1974 Details : In case of memory pressure, list_del() can be called twice on
1975 req->rq_history_list, causing a kernel oops.
1979 Description: (ptllnd_peer.c:557:kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1980 Details : fix stack overflow in the distributed lock manager by defering
1981 export eviction after a failed ast to the elt thread instead of
1982 handling it in the dlm interpret routine.
1984 Severity : enhancement
1986 Description: More exported tunables for mballoc
1987 Details : Add support for tunable preallocation window and new tunables for
1988 large/small requests
1992 Description: Detect corruption of block bitmap and checking for preallocations
1993 Details : Checks validity of on-disk block bitmap. Also it does better
1994 checking of number of applied preallocations. When corruption is
1995 found, it turns filesystem readonly to prevent further corruptions.
1999 Frequency : only for big-endian servers
2000 Description: Check for big-endian system while mounting fs with extents feature
2001 Details : Mounting a filesystem with extents feature will fail on big-endian
2002 systems since ext3-based ldiskfs is not supported on big-endian
2003 systems. Can be over-ridden with "bigendian_extents" mount option.
2007 Description: Excessive recovery window
2008 Details : With AT enabled, the recovery window can be excessively long (6000+
2009 seconds). To address this problem, we no longer use
2010 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2011 timeout no longer depends on the service time, it is set to
2012 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2013 time via pb_service_time.
2016 Descriptoin: Don't sync journal after every i/o
2017 Details : Implement write RPC replay to allow server replies for write RPCs
2018 before data is on disk. However, this feature is disabled by
2019 default since some issues leading to data corruptions have been
2020 found during recovery (e.g. bug 19128). This feature can be enabled
2021 by running the following command on the OSSs:
2022 lctl set_param obdfilter.*.sync_journal=0
2026 Description: Watchdog triggered on MDS failover
2027 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2028 the OSTs know that the MDS "UUID" can be reused for the same export
2029 from a different NID, so we do not need to wait for the export to
2034 Descriptoin: Lustre detected file system corruption with inode out of bounds
2035 Details : don't update i_size on MDS_CLOSE for directories. This causes
2036 directory corruptions on the MDT.
2038 -------------------------------------------------------------------------------
2040 2008-05-26 Sun Microsystems, Inc.
2042 * Support for kernels:
2043 2.6.5-7.311 (SLES 9),
2044 2.6.9-67.0.7.EL (RHEL 4),
2045 2.6.16.54-0.2.5 (SLES 10),
2046 2.6.18-53.1.14.el5 (RHEL 5),
2047 2.6.22.14 vanilla (kernel.org)
2048 * Client support for unpatched kernels:
2049 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2050 2.6.16 - 2.6.22 vanilla (kernel.org)
2051 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
2052 we do not recommend using patchless RHEL4 clients with kernels
2053 prior to 2.6.9-55EL (RHEL4U5).
2054 * Recommended e2fsprogs version: 1.40.7-sun1
2055 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2056 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
2057 removed cwd "./" (refer to Bugzilla 14399).
2058 * A new quota file format has been introduced in 1.6.5.
2059 The format conversion from prior releases is handled transparently,
2060 but releases older than 1.4.12/1.6.5 will not understand this new
2061 format. The automatic format conversion can be avoided by running
2062 the following command on the MDS before upgrading:
2063 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
2064 For more information, please refer to bugzilla 13904.
2068 Description: quota performance fix
2069 Details : quota data is written in journalled mode instead of ordered to
2070 increase performance
2074 Description: lfs support for human-readable quota grace time strings
2075 Details : lfs setquota -t and lfs quota -t represent quota grace times
2076 in "XXwXXdXXhXXmXXs" format instead of large values in seconds
2079 Frequency : always with o2ib 1.3 and sles10
2081 Description: fix build with SLES10 and o2ib v3.
2082 Details : sles10 uses diffrent name for Module.symver file but configure
2083 assume this file has same name on RHEL/SLES/vanila kernels.
2086 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
2088 Description: MDS may lose file striping (and hence file data) in some cases
2089 Details : If there are additional extended attributes stored on the MDS,
2090 in particular ACLs, SELinux, or user attributes (if user_xattr
2091 is specified for the client mount options) then there is a risk
2092 of attribute loss. Additionally, the Lustre file striping
2093 needs to be larger than default (e.g. striped over all OSTs),
2094 and an additional attribute must be stored initially in the
2095 inode and then increase in size enough to be moved to the
2096 external attribute block (e.g. ACL growing in size) for file
2099 Severity : enhancement
2101 Description: add message levels for liblustreapi
2106 Description: MDT cannot be unmounted, reporting "Mount still busy"
2107 Details : Mountpoint references were being leaked during open reply
2108 reconstruction after an MDS restart. Drop mountpoint reference
2109 in reconstruct_open() and free dentry reference also.
2114 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
2115 Details : Move the 'good_osts' check before the 'total_bavail' check. This
2116 will result in an -EAGAIN and in the exit call path we call
2117 alloc_rr() which will with increasing aggressiveness attempt to
2118 aquire precreated objects on the minimum number of required OSCs.
2122 Description: Use old size assignment to avoid deadlock
2123 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
2124 the scheduling while holding a spinlock. We do not need locking
2125 for size in ll_update_inode() because size is only updated from
2126 the MDS for directories or files without objects, so there is no
2127 other place to do the update, and concurrent access to such inodes
2128 are protected by the inode lock.
2132 Description: Use __u64 instead of int for valid bits
2136 Description: resolve "_IOWR redefined" build error on SLES10
2140 Description: dump the memory debugging after all modules are unloaded to
2141 suppress false negative in conf_sanity test 39
2145 Description: the recovery timer never expires
2146 Details : for new client connect request, the recovery timer should not be
2147 reset, otherwise recovery timer will never expired, if the old
2148 client never come. Only old client connect and first connection
2149 req should trigger recovery timer reset.
2153 Description: the min numbers of lproc stats are wrong
2154 Details : adding a new constant LC_MIN_INIT and use it for initialization
2158 Frequency : always with interactive lfs
2160 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
2163 Frequency : with multiple concurrent readdir processes in same directory
2164 Bugzilla : 15406, 15169, 15175
2165 Description: misc fixes for directory readahead.
2166 Details : prevent previous statahead async RPC callback from processing the
2167 current "statahead_info", race condition between async RPC callback
2168 add dentry into dentry hash table and "ls" thread revalidate such
2169 dentry, statahead his/miss control for hidden items, and so on.
2171 Severity : enhancement
2173 Description: build kernel-ib packages for OFED 1.3 in our release cycle
2177 Description: incore types cleaning in quota code (with respect to 64-bit limits)
2178 Details : several u32 variables declarations are replaced with u64 declarations
2183 Description: fix SLES kernel versioning
2184 Details : the kernel version for our SLES 10 kernel did not include a "-"
2185 before the "smp" at the end. while this was not a problem in
2186 general, it did mean that software trying to use the kernel
2187 version to try to detect a vendor specific kernel would fail.
2188 this was most evident by the OFED build scripts.
2193 Description: Don't update lov_desc members until making sure they are valid
2194 Details : When updating lov_desc members via proc fs, need fix their
2195 validities before doing the real update.
2198 Frequency : very rare
2200 Description: don't put request into delay list while invalidate in flight.
2201 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
2202 import in flight. this produce timeout for invalidate and sometimes
2203 can cause stale data.
2205 Severity : enhancement
2207 Description: Update kernel to SLES9 2.6.5-7.311.
2209 Severity : enhancement
2211 Description: Update kernel to RHEL4 2.6.9-67.0.7.
2216 Frequency : on PPC only
2217 Description: not convert ost objects for directory because it's not exist.
2218 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
2220 Severity : enhancement
2222 Description: Fix warnings with compile liblustre at sles10/rhel5 which have
2223 __u64 as usingied long long type.
2226 Frequency : rare, on shutdown
2228 Description: race process ast vs remove callback
2229 Details : removing callback before disconnect import open race with
2230 processing callback.
2232 Severity : enhancement
2234 Description: Update kernel to SLES9 2.6.5-7.311.
2236 Severity : enhancement
2238 Description: Files open for execute are not marked busy on SLES10
2239 Details : Add FMODE_EXEC to SLES10 SP1 server kernel series.
2241 Severity : enhancement
2243 Description: Add server support for vanilla-2.6.22.14.
2246 Frequency : occasional
2248 Description: Avoid lov_create() getting stuck in obd_statfs_rqset()
2249 Details : If an OST is down the MDS will hang indefinitely in
2250 obd_statfs_rqset() waiting for the statfs data. While for
2251 MDS QOS usage of statfs, it should not stuck in waiting.
2253 Severity : enhancement
2255 Description: Disable adaptive timeouts by default
2258 Frequency : on network error
2260 Description: panic with double free request if network error
2261 Details : mdc_finish_enqueue is finish request if any network error occuring,
2262 but it's true only for synchronus enqueue, for async enqueue
2263 (via ptlrpcd) this incorrect and ptlrpcd want finish request
2267 Frequency : rare, on recovery
2269 Description: read procfs can produce deadlock in some situation
2270 Details : Holding lprocfs lock with send rpc can produce block for destroy
2271 obd objects and this also block reconnect with -EALREADY.
2272 This isn't fix all lprocfs bugs - but make it rare.
2274 Severity : enhancement
2276 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
2279 Frequency : frequent on X2 node
2281 Description: mdc_set_open_replay_data LBUG
2282 Details : Set replay data for requests that are eligible for replay.
2287 Description: lustre_mgs: operation 101 on unconnected MGS
2288 Details : When MGC is disconnected from MGS long enough, MGS will evict the
2289 MGC, and late on MGC cannot successfully connect to MGS and a lot
2290 of the error messages complaining that MGS is not connected.
2293 Frequency : rare, depends on device drivers and load
2295 Description: MDS or OSS nodes crash due to stack overflow
2296 Details : Code changes in 1.6.4 increased the stack usage of some functions.
2297 In some cases, in conjunction with device drivers that use a lot
2298 of stack, the MDS (or possibly OSS) service threads could overflow
2299 the stack. One change which was identified to consume additional
2300 stack has been reworked to avoid the extra stack usage.
2302 Severity : enhancement
2304 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
2306 Severity : enhancement
2308 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
2310 Severity : enhancement
2312 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
2314 Serverity : enhancement
2316 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
2319 Frequency : rare on shutdown OST
2321 Description: Don't allow skipping OSTs if index has been specified.
2322 Details : Don't allow skipping OSTs if index has been specified, make locking
2323 in internal create lots better.
2328 Description: ASSERTION(!PageDirty(page)) failed
2329 Details : Wrong check could lead to an assertion failure under specific
2335 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
2336 Details : access to bitfield in structure is always rounded to long
2337 and this produce problem with not atomic change any bit.
2342 Description: Lustre make rpms failed.
2343 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
2344 builds Lustre rpms from tarball.
2346 Severity : enhancement
2348 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
2351 Frequency : rare on shutdown OST
2353 Description: If llog cancel was not send before clean_exports phase, this can
2354 produce deadlock in llog code.
2355 Details : If llog thread has last reference to obd and call class_import_put
2356 this produce deadlock because llog_cleanup_commit_master wait when
2357 last llog_commit_thread exited, but this never success because was
2358 called from llog_commit_thread.
2361 Frequency : only if OST index is skipped
2363 Description: NULL lov_tgts causing MDS oops
2364 Details : more safe checks for NULL lov_tgts for avoid oops.
2366 Severity : enhancement
2368 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
2370 Severity : enhancement
2372 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
2377 Description: make mgs_setparam() handle fsname containing dash
2378 Details : fsname containing a dash does not work with lctl conf_param
2380 Severity : enhancement
2382 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
2385 Frequency : rare, in recovery and (or) destroy lovobjid file.
2387 Description: rewrite lov objid code.
2388 Details : Cleanup for lov objid code, remove scability problems and wrong
2389 locking. Fix sending last_id into ost.
2391 Severity : enhancement
2393 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
2395 Severity : enhancement
2397 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
2398 Details : Use d_move_locked instead of __d_move.
2401 Frequency : rare, at shutdown
2403 Description: access already free / zero obd_namespace.
2404 Details : if client_disconnect_export was called without force flag set,
2405 and exist connect request in flight, this can produce access to
2406 NULL pointer (or already free pointer) when connect_interpret
2407 store ocd flags in obd_namespace.
2410 Frequency : only at startup
2412 Description: not alloc memory with spinlock held.
2413 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
2414 if any spinlock held.
2419 Description: lfs find does not continue on file error
2420 Details : Continue other files processing when a file/dir is absent.
2424 Description: Inconsistent usage of lustre_pack_reply()
2425 Details : Standardize the usage of lustre_pack_reply() such that it
2426 always generate a CERROR on failure.
2429 Frequency : very rare
2431 Description: Fix replay if there is an un-replied request and open
2432 Details : In some cases, older replay request will revert the
2433 mcd->mcd_last_xid on MDS which is used to record the client's
2434 latest sent request.
2436 Severity : enhancement
2438 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
2440 Severity : enhancement
2442 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
2444 Severity : enhancement
2446 Description: Update to SLES9 kernel-2.6.5-7.287.3.
2448 Severity : enhancement
2450 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
2452 Severity : enhancement
2454 Description: Build SLES10 patchless client fails
2455 Details : The configure was broken by run ./configure with
2456 --with-linux-obj=.... argument for patchless client. When the
2457 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
2458 can't search header adequately. Use absolute path such as
2459 -I($LINUX)/include instead.
2461 Severity : enhancement
2463 Description: Lustre Page Accounting
2464 Details : New macros for page alloc and free which enable accounting
2465 of page allocation of Lustre. Use percpu counters to store memory
2466 and page statistics.
2469 Frequency : only if debugging is disabled
2471 Description: LASSERT_{REQ,REP}SWAB macros are buggy
2472 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
2473 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
2474 these macros and replace them with their definitions instead.
2479 Description: interrupt oig_wait produce painc on resend.
2480 Details : brw_redo_request can be used for resend requests from ptlrpcd and
2481 private set, and this produce situation when rq_ptlrpcd_data not
2482 copyed to new allocated request and triggered LBUG on assert
2483 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
2484 ptlrpcd set if request is changed and can be safety changed to use
2487 Severity : enhancement
2489 Description: organize the server-side client stats on per-nid basis
2490 Details : Change the structure of stats under obdfilter and mds to
2498 The "uuid"s file would list the uuids of _active_ exports.
2499 And the clear entry is to clear all stats and stale nids.
2504 Description: Processes looping in ll_readdir() on Lustre clients finally causing
2505 a full node pseudo-hang
2506 Details : Concurrent access to the same directory from multiple clients with
2507 intensive file creation/removal can cause a client node to spin in
2508 ll_readdir(). i_version must be increased every time the lock is
2509 cancelled to ensure a revalidate is done.
2514 Description: touch file failed when fs is not full
2515 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
2516 otherwise we can get ENOSP while fs is not full.
2519 Frequency : only for Cray XT3
2520 Bugzilla : 12829/13455
2521 Description: Changing primary group doesn't change the group lustre assigns to
2523 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
2524 overridden with the primary group provided by the group upcall,
2525 whereas the supplied fsgid can be trusted if it is in the list of
2526 supplementary groups returned by the group upcall.
2528 Severity : enhancement
2530 Description: Root Squash Functionality
2531 Details : Implementation of NFS-like root squash capability. Specifically,
2532 don't allow someone with root access on a client node to be able
2533 to manipulate files owned by root on a server node.
2535 Severity : enhancement
2537 Description: Slow trucate/writes to huge files at high offsets.
2538 Details : Directly associate cached pages to lock that protect those pages,
2539 this allows us to quickly find what pages to write and remove
2540 once lock callback is received.
2545 Description: Too many locks accumulating on client during NFS usage
2546 Details : mds_open improperly used accmode to find out access mode to a
2547 file. Also mdc_intent_lock logic to find out if we already have
2548 lock similar to just received was flawed since introduction of
2549 skiplists - locks are now added to the front of the granted
2554 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
2556 Details : In obd setup failure handler, obd_stopping will not necessarily to
2557 be 1, and obd_set_up should also be checked to make sure whether
2558 obd is completely setup.
2560 Severity : enhancement
2562 Description: Allow masking D_WARNING, D_ERROR messages from console
2563 Details : Console messages can now be disabled via lnet.printk.
2568 Description: User code with malformed file open parameter crashes client node
2569 Details : Before packing join_file req, all the related reference should be
2570 checked carefully in case some malformed flags cause fake join_file
2576 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
2577 grain of quota doing harm to quota's accuracy
2578 Details : qunit size will be changed when quota limitation is too low/high;
2579 record the pending quota write in order to get more accureate
2580 quota; delete the patch for bug12588, which is unnecessary when
2581 this patch is landed. This bug also contains fixes for bug 14526,
2582 14299, 14601 and 13794.
2586 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
2587 Details : ldlm_completion_ast() assumes that a lock is granted when the req
2588 mode is equal to the granted mode. However, it should also check
2589 that LDLM_FL_CP_REQD is not set.
2593 Description: Heavy nfs access might result in deadlocks
2594 Details : After ELC code landed, it is now improper to enqueue any mds
2595 locks under och_sem, because enqueue might want to decide to
2596 cancel open locks for same inode we are holding och_sem for.
2600 Description: 35% write performance drop with ldiskfs2 when quotas are on
2601 Details : Enable ext3 journalled quota by default to improve performance
2602 when quotas are turned on.
2606 Description: Client eviction while running blogbench
2607 Details : A lot of unlink operations with concurrent I/O can lead to a
2608 deadlock causing evictions. To address the problem, the number of
2609 oustanding OST_DESTROY requests is now throttled to
2610 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
2611 callbacks are processed in priority.
2614 Frequency : RHEL4 only
2616 Description: mkfs is very slow on IA64/RHEL4
2617 Details : A performance regression has been discovered in the MPT Fusion
2618 driver between versions 3.02.73rh and 3.02.99.00rh. As a
2619 consequence, we have downgraded the MPT Fusion driver in the RHEL4
2620 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
2623 Frequency : PPC/PPC64 only
2625 Description: conflicts between asm-ppc64/types.h and lustre_types.h
2626 Details : fix duplicated definitions between asm-ppc64/types.h and
2627 lustre_types.h on PPC.
2630 Frequency : PPC/PPC64 only
2632 Description: asm-ppc/segment.h does not exist
2633 Details : fix compile issue on PPC.
2637 Description: data checksumming impacts single node performance
2638 Details : add support for several checksum algorithms. Currently, CRC32 and
2639 Adler-32 are supported. The checksum type can be changed on the fly
2640 through /proc/fs/lustre/osc/*/checksum_type.
2644 Description: use adler32 for page checksums
2645 Details : when available, use the Adler-32 algorithm instead of CRC32 for
2650 Description: better handle error messages in extents code
2652 Severity : enhancement
2654 Description: SNMP support enhancement
2655 Details : Adding total number of sampled request for an MDS node in snmp
2658 Severity : enhancement
2660 Description: Optimize ldlm waiting list processing for PR extent locks
2661 Details : When processing waiting list for read extent lock and meeting read
2662 lock that is same or wider to it that is not contended, skip
2663 processing rest of the list and immediatelly return current
2664 status of conflictness, since we are guaranteed there are no
2665 conflicting locks in the rest of the list.
2669 Description: Time out and refuse to reconnect
2670 Details : When the failover node is the primary node, it is possible
2671 to have two identical connections in imp_conn_list. We must
2672 compare not conn's pointers but NIDs, otherwise we
2673 can defeat connection throttling.
2677 Description: Client not clear own cache if answer to reconnect is lost.
2678 Details : Client gets evicted from server. Now client also thinks it is
2679 disconnected (or gets ENOTCONN on its operation) and decides to
2680 reconnect. Server receives reconnect message, but cannot find
2681 export. New export is created that is fully valid (new cookie!),
2682 but reply is lost and not reported to client. Client reconnects
2683 again and gets back a just-created connection, but it is not new
2684 so client thinks it was not evicted and does not do recovery.
2688 Description: Detect stride IO mode in read-ahead
2689 Details : When a client does stride read, read-ahead should detect that and
2690 read-ahead pages according to the detected stride pattern.
2694 Description: build for x2 fails
2695 Details : fix compile issue on Cray systems.
2697 Severity : enhancement
2699 Description: implement readv/writev APIs(aio_read/aio_writes in newer kernels)
2700 Details : This greatly improves speed of NFS writes on 2.6 kernels.
2703 Frequency : only on PPC/SLES10
2705 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
2706 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
2707 be defined. Add a hack in mkfs_lustre.c to work around this compile
2712 Description: LASSERT on MDS when client holding flock lock dies
2713 Details : ldlm pool logic depends on number of granted locks equal to
2714 number of released locks which is not true for flock locks, so
2715 just exclude such locks from consideration.
2719 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
2720 Details : Use fsfilt_sync() for both the whole filesystem sync and
2721 individual file sync to eliminate dangerous inode locking
2722 with I_LOCK that can lead to a deadlock.
2726 Description: Update an obsolete wirecheck.c generator
2727 Details : Update wirecheck.c/wirehdr.c and regenerate wiretest.c
2731 Description: Client can panic on open sometimes
2732 Details : It is possible that we try to free already freed request in
2733 ll_file_open in some error cases when we send request from
2738 Description: performance in 1.6.3
2739 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
2740 sure that 1MB requests are not fragmented by the block layer.
2744 Description: LDLM soft lockups - improvement
2745 Details : It is be possible to send the lock handle along with each read
2746 or write request because the client is already doing a lock match
2747 itself so there isn't any reason the OST should have to re-do that
2753 Description: lfs quota fails with deactivated OSTS
2754 Details : With this patch, three improvements are included:
2755 1. detete the softlimit in mds and osts when use "lfs quota".
2756 2. display the inaccurate data in the output of "lfs quota".
2757 3. try to get quota info when "lfs quota" is executed.
2762 Description: Extent locks not granted with no conflicts sometimes.
2763 Details : When race occurs in glimpse handler and nothing is returned,
2764 we do not reprocess the queue after lock cancel, and that leads
2765 to a stall until next activity on a resource
2768 Frequency : failover with quotaon
2770 Description: during mds failovers with quota on, OSTs got into deadlock state
2771 and causing dumpstack.
2772 Details : for every quota slave, at any time, there is only one quota req
2773 is sent to quota master for every uid/gid. Before that quota req
2774 returns, all the thread relative to the same uid/gid will wait.
2775 So if the quota req is lost because mds failovers or any other
2776 reasons, this bug will be hit. Now, dqacq_interpret() will handle
2777 quota reqs who time out.
2779 Severity : enhancement
2782 Description: when quota slave checks if quota is enough, there is an unnecessary
2784 Details : place this wait on necessary place instead of always waiting.
2786 --------------------------------------------------------------------------------
2788 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
2790 * Support for kernels:
2791 2.6.5-7.286 (SLES 9),
2792 2.6.9-55.0.9.EL (RHEL 4),
2793 2.6.16.53-0.8 (SLES 10),
2794 2.6.18-8.1.14.el5 (RHEL 5),
2795 2.6.18.8 vanilla (kernel.org)
2796 * Client support for unpatched kernels:
2797 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2798 2.6.16 - 2.6.22 vanilla (kernel.org)
2799 * Due to recently discovered recovery problems, we do not recommend
2800 using patchless RHEL 4 clients with this or any earlier release.
2801 * Recommended e2fsprogs version: 1.40.2-cfs1
2802 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2805 Frequency : occasional
2807 Description: MDS hang or stay in waiting lock
2808 Details : If client receive lock with CBPENDING flag ldlm need send lock
2809 cancel as separate rpc, to avoid situation when cancel request
2810 can't processed due all i/o threads stay in wait lock.
2813 Frequency : occasional
2815 Description: Do not fail import if osc_interpret_create gets -EAGAIN
2816 Details : If osc_interpret_create got -EAGAIN it immediately exits and
2817 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
2818 oscc_has_objects and see OSC has no objests and call
2819 oscc_internal_create to resend create request.
2821 Severity : enhancement
2823 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
2824 Details : Update which_patch & target file for SLES10 latest kernel.
2826 Severity : enhancement
2828 Description: add --type and --size parameters to lfs find
2829 Details : Enhance lfs find by adding filetype and filesize parameters. Also
2830 multiple OBDs can now be specified for the --obd option.
2832 Severity : enhancement
2834 Description: eliminate client locks in face of contention
2835 Details : file contention detection and lockless i/o implementation
2836 for contended files.
2838 Severity : enhancement
2840 Description: Remove client patches from SLES 10 kernel.
2841 Details : This causes SLES 10 clients to behave as patchless clients
2842 even on a Lustre-patched (server) kernel.
2844 Severity : enhancement
2846 Description: use i_size_read and i_size_write in 2.6 port
2847 Details : replace inode->i_size access with i_size_read/write()
2850 Frequency : when removing large files
2852 Description: scheduling issue during removal of large Lustre files
2853 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
2854 It causes scheduling issues when removing large files (17TB in the
2860 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
2861 Details : Impossible (EPROTO is returned) to access a directory that has a
2862 non-default striping and ACLs.
2865 Frequency : only on ppc
2867 Description: /proc/fs/lustre/devices broken on ppc
2868 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
2869 We should make sure the type of 'index' is loff_t and then cast
2870 explicitly as needed below. Do not assign an explicitly cast
2874 Frequency : only for rhel5
2876 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
2877 Details : Modify the target file & which_kernel.
2880 Frequency : if the uninit_groups feature is enabled on ldiskfs
2882 Description: e2fsck reports "invalid unused inodes count"
2883 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
2884 feature and only a single inode is created in a group then the
2885 "bg_unused_inodes" count is incorrectly updated. Creating a
2886 second inode in that group would update it correctly.
2891 Description: buffer overruns could theoretically occur
2892 Details : llapi_semantic_traverse() modifies the "path" argument by
2893 appending values to the end of the origin string, and a buffer
2894 overrun may occur. Adding buffer overrun check in liblustreapi.
2896 Severity : enhancement
2898 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
2901 Frequency : only if filesystem is inconsistent
2903 Description: handle "serious error: objid * already exists" more gracefully
2904 Details : If LAST_ID value on disk is smaller than the objects existing in
2905 the O/0/d* directories, it indicates disk corruption and causes an
2906 LBUG(). If the object is 0-length, then we should use the existing
2907 object. This will help to avoid a full fsck in most cases.
2909 Severity : enhancement
2911 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
2912 Details : Modify vm-tunables-rhel4.patch.
2914 Severity : enhancement
2916 Description: Kernel config for 2.6.18-vanilla.
2917 Details : Modify targets/2.6-vanilla.target.in.
2918 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
2919 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
2920 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
2921 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
2924 Frequency : occasional
2926 Description: improve handling recoverable errors
2927 Details : If request processed with error which can be recoverable on server
2928 request should be resend, otherwise page released from cache and
2934 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
2935 Details : Modify target file & which_patch.
2936 A flaw was found in the IA32 system call emulation provided
2937 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
2938 value could be stored in the %RAX register, which could trigger an
2939 out-of-bounds system call table access. An untrusted local user
2940 could exploit this flaw to run code in the kernel
2941 (ie a root privilege escalation). (CVE-2007-4573).
2945 Description: change order of libsysio includes
2946 Details : '#include sysio.h' should always come before '#include xtio.h'
2948 Severity : enhancement
2950 Description: adapt the lustre_config script to support the upgrade case
2951 Details : Add "-u" option for lustre_config script to support upgrading 1.4
2952 server targets to 1.6 in parallel.
2957 Description: To avoid grant space > avaible space when the disk is almost
2958 full. Without this patch you might see the error "grant XXXX >
2959 available" or some LBUG about grant, when the disk is almost
2961 Details : In filter_check_grant, for non_grant cache write, we should
2962 check the left space by if (*left > ungranted + bytes), instead
2963 of (*left > ungranted), because only we are sure the left space
2964 is enough for another "bytes", then the ungrant space should be
2965 increase. In client, we should update cl_avail_grant only there is
2966 OBD_MD_FLGRANT in the reply.
2971 Description: Update RHEL 4 kernel to fix local root privilege escalation.
2972 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
2973 described in CVE-2007-4573. This problem could allow untrusted
2974 local users to gain root access.
2977 Frequency : when using O_DIRECT and quotas
2979 Description: Incorrect file ownership on O_DIRECT output files
2980 Details : block usage reported by 'lfs quota' does not take into account
2981 files that have been written with O_DIRECT.
2985 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2986 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2987 window to 0. If it does hit cache miss, and read-pattern is still
2988 stride-io mode, does not reset the stride window, but also does not
2989 increase the stride window length in this case.
2991 --------------------------------------------------------------------------------
2993 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
2995 * Support for kernels:
2996 2.6.5-7.286 (SLES 9),
2997 2.6.9-55.0.2.EL (RHEL 4),
2998 2.6.16.46-0.14 (SLES 10),
2999 2.6.18-8.1.8.el5 (RHEL 5),
3000 2.6.18.8 vanilla (kernel.org)
3001 * Client support for unpatched kernels:
3002 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3003 2.6.16 - 2.6.21 vanilla (kernel.org)
3004 * Due to recently discovered recovery problems, we do not recommend
3005 using patchless RHEL 4 clients with this or any earlier release.
3006 * Recommended e2fsprogs version: 1.40.2-cfs1
3007 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3011 Description: Fix errors in lfs documentation
3012 Details : Fixes man pages
3014 Severity : enhancement
3016 Description: Adaptive timeouts
3017 Details : RPC timeouts adapt to changing server load and network
3018 conditions to reduce resend attempts and improve recovery time.
3020 Severity : enhancement
3022 Description: llapi_file_create() does not allow some changes
3023 Details : add llapi_file_open() that allows specifying the file creation
3024 mode and open flags, and also returns an open file handle.
3026 Severity : enhancement
3028 Description: df doesn't work properly if diskfs blocksize != 4K
3029 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
3032 Frequency : always on directories with default striping set
3034 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
3035 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
3037 Severity : enhancement
3039 Description: merge and cleanup kernel patches.
3040 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
3042 Severity : enhancement
3044 Description: RedHat Update kernel for RHEL5
3045 Details : Kernel config file for RHEL5.
3047 Severity : enhancement
3049 Description: OSS needs mutliple precreate threads
3050 Details : Add ability to start more than one create thread per OSS.
3053 Frequency : only with quota on the root user
3055 Description: mds_obd_create error creating tmp object
3056 Details : When the user sets quota on root, llog will be affected and can't
3057 create files and write files.
3060 Frequency : Always on ia64 patchless client, and possibly others.
3062 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
3063 Details : This allows the patchless client to be loaded on architectures
3064 without this export.
3069 Description: disorder of journal start and llog_add cause deadlock.
3070 Details : in llog_origin_connect, journal start should happen before
3071 llog_add keep the same order as other functions to avoid
3074 Severity : enhancement
3076 Description: RedHat Update kernel for RHEL5
3077 Details : Modify the kernel config file more closer RHEL5.
3079 Severity : enhancement
3081 Description: Build failure against Centos5 (RHEL5)
3082 Details : Define PAGE_SIZE when it isn't present.
3085 Frequency : occasionally when using NFS
3087 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
3088 Details : with NFS, the anon dentry's parent was set to itself in
3089 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
3090 corresponding dentry other than use rec->ur_name.
3092 Severity : enhancement
3094 Description: enable data checksumming by default
3095 Details : enable checksum by default, allow --disable-checksum
3096 configure option and "-o nochecksum" mount option. Checksums
3097 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
3100 Frequency : Occasionally with failover
3102 Description: Client eviction due to failover config
3103 Details : after a connection loss, the lustre client should attempt to
3104 reconnect to the last active server first before trying the
3105 other potential connections.
3107 Severity : enhancement
3109 Description: client-side metadata stat-ahead during readdir(directory readahead)
3110 Details : perform client-side metadata stat-ahead when the client detects
3111 readdir and sequential stat of dir entries therein
3116 Description: when mds and osts use different quota unit(32bit and 64bit),
3117 quota will be released repeatly.
3118 Details : void sending multiple quota reqs to mds, which will keep the
3119 status between the reqs.
3122 Frequency : only with liblustre clients on XT3
3124 Description: evictions taking too long
3125 Details : allow llrd to evict clients directly on OSTs
3129 Description: osts not allocated evenly to files
3130 Details : change the condition to increase offset_idx
3132 Severity : enhancement
3134 Description: self-adjustable client's lru lists
3135 Details : use adaptive algorithm for managing client cached locks lru
3136 lists according to current server load, other client's work
3137 pattern, memory activities, etc. Both, server and client
3138 side namespaces provide number of proc tunables for controlling
3143 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
3144 Details : rewrite inherited code (uuid parsing code from ext2 utils and
3145 readdir code from ext3) from scratch preserving functionality.
3149 Description: Only those disconnect error should be returned by rq_status.
3150 Details : In open/enqueue processs, Some errors, which will cause client
3151 disconnected, should be returned by rq_status, while other
3152 errors should still be returned by intent, then mdc or llite will
3155 Severity : enhancement
3157 Description: Tune the kernel for good SCSI performance.
3158 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
3159 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
3163 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
3164 Bugzilla : 13547, 13627
3165 Description: Data corruption for OSTs that are formatted larger than 2TB
3167 Details : When generating the bio request for lustre file writes the
3168 sector number would overflow a temporary variable before being
3169 used for the IO. The data reads correctly from Lustre (which
3170 will overflow in a similar manner) but other file data or
3171 filesystem metadata may be corrupted in some cases.
3175 Description: "lfs find -obd UUID" prints directories
3176 Details : "lfs find -obd UUID" will return all directory names instead
3177 of just file names. It is incorrect because the directories
3178 do not reside on the OSTs.
3181 Bugzilla : 13596, 13614
3182 Description: MDS hang after unclean shutdown of lots of clients
3183 Details : Never resend AST requests.
3187 Frequency : Always, for kernels after 2.6.16
3188 Description: Fix warning idr_remove called for id=.. which is not allocated.
3189 Details : Last kernels save old s_dev before kill super and not allow
3190 to restore from callback - restore it before call kill_anon_super.
3194 Description: Fix error on 'ls .' at the top of the Lustre mount.
3195 Details : Don't revalidate dentry if it is a root dentry.
3197 --------------------------------------------------------------------------------
3199 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
3201 * Support for kernels:
3202 2.6.5-7.286 (SLES 9),
3203 2.6.9-55.0.2.EL (RHEL 4),
3204 2.6.16.46-0.14 (SLES 10),
3205 2.6.18-8.1.8.el5 (RHEL 5),
3206 2.6.18.8 vanilla (kernel.org)
3207 * Client support for unpatched kernels:
3208 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3209 2.6.16 - 2.6.22 vanilla (kernel.org)
3210 * Due to recently discovered recovery problems, we do not recommend
3211 using patchless RHEL 4 clients with this or any earlier release.
3212 * Recommended e2fsprogs version: 1.39.cfs8
3213 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3218 Description: block reactivating mgc import until all deactivates complete
3219 Details : Fix race when failing back MDT/MGS to itself (testing)
3221 Severity : enhancement
3223 Description: lfs setstripe enhancement
3224 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
3227 Frequency : mds/oss recovery
3229 Description: llog ctxt is refrenced after it has been freed.
3230 Details : llog ctxt refcount was added to avoide the race between ctxt free
3231 and llog recovery process. Each llog user must hold ctxt refcount
3232 before it access the llog. And the llog ctxt can only be freed
3233 when its refcount is zero.
3235 Severity : enhancement
3237 Description: randomly memory allocation failure util
3238 Details : Make lustre randomly failed allocating memory for testing purpose.
3241 Frequency : only for SLES10
3243 Description: Update kernel patch for SLES10 SP1
3244 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
3246 Severity : enhancement
3248 Description: omit set fsid for export NFS
3249 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
3250 as FS_REQUIRES_DEV for avoid problems with generate fsid.
3253 Frequency : after network failures
3255 Description: Add sync option to mount_lustre.c
3256 Details : Client loses data written to lustre after a network interruption.
3258 Severity : enhancement
3260 Description: Error message improvement.
3261 Details : Merging of two LCONSOLE_ERROR_MSG into one.
3263 Severity : enhancement
3265 Description: don't use GFP_* in generic Lustre code.
3266 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
3268 Severity : enhancement
3270 Description: obdclass is limited by single OBD_ALLOC(idarray)
3271 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
3273 Severity : enhancement
3275 Description: updated patchess for new RHEL4 kernel
3276 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
3278 Severity : enhancement
3280 Description: warnings with build patchless client with vanila 2.6.19 and up
3281 Details : change old ctl_table style and replace ctl_table/ctl_table_header
3282 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
3284 Severity : enhancement
3286 Description: O_DIRECT bypasses client statistics.
3287 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
3288 read_ahead_stats were updated. Copied stats section from
3289 osc_send_oap_rpc() into async_internal().
3292 Frequency : only for Cray XT3
3294 Description: peer credits not enough on many OST per OSS systems.
3295 Details : Use new lnet way to add credits as we need those for pings and ASTs
3298 Frequency : only with liblustre
3300 Description: Liblustre is not releasing flock locks on file close.
3301 Details : Release flock locks on file close.
3304 Frequency : only for RHEL4
3306 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
3307 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
3308 Update target file and kernel config.
3312 Description: lustre support for RHEL5
3313 Details : Add support for RHEL5.
3318 Description: ASSERTION(export != NULL) failed in target_handle_connect
3319 Details : Assetion hit is result of rare race between disconnect and connet
3320 to same nid. target_handle_connect found old connect cockie and
3321 tried to reconnect, but can't find export for this cockie.
3326 Description: umount blocks forever on error
3327 Details : In result of wrong using obd_no_recov and obd_force flags client
3328 can hand if cancel or some other requests is lost.
3331 Frequency : Only for SLES
3333 Description: sanity_quota fail test_1
3334 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
3335 file, which makes TSTID[2] inunique.
3337 Severity : enhancement
3339 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
3340 Details : Update target/ChangeLog/which_patch .
3342 Severity : enhancement
3344 Description: jbd statistics
3345 Details : Port older jbd statistics patch for sles10
3350 Description: lvbo_init failed for resource with missing objects.
3351 Details : Fix returning error if we do stat for file with missing/corrupted
3352 objects and i_size set to all sum of size of all avaible objects.
3353 if we to truncate/write to missing object - it is recreated.
3358 Description: Oops in read and write path when failing to allocate lock.
3359 Details : Check if lock allocation failed and return error back.
3362 Frequency : When flocks are used.
3364 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
3365 Details : Flock locks might destroy just granted lock if it could be merged
3366 with another existing flock, this is done in completion handler,
3367 so teach ldlm_cli_enquque_fini that this is a valid case for
3373 Description: reply_lock_interpret crash due to race with it and lock cancel.
3374 Details : Do not replay locks that are being cancelled. Do not reference
3375 locks by their address during replay, just by their handle.
3378 Frequency : only with deactivated OSTs
3380 Description: lstripe command fails for valid OST index
3381 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
3382 instead of lov->desc.ld_active_tgt_count.
3384 Severity : enhancement
3386 Description: Build failure against Centos5 (RHEL5)
3387 Details : Use getpagesize() instead of PAGE_SIZE.
3389 --------------------------------------------------------------------------------
3391 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
3393 * Support for kernels:
3394 2.6.5-7.283 (SLES 9),
3395 2.6.9-55.EL (RHEL 4),
3396 2.6.16.46-0.14 (SLES 10),
3397 2.6.18.8 vanilla (kernel.org)
3398 * Client support for unpatched kernels:
3399 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
3400 2.6.16 - 2.6.22 vanilla (kernel.org)
3401 * Due to recently discovered recovery problems, we do not recommend
3402 using patchless RHEL 4 clients with this or any earlier release.
3403 * Recommended e2fsprogs version: 1.39.cfs8
3404 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
3405 * Starting with this release, the ldiskfs backing filesystem required
3406 by Lustre is now in its own package, lustre-ldiskfs. This package
3407 should be installed. It is versioned separately from Lustre and
3408 may be released separately in future.
3413 Description: server LBUG when shutting down
3414 Details : Block umount forever until the mount refcount is zero rather
3415 than giving up after an arbitrary timeout.
3417 Severity : enhancement
3419 Description: add optional extra BUILD_VERSION info
3420 Details : add a new environment variable (namely LUSTRE_VERS) which allows
3421 to override the lustre version.
3424 Frequency : 2.6.18 servers only
3426 Description: ll_kern_mount() doesn't release the module reference
3427 Details : The ldiskfs module reference count never drops down to 0
3428 because ll_kern_mount() doesn't release the module reference.
3433 Description: server LBUG when using old ost_num_threads parameter
3434 Details : Accept the old ost_num_threads parameter but warn that it
3435 is deprecated, and fix an off-by-one error that caused an LBUG.
3440 Description: Transient SCSI error results in persistent IO issue
3441 Details : iobuf->dr_error is not reinitialized to 0 between two
3445 Frequency : sometimes when underlying device returns I/O errors
3447 Description: OSTs not going read-only during write failures
3448 Details : OSTs are not remounted read-only when the journal commit threads
3449 get I/O errors because fsfilt_ext3 calls journal_start/stop()
3450 instead of the ext3 wrappers.
3453 Frequency : only with 10000 clients or more
3455 Description: poor connect scaling with increasing client count
3456 Details : Don't run filter_grant_sanity_check for more than 100 exports
3457 to improve scaling for large numbers of clients.
3460 Frequency : SLES10 only
3462 Description: sanity-quota.sh quotacheck failed: rc = -22
3463 Details : Quotas cannot be enabled on SLES10.
3466 Frequency : liblustre clients only
3468 Description: getdirentries does not give error when run on compute nodes
3469 Details : getdirentries does not fail when the size specified as an argument
3470 is too small to contain at least one entry
3472 Severity : enhancement
3474 Description: Add LNET router traceability for debug purposes
3475 Details : If a checksum failure occurs with a router as part of the
3476 IO path, the NID of the last router that forwarded the bulk data
3477 is printed so it can be identified.
3482 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
3483 Details : Due to a race condition, liblustre clients were occasionally
3484 evicted incorrectly.
3486 Severity : enhancement
3488 Description: lfs setstripe use optional parameters instead of postional
3491 Severity : enhancement
3493 Description: Nanosecond timestamp support for ldiskfs
3494 Details : The on-disk ldiskfs filesystem has added support for nanosecond
3495 resolution timestamps. There is not yet support for this at
3496 the Lustre filesystem level.
3499 Frequency : during server recovery
3501 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
3502 Details : request with rq_no_resend flag not awake l_wait_event if they get a
3506 Frequency : nfs export on patchless client
3508 Description: connectathon hang when test nfs export over patchless client
3509 Details : Disconnected dentry cannot be found with lookup, so we do not need
3510 to unhash it or make it invalid
3513 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
3514 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
3515 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
3520 Description: open req refcounting wrong on reconnect
3521 Details : If reconnect happened between getting open reply from server and
3522 call to mdc_set_replay_data in ll_file_open, we will schedule
3523 replay for unreferenced request that we are about to free.
3524 Subsequent close will crash in variety of ways.
3525 Check that request is still eligible for replay in
3526 mdc_set_replay_data().
3531 Description: disable writes to filesystem when reading health_check file
3532 Details : the default for reading the health_check proc file has changed
3533 to NOT do a journal transaction and write to disk, because this
3534 can cause reads of the /proc file to hang and block HA state
3535 checking on a healthy but otherwise heavily loaded system. It
3536 is possible to return to the previous behaviour during configure
3537 with --enable-health-write.
3539 Severity : enhancement
3541 Description: 64-bit inode version
3542 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
3543 to the inode. This will be required for version-based recovery.
3548 Description: MDS fails to start if a duplicate client export is detected
3549 Details : in some rare cases it was possible for a client to connect to
3550 an MDS multiple times. Upon recovery the MDS would detect this
3551 and fail during startup. Handle this more gracefully.
3553 Severity : enhancement
3555 Description: Add -o localflock option to simulate old noflock behaviour.
3556 Details : This will achieve local-only flock/fcntl locks coherentness.
3561 Description: log_commit_thread vs filter_destroy race leads to crash
3562 Details : Take import reference before releasing llog record semaphore
3567 Description: Wrong request locking in request set processing
3568 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
3569 imp_delayed_list, in this place should be used imp_lock.
3572 Frequency : when reconnecting
3574 Description: Grant leak when OSC reconnect to OST
3575 Details : When osc reconnect ost, OST(filter) should check whether it
3576 should grant more space to client by comparing fed_grant and
3577 cl_avail_grant, and return the granted space to client instead
3578 of "new granted" space, because client will call osc_init_grant
3579 to update the client grant space info.
3582 Frequency : when client reconnects to OST
3584 Description: Grant leak when OSC does a resend and replays bulk write
3585 Details : When osc reconnect to OST, OST(filter) should clear grant info of
3586 bulk write request, because the grant info while be sync between
3587 OSC and OST when reconnect, and we should ignore the grant info
3588 these of resend/replay write req.
3593 Description: Grant space more than avaiable space sometimes.
3594 Details : When then OST is about to be full, if two bulk writing from
3595 different clients came to OST. Accord the avaliable space of the
3596 OST, the first req should be permitted, and the second one
3597 should be denied by ENOSPC. But if the seconde arrived before
3598 the first one is commited. The OST might wrongly permit second
3599 writing, which will cause grant space > avaiable space.
3602 Frequency : when client is evicted
3604 Description: Grant might be wrongly erased when osc is evicted by OST
3605 Details : when the import is evicted by server, it will fork another
3606 thread ptlrpc_invalidate_import_thread to invalidate the
3607 import, where the grant will be set to 0. While the original
3608 thread will update the grant it got when connecting. So if
3609 the former happened latter, the grant will be wrongly errased
3610 because of this race.
3615 Description: Checking Stale with correct fid
3616 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
3617 to check whether it is stale, because sometimes, we want the
3618 enqueue happened anyway, and op_data.fid2 will not be initialized.
3620 Severity : enhancement
3622 Description: update patchless client
3623 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
3626 Frequency : only with 2.4 kernel
3628 Description: random memory corruption
3629 Details : size of struct ll_inode_info is to big for union inode.u and this
3630 can be cause of random memory corruption.
3635 Description: Memory leak in recovery
3636 Details : Lov_mds_md was not free in an error handler in mds_create_object.
3637 It should also check obd_fail before fsfilt_start, otherwise if
3638 fsfilt_start return -EROFS,(failover mds during mds recovery).
3639 then the req will return with repmsg->transno = 0 and rc = EROFS.
3640 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
3641 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
3642 be freed no matter whether fsfilt_commit success or not.
3645 Frequency : only with large numbers of cores on a single node
3647 Description: Prevents from taking the superblock lock in llap_from_page for
3649 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
3650 for llap_from_page call in ll_removepage() prevents from taking
3651 the superblock lock for a soon died page.
3656 Description: Not check open intent error before release open handle
3657 Details : in some rare cases, the open intent error is not checked before
3658 release open handle, which may cause
3659 ASSERTION(open_req->rq_transno != 0), because it tries to release
3660 the failed open handle.
3665 Description: Set cat log bitmap only after create log success.
3666 Details : in some rare cases, the cat log bitmap is set too early. and it
3667 should be set only after create log success.
3671 Description: Accessing a block bevice can re-enable I/O when Lustre is
3672 tearing down a device.
3673 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
3678 Description: service threads may hog cpus when there are a lot of requests
3679 Details : Insert cond_resched to give other threads a chance to use some CPU
3684 Description: the cat log was not initialized in recovery
3685 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
3686 unlink log on mds will not be initialized until mds post
3687 recovery. And also in mds post recovery, the unlink log will
3688 initialization will be done asynchronausly, so there will be race
3689 between add unlink log and unlink log initialization.
3693 Description: brw_stats were being printed incorrectly
3694 Details : brw_stats were being printed as log2 but all of them were not
3695 recorded as log2. Also remove some code duplication arising from
3696 filter_tally_{read,write}.
3700 Frequency : rare, only in recovery.
3701 Description: ASSERTION(req->rq_type != LI_POISON) failed
3702 Details : imp_lock should be held while iterating over imp_sending_list for
3703 prevent destroy request after get timeout in ptlrpc_queue_wait.
3707 Description: replay-single.sh test 52 fails
3708 Details : A lock's skiplist need to be cleanup when it being unlinked
3709 from its resource list.
3714 Description: Short directio read returns full requested size rather than
3716 Details : Direct I/O operations should return actual amount of bytes
3717 transferred rather than requested size.
3719 Severity : enhancement
3721 Description: metadata RPC reduction (e.g. for rm performance)
3722 Details : decrease the amount of synchronous RPC between clients and servers
3723 by canceling conflicing lock before the operation on the client
3724 and packing thier handles into the main operation RPC to server.
3726 Severity : enhancement
3728 Description: add #ifdef HAVE_KERNEL_CONFIG_H
3729 Details : kernels from 2.6.19 not need include linux/config.h, but add
3730 include linux/autoconf.h in commpiler command line.
3732 Severity : enhancement
3734 Description: patchless client support for 2.6.22 kernel
3735 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
3736 removed. In this case we need drop using os depended interface to
3737 kmem_cache and use cfs_mem_cache API.
3742 Description: fix mal-formatted messages
3743 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
3747 Frequency : always in liblustre
3748 Description: wrong IS_ERR implementation in liblustre.h
3749 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
3754 Description: Correct condition for output debug message.
3755 Details : inode i_nlink equal zero is not enough for output message about
3756 disk corruption, i_ctime and i_mode should be also checked.
3760 Frequency : always in patchless client
3761 Description: add configure check for truncate_complete_page
3762 Details : improve checks for exported symbols. This allow run check without
3763 sources, but with Module.symvers shipped with kernel distribution.
3764 add check for truncate_complete_page used by patchless client.
3768 Description: sanity.sh test_77h fails with "test_77h file compare failed"
3769 Details : test_77h uses a file which was messed by other test case.
3773 Description: Not Check whether lov_tgts is NULL in some lov functions
3774 Details : Checking whether lov_tgts is NULL in some functions.
3778 Description: replace obdo_alloc() with OBDO_ALLOC macro
3779 Details : nothing special is done in obdo_alloc() function, and for
3780 debugging purpose, it needs to be replaced with macros.
3784 Description: bad return value and errno from fcntl call
3785 Details : In liblustre API, errno should be a negative value if error
3790 Description: ptlrpc_check_set() LBUG
3791 Details : In case of positive reply from server and failed client bulk
3792 callback after bulk transfer shouldn't LBUG, but process this
3793 request as erroneous.
3795 Severity : enhancement
3797 Description: VFS operations stats tool.
3798 Details : Tool which collects stats by tracking value written in pid,
3799 ppid, gid and uses llstat to generate output to plot graph using
3801 Updated lustre/utils/Makefile.am
3802 Added lustre/utils/ltrack_stats.c
3804 Severity : enhancement
3806 Description: 2.6.18 server support (lustre 1.6.1)
3807 Details : Support for 2.6.18 kernels on the server side.
3812 Description: ASSERTION(imp->imp_conn_current) failed
3813 Details : an assertion failure is hit if a client node boots and attempts to
3814 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
3817 Frequency : only for i686
3819 Description: 1.4.11 RC1 build fails for RHEL 4, i686
3820 Details : Fixed config variable for build.
3825 Description: Updated patchess for new RHEL4 kernel
3826 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
3827 Updated patch jbd-stats-2.6.9.patch
3828 Updated patch qsnet-rhel4-2.6.patch
3829 Updated patch quota-deadlock-on-pagelock-core.patch
3830 Updated patch vfs_intent-2.6-rhel4.patch
3831 Updated patch vfs_races-2.6-rhel4.patch
3832 Updated series file 2.6-rhel4-titech.series
3833 Updated series file 2.6-rhel4.series
3834 Updated kernel config files
3839 Description: lquota slave complains LBUG when reconnecting with mds
3841 Details : quota slave depends qctxt->lqc_import to send its quota request.
3842 This pointer will be invalid if mds did failover or broke its
3843 connect to osts, which leads to LBUG.
3846 Frequency : when qunit size is too small(less than 20M)
3848 Description: write is stopped by improper -EDQUOT
3849 Details : If the master is busy and qunit size is small enough(let's say 1M),
3850 the slave can not get quota from master on time, which will lead
3851 slave to trigger a -EQUOTA to client.
3856 Description: Deadlock during metadata tests
3857 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
3858 called with the per-dentry lock held.
3861 Frequency : SLES9 only
3863 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
3864 Details : set CONFIG_NR_CPUS to 128 instead of 8.
3866 Severity : enhancement
3868 Description: remove fs_prep_san_write operation and related patches
3869 Details : remove the ext3-san-jdike patches which are no longer useful.
3874 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
3875 Details : a disk corruption can cause the mballoc code to assert on a
3876 double free or other extent corruptions. Handle these with
3877 ext3_error() instead of assertions.
3880 Frequency : only with mballoc3 code and deep extent trees
3882 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
3883 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
3884 can cause the filesystem to be remounted read-only.
3889 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
3890 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
3891 few external users of this ioctl number is lfsck's call to
3892 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
3893 Add the old ioctl number to the handler so both old and new
3898 Description: parallel lock callbacks
3899 Details : Instead of sending blocking and completion callbacks as separated
3900 requests, adding them to a set and sending in parallel.
3904 Description: Disable most debugging by default
3905 Details : To improve performance, disable most logging (for debug purposes)
3906 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
3907 default, and HA includes fewer messages.
3911 Frequency : only run on patchless client.
3912 Description: use do_facet on sanity.sh for test handling recoverables errors
3913 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
3917 Description: hash tables for lists of nids, connections and uuids
3918 Details : Hash tables noticeably help when a lot of clients connect to a
3919 server, to faster identify duplicate connections or reconnects,
3920 also to faster find export to evict in manual eviction case.
3924 Description: Sometimes, when the server evict a client, and the client will
3925 not be evicted as soon as possible.
3926 Details : In enqueue req, the error was returned by intent, instead of
3927 rq_status which make ptlrpc layer not detect this error, and
3928 does not evict the client. So enqueue error should be returned
3932 Frequency : only at startup
3934 Description: Delay client connections to MDT until fist MDT->OST connect
3935 Details : If a client tried to create a new file before the MDT had
3936 connected to any OSTs, the create would return EIO. Now
3937 the client will simply block until the MDT connects to the
3938 first OST and the create can succeed.
3941 Frequency : only for SLES9
3943 Description: Routinely utilize latest Quadrics drivers in CFS releases
3944 Details : Update patch qsnet-suse-2.6.patch.
3947 Frequency : only for sles10
3949 Description: Update patches for SLES 10 SP1 kernel.
3950 Details : Update the patch vfs_intent-2.6-sles10.patch.
3955 Description: Routinely utilize latest Quadrics drivers in CFS releases
3956 Details : Update patch qsnet-rhel4-2.6.patch.
3959 Frequency : at statup only
3961 Description: mds_lov_synchronize race leads to various problems
3962 Details : simultaneous MDT->OST connections at startup can cause the
3963 sync to abort, leaving the OSC in a bad state.
3968 Description: Using wrong pointer in osc_brw_prep_request
3969 Details : Access to array[-1] can produce panic if kernel compiled with
3970 CONFIG_PAGE_ALLOC enabled
3972 Severity : enhancement
3974 Description: Async OSC create to avoid the blocking unnecessarily.
3975 Details : If a OST has no remain object, system will block on the creating
3976 when need to create a new object on this OST. Now, ways use
3977 pre-created objects when available, instead of blocking on an
3978 empty osc while others are not empty. If we must block, we block
3979 for the shortest possible period of time.
3983 Frequency : only in recovery
3984 Description: Mark OST as early accessible if his start SYNC.
3985 Details : osc_precreate return flag early accessible if oscc marked as
3986 OSCC_FLAG_SYNC_IN_PROGRESS.
3991 Description: Sometimes precreate code can triger create object on wrong ost
3992 Details : Wrong protected or not not restored variables aftre precreate loop
3993 can produce creation object on wrong ost.
3996 Frequency : oss recovery
3998 Description: llog_commit_thread cleanup should sync with llog_commit_thread
4000 Details : llog_commit_thread_count should be synced between llog_commit
4001 start and cleanup, so new llog_commit thread should not be started
4002 when llog_commit threads being stopped to avoid accessing some
4005 Severity : enhancement
4007 Description: Add printing inode info into message about error in writepage.
4009 --------------------------------------------------------------------------------
4011 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
4016 Frequency : on some architectures
4018 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
4019 Details : Uninitialized flags sometimes cause configuration commands to
4023 Frequency : patchless clients only
4025 Description: missing __iget() symbol export
4026 Details : The __iget() symbol export is missing. To avoid the need for
4027 this on patchless clients the deathrow inode reaper is turned
4028 off, and we depend on the VM to clean up old inodes. This
4029 dependency was during via the fix for bug 12181.
4034 Description: sanity.sh fail: test_52b
4035 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
4036 incorrect inode's flags to client.
4038 --------------------------------------------------------------------------------
4040 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
4042 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
4043 INTEROPERATE with older versions automatically. In many cases a
4044 special upgrade step is needed. Please read the
4045 user documentation before upgrading any part of a 1.4.x system.
4046 * WARNING: Lustre configuration and startup changes are required with
4047 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
4049 * Support for kernels:
4050 2.4.21-47.0.1.EL (RHEL 3),
4051 2.6.5-7.283 (SLES 9),
4052 2.6.9-42.0.10.EL (RHEL 4),
4053 2.6.12.6 vanilla (kernel.org),
4054 2.6.16.27-0.9 (SLES10)
4055 * Client support for unpatched kernels:
4056 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
4057 2.6.16 - 2.6.19 vanilla (kernel.org),
4058 2.6.9-42.0.8EL (RHEL 4)
4059 * Recommended e2fsprogs version: 1.39.cfs6
4060 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4063 Severity : enhancement
4065 Description: Async OSC create to avoid the blocking unnecessarily.
4066 Details : If a OST has no remain object, system will block on the creating
4067 when need to create a new object on this OST. Now, ways use
4068 pre-created objects when available, instead of blocking on an
4069 empty osc while others are not empty. If we must block, we block
4070 for the shortest possible period of time.
4072 Severity : enhancement
4074 Description: MountConf
4075 Details : Lustre configuration is now managed via mkfs and mount
4076 commands instead of lmc and lconf. New obd types (MGS, MGC)
4077 are added for dynamic configuration management. See
4078 https://mail.clusterfs.com/wikis/lustre/MountConf for
4081 Severity : enhancement
4083 Description: dynamic OST addition
4084 Details : OSTs can now be added to a live filesystem
4086 Severity : enhancement
4088 Description: startup order invariance
4089 Details : MDTs and OSTs can be started in any order. Clients only
4090 require the MDT to complete startup.
4092 Severity : enhancement
4094 Description: parallel, asynchronous orphan cleanup
4095 Details : orphan cleanup is now performed in separate threads for each
4096 OST, allowing parallel non-blocking operation.
4098 Severity : enhancement
4100 Description: optimized stripe assignment
4101 Details : stripe assignments are now made based on ost space available,
4102 ost previous usage, and OSS previous usage, in order to try
4103 to optimize storage space and networking resources.
4105 Severity : enhancement
4107 Description: Permanently set tunables
4108 Details : All writable /proc/fs/lustre tunables can now be permanently
4109 set on a per-server basis, at mkfs time or on a live
4112 Severity : enhancement
4114 Description: Lustre message v2
4115 Details : Add lustre message format v2.
4117 Severity : enhancement
4119 Description: client OST exclusion list
4120 Details : Clients can be started with a list of OSTs that should be
4121 declared "inactive" for known non-responsive OSTs.
4125 Description: ENOENT returned for valid filehandle during dbench.
4126 Details : Check if a directory has children when invalidating dentries
4127 associated with an inode during lock cancellation. This fixes
4128 an incorrect ENOENT sometimes seen for valid filehandles during
4129 testing with dbench.
4132 Frequency : SFS test only (otherwise harmless)
4134 Description: SPEC SFS validation failure on NFS v2 over lustre.
4135 Details : Changes the blocksize for regular files to be 2x RPC size,
4136 and not depend on stripe size.
4138 Severity : enhancement
4140 Description: fine-grained SMP locking inside DLM
4141 Details : Improve DLM performance on SMP systems by removing the single
4142 per-namespace lock and replace it with per-resource locks.
4144 Severity : enhancement
4146 Description: don't hold multiple extent locks at one time
4147 Details : To avoid client eviction during large writes, locks are not
4148 held on multiple stripes at one time or for very large writes.
4149 Otherwise, clients can block waiting for a lock on a failed OST
4150 while holding locks on other OSTs and be evicted.
4152 Severity : enhancement
4154 Description: Multiple MD RPCs in flight.
4155 Details : Further unserialise some read-only MDT RPCs - learn about intents.
4156 To avoid overly-overloading MDT, introduce a limit on number of
4157 MDT RPCs in flight for a single client and add /proc controls
4158 to adjust this limit.
4160 Severity : enhancement
4162 Description: client read/write statistics
4163 Details : Add client read/write call usage stats for performance
4164 analysis of user processes.
4165 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
4166 file access. extents_stats shows chunk size distribution.
4167 extents_stats_per_process show chunk size distribution per
4170 Severity : enhancement
4172 Description: per-client statistics on server
4173 Details : Add ldlm and operations statistics for each client in
4174 /proc/fs/lustre/mds|obdfilter/*/exports/
4176 Severity : enhancement
4178 Description: improved MDT statistics
4179 Details : Add detailed MDT operations statistics in
4180 /proc/fs/lustre/mds/*/stats
4182 Severity : enhancement
4184 Description: VFS operations stats
4185 Details : Add client VFS call stats, trackable by pid, ppid, or gid
4186 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
4191 Description: Fix client-side osc byte counters
4192 Details : The osc read/write byte counters in
4193 /proc/fs/lustre/osc/*/stats are now working
4196 Frequency : always as root on SLES
4198 Description: Failure of copying files with lustre special EAs.
4199 Details : Client side always return success for setxattr call for lustre
4200 special xattr (currently only "trusted.lov").
4205 Description: Refcount LNET uuids
4206 Details : The global LNET uuid list grew linearly with every startup;
4207 refcount repeated list entries instead of always adding to
4210 Severity : enhancement
4212 Description: Dynamic service threads
4213 Details : Within a small range, start extra service threads
4214 automatically when the request queue builds up.
4217 Frequency : mixed-endian client/server environments
4219 Description: mixed-endian crashes
4220 Details : The new msg_v2 system had some failures in mixed-endian
4223 Severity : enhancement
4225 Description: Easy OST removal
4226 Details : OSTs can be permanently deactivated with e.g. 'lctl
4227 conf_param lustre-OST0001.osc.active=0'
4229 Severity : enhancement
4231 Description: MGS proc entries
4232 Details : Added basic proc entries for the MGS showing what filesystems
4235 Severity : enhancement
4237 Description: provide MGS failover
4238 Details : Added config lock reacquisition after MGS server failover.
4240 Severity : enhancement
4242 Description: add Linux 2.4 support
4243 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
4247 Description: a large application tries to do I/O to the same resource and dies
4248 in the middle of it.
4249 Details : Check the req->rq_arrival time after the call to
4250 ost_brw_lock_get(), but before we do anything about
4251 processing it & sending the BULK transfer request. This
4252 should help move old stale pending locks off the queue as
4253 quickly as obd_timeout.
4256 Frequency : when an incorrect nid is specified during startup
4258 Description: ptlrpc connect to non-existant node causes kernel crash
4259 Details : LNET can't be re-entered from an event callback, which
4260 happened when we expire a message after the export has been
4261 cleaned up. Instead, hand the zombie cleanup off to another
4264 Severity : enhancement
4266 Description: plain/inodebits lock performance improvement
4267 Details : Grouping plain/inodebits in granted list by their request modes
4268 and bits policy, thus improving the performance of search through
4272 Frequency : only if OST filesystem is corrupted
4274 Description: client incorrectly hits assertion in ptlrpc_replay_req()
4275 Details : for a short time RPCs with bulk IO are in the replay list,
4276 but replay of bulk IOs is unimplemented. If the OST filesystem
4277 is corrupted due to disk cache incoherency and then replay is
4278 started it is possible to trip an assertion. Avoid putting
4279 committed RPCs into the replay list at all to avoid this issue.
4282 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
4284 Description: System hang on startup
4285 Details : This bug allowed the liblustre (e.g. catamount) client to
4286 return to the app before handling all startup RPCs. This
4287 could leave the node unresponsive to lustre network traffic
4288 and manifested as a server ptllnd timeout.
4290 Severity : enhancement
4292 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
4293 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
4294 boolean to control whether to print peer debug info when a
4295 client's RPC times out.
4298 Frequency : only for kernels with patches from Lustre below 1.4.3
4300 Description: Remove old rdonly API
4301 Details : Remove old rdonly API which unused from at least lustre 1.4.3
4304 Frequency : only for devices with external journals
4306 Description: Set external device read-only also
4307 Details : During a commanded failover stop, we set the disk device
4308 read-only while the server shuts down. We now also set any
4309 external journal device read-only at the same time.
4312 Frequency : when upgrading from 1.4 while trying to change parameters
4314 Description: The wrong (new) MDC name was used when setting parameters for
4315 upgraded MDT's. Also allows changing of OSC (and MDC)
4316 parameters if --writeconf is specified at tunefs upgrade time.
4319 Frequency : when setting specific ost indicies
4321 Description: QOS code breaks on skipped indicies
4322 Details : Add checks for missing OST indicies in the QOS code, so OSTs
4323 created with --index need not be sequential.
4325 Severity : enhancement
4327 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
4328 Details : The uninit_groups feature works in conjunction with the kernel
4329 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
4330 up the pass1 processing of e2fsck. This is a read-only feature
4331 in ldiskfs2 only, so older kernels and current ldiskfs cannot
4332 mount filesystems that have had this feature enabled.
4334 Severity : enhancement
4336 Description: Improve multi-block allocation algorithm to avoid fragmentation
4337 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
4338 allocation locality and avoid filesystem fragmentation.
4340 ------------------------------------------------------------------------------
4342 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
4344 * Support for kernels:
4345 2.4.21-47.0.1.EL (RHEL 3)
4346 2.6.5-7.283 (SLES 9)
4347 2.6.9-42.0.10.EL (RHEL 4)
4348 2.6.12.6 vanilla (kernel.org)
4349 2.6.16.27-0.9 (SLES 10)
4350 * Recommended e2fsprogs version: 1.39.cfs5
4352 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4356 Frequency : occasional, depends on client load and configuration
4357 Bugzilla : 12181, 12203
4358 Description: data loss for recently-modified files
4360 Details : In some cases it is possible that recently written or created
4361 files may not be written to disk in a timely manner (this should
4362 normally be within 30s unless client IO load is very high).
4363 The problem appears as zero-length files or files that are a
4364 multiple of 1MB in size after a client crash or client eviction
4365 that are missing data at the end of the file.
4367 This problem is more likely to be hit on clients where files are
4368 repeatedly created and unlinked in the same directory, clients
4369 have a large amount of RAM, have many CPUs, the filesystem has
4370 many OSTs, the clients are rebooted frequently, and/or the files
4371 are not accessed by other nodes after being written.
4373 The presence of the problem can be detected by looking at
4374 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
4375 smaller than the second (nr_unused) then dirty files will not
4376 be flushed automatically to disk. "sync; sleep 10" should be
4377 run several times on the node before unmounting it to update
4378 Lustre (this is also safe to run on nodes without this problem).
4380 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
4381 that can cause this same problem, so customers using that kernel
4382 also need to update the kernel in addition to Lustre. In order
4383 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
4385 It is normal that files written just before a client crash (less
4386 than 30s) may not yet have been flushed to disk, even for local
4390 Frequency : frequent on thin XT3 nodes
4392 Description: UUID collision on thin XT3 Linux nodes
4393 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
4394 randomly, since we relied on an insufficiently-seeded PRNG.
4399 Description: OSS hangs after "All ost request buffers busy"
4400 Details : A deadlock between quota and journal operations caused OSS
4401 hangs after printing "All ost request buffers busy."
4404 Frequency : always on liblustre builds
4406 Description: Cleanup compiler warnings on liblustre
4409 Frequency : always on liblustre builds on XT3
4411 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
4415 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
4416 > 32000 subdirectories
4417 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
4418 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
4419 subdirectory count crosses 32000. This will aid e2fsck to
4420 correctly handle more than 32000 subdirectories.
4423 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
4425 Description: System hang on startup
4426 Details : This bug allowed the liblustre (e.g. catamount) client to
4427 return to the app before handling all startup RPCs. This
4428 could leave the node unresponsive to lustre network traffic
4429 and manifested as a server ptllnd timeout.
4431 Severity : enhancement
4433 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
4434 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
4435 boolean to control whether to print peer debug info when a
4436 client's RPC times out.
4441 Description: make O_SYNC working on 2.6 kernels
4442 Details : 2.6 kernels use different method for mark pages for write,
4443 so need add a code to lustre for O_SYNC work.
4448 Description: Failure to close file and release space on NFS
4449 Details : Put inode details into lock acquired in ll_intent_file_open.
4450 Use mdc_intent_lock in ll_intent_open to properly
4451 detect all kind of errors unhandled by mdc_enqueue.
4456 Description: proc file read during shutdown sometimes raced obd removal,
4458 Details : Add lock to prevent obd access after proc file removal.
4461 Frequency : Only for files larger than 4GB on 32-bit clients.
4463 Description: improperly doing page alignment of locks
4464 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
4465 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
4470 Description: RPCs being resent when they shouldn't be
4471 Details : Some RPCs that should not be resent are being resent. This
4472 can cause inconsistencies in the RPC state machine. Do not
4473 resend such requests.
4476 Frequency : rare, only with NFS export
4478 Description: Crash on NFS re-export node
4479 Details : under very unusual load conditions an assertion is hit in
4480 ll_intent_file_open()
4483 Frequency : only if OST filesystem is corrupted
4485 Description: client incorrectly hits assertion in ptlrpc_replay_req()
4486 Details : for a short time RPCs with bulk IO are in the replay list,
4487 but replay of bulk IOs is unimplemented. If the OST filesystem
4488 is corrupted due to disk cache incoherency and then replay is
4489 started it is possible to trip an assertion. Avoid putting
4490 committed RPCs into the replay list at all to avoid this issue.
4495 Description: large O_DIRECT requests fail under memory pressure/fragmentation
4496 Details : Large single O_DIRECT read and write calls can fail to allocate
4497 a sufficiently large buffer to process the request. In case of
4498 allocation failure the allocation is retried with a smaller
4499 buffer and broken into smaller requests.
4501 Severity : enhancement
4503 Description: Add -o localflock option to simulate old noflock behaviour.
4504 Details : This will achieve local-only flock/fcntl locks coherentness.
4509 Description: versioning check is incomplete
4510 Details : Checking the version difference of client vs. server, report
4511 error if the gap is too big.
4516 Description: add support PG_writeback bit
4517 Details : add support for PG_writeback bit for Lustre, for more carefull
4518 work with page cache in 2.6 kernel. This also fix some deadlocks
4519 and remove hack for work O_SYNC with 2.6 kernel.
4521 Severity : enhancement
4523 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
4524 Details : The uninit_groups feature works in conjunction with the kernel
4525 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
4526 up the pass1 processing of e2fsck. This is a read-only feature
4527 in ldiskfs2 only, so older kernels and current ldiskfs cannot
4528 mount filesystems that have had this feature enabled.
4530 Severity : enhancement
4532 Description: Improve multi-block allocation algorithm to avoid fragmentation
4533 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
4534 allocation locality and avoid filesystem fragmentation.
4536 ------------------------------------------------------------------------------
4538 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
4540 * Support for kernels:
4541 2.6.9-42.0.3.EL (RHEL 4)
4542 2.6.5-7.276 (SLES 9)
4543 2.4.21-47.0.1.EL (RHEL 3)
4544 2.6.12.6 vanilla (kernel.org)
4545 2.6.16.21-0.8 (SLES10)
4546 * Recommended e2fsprogs version: 1.39.cfs2-0
4548 * The backwards-compatible /proc/sys/portals symlink has been removed
4549 in this release. Before upgrading, please ensure that you change
4550 any configuration scripts or /etc/sysctl.conf files that access
4551 /proc/sys/portals/* or sysctl portals.* to use the corresponding
4552 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
4553 in advance of the upgrade on any system running Lustre 1.4.6 or
4554 newer, since /proc/sys/lnet was added in that version.
4555 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
4559 Frequency : only when quota is used
4561 Description: avoid scanning export list for quota master
4562 Details : Change the algorithms to avoid scanning export list in order
4563 to improve the efficiency.
4566 Frequency : MDS failover only, very rarely
4568 Description: "went back in time" messages on mds failover
4569 Details : The greatest transno may be lost when the current operation
4570 finishes with an error (transno==0) and the client's last_rcvd
4571 record is over-written. Save the greatest transno in the
4572 mds_last_transno for this case.
4575 Frequency : always for specific kernels and striping counts
4577 Description: client may get "Matching packet too big" without ACL support
4578 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
4579 when trying to access files in certain configurations. The
4580 clients should in fact be denied when mounting because they do
4581 not understand ACLs.
4584 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
4586 Description: many clients connecting with IO in progress causes connect timeouts
4587 Details : Avoid synchronous journal commits to avoid delays caused by many
4588 clients connecting/disconnecting when bulk IO is in progress.
4589 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
4590 OST_IO_PORTAL to avoid delays behind potentially many pending
4594 Frequency : occasionally with multiple writers to a single file
4596 Description: shared writes to file may result in wrong size reported by stat()
4597 Details : Allow growing of kms when extent lock is cancelled
4600 Frequency : always with random mmap IO to multi-striped file
4602 Description: mmap write might be lost if we are writing to a 'hole' in stripe
4603 Details : Only if the hole is at the end of OST object so that kms is too
4604 small. Fix is to increase kms accordingly in ll_nopage.
4607 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
4609 Description: writes to a missing object would leak memory on the OST
4610 Details : If there is an inconsistency between the MDS and OST filesystems,
4611 such that the MDS references an object that doesn't exist, writes
4612 to that object will leak memory due to incorrect cleanup in the
4613 error handling path, eventually running out of memory on the OST.
4618 Description: Creating too long symlink causes lustre errors
4619 Details : Check symlink and name lengths before sending requests to MDS.
4622 Frequency : only if flock is enabled (not on by default)
4624 Description: posix locks not released on fd closure on 2.6.9+
4625 Details : We failed to add posix locks to list of inode locks on 2.6.9+
4626 kernels, this caused such locks not to be released on fd close and
4627 then assertions on fs unmount about still used locks.
4630 Frequency : MDS failover only, very rarely
4632 Description: clients may get ASSERTION(granted_lock != NULL)
4633 Details : When request was taking a long time, and a client was resending
4634 a getattr by name lock request. The were multiple lock requests
4635 with the same client lock handle and
4636 mds_getattr_name->fixup_handle_for_resent_request found one of the
4637 lock handles but later failed with ASSERTION(granted_lock != NULL).
4642 Description: handle->h_buffer_credits > 0, assertion failure
4643 Details : h_buffer_credits is zero after truncate, causing assertion
4644 failure. This patch extends the transaction or creates a new
4648 Frequency : NFS re-export or patchless client
4649 Bugzilla : 11179, 10796
4650 Description: Crash on NFS re-export node (__d_move)
4651 Details : We do not want to hash the dentry if we don't have a lock.
4652 But if this dentry is later used in d_move, we'd hit uninitialised
4653 list head d_hash, so we just do this to init d_hash field but
4654 leave dentry unhashed.
4657 Frequency : NFS re-export or patchless client
4659 Description: NFS exports has problem with symbolic link
4660 Details : lustre client didn't properly install dentry when re-exported
4661 to NFS or running patchless client.
4664 Frequency : NFS re-export or patchless client
4666 Description: Various nfs/patchless fixes.
4667 Details : fixes reuse disconected alias for lookup process - this fixes
4668 warning "find_exported_dentry: npd != pd",
4669 fix permission error with open files at nfs.
4670 fix apply umask when do revalidate.
4673 Frequency : occasional
4675 Description: Crash on NFS re-export node
4676 Details : calling clear_page() on the wrong pointer triggered oops in
4677 generic_mapping_read().
4680 Frequency : rarely, using O_DIRECT IO
4682 Description: unaligned directio crashes client with LASSERT
4683 Details : check for unaligned buffers before trying any requests.
4686 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
4688 Description: stale data returned from RAID cache
4689 Details : If only a small amount of IO is done to the RAID device before
4690 reading it again it is possible to get stale data from the RAID
4691 cache instead of reading it from disk.
4694 Frequency : always for sles10 kernel
4696 Description: sles10 support
4697 Details : ll_follow_link: compile fixes and using of nd_set_link
4698 under newer kernels.
4701 Frequency : depends on arch, kernel and compiler version, always on sles10
4704 Description: recursive or deep enough symlinks cause stack overflow
4705 Details : getting rid of large stack-allocated variable in
4709 Frequency : depends on hardware
4711 Description: lustre write performance loss in the SLES10 kernel
4712 Details : the performance loss is caused by using of write barriers in the
4713 ext3 code. The SLES10 kernel turns barrier support on by
4714 default. The fix is to undo that change for ldiskfs.
4716 ------------------------------------------------------------------------------
4718 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
4720 * Support for kernels:
4721 2.6.9-42.0.3EL (RHEL 4)
4722 2.6.5-7.276 (SLES 9)
4723 2.4.21-47.0.1.EL (RHEL 3)
4724 2.6.12.6 vanilla (kernel.org)
4728 Frequency : quota enabled and large files being deleted
4730 Description: releasing more than 4GB of quota at once hangs OST
4731 Details : If a user deletes more than 4GB of files on a single OST it
4732 will cause the OST to spin in an infinite loop. Release
4733 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
4738 Description: statfs data retrieved from /proc may be stale or zero
4739 Details : When reading per-device statfs data from /proc, in the
4740 {kbytes,files}_{total,free,avail} files, it may appear
4741 as zero or be out of date.
4744 Frequency : systems with MD RAID1 external journal devices
4746 Description: lconf's call to blkid is confused by RAID1 journal devices
4747 Details : Use the "blkid -l" flag to locate the MD RAID device instead
4748 of returning all block devices that match the journal UUID.
4751 Frequency : always, for aggregate stripe size over 4GB
4753 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
4754 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
4755 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
4758 Frequency : always if "lfs find" used on a local file/directory
4760 Description: "lfs find" segfaults if used on a local file/directory
4761 Details : The case where a directory component was not specified wasn't
4762 handled correctly. Handle this properly.
4765 Frequency : always on ppc64
4767 Description: the write to an ext3 filesystem mounted with mballoc got stuck
4768 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
4769 perform endianness conversion.
4772 Frequency : rarely (truncate to non-zero file size after write under load)
4773 Bugzilla : 10730, 10687
4774 Description: Files padded with zeros to next 4K multiple
4775 Details : With filesystems mounted using the "extents" option (2.6 kernels)
4776 it is possible that files that are truncated to a non-zero size
4777 immediately after being written are filled with zero bytes beyond
4778 the truncated size. No file data is lost.
4780 Severity : enhancement
4782 Description: Allow recovery/failover for liblustre clients.
4783 Details : liblustre clients were unaware of failover configurations until
4786 Severity : enhancement
4788 Description: user file locks should fail when not mounting with flock option
4789 Details : Set up an error-returning stub in ll_file_operations.lock field
4790 to prevent incorrect behaviour when client is mounted without
4791 flock option. Also, set up properly f_op->flock field for
4795 Frequency : always on ia64
4797 Description: "lfs df" loops on printing out MDS statfs information
4798 Details : The obd_ioctl_data was not initialized and in some systems
4799 this caused a failure during the ioctl that did not return
4800 an error. Initialize the struct and return an error on failure.
4803 Frequency : SLES 9 only
4805 Description: Error of copying files with lustre special EAs as root
4806 Details : Client side always return success for setxattr call for lustre
4807 special xattr (currently only "trusted.lov").
4810 Frequency : rarely on clusters with both ia64+i386 clients
4812 Description: ia64+i686 clients doing shared IO on the same file may LBUG
4813 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
4814 clients are doing concurrent writes to the same file it is
4815 possible that the ia64 clients may LASSERT because the OST
4816 extent locks are not PAGE_SIZE aligned. Ensure that grown
4817 locks are always aligned on the request boundary.
4820 Frequency : specific use, occasional
4822 Description: Overwriting in use executable truncates on-disk binary image
4823 Details : If one node attempts to overwrite an executable in use by
4824 another node, we now correctly return ETXTBSY instead of
4825 truncating the file.
4827 Severity : enhancement
4829 Description: Async OSC create to avoid the blocking unnecessarily.
4830 Details : If a OST has no remain object, system will block on the creating
4831 when need to create a new object on this OST. Now, ways use
4832 pre-created objects when available, instead of blocking on an
4833 empty osc while others are not empty. If we must block, we block
4834 for the shortest possible period of time.
4839 Description: chmod on Lustre root is propagated to other clients
4840 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
4841 invalid by the follow_mount time.
4844 Frequency : liblustre clients only
4846 Description: Race in 'instant cancel' lock handling could lead to such locks
4847 never to be granted in case of SMP MDS
4848 Details : Do not destroy not yet granted but cbpending locks in
4852 Frequency : replay/resend of open
4854 Description: non null lock assetion failure in mds_intent_policy
4855 Details : Trying to replay/resend lockless open requests resulted in
4856 mds_open() returning 0 with no lock. Now it sets a flag if
4857 a lock is going to be returned.
4859 Severity : enhancement
4861 Description: Checksum enhancements
4862 Details : New checksum enhancements allow for resending RPCs that failed
4865 Severity : enhancement
4867 Description: Tunables on number of dirty pages in cacche
4868 Details : Allow to set limit on number of dirty pages cached.
4873 Description: client crash on unmount - lock still has references
4874 Details : In some error handling cases it was possible to leak a lock
4875 reference on a client while accessing a file. This was not
4876 harmful to the client during operation, but would cause the
4877 client to crash when the filesystem is unmounted.
4880 Frequency : specific case, rare
4882 Description: ETXTBSY on mds though file not in use by client
4883 Details : ETXTBSY is no longer incorrectly returned when attempting to
4884 chmod or chown a directory that the user previously tried to
4885 execute or a currently-executing binary.
4888 Frequency : extremely rare except on liblustre-based clients
4890 Description: Lustre space not freed when files are deleted
4891 Details : Clean up open-unlinked files after client eviction. Previously
4892 the unlink was skipped and the files remained as orphans.
4897 Description: OST failure "would be an LBUG" in waiting_locks_callback()
4898 Details : In some cases it was possible to send a blocking callback to
4899 a client doing a glimpse, even though that client didn't get
4900 a lock granted. When the glimpse lock is cancelled on the OST
4901 the freed lock is left on the waiting list and corrupted the list.
4904 Frequency : all core dumps
4906 Description: Broke core dumps to lustre
4907 Details : Negative dentry may be unhashed if parent does not have UPDATE
4908 lock, but some callers, e.g. do_coredump, expect dentry to be
4909 hashed after successful create, hash it in ll_create_it.
4911 ------------------------------------------------------------------------------
4913 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
4915 * Support for kernels:
4916 2.6.9-42.0.2.EL (RHEL 4)
4917 2.6.5-7.276 (SLES 9)
4918 2.4.21-40.EL (RHEL 3)
4919 2.6.12.6 vanilla (kernel.org)
4923 Frequency : always on RHEL 3
4925 Description: Number of open files grows over time
4926 Details : The number of open files grows over time, whether or not
4927 Lustre is started. This was due to a filp leak introduced
4928 by one of our kernel patches.
4930 ------------------------------------------------------------------------------
4932 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
4934 * Support for kernels:
4935 2.6.9-42.EL (RHEL 4)
4936 2.6.5-7.267 (SLES 9)
4937 2.4.21-40.EL (RHEL 3)
4938 2.6.12.6 vanilla (kernel.org)
4943 Bugzilla : 5719, 9635, 9792, 9684
4944 Description: OST (or MDS) trips assertions in (re)connection under heavy load
4945 Details : If a server is under heavy load and cannot reply to new
4946 connection requests before the client resends the (re)connect,
4947 the connection handling code can behave badly if two service
4948 threads are concurrently handing separate (re)connections from
4949 the same client. Add better locking to the connection handling
4950 code, and ensure that only a single connection will be processed
4951 for a given client UUID, even if the lock is dropped.
4953 Severity : enhancement
4955 Description: add TCP zero-copy support to kernel
4956 Details : Add support to the kernel TCP stack to allow zero-copy bulk
4957 sends if the hardware supports scatter-gather and checksumming.
4958 This allows socklnd to do client-write and server-read more
4959 efficiently and reduce CPU utilization from skbuf copying.
4962 Frequency : only if NFS exporting from client
4964 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
4965 Details : If setting attributes on a file created under NFS that had
4966 never been opened it would be possible to oops the client
4967 if the file had no objects.
4971 Bugzilla : 9326, 10402, 10897
4972 Description: client crash in ptlrpcd_wake() thread when sending async RPC
4973 Details : It is possible that ptlrpcd_wake() dereferences a freed async
4974 RPC. In rare cases the ptlrpcd thread alread processed the RPC
4975 before ptlrpcd_wake() was called and the request was freed.
4978 Frequency : always for liblustre
4980 Description: liblustre client does MDS+OSTs setattr RPC for each write
4981 Details : When doing a write from a liblustre client, the client
4982 incorrectly issued an RPC to the MDS and each OST the file was
4983 striped over in order to update the timestamps. When writing
4984 with small chunks and many clients this could overwhelm the MDS
4985 with RPCs. In all cases it would slow down the write because
4986 these RPCs are unnecessary.
4988 Severity : enhancement
4990 Description: allow number of MDS service threads to be changed at module load
4991 Details : It is now possible to change the number of MDS service threads
4992 running. Adding "options mds mds_num_threads={N}" to the MDS's
4993 /etc/modprobe.conf will set the number of threads for the next
4994 time Lustre is restarted (assuming the "mds" module is also
4995 reloaded at that time). The default number of threads will
4996 stay the same, 32 for most systems.
5001 Description: OST crash if filesystem is unformatted or corrupt
5002 Details : If an OST is started on a device that has never been formatted
5003 or if the filesystem is corrupt and cannot even mount then the
5004 error handling cleanup routines would dereference a NULL pointer.
5009 Description: NULL pointer deref in llap_from_page.
5010 Details : get_cache_page_nowait can return a page with NULL (or otherwise
5011 incorrect) mapping if the page was truncated/reclaimed while it was
5012 searched for. Check for this condition and skip such pages when
5013 doing readahead. Introduce extra check to llap_from_page() to
5014 verify page->mapping->host is non-NULL (so page is not anonymous).
5017 Frequency : Sometimes when using sys_sendfile
5019 Description: "page not covered by a lock" warnings from ll_readpage
5020 Details : sendfile called ll_readpage without right page locks present.
5021 Now we introduced ll_file_sendfile that does necessary locking
5022 around call to generic_file_sendfile() much like we do in
5026 Frequency : with certain MDS communication failures at client mount time
5028 Description: NULL pointer deref after failed client mount
5029 Details : a client connection request may delayed by the network layer
5030 and not be sent until after the PTLRPC layer has timed out the
5031 request. If the client fails the mount immediately it will try
5032 to clean up before the network times out the request. Add a
5033 reference from the request import to the obd device and delay
5034 the cleanup until the network drops the request.
5037 Frequency : occasionally during client (re)connect
5039 Description: assertion failure during client (re)connect
5040 Details : processing a client connection request may be delayed by the
5041 client or server longer than the client connect timeout. This
5042 causes the client to resend the connection request. If the
5043 original connection request is replied in this interval, the
5044 client may trip an assertion failure in ptlrpc_connect_interpret()
5045 which thought it would be the only running connect process.
5048 Frequency : only with obd_echo servers and clients that are rebooted
5050 Description: kernel BUG accessing uninitialized data structure
5051 Details : When running an obd_echo server it did not start the ping_evictor
5052 thread, and when a client was evicted an uninitialized data
5053 structure was accessed. Start the ping_evictor in the RPC
5054 service startup instead of the OBD startup.
5056 Severity : enhancement
5057 Bugzilla : 10193 (patchless)
5058 Description: Remove dependency on various unexported kernel interfaces.
5059 Details : No longer need reparent_to_init, exit_mm, exit_files,
5060 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
5063 Frequency : rare (only users of deprecated and unsupported LDAP config)
5065 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
5066 Details : LDAP apparently contains 'lustreName' attributes instead of
5067 'name'. A simple remapping of the name is sufficient.
5070 Frequency : rare (only with non-default dump_on_timeout debug enabled)
5072 Description: waiting_locks_callback trips kernel BUG if client is evicted
5073 Details : Running with the dump_on_timeout debug flag turned on makes
5074 it possible that the waiting_locks_callback() can try to dump
5075 the Lustre kernel debug logs from an interrupt handler. Defer
5076 this log dumping to the expired_lock_main() thread.
5078 Severity : enhancement
5080 Description: Support NFS exporting on 2.6 kernels.
5081 Details : Implement non-rawops metadata methods for NFS server to use without
5082 changing NFS server code.
5085 Frequency : very rare (synthetic metadata workload only)
5087 Description: two racing renames might cause an MDS thread to deadlock
5088 Details : Running the "racer" program may cause one MDS thread to rename
5089 a file from being the source of a rename to being the target of
5090 a rename at exactly the same time that another thread is doing
5091 so, and the second thread has already enqueued these locks after
5092 doing a lookup of the target and is trying to relock them in
5093 order. Ensure that we don't try to re-lock the same resource.
5096 Frequency : only very large systems with liblustre clients
5098 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
5099 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
5100 all OSTs in parallel. This allows the eviction of stale liblustre
5101 clients to proceed much faster than if they were done in series,
5102 and also offers similar improvements for other set_info RPCs.
5107 Description: excessive CPU usage during initial read phase on client
5108 Details : During the initial read phase on a client, it would agressively
5109 retry readahead on the file, consuming too much CPU and impacting
5110 performance (since 1.4.5.8). Improve the readahead algorithm
5111 to avoid this, and also improve some other common cases (read
5112 of small files in particular, where "small" is files smaller than
5113 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
5118 Description: MDS crash when receiving packet with unknown intent.
5119 Details : Do not LBUG in unknown intent case, just return -EFAULT
5121 Severity : enhancement
5122 Bugzilla : 9293, 9385
5123 Description: MDS RPCs are serialised on client. This is unnecessary for some.
5124 Details : Do not serialize getattr (non-intent version) and statfs.
5127 Frequency : occasional, when OST network is overloaded/intermittent
5129 Description: client evicted by OST after bulk IO timeout
5130 Details : If a client sends a bulk IO request (read or write) the OST
5131 may evict the client if it is unresposive to its data GET/PUT
5132 request. This is incorrect if the network is overloaded (takes
5133 too long to transfer the RPC data) or dropped the OST GET/PUT
5134 request. There is no need to evict the client at all, since
5135 the pinger and/or lock callbacks will handle this, and the
5136 client can restart the bulk request.
5139 Frequency : Always when mmapping file with no objects
5141 Description: client crashes when mmapping file with no objects
5142 Details : Check that we actually have objects in a file before doing any
5143 operations on objects in ll_vm_open, ll_vm_close and
5149 Description: Request leak when working with deleted CWD
5150 Details : Introduce advanced request refcount tracking for requests
5151 referenced from lustre intent.
5153 Severity : Enhancement
5155 Description: Cache open file handles on client.
5156 Details : MDS now will return special lock along with openhandle, if
5157 requested and client is allowed to hold openhandle, even if unused,
5158 until such a lock is revoked. Helps NFS a lot, since NFS is opening
5159 closing files for every read/write openration.
5161 Severity : Enhancement
5163 Description: Cache open negative dentries on client when possible.
5164 Details : Guard negative dentries with UPDATE lock on parent dir, drop
5165 negative dentries on lock revocation.
5170 Description: Remounting a client read-only wasn't possible with a zconf mount
5171 Details : It wasn't possible to remount a client read-only with llmount.
5173 Severity : enhancement
5174 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
5175 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
5176 MPICH is included to add Lustre-specific ADIO interfaces.
5177 This is based closely on the UFS ADIO layer and only differs
5178 in file creation, in order to allow the OST striping to be set.
5179 This is user-contributed code and not supported by CFS.
5184 Description: extended inode attributes (immutable, append-only) work improperly
5185 when 2.4 and 2.6 kernels are used on client/server or vice versa
5186 Details : Introduce kernel-independent values for these flags.
5188 Severity : enhancement
5191 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
5192 Details : Many of the /proc/ tunables can only be tuned at a megabyte
5193 granularity. Now, Fractional MB granularity is be supported,
5194 this is very useful for low memory system.
5196 Severity : enhancement
5198 Description: Getattr by fid
5199 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
5200 locks, avoids extra getattr rpc requests to MDS, allows '/' to
5201 have locks and avoids getattr rpc requests for it on every stat.
5204 Frequency : Always, for filesystems larger than 2TB
5206 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
5207 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
5208 filesystems larger than 2TB to be mounted with mballoc enabled.
5211 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
5213 Description: filesystem corruption for non-standard kernels and very large OSTs
5214 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
5215 filesystems larger than 2TB is mounted then the kernel will
5216 silently corrupt the start of the filesystem. CONFIG_LBD is
5217 enabled for all CFS-supported kernels, but the possibility of
5218 this happening with a modified kernel config exists.
5220 Severity : enhancement
5222 Description: add client O_DIRECT support for 2.6 kernels
5223 Details : It is now possible to do O_DIRECT reads and writes to files
5224 in the Lustre client mountpoint on 2.6 kernel clients.
5226 Severity : enhancement
5228 Description: parallel glimpse, setattr, statfs, punch, destroy requests
5229 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
5230 parallel, not waiting for response from every OST before sending
5231 a rpc to the next OST.
5236 Description: setattr vs write race when updating file timestamps
5237 Details : Client processes that update a file timestamp into the past
5238 right after writing to the file (e.g. tar) it is possible that
5239 the updated file modification time can be reset to the current
5240 time due to a race between processing the setattr and write RPC.
5242 Severity : enhancement
5244 Description: Bring 'lfs find' closer in line with regular Linux find.
5245 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
5246 -print0 options and obtains all the needed info through the lustre
5249 Severity : enhancement
5251 Description: support up to 1024 configured devices on one node
5252 Details : change obd_dev array from statically allocated to dynamically
5253 allocated structs as they are first used to reduce memory usage
5258 Description: Flush dirty partially truncated pages during truncate
5259 Details : Immediatelly flush partially truncated pages in filter_setattr,
5260 this way we completely avoid having any pages in page cache on OST
5261 and can retire ugly workarounds during writes to flush such pages.
5266 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
5267 Details : protect inode from truncation within vfs_unlink() context
5268 just take a reference before calling vfs_unlink() and release it
5269 when parent's i_sem is free.
5272 Frequency : always, if extents are used on OSTs
5274 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
5275 Details : OSTs using the extents format would not zero the high 16 bits of
5276 the index physical block number. This is not a problem for any
5277 OST filesystems smaller than 16TB, and no kernels support ext3
5278 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
5279 new/modified files) and can be fixed for existing filesystems
5280 with e2fsprogs-1.39-cfs1.
5285 Description: import connection selection may be incorrect if timer wraps
5286 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
5287 import connections to be ignored if the 32-bit jiffies counter
5288 wraps. Use a 64-bit jiffies counter.
5291 Frequency : during server recovery
5293 Description: crash after server is denying duplicate export
5294 Details : If clients are resending connect requests to the server, the
5295 server refuses to allow a client to connect multiple times.
5296 Fixed a bug in the handling of this case.
5299 Frequency : very large clusters immediately after boot
5301 Description: LNET request buffers exhausted under heavy short-term load
5302 Details : If a large number of client requests are generated on a service
5303 that has previously never seen so many requests it is possible
5304 that the request buffer growth cannot keep up with the spike in
5305 demand. Instead of dropping incoming requests, they are held in
5306 the LND until the RPC service can accept more requests.
5309 Frequency : Sometimes during replay
5311 Description: Assertion failure in ll_local_open after replay.
5312 Details : If replay happened on an open request reply before we were able
5313 to set replay handler, reply will become not swabbed tripping the
5314 assertion in ll_local_open. Now we set the handler right after
5315 recognising of open request
5318 Frequency : very rare
5320 Description: kernel reports "badness in vsnprintf"
5321 Details : Reading from the "recovery_status" /proc file in small chunks
5322 may cause a negative length in lprocfs_obd_rd_recovery_status()
5323 call to vsnprintf() (which is otherwise harmless). Exit early
5324 if there is no more space in the output buffer.
5326 Severity : enhancement
5328 Description: clear OBD RPC statistics by writing to them
5329 Details : It is now possible to clear the OBD RPC statistics by writing
5330 to the "stats" file.
5335 Description: Client mtime is not the same on different clients after utimes
5336 Details : In some cases, the client was using the utimes() syscall on
5337 a file cached on another node. The clients now validate the
5338 ctime from the MDS + OSTs to determine which one is right.
5343 Description: Inability to activate failout mode
5344 Details : lconf script incorrectly assumed that in python string's numeric
5345 value is used in comparisons.
5348 Frequency : always with multiple stripes per file
5350 Description: Inefficient object allocation for mutli-stripe files
5351 Details : When selecting which OSTs to stripe files over, for files with
5352 a stripe count that divides evenly into the number of OSTs,
5353 the MDS is always picking the same starting OST for each file.
5354 Return the OST selection heuristic to the original design.
5359 Description: mount failures may take full timeout to return an error
5360 Details : Under some heavy load conditions it is possible that a
5361 failed mount can wait for the full obd_timeout interval,
5362 possibly several minutes, before reporting an error.
5363 Instead return an error as soon as the status is known.
5365 ------------------------------------------------------------------------------
5367 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
5369 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
5370 INTEROPERATE with older versions automatically. Please read the
5371 user documentation before upgrading any part of a live system.
5372 * WARNING: Lustre networking configuration changes are required with
5373 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
5376 * Support for kernels:
5377 2.6.9-22.0.2.EL (RHEL 4)
5378 2.6.5-7.244 (SLES 9)
5379 2.6.12.6 vanilla (kernel.org)
5382 Severity : enhancement
5383 Bugzilla : 7981/8208
5384 Description: Introduced Lustre Networking (LNET)
5385 Details : LNET is new networking infrastructure for Lustre, it includes
5386 a reorganized network configuration mode (see the user
5387 documentation for full details) as well as support for routing
5388 between different network fabrics. Lustre Networking Devices
5389 (LNDs) for the supported network fabrics have also been
5390 created for this new infrastructure.
5392 Severity : enhancement
5393 Description: Introduced Access control lists
5394 Details : clients can set ACLs on files and directories in order to have
5395 more fine-grained permissions than the standard Unix UGO+RWX.
5396 The MDS must be started with the "-o acl" mount option.
5398 Severity : enhancement
5399 Description: Introduced filesystem quotas
5400 Details : Administrators may now establish per-user quotas on the
5403 Severity : enhancement
5405 Description: Configuration change for the XT3
5406 The PTLLND is now used to run Lustre over Portals on the XT3
5407 The configure option(s) --with-cray-portals are no longer used.
5408 Rather --with-portals=<path-to-portals-includes> is used to
5409 enable building on the XT3. In addition to enable XT3 specific
5410 features the option --enable-cray-xt3 must be used.
5415 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
5416 Details : A race between allocating a new llcd and re-getting the llcd_lock
5417 allowed another thread to grab newly-allocated llcd.
5419 Severity : enhancement
5421 Description: 2.6 OST async journal commit and locking fix to improve performance
5422 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
5423 2.6 kernels are now async as they already were in 2.4 kernels so
5424 that they can commit concurrently with the network bulk transfer.
5425 For block-allocated files the filter allocation semaphore is held
5426 to avoid filesystem fragmentation during allocation. BKL lock
5427 removed for 2.6 xattr operations where it is no longer needed.
5432 Description: lconf incorrectly determined whether two IP networks could talk
5433 Details : In some more complicated routing and multiple-network
5434 configurations, lconf will avoid trying to make a network
5435 connection to a disjoint part of the IP space. It was doing the
5436 math incorrectly for one set of cases.
5441 Description: Fix for potential infinite loop processing records in an llog.
5442 Details : If an llog record is corrupted/zeroed, it is possible to loop
5443 forever in llog_process(). Validate the llog record length
5444 and skip the remainder of the block on error.
5447 Frequency : occasional (liblustre only)
5449 Description: liblustre could not open files whose last component is a symlink
5450 Details : sysio_path_walk() would incorrectly pass the open intent to
5451 intermediate path components.
5454 Frequency : rare (liblustre only with non-standard tuning)
5455 Bugzilla : 7201 (7350)
5456 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
5457 Details : llu_lookup_finish_locks() tries to set lock data on a lock
5458 after it has been released, only do this for referenced locks
5460 Severity : enhancement
5462 Description: specifying an (invalid) directory default stripe_size of -1
5463 would reset the directory default striping
5464 Details : stripe_size -1 was used internally to signal directory stripe
5465 removal, now use "all default" to signal dir stripe removal
5466 as a directory striping of "all default" is not useful
5469 Frequency : common for large clusters running liblustre clients
5471 Description: doing an ls when liblustre clients are running is slow
5472 Details : sending a glimpse AST to a liblustre client waits for every AST
5473 to time out, as liblustre clients will not respond. Since they
5474 cannot cache data we refresh the OST lock LVB from disk instead.
5476 Severity : enhancement
5478 Description: doing an ls at the same time as file IO can be slow
5479 Details : enqueue and other "small" requests can be blocked behind many
5480 large IO requests. Create a new OST IO portal for non-IO
5481 requests so they can be processed faster.
5484 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
5486 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
5487 created file with mode 000
5488 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
5489 allow an ftruncate() as a non-root user to a file with mode 000.
5490 Lustre now respects this flag to disable mode checks when
5491 truncating a file owned by the user
5494 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
5497 Description: Revoking locks from clients that went dead or catatonic might take
5499 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
5500 cancellation of such locks instant on servers without waiting for
5501 any reply from clients. Clients drops these locks when cancel
5502 notification from server is received without replying.
5505 Frequency : liblustre-only, when liblustre client dies or becomes busy
5507 Description: Doing ls on Linux clients can take a long time with active
5509 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
5510 granting such locks to it in the first place if possible. Locks
5511 are taken by proxy on the OST during the read or write and
5512 dropped immediately afterward. Add connect flags handling, do
5513 not grant locks to liblustre clients for glimpse ASTs.
5515 Severity : enhancement
5517 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
5518 Details : Existing read-ahead algorithm is tuned for the case of streamlined
5519 sequential reads and behaves badly with applications doing random
5520 reads. Improve it by reading ahead at least read region, and
5521 avoiding excessive large RPC for small reads.
5523 Severity : enhancement
5525 Description: Creating more than 1000 files for a single job may cause a load
5526 imbalance on the OSTs if there are also a large number of OSTs.
5527 Details : qos_prep_create() uses an OST index reseed value that is an
5528 even multiple of the number of available OSTs so that if the
5529 reseed happens in the middle of the object allocation it will
5530 still utilize the OSTs as uniformly as possible.
5535 Description: OST or MDS may oops in ping_evictor_main()
5536 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
5537 but doesn't restart at beginning of obd_exports_timed list
5540 Severity : enhancement
5542 Description: improve by-nid export eviction on the MDS and OST
5543 Details : allow multiple exports with the same NID to be evicted at one
5544 time without re-searching the exports list.
5547 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
5549 Description: MDS may oops in groups_free()
5550 Details : in rare race conditions a newly allocated group_info struct is
5551 freed again, and this can be NULL. The 2.4 compatibility code
5552 for groups_free() checked for a NULL pointer, but 2.6 did not.
5555 Frequency : common for liblustre clients doing little filesystem IO
5556 Bugzilla : 9352, 7313
5557 Description: server may evict liblustre clients accessing contended locks
5558 Details : if a client is granted a lock or receives a completion AST
5559 with a blocking AST already set it would not reply to the AST
5560 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
5563 Frequency : lfs setstripe, only systems with more than 160 OSTs
5565 Description: unable to set striping with a starting offset beyond OST 160
5566 Details : llapi_create_file() incorrectly limited the starting stripe
5567 index to the maximum single-file stripe count.
5570 Frequency : LDAP users only
5572 Description: lconf did not handle in-kernel recovery with LDAP properly
5573 Details : lconf/LustreDB get_refs() is searching the wrong namespace
5575 Severity : enhancement
5577 Description: bind OST threads to NUMA nodes to improve performance
5578 Details : all OST threads are uniformly bound to CPUs on a single NUMA
5579 node and do their allocations there to localize memory access
5581 Severity : enhancement
5583 Description: llmount can determine client NID directly from Myrinet (GM)
5584 Details : the client NID code from gmnalnid was moved directly into
5585 llmount, removing the need to use this or specifying the
5586 client NID explicitly when mounting GM clients with zeroconf
5589 Frequency : if client is started with down MDS
5591 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
5592 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
5593 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
5594 ptlrpc_import_delayed_req() because we didn't check if the
5595 request was interrupted, and we also didn't break out of the
5596 event loop if there was no timeout
5601 Description: data loss during non-page-aligned writes to a single file from
5602 both multiple nodes and multiple threads on one node at same time
5603 Details : updates to KMS and lsm weren't protected by common lock. Resulting
5604 inconsistency led to false short-reads, that were cached and later
5605 used by ->prepare_write() to fill in partially written page,
5606 leading to data loss.
5609 Frequency : always, if lconf --abort_recovery used
5611 Description: lconf --abort_recovery fails with 'Operation not supported'
5612 Details : lconf was attempting to abort recovery on the MDT device and not
5615 Severity : enhancement
5617 Description: remove cleanup logs
5618 Details : replace lconf-generated cleanup logs with lustre internal
5619 cleanup routines. Eliminates the need for client-cleanup and
5622 Severity : enhancement
5624 Description: add support for EAs (user and system) on lustre filesystems
5625 Details : it is now possible to store extended attributes in the Lustre
5626 client filesystem, and with the user_xattr mount option it
5627 is possible to allow users to store EAs on their files also
5629 Severity : enhancement
5631 Description: Add possibility (config option) to show minimal available OST free
5633 Details : When compiled with --enable-mindf configure option, statfs(2)
5634 (and so, df) will return least minimal free space available from
5635 all OSTs as amount of free space on FS, instead of summary of
5636 free spaces of all OSTs.
5638 Severity : enhancement
5640 Description: do not expand extent locks acquired on OST-side
5641 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
5642 by server: they are not cached anyway.
5645 Frequency : when mmap is used/binaries executed from Lustre
5647 Description: Unmmap pages before throwing them away from read cache.
5648 Details : llap_shrink cache now attempts to unmap pages before discarding
5649 them (if unmapping failed - do not discard). SLES9 kernel has
5650 extra checks that trigger if this unmapping is not done first.
5655 Description: lconf didn't resolve symlinks before checking to see whether a
5656 given mountpoint was already in use
5659 Frequency : when migrating failover services
5660 Bugzilla : 6395, 9514
5661 Description: When migrating a subset of services from a node (e.g. failback
5662 from a failover service node) the remaining services would
5663 time out and evict clients.
5664 Details : lconf --force (implied by --failover) sets the global obd_timeout
5665 to 5 seconds in order to quickly disconnect, but this caused
5666 other RPCs to time out too quickly. Do not change the global
5667 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
5669 Severity : enhancement
5670 Frequency : if MDS is started with down OST
5671 Bugzilla : 9439,5706
5672 Description: Allow startup/shutdown of an MDS without depending on the
5673 availability of the OSTs.
5674 Details : Asynchronously call mds_lov_synchronize during MDS startup.
5675 Add appropriate locking and lov-osc refcounts for safe
5676 cleaning. Add osc abort_inflight calls in case the
5677 synchronize never started.
5680 Frequency : occasional (Cray XT3 only)
5682 Description: root not authorized to access files in CRAY_PORTALS environment
5683 Details : The client process capabilities were not honoured on the MDS in
5684 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
5685 been accessed by an authorized user then root was able to access
5686 the file on the local client also. The root user capabilities
5687 are now allowed on the MDS, as this environment has secure UID.
5690 Frequency : occasional
5692 Description: ldiskfs "too long searching" message happens too often
5693 Details : A debugging message (otherwise harmless) prints too often on
5694 the OST console. This has been reduced to only happen when
5695 there are fragmentation problems on the filesystem.
5700 Description: Division by zero in statfs when all OSCs are inactive
5701 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
5702 lov_statfs divides by value returned by lov_get_stripecnt().
5706 Bugzilla : 9489, 3273
5707 Description: First write from each client to each OST was only 4kB in size,
5708 to initialize client writeback cache, which caused sub-optimal
5709 RPCs and poor layout on disk for the first writen file.
5710 Details : Clients now request an initial cache grant at (re)connect time
5711 and so that they can start streaming writes to the cache right
5712 away and always do full-sized RPCs if there is enough data.
5713 If the OST is rebooted the client also re-establishes its grant
5714 so that client cached writes will be honoured under the grant.
5719 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
5720 Details : Now I/O RPCs go to different portal number and (presumably) fast
5721 lock requests (and glimses) and other RPCs get their own service
5722 threads pool that should be able to service those RPCs
5725 Severity : enhancement
5727 Description: Ability to exchange lustre version between client and servers and
5728 issue warnings at client side if client is too old. Also for
5729 liblustre clients there is ability to refuse connection of too old
5731 Details : New 'version' field is added to connect data structure that is
5732 filled with version info. That info is later checked by server and
5736 Frequency : rare, liblustre only.
5737 Bugzilla : 9296, 9581
5738 Description: Two simultaneous writes from liblustre at offset within same page
5739 might proceed at the same time overwriting eachother with stale
5741 Details : I/O lock withing llu_file_prwv was released too early, before data
5742 actually was hitting the wire. Extended lock-holding time until
5743 server acknowledges receiving data.
5746 Frequency : extremely rare. Never observed in practice.
5748 Description: avoid generating lustre_handle cookie of 0.
5749 Details : class_handle_hash() generates handle cookies by incrementing
5750 global counter, and can hit 0 occasionaly (this is unlikely, but
5751 not impossible, because initial value of cookie counter is
5752 selected randonly). Value of 0 is used as a sentinel meaning
5753 "unassigned handle" --- avoid it. Also coalesce two critical
5754 sections in this function into one.
5756 Severity : enhancement
5758 Description: allow liblustre clients to delegate truncate locking to OST
5759 Details : To avoid overhead of locking, liblustre client instructs OST to
5760 take extent lock in ost_punch() on client's behalf. New connection
5761 flag is added to handle backward compatibility.
5763 Severity : enhancement
5764 Bugzilla : 4928, 7341, 9758
5765 Description: allow number of OST service threads to be specified
5766 Details : a module parameter allows the number of OST service threads
5767 to be specified via "options ost ost_num_threads={N}" in the
5768 OSS's /etc/modules.conf or /etc/modprobe.conf.
5772 Bugzilla : 6146, 9635, 9895
5773 Description: servers crash with bad pointer in target_handle_connect()
5774 Details : In rare cases when a client is reconnecting it was possible that
5775 the connection request was the last reference for that export.
5776 We would temporarily drop the export reference and get a new
5777 one, but this may have been the last reference and the export
5778 was just destroyed. Get new reference before dropping old one.
5780 Severity : enhancement
5781 Frequency : if client is started with failover MDS
5783 Description: Allow multiple MDS hostnames in the mount command
5784 Details : Try to read the configuration from all specified MDS
5785 hostnames during a client mount in case the "primary"
5788 Severity : enhancement
5790 Description: Stop sending data to evicted clients as soon as possible.
5791 Details : Check if the client we are about to send or are sending data to
5792 was evicted already. (Check is done every second of waiting,
5793 for which l_wait_event interface was extended to allow checking
5794 of exit condition at specified intervals).
5797 Frequency : rare, normally only when NFS exporting is done from client
5799 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
5801 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
5804 Frequency : always, if srand() is called before liblustre initialization
5806 Description: Liblustre uses system PRNG disturbing its usage by user application
5807 Details : Introduce internal to lustre fast and high-quality PRNG for
5808 lustre usage and make liblustre and some other places in generic
5809 lustre code to use it.
5811 Severity : enhancement
5812 Bugzilla : 9477, 9557, 9870
5813 Description: Verify that the MDS configuration logs are updated when xml is
5814 Details : Check if the .xml configuration logs are newer than the config
5815 logs stored on the MDS and report an error if this is the case.
5816 Request --write-conf, or allow starting with --old_conf.
5818 Severity : enhancement
5820 Description: Handle symlinks in the path when checking if Lustre is mounted.
5821 Details : Resolve intermediate symlinks when checking if a client has
5822 mounted a filesystem to avoid duplicate client mounts.
5827 Description: lconf can hit an error exception but still return success.
5828 Details : The lconf command catches the Command error exception at the top
5829 level script context and will exit with the associated exit
5830 status, but doesn't ensure that this exit status is non-zero.
5835 Description: failure of ptlrpc thread startup can cause oops
5836 Details : Starting a ptlrpc service thread can fail if there are a large
5837 number of threads or the server memory is very fragmented.
5838 Handle this without oopsing.
5841 Frequency : always, only if liblustre and non-default acceptor port was used
5843 Description: liblustre cannot connect to servers with non-default acceptor port
5844 Details : tcpnal_set_default_params() was not called and was therefore
5845 ignoring the environment varaible TCPNAL_PORT, as well as other
5846 TCPNAL_ environment variables
5851 Description: two objects could be created on the same OST for a single file
5852 Details : If an OST is down, in some cases it was possible to create two
5853 objects on a single OST for a single file. No problems other
5854 than potential performance impact and spurious error messages.
5858 Bugzilla : 5681, 9562
5859 Description: Client may oops in ll_unhash_aliases
5860 Details : Client dcache may become inconsistent in race condition.
5861 In some cases "getcwd" can fail if the current directory is
5867 Description: Inode refcounting problems in NFS export code
5868 Details : link_raw functions used to call d_instantiate without obtaining
5869 extra inode reference first.
5873 Bugzilla : 9942, 9903
5874 Description: Referencing freed requests leading to crash, memleaks with NFS.
5875 Details : We used to require that call to ll_revalidate_it was always
5876 followed by ll_lookup_it. Also with revalidate_special() it is
5877 possible to call ll_revalidate_it() twice for the same dentry
5878 even if first occurence returned success. This fix changes semantic
5879 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
5880 reference on a request referred from the intent.
5881 ll_intent_release() then releases such a request.
5884 Frequency : rare, normally benchmark loads only
5886 Description: unlinked inodes were kept in memory on the client
5887 Details : If a client is repeatedly creating and unlinking files it
5888 can accumulate a lot of stale inodes in the inode slab cache.
5889 If there is no other client load running this can cause the
5890 client node to run out of memory. Instead flush old inodes
5891 from client cache that have the same inode number as a new inode.
5894 Frequency : SLES9 2.6.5 kernel and long filenames only
5895 Bugzilla : 9969, 10379
5896 Description: utime reports stale NFS file handle
5897 Details : SLES9 uses out-of-dentry names in some cases, which confused
5898 the lustre dentry revalidation. Change it to always use the
5902 Frequency : rare, unless heavy write-truncate concurrency is continuous
5903 Bugzilla : 4180, 6984, 7171, 9963, 9331
5904 Description: OST becomes very slow and/or deadlocked during object unlink
5905 Details : filter_destroy() was holding onto the parent directory lock
5906 while truncating+unlinking objects. For very large objects this
5907 may block other threads for a long time and slow overall OST
5908 responsiveness. It may also be possible to get a lock ordering
5909 deadlock in this case, or run out of journal credits because of
5910 the combined truncate+unlink. Solution is to do object truncate
5911 first in one transaction without parent lock, and then do the
5912 final unlink in a new transaction with the parent lock. This
5913 reduces the lock hold time dramatically.
5916 Frequency : rare, 2.4 kernels only
5918 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
5919 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
5920 before unmounting their filesystem in order to avoid deadlock.
5921 The kernel_locked() function in 2.4 kernels only checks whether
5922 the kernel lock is held, not whether it is this process that is
5923 holding it as 2.6 kernels do.
5928 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
5929 Details : The client ptlrpc code may be trying to reconnect to a down
5930 server before a previous connection attempt has timed out.
5931 Increase the reconnect interval to be longer than the connection
5932 timeout interval to avoid sending duplicate connections to
5936 Frequency : echo_client brw_test command
5938 Description: fix echo_client to work with OST preallocated code
5939 Details : OST preallocation code (5137) didn't take echo_client IO path
5940 into account: echo_client calls filter methods outside of any
5941 OST thread and, hence, there is no per-thread preallocated
5942 pages and buffers to use. Solution: hijack pga pages for IO. As
5943 a byproduct, this avoids unnecessary data copying.
5947 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
5948 Description: Client can oops in mdc_commit_close() after open replay
5949 Details : It was possible for the MDS to return an open request with no
5950 transaction number in mds_finish_transno() if the client was
5951 evicted, but without actually returning an error. Clients
5952 would later try to replay that open and may trip an assertion
5953 Simplify the client close codepath, and always return an error
5954 from the MDS in case the open is not successful.
5957 Frequency : rare, 2.6 OSTs only
5959 Description: OST may deadlock under high load on fragmented files
5960 Details : If there was a heavy load and highly-fragmented OST filesystems
5961 it was possible to have all the OST threads deadlock waiting on
5962 allocation of biovecs, because the biovecs were not released
5963 until the entire RPC IO was completed. Instead, release biovecs
5964 as soon as they are complete to ensure forward IO progress.
5966 Severity : enhancement
5968 Description: Support for specifying external journal device at mount
5969 Details : If an OST or MDS device is formatted with an external journal
5970 device, this device major/minor is stored in the ext3 superblock
5971 and may not be valid for failover. Allow detecting and
5972 specifying the external journal at mount time.
5977 Description: Mounting an MDS with pending unlinked files may cause oops
5978 Details : target_finish_recovery() calls mds_postrecov() which returned
5979 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
5980 considers this an error and immediately begins cleaning up the
5981 lov, just after starting the mds_lov process
5983 Severity : enhancement
5985 Description: Implement 'lfs df' to report actual free space on per-OST basis
5986 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
5987 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
5988 usage of objects; '-h' to report in human readable format.
5990 ------------------------------------------------------------------------------
5992 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
5999 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
6000 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
6001 at startup during orphan unlinking. Instead, skip buddy update
6002 before setup, as it will be regenerated after recovery is complete.
6007 Description: If an OST is inactive, its locks might reference stale inodes.
6008 Details : lov_change_cbdata() must iterate over all namespaces, even if
6009 they are inactive to clear inode references from the lock.
6011 Severity : enhancement
6012 Frequency : occasional, if non-standard max_dirty_mb used
6014 Description: Client will block write RPCs if not enough grant
6015 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
6016 then the client will block writes while waiting for another RPC
6017 to complete instead of consuming its dirty limit. With change
6018 we get improved performance when max_dirty_mb is small.
6020 Severity : enhancement
6021 Bugzilla : 3389, 6253
6022 Description: Add support for supplementary groups on the MDS.
6023 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
6024 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
6025 lookups for user supplementary groups into a cache.
6029 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
6030 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
6031 for liblustre O_EXCL is enforced by MDS.
6035 Description: readdir never returns NULL in liblustre.
6036 Details : Corrected llu_iop_getdirentries logic, to return offset of next
6037 dentry in struct dirent.
6041 Frequency : liblustre only, depends on application IO pattern
6042 Description: liblustre clients evicted if not contacting servers
6043 Details : Don't put liblustre clients into the ping_evictor list, so
6044 they will not be evicted by the pinger ever.
6046 Severity : enhancement
6048 Description: Add ability to evict clients by NID from MDS.
6049 Details : By echoing "nid:$NID" string into
6050 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
6051 $NID would be instantly evicted from this MDS and from all active
6052 OSTs connected to it.
6056 Description: Do not query file size twice, somewhat slowing stat(2) calls.
6057 Details : lookup_it_finish() used to query file size from OSTs that was not
6062 Description: service threads change working directory to that of init
6063 Details : Starting lustre service threads may pin the working directory
6064 of the parent thread, making that filesystem busy. Threads
6065 now change to the working directory of init to avoid this.
6069 Frequency : during shutdown only
6070 Description: shutdown with a failed MDS or OST can cause unmount to hang
6071 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
6076 Frequency : 2.6 only
6077 Description: chmod/chown may include an extra supplementary group
6078 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
6079 supplementary group and if none is specified this is used.
6082 Bugzilla : 5479 (6816)
6084 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
6085 Details : If lookup is in progress on a file that is unlinked we might try
6086 to revalidate the inode and fail in revalidate after lookup is
6087 complete and ll_file_open() enqueues the open again but
6088 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
6091 Frequency : always, if lconf --abort_recovery used
6093 Description: lconf --abort_recovery fails with 'Operation not supported'
6094 Details : lconf was attempting to abort recovery on the MDT device and not
6097 ------------------------------------------------------------------------------
6099 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
6104 Frequency : rare (only unsupported configurations with a node running as an
6106 Bugzilla : 6514, 5137
6107 Description: Mounting a Lustre file system on a node running as an OST could
6109 Details : OSTs now preallocates memory needed to write out data at
6110 startup, instead of when needed, to avoid having to
6111 allocate memory in possibly low memory situations.
6112 Specifically, if the file system is mounted on on OST,
6113 memory pressure could force it to try to write out data,
6114 which it needed to allocate memory to do. Due to the low
6115 memory, it would be unable to do so and the node would
6116 become unresponsive.
6118 Severity : enhancement
6120 Description: Addition of lconf --service command line option
6121 Details : lconf now accepts a '--service <arg>' option, which is
6122 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
6124 Severity : enhancement
6126 Description: Failover mode is now the default for OSTs.
6127 Details : By default, OSTs will now run in failover mode. To return to
6128 the old behaviour, add '--failout' to the lmc line for OSTs.
6130 Severity : enhancement
6132 Description: Health checks are now provided for MDS and OSTs
6133 Details : Additional detailed health check information on MSD and OSTs
6134 is now provided through the procfs health_check value.
6137 Frequency : occasional, depends on IO load
6139 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
6140 after numerous create/delete cycles
6141 Details : The ext3 inode allocation policy would not allocate new inodes
6142 very well on the OSTs because there are no new directories
6143 being created. Instead we look for groups with free space if
6144 the parent directories are nearly full.
6149 Description: Network or server problems during mount may cause partially
6150 mounted clients instead of returning an error.
6151 Details : The config llog parsing code may overwrite the error return
6152 code during mount error handling, returning success instead
6158 Description: MDS can fail to allocate large reply buffers
6159 Details : After long uptimes the MDS can fail to allocate large reply
6160 buffers (e.g. zconf client mount config records) due to memory
6161 fragmentation or consumption by the buffer cache. Preallocate
6162 some large reply buffers so that these replies can be sent even
6163 under memory pressure.
6167 Frequency : rare (liblustre)
6168 Description: fsx running with liblustre complained that using truncate() to
6169 extend the file doesn't work. This patch corrects that issue.
6170 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
6171 ATTR_SIZE and lsm use in llu_setattr_raw.
6175 Frequency : rare, only 2.6 kernels
6176 Description: Unusual file access patterns on the MDS may result in inode
6177 data being lost in very rare circumstances.
6178 Details : Bad interaction between the ea-in-inode patch and the "no-read"
6179 code in the 2.6 kernel caused the inode and/or EA data not to
6180 be read from disk, causing single-file corruption.
6184 Frequency : rare, only 2.6 filesystems using extents
6185 Description: Heavy concurrent write and delete load may cause data corruption.
6186 Details : It was possible under high-load situations to have an extent
6187 metadata block in the block device cache from a just-unlinked
6188 file overwrite a newly-allocated data block. We now unmap any
6189 metadata buffers that alias just-allocated data blocks.
6193 Frequency : filesystems with default stripe_count larger than 77
6194 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
6195 Details : lconf specifies an inode size of 4096 bytes when the default
6196 stripe_count is larger than 77. This conflicts with the default
6197 inode density of 1 per 4096 bytes. Allocate smaller inodes in
6198 this case to avoid pinning too much memory for large EAs.
6200 ------------------------------------------------------------------------------
6202 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
6207 Frequency : rare (extremely heavy IO load with hundreds of clients)
6209 Description: Client is evicted, gets IO error writing to file
6210 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
6211 caused clients to be evicted for AST timeouts. The fixes in
6212 bug 5192 mean we no longer need to have such short AST timeouts
6213 so ldlm_timeout has been increased.
6216 Frequency : occasional during --force or --failover shutdown under load
6217 Bugzilla : 5949, 4834
6218 Description: Server oops/LBUG if stopped with --force or --failover under load
6219 Details : a collection of import/export refcount and cleanup ordering
6220 issues fixed for safer force cleanup
6223 Frequency : only filesystems larger than 120 OSTs
6224 Bugzilla : 5990, 6223
6225 Description: lfs getstripe would oops on a very large filesystem
6226 Details : lov_getconfig used kfree on vmalloc'd memory
6229 Frequency : only filesystems exporting via NFS to Solaris 10 clients
6230 Bugzilla : 6242, 6243
6231 Description: reading from files that had been truncated to a non-zero size
6232 but never opened returned no data
6233 Details : ll_file_read() reads zeros from no-object files to EOF
6238 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
6239 Details : There's little harm in aborting MDS/OSS recovery and letting it
6240 try again, so I removed the LASSERT and return an error instead.
6242 Severity : enhancement
6244 Description: New debugging infrastructure for tracking down data corruption
6245 Details : The I/O checksum code was replaced to: (a) control it at runtime,
6246 (b) cover more of the client-side code path, and (c) try to narrow
6247 down where problems occurred
6251 Bugzilla : 3819, 4364, 4397, 6313
6252 Description: Racing close and eviction MDS could cause assertion in mds_close
6253 Details : It was possible to get multiple mfd references during close and
6254 client eviction, leading to one thread referencing a freed mfd.
6256 Severity: : enhancement
6257 Bugzilla : 3262, 6359
6258 Description: Attempts to reconnect to servers are now more aggressive.
6259 Details : This builds on the enhanced upcall-less recovery that was added
6260 in 1.4.2. When trying to reconnect to servers, clients will
6261 now try each server in the failover group every 10 seconds. By
6262 default, clients would previously try one server every 25 seconds.
6267 Description: After recovery, certain operations trigger a failed
6268 assertion on a client.
6269 Details : Failing over an mds, using lconf -d --failover, while a
6270 client was doing a readdir() call would cause the client to
6271 LBUG after recovery completed and the readdir() was resent.
6273 Severity : enhancement
6275 Description: Default groups are now added by lconf
6276 Details : You can now run lconf --group <servicename> without having to
6277 manually add groups with lmc.
6280 Frequency : occasional
6282 Description: Nodes with an elan id of 0 trigger a failed assertion
6285 Frequency : always when accessing e.g. tty/console device nodes
6287 Description: tty and some other devices nodes cannot be used on lustre
6288 Details : file's private_data field is used by device data and lustre
6289 values in there got lost. New field was added to struct file to
6290 store fs-specific private data.
6293 Frequency : when exporting Lustre via NFS
6295 Description: NFSD failed occasionally when looking up a path component
6296 Details : NFSD is looking up ".." which was broken in ext3 directories
6297 that had grown large enough to become hashed.
6300 Frequency : Clusters with multiple interfaces not on the same subnet
6302 Description: Nodes will repeatedly try to reconnect to an interface which it
6303 cannot reach and report an error to the log.
6304 Details : Extra peer list entries will be created by lconf with some peers
6305 unreachable. lconf now validates the peer before adding it.
6308 Frequency : Only if a default stripe is set on the filesystem root.
6310 Description: Setting a default stripe on the filesystem root prevented the
6311 filesystem from being remounted.
6312 Details : The client was sending extra request flags in the root getattr
6313 request and did not allocate a reply buffer for the dir EA.
6316 Frequency : occasional, higher if lots of files are accessed by one client
6317 Bugzilla : 6159, 6097
6318 Description: Client trips assertion regarding lsm mismatch/magic
6319 Details : While revalidating inodes the VFS looks up inodes with ifind()
6320 and in rare cases can find an inode that is being freed.
6321 The ll_test_inode() code will free the lsm during ifind()
6322 when it finds an existing inode and then the VFS later attaches
6323 this free lsm to a new inode.
6327 Bugzilla : 6422, 7030
6328 Description: MDS deadlock between mkdir and client eviction
6329 Details : Creating a new file via mkdir or mknod (starting a transaction
6330 and getting the ns lock) can deadlock with client eviction
6331 (gets ns lock and trying to finish a synchronous transaction).
6334 Frequency : occasional
6335 Description: While starting a server, the fsfilt_ext3 module could not be
6337 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
6338 kernels. Previously, lconf would still use the ext3 name
6339 when trying to load modules. Now, it will correctly use
6340 ext3 on 2.4 and ldiskfs on 2.6.
6342 Severity : enhancement
6343 Description: The default stripe count has been changed to 1
6344 Details : The interpretation of the default stripe count (0, to lfs
6345 or lmc) has been changed to mean striping across a single
6346 OST, rather than all available. For general usage we have
6347 found a stripe count of 1 or 2 works best.
6349 Severity : enhancement
6350 Description: Add support for compiling against Cray portals.
6351 Details : Conditional compiling for some areas that are different
6355 Frequency : occasional
6356 Bugzilla : 6409, 6834
6357 Description: Creating files with an explicit stripe count may lead to
6358 a failed assertion on the MDS
6359 Details : If some OSTs are full or unavailable, creating files may
6360 trigger a failed assertion on the MDS. Now, Lustre will
6361 try to use other servers or return an error to the
6365 Frequency : occasional
6367 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
6368 caused lock timeout during readahead (since 1.4.2).
6369 Details : Processes doing readahead might match a lock that hasn't been
6370 granted yet if there are overlapping and conflicting lock
6371 requests. The readahead process waits on ungranted lock
6372 (original lock is CBPENDING), while OST waits for that process
6373 to cancel CBPENDING read lock and eventually evicts client.
6375 Severity : enhancement
6377 Description: Initial enabling of flock support for clients
6378 Details : Implements fcntl advisory locking and file status functions.
6379 This feature is provided as an optional mount flag (default
6380 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
6381 locking are implemented yet, and those that are are not guaranteed
6382 to be completely correct in production environments.
6383 mount -t lustre -o [flock|noflock] ...
6386 Frequency : occasional
6388 Description: OSTs running 2.4 kernels but with extents enabled might trip an
6389 assertion in the ext3 JBD (journaling) layer.
6390 Details : The b_committed_data struct is protected by the big kernel lock
6391 in 2.4 kernels, serializing journal_commit_transaction() and
6392 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
6393 access to this struct. In 2.6 kernels there is finer grained
6394 locking to improve SMP performance of the JBD layer.
6398 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
6400 -----------------------------------------------------------------------------
6402 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
6404 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
6405 versions of Lustre. Please update all servers and clients to
6406 version 1.4.2 or later at the same time. You must also run
6407 "lconf --write-conf {config}.xml" on the MDS while it is stopped
6408 to update the configuration logs.
6410 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
6411 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
6412 - swab llog records of type '0' so we get proper header size/idx (5861)
6413 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
6414 - fix rename of one directory over another leaking an inode (5953)
6415 - avoid SetPageDirty on 2.6 (5981)
6416 - don't re-add just-being-destroyed locks to the waiting list (5653)
6417 - when creating new directories, inherit the parent's custom
6418 striping settings if present parent (3048)
6419 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
6420 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
6421 - don't hold client locks on temporary worklist from l_lru (5666)
6422 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
6423 - automatically evict dead clients (5921)
6424 - Update file size properly in create+truncate+fstat case (6196)
6425 - Do not unhash mountpoint dentries, do not allow removal of
6427 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
6428 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
6429 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
6430 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
6431 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
6432 - Correctly handle reads of files with no objects (6243)
6433 - lctl recover will also mark a device active if deactivate used (5933)
6435 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
6436 - allow --write-conf on an MDS with different nettype than client (5619)
6437 - don't write config llogs to MDS for mounts not from that MDS (5617)
6438 - lconf should create multiple TCP connections from a client (5201)
6439 - init scripts are now turned off by default; run chkconfig --on
6440 lustre and chkconfig --on lustrefs to use them
6441 - upcalls are no longer needed for clients to recover to failover
6443 - add --abort-recovery option to lconf to abort recovery on device
6445 - add support for an arbitrary number of OSTs (3026)
6446 - Quota support protocol changes.
6447 - forward compatibility changes to wire structs (6007)
6448 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
6449 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
6450 - improved "lustre status" script
6451 - initialize blocksize for non-regular files (6062)
6452 - added --disable-server and --disable-client configure options (5782)
6453 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
6454 - Vanilla 2.4.29 support
6455 - increase maximum number of obd devices to 520 (6242)
6456 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
6457 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
6458 SLES 9 2.6.5 kernels
6459 - limit stripes per file to 160 (the maximum EA size) (6093)
6461 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
6464 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
6465 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
6466 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
6467 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
6468 - small ext3 extents cleanups and fixes (5733)
6469 - improved mballoc code, several small races and bugs fixed (5733, 5638)
6470 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
6471 - avoid needless client->OST connect, fix handle mismatch (5317)
6472 - fix DLM error path that led to out-of-sync client, long delays (5779)
6473 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
6474 - fix several locking issues related to i_size (5492,5624,5654,5672)
6475 - don't move pending lock onto export if it is already evicted (5683)
6476 - fix kernel oops when creating .foo in unlinked directory (5548)
6477 - fix deadlock in obdfilter statistics vs. object create (5811)
6478 - use time_{before,after} to avoid timer jiffies wrap (5882)
6479 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
6480 - Do not leak request if server was not able to process it (5154)
6481 - If mds_open unable to find parent dir, make that negative lookup(5154)
6482 - don't create new directories with extent-mapping (5909, 5936)
6484 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
6485 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
6486 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
6487 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
6488 - support for RHEL4 kernel on i686 (5773)
6489 - provide error messages when incompatible logs are encountered (5898)
6491 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
6492 * version 1.4.0.10 (1.4.1 release candidate 1)
6494 - don't keep a lock reference when lock is not granted (4238)
6495 - unsafe list practices (rarely) led to infinite eviction loop (4908)
6496 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
6497 - drop import inflight refcount on signal_completed_replay error (5255)
6498 - unlock page after async write error during send (3677)
6499 - handle missing objects in filter_preprw_read properly (5265)
6500 - no transno return for symlink open, don't save no-trasno open (3440)
6501 - don't try to complete elan receive that already failed (4012)
6502 - free RPC server reply state on error (5406)
6503 - clean up thread from ptlrpc_start_thread() on error (5160)
6504 - readahead could read extra page into cache that wasn't ejected (5388)
6505 - prevent races in class_attach/setup/cleanup/detach (5260)
6506 - don't dereference de->d_inode after l_dput of de (5458)
6507 - use "int" for stripe value returned from lock_to_stripe (5544)
6508 - mballoc allocation and error-checking fixes in 2.6 (5504)
6509 - block device patches to fix I/O request sizes in 2.6 (5482)
6510 - look up hostnames for IB nals (5602)
6511 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
6512 - don't start multiple acceptors for the same port (5277)
6513 - fix incorrect LASSERT in mds_getattr_name (5635)
6514 - export a proc file for general "ping" checking (5628)
6515 - fix "lfs check" to not block when the MDS is down (5628)
6517 - service request history (4965)
6518 - put {ll,lov,osc}_async_page structs in a single slab (4699)
6519 - create an "evict_client" /proc entry on OSTs, like the MDS has
6520 - fix mount usage message, return errors per mount(8) (5168)
6521 - change grep [] to grep "[]" in tests so they work in more UMLs
6522 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
6523 - remove ancient LOV_MAGIC_V0 EA support (5047)
6524 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
6525 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
6526 - allow readahead allocations to fail when low on memory (5383)
6527 - mmap locking landed again, after considerable improvement (2828)
6528 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
6530 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
6533 - send OST transaction number in read/write reply to free req (4966)
6534 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
6535 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
6536 - the watchdog thread now runs as interruptible (5246)
6537 - flock/lockf fixes (but it's still disabled, pending 5135)
6538 - don't use EXT3 constants in llite code (5094)
6539 - memory shortage at startup could cause assertion (5176)
6541 - reorganization of lov code
6542 - single portals codebase
6544 - add extents/mballoc support (5025)
6545 - direct I/O reads in the obdfilter (4048)
6546 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
6548 tbd Cluster File Systems, Inc. <info@clusterfs.com>
6551 - send OST transaction number in read/write reply to free req (4966)
6552 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
6553 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
6554 - the watchdog thread now runs as interruptible (5246)
6555 - handle missing objects in filter_preprw_read properly (5265)
6556 - unsafe list practices (rarely) led to infinite eviction loop (4908)
6557 - drop import inflight refcount on signal_completed_replay error (5255)
6558 - unlock page after async write error during send (3677)
6559 - return original error code on reconstructed replies (3761)
6560 - no transno return for symlink open, don't save no-trasno open (3440)
6562 - add pid to ldlm debugging output (4922)
6563 - bump the watchdog timeouts -- we can't handle 30sec yet
6564 - extra debugging for orphan dentry/inode bug (5259)
6566 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
6569 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
6570 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
6571 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
6572 - fix NULL dereference in /proc/sys/portals/routes (4827)
6573 - allow failed mdc_close() operations to be interrupted (4561)
6574 - stop precreate on OST before MDS would time out on it (4778)
6575 - don't send partial-page writes before EOF from client (4410)
6576 - discard client grant for sub-page writes on large-page clients (4520)
6577 - don't free dentries not owned by NFS code, check generation (4806)
6578 - fix lsm leak if mds_create_objects() fails (4801)
6579 - limit debug_daemon file size, always print CERROR messages (4789)
6580 - use transno after validating reply (3892)
6581 - process timed out requests if import state changes (3754)
6582 - update mtime on OST during writes, return in glimpse (4829)
6583 - add mkfsoptions to LDAP (4679)
6584 - use ->max_readahead method instead of zapping global ra (5039)
6585 - don't interrupt __l_wait_event() during strace
6587 - add software watchdogs to catch hung threads quickly (4941)
6588 - make lustrefs init script start after nfs is mounted
6589 - fix CWARN/ERROR duplication (4930)
6590 - return async write errors to application if possible (2248)
6591 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
6592 - print NAL number in %x format (4645)
6593 - update barely-supported suse-2.4.21-171 series (4842)
6594 - support for sles 9 %post scripts
6595 - support for building 2.6 kernel-source packages
6596 - support for sles km_* packages
6598 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
6601 - ignore -ENOENT errors in osc_destroy (3639)
6602 - notify osc create thread that OSC is being cleaned up (4600)
6603 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
6604 - reconstruct ost_handle() like mds_handle() (4657)
6605 - create a new thread to do import eviction to avoid deadlock (3969)
6606 - let lconf resolve symlinked-to devices (4629)
6607 - don't unlink "objects" from directory with default EA (4554)
6608 - hold socknal file ref over connect in case target is down (4394)
6609 - allow more than 32000 subdirectories in a single directory (3244)
6610 - fix blocks count for O_DIRECT writes (3751)
6611 - OST returns ENOSPC from object create when no space left (4539)
6612 - don't send truncate RPC if file size isn't changing (4410)
6613 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
6614 - bind to privileged port in socknal and tcpnal (3689)
6616 - rate limit CERROR/CWARN console message to avoid overload (4519)
6617 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
6618 - basic mmap support (3918)
6619 - kernel patch series update from b1_4 (4711)
6621 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
6624 - avoid crash during MDS cleanup with OST shut down (2775)
6625 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
6626 - don't use bad inodes on the MDS (3744)
6627 - dynamic object preallocation to improve recovery speed (4236)
6628 - don't hold spinlock over lock dumping or change debug flags (4401)
6629 - don't zero obd_dev when it is force cleaned (3651)
6630 - print grants to console if they go negative (4431)
6631 - "lctl deactivate" will stop automatic recovery attempts (3406)
6632 - look for existing locks in ldlm_handle_enqueue() (3764)
6633 - don't resolve lock handle twice in recovery avoiding race (4401)
6634 - revalidate should check working dir is a directory (4134)
6636 - don't always mark "slow" obdfilter messages as errors (4418)
6638 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
6641 - don't close LustreDB during write_conf until it is done (3860)
6642 - fix typo in lconf for_each_profile (3821)
6643 - allow dumping logs from multiple threads at one time (3820)
6644 - don't allow multiple threads in OSC recovery (3812)
6645 - fix debug_size parameters (3864)
6646 - fix mds_postrecov to initialize import for llog ctxt (3121)
6647 - replace config semaphore with spinlock (3306)
6648 - be sure to send a reply for a CANCEL rpc with bad export (3863)
6649 - don't allow enqueue to complete on a destroyed export (3822)
6650 - down write_lock before checking llog header bitmap (3825)
6651 - recover from lock replay timeout (3764)
6652 - up llog sem before sending rpc (3652)
6653 - reduce ns lock hold times when setting kms (3267)
6654 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
6655 - fix NULL deref and obd_dev leak on setup error (3312)
6656 - replace some LBUG about llog ops with error handling (3841)
6657 - don't match INVALID dentries from d_lookup and spin (3784)
6658 - hold dcache_lock while marking dentries INVALID and hashing (4255)
6659 - fix invalid assertion in ptlrpc_set_wait (3880)
6661 - add libwrap support for the TCP acceptor (3996)
6662 - add /proc/sys/portals/routes for non-root route listing (3994)
6663 - allow setting MDS UUID in .xml (2580)
6664 - print the stack of a process that LBUGs (4228)
6666 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
6669 - don't cleanup request in ll_file_open() on failed MDS open (3430)
6670 - make sure to unset replay flag from failed open requests (3440)
6671 - if default stripe count is 0, use OST count for inode size (3636)
6672 - update parent mtime/ctime on client for create/unlink (2611)
6673 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
6674 - free recovery state on server during a forced cleanup (3571)
6675 - unregister_reply for resent reqs (3063)
6676 - loop back devices mounting and status check on 2.6 (3563)
6677 - fix resource-creation race that can provoke i_size == 0 (3513)
6678 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
6679 - more debugging for page-accounting assertion (3746)
6680 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
6681 - avoid deadlock after precreation failure (3758)
6682 - fix race and lock order deadlock in orphan handling (3450, 3750)
6683 - add validity checks when grabbing inodes from l_ast_data (3599)
6685 - add /proc/.../recovery_status to obdfilter (3428)
6686 - lightweight CDEBUG infrastructure, debug daemon (3668)
6687 - change default OSC RPC parameters to be better on small clusters
6688 - turn off OST read cache for files smaller than 32MB
6689 - install man pages and include them in rpms (3100)
6690 - add new init script for (un)mounting lustre filesystems (2593)
6691 - run chkconfig in %post for init scripts (3701)
6692 - drop scimac NAL (unmaintained)
6694 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
6697 - clean kiobufs before and after use (3485)
6698 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
6699 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
6700 - change init script to fail more clearly if not run as root (1528)
6701 - allow clients to reconnect during replay (1742)
6702 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
6703 - don't do DNS lookups on NIDs too small for IP addresses (3442)
6704 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
6705 - fix cond_resched (3554)
6706 - only evict unfinished clients after recovery (3515)
6707 - allow bulk resend, prevent data loss (3570)
6708 - dynamic ptlrpc request buffer allocation (2102)
6709 - don't allow unlinking open directory if it isn't empty (2904)
6710 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
6711 - remove extraneous obd dereference causing LASSERT failure (3334)
6712 - don't use get_cycles() when creating temp. files on the mds (3156)
6713 - hold i_sem when setting i_size in ll_extent_lock() (3564)
6714 - handle EEXIST for set-stripe, set proper directory name (3336)
6716 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
6717 - fix ksocknal_fmb_callback() error messages (2918)
6719 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
6722 - don't copy lvb into (possibly NULL) reply on error (2983)
6723 - don't deref dentry after dput, don't free lvb on error (2922)
6724 - use the kms to determine writeback rpc length (2947)
6725 - increment oti_logcookies when osc is inactive (2948)
6726 - update client's i_blocks count via lvb messages (2543)
6727 - handle intent open/close of special files properly (1557)
6728 - mount MDS with errors=remount-ro, like obdfilter (2009)
6729 - initialize lock handle to avoid ASSERT on error cleanup (3057)
6730 - don't use cancelling-locks' kms values (2947)
6731 - use highest lock extent for kms, not last one (2925)
6732 - don't dereference ERR_PTR() dentry in error handling path (3107)
6733 - fix thread race in portals_debug_dumplog() (3122)
6734 - create lprocfs device entries at setup instead of at attach (1519)
6735 - common AST error handler, don't evict client on completion race (3145)
6736 - zero nameidata in detach_mnt in 2.6 (3118)
6737 - verify d_inode after revalidate_special is valid in 2.6 (3116)
6738 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
6739 - initialize RPC timeout timer earlier for 2.6 (3219)
6740 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
6741 - print nal/nid for unknown nid (3258)
6742 - additional checks for oscc recovery before doing precreate (3284)
6743 - fix ll_extent_lock() error return code for 64-bit systems (3043)
6744 - don't crash in mdc_close for bad permissions on open (3285)
6745 - zero i_rdev for non-device files (3147)
6746 - clear page->private before handing to FS, better assertion (3119)
6747 - tune the read pipeline (3236)
6748 - fix incorrect decref of invalidated dentry (2350)
6749 - provide read-ahead stats and refine rpc in flight stats (3328)
6750 - don't hold journal transaction open across create RPC (3313)
6751 - update atime on MDS at close time (3265)
6752 - close LDAP connection when recovering to avoid server load (3315)
6753 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
6754 - don't leak open file on MDS after open resend (3325)
6755 - serialize filter_precreate and filter_destroy_precreated (3329)
6756 - loop device shouldn't call sync_dev() for nul device (3092)
6757 - clear page cache after eviction (2766)
6758 - resynchronize MDS->OST in background (2824)
6759 - refuse to mount the same filesystem twice on same mountpoint (3394)
6760 - allow llmount to create routes for mounting behind routers (3320)
6761 - push lock cancellation to blocking thread for glimpse ASTs (3409)
6762 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
6763 - fix rare problem with rename on htree directories (3417)
6765 - allow default OST striping configuration per directory (1414)
6766 - fix compilation for qswnal for 2.6 kernels (3125)
6767 - increase maximum number of MDS request buffers for large systems
6768 - change liblustreapi to be useful for external progs like lfsck (3098)
6769 - increase local configuration timeout for slow disks (3353)
6770 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
6772 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
6775 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
6776 - don't overwrite extent policy data in reply if lock was blocked (2901)
6777 - drop filter export grants atomically with removal from device (2663)
6778 - del obd_self_export from work_list in class_disconnect_exports (2908)
6779 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
6780 - swab reply message in mdc_close, other PPC fixes (2464)
6781 - fix destroying of named logs (2325)
6782 - overwrite old logs when running lconf --write_conf (2264)
6783 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
6784 - fix race in target_handle_connect (2898)
6785 - mds_reint_create() should take same inode create lock (2926)
6786 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
6787 - don't close files for self_export to avoid uninitialized obd (2936)
6788 - allow MDS with the same name as client node (2939)
6789 - hold dentry reference for closed log files for unlink (2325)
6790 - reserve space for all logs during transactions (2059)
6791 - don't evict page beyond end of stripe extent (2925)
6792 - don't oops on a deleted current working directory (2399)
6793 - handle hard links to targets without a parent properly (2517)
6794 - don't dereference NULL lock when racing during eviction (2867)
6795 - don't grow lock extents when lots of conflicting locks (2919)
6797 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
6800 - account for cache space usage on clients to avoid data loss (974)
6801 - lfsck support in lustre kernel code (2349)
6802 - reduce journal credits needed for BRW writes (2370)
6803 - orphan handling to avoid losing space on client/server crashes
6804 - ptlrpcd can be blocked, stopping ALL progress (2477)
6805 - use lock value blocks to assist in proper KMS, faster stat (1021)
6806 - takes i_sem instead of DLM locks internally on obdfilter (2720)
6807 - recovery for initial connections (2355)
6808 - fixes for mds_cleanup_orphans (1934)
6809 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
6810 - block all file creations until orphan recovery completes (1901)
6811 - client remove rq_connection from request struct (2423)
6812 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
6813 - recovery timer race (2670)
6814 - mdc_close recovey bug (2532)
6815 - ptlrpc cleanup bug (2710)
6816 - mds timeout on local locks (2588)
6817 - namespace lock held during RPCs (2431)
6818 - handle interrupted sync write properly (2503)
6819 - don't try to handle a message that hasn't been replied to (2699)
6820 - client assert failure during cleanup after abort recovery (2701)
6821 - leak mdc device after failed mount (2712)
6822 - ptlrpc_check_set allows timedout requests to complete (2714)
6823 - wait for inflight reqs when ptlrpcd finishes (2710)
6824 - make sure unregistered services are removed from the srv_list
6825 - reset bulk XID's when resending them (caught by 1138 test)
6826 - unregister_bulk after timeout
6827 - fix lconf error (2694)
6828 - handle write after unfinished setstripe, stripe-only getstripe (2388)
6829 - readahead locks pages, leaves pending causing memory pressure (2673)
6830 - increase OST request buffers to 4096 on large machines (2729)
6831 - fix up permission of existing directories in simple_mkdir (2661)
6832 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
6833 - don't assign transno to errored transactions (2742)
6834 - don't delete objects on OST if given a bogus objid from MDS (2751)
6835 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
6836 - if rq_no_resend, then timeout request after recovery (2432)
6837 - fix MDS llog_logid record size, 64-bit array alignment (2733)
6838 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
6839 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
6840 - MDS recovery shouldn't delete objects that it has given out (2730)
6841 - if enqueue arrives after completion, don't clobber LVB (2819)
6842 - don't unlock pages twice when trigger_group_io returns error (2814)
6843 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
6844 - don't write pages to disk if there was an error (1450)
6845 - don't ping imports that have recovery disabled (2676)
6846 - take buffered bytes into account when balancing socknal conn (2817)
6847 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
6848 - reconnect unlink llog connection after MDS reconnects to OST (2816)
6849 - remove little-endian swabbing of llog records (1987)
6850 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
6851 - retry reposting request buffers if they fail (1191)
6852 - grow extent at grant time to avoid granting a revoked lock (2809)
6853 - lock revoke doesn't evict page if covered by a second lock (2765)
6854 - disable VM readahead to avoid reading outside lock extents (2805)
6856 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
6857 - updated kernel patches for hp-2.4.20 kernel (2681)
6859 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
6862 - fix truncated write corruption (2366)
6863 - fix for failed assertion in iopen_connect_dentry (1792,2517)
6865 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
6866 - ensure len(uuid) < 37 in lmc (1171)
6867 - fix ia64 OOPS in llog_test (2255)
6868 - zero end of page at obdfilter for partial page writes (2648)
6869 - don't leave stale dentries around after renames (bug 2428)
6870 - fix timeouts when evicting a client with a single lock held (2642)
6871 - set deadline for the initial HELLO message to drain (2634)
6872 - print out dotted-quad IP addresses in the socknal (2302)
6874 - additional debugging for MDS client eviction problem (2443)
6875 - fix mkfsoptions support for osts (2603, 2604)
6877 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
6880 - add series for the vanilla 2.6.0 kernel
6881 - add series for the vanilla 2.4.24 kernel
6882 - add series for a cray x86/64 UL kernel drop
6883 - fix xattr patches for the vanilla 2.4.19 series
6885 - generate true UUIDs in lmc (1171)
6886 - have portals stack dumping break in UML (2466)
6887 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
6888 - allocate with _NOFS in ldlm to avoid deadlock (1933)
6889 - wake callback waiting threads on client eviction (2460)
6890 - Add --ptldebug and --subsystem to lmc (1719)
6891 - update assertion to allow safe interrupt allocation
6892 - set rq_no_resend for cancel requests (2432)
6893 - recalculate ptlrpcd timeout after resend (2494)
6894 - call vfs_rmdir when removing pending directories (2368)
6895 - fix renaming a file to itself (2429)
6896 - lmc creates a default one-stripe lov (2454)
6897 - expand procfs space to handle large clusters (2326)
6898 - increase UML stack to avoid overflow
6899 - update lconf's list of debug and subsystem masks
6900 - fix lfs find --obd (2510)
6901 - /proc tunable for disabling filter read caching (2591)
6902 - stop rpm packages from altering slapd.conf (2301)
6903 - disable nagle in the socknal under 0conf (2578)
6904 - choose mds inode size based on stripe count (2572)
6905 - fix kernel-source rpm problems (2516)
6907 - add --disable-doc to avoid pdf generation (2421)
6908 - update documentation, tests, type-os, comments
6909 - avoid format warnings on ia64
6910 - remove the TOE NAL
6911 - tiny code cleanups by removing unused fields
6913 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
6916 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
6917 - remove the most blatant lies from BUILDING file (2371)
6918 - change default debug level to reasonable production setting
6919 - reduce client side cache size to reduce cache flush time
6920 - reduce max RPCs in flight to avoid unnecessary file fragmentation
6921 - make TCP zerocopy and pinger support enabled by default (2476)
6922 - sync writes completed after process exits caused crashes (2319)
6923 - maintain correct mount count on the MDS (2356)
6924 - backout 1557, because 2316 wasn't really fixed
6925 - better file I/O statistics gathering in /proc
6926 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
6927 - another kernel patch to fix zero-copy TCP function export
6928 - don't take duplicate lock when processing re-sent getattr (2420)
6929 - lctl uses obd_self_export instead of creating new conn (2353)
6930 - MDS/OST recovery case which requires object creation asserted (2425)
6931 - move lfs from /usr/sbin to /usr/bin in packages
6932 - fix race between mds_client_add and mds_client_free (2417)
6933 - use kmalloc instead of slabs in portals (2430)
6934 - don't create duplicate records when a failover MDS is present (2442)
6935 - remove unnecessary mount age check (2332)
6936 - don't remove directory inodes from locks prematurely (2451)
6937 - don't break if MDS service name is the same as hostname (2103)
6938 - fix races in client write RPC generation when cache full (2482)
6940 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
6943 - remove now-unused request->rq_obd (278)
6944 - if an allocation fails, print out how much memory we've used (1933)
6945 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
6946 - add the "configurable stack size" patch to most series files (1256)
6947 - ability to write large log records, for 100+ OST configs (2306)
6948 - fix NULL deref when filter_prep fails (2314)
6949 - fix operator precedence error in filter_sync
6950 - dynamic allocation of socknal TX descriptors (2315)
6951 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
6952 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
6953 - ensure that CWARN messages in Portals always get to the syslog
6954 - __init/__exit are not for prototype decls (ldlm_init/exit)
6955 - x86-64 compile warning fixes
6956 - fix gateway LMC keyword conflict (2318)
6957 - fix MDS lock inversions in getattr/reint paths (1844)
6958 - fix a rare lock re-ordering bug, which caused deadlock (2322)
6959 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
6960 - DLM race condition prevented some lock evictions (2328)
6961 - ENOMEM detection and retry on socknal sends (2230)
6962 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
6963 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
6964 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
6965 - better error messages when a client is rejected during recovery (1505)
6966 - avoid cancelling locks which were never granted, after failure (2330)
6967 - fix i_sem/journal inversion in mds_client_add (2333)
6968 - fix truncate/getattr lock cycle deadlock (2334)
6969 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
6970 - fix two rare exit paths which could leak an l_lock() ref (2321)
6971 - fencepost error in MDS/OST orphan recovery (2226)
6972 - make log record alignment 8 bytes (1988)
6973 - lstripe now fails when requested offset > ost_count (2237)
6974 - ensure that all kernel series have a complete list.h (1607)
6975 - fix crashes in special-file operations (2316)
6976 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
6978 - allow configurable automake binary, for testing new versions
6979 - small update to the lfs documentation
6981 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
6983 * fix negative export reference count in fsfilt_sync (2312)
6985 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
6986 * release candidate 0.9.1
6988 - orphans are moved into the PENDING directory for possible recovery
6989 - replayed opens now open by fid for orphan/rename safety (1042)
6990 - last close of an orphan inode generates a transno (683)
6991 - chdir() and mount() now pin the directory entry (1020)
6992 - avoid CERROR in normal ll_setattr_raw() error case (1500)
6993 - discard very old requests without processing them (1502)
6994 - remove some common, well-understood CERRORs (1505)
6995 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
6996 - clear "grant" flags in OST replies until OST grant code lands (1644)
6997 - fix read performance by not clobbering i_blksize on client (1598)
6998 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
6999 - make LRU size a /proc tunable, clears locks when reduced (707)
7000 - fix some lprocfs rot that prevented ptlbd from loading (1732)
7001 - server locks take references on exports now (1558)
7002 - build fixes for 2.4.20-rh trees (1663)
7003 - return an error from lov_create if all OSCs are inactive (1751)
7004 - fix import levels when a reconnect happens without a timeout (1597)
7005 - exit early from mds_open if we get a lookup error (1749)
7006 - partial page read at EOF wouldn't wait for disk before sending (1642)
7007 - avoid NULL deref in obdfilter when reading page past EOF (1592)
7008 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
7009 - fix LBUG in ll_it_open_error with rc = -2 (1861)
7010 - write/truncate lock inversion (1639)
7011 - Don't auto-load obdclass, portals modules during cleanup (1495)
7012 - fix timestamps from jumping to "now" (1763)
7013 - extra journal assertions (1648)
7014 - add an extra multiunlink test (1771)
7015 - fix read_record/write_record API (1776)
7016 - fix leak of offset_extent, possible incorrect i_size later (1772)
7017 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
7018 - Debugging for the kqswnal_get_idle_tx problems (1820)
7019 - Allow recovery to be attempted multiple times (1536)
7020 - Write out MDS last_rcvd file after it is first created (1600)
7021 - Fix tx_descriptor leak in failed transmit situations (1827)
7022 - ext3 journaling fixes for assertion failure after IO error (1871)
7023 - class_export_put() on freed export after completion AST error (1896)
7024 - Fix revalidate looping in VFS (1322)
7025 - Don't access a freed export during MDS_REINT timeout (1521)
7026 - Add open-unlink recovery support on the MDS (1673,1764)
7027 - Return an error if no MDS data was read from last_rcvd (1946)
7028 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
7029 - Don't setup a disk device that doesn't match exported UUID (317)
7030 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
7031 - avoid committing NULL handle in force close
7032 - local.sh is now a one-stripe LOV configuration
7033 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
7034 - fix ext3 htree duplicate directory entry corruption (1516)
7035 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
7036 - update to new LOV EA format (2097)
7037 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
7039 2003-06-15 Phil Schwan <phil@clusterfs.com>
7042 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
7043 - per-import/export recovery handling (958, 931, 959)
7044 - multiple last-rcvd slots, for serving multiple FSes (949)
7045 - connections are again shared between multiple imp/exports (963, 964)
7046 - "umount -f" would hang if any requests needed to be sent (393, 978)
7047 - avoid pinning large req buffer by copying for queued messages (989)
7048 - add "uuid" to "lctl device" command to help upcalls (991)
7049 - "open" RPCs with transnos would confuse recovery counters (1037)
7050 - do proper endian conversion of all wire messages (288, 340, 891)
7051 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
7052 - call waiting locks callback from LDLM recovery thread (1127, 1151)
7053 - fix ptlrpc_connection leak in target_handle_connect (1174)
7054 - fix import refcounting bug in OST and MDS cleanup (1134)
7055 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
7056 - fix very unlikely obd_types race condition (501)
7057 - remove osc_open hack for echo_client (1187)
7058 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
7059 - a failure in read_inode2 leads to deadlock (1139)
7060 - cancel ack-locks as soon as transaction is committed (1072)
7061 - fix major leaks and crashes in the bulk I/O path (937, 1057)
7062 - make sure to commitrw after any preprw to avoid deadlock (1162)
7063 - failing to execute a file in a lustre FS would lock inode (1203)
7064 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
7065 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
7066 - obd_uuid2tgt would walk off the end of the list (1255)
7067 - on IA64 the osc would give portals incorrect bulk size (1258)
7068 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
7069 - fix lock inversion caused by new llite matching code (1282)
7070 - limit the number of dirty pages on a client to 10MB (1286)
7071 - timed out locks were not being corrected cancelled (1289)
7072 - fix O_DIRECT above 4GB on IA-32 (1292)
7073 * major user-visible changes
7074 - fail out/fail over policy now controlled by the upcall (993)
7076 - add OBD_PING to check server availability and failure (954)
7077 - lustre messages are now sent in sending host order (288, 340, 891)
7078 - add eadatalen to MDS getattr reply (340)
7079 - OST read replies may contain second buffer, with per-page status (593)
7081 2003-03-11 Phil Schwan <phil@clusterfs.com>
7084 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
7085 - failed open()s could cause deadlock; fixed (867, 869)
7086 - stop cancelling OST locks when files are closed (481)
7087 - overlapping XID spaces caused network corruption (851, 853)
7088 - fix unsafe fsfilt counter arithmetic; change to atomic_t
7089 - setattr_raw added, to do single-RPC, server-side setattrs
7090 - lmc/lconf syntax change for OST UUIDs
7091 - fix crashy race condition between ptlrpc_free_req and osc_close
7092 - don't use request in mdc_enqueue if we hit a timeout (889)
7093 - don't set the inode i_size for regular files from the MDS (896)
7094 - handle out of order completion AST (842)
7095 - don't LBUG if a lock request times out after receiving AST (913)
7096 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
7097 - if a bad lock AST arrives, send an error instead of dropping entirely
7098 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
7099 - fix leak in bulk IO when only partially completed (899, 900, 926)
7100 - fix O_DIRECT for ia64 (55)
7101 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
7102 - C-z after timeout could hang a process forever; fixed (977)
7104 - client-side I/O cache (678, 924, 929, 941, 970)
7106 - READPAGE and SETATTRs which don't take server-side locks get
7109 2003-02-11 Phil Schwan <phil@clusterfs.com>
7112 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
7113 - Fix fsfilt_extN_readpage() to read a full page of directory
7114 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
7115 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
7116 - Fix imbalanced LOV object allocation and out-of-bound access (469)
7117 - Most intent operations were removed, in favour of a new RPC mode
7118 that does a single RPC to the server and bypasses most of the VFS
7119 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
7120 - Aggressively cancel local locks on DLM servers
7121 - mds_reint_unlink sends EA to the client if it's the last nlink.
7122 client uses that EA to unlink OST objects.
7123 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
7124 - recursive symlinks were fixed (439)
7125 - fixed NULL deref in DEBUG_REQ
7126 - filter_update_lastobjid no longer calls sync, which annoyed extN
7127 - fixed multi-client small-writes to a single file problem (445)
7128 - fixed mtime updates during file writes (607)
7129 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
7130 - fixed bug in obd_brw_read/write() (under guise of testing 367)
7131 - fixed Linux OST size reporting problem (444, 656)
7132 - OST now updates object mtime with writes or setattr (607, 619)
7133 - client verifies file size before zeroing page past EOF (445)
7134 - OST now writes last allocated objid to disk with allocation (108)
7135 - LOV on echo now works (409)
7137 - mds_reint_unlink sends a new buffer, with the EA included. this
7138 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
7139 set if a regular file was being unlinked, and it was the last link
7140 - use PtlGet from the target for bulk writes (315)
7141 - OST now updates object mtime with writes or setattr (607, 619)
7142 - LDLM now has a grant-time callback to revalidate locked items, if
7144 - Many MDS operations were reorganized to combat race conditions
7146 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
7147 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
7149 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
7152 - Fully reactivate OST imports after reconnection (512, others)
7153 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
7154 - More graceful error handling for truncating on dead OST (515)
7155 - Don't error out unless we're actually accessing dead stripes (474)
7156 - Fix garbage sizes when stripes are missing (410)
7157 - LRU counters were broken, causing constant lock purge (433, 432)
7158 - garbage on read from stripes with failed OSTs (441)
7159 - mark OSCs as active before reconnecting during recovery (438)
7160 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
7161 - lfind did not preserve OST order in output (443)
7162 - symlinks cause hung clients, incorrect data (439)
7163 - stop dereferencing request after dropping refcount (457)
7164 - don't LASSERT(spin_is_locked) on non-SMP (455)
7165 - fixes for many rename() bugs
7166 - fstat didn't correctly synchronize attributes (399)
7167 - server must handle lock cancellation during blocking AST prep (487)
7168 - bulk descriptors were free()d too soon (511)
7169 - fix paths in lconf, which would load incorrect modules (451, 507)
7170 - fix confusing lconf 'host not found' error message (386)
7171 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
7172 - fix race condition in mdc_blocking_ast() for inode access (526)
7173 - fix lov_unpackmd() unpacking wrong number of stripes (537)
7174 - fix lov_set_osc_active() marking wrong OSC inactive (440)
7175 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
7176 - fix multiple writes of stripe MD to MDS (358, maybe 519)
7177 - fix lstripe in several ways (kernel side) (527)
7178 - fix request leak in ldlm_cli_enqueue (262)
7179 - incorrect OSC was marked inactive after OST failure
7180 - call mds_fs_cleanup before unmounting filesystem (524)
7181 - fix races between taking ns_lock and ldlm_lock_change_resource
7182 - fix races updating LOV export open file list
7183 - fix lov_enqueue error path, avoid decref-ing bad lock handle
7184 - fix recovery NULL deref in ldlm_cli_cancel_unused
7185 - fix some DLM races by using new hash table for lock handles (419)
7186 - permit the client to specify desired inodes, at replay
7187 - duplicate requests when we queue them for replay reintegration
7188 - fix last_rcvd offset calculation
7189 - sync after each recovered transaction, so we always make progress
7190 - never, not always, ERESTART requests without transnos
7191 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
7193 - skip replay if the MDS says that the client is already connected
7194 - don't check for a recovery-enabled export to match lctl's UUID
7195 - don't INC_USE_COUNT for phantom exports
7196 - don't crash when cleaning up phantom exports (567)
7197 - don't double-finish or set replay data for errored mdc_open requests
7198 - abort requests when they time out, so we don't get old replies
7199 - send/receive replies for AST messages again
7200 - if the client says that it doesn't have the lock, cancel it on the
7202 - if we timeout during I/O, don't try to cancel an in-use lock; instead
7203 mark it as destroyed, it will all work out when decref is called
7204 - fix module use counts (22, 581)
7206 - ASTs now expect a reply (server cancels lock on error reply)
7208 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
7211 - fix many simultaneous client startup (392)
7212 - fix dentry->d_it clobbering
7213 - credentials weren't being shipped for readdir/getattr operations
7214 - remove invalid assertions triggered during some concurrent MD
7216 - proper Lustre versions added (336, 389)
7217 - fix memory leak for create error case (398)
7218 - fix LOV locking bug that would get cli/srv out of sync
7219 - fix echo client over LOV (409)
7220 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
7221 - fix double-O_EXCL intent crash (424)
7222 - avoid sending multiple lock CANCELs (352)
7224 - MDS can do multi-client recovery (modulo bugs in new code)
7226 - many updates, edits, cleanups
7228 2002-11-18 Phil Schwan <phil@clusterfs.com>
7231 - fix null d_it dereference (346)
7232 - fix full OST/dbench hang (333)
7233 - fix permission problem with file removal (286)
7234 - fix removal of OSCs from LOV when they fail
7235 - fix NULL deref during bulk timeout (214)
7236 - fix problems related to multiple filesystems on one MDS (241)
7237 - fixed serious subtle metadata locking bugs
7238 - free locks on clients when inodes are removed due to memory
7240 - fix inode pointer in lock data (285)
7241 - partial support for multiple MDS on a single host (241)
7242 - data locks weren't cancelled at clear_inode time (290, 311)
7243 - intent locks could lead to unbounded lock growth (205)
7244 - added a maximum lock count, an LRU list, and a flusher
7245 - fix multiple rename (365)
7246 - properly abstracted the echo client
7247 - OSC locked 1 byte too many; fixed
7248 - rewrote brw callback code:
7249 - fixed recovery bugs related to LOVs (306)
7250 - fixed too-many-pages-in-one-write crash (191)
7251 - fixed (again) crash in sync_io_timeout (214)
7252 - probably fixed callback-related race (385)
7254 - Add capability to MDS protocol
7255 - LDLM cancellations and callbacks on different portals
7257 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
7260 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
7261 - defer open object destruction to close time (601981, 138)
7262 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
7263 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
7264 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
7265 - don't call obd_setattr in ll_file_release for destroyed objects
7266 * protocol change to lustre_msg: move |version| and add |flags|
7267 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
7268 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
7269 * added replay of create, unlink, link and rename operations during
7270 MDS failover; recovery should be much more robust now
7271 * remove failed OSCs from LOVs (only lov_create uses this so far)
7272 * the lustre-HOWTO was brought (more) up to date (582544)
7274 2002-10-23 Phil Schwan <phil@clusterfs.com>
7277 - in-use dentries weren't being reused properly (617851)
7278 - prevent multiple LDLM setup (599178)
7279 - fix LOV size calculations for truncate (617853)
7280 - fix client handling of MDS intent errors (POSIX)
7281 - fix permission bug in lovstripe.c test (624321)
7282 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
7283 - truncate past end of file could corrupt data
7284 - proper cleanup after timeouts, crashes, etc (592524, 550815)
7285 - a race in recovery could return ETIMEDOUT to apps (623947)
7286 - building outside the source directory was fixed
7287 * the lustre-HOWTO was brought (more) up to date (582544)
7288 * major progress was made on recovery functionality
7290 2002-10-10 Phil Schwan <phil@clusterfs.com>
7293 - recovery deadlock fix
7294 - rm -rf causes LBUG fix (617817)
7295 - file open by multiple tasks fix (618962)
7296 - directory permissions bugs (602707 and 620007)
7297 - journal_stop fixed with locking (611313)
7298 - O_APPEND failures resolved (618273, perhaps 614459)
7299 - lconf PATH fix (619770)
7300 - IA64 build fix (621450)
7301 - RPC buffer sizes scale with amount of memory
7303 2002-10-01 Phil Schwan <phil@clusterfs.com>
7306 - locks would be cancelled without throwing away data pages,
7307 resulting in inconsistent data (605627)
7308 - inode attributes were not always being refreshed (605627, 612449)
7309 - lconf now continues to cleanup after lctl reports an error
7310 - MDS now enforces user permissions (602707)
7311 - lprocfs cleanup fixed, but not yet enabled (614157)
7312 - fixed infinite server hang, should a client not respond to an AST
7313 - avoid going into recovery if user calls readlink() with a buffer
7314 that's too small (613941)
7315 - AST RPCs no longer require replies (614867) -- this may be changed
7316 - don't crash server if client sends an IOV that's too big (611336)
7317 - fixed lock conversion deadlock (611892)
7318 - fixed the following of symlinks (614622)
7319 * recovery: the server can remove locks from a client that dies, other
7320 clients can make progress
7321 * more extN patch fixes
7322 * compile-time configurable ptlrpc buffer allocations
7324 - collaborative read cache document
7325 - Lustre Lite Performance CDR document-in-progress
7327 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
7330 - fix typo in patch-2.4.18
7332 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
7335 - clear ptlrpc request each time in handle_incoming_request()
7336 - unlink of files now destroys the object on the OST
7338 2002-09-19 Peter Braam <braam@clusterfs.com>
7340 * add hard link support
7341 * change obdfile creation method
7342 * kernel patch changed
7344 2002-09-19 Peter Braam <braam@clusterfs.com>
7347 - stack overflow bug in extN fixed
7349 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
7351 * documentation updates
7352 - add man pages for config tools
7353 - update tests/README to describe testing with new config tools
7354 - finish metadata API descriptions
7355 * bug fixes and cleanups
7356 - statfs workaround for 16TB limit
7357 - LOV stripe allocation improved, can stripe on subset of OSTs
7358 - LOV file size/IO offset was wrong for files > 4GB in size
7359 - object EA data was being dropped, caused files to be unreadable
7360 - memory overflow with non-LOV OST caused memory corruption
7361 - fixed regression tests to work with new config tools, obdfilter
7362 - fixed bug when directory size became larger than 1 block
7363 - fixed bug (for single client case) when PWD was deleted
7364 - invalidate local directory pages when doing intent-based ops
7365 - avoid LDLM oops when lock callback contained bad data
7367 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
7369 * documentation updates
7370 * bug fixes and cleanups
7371 - configuration tools
7374 - 64-bit compile warnings
7375 - 64-bit internal statfs data
7377 * test_brw on persistent OST devices
7379 * lprocfs (disabled)
7381 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
7383 * documentation updates
7384 * bug fixes and cleanups
7385 * configuration tools
7387 2002-08-30 Peter J. Braam <braam@clusterfs.com>
7390 * many small fixes to 0_5_4
7391 * io/network handling
7392 * thinkos in MDS operations
7394 2002-08-24 Peter J. Braam <braam@clusterfs.com>
7397 * crucial basic fixes to 0.5.3
7398 * IOR, Iozone work over Elan
7401 2002-08-07 Phil Schwan <phil@clusterfs.com>
7402 * version 0_5_3, our first alpha
7403 * we use the new Portals iovs
7404 * documentation updates
7405 * bug fixes and cleanups
7406 * small changes in the DLM wire protocol
7408 2002-07-25 Peter J. Braam <braam@clusterfs.com>
7409 * version 0_5_1 with some initial stability,
7410 * locking on MD and file I/O.
7411 * documentation updates
7412 * several bug fixes since 0.5.0
7413 * small changes in wire protocol
7415 2002-07-18 Phil Schwan <phil@clusterfs.com>
7417 * delivered as Lustre Light Alpha
7418 * fixed a crash after handling invalid MDS requests
7419 * fixed directory pages for architectures with non-4k pages sizes
7421 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
7422 * release version v0_4_4
7423 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
7425 * Fixes a number of interruption problems with OST operations.
7426 * Update documentation for portals header changes
7427 * Move all wire protocol structs/defines to lustre_idl.h
7428 * Fixes symlink length bug.
7429 * Add tcpdump to repository.
7431 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
7432 * release version v0_4_3
7433 * Fixes statfs for inodes on extN.
7434 * Fixes bug in runtests which would delete /etc/hosts.
7435 * Use 64-bit object IDs wherever possible (not into VFS though)
7436 Remove ost_get_info, which is unused by lustre, and out of date.
7438 2002-07-03 Peter Braam <braam@clusterfs.com>
7439 * release version v0_4_2 Fixes a lookup error (type not passed)
7440 * move forward to head of Portals
7441 * move forward to latest Lustre kernel
7443 2002-06-25 Peter Braam <braam@clusterfs.com>
7444 * release version v0_4_1. Hopefully stable on single node use.